Chalcogenide Photovoltaics

Scheer · Schock T his first comprehensive description of the most important material properties and device aspects clos...

Author: Scheer R.

284 downloads 1421 Views 3MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Scheer · Schock

T

his first comprehensive description of the most important material properties and device aspects closes the gap between general books on solar cells and journal articles on chalcogenide-based photovoltaics. Written by two very renowned authors with years of practical experience in the field, the book covers II-VI and I-III-VI2 materials as well as energy conversion at heterojunctions. It also discusses the latest semiconductor heterojunction models and presents modern analysis concepts. Thin film technology is explained with an emphasis on current and future techniques for mass production, and the book closes with a compendium of failure analysis in photovoltaic thin film modules. With its overview of the semiconductor physics and technology needed, this practical book is ideal for students, researchers, and manufacturers, as well as for the growing number of engineers and researchers working in companies and institutes on chalcogenide photovoltaics.

Roland Scheer, Hans-Werner Schock

Chalcogenide Photovoltaics Physics, Technologies, and Thin Film Devices

Hans-Werner Schock became head of the Institute of Technology at the Helmholtz Zentrum Berlin für Materialien und Energie (former Hahn-Meitner Institute) in the division “Solar Energy Research” in late 2004. He received his diploma in electrical engineering in 1974 and obtained his PhD in electrical engineering from Stuttgart University, Germany, in 1986. Starting in the early 70s, he has taken the development of chalcogenide solar cells from basic investigations to the transfer to a pilot fabrication plant. From 1986 to 2003 he coordinated the research on chalcopyrite based solar cells in the framework of the European photovoltaic program. From 1982 to 2004 he was head of the compound semiconductor thin film group of the Institute of Physical Electronics at the University of Stuttgart. He is author or co-author of more than 300 contributions in books, scientific journals and published conference proceedings. For his achievements in the development of chalcopyrite based solar cells he received the prestigious “Becquerel Prize” in 2010.

www.wiley-vch.de

Chalcogenide Photovoltaics

Roland Scheer received his diploma degree in electronic engineering from the University of Applied Sciences Berlin, Germany. He joined AEG Konstanz, Germany, in 1983 as electronic engineer. In parallel, he studied physics at the University of Konstanz and at the Technical University of Berlin where he received his physics diploma in 1990. In 1994, he finished his doctoral thesis in physics and joined the Helmholtz-Centre (former Hahn-Meitner Institute) in Berlin. His R&D activities are focused on thin film solar cells where he invented a new solar cell based on CuInS2 . In 2002, Roland Scheer was visiting scientist at the Advanced Institute for Science and Technology (AIST) in Tsukuba, Japan. He was lecturer at the University of Potsdam until in 2010 he became full professor holding the endowed chair for photovoltaics at the Martin-Luther-University in Halle-Wittenberg.

Roland Scheer and Hans-Werner Schock Chalcogenide Photovoltaics

Related Titles Klauk, H. (ed.)

Brabec, C., Scherf, U., Dyakonov, V. (eds.)

Organic Electronics II

Organic Photovoltaics

More Materials, Manufacturing and Applications

Materials, Device Physics, and Manufacturing Technologies

2011

2008

ISBN: 978-3-527-32647-1

ISBN: 978-3-527-31675-5

Klauk, H. (ed.)

Mullen, ¨ K., Scherf, U. (eds.)

Organic Electronics

Organic Light Emitting Devices

Materials, Manufacturing and Applications

Synthesis, Properties and Applications

2006

2006

ISBN: 978-3-527-31264-1

ISBN: 978-3-527-31218-4

Luque, A., Hegedus, S. (eds.)

Brendel, R.

Handbook of Photovoltaic Science and Engineering

Thin-Film Crystalline Silicon Solar Cells

2nd Ed.

Physics and Technology

2011

2002

ISBN: 978-0-470-721698-8

ISBN: 978-3-527-40376-9

Wurfel, ¨ P.

Physics of Solar Cells From Basic Principles to Advanced Concepts 2009 978-3-527-40857-3

Roland Scheer and Hans-Werner Schock

Chalcogenide Photovoltaics Physics, Technologies, and Thin Film Devices

Prof. Dr. Roland Scheer Martin-Luther Universität Halle-Wittenberg Naturwissenschaftliche Fakultät II Institut für Physik/Fachgruppe Photovoltaik Von Dankelmann-Platz 3 06120 Halle (Saale) Germany Prof. Dr. Hans-Werner Schock Helmholtz-Zentrum Berlin Glienicker Straße 100 14109 Berlin Germany

All books published by Wiley-VCH are carefully produced. Nevertheless, authors, editors, and publisher do not warrant the information contained in these books, including this book, to be free of errors. Readers are advised to keep in mind that statements, data, illustrations, procedural details or other items may inadvertently be inaccurate. Library of Congress Card No.: applied for British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at .  2011 WILEY-VCH Verlag & Co. KGaA, Boschstr. 12, 69469 Weinheim, Germany

All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form – by photoprinting, microfilm, or any other means – nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not specifically marked as such, are not to be considered unprotected by law. Typesetting Laserwords Private Ltd., Chennai, India Printing and Binding Fabulous Printers Pte Ltd Singapore Cover Design Grafik-Design Schulz, Fußgönheim Printed in Singapore Printed on acid-free paper ISBN: 978-3-527-31459-1

V

Contents Preface XI Symbols and Acronyms XIII 1 1.1 1.1.1 1.2 1.2.1 1.3

Introduction 1 History of Cu(In,Ga)(S,Se)2 Solar Cells 1 Milestones of Cu(In,Ga)(S,Se)2 Development 3 History of CdTe Solar Cells 5 Milestones of CdTe Development 6 Prospects of Chalcogenide Photovoltaics 7

2 2.1 2.2 2.3 2.3.1 2.3.2 2.3.3 2.3.4 2.3.5 2.3.6 2.3.7 2.4 2.4.1 2.4.2 2.4.3 2.4.4 2.4.5 2.4.5.1 2.4.5.2 2.4.5.3 2.4.5.4 2.4.6 2.4.6.1

Thin Film Heterostructures 9 Energies and Potentials 9 Charge Densities and Fluxes 11 Energy Band Diagrams 13 Rules and Conventions 13 Absorber/Window 17 Absorber/Buffer/Window 20 Interface States 24 Interface Dipoles 29 Deep Bulk States 29 Bandgap Gradients 32 Diode Currents 36 Superposition Principle and Shifting Approximation 36 Regions of Recombination 37 Radiative Recombination 40 Auger Recombination 43 Defect Related Recombination 44 SCR Recombination 50 QNR Recombination 60 Back Surface Recombination 62 Interface Recombination 63 Parallel Processes 73 SCR and QNR Recombination 73

Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

VI

Contents

2.4.6.2 2.4.7 2.4.8 2.4.9 2.5 2.5.1 2.5.2 2.5.3 2.5.4 2.5.4.1 2.5.4.2 2.5.4.3 2.5.4.4 2.5.4.5 2.5.4.6 2.5.5 2.5.6 2.5.7 2.5.7.1 2.5.7.2 2.5.7.3 2.6 2.6.1 2.6.1.1 2.6.1.2 2.6.1.3 2.6.2 2.6.2.1 2.6.2.2 2.6.2.3 2.6.2.4 2.6.3 2.6.4

SCR and IF Recombination 75 Barriers for Diode Current 76 Bias Dependence 78 Non-Homogeneities 79 Light Generated Currents 80 Generation Currents 81 Generation Function 84 Photo Current 86 Collection Function 87 Absorber Quasi Neutral Region 87 QNR with Graded Bandgap 90 QNR with Back Surface Field 91 Absorber Space Charge Region 92 Buffer Layer 94 Simulating the Collection Function 95 Quantum Efficiency and Charge Collection Efficiency 96 Barriers for Photo Current 97 Voltage Dependence of Photo Current 99 Width of SCR 99 Interface Recombination 100 Photo Current Barriers 100 Device Analysis and Parameters 101 Equivalent Circuits 101 DC Equivalent Circuit 101 AC Equivalent Circuit 103 Module Equivalent Circuit 105 Current–Voltage Analysis 107 External Collection Efficiency 107 Diode Parameters 108 Open Circuit Voltage 112 Fill Factor 116 Capacitance–Voltage Analysis 119 Admittance Spectroscopy 122

3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10

Design Rules for Heterostructure Solar Cells and Modules 129 Absorber Bandgap 129 Band Alignment 131 Emitter Doping and Doping Ratio 137 Fermi Level Pinning 140 Absorber Doping 142 Absorber Thickness 147 Grain Boundaries 150 Back Contact Barrier 156 Buffer Thickness 159 Front Surface Gradient 162

Contents

3.11 3.12

Back Surface Gradients 165 Monolithic Series Interconnection 171

4 4.1 4.1.1 4.1.2 4.1.3 4.1.3.1 4.1.3.2 4.1.3.3 4.1.4 4.1.4.1 4.1.4.2 4.1.5 4.1.6 4.2 4.2.1 4.2.1.1 4.2.1.2 4.2.1.3 4.2.2 4.2.3 4.2.3.1 4.2.3.2 4.2.3.3 4.2.3.4 4.2.3.5 4.2.3.6 4.2.4 4.2.4.1 4.2.4.2 4.2.5 4.2.5.1 4.2.5.2 4.2.6 4.3 4.4 4.4.1 4.4.2 4.5

Thin Film Material Properties 175 AII -BVI Absorbers 175 Physico-Chemical Properties 175 Lattice Dynamics 179 Electronic Properties 180 Practical Doping Limits 180 Defect Spectroscopy 182 Minority Carrier Lifetime 182 Optical Properties 185 CdTe 185 Multinary Phases 187 Surface Properties 188 Properties of Grain Boundaries 189 AI -BIII -CVI 2 Absorbers 192 Physico-Chemical Properties 193 Ternary Phase Diagrams 193 Multinary Phases 198 Diffusion Coefficients 200 Lattice Dynamics 201 Electronic Properties 203 Single Point Defects 204 Defect Complexes 209 Defect Spectroscopy 210 Practical Doping Limits 214 Carrier Mobility 215 Minority Carrier Lifetime 216 Optical Properties 216 Ternary Semiconductors 217 Multinary Semiconductors 218 Surface Properties 220 Surface Composition 220 Surface Electronics 222 Properties of Grain Boundaries 224 Buffer Layers 226 Window Layers 226 Low Resistance Windows 226 High Resistance Windows 230 Interfaces 230

5 5.1 5.1.1

Thin Film Technology 235 CdTe Cells and Modules 235 Substrates 235

VII

VIII

Contents

5.1.2 5.1.3 5.1.4 5.1.5 5.1.6 5.1.7 5.1.8 5.1.8.1 5.1.8.2 5.1.9 5.2 5.2.1 5.2.2 5.2.3 5.2.3.1 5.2.3.2 5.2.3.3 5.2.3.4 5.2.4 5.2.5 5.2.6 5.2.7 5.2.8 5.2.8.1 5.2.8.2 5.2.9 5.2.10

Window Layers for CdTe Cells 236 Buffer Layers for CdTe Cells 237 CdTe Absorber Layer 240 Activation by Chlorine Treatment 243 Influence of Oxygen 245 Influence of Copper 246 Back Contact 249 Surface Modification 249 Primary and Secondary Contacts 250 Module Fabrication and Life Cycle Analysis 251 Cu(In,Ga)(S,Se)2 Cells and Modules 252 Substrates 253 Back Contacts 253 Cu(In,Ga)(S,Se)2 Absorber Layers 255 Co-evaporation 257 Deposition Reaction 260 Sputtering 262 Epitaxy, Chemical Vapor Deposition, and Vapor Transport Processes 262 Influence of Sodium 262 Influence of Gallium 265 Influence of Sulfur 265 Influence of Oxygen 266 Buffer Layers of CIGS 268 Chemical Bath Deposited CdS 268 Alternative Buffer Layers 270 Window Layers of CIGS 272 Module Fabrication 273

6 6.1 6.1.1 6.1.2 6.1.3 6.1.4 6.1.5 6.1.6 6.1.7 6.2 6.2.1 6.2.2 6.2.3 6.2.4 6.2.4.1 6.2.4.2

Photovoltaic Properties of Standard Devices CdTe Device Properties 277 Solar Cell Parameters 277 Diode Currents 277 Collection Functions 279 Device Anomalies 280 Transient Effects and Metastability 281 Device Model 282 Stability 285 AI -BIII -CVI 2 Device Properties 286 Solar Cell Parameters 286 Diode Currents 288 Collection Function 290 Transient Effects and Metastability 292 Relaxed State 292 Models for Relaxed State 293

277

Contents

6.2.4.3 6.2.4.4 6.2.4.5 6.2.4.6 6.2.4.7 6.2.4.8 6.2.4.9 6.2.5 6.2.6

Red Light Effect 294 Forward Bias Effect 295 Blue Light Effect 295 White Light Effect 296 Reverse Bias Effect 296 Models for Metastability 297 Implications for Module Testing Device Model 299 Stability 302

7 7.1 7.1.1 7.1.2 7.1.3 7.1.4 7.2 7.2.1 7.2.2 7.2.3 7.3 7.3.1 7.3.2 7.3.3 7.3.4 7.4 7.4.1 7.4.2 7.4.3 7.4.4 7.4.5 7.4.6 7.5 7.5.1 7.5.2

Appendix A: Frequently Observed Anomalies 305 JV Curves 305 Roll Over Effect 305 Cross Over 306 Kink in Light JV Curve 306 Violation of Shifting Approximation 307 Solar Cell Parameters 308 Reduced Jsc but High Voc 308 Reduced Voc but High Jsc 308 High Jsc but Low FF 309 Diode Parameters 309 Diode Parameter A > 2 309 Activation Energy Ea < Eg,a 309 Diode Quality Factor Illumination Dependent 310 Diode Quality Factor Temperature-Dependent 310 Quantum Efficiency 311 High Jsc but Low EQE 311 Low Jsc but High EQE 311 Low Blue Response in IQE 311 Low Red Response in IQE 312 Quantum Efficiency Low at All Wavelengths 312 Apparent Quantum Efficiency 313 Transient Effects 313 Voc Time-Dependent with dVoc /dt > 0 313 Voc Time-Dependent with dVoc /dt < 0 314

8

Appendix B: Tables References 321 Index 361

315

298

IX

XI

Preface The subjects of this book are thin film solar cells based on sulfide, selenide, or telluride semiconductors. We call them the chalcogenides. The most prominent members – at least today – are CdTe and CuInSe2 . These solar cells have been in production since the beginning of the twenty-first century, and production volume is growing rapidly. More companies are entering the field, each company expecting efficiency and cost advantages from their specific cell design or production process. This commercialization period, however, had a long forerunner. Chalcogenide photovoltaics has been the subject of research and development for four decades. The hope was that the use of polycrystalline films – instead of single crystal materials – would allow the fabrication of low-cost solar cells. Counting from 1980, there have been over 80 international scientific events dealing with this material group and its application in solar cells. The authors do remember enthusiastic periods when exploitation appeared just ahead and then again pragmatic periods when more research was shown to be needed. However, constantly important findings have been made – several of them producing unexpected results. The method was empirical science where in the best case theoretical understanding was the second step. The findings brought out several so-called magic steps which now are executed in production. Today it is fair to say that the achievements have been remarkable: chalcogenide-based photovoltaic devices achieve the highest energy conversion efficiencies among the thin film technologies – more than 20%. This value we still consider as a miracle, given that so many basic material properties are not well understood. The topics in this book are the physics and technology of chalcogenide-based photovoltaics. Considerable space is given to heterostructures (stacks of different semiconductor materials) which build the working core of the solar cell. CdTe- and CuInSe2 -based solar cells are heterostructures but the concepts are valid for any type of heterostructure. Important topics are band diagrams, diode currents, and photo currents in heterostructures where we discuss state of the art models. We provide only a limited section on device analysis which treats with specific aspects of non-ideal solar cells. Non-ideal solar cells are also the subject of Chapter 7, Appendix A, entitled Frequently observed anomalies, which gives a compendium of phenomena and the models which help to explain these anomalies. Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

XII

Preface

Material aspects of the chalcogenides for photovoltaics are treated with focus on II–VI and I–III–VI2 materials. The reader will find tables and figures on cell-relevant material properties related to solar cells which normally are found in a handbook. Within Chapter 5, entitled Thin film technology, the clear focus is on CdTe and CuIn1–x Gax Se1–y Sy , the chalcogenides currently most exploited. We discuss every technological step and every relevant chemical impurity. We hope that the reader will find enough information to make his own highly efficient chalcogenide solar cell. In Chapter 6, we then describe the photovoltaic properties of CdTe and CuIn1–x Gax Se1–y Sy standard devices also regarding the question of metastability and stability. We develop electronic models and band diagrams for the cells – admittedly based partly on secured knowledge and partly on the authors’ opinion. This book is the result of five years’ work. During this period, the authors sought discussions with several scientists all over the world. The feedback from those colleagues was very valuable. We want to thank Dr Abou-Ras, Prof. Burgelman, Prof. Cahen, Dr Eisenbarth, Prof. Ferekides, Dr Fränzel, Dr Igalson, Prof. Jägermann, Prof. Klein, Prof. Mönch, Prof. Romeo, Prof. Rau, Prof. Siebentritt, B.A. Schubert, Dr Visoly-Fischer, H. Wilhelm, and Prof. Würfel. We further thank our families for their encouragement and their forbearance. Roland Scheer Hans-Werner Schock

XIII

Symbols and Acronyms

Symbol

Unit

A AM CBM Cc Dn Dp Dd dtot , dδ

cm –2 eV−1 cm2 s –1 cm2 s –1 cm –3 eV –1 cm

Ec,δ

eV

Ea EF EFn EFp Eg,δ

eV eV eV eV eV

Ei

eV

En,δ

eV

Ep,δ

eV

Ed

eV

Ev,δ

eV

En

eV

Ep

eV

EQE En , Ep

eV

Diode quality factor Air mass Conduction band minimum Band gap gradient defect factor Electron diffusion coefficient Hole diffusion coefficient Defect distribution Total thickness of heterostructure and thicknesses of absorber, buffer, window layers (δ = a, b, w) Conduction band edge of absorber, buffer, window layers (δ = a, b, w) Activation energy of Jo Equilibrium Fermi energy Quasi Fermi energy of electrons Quasi Fermi energy of holes Energy bandgap of absorber, buffer, window, or interface (δ = a, b, w, IF) Intrinsic position of the Fermi energy with respect to a reference level Absolute value of ζn in absorber, buffer, window (δ = a, b, w) Absolute value of ζp in absorber, buffer, window (δ = a, b, w) Defect level referenced to the valence band maximum Valence band edge of absorber, buffer, window layers (δ = a, b, w) Fermi energy referenced to the conduction band or quasi-Fermi energy of electrons Fermi energy referenced to the valence band or quasi-Fermi energy of holes External quantum efficiency Demarcation levels


XIV

Symbols and Acronyms Symbol

Unit

EFn EFp Ec,back , Ev,back

eV eV eV

Ec,front , Ev,front

eV

Ec b,a , Ev b,a

eV

Ec w,b , Ev w,b

eV

F FF FLP G GB HRW J J0 J00 JCBM

V/cm

Jsc Jζ

A cm –2 A cm –2

LO LRW n

cm –3

Nv Nc Nd NIF NP q QIF QNR Rbc SCR SRH Sn0 , Sp0 TO U UHV V VBM

cm –3 s –1

A cm –2 A cm –2 A cm –2 A cm –2

cm –3 cm –3 cm –3 cm –2 As As cm –2

cm s –1 cm –3 s –1 V

Discontinuity of the electron quasi Fermi energy Discontinuity of the hole quasi Fermi energy Bandgap variation at the back surface of a layer realized by a variation of the conduction band edge or valence band edge, respectively Bandgap variation at the front surface of a layer realized by a variation of the conduction band edge or valence band edge, respectively Conduction band and valence band offset, respectively, at the buffer/absorber interface (see sign convention in Section 2.3.1) Conduction band and valence band offset, respectively, at the buffer/window interface (see sign convention in Section 2.3.1) Electric field Fill factor Fermi level pinning Generation rate Grain boundaries High resitance window Current density Saturation current density Reference current density Current density due to a gradient in the conduction band minimum Short circuit current Current density due to a gradient in the reduced chemical potential Longitudinal optic Low resistance window Electron density Refractive index Effective density of states in valence band Effective density of states in conduction band Bulk defect density Interface defect density Nitric–phosphoric acid Elementary charge Interface charge density Quasi neutral region Back contact reflectivity Space charge region Shockley, Read, Hall (recombination) Surface or interface recombination velocity Transversal optic Recombination rate Ultrahigh vacuum Voltage Valence band minimum

Symbols and Acronyms

Symbol

Unit

vn vp Voc wtot , wδ

cm s –1 cm s –1 V cm

x y α χ

K –1 eV

φb p φb n

As V –1 cm –1 V V

η η(V)

%

ηc (z,V)

µn µp τn0 τp0 ζn , ζp

cm2 V –1 s –1 cm2 V –1 s –1 s s eV

ζn , ζp

eV

ζn0 , ζp0

eV

Electron thermal velocity Hole thermal velocity Open circuit voltage Total width of SCR and SCR of absorber, buffer, window layers (δ = a, b, w) Stoichiometry deviation [Eq. (4.5)] Molecularity deviation [Eq. (4.5)] Thermal expansion coefficient Electron affinity; corresponds to the absolute value of ζn0 Dielectric constant Hole barrier Electron barrier Charge collection efficiency in an electron beam induced current experiment (EBIC) Solar cell efficiency External collection efficiency; describes the dependence of the photo current on the voltage Collection function; describes the collection probability of charge carriers generated in the solar cell Electron mobility Hole mobility Minimum lifetime of the electron Minimum lifetime of the hole Reduced chemical potential of electrons and holes; the part of the chemical potential which reflects the number density of species Chemical potential of electrons and holes, respectively Concentration independent parts of the chemical potential of electrons and holes; quantities which reflect the chemical environment of the species

XV

1

1 Introduction The subject of this book is sketched in Figure 1.1. It is a heterostructure solar cell based on polycrystalline thin films comprising a metallic back contact, a semiconducting absorber layer, and some kind of emitter which is transparent for most of the solar spectrum. Typical film thicknesses are around 1–3 µm – enough to gather most of the photons incident from the sun. The film thickness of the absorber layer is in the size range of the polycrystalline grain and the absorption length of sunlight. All films are deposited by astonishingly simple methods, given that chalcogenide based cells today have the highest efficiency among the thin film solar cells. In the following, we give a historical digest on solar cell development. We try to label some milestones of progression and understanding. Certainly, the prospect of these cells is more difficult. Hence, we only outline some facts and useful estimations. Figure 1.1 Sketch of a polycrystalline thin film cell with window layer and absorber layer. The boundaries of the large grains of the absorber layer are outlined. The thickness of the window and absorber layers are dw and da . wa is the width of the space charge layer within the absorber layer. The letters (a) and (b) indicate perpendicular (a) and parallel (b) grain boundaries with respect to the space charge layer.

a Emitter

Back contact

Absorber

b

y -da

-wa

x z

1.1 History of Cu(In,Ga)(S,Se)2 Solar Cells

The history of Cu(In,Ga)Se2 (CIGSe)-based solar cells began in 1975 when Bell laboratory scientists achieved 12% solar energy conversion efficiency with a cell where a layer of CdS was evaporated onto a CuInSe2 single crystal [1]. Stimulated by this Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

1 Introduction

result, the development of CuInSe2 thin films started. At the University of Maine (see Figure 1.2), thin films were deposited by dual source evaporation of CuInSe2 and Se [2]. Small area devices in the substrate configuration were built with an efficiency in the range of 5%. In the early 1980s, companies started developing CIGSe solar cells (see Figure 1.2). At this stage, two different preparation concepts for CIGSe were pursued. Boeing used a co-evaporation process where Cu, In, and Se were deposited from separate evaporation sources. The Boeing group was the first to demonstrate a more than 10% efficient thin film device [3]. Arco Solar, in contrast, developed a deposition–reaction process using Cu and In metallic precursors, with H2 Se as the reacting chalcogen source. Surprisingly, for a period of time the deposition–reaction technology gave better device efficiencies than the co-evaporation approach (see Figure 1.2). The reason was that, during the reaction step, the absorber layer was doped with Na – a fact that was overlooked by the rest of the research community. In 1995, Na-containing glasses became standard and both technologies – co-evaporation and deposition–reaction – gave almost similar efficiencies of around 14% on the cell level. The deposition–reaction technology, however, was much easier to scale up: sputtering of metal films was well established in the glass industry. The limitation was rather in the width of available quartz tubes for the selenization reactor. Accordingly, the first modules developed by Arco Solar were only 1 ft (305 mm) wide. Parallel selenization of up to 20 substrates was a batch process with around 6 h process time.

NREL

ZSW

20 Efficiency [%]

2

EuroCIS SSI Arco (3)

15

(4) Uni Johbg Showa

(2) SSG

10 (1)

5

Boeing

Wuerth

SSI

Univ. of Maine

1980

1990 Year

Figure 1.2 Solar energy conversion efficiency of thin film cells and modules based on Cu(In,Ga)(S,Se)2 (CIGSSe). Full circles – cells by co-evaporation, open circles – full modules by co-evaporation, full squares – cells from deposition–reaction technology, open squares – full modules from deposition–reaction. Abbreviations: Arco – Arco Solar, SSI – Siemens Solar Industries (USA), EuroCIS – University

2000

2010

Stuttgart and University Uppsala, SSG – Siemens Solar AG (Germany), Wuerth – Wuerth Solar (Germany), Showa – Showa Shell (Japan), Uni Johbg – University of Johannisburg (South Africa), NREL – National Renewable Energy Laboratories (USA), ZSW (Germany). Labels (1) to (4) refer to milestones of (CIGSSe) development which are discussed in the text.

1.1 History of Cu(In,Ga)(S,Se)2 Solar Cells

The headstart achieved by co-evaporated solar cells (not modules) began with employing more complicated evaporation processes. The so-called three-stage process invented in 1994 by the National Renewable Energy Laboratories (NREL) group [4] provided the basis for an efficiency of 20%. Around 2000, an effort was made at Wuerth Solar to bring the co-evaporation technology into production. Wuerth Solar designed line-evaporation sources that were able to deposit on substrates 2 ft wide. As depicted in Figure 1.2, co-evaporated module efficiency could rapidly compete with modules from the deposition–reaction technology. In 2008, several laboratories around the world had the experience to prepare a CIGSe solar cell with an efficiency above 19%. Concerning large area modules (>0.4 m2 ), both co-evaporation and deposition–reaction technologies reached above 12% efficiency. While writing these lines, submodules of Cu(In,Ga)(S,Se)2 formed by the deposition–reaction process have been reported with module efficiencies approaching 16% [5]. Both process schemes are still in competition and may further be pursued in parallel. Today, a large number of subvariants of these process schemes are being investigated by research laboratories and companies. They all have the medium-term goal of a thin film module with more than 15% efficiency produced with costs below 0.5 ¤/Wp . 1.1.1 Milestones of Cu(In,Ga)(S,Se)2 Development

As the first evaporated CuInSe2 films (from CuInSe2 and Se evaporation sources) suffered from poor control of the ratio of copper to group III element, Cu/In (in early days the group III element was only indium), the first milestone, labeled (1) in Figure 1.2, was to apply a triple-source evaporation geometry in which the copper and indium rates during deposition and thus the Cu/In ratio in the film could deliberately be adjusted. Soon it was recognized that the best cell efficiencies were achieved by a copper-poor composition, that is, Cu/In < 1. Around 1988, two innovations, summarized under milestone (2), led to devices with ∼12% efficiency. The first innovation was the use of a thin CdS layer prepared in a chemical bath. This thin buffer layer (see Figure 1.3) rendered a reduction of the optical absorption and a better coverage of the polycrystalline CuInSe2 surface. Due to the complete coverage of the absorber by a CdS layer ∼50 nm thick, shunting i-ZnO/n-ZnO 0.25 – 1 µm Buffer

50 nm

Cu(In,Ga)(S,Se)2 1 – 3 µm Mo 2 µm

0.5 – 1 µm

Glass, metal, polymer

Figure 1.3 Schematic outline of a substrate-type solar cell based on Cu(In,Ga)(S,Se)2 (CIGSSe).

3

1 Introduction

paths could be eliminated. Chemical bath deposition of the buffer at that time became standard, and today turnkey equipment for this process step is available. The second innovation was the employment of a Cu-rich growth period during CuInSe2 film formation. Although CuInSe2 films require a Cu-poor composition in the end, their morphology and electronic properties are superior if they are partly grown under an excess of Cu. CIGSe films grown by the deposition–reaction process pass the Cu-rich growth regime automatically. In the beginning of the 1990s, it became clear that CuInSe2 solar cells perform better if the cells are prepared on sodium (Na)-containing glass and if part of the indium is replaced by gallium (Ga). (As a speciality of the deposition–reaction process, also a part of the selenium was replaced by sulfur.) Thus, the absorber became a multinary compound semiconductor. This established milestone (3) which allowed demonstrating efficiencies of around 14%. Today, Na and Ga doping is an issue for all types of CIGSe preparation. With the introduction of Ga (and S), the CuInSe2 cell became a CIGSSe cell with a material stack as shown in Figure 1.3 – a molybdenum back contact, a Cu(In,Ga)(S,Se)2 absorber layer, a CdS buffer layer, and ZnO formed by two separate layers. We note, however, that the principle outline of the cell (Mo/absorber/CdS/ZnO) was invented in 1987 by the ARCO Company [6]. Although in the mid-1990s there was still room for improvement for all types of CIGSe preparation techniques, it was a very particular innovation which yielded around 18% devices and brought up milestone (4) at the end of the 1990s. The use of a complicated three-stage co-evaporation process which until then had only been realized in the laboratory. The principle of this process is explained in Section 5.2.3. Subtle parameter variations and the optimization of the contact scheme for single 0.5 cm2 cells brought up a cell of 20% air mass (AM) 1.5 efficiency [7]. Figure 1.4 shows the current–voltage curve of such a highly efficient CIGSe solar cell (redrawn from [8]). This laboratory cell shows the enormous potential of CIGSe

Current density [mA cm−2]

4

0 −10 −20

(2) 20

−30 −40 0.0

%

(1) 0.2

0.4 0.6 Voltage [V]

0.8

Figure 1.4 Current density versus voltage curve of (1) a high efficiency Mo/Cu(In,Ga)Se2 /CdS/ZnO solar cell and (2) a high efficiency BC/CdTe/CdS/Zn2 SnO4 /Cd2 SnO4 solar cell (redrawn after Ref. [8]).

1.2 History of CdTe Solar Cells

solar cells. Also, cells made by the deposition reaction process have been improved in the past decade due to parameter optimizations (Figure 1.2). 1.2 History of CdTe Solar Cells

The development of CdTe solar cells was initially motivated by their potential space application in communication satellites. In 1963, the first thin film solar cell of the type n-CdTe/p-Cu2 Te was demonstrated by a General Electric Research Laboratory [9]. The cell had an efficiency of 6%, but device instability (caused by Cu diffusion) led to the quest for an alternative heteropartner for CdTe. In 1972, Bonnet and Rabenhorst presented a p-CdTe/n-CdS heterojunction solar cell which also had 6% efficiency [10]. The CdTe layer was deposited by vapor transport deposition (cf. Section 5.1.4), a method that today is used in mass production. These cells were of the substrate type, having a Mo back contact and a CdTe thickness of more than 10 µm [10]. Ten years later, Tyan et al. [11] presented a superstrate thin film cell with more than 10% efficiency (Figure 1.5). This cell was grown by closed-space sublimation – the leading method of later champion devices. Encouraged by this achievement, several companies started inhouse developments. The goal was to find the best low-cost deposition method that allowed high throughput and that could deliver more than 10% efficient modules. Surprisingly, this goal was reached using very diverse deposition methods such as vapor transport deposition, electrodeposition, spray pyrolysis, and screen printing (cf. Table 5.2). CdTe was 18 (4)

USF

16

NREL

Efficiency [%]

NREL (3)

14

BP Solar

Photon Energy

12

BP Solar

Kodak 10 8 6

(2) (1)

AMETEK

BP Solar

Matsushita Batteries

FS

BP Solar Matsushita

Batelle Inst. 1980

1990 Year

2000

Figure 1.5 Solar energy conversion efficiencies of thin film cells (full marker) and modules (open markers) based on CdTe. Abbreviations: USF – University of South Florida, NREL – National Renewable Energy Laboratories (USA), FS – First Solar.

2010

5

6

1 Introduction

considered as an extraordinary forgiving material. Nevertheless, it was not earlier than 2002 that CdTe modules became a mass product on basis of vapor transport deposition while the so-called low-cost methods such as spray pyrolysis and screen printing have been ruled out – at least for the time being. Reasons for the stepback of early industrial players were probably rather in strategic decisions than in technological obstacles. On the cell level, the NREL (USA) in 2001 achieved the long-standing record of 16.5% efficiency [8]. They used close-space sublimation for the absorber preparation. Nowadays, CdTe modules are produced on the GWp /year level and currently are the cost leader in the photovoltaic industry. In a sense, the CdTe technology exemplifies the cost advantage of thin film photovoltaics over wafer-based technologies – an advantage that has been claimed for quite some time. 1.2.1 Milestones of CdTe Development

Soon after realization of the first CdS/CdTe device it was recognized that this type of cell is more efficient if formed in the superstrate configuration. Use of this configuration (Figure 1.6) was the first milestone, as shown in Figure 1.5. Interdiffusion between CdS and CdTe which can easily happen in the process sequence was believed to be a reason for the superior performance of the superstrate configuration. The second milestone in 1982 was the incorporation of oxygen during film deposition. Several effects of oxygen on film properties are discussed in Section 5.1.6. Use of chlorine may have been implicit for several preparation techniques such as screen printing and electrodeposition. But it was the third milestone when researchers after 1990 became aware that the activation of the CdTe by chlorine treatment substantially increases the device efficiency. Since that time, chlorine activation using specific and critical process parameters has always been part of the process. The result of such parameter optimization were laboratory-scale solar cells with efficiencies exceeding 15% – first demonstrated by Ferekides et al. [12]. It took almost 10 years before the fourth milestone led to a 16.5% device. We identify this Glass

1 µm

TCO

0.5 – 1 µm

CdS

100 nm

CdTe

2 – 8 µm

Back contact

Figure 1.6 Schematic outline of a superstrate-type solar cell based on CdTe with a ZnTe:Cu/Ti back contact.

1.3 Prospects of Chalcogenide Photovoltaics

milestone with the insertion of a high mobility window layer of Cd-stannate. The current–voltage (JV) curve of such a highly efficient device is shown in Figure 1.4. Highest module efficiencies are around 11% since the champion cell technology has not yet been completely transferred to production. 1.3 Prospects of Chalcogenide Photovoltaics

If chalcogenide photovoltaics are to substantially contribute to future energy supply, we speak of some terawatts of installed capacity. Currently, chalcogenide module production is in the gigawatts/year range, mostly driven by one CdTe module producer (First Solar). We want to address the question about possible hurdles for future expansion. Zweibel assessed future cost reduction potentials of thin film photovoltaics [13]. Table 1.1 condenses this work and reveals that both technologies [CdTe and Cu(In,Ga)(S,Se)2 ] have about the same cost perspective for large scale production of around 0.5 $/Wp . Already today (2009) CdTe module production is reported with costs of 0.85 $/Wp at an efficiency of 11%. Thus, it appears realistic that future cost reduction and efficiency increase will lead to the figures in Table 1.1. With an anticipated module lifespan of 30 years and the listed costs for ground-mounted and rooftop installations, these numbers translate into about 0.06–0.09 $/kWh AC electricity costs in sunny climates with around 1800 kWh m−2 year−1 [13]. Even lower-cost figures are derived in a study by Keshner and Arya (Hewlett Packard) [14]. In the climate referenced above, the energy payback time of today’s CdTe modules is around 1.1 years [15]. Thus, we see no hurdles for CdTe and Cu(In,Ga)(S,Se)2 technology expansions in terms of costs and energy payback time. A chalcogenide thin film module contains, depending on the chalcogenide film thickness, some grams per square meter of metals and chalcogens (see Table 1.2). Among the materials listed in Table 1.2, indium and tellurium are considered most critical in terms of availability. In order to estimate the maximum installation possible based on each material, two different approaches are pursued: (i) use of published material reserves and (ii) extrapolation of current annual material Cost projections of CdTe and Cu(In,Ga)(S,Se)2 module production according to Ref. [13] assuming gigawatts/year production level. Table 1.1

CdTe/glass CIGSSe/glass

Projected efficiency (%)

Module costs ($/Wp )

Systems costs for large ground mounted systems ($/Wp )

Systems costs for large commercial roof top systems ($/Wp )

14 15.5

0.47 0.51

1.12 1.18

1.61 1.67

7

8

1 Introduction Table 1.2 Material consumption for chalcogenide modules with absorber thickness 1 µm and material resources after United States Geological Survey data for 2008. Maximum possible installations of CdTe and CuIn0.7 Ga0.3 (S,Se)2 single junction solar cells assuming efficiencies of 14 and 15.5%, respectively, a module lifespan of 30 years, and complete use of the resource.

Cd Te In Ga Se

g/m2

Reserves (106 kg)

Reserve base (106 kg)

Maximum installation (TWp ) according to Anderson [16]

Maximum installation (TWp ) according to Schubert et al. [17]

Maximum installation (TWp ) according to Zweibel [13]

3.2 3.3 1.5 0.27 2.4

490 22 11 No data 86

1200 48 16 >1000 172

12.5 1.51 3.8 No data 12.5

– – 1.29 – –

– 15 8.5 – 150

extraction. Both approaches are reflected in Table 1.2. Anderson [16] uses reported reserves for indium and tellurium and derives 1.51 and 3.8 TWp maximum installation for CdTe and Cu(In,Ga)(S,Se)2 , respectively. About half this number for Cu(In,Ga)(S,Se)2 is obtained by Schubert et al. [17] but there only about 50% use of indium for PV is assumed. The original figures of Anderson need correction because indium reserve estimations increased by a factor of ∼2.5 (!) between 2007 and 2008. Interestingly, for 2009 no data are given by the U.S. Geological Survey, reflecting the fact that the reserves data for indium contain large uncertainties. This is why Zweibel [13] uses a different approach. Today, indium is exploited from Zn ore (although also contained in copper, lead, tin, and tungsten ores). Extrapolating the Zinc mining capacity with 1% annual growth gives an alternative estimation for indium reserves. This approach (and the corresponding one for tellurium) is used by Zweibel in Ref. [13] and in the respective column of Table 1.2. After correction to 1 µm absorber thickness (for the sake of comparability) we arrive at around 10 TWp for each technology. Thinning the chalcogenide absorber to 0.5 µm simply doubles these values. (Today, the world average power demand is around 10 TW.) We may conclude that – based on the current estimates – chalcogenide thin film solar cells based on CdTe and Cu(In,Ga)(S,Se)2 can substantially add to the world power supply. However, it is questionable whether the reserves are sufficient to make them the only photovoltaic technologies of the future.

9

2 Thin Film Heterostructures Today, thin film chalcogenide photovoltaic devices are fabricated in the form of heterostructures. An example is p-type CdTe in contact with n-type CdS and n-type SnO2 . CdTe is the absorber, CdS the buffer, and SnO2 the window. We call this device an absorber/buffer/window heterostructure although there are even more components to mention, namely the metal contacts and interfaces. Slightly simpler is an absorber/window heterojunction. In this chapter we present the physical concepts to describe the band diagram and the electrical transport phenomena in heterostructure solar cells. In the last section, we discuss particular analysis aspects. We use device simulation extensively – since many physical relationships can be explained more easily by graphs than by formulas. For the simulation which is accomplished mostly in one dimension, we designed a generic absorber/buffer/window device. Table 8.1 gives the default values of the physical quantities of this device. Deviations from the default values are highlighted in the text.

2.1 Energies and Potentials

First, we briefly introduce definitions for energies and potentials in as far as they are used throughout this book. For a more detailed description, the reader is referred to other works [18, 19]. Very generally, Figure 2.1 depicts the valence band edge Ev and conduction band edge Ec of a non-homogeneous semiconductor under non-equilibrium condition. The non-equilibrium condition may be induced by illumination, carrier injection or both. The reference energy is the potential energy E = 0. (In the heterostructure’s energy band diagram, we assign the zero of energy to the lowest lying valence band edge of all semiconductors of the heterostructure.) The absolute values of the band edges are not known experimentally (see Ref. [20] for a detailed discussion). Note, that we do not define a vacuum level inside the semiconductor. If band offsets at a heterostructure interface are required, we can use calculated or measured values of Ec and Ev . Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

10

2 Thin Film Heterostructures E=0 Ec

-qϕ -qϕ

En

EFn Ec

Ev

ζn0

Ei EFp

ζ′n

ζ′p

ζp0 χ

EFn

ζn Eg ζp

EFp Ep

Ed Ev

E z

Figure 2.1 Schematic band diagram of a semiconductor. In order to demonstrate a very general case, we allow the semiconductor to have a varying electron affinity and varying energy gap. Furthermore, the semiconductor is in non-equilibrium as revealed by the energy separation of the electron and hole quasi Fermi levels.

As for the reference level, the absolute value of the electrostatic potential ϕ is also not known. In band diagrams we plot it on a separate energy scale. Analytically, we only make use of variations of the potential energy −q dϕ/dz. These variations in −qϕ can be due to space charge within the semiconductor or due to an external electric field. The semiconductor in Figure 2.1 exhibits a bandgap energy Eg and an electron affinity χ, both varying as a function of position. The physical reason may be that the semiconductor exhibits varying material properties such as element composition or strain. Gradients of Eg and χ are independent of the gradient in the electrostatic potential. Since the semiconductor is depicted under non-equilibrium condition, each carrier has its own quasi Fermi level, EFn and EFp . In the case of thermodynamic equilibrium, both quasi Fermi levels join to the equilibrium Fermi energy EF . The quasi Fermi levels can be identified with the electrochemical potentials of electrons and holes. This is reflected in the relations EFn = −qϕ + ζ n

(2.1)

EFp = −qϕ − ζ p

(2.2)

as can be inferred from Figure 2.1. The quantities ζ n and ζ p are the chemical potentials of electrons and holes, respectively. If the chemical potentials are related to the band edges, we write the nonprimed variables ζn and ζp which we denote as reduced chemical potentials. In the case that −ζn and −ζp are larger than 3kT, that is, if the semiconductor is nondegenerate, we can write ζn = kTln{n/Nc } and ζp = kTln{p/Nv }. From these expressions we immediately see that the reduced

2.2 Charge Densities and Fluxes

chemical potentials ζn and ζp depend on the particle number densities n, p, and the density of states in the conduction and valence band, Nc and Nv . Thus, the reduced chemical potentials reflect those parts of the chemical potential which are concentration dependent. However, there are those parts ζn0 and ζp0 which depend on the chemistry of the material and not on the concentrations. That is ζn = ζ n − ζn0

ζp = ζ p − ζp0

(2.3) (2.4)

The absolute value of ζn0 is denoted as electron affinity. The reduced chemical potentials ζn and ζp both have negative values as reflected in the orientation of the arrows in Figure 2.1. Thus, it is convenient to introduce the absolute quantities En = −ζn and Ep = −ζp . Their definition is sketched on the right hand side of Figure 2.1 with arrows going up (for positive energies). In non-equilibrium, En and Ep are the quasi Fermi energies referenced to the bound edges. In equilibrium, En and Ep denote the same Fermi energy but referenced to the conduction band and valence band, respectively. The semiconductor may contain deep or shallow impurity levels of density Nd . Their electronic transition energies are quoted as Ed in general and ED and EA in the case of donors and acceptors. Note, that we use the term electronic transition energy instead of energy level in agreement with the definition given in Ref. [21]. The transition energy of a single donor state, for instance, describes the position of the Fermi energy at which the donor changes from D0 to D+ . Transition energies of all energies are given with respect to the valence band maximum. With the definitions above, the quasi Fermi levels become EFn (z) = −qϕ(z) − χ(z) − En (z)

(2.5)

EFp (z) = −qϕ(z) − χ(z) − Eg (z) + Ep (z)

(2.6)

z is the space coordinate as depicted in Figure 2.1.

2.2 Charge Densities and Fluxes

Three sets of equations are sufficient to describe the transport of carriers within the semiconductor or within heterojunctions: the transport equations, the continuity equations, and the Poisson equation. The transport equations for electrons and holes are given by dEFn (z) (2.7) dz dEFp (z) Jp (z) = µp p(z) (2.8) dz where the electric current densities Jn (z) and Jp (z) are expressed in units of A cm−2 . n, p are the carrier densities of electrons and holes and µn and µp are their mobilities. Jn (z) = µn n(z)

11

12

2 Thin Film Heterostructures

With the aid of Figure 2.1, it may be instructive to verify the sign convention in this book. Due to the orientation of the coordinate z in Figure 2.1, dEFn /dz and dEFp /dz are positive and thus the current densities Jn and Jp are positive. Electrons are moving toward smaller values of EFn (to the left) and holes are moving toward larger values of EFp (to the right). The total electric current density J(z) = Jp (z) + Jn (z) is positive if flowing from left to right. Using Eqs. (2.5) and (2.6), the current densities can be expressed as dϕ(z) dχ(z) dEn (z) − − (2.9) Jn (z) = µn n(z) −q dz dz dz dϕ(z) dχ(z) dEg (z) dEp (z) Jp (z) = µp p(z) −q − − + (2.10) dz dz dz dz In Eqs. (2.9) and (2.10) we assume that the mobilities of electrons and holes, µn and µp , are constant and do not change with position. Furthermore, we do not allow a variation of temperature within the semiconductor. (For the discussion of thermoelectric effects see Ref. [18]). Equations (2.9) and (2.10) state that the electric current in a semiconductor device such as a solar cell is driven by gradients of the electrostatic potential, gradients of the electron affinity and bandgap, and gradients of En and Ep (the absolute of the reduced chemical potentials). Equations (2.9) and (2.10) are also valid for degenerate semiconductors. In the case of nondegenerate semiconductors (En,p > 3kT) we can write dEn kT dn(z) kT dNc (z) =− + dz n(z) dz Nc (z) dz

(2.11)

dEp kT dp(z) kT dNv (z) =− + dz p(z) dz Nv (z) dz

(2.12)

From Eqs. (2.11) and (2.12) we recognize that electric current can be driven by gradients of the carrier density and the density of states. In a homojunction solar cell, the current is only driven by gradients of En and Ep (in the quasi neutral region, QNR) and a gradient of −qϕ (in the field zone). However, in the heterojunction solar cells, gradients of the state density and the band edges give rise to additional currents. These gradients are sometimes denoted as effective force fields [22]. Later we have to quantify the different sources of current flow as included in Eqs. (2.9−2.12). The continuity equations for electrons and holes are given by ∂n(z) 1 dJn (z) = Gn (z) − Un (z) + ∂t q dz

(2.13)

1 dJp (z) ∂p(z) = Gp (z) − Up (z) − ∂t q dz

(2.14)

Here we have introduced the particle flux densities of electrons and holes, Jn /q and Jp /q. The continuity equations (Eqs. (2.13) and (2.14)) describe particle number densities and not charge densities. Gn (z), Gp (z), Un (z), and Up (z) are the generation and recombination rates of electrons and holes. In the case of recombination

2.3 Energy Band Diagrams

and generation via trap states, the generation rates as well as the recombination rates can be different for electrons and holes. In the absence of traps, however, generation of an electron into the conduction band necessarily leaves a hole in the valence band. In the latter case, generation of electrons and holes take place simultaneously and we can write Gn = Gp as well as Un = Up . Note that we differentiate the external rate of generation by photons and the internal, thermal generation rate. The continuity equations (Eqs. (2.13) and (2.14)) describe dynamic processes. Throughout this book, however, we mostly deal with the stationary state where we can put ∂n(z)/∂t = ∂p(z)/∂t = 0. The Poisson equation relates charge densities and the electrostatic potential. The most general form is derived from the Maxwell equation d/dz · D(z) = ρ(z) with D(z) = −ε(z)d/dzϕ(z). Then the Poisson equation reads −ρ(z) dln{ε(z)} dϕ(z) d2 ϕ(z) = + dz dz ε(z) dz2

(2.15)

Here, we allow the dielectric constant to be dependent on position. In the case of ε = const, Eq. (2.15) simplifies to d2 ϕ(z)/dz2 = −ρ(z)/ε. The space charge ρ(z) is given by ρ(z) = q/ε(z) · (p(z) − n(z) + ND + (z) − NA − (z))

(2.16)

where p(z), n(z) represent the mobile charges and ND + (z), NA − (z) the fixed charges in the form of charged donors and acceptors. Within the scope of this book, we are interested in the short circuit current density, in the open circuit voltage, and in the fill factor of the chalcogenide solar cell. Only implicitly, these quantities are included in Eqs. (2.9), (2.10), (2.13), (2.14), and (2.16).


The energy band diagram of an energy converting heterostructure is very important as it defines the carrier transport, carrier recombination, and Fermi level splitting. In this section, we need to introduce the definitions of band offsets, discuss the distribution of the electrostatic potential and consider the effects of interface charge, interface dipoles, and deep bulk states. Furthermore, bandgap gradients shall be taken into account. 2.3.1 Rules and Conventions

In general, we consider solar cells consisting of the back contact (bc), the absorber layer (a), the buffer layer (b), the window layer (w), and the front contact (fc). Between each component there will be an interface (IF). But for the beginning, let

13

14


(1) Ec EF Ev

∆Ec

(2) Ec EF

∆Ev Ev

(3)

(4)

Figure 2.2 Schematic steps of a four-step recipe on how to construct a heterojunction band diagram (see text).

us discuss a simple heterojunction between the two semiconductors absorber (a) and window (w). In order to construct the band diagram, we need to know the band offsets Ev and Ec as well as the equilibrium Fermi level positions Ep,a and En,w where the subscript characters a and w refer to absorber and window, respectively. The heterojunction in equilibrium is defined by two conditions: (i) the band offsets remain preserved on a relative scale and (ii) the equilibrium Fermi level is flat. With aid of an example, we show the four steps of a simple recipe how to construct the band diagram in Figure 2.2: • The basis is a picture of Ec , Ev , and EF for both semiconductors before contact, together with their band offsets as given by vertical bars. • In the next step a flat Fermi energy is drawn. Far from the junction, Ev and Ec have their positions (with respect to the Fermi energy) as it was before contact formation. • The band offsets are placed depending on the doping ratio of the two semiconductors. If the n-type semiconductor is much higher doped (as assumed in the example), the offset Ec remains close to the Fermi energy. Vice versa, if the p-type semiconductor is much higher doped, the offset Ev remains close to EF . If the doping levels are about identical, the offsets are placed between the two extreme cases, that is, both offsets have about the same distance to EF . • The band edges are connected with convex shape for the p-type semiconductor and concave shape for the n-type. Thereby, it is recommended to draw the majority carrier related band edge first, that is, the valence band of the p-type and conduction band of the n-type semiconductor. The procedure may be extended to a heterostructure with three semiconductors. Figure 2.3a shows the band edges of three semiconductors plus two metals and their Fermi energies before contact. Here, it is assumed that each semiconductor


-qϕ -qϕ(-wa) = qVbi -qϕ(0) -qϕ = (0)

E -qϕ ∆ECa,b

∆ECb,w

EC

E

EC

Ep,a

(fc)

z=0

(bc)

EV absorber buffer window z

(a)

Figure 2.3 Schematic of the heterostructure formation process between a window layer, w, a buffer layer, b, and an absorber layer, a. The heterostructure is contacted by metallic front, fc, and back, bc, contacts. Before contact formation (a), the band edges are aligned according to experimental or theoretical band offsets. Fermi energies differ while the electric potential is constant. After

E=0 (b)

EV -da

-wa 0 db db+dw

z

contact (b), the valence band and conduction band offsets are preserved and appear as band discontinuities. Charge transfer between different layers has led to the equilibration of the Fermi energy. Note that the electric potential does not exhibit any discontinuities since the heterostructure’s interfaces are free of electric dipoles.

is homogeneous with respect to effective densities of states, doping densities, static dielectric constant, bandgap, and electron affinity. If it comes to the band offsets, we use the following sign convention: the bandgap difference Eg is a positive quantity if we write at each interface Eg = Eg l − Eg s where the superscripts denote large (l) and small (s) bandgaps. If we define for the band offsets Ev = Ev s − Ev l and Ec = Ec l − Ec s , it is at each semiconductor/semiconductor interface Eg = Ec + Ev

(2.17)

It is common practice to speak of positive and negative band offsets. With the above definitions, a band offset is positive if a charge carrier has to spent kinetic energy in order to overcome an energy barrier from the small gap to the large gap semiconductor. Thus, the example in Figure 2.2 includes positive offsets for both the conduction band and the valence band. Vice versa, a carrier gains kinetic energy at a negative offset. Band offsets are also classified by the terms spike (for the positive barrier) and cliff (for the negative barrier). In Figure 2.3, electrons coming from the absorber must pass a conduction band spike while entering the buffer layer and must pass a conduction band cliff while entering the window layer. And finally, there is a convention to denote different types of heterojunctions. If both band extrema (valence band maximum and conduction band minimum) are located in one semiconductor, this interface is referred to as a type I interface. In Figure 2.2 this is the case and in Figure 2.3 this is the case for the buffer/absorber interface. A type II interface is found for the buffer/window interface of Figure 2.3.

15

16


Here the valence band maximum is in one semiconductor and the conduction band minimum is in the other. The band line up is a property of the two semiconductors forming an interface. The valence band offset, Ev , for a particular interface is obtained from experiment or is calculated from theory. Linear models have been proposed which assume that the band offsets are suspect to a transitivity rule: Knowledge of Ev for the heterojunctions AB as well as BC (A, B, C are different semiconductors) allows to calculate Ev for the AC heterojunction. These linear models, however, are still under debate (see Ref. [20]). In the case that chemical intermixing occurs at an interface, the band offset of the heterojunction of interest may differ from the one found in a particular experiment or from the theoretical band offset. In Table 4.15 a compilation of experimental valence band offsets for chalcogenide materials is given. Knowing the valence band offset Ev , Eq. (2.17) may be applied to calculate the conduction band offset Ec , provided the bandgaps of the semiconductors at the interface are known. In Figure 2.3a we have drawn the band edges according to the band offsets at each particular interface. This is also the case for the semiconductor/metal interfaces where in order to describe the band alignment, we need to know the difference between the semiconductor valence band edge and the metal Fermi energy from experiment or theory. In the example of Figure 2.3a, all Fermi energies before equilibration are on different energy levels. The establishment of equilibrium requires mobile carriers to be transferred across each newly formed interface. In the example in Figure 2.3: 1) Electrons from the window to the front contact form a narrow depletion region of tunneling length. 2) Electrons from buffer to window and from buffer to absorber lead to complete depletion of the buffer layer. The buffer layer acts as a dielectric in the specific example of Figure 2.3. 3) Holes from absorber to buffer form the absorber depletion region. 4) Holes from absorber to back contact form a back surface band bending which, if too large, acts as a barrier for the diode current. Ionized impurities as well as free carriers form the space charge which enters Eq. (2.16). For the densities of mobile carriers we have to write Ep (z) Ev (z) − EF = Nv (z)F1/2 − p(z) = Nv (z)F1/2 kT kT EF − EC (z) En (z) = Nc (z)F1/2 − (2.18) and n(z) = Nc (z)F1/2 kT kT where F1/2 {η} is the Fermi–Dirac integral of order 1/2 √ ∞ βdβ 2 F1/2 {η} = √ π 0 1 + exp{β − η}

(2.19)

Even for heterostructures of nondegenerate semiconductors, the band offsets can bring the Fermi level very close to the band edges. Therefore, the Boltzmann


approximation is not necessarily applicable and a correct description of the position dependent carrier concentrations is only possible through Eq. (2.18).1) According to Poisson’s equation (Eq. (2.15)), the space charge is connected with the electrostatic potential ϕ(z) and the dielectric function ε(z) which may change at the semiconductor interfaces. Inserting Eq. (2.16) into Eq. (2.15) gives dln{ε(z)} dϕ(z) d2 ϕ(z) + dz dz dz2 Ep En − − Nc (z)F1/2 − (2.20) = −q N+ D (z) − NA (z) + NV (z)F1/2 − kT kT Equation (2.20) is the screening equation for heterostructures with varying band edges Ec,v (z) and electric potential ϕ(z). It is a nonlinear differential equation for ϕ(z). As the complete heterostructure of volume V will be charge neutral, the boundary condition for Eq. (2.20) is

db +dw

−da

ρ(z)dz = 0

(2.21)

Equation (2.20) together with the boundary condition Eq. (2.21) is best solved numerically by a device simulation program. However, simple cases can be solved analytically as will be shown in the next section. In Figure 2.3b we see the electric potential ϕ(z) as a function of position. It is a smooth function which does not include discontinuities due to band offsets. In contrast, the functions Ec,v (z) are subject both to band offsets and to the varying potential. Finally, we want to comment on the orientation of our band diagrams. In contrast to other publications, here the light enters the cell from the right hand side. Thus, the window layer in a band diagram is on the right hand side and the absorber on the left hand side. The reason is that, with this convention, we are able to write the diode equation in such a way that the electric diode current is positive under forward bias and the photo current is negative; at the same time we can respect Eqs. (2.7) and (2.8). Furthermore, by regarding this convention we can speak of a positive electric current if flowing from left to right in an energy band diagram. 2.3.2 Absorber/Window

In the following, we consider a device where the emitter only contains the window layer. We are interested in the potential distribution within such a device. For a heterojunction between two semiconductors without interface charge, an analytical 1) We note, however, that at present time

the Fermi–Dirac integral may not be implemented in a simulation software. Then the integral in Eq. (2.18) is nevertheless

approximated by the exponential of the Boltzmann statistic. This can be a source of (small) deviations in device simulation.

17

18


-qϕ -qϕ(-wa)= qVbi -qϕ(0) -qϕ= 0

-qϕ -qϕ(-wa) =qVbi -qϕ =0 Ec

E

Ec

E

(bc) EV Ep,a

(fc)

EV

(bc)

z= 0

(fc) Ep,a

z=0

E= 0 (a)

absorber -da

E =0

window

-wa0

dw

z

(b)

absorber window -da

-wa0

dw

z

Figure 2.4 Schematic of absorber/window heterojunction band diagrams with (a) symmetric doping, that is, εa NA,a = εw ND,w and (b) asymmetric doping, that is, εw ND,w εa NA,a .

solution of Eq. (2.20) can be derived [23]. The following approximations are employed: 1) Charge contributions due to mobile carriers are omitted. 2) Impurity atoms are completely ionized in the space charge regions (SCRs) (depletion approximation). 3) The dielectric functions are constant within each layer, that is, dεm /dz = 0 with m = a, b, w. We consider the heterojunction between the n-type window and the p-type absorber of Figure 2.4. The built-in voltage, Vbi , of these structures is derived from the balance qVbi = Eg,a − Ep,a − En,w + Ec

(2.22)

where Eg,a is the bandgap of the absorber. Note that Vbi shall be a positive quantity. The moduli of the reduced chemical potentials of holes in the absorber (Ep,a = EFp − Ev,a ) and electrons in the window (En,w = Ec,w − EFn ) are defined in Figure 2.1. The conduction band offset Ec may be positive (type I) or negative (type II) or zero as in the present example. We put the interface at the position z = 0. The potential drop in the window layer is given by −ϕ(0) = Vbi εa NA,a /

(2.23)

where we set = εa NA,a + εw ND,w . As becomes apparent in Section 2.6.3, a critical quantity for the performance of a heterostructure solar cell is the type inversion of the absorber at the absorber/emitter interface. Type inversion in a p-type absorber means that, in the cell’s equilibrium, the quantity Ep,a is small in the bulk and is large at the interface. In the best case this would mean that Ep,az=0 approaches Eg,a . In general, we want to use the quantity Ep,az=0 for the degree of absorber


inversion. An equivalent notation used in other works is the hole barrier φp . For a heterojunction of an n-type window and p-type absorber it is Ep,az=0 (V) = Ep,a + q(Vbi − V)εw ND,w /

(2.24)

We see from Eq. (2.24) that Ep,az=0 becomes large if εw ND,w is much larger than εa NA,a . Through Vbi in Eq. (2.22), Ep,az=0 increases with increasing Ec . A large value of Ep,az=0 is realized in Figure 2.4b. This is the nonsymmetric n+ p-heterojunction. The depletion region mostly extends into the absorber. The interface between window and absorber may contain interface charge. For the beginning, we assume that the interface charge density is independent of the Fermi energy at the interface. Thus, the interface charge density QIF = qNIF is fixed by the assumption of NIF charged interface states. In reality, this is not the case since the Fermi energy dictates the occupation of the states NIF . Nevertheless, for the time being it is instructive to discuss fixed charge densities. Moreover, the absolute interface charge density may be the measured quantity of a capacitance experiment [24]. The neutrality condition of an absorber/window heterostructure with interface charge is −qNA,a wa + zqNIF + qND,w ww = 0

(2.25)

where wa and ww are the depletion widths of absorber and window. It is z = +1 for a positive interface charge and z = −1 for a negative one. Equation (2.25) is valid as long as free carriers can be neglected. With the neutrality condition in Eq. (2.25), an analytical solution of Eq. (2.20) can be obtained and we can calculate the Fermi level position at the absorber surface

qε ND,w qN2IF Ep,az=0 (V) = Ep,a + W (Vbi − V) − εw ND,w − εa NA,a 2 2 2 zqNIF 2 (V ε N ε N − V) − N + w D,w a A,a bi IF 2 q

(2.26)

The first two terms of Eq. (2.26) are identical to Eq. (2.24) while the last two terms describe the influence of a positive or negative charge at the interface. In Figure 2.5 we investigate an n+ /p heterostructure where the negative interface charge has been varied. If not otherwise stated the device parameters are as given in Table 8.1. Due to the higher doping of the window, the Fermi level of the absorber at the interface with low interface charge is close to the conduction band, that is, Ep,az=0 > Eg,a /2. A density of negatively charged states of NIF = 1012 cm−2 can shift Ep,az=0 from close to the conduction band up to close to the valence band. Negative interface charge removes the inversion. The influence of the interface states, however, depends on the window doping: the higher the window doping, the less sensitive is the heterostructure to interface states. The width of the absorber’s depletion region is given by

2 2

zεa NIF εa εw ND,w q (Vbi − V) − NIF + (2.27) wa (V) = NA,a 2

19

2 Thin Film Heterostructures -Qi [C cm−2] 10−7

+Qi [C cm−2]

10−8

8 6

1.2

4

2

10−8

8 6

4

2

2

4

6 8

10−7 2

4

6 8

1019 1018

z =0

[eV]

1.0

Ep,a

20

0.8

NA,a =1016 cm−3

ND,w = 1017 cm−3

0.6 1017

0.4 0.2 0.0

1018 8 6

1012 (a)

4

2

8 6

1011

4

2

2

1010 −2

-zNi [cm ]

Figure 2.5 Calculated Fermi level position at the absorber surface Ep,az=0 as a function of the interface charge density of an absorber/window heterostructure. In (a) it is NA,a = 1016 cm−3 and ND,w is the varying

4

6 8

1010 (b)

2

1011

4

6 8

1012 −2

+zNi [cm ]

parameter. In (b) it is ND,w = 1016 cm−3 and NA,a is the varying parameter. Conduction band offset Ec = 0. Other parameters as in default Table 8.1.

where again it is = εa NA,a + εw ND,w and z gives the charge state of the interface state density NIF . For NIF = 0 and ND,W NA,a , we have 2(Vbi − V)εa wa (V) = (2.28) qNA,a For completeness we give the window’s depletion layer width

2 2

zεw NIF εa εw NA,a q (Vbi − V) − NIF + ww (V) = − ND,w 2

(2.29)

2.3.3 Absorber/Buffer/Window

Next we turn to a heterostructure of three semiconductors of the kind n-type window/n-type buffer/p-type absorber which was introduced in Figure 2.3a,b. We again put the focus on the Fermi level position at the absorber surface and will vary the doping densities and interface charge. This holds Ep,az=0 = Eg,a + Ec w,b + Ec b,a − En,w + qϕ(0)

(2.30)

where Ec w,b and Ec b,a are the conduction band offsets at the buffer/window and buffer/absorber interface. In the example of Figure 2.3 it is Ec w,b < 0 and Ec b,a > 0. ϕ(0) is the potential drop in the window and buffer layer (qϕ(0) < 0 in


Figure 2.3). If we put χ = Ec w,b + Ec b,a , Eq. (2.30) indicates that χ must be large in order to obtain a large Ep,az=0 . The quantity ϕ(0) is best found by numerical solution of Eq. (2.20). For the example in Figure 2.3, however, ϕ(0) may be approximated by an analytical solution. Approximations (1–3) are as given in Section 2.3.2, but in addition the window doping (4) is much higher than the buffer and absorber doping. Therefore, the potential drop in the window layer is negligible. Also, the buffer layer (5) is completely depleted of mobile charge carriers; and the Fermi energy (6) is flat within all layers. Thus, ϕ(0) can be calculated by integration of the charge density per area in the window qND,w ww and in the buffer qND,b wb ϕ(0) = −

q q ND,b w2b − ND,w wb ww 2εb εb

(2.31)

where wb , ww are the widths of the depletion regions in buffer and window and ND,b , ND,w are the donor densities of the respective layers. If we add up the energies from the left side of the buffer/absorber interface, we obtain a second equation for Ep,az=0 Ep.az=0 = Ep,a +

q2 NA,a w2a 2εa

(2.32)

where wa is the width of the absorber depletion region and NA,a the acceptor density in the absorber. Together with the charge neutrality condition, we have three equations with the three unknowns: Ep,az=0 , wa , and ww . According to approximation (5), wb is given by the total width of the buffer layer, that is, wb = db . The band diagram in Figure 2.6a depicts a heterostructure without interface charge in equilibrium. We have selected the special case of ND,b = NA,a = 1016 cm−3 . Also given in Figure 2.6a are carrier densities, electric field, and volume charge. Due to the high n-type doping of the window layer of ND,w = 1018 cm−3 the band bending in the window indeed is negligible. If we inspect the density of free carriers in the buffer layer in Figure 2.6a, we find that electron and hole concentrations are below 1015 cm−3 everywhere in the buffer layer. It are the band and Eb,a offsets, Ew,b c c , which predefine the conduction band of the buffer layer and support the buffer depletion. Therefore, even for ND,b = NA,a the buffer is fully depleted and all the approximations listed above can be applied. The built-in voltage of the cell in Figure 2.6a is given by qVbi = Eg,a − Ep,a − En,w + χ

(2.33)

The charge neutrality condition of the cell in Figure 2.6a without interface charge may be expressed as qND,w ww + qND,b wb − qNA,a wa = 0

(2.34)

Equation (2.34) is only valid as long as wa does not extend up to the SCR of the back contact. With the approximations listed above we can solve Eqs. (2.30–2.32)

21


22

EF

Ec

Ec

Ev

EA,d EF Ev

p

1015 1010 105

p n

n 2

+ 1017

+

14

10 10

(a)

−

(b)

−

(c)

−

+ (d)

−

1012 + 1010 1012

−

17

+

+

−0.6 −0.4−0.2 0.0 0.2

−0.6−0.4−0.2 0.0 0.2

−0.6 −0.4 −0.2 0.0 0.2

−0.6−0.4 −0.2 0.0 0.2

Area Charge [cm−2]

0 −2× 105

−

Volume Charge [cm−3] F [V cm−1] n,p [cm−3]

E [eV]

4 3 2 1 0

Position z [µm]

Figure 2.6 Simulated solution curves of an absorber/buffer/window heterostructure: (a) without interface charge, (b) with area charge of +8 × 1011 q cm−2 at the buffer/absorber interface, (c) with area charge of 2 × 1012 q cm−2 at the buffer/window interface, and (d) with acceptor states of density 5 × 1017 cm−3 at EA,d = 0.6 eV (dashed-dotted line) in the absorber layer. Other parameters as in default Table 8.1. Note that (a) shows the default device from Table 8.1 in equilibrium. From bottom to top: total electric volume charge

S(z) according to Eq. (2.16) on a logarithmic scale (left axis) and electric area charge Qss (right axis). Area charge is given as a full line. The zero line was set to a concentration of 1014 cm−3 for the volume charge and 1010 cm−2 for the area charge. Electric field F(z). Small jumps in the electric field in (a) indicate differences in the dielectric constant. Densities of mobile electrons and holes. In the absorber, the densities exhibit a crossing point of n = p which locates the electronic junction of the heterostructure.

and (2.34) for Ep,az=0 thus

Ep.az=0 (V) = Eg,a 1 − κ−1 IF +

κ−1 IF

q(Vbi − V) + Ep.a−νIF z1

q2 ND,b d2b q2 NA,a db + − wa (V) 2εb εb

(2.35)

where κIF = 1 and νIF = 0 in the case of absent interface charge. The width of the absorber SCR, wa (V), is given by

2 2N

εa d b 2 d q εa db 2ε D,b a b wa (V) = − q(Vbi − V) + + + 2 + νIF z2 εb κIF εb κIF q NA,a κIF 2εb (2.36)

2.3 Energy Band Diagrams 1.2

Ep,a z = 0 [eV]

1.0 0.8

1015 cm−3

1015 1016

1015

0.6 0.4

1016 10

NA,a = 10

17

(a)

ND,b = 1017 cm−3

1016 ND,b

cm

−3

cm

0.2 0.0 1015

16

−3

[cm−3]

1017

1015 (b)

1016 NA,a

Figure 2.7 Fermi energy referenced to valence band maximum, Ep,az=0 , at the buffer/absorber interface of an absorber/buffer/window heterostructure without interface charge (NIF = 0). The data shown are: simulated values (symbols) and calculated values after Eq. (2.35) (lines). (a) Data

[cm−3]

1017

NA,a = 1017 cm−3

10−7 (c)

10−6

10−5

10−4

db [cm]

shown as a function of buffer donor density with parameter NA,a . (b) Data shown as a function of the absorber acceptor density with parameter ND,b . (c) Data shown as a function of the buffer thickness with ND,b = 1015 cm−3 and with parameter NA,a . Other parameters as in default Table 8.1.

and the electric field at the buffer/absorber interface is given by

qNA,a db 2 2NA,a qNA,a db q2 ND,b d2b Fz=0 (V) = − + + + νIF z2 q(Vbi − V) + εb κIF εb κIF εa κIF 2εb (2.37) Figure 2.7a gives values of Ep,az=0 calculated after Eq. (2.30) as a function of the buffer donor concentration for different absorber acceptor concentrations (full line). The calculated values of Ep,az=0 are compared to values for Ep,az=0 obtained by device simulation (markers). In Figure 2.7b, Ep,az=0 is plotted as a function of the absorber acceptor concentration for different buffer donor concentrations. The following trends for achieving a large value of Ep,az=0 may be observed in Figure 2.7a,b: At low buffer doping of ND,b = 1015 cm−3 and fixed thickness of the buffer layer, Ep,az=0 only depends on the absorber doping. The buffer acts as a dielectric and the potential drop in the buffer is determined by the charge in the window and absorber. This charge increases with increasing absorber doping and as a result Ep,az=0 decreases. However, the potential drop in the (lowly doped) buffer layer also depends on the buffer thickness. The formula of the parallel plate capacitor, V = qdb /εb ε0 , shows that the potential drop linearly increases for a larger buffer thickness. For a larger buffer thickness, the buffer starts to counterbalance the complete absorber charge and the value of Ep,az=0 can be approximated according to the simple absorber/emitter heterojunction model of Eq. (2.24). If the buffer doping approaches the absorber doping, Ep,az=0 starts to rise in Figure 2.7a. Now the charge in the buffer layer becomes sufficient to partly counterbalance the charge in the absorber. Thus, if we aim for a low value of Ep,az=0 due to device reasons, a higher value of NA,a is tolerable for high ND,b . Certainly, Ep,az=0 becomes larger if the absorber acceptor concentration is

23

2 Thin Film Heterostructures 120 × 10−6 100

ND,b = 1017 cm−3 1016 cm−3

80 Wa [cm]

24

1015 cm−3

60 40 20 0 1015

1016

1017

NA,a [cm−3]

Figure 2.8 Width of absorber space charge region calculated after Eq. (2.36) (lines) and simulated (symbols) as a function of absorber acceptor density for fixed buffer donor densities of ND,b = 1015 , 1016 , and 1017 cm−3 .

decreased. The conclusion of Figure 2.7a,b can be expressed in the rule that ND,b has to be larger than NA,a for a large Ep,az=0 . Following this rule transforms the absorber/buffer/window heterostructure in Figure 2.6 into a simple heterojunction between an n-type emitter and a p-type absorber. Based on these considerations, we understand that in Figure 2.6a the inversion is not complete since the condition ND,b > NA,a is not fulfilled. Although in equilibrium electrons are still majority carriers at the absorber surface, this will change for small electric forward bias. The volume charge within the window layer is high but restricted to a small region. As a result, the potential drop in the window layer can be neglected [which was conditional for Eq. (2.31)]. The small offsets of the electric field at the buffer/window and buffer/absorber interfaces are due to differences in the dielectric constant. Figure 2.8 depicts the width of the SCR calculated after Eq. (2.36) as well as simulated data for wa . Calculated and simulated values are in good agreement. 2.3.4 Interface States

Interface states may add additional charge to the charge neutrality equation and thus may change the potential distribution. If interface states are charged or not, depends on the position of the Fermi energy at the respective interface. Before we take this into account we want to simplify matters and to consider the effect of interface charge QIF only. We do this by assuming that a given density of interface states NIF is fully charged – either negative or positive. For interface charge at the buffer/window or the buffer/absorber interface the charge neutrality condition reads qND,w ww + qND,b wb + zqNIF − qNA,a wa = 0

(2.38)


−QIF [C cm−2] 10−7 1.2

4

2

10−8

8 6

4

2

2

4

6 8

10−7 2

4

6 8

1015

1.0

Ep,a z= 0 [eV]

+QIF [C cm−2]

10−8

8 6

25

0.8 1016 NA,a = 1017 cm−3

0.6 0.4 0.2 0.0

8 6

1012 (a)

4

2

8 6

4

2

1011

2

1010 −2

−zNIF [cm ]

Figure 2.9 Device as described in caption of Figure 2.7 but with (a) negative charge density −qNIF and (b) positive charge density +qNIF at the buffer/absorber interface. Values of Ep,az=0 obtained by device simulation (symbols) are given as a function of

1010 (b)

4

6 8

2

4

1011

6 8

1012 −2

+zNIF [cm ]

the charged interface state density for absorber acceptor densities of 1016 , 5 × 1016 , and 1017 cm−3 . Other parameters as in default Table 8.1. Solid lines represent Ep,az=0 calculated according to Eq. (2.35).

In the case of charge at the buffer/absorber interface, Eq. (2.35) is the solution for Ep,az=0 , if we insert as parameters κIF = 1, νIF = q2 NIF db /εb (where NIF is given in units of cm−2 ). We further put z1 = z2 = +1 in the case of charged interface donors and z1 = z2 = −1 in the case of charged interface acceptors. Figure 2.9 shows simulated and calculated values of Ep,az=0 as a function of negative (a) and positive (b) charge densities at the buffer/absorber interface. Note that NIF on the abscissa of Figure 2.9 is the density of fixed charge divided by the element charge. Thence, NIF is the density of charged interface states. In Figure 2.9, we find that Ep,az=0 can strongly be influenced by a charged state density exceeding ±1011 cm−2 . In absolute terms, Ep,az=0 changes if the interface charge density becomes comparable to the buffer and absorber charge per area, the latter given by the product of doping density and depletion width. Negative charge at the buffer/absorber interface has the same effect as a higher absorber acceptor concentration: It decreases Ep,az=0 and reduces the inversion. Vice versa, a positive charge resembles a higher buffer donor concentration. It increases Ep,az=0 and reinforces the inversion. Ep,az=0 values calculated after Eq. (2.35) follow the simulated data points as long as the condition ND,b ≤ NA,a is fulfilled. Also the charge at the buffer/window interface can influence the band diagram. In Figure 2.10, we plot simulated values of Ep,az=0 as a function of negatively and positively charged interface state densities. A negative charge at the buffer/window

2 Thin Film Heterostructures −QIF [C cm−2] 10 1.2

6 5 4 3

z =0

[eV]

1.0

Ep,a

26

2

10−7 76 5 4 3

2

2

3

4 5 67

2

3 4 5 6

1015

0.8 1016

0.6 0.4

NA,a = 1017 cm−3

0.2 0.0

4 3

2

76 5 4 3

1012 (a)

+QIF [C cm−2]

−7

2

2

1011

−zNIF [cm−2]

Figure 2.10 Device as described in the caption of Figure 2.7 but with (a) negative charge density −qNIF at the buffer/window interface and (b) positive charge density +qNIF . Values of Ep,az=0 as obtained by

1011 (b)

3

4 5 67

2

3 4

1012 +zNIF [cm−2]

device simulation are given as a function of the charged interface state density with NAa as parameter. Other device parameters as in default Table 8.1.

interface has the same effect as negative charge at the buffer/absorber interface. It decreases Ep,az=0 and reduces the inversion. Compared to the buffer/absorber interface, however, a higher charge density is required at the buffer/window interface. This is due to the fact, that the positive charge density in the window layer is higher than in the buffer layer. No influence on the absorber inversion has a positive charge at the buffer/window interface (in contrast to positive charge at the buffer/absorber interface). Such a positive charge is screened by electrons in an accumulation region of the window layer. In the case of buffer/window interface charge, Ep,az=0 cannot be calculated by use of Eq. (2.35) since the potential drop in the window layer can no longer be omitted. In Figure 2.6b, the band diagram, carrier densities, and so on, in the case of interface charge at the absorber/buffer interface are depicted. While without interface charge (Figure 2.6a) the absorber inversion was not complete, it becomes complete in the case of 1 × 1012 cm−2 positive charge density (see right axis) at the buffer/absorber interface. Now the electron concentration at the absorber surface exceeds the hole concentration by more than 15 orders of magnitude. Electrons remain majority carriers at the absorber surface even for not too large a forward bias. The electric field shows a jump at the absorber/buffer interface due to the fixed interface charge. The electric field region mostly is located within the absorber. An opposite effect has a negative charge either at the absorber/buffer or at the buffer/window interface: It decreases the inversion of the absorber surface. In Figure 2.6c, we give the example for negative charge at the buffer/window interface. Looking at the volume charge, we see that a positive volume charge in window and buffer screens the negative area charge. As a result of charge balance, less space charge is allocated in the absorber and the potential drop in the absorber is lower. The absorber is not in inversion and holes are majority carriers at the absorber


surface. Thus, in an absorber/buffer/window heterostructure, the buffer/window interface can be extremely important. Now, we consider the more realistic case that the Fermi energy at the interface determines the charge in a given interface state density. The interface charge is given by EA,IF − EF QIF = −qNA,IF F1/2 KT ED,IF − EF and QIF = +qND,IF [1 − F1/2 (2.39) KT in the case of acceptor and donor states of density NA,IF and ND,IF . EA,IF and ED,IF are the energy levels of the interface acceptor and donor states and F1/2 is the Fermi–Dirac function. Thus, the charge neutrality condition for discrete interface states is qND,w ww + qND,b wb + qIF − qNA,a wa = 0

(2.40)

Due to the exponential term in the Fermi–Dirac function (and also in the Boltzmann approximation), Eqs. (2.30–2.32) and (2.40) cannot be solved analytically. In the following, we continue to focus on the absorber surface and its degree of inversion. As an example, we introduce discrete interface states of acceptor and donor type and of varying density at the buffer/absorber interface of a heterostructure as depicted in Figure 2.6. Two critical cases are investigated: First, acceptor states at different energies for a heterostructure with ND,b > NA,a . As we learn from Figure 2.7, without interface states this heterostructure has an inverted absorber. And second, donor states at different energies for a heterostructure with noncomplete inversion due to ND,b < NA,a . Figure 2.11 gives Ep,az=0 for (a) acceptor and (b) donor states of different energy levels as a function of the interface defect density, respectively. The defect states are located at 0.9, 0.7, and 0.5 eV above the valence band of the absorber. We select different ratios of absorber/buffer doping in order to illustrate the effects of acceptor and donor interface states. In Figure 2.11a Ep,az=0 is large or the inversion is strong with 1010 cm−2 acceptor states present at the interface. Acceptor states of density > 1011 cm−2 reduce the inversion of the absorber and Ep,az=0 decreases. A different starting point is chosen in Figure 2.11b where, due to the absorber/buffer doping ratio, the inversion is weak and Ep,az=0 is ∼0.6 eV. Now donor states located close to the conduction band enhance the inversion. The closer their energy level to the conduction band, the stronger is the inversion. As an example, for a state density of 1013 cm−2 the Fermi energy is more than 1 eV above the valence band at the interface while the donor defect is 0.9 eV. Thus, even though the donor states are only partially ionized (empty) the positive interface charge is sufficient to substantially add to the positive charge of buffer and window. As a result, the absorber SCR is enlarged and the potential drop in the absorber becomes higher. Another variation explains the phrase ‘Fermi level pinning’ for a heterostructure with interface states. For a fixed density of discrete interface states energetically

27

28


Ep,a z = 0 [eV]

−QIF [C cm−2]

(a)

10−5 10−6 10−7 10−8 1.2 1.0 EA,IF = 0.9 eV 0.8 0.7 eV 0.6 0.5 eV 0.4 0.2 0.0 1014 1013 1012 1011 1010 NA,IF

[cm−2]

QIF [C cm−2] 10−8 10−7 10−6 10−5 0.9 eV 0.7 eV

No IF states

EA,IF = 0.9 eV 0.7 eV 0.5 eV

ED,IF = 0.5 eV

ND,b =5× 1016 cm−3

1010 1011 1012 1013 1014 (b)

ND,IF

Figure 2.11 Device as described in the caption of Figure 2.7 but with varied buffer and absorber doping. (a) ND,b = 5 × 1016 cm−3 and NA,a = 1 × 1016 cm−3 and acceptor states NA,IF at the buffer/absorber interface. The acceptor states are at different discrete energy levels Ec − EA,IF in the bandgap of the absorber. (b) Buffer and absorber

[cm−2]

1015 (c)

1016 NA,a

1017

[cm−3]

doping of ND,b = 5 × 1016 cm−3 and NA,a = 1 × 1017 cm−3 and interface donor states ND,IF at different energy levels. (c) Buffer doping of ND,b = 5 × 1016 cm−3 and varying absorber acceptor doping with discrete acceptor states of density NA,IF = 1013 cm−3 at different energy levels in the absorber bandgap.

located at a certain energy, a variation of the absorber acceptor concentration has no influence on the Fermi level position at the interface or, in other words, no influence on Ep,az=0 . This is visible in Figure 2.11c, were we plot Ep,az=0 for discrete acceptor levels of density 1013 cm−2 located at EA,IF = 0.5, 0.7, and 0.9 eV. We see that with decreasing absorber acceptor concentration, the inversion cannot further be increased. Instead, the Fermi level is ‘pinned’ at the acceptor interface state energy. Like in the bulk of a semiconductor, an interface may exhibit different states of both donor and acceptor type. Unlike in the bulk, their occupation is not easy to predict and depends on the doping of the adjacent layer. Consider, for instance, the case of ND,b > NA,a and acceptor states at EA,IF = 0.5 eV in Figure 2.11a. Due to an acceptor state density of NA,IF = 1013 cm−2 the Fermi energy lies at Ep,az=0 = 0.4 eV. Higher lying donor states of density comparable with the acceptor will improve the situation and will pin the Fermi energy at a larger value of Ep,az=0 due to their positive charge. In contrast, for the other case ND,b < NA,a (Figure 2.11b) the same combination of acceptor and donor states has no influence on the Fermi level position. In that case, the Fermi energy stays at a large distance from the conduction band since the charge in the donor states is not sufficient. The case of more than one donor (acceptor) level at different energies at the interface is more simple. Usually, it is the donor (acceptor) energetically located nearer to (farer from) the conduction band edge which determines the band bending and thus the value of Ep,az=0 . Lower (higher) lying donors (acceptors) may not be charged and, therefore, may not contribute to the overall charge balance. The extreme variant of multiple interface states of different ionization character (acceptor and donor) is a continuous interface defect density (per area and energy). If this state density derives from the complex band structure [25], it exhibits


acceptor character close to the conduction band and donor character close to the valence band. Depending on the position of the Fermi level, the interface either exhibits a net positive charge if the Fermi level lies below a certain charge neutrality level or exhibits a net negative charge if the Fermi level lies above a charge neutrality level. If the Fermi level coincides with the charge neutrality level, the net charge in the interface states is zero. The shape of the interface state density as a function of energy in the gap may be ∪-type with increasing density toward the band edges. In the vicinity of the charge neutrality level it may be approximated by a constant density. A constant interface state density with charge neutrality level has been assumed in Ref. [26]. With the parameters νIF = q2 NIF wb /εb , κIF = 1 + νIF , z1 = φn − Eg,a , and z2 = φn − Ep,a Eq. (2.35) can be used to calculate the quantity Ep,az=0 for charge neutrality levels φn located at different energy positions in the bandgap. φn is measured from the valence band edge. The results are similar as for discrete interface states: at a sufficient large density of interface states, the Fermi level is pinned at the charge neutrality level, regardless of the doping ratio between buffer and absorber which dictates the position of the Fermi level at the interface without interface states. In conclusion, acceptor states at the buffer/absorber (or buffer/window) interface push the Fermi energy downwards and diminish the inversion. In contrast, donor states energetically located close to the conduction band can enhance the inversion. Certainly, interface states not only modify the band bending but also can contribute to recombination. For instance, there could be a trade off between beneficial donor states in terms of inversion and detrimental donor states in terms of recombination. 2.3.5 Interface Dipoles

Interface dipoles can change the band offset at an interface. They can be introduced deliberately by use of ‘band offset engineering’ [20] or they may be the result of differences in the fabrication of an interface. To date chalcogenide heterostructures exhibit isovalent or pseudo-isovalent interfaces. As pointed out by Franciosi and van de Walle [20], isovalent interfaces are difficult to engineer and only small deviations from the canonical band line up are to be expected. The implication of a possible interface dipole may easily be conceived based on the value of Ec at the buffer/absorber interface in Figure 2.3. A dipole with negative charge at the buffer side and positive charge at the absorber side would increase the value of Ec . Thereby, a type II interface could be inverted to a type I interface. 2.3.6 Deep Bulk States

The semiconductors of our heterostructure may contain defect states within the bandgap. If the defect states become charged, they add to the space charge of the heterostructure, similar as interface states. Thus, they influence the potential

29

30


drop in the layers. Certainly, defect states in the bandgap also give rise to carrier recombination. In equilibrium, the charge of the defect states is governed by the Fermi level. Donor-like (acceptor-like) defect states sufficiently above (below) the Fermi level contain positive (negative) charge. Under non-equilibrium, that is, voltage or light bias, the occupation of defects states is governed by the kinetics of electron and hole capture. In the following, we investigate the influence that deep states in the p-type absorber layer have on the band diagram. As an example, we select the standard heterostructure of window, buffer, and absorber, as depicted in equilibrium in Figure 2.6a. Without deep states, the absorber is in inversion. Now we introduce deep acceptor states of density NA,d = 5 × 1017 cm−3 (σp mp ∗ /σn mn = 103 ) in the absorber layer (Figure 2.6d). Due to the position of the Fermi energy, the deep acceptor states are not charged in the QNR. Thus, the hole concentration about equals the doping concentration of NA,a = 1016 cm−3 at z = da . At the buffer/absorber junction, however, the deep acceptors become charged due to absorber band bending. Consequently, they add to the space charge of the absorber. Band bending in the absorber is enforced close to the junction but in total the inversion is reduced. A higher interface charge would be required in order to preserve inversion of the absorber surface. In total, the deep acceptor states have modified the band diagram at the buffer/absorber junction. The SCR of the absorber in Figure 2.6d is divided into two subregions of volume charge density: ρ/q = −5 × 1017 cm−3 and −1 × 1016 cm−3 . In order to solve the screening equation (Eq. (2.20)), the absorber would require to be calculated as two different layers. Thus, Eqs. (2.35) and (2.36) can no longer be applied. Meyer et al. developed [27] a differential model to calculate small variations of a band diagram caused by a differential supplement charge in the absorber and emitter layer. Using this model, voltage and light induced effects on the diode current and on the space charge capacitance can be described. Subregions of different volume charge can also be the effect of compensation doping. Consider the case with NA,a = 5 × 1017 cm−3 and ND,d = 4.9 × 1017 cm−3 . In the vicinity of the buffer/absorber junction, the deep donors are not charged and do not compensate the high p-type doping of NA,a = 5 × 1017 cm−3 . The band diagram would be similar to Figure 2.6d. How could the cases of charged deep traps experimentally be diagnosed? The absorber width of the SCR with deep states is only slightly reduced to about 0.2 µm compared to the case without deep states (see Figure 2.6a). Thus, the zero voltage capacitance of all devices would be rather similar. Capacitance–voltage measurements, however, would reveal a change in slope of a C−2 (V) plot indicating two regimes of different doping concentrations [28]. A possible compensation doping can be determined from the temperature dependence of the dark carrier concentration [29]. Furthermore, deep trap levels can be analyzed by deep-level transient spectroscopy [30]. Compared to the case without deep states, the electric field of the device in Figure 2.6d is much enlarged at the buffer/absorber interface. This can have a beneficial effect on the Voc since the effective recombination zone in the SCR


is reduced (see Section 2.4.5.1). However, a field strength of 2 × 105 V cm−1 can cause tunneling enhancement of recombination (see Section 2.4.5.1). The tunneling would increase the diode current under solar cell operation and thereby reduce Voc . Certainly, deep states may also act as efficient recombination centers (RCs) (depending on the capture cross sections) and thereby reduce Voc . Cases where the occupation of deep states is illumination and bias dependent is treated in Section 2.4.8. In the dark, the occupation of deep states in the bandgap of a semiconductor is governed by the Fermi–Dirac function. Under non-equilibrium conditions such as voltage and light bias, however, the occupation of deep defects depends on transition rates between the defects and the conduction and valence bands. The transition rates are proportional to the capture cross sections. If the capture cross sections for electrons and holes are largely different (which is normally the case) a steady state occupancy with one charge carrier type may occur. The trapped photo generated charge carriers then may alter the band diagram with respect to the case without illumination. An example is given in Figure 2.12 where the absorber of 1016 cm−3 p-type doping contains 5 × 1017 cm−3 deep acceptors energetically located at midgap energy. If the deep acceptors exhibit equal capture cross sections (σn = σp = 10−15 cm−2 ), the band diagram under illumination and in short circuit condition resembles the case in the dark (see Figure 2.6d). If it is σp σn , however, holes photo generated within the absorber layer will become trapped in the deep acceptor states thus reducing the net negative charge close to the buffer/absorber junction. As a result, the charge density in the absorber close to the buffer/absorber junction is reduced and the band diagram under illumination approaches the case without deep states (see Figure 2.6a). As a second example of the deep states’ influence on the band diagram we introduce deep acceptors in the buffer layer. We assume that the buffer layer is compensated. The net n-type doping of 1015 cm−3 is realized by a background doping with ND,b = 5 × 1018 cm−3 donors and a density of NA,d = 4.999 × 1018 cm−3 compensating deep acceptors. The band diagram in the dark is depicted in Figure 2.13a. The net space charge of ρ/q = +1015 cm−3 in the buffer is not sufficient to balance the negative absorber space charge. Thus, the buffer acts as a dielectric and, due to the missing interface charge, the absorber is only in slight inversion. The majority of deep acceptor states in the buffer is occupied or, in other words, is negatively charged in the dark. This may change, however, under illumination where photo generated holes may discharge the deep acceptor states. If the capture cross section of holes largely exceeds the capture cross section of electrons, under steady state conditions the deep acceptors become empty. The compensation is reduced and the positively charged donors add to the space charge. This case is reflected in Figure 2.13b. Under illumination, the buffer space charge can partly balance the absorber space charge. As a result, the band bending in the buffer layer is reduced. If we consider the electron barrier φb n from the window to the absorber in Figure 2.13, we see that this barrier is reduced for the case σp σn . Such a barrier reduction can be the origin of a cross over between JV curves measured in the dark and under illumination (see Section 7.1.2).

31


4 E [eV]

3

Ec EFp

EA,d EFn

EFn

EFp

2 1 Ev

0

F [V cm−1]

n,p [cm−3]

p

Volume Charge [cm−3]

32

1015

p

1010

n

n

105

2 0 −2 × 105

1017 10

+

+

14

−

−

1017 −0.6

−0.4

(a)

−0.2

0.0

0.2

Position z [µm]

Figure 2.12 Solution curves to show the influence of capture cross sections of deep acceptor states in the absorber. Device under AM1.5G illumination in short circuit. (a) Equal capture cross sections of electrons and holes of σp = σn = 10−15 cm−2 . (b) Hole capture cross section σp = 10−12 cm−2

−0.6 (b)

−0.4

−0.2

0.0

0.2

Position z [µm]

and electron capture cross section σn = 10−15 cm−2 . Deep acceptor state density NA,d = 5 × 1017 cm−3 . Other parameters as in default Table 8.1. For comparison, Figure 2.39a gives the default device (without deed acceptor states) under short circuit.

2.3.7 Bandgap Gradients

Bandgap gradients may occur in any semiconductor constituting the chalcogenide solar cell. They may occur as gradients in the conduction band, in the valence band or in both bands. In this section, we present some possible bandgap gradients in the absorber layer. By aid of simulated examples we want to exemplify Eqs. (2.9) and (2.10) which state that bandgap gradients can add to the current density of the device. Figure 2.14 sketches absorbers with normal and double bandgap gradients. Only linear gradients occurring within a limited depth of the absorber are considered.


Ec

E [eV]

4

Ec

φbn

φbn

3 2

EA,d

EA,d

Ev

Ev

1

Volume Charge [cm−3]

F [V cm−1]

n,p [cm−3]

0

(a)

1015

p

1010 105

n

2 0 −2 × 105 +

1017 1014

+

−

−

1017 −0.4

−0.2

0.0

0.2

Position z [µm]

Figure 2.13 Solution curves to show the influence of capture cross sections of deep donors in buffer. Default heterostructure but with high degree of compensation in the buffer layer. Background doping in the buffer is ND,b = 5 × 1018 cm−3 .

0.4

−0.4

−0.2

(b)

Position z [µm]

0.0

0.2

0.4

Compensating acceptors in the buffer of density 4.999 × 1018 cm−3 at midgap with capture cross section for electrons and holes of σp = 10−12 cm−2 and σn = 10−18 cm−2 , respectively, (a) in the dark and (b) under AM1.5G illumination in short circuit.

Normal bandgap gradients widen the energy gap toward the back surface (a,b) or toward the front surface (c,d) of the semiconductor. A V-shaped bandgap is the origin of a double bandgap gradient (e,f). If we allow only one change in slope of the bandgap located at dmin , the number of possible combinations of front and back surface gradients in the conduction and valence bands amounts to 15. Not all combinations, however, are practical and useful. The bandgap gradients in Figure 2.14 can be described by the parameters Ec , Ev , and dmin . The effect of a back surface gradient is to decrease minority carrier recombination at the back contact. This leads to an increased photo current and decreased diode current. The impact of a back surface gradient becomes larger for a smaller distance between back contact and collecting junction. The relevant quantity for the discussion of bandgap gradients is the current density of the minority carriers.

33

34


∆Ec,back ∆Ev,back

Ev (a)

(b)

-dmin

∆Ec,front (c)

(d)

(e)

(f)

E

∆Ev,front

-dmin

-dmin,f z

(g)

-dmin,b

(h)

Figure 2.14 Eight examples (a–h) of bandgap gradients in absorber layers with light entering from the right side. dmin is referenced to z = 0 at the right hand side of the absorber and is a positive quantity.

In the p-type absorber this is the electron current density Jn . According to Eq. (2.9) (for a one-dimensional problem), Jn is composed of the field term Jn,E (z) = −µn (z)n(z) dqϕ(z)/dz, the effective force field term Jn,χ (z) = −µn (z)n(z)dχ(z)/dz and a term which depends on the chemical potential of the electrons Jn,ζ (z) = −µn (z)n(z) dEn (z)/dz. For a nondegenerate semiconductor, dEn (z)/dz is given by Eq. (2.11). If the effective state density, Nc , in the conduction band is constant, the term of the reduced chemical potential becomes a pure diffusion term of −kTµn (z) dn(z)/dz. Equation (2.9) then reads Jn = Jn,E (z) + Jn,CBM (z) + Jn,ζ (z). In Figure 2.15 we show examples of cells (a) without and (b) with a back surface gradient under short circuit condition. The different current contributions according to Eq. (2.9) have been plotted separately. At the back surface at x = 1.5 µm the default device (Table 8.1) exhibits a high surface recombination velocity of Sn0 = Sp0 = 107 cm s−1 . Without back surface gradient, the cell collects electrons from about −0.6 µm. The starting point of the collection region is indicated by a sign reversal of the current density Jn . For z < −0.6 µm, photo generated minority electrons flow to the back contact and recombine. The photo current is composed out of a field component, Jn,E , and a diffusion component, Jn,ζ . With aid of a back


E [eV]

4

35

EFn

3

EFp

J [A cm-2]

n [cm-3]

2 1 1013

109 + 10−1

10−1

10−1

n

n

Jn

Jn

10−5

+ 103 J [A cm-2]

n

1011

Jn Jp

Jp

Jp Jn,z

Jn,z

Jn,z

10−5 10−1 103

−1.5 −1.0 −0.5 (a)

Jc

Jn,E

Jn,E

0.0

0.5

−1.5 −1.0 −0.5 (b)

Figure 2.15 Solution curves for devices (a) without bandgap gradient, (b) with back surface gradient (Ec,back = 0.2 eV, dmin = 0.3 µm), and (c) with front surface gradient (Ev,front = 0.2 eV, dmin = 0.3 µm). AM1.5

z [µm]

Jn,E

0.0

0.5

−1.5 −1.0 −0.5 (c)

illumination under short circuit condition. Semiconductor parameters as listed in default Table 8.1. Different current contributions are calculated according to the terms in Eqs. (2.9) and (2.10).

surface gradient (Figure 2.15b) the position of sign reversal of Jn shifts almost up to the back contact. Thus more minority carriers are collected at the junction. The driving force for the extended current collection is not an electric field as the gradient of the electric potential is zero outside of the SCR. Instead, it is the gradient of the chemical potential and, to be more specific, of the concentration independent part ζn,0 = −χ. This leads to the current contribution Jn,χ . Note also that the diffusion component is changed. Diffusion toward the back surface sets in at larger z. Thus, the diffusion component partly diminishes the positive effect of the bandgap gradient. This is the reason why the additional field of a back surface gradient is not simply given by Ec /q(wa − dmin ) but must be calculated from the gradient of EFn . The increase in bandgap toward the back surface of the absorber can also be realized by a slope in the valence band edge (Ev,back < 0) as sketched in Figure 2.14b. If the p-type doping concentration of the absorber is constant in depth, the gain in solar cell parameters is identical to the case of Ec,front > 0. In that case the graded region of the absorber contains a constant electric field.

0.0

0.5

36


The benefit of a bandgap gradient at the back surface of the absorber in terms of improved cell performance depends on Eg,back and dmin . Rules for the design of a back surface gradient are presented in Section 3.11. Bandgap widening at the front surface aims at reducing the diode current in the graded region of the absorber while maintaining the photo current in the nongraded region. Figure 2.14c sketches bandgap widening at the front surface by an increased conduction band edge and (Figure 2.14d) by a decreased valence band edge. A numerical example for Ev,front = −0.2 eV(dmin = 0.3 µm) is given in Figure 2.15c. For devices with a dominant interface recombination or interface-near recombination, the introduction of a front surface gradient can shift the recombination from positions close to the interface to positions within the bulk. We will see in Section 3.10 that in this case the front surface gradient can improve the cell. For a cell without interface recombination, the benefit of a front surface gradient will be small. The example in Figure 2.15c shows that the bandgap gradient leads to a positive gradient of EFn below z = −0.77 µm (the sign reversal of Jn is at slightly smaller z than in Figure 2.15a). This positive slope of dEFn /dz > 0 reduces the photo current and fill factor of the device, in particular for small lifetimes. It also occurs for a surface bandgap gradient realized by a surface gradient of the conduction band. Finally, Figure 2.14e,f sketches the cases of double bandgap gradients. In certain cases, the benefit for the cell performance may be larger than with front or back surface gradients alone.

2.4 Diode Currents

In this section, we discuss different recombination processes which may dominate the diode current of chalcogenide solar cells. We want to answer the questions: 1) What are the parameters J00 , Ea , and A for each recombination processes. 2) How does the cell behave if parallel processes are active. 3) What is the influence of barriers and non-homogeneities. Although recombination processes determine the diode current of the cell, we speak of diode current and not of recombination current. The reason is that recombination also reduces the photo current. We discriminate the diode current Jdiode from the photo current Jph which is treated in Section 2.4.9. The motivation behind this discrimination is to apply the superposition principle wherever possible. Other expressions for the diode current are bucking current or reverse current. 2.4.1 Superposition Principle and Shifting Approximation

If the solar cell is voltage biased in the dark, a pure diode current is flowing. This diode current can easily be determined but it may be different from the

2.4 Diode Currents

diode current if the cell is under illumination. Reasons for this difference, among others, may be an illumination dependent carrier lifetime, a large series resistance changing the boundary condition for QNR recombination [31], a change from low to high injection condition under illumination. In order to understand the limitations of the solar cell we are mainly interested in the diode current under illumination. Knowing this current allows calculating the Voc of the device. In practice, one often analyzes both cases, with and without illumination. The superposition principle states that the total current can be written as a sum of diode and photo currents. Also the photo current may be voltage dependent. Jlight (V) = Jdiode (V) + Jph (V)

(2.41)

It is justified to apply the superposition principle as long as we make sure that the diode current is the one flowing under illumination. Experimentally, the diode current under illumination shall be obtained by subtracting the (voltage dependent) photo current Jph (V) from the light current, Jlight (V). The problem of evaluation of the light JV curve is discussed in Section 2.6.2. The superposition principle shall be discriminated from the shifting approximation [31] which reads Jlight (V) = Jdark (V) + Jsc

(2.42)

This shifting approximation is rarely valid in chalcogenide solar cells. This is unfortunate as the validity of the shifting approximation would allow to measure two simple items to describe the complete device: the dark curve Jdark (V) and the short circuit current Jsc . 2.4.2 Regions of Recombination

The diode current of a heterostructure can be based on different recombination paths which are depicted in Figure 2.16. Recombination may take place in the SCR (path 4 in Figure 2.16), in the QNR (path 2) and at the back contact (path 3). Specific for heterostructure solar cells is the recombination path (1) at the absorber surface or, in other words, at the interface to the emitter. At this interface, recombination active defects will be present due to lattice mismatch or segregation of impurities. The possibility of recombination path (1) clearly is a disadvantage of a heterostructure solar cell against a homojunction. But there is also an advantage: Due its large bandgap recombination in the window and in the buffer layer (path 5) can be neglected for the calculation of the diode current. This is why in the following we deal with recombination in the absorber or at its surfaces. The diode current density comprises the hole current and electron current according to Jdiode = Jn (z) + Jp (z)

(2.43)

where Jn and Jp are given by Eqs. (2.9) and (2.10). Jn and Jp are dependent on z but Jdiode is independent of z. Since the recombination rate is highest in the absorber

37

38


Ec

(a) 2

4

3

φbn

EF Ev

Back contact

1

φbp

(b) 5 4* 1*

Ep,a

z=0

z=0 Figure 2.16 (a) Schematic presentation of the most critical recombination paths in a heterostructure solar cell. Recombination (1) at the absorber surface, (2) in the absorber bulk, (3) at the absorber back contact, and (4) in the absorber space charge region. Also, a recombination path in the buffer

layer is indicated (5). (b) Sketch of tunneling enhanced recombination (1∗ ) for interface and (4∗ ) SCR recombination. By tunneling, the density of holes which contribute to a particular recombination path may be increased.

layer and since the absorber layer is a p-type semiconductor, we can identify Jdiode with the electron (minority carrier) current density at the absorber/emitter interface, Jnz=0 . As an example, we refer to Figure 2.46a where Jdiode , Jn (z), and Jp (z) are given for the default device under forward bias. Thus, the following discussion concentrates on the electron current density. The starting point for the calculation of Jn (z) is the continuity equation, Eq. (2.13). If we integrate over the SCR of the absorber which extends from −wa to 0, we obtain 0 0 dJn (z) = q (Un (z) − G0 (z))dz + Jdiode,lF (2.44) −wa

−wa

Here, the thermal generation rate Gn = G0 has been introduced. The integral, although including the position z = 0, does not account for the interface absorber/emitter. Interface recombination is treated separately and accounted for by Jdiode,IF . The current density related to recombination in the QNR is given by Jnz=−wa . Here, in the field free region, we can make use of Eqs. (2.9) and (2.11) in order to write the well known expression for the diffusion current dn(z) (2.45) Jnz=−wa = qDn dz z=−wa As recombination in the QNR and at the back contact requires transport of carriers to the recombination sites by diffusion, one may also speak of diffusion limited current. However, diffusion is governed by the carrier density gradient which in turn depends on the recombination rates in the field free regions.

2.4 Diode Currents

Finally, the total diode current is given by

Jdiode = Jnz=0 = qDn

dn(z) +q dz z=−wa

0 −wa

(Un (z) − G0 (z))dz + Jdiode,lF (2.46)

The right hand side of Eq. (2.46) shows terms due to recombination in the QNR, in the SCR, and at the interface which we denote as Jdiode,QNR, Jdiode,SCR , and Jdiode,IF , respectively. It turns out that the general mathematical form of any contribution to the diode current reads qV Jdiode = J0 exp −1 with AkT −Ea J0 = J00 exp AkT

(2.47) (2.48)

J0 is the saturation current density which is activated by the activation energy Ea . Ea describes the temperature dependence of the saturation current density. J00 may be called reference current density which is only weakly temperature dependent. The diode quality factor A moderates the voltage dependence of the current density. Several recombination mechanisms may be operative in parallel, each carrying a certain share of the current. The magnitude or, in other words, the contribution of a particular process depends on its parameters J0 and A. If, for instance, a recombination process has a low J0 and a low A, it may not carry much current at low voltage. At higher voltage, due to its strong voltage dependence (small A), this process may dominate the diode. Parallel recombination processes are discussed in Section 2.4.6. Experimentally, the diode quality factor can be determined from the voltage dependence of the diode current. However, the exponential character of the JV relationship in Eq. (2.47) originates from the recombination rate and its voltage dependence. For all recombination processes considered here, the exponential voltage dependence finally results from, or can be expressed by, the voltage dependence of the carrier densities. We make use of this observation in order to derive the diode quality factors for different diode currents from the voltage dependence of the carrier densities. Recombination can be classified with respect to the three particles which can take on the recombination energy. These particles are photons (radiative recombination), electrons (Auger recombination), and phonons (phonon recombination). Another classification scheme discriminates between band to band and defect related recombination. Which of these classifications is useful for recombination in chalcogenide devices? Recombination by the emission of phonons requires defects in the bandgap to be present. It turns out that current chalcogenide materials contain enough defects to dominate the diode current. This is why we treat defect related recombination separately from radiative and Auger recombination.

39

40


2.4.3 Radiative Recombination

The inverse process of optical absorption is radiative recombination. Here, we mean direct recombination of electrons and holes. It is this process which cannot be avoided and which defines the Shockley–Queisser limit of an ideal pn junction [32]. The net recombination rate under non-equilibrium conditions is given by R = Un − G0 = B(np − n2i )

(2.49)

where B is the radiative recombination constant. In equilibrium, according to the principle of detailed balance, the equilibrium generation rate G0 equals the equilibrium recombination rate U0 . Consider radiative recombination in the SCR of an asymmetric absorber/window heterostructure with ND,w NA,a . We want to use Eq. (2.46) in order to derive the JV characteristic. First, we assume that each quasi Fermi level is flat from the injecting contact through the majority carrier regions and throughout the SCR.2) Only in the minority carrier regions, the quasi Fermi levels shall be allowed to vary. (As an example, see the simulated quasi Fermi levels in Figure 2.46a.) In the complete SCR the splitting of the quasi Fermi levels equals the applied bias voltage and it holds

np =

n2i

EFn − EFp exp kT

=

n2i

qV exp kT

(2.50)

With Eqs. (2.49) and (2.50), the recombination rate in the SCR, RSCR , is known and we can apply Eq. (2.44) (without the term Jdiode,IF ) to calculate the diode current density qV R Bn2i dz = q Jdiode,SCR = q − 1 dz exp 2 2 kT wa nr wa nr,a −Eg,a B qV = qwa (V) 2 Nc,a Nv,a exp exp −1 (2.51) nr,a kT kT The relative refraction index nr,a accounts for the fact that, due to total reflection, not all photons leave the cell [33]. A comparison of Eq. (2.51) with Eq. (2.47) shows that the diode quality factor for radiative recombination in the SCR is A = 1. Furthermore, it is Ea = Eg and J0 = qwa (V)Bn−2 r,a Nc,a Nv,a exp{−Eg,a /kT}. The width of the absorber SCR is given in Eq. (2.28). The weak voltage dependence of wa can safely be neglected against the exponential voltage dependence. From Eq. (2.51) we can learn that the recombination rate is constant in the SCR. Accordingly, the diode current increases with increasing wa . 2) The assumption of flat majority carrier

quasi Fermi levels up to the point of maximum recombination at first appears to be in contradiction to Eqs. (2.7) and (2.8). Without a gradient of the quasi Fermi levels

there would be no current flow. However, since we speak of majority carriers, which normally exhibit high densities, very small gradients of EFn and EFp can drive high current densities.

2.4 Diode Currents

Since, however, wa normally is small compared to the diffusion length in the limit of radiative recombination, Jdiode,SCR is very small and does not limit the diode current of the solar cell. Concerning our aim to derive a general routine for the determination of the diode quality factors for different recombination mechanisms, we conclude that the integration step in Eq. (2.51) does not add a voltage dependence stronger than the exponential one which originates from the voltage dependence of the carrier densities. This forms the motivation to generalize Eq. (2.50) as Eg qV qV 2/A 1/A 1/A = (Nc Nv ) exp − exp (np) = ni exp AkT AkT AkT (2.52) where we use the diode quality factor in order to describe a possible exponent of the product of the carrier densities. If the recombination rate is proportional to the product np, it holds A = 1 as in Eq. (2.51). If for other recombination processes the recombination rate depends on (np)κ , we can directly read out the diode quality factor from κ = 1/A. In order to read out the voltage characteristic of the recombination rate, it is often convenient to consider the case of qV kT. In that case the product np is much larger than n2i , which arises from the generation rate, the term n2i in Eq. (2.49) can be neglected. For radiative recombination in the QNR, we have to calculate the first term on the right hand side of Eq. (2.46). Thus, we have to find the carrier density and its derivative as a function of voltage. In order to do so, we want to anticipate Section 2.5.4 where we introduce the collection function ηc . Between the collection function and the carrier density there is a reciprocity relationship expressed in Eq. (2.148). Using this relationship, we can write the carrier density at the position z (−da < z < −wa ) as n(z) − n0 = ηc (z)(n(−wa ) − n0 )

(2.53)

n0 is the dark electron concentration which is the minority carrier concentration. n0 shall not be position dependent. The carrier density under applied bias at the edge of the SCR of a p-type semiconductor can be obtained from Eq. (2.50) which is divided by p. This gives n2 qV qV = n0 exp (2.54) n(−wa ) = i exp NA,a kT kT with the equality p = NA,a and n0 = n2i /NA,a . Equation (2.54) again assumes that the quasi Fermi levels are flat throughout the SCR. This is denoted as the ‘quasi equilibrium assumption.’ The assumption is valid for not too large recombination in the SCR and at the interface. One can argue that if IF or SCR recombination is so strong that the quasi Fermi levels are not flat, then recombination in the QNR will not be the dominant recombination path. Only when the splitting of the quasi Fermi levels EFn − EFp at the position −wa equals the applied voltage, recombination in the QNR can be a relevant path and contribute to the diode

41

42


current. Assuming the validity of Eqs (2.53) and (2.54) then reads qV − 1 ηc (z) + n0 n(z) = n0 exp kT with the derivative dη(z) qV dn(z) = n − 1 exp 0 dz z=−wa kT dz z=−wa

(2.55)

(2.56)

This expression introduced in Eq. (2.45) gives the diode current for radiative recombination in the QNR as well as for any other type of recombination in the QNR n2i qV dη(z) Jdiode,QNR = qdn (2.57) exp −1 NA,a kT dz z=−wa In order to specify for the radiative recombination limit, we use the expression for the electron diffusion length L2n = Dn τrad and replace Dn in Eq. (2.57). Next, let us assume that the absorber thickness is much larger than the electron diffusion length. Then, the quasi Fermi levels fall off linearly in the QNR and the collection function is given by z + wa ηc (z) = exp (2.58) Ln An example for a collection function in the QNR can be seen in Figure 2.36. The a is the solution of a linear differential equation for η such as Eq. term exp z+w c Ln dη(z) z+wa 3) = 0. exp Ln indicates that the electron (2.150) with the condition dz z=−da quasi Fermi level and thus the splitting of the quasi Fermi levels decays linearly from the point of carrier injection which is the end of the SCR, −wa . With the dη(z) = L−1 derivation n , the diode current for radiative recombination in the dz z=−wa QNR becomes qLn n2i qV −1 (2.59) exp Jdiode,QNR = τrad NA,a kT We are interested in the diode quality factor and see by comparison with Eq. (2.47) that for radiative recombination in the QNR also it is A = 1. Obviously, the variation of the quasi Fermi level splitting and thus the collection function does not add a further voltage dependence. This also holds for more realistic conditions, for instance, if the diffusion length approaches the film thickness. Then the derivative of the collection function is given by Eq. (2.153). Does radiative recombination limit the performance of chalcogenide solar cells? In order to answer this question, we need to know the radiative lifetime τrad . For a 3) The fact that we can solve a linear

differential equation for minority carriers injected into the QNR is an outcome of linearity of the recombination rate. This is not trivial and not valid for all

conditions of non-equilibrium. In the case of low injection it holds p ≈ NA,a , and the recombination rate is given by R = BNA,a (n − n0 ). Thus, the rate is linear in n.

2.4 Diode Currents

p-type absorber, the constant B is connected with the lifetime by 1/τrad = BNA,a . It further holds [29] U0 =

2πn2r c2 h3

∞ Eg

α(E)E2 dE E − 1) (exp kT

(2.60)

where nr is the refractive index. With B = U0 /n2i the radiative lifetime τrad = n2i /U0 NA can be calculated if the absorption coefficient is known as a function of photon energy. Using the absorption coefficients as plotted in Figures 4.6 and 4.20, the carrier lifetimes in the limit of radiative recombination were calculated after Eq. (2.60).4) The values of τrad are given in Tables 4.1 and 4.8 (see also Ref. [33]). It can be seen that the listed chalcogenide thin films exhibit radiative lifetimes in the range of 10−6 s. As the measured lifetimes of current chalcogenide devices at room temperature (RT) are 10−6 s for NA < 1017 cm−3 ) it becomes evident that, for state of the art chalcogenide thin film absorbers, defect recombination is the limiting process. Only to a minor extend radiative recombination may contribute to the total recombination rate. Auger recombination only dominates at high doping levels (see Section 3.5). Defect related recombination can take place at different sites within the heterostructure (see Figure 2.16): In the absorber bulk (2), in the absorber SCR (4), and at the surfaces of the absorber (1) and (3). Defect related recombination also takes place in the buffer (5) and window layer, but we show below that there it adds little to the diode current. Processes (1), (3), and (4), which include transport of carriers over energy barriers, can be enhanced by tunneling. We may further discriminate between single defect states with well defined energy levels in the bandgap, multiple defects, and defect distributions with a quasi continuous state density in the bandgap. Defect related recombination via a single defect state of density Nd has been analyzed by Shockley and Read [43] and Hall [44]. We refer to this type of recombination as Shockley, Read, Hall (SRH) recombination. Figure 2.17 shows the principal transitions involved in SRH recombination: the capture of an electron (I) and hole (II) and the re-emission of an electron (III) and hole (IV). If we write the density of occupied defect states as nd , the capture and emission rates can be expressed as (I)

n/τc,n = n(Nd −nd )σn vn ,

(II)

p/τc,p = pnd σp vp ,

(III)

nd /τe,n = nd σn vn NC exp{−(Eg −Ed )/kT},

(IV)

(Nd −nd )/τe,p = (Nd −nd )σp vp NV exp{(−Ed )/kT}

p-type Ec

n-type

Ec EFn

En

En

I

Demarkation levels

EFn Ed EFp Ec

(2.65)

Ep

(a) Figure 2.17 Semiconductors (a) p-type and (b) n-type band edges in non-equilibrium with a defect level Ed . The demarkation energies Ep and En are indicated by arrows. Defects within the shaded energy regions are efficient recombination centers. Defects

III EFp Ed

Ep

II

IV Ec

(b) above are electron traps and below are hole traps. Process I is the capture of an electron by the defect level Ed and III the re-emission of an electron. Process II is the capture of a hole and IV the re-emission of the hole.

45

46


where σn,p are the capture cross sections and τc , τe are the capture and emission time constants, respectively. We discriminate recombination states and trap states as two possible characters of the defect states. The character can change with the bias condition. Criteria for the discrimination between recombination and trap states can be formulated by aid of so-called demarcation levels En and Ep . These demarcation levels are derived from the following consideration: they give the energy at which a defect must be energetically situated in order to provide the equal probability for a carrier to be re-emitted or to be annihilated by the opposite carrier type. In other words, for a defect at the demarcation level, process III in Figure 2.17b has the same rate as process II (p-type semiconductor) and process IV has the same rate as process I (n-type semiconductor). This brings up the equations for En and Ep [38] σp m∗p Ep = En + kT ln and σn m∗n σn m∗n En = Ep + kT ln (2.66) σp m∗p The character of a defect state Nd at energy Ed in the bandgap can be derived from these conditions: • Ed < Ep : hole trap (HT) which communicates with the valence band. • Ed > Eg − En : electron trap (ET) which communicates with the conduction band. We want to visualize these conditions by aid of Figure 2.17. Consider a p-type semiconductor where holes are majority carriers even under non-equilibrium conditions. What is the energetic range where we find efficient RCs? According to Figure 2.17a, efficient RCs lie between Ep and Eg − En . Centers which lie above Eg − En act as ETs, which means that it is more likely that an electron is re-emitted to the conduction band than that a hole completes the recombination process and annihilates the electron. If Ep is smaller, then even more holes exist to complete the recombination process and accordingly En is smaller. This makes it plausible that for the calculation of the electron demarcation level, we employ the hole quasi Fermi level and vice versa – a fact which, at first glance, may be surprising. Defects which lie below Ep act as HTs. Thus in a p-type semiconductor, those defects are efficient RCs which lie between Ep and Eg − En . The analog considerations hold for an n-type semiconductor in Figure 2.17b. In Figure 2.17 we have shaded the region of efficient recombination. Due to the correction terms in Eq. (2.66), the demarcation levels are not simply mirrored at the intrinsic Fermi level position or, in other words, En does not equal Ep and Ep does not equal En . From the rates in Eq. (2.65), the fundamental expression for the net recombination rate R can be derived [45]: R = U − G0 =

np − ni 2

γp (n + n∗ ) + γn p + p∗

(2.67)

2.4 Diode Currents

Equation (2.67) is valid for electrons and holes. It describes the complete recombination process. The quantities n∗ = NC exp{−(Eg − Ed )/kT} and p∗ = NV exp{(−Ed )/kT} are auxiliary densities of carriers which would be realized if the Fermi energy is located at the defect level Ed . These auxiliary densities stand for carrier emission from the trap states. The quantities γp and γn in Eq. (2.67) at the moment are placeholders. In the case of bulk recombination without tunneling, γp and γn equal τn0 and τp0 , the minimum lifetimes of electrons and holes. γn,p have other meanings in the case of tunneling enhanced recombination and interface recombination. The minimum lifetime τn0 (τp0 ) can be measured if electrons (holes) are injected into a highly p-type (n-type) semiconductor with the same defect density as the semiconductor under consideration. The concentrations n and p in Eq. (2.67) represent the non-equilibrium carrier concentrations which may be realized by voltage bias or light bias. In equilibrium, it is np = ni 2 . Then, the thermal generation rate equals the recombination rate and, accordingly, the net recombination rate becomes zero. The net recombination rate sometimes is simply referred to as ‘recombination rate’ which can lead to misunderstandings. The energetic position of a defect center within the bandgap of the semiconductor can be from close to the valence band up to close to the conduction band. Above, we explain that a RC is efficient if it communicates with both bands, in other words, if Ed lies between the majority carrier quasi Fermi level and the minority carrier demarcation level. Let us investigate if this condition for efficient recombination is also reflected in Eq. (2.67). The equation indeed predicts that the net recombination rate becomes small when n∗ or p∗ are large, that is, if the defect level is close to one of the band edges. In contrast, the recombination rate can become large if n∗ and p∗ are small, that is, if the recombination level is close to midgap. In Figure 2.17, we have shaded the regions of efficient recombination. In the next step, it may be asked at what position within the device the recombination rate has its maximum. We use the relationship between n and p under voltage bias, np = ni 2 exp{qV/kT}. Tunneling may be excluded, thus it is γn,p = τn0,p0 . Insertion into Eq. (2.67) yields − 1 n2i exp qV kT R = U − G0 = (2.68) n2i ∗ τp0 n + τp0 n∗ + τn0 n exp qV p + τ n0 kT The first notion is that the net recombination rate depends on the intrinsic carrier concentration n2i = Nv Nc exp{−Eg /kT}. Thus, in a heterostructure solar cell the layer with the smallest bandgap will generate the main part of the diode current. Normally, this is the absorber layer. Only to a minor part, the diode current results from recombination in the window or buffer layer. We can turn this argument around in order to point out that regions in the solar cell with a bandgap smaller than the absorber layer carry the highest recombination traffic. If, for instance, the absorber layer in the solar cell is highly defective, bandgap fluctuations may reduce the effective bandgap of the absorber. This increases the diode current (see Section 2.4.9).

47


Ed1-Ed2

−1

−1

τn 0 n = τp−01 p

τn 0 n = τp−01 p EFn Ec

Energy

48

Ev (a)

Position z

(b)

EFp Position z

Figure 2.18 Energy region of efficient recombination centers within the bandgap of a heterostructure at two different values (a, b) of forward voltage bias.

In the bulk of the absorber, maximum recombination can occur either in the SCR or in the QNR. In order to determine the maximum of recombination we form the derivation of R with respect to n qV 2 −n exp − 1 i kT n2i qV dR = τ − τ exp 2 p n 2 dn n kT n2i ∗ τp0,a n + τp0,a n∗ + τn0,a n exp qV kT + τn0,a p (2.69) dR/dn vanishes if the bracket on the right hand side of Eq. (2.69) becomes zero, that is, for τn0,a −1 n = τp0,a −1 p. (We see this if we replace n2i exp{qV/kT} by the product np.) Thus, where the condition τn0,a −1 n = τp0,a −1 p

(2.70)

is fulfilled, the maximum of recombination takes place. Certainly, Eq. (2.70) only in that case marks the position of maximum recombination if τn0,a and τp0,a are approximately constant within the absorber layer.5) If we assume equal lifetimes for electrons and holes, τn0,a ≈ τp0,a , the location of maximum recombination is at the position of n = p. For a reasonably doped absorber, the position of τn0,a −1 n = τp0,a −1 p is within the SCR. Thus, the position of maximum recombination normally is within the SCR. Below we show, however, that the position of maximum recombination does not necessarily dominate the diode current of the device. Although having a small maximum in the SCR, substantial recombination can also take place in the QNR. With the knowledge that the maximum recombination takes place at τp0,a n = τn0,a p, we may once more consider the region of effective RCs within the bandgap. Figure 2.18 schematically exemplifies this region for two different voltage biases together with the position of τn0,a −1 n = τp0 −1 p. (Note that the position of τn0 −1 n = τp0 −1 p changes with voltage bias.) The upper limit of the region of effective RCs 5) Particularly for polycrystalline thin film

absorbers, the lifetime may be depth dependent. Since the grain size increases

with film thickness, regions, which have been deposited in a later state of the growth process, may exhibit lower defect densities.

2.4 Diode Currents

is the demarcation level of electrons where it is p > n and the electron quasi Fermi level where it is n > p. The lower limit is the hole quasi Fermi level (p > n) and the demarcation level of holes (n > p). For small forward bias, the region of efficient RCs at the position of τn0 −1 n = τp0 −1 p is very narrow. A defect level Ed1 generates only a small diode current. At higher forward bias, however, the region of efficient recombination at n = p is much wider. Both defects Ed1 and Ed2 bring approximately the same current. If there would be a quasi continuous defect density in the bandgap, the density of states which contribute to recombination change with voltage bias. This influences the shape of the JV curve in the dark and under illumination. We can summarize that if the RC is not within the region of efficient recombination at the position τn0 −1 n = τp0 −1 p, the recombination can still have a maximum at τn0 −1 n = τp0 −1 p, but this maximum is small and the total diode current is small. Finally, we come back to the question if recombination at τn0,a −1 n = τp0,a −1 p necessarily dominates the diode current of the device. A simple simulation of an absorber/window heterostructure can exemplify this question. In Figure 2.19, 4

Ec EFn

3 E [eV]

Ec

EFn

EFp

Ev

EFp

2 1 Ev

0

n

p

n

1 × 10−2

100

0.5

10−3

0

10−6

30 20

Jtot

Jtot

Jp

10 0 −1.5

(a)

p

1016 1014 1012 1010

JN −1.0

−0.5

JN −1.0

0.0

Position z [µm]

(b)

−0.5

Jp 0.0

Position z [µm]

Figure 2.19 Simulation of an absorber/window device with carrier lifetimes of (a) 10−8 s and (b) 10−11 s for electrons and holes in the absorber and window. The devices are under electric forward bias where the bias voltage corresponds to the Voc under illumination. Simulation was performed using SCAPS [925]. Other device parameters as in Table 8.1.

0.5

Eff. Rec. [cm−3 s−1]

J [mA cm−2]

Cum. Rec. [cm−2 s−1]

n,p [cm−3]

1018

49

50


recombination rates in the dark for devices with τn0,p0 = 10−8 s (a) and τn0,p0 = 10−11 s (b) are given as a function of position. The devices are under forward bias where the bias voltage is selected to match the Voc of the device under illumination. For both carrier lifetimes, recombination in the absorber is by far dominating over recombination in the window layer. For τn0,p0 = 10−11 s (Figure 2.19b), the recombination rate has a distinct maximum in the SCR of the absorber. However, it is very small in the QNR. The cumulative recombination rate confirms that, for such a short lifetime, the SCR is the main recombination zone. In there, the location of n ≈ p and its surroundings carry the main recombination traffic. The recombination rate drops off exponentially for n > p and p > n. In the case of τn0,p0 = 10−8 s (Figure 2.19a), the recombination rate is approximately constant in the QNR. The recombination maximum in the SCR is small. The cumulative recombination rate shows that more than half of the recombination takes place in the QNR. However, both recombination types are active in parallel. Thus, the predominance of QNR recombination and SCR recombination changes with carrier lifetime. The recombination regimes and their dominance can be differentiated by the voltage and temperature dependencies of the diode current. It is the diode quality factor A which expresses this voltage dependence. In the following, we deal separately with defect recombination in the SCR, in the QNR, and at the IF. 2.4.5.1 SCR Recombination As for all defect related recombination, the SRH formula Eq. (2.67) governs the recombination rate in the SCR. In the preceding section, we show that recombination in the SCR has a distinct maximum at the position τp0 n = τn0 p. We can make use of this equality in order to rewrite Eq. (2.52) as

n2/A = (τn0 Nv Nc /τp0 )1/A exp{−Eg /AkT}exp{qV/AkT} or p2/A = (τp0 Nv Nc /τn0 )1/A exp{−Eg /AkT}exp{qV/AkT}

(2.71) κ

κ

If now the recombination rate can be written as a function of n or p , the diode quality factor can be directly deduced from the exponent of the carrier concentrations, that is, from κ = 2/A. For instance, let the recombination rate depend on p or n where the exponent is κ = 1, then the diode quality factor is 2. We start our consideration of SCR recombination with the (historical) case where the absorber layer has a defect level approximately at midgap, Ed2 ≈ Ei , as depicted in Figure 2.20a. Such a defect level is within the energetic region of efficient recombination for all voltage biases. Close to the point of maximum recombination it holds n, p n∗ , p∗ . Thus re-emission of trapped carriers can be neglected and Eq. (2.67) can be simplified to R = np − n2i (τp0,a n + τn0,a p)−1

(2.72)

At sufficient forward bias of qV kT the term arising from equilibrium generation can be neglected. Directly at the point of maximum recombination τp0,a n = τn0,a p the recombination rate is n2i

Rm = p/2τp0,a

(2.73)

2.4 Diode Currents

Ed1 Ed3 Ed2

Ed1 Ed3 Ed2

EFn EFp

En

Ec

Nd0 Ed1

EFn

Ed3 Ed2

Ev

EFp

Nd2 Nd1

0

z=0

Ev

z=0

(a)

(b)

(c)

(d)

Figure 2.20 Defect related recombination via defect states Ed1 and Ed2 (a) in the SCR, (b) in the QNR, and (c) at the IF of an absorber/window heterojunction. (d) Quasi continuous defect density with maximum at Ec .

Comparing the exponential factor of the hole density with the one in Eq. (2.71) we see that A = 2. If SCR recombination is by far the dominant process, the diode quality factor for the particular device is 2. Next, we will show once more that calculating the diode current does not add a voltage dependence stronger than the exponential one of Eq. (2.73). Equation (2.72) gives the maximum of the recombination rate at the position −zm in the SCR. In the vicinity of −zm , the electron and hole concentrations vary exponentially with distance to −zm , that is n(z ) = nm exp{−qFm z /kT} and p(z ) = pm exp{qFm z /kT}

(2.74)

where z = −(z + zm ). The exponential variation of n and p can be seen in Figure 2.19. Fm is the electric field at the position of maximum recombination. Thus, the recombination rate may be written as R(z ) =

np qF z qFm z τp0,a nm exp − m + τn0,a pm exp kT kT

(2.75)

Here, we assume that the product np is independent of z since, as stated in Section 2.4.5, the product is approximately constant up to (and also sufficiently beyond) the point of maximum recombination: This is equivalent to the assumption that the quasi Fermi levels are flat within the SCR. The current density is Jdiode,SCR = q

wa −zm

R(z )dz

(2.76)

−zm

With the substitution u =

τ

n0,a pm τp0,a nm

1/2

−1/2 R(u) = np τn0,a pm τp0,a nm

qFm z exp kT

1 u−1 + u

51

it results

(2.77)

Ec

52


Integration of R(u) from u(−zm ) to u(wa − zm ) or, which is a good approximation, from 0 to ∞ yields the diode current ∞ ∞ kT R(u) np 1 kT Jdiode,SCR = du = −q du (2.78) 2 Fm u Fm τn0,a pm τp0,a nm 0 u + 1 0 The integral is listed as π/2 and with nm = n, pm = p, and τp0,a n = τn0,a p we obtain

Jdiode,SCR =

π/ kT Nc,a Nv,a 1/2 Eg qV 2 exp exp − Fm τn0,a τp0,a 2kT 2kT

(2.79)

We notice in Eq. (2.79) that indeed the diode quality factor is A = 2. Thus, the main voltage dependence of Jdiode,SCR arises from the recombination rate Rm in Eq. (2.73). The term πkT/qFm can be interpreted as an effective width of the recombination zone. Often, Fm is approximated by the maximum field which, in the case of an asymmetric heterojunction, reads Fm ≈ Fz=0 = 2(Vbi −V)/wa

(2.80)

With increasing forward bias, Fm decreases and, accordingly, 1/Fm increases. This additional voltage dependence is the reason why for experimental solar cells which are limited by SCR recombination, the diode quality factor is slightly smaller than 2. A compilation of diode quality factors for different recombination processes is given in Table 2.1. In the following, we derive most of them in a similar fashion as above. A diode quality factor of A ≈ 2 is diagnosed when the current density in a semi-log plot varies with qV/2kT. The origin is that the recombination rate varies with exp{qV/2kT}. But the physical reason for A = 2 may be rationalized as follows: the quasi Fermi levels at the position τp0,a n = τn0,a p vary symmetrically with respect to the defect level. Since the splitting of the quasi Fermi levels is given by qV, each Fermi level varies as qV/2. Whenever a diode quality factor A = 2 (or close to 2) Table 2.1 List of diode quality factors 1/A for different recombination processes in the space charge region (SCR), the quasi neutral region (QNR), and at the interface (IF) of an absorber/window heterojunction. For an absorber/buffer/window heterostructure see Eq. (2.113).

Recombination region

Defect

Without tunneling

With tunneling

SCR

Ec,a − Ed ≈ Eg,a /2 Nd (E) Ec,a − Ed ≈ Eg,a /2 Nd (E) Ec,a − Ed ≈ Eg,i /2 Nd (E)

1/2 1/2 1 + TT∗ 1 1 1−θ 1 − θ + θ TT∗

1/2 [1 − ] 1/2 1 + TT∗ − – – (1 − θ) (1 − ) (1 − θ) (1 − ) + θ TT∗

QNR IF

2.4 Diode Currents

is diagnosed for a solar cell device, we may consider the quasi Fermi levels of electrons and holes moving symmetrically with respect to the dominating defect level under applied voltage bias. In the case that there are more than one deep defect levels in the bandgap, the recombination rates are added up as well as their contributions to the current densities. Since all current components due to deep defect levels exhibit the same voltage dependence, the A factor is still approximately 2. Thus, the density of defect states located within the zone of efficient recombination can be added up in order i to write one rate equation for the recombination. It is τn0,p0,a −1 = νn,p i σn,p Nid for i defect levels. In the SCR, a defect level may also be located close to one of the band edges, as depicted by the defect Ed1 in Figure 2.20a. Then at the position τp0,a n = τn0,a p it holds n∗ n, p, p∗ and Eq. (2.67) becomes R = np/τp0,a n∗

(2.81)

By application of Eq. (2.52) it becomes apparent that A = 1. Since the product np is constant in the SCR, the recombination rate is also constant. Integration of Eq. (2.81) over the SCR is equivalent to multiplication with wa . Figure 2.21 shows the simulated JV curve for a flat defect level Ed1 . In agreement with Eq. (2.81), the diode quality factor for the flat defect Ed1 is close to unity. A diode with a flat defect (A = 1) exhibits a stronger voltage dependence of the diode current as compared to a diode with a deep defect (A = 2). As pointed out in Section 2.4.5, however, the level Ed1 is not within the region of efficient recombination. Thus, the total recombination rate of a flat defect is small. Accordingly, recombination in an absorber having a combination of flat and deep defect levels is dominated by the deep defect. As long as the forward bias is not too large, the diode quality factor will be A = 2.

Current density [A cm−2]

103 100 10−3 10−6

A = 1.78 Ed2 Ed1 Ed3

A = 1.1

10−9 10−12 0.0

0.2

0.4 0.6 Voltage [V]

0.8

1.0

Figure 2.21 Simulated JV curves of an absorber/window heterojunction assuming recombination via single defect levels Ed1 , Ed2 , and Ed3 as shown in Figure 2.22a. The diode quality factors where obtained by curve fitting. After Ref. [46].

53

54


If the defect level is at an intermediate energy depth as depicted by Ed3 in Figure 2.20a, the diode factor will not lie between 1 and 2 as one could initially assume. Figure 2.21 shows that the JV curve for a diode with defect Ed3 continuously changes slope from 1.1 up to 1.78. The reason is that with increasing voltage bias, the region of efficient recombination widens and finally envelopes the intermediate defect Ed3 . Thus, a change in slope of a log J(V) curve can indicate a defect at intermediate depth in the energy gap. After discussing single defects and defect combinations, we may consider the case where there is a continuous defect density in the bandgap. This opens the possibility for numerous variations. The defect distribution may be (among other possibilities) constant within the bandgap, linearly decreasing from the band edges, or exponentially decreasing. We are particularly interested in such a defect distribution that can produce a voltage independent diode quality factor. Since the carrier concentrations at the point of maximum recombination vary exponentially with voltage bias, it is an exponential defect distribution which has a strong impact on the diode quality factor [46]. A defect density which is constant or a defect distribution which is linearly decreasing from the band edges will lead to a voltage dependence of the diode factor. Examples of exponential defect distributions are shown in Figure 2.20d. Without loss of generality, the defect distribution may have its maximum Nd0 at the conduction band edge and decay into the bandgap with the characteristic energy kT∗ . Such an exponential defect distribution can be written as dNd (E) = Nd0

exp

−(Ec −E) kT∗ ∗

kT

dE (2.82)

where Nd (E) is in units of cm−3 . Then, by neglecting the term γn p∗ = τn0,a p∗ in Eq. (2.67) the recombination rate becomes c −E N (Ec ) d exp − EkT np Ec R(E) ∗ dNd (E) = R= dE kT∗ Ev τp0,a n + τn0,a p + τp0,a Nc exp − EckT−E Nd (Ev ) Nd0 (2.83) Note that the absolute number of RCs is factored in the values of τn0,a and Ec τp0,a . It holds τ−1 n0,p0,a = vn,p Ev σn,p (E)Nd (E)dE. We make use of the substitution ν = γp Nc (γp n + γn p)−1 exp{−(Ec − E)/kT} and obtain [47] ∗−1

T/T∗ −1

T/T∗ ν(Ec ) νT/T T R = np ∗ τp0,a n + τn0,a p τp0 Nc dν (2.84) T 1+ν ν(Ev ) It can be shown that also for an exponential defect distribution, the maximum recombination is located at τp0,a n = τn0,a p. The parameter 1/T∗ denotes the damping of the exponential defect distribution. A defect distribution with T∗ > T is the interesting one since it noticeable extends into the region of efficient recombination. An example is given as dNd2 /dE in Figure 2.20d. To the contrary, a defect distribution dNd1 /dE with T∗ < T may hardly exceed the demarcation levels. Such a fast decaying distribution does not

2.4 Diode Currents

induce efficient recombination – similar to a discrete level close to the band edges. We consider the case T∗ > T and find by application of Eq. (2.70) R=

T 1 T∗ 2τp0,a

T/T∗

2 Nv,a

pT/T

∗+1

∗

πT/T

sin πT/T∗

(2.85)

By use of Eq. (2.71) the diode quality factor can be derived from the exponent of p which is κ = 1 + T/T∗ . Thus, 1/A = 1/2(1 + T/T∗ ). This is an important result as it delivers diode factors between 1 and 2. It is A = 2 for T∗ T and A → 1 for T∗ → T. A separate evaluation of Eq. (2.84) shows that A ≈ 1 for T∗ ≤ T. The calculation of the diode current density is similar as for a discrete level where again we make use of an exponential decay of the carrier densities around the point of maximum recombination. The result is [47]

Jdiode,SCR

T/2T∗ 2 τp0,a T Nc,a Nv,a 1/A , 1 τn0,a τp0,a Nc,a Nv,a T∗ −Eg,a qV × exp exp AkT AkT

kT = Fm

(2.86)

∗

with 1/A = 1/2(1 + T/T ). The integral (ϑ, ξ) is defined by

(ϑ,ξ) =

πϑ sin(πϑ)

∞

0

(ϑ−1) u−1 + u du uξ

(2.87)

Formally, Eq. (2.86) can be written as Eq. (2.47) which allows to identify J00 =

kT Fm

Nc,a Nv,a τn0,a τp0,a

1/2

τp0,a Nc,a τn0,a Nv,a

T/2T∗

T/T∗ , 1

(2.88)

and J0 = J00 exp{−Eg /AkT}. The function (ϑ, ξ) must be evaluated numerically but has been plotted as a function of T/T∗ in Ref. [47] (curve E00 /kT∗ = 0). Equation (2.86) reveals that beside the exponential voltage dependence there is again only the linear voltage dependence of Fm which causes slightly reduced A factors. The activation energy of J0 is Eg . J00 exhibits only a weak temperature dependence. Figure 2.22 shows calculated values of J0 as a function of NA,a for an asymmetric absorber/window heterostructure (ND,w NA,a ). The defect distribution decays with T∗ /T = 3 at RT. With increasing NA,a , J0 becomes smaller. The mathematical reason is the increased electric field Fm , but the physical reason is the steeper drop of n and p close to the point of maximum recombination. This reduces the zone of highest recombination within the SCR upon an increase of NA,a . As a comparison, Figure 2.22 gives J0 for a single defect level at Ed = Ei . The defect volume density is identical to the defect distribution. A higher saturation current is an outcome of the fact that the deep defects all are within the region of efficient recombination. This is not the case for the defect distribution with a relatively low value of T∗ . However, the principal drop of J0 with increasing

55


Current Density J0 [mA cm−2]

56

10−2

τn0,p0 = 10−8 s with Tunneling

10−4 10−6 10−8 10−10 1013

1014

1015

1016

1017

1018

NA,a [cm−3] Figure 2.22 Calculated saturation current densities for an asymmetric heterostructure of the absorber/window type. Device parameters as in default Table 8.1. Full lines describe J0 as a function of NA,a for recombination via a deep defect without [Eq. (2.79)] and with tunneling enhancement [Eq. (2.47)] with 1/A from Table 2.1 and

J00 from Table 2.2). Dashed lines are for an exponential defect distribution without [Eq. (2.86)] and with tunneling enhancement [Eq. (2.47)] with 1/A from Table 2.1 and J00 from Table 2.2). The tunneling enhancement factor is F = 3.6 × 105 V cm−1 . The decay parameter of the exponential defect distribution is T∗ = 900 K.

NA,a is identical for both defect scenarios. The influence of NA,a on the solar cell parameters is discussed in Section 3.5. In the SCR, recombination may be enhanced by tunneling. The principal idea is depicted as path (4∗ ) of Figure 2.16: if a strong electric field is present, the density of carriers available for recombination at a certain location within the SCR is increased due to the finite probability of carriers for tunneling into the bandgap. At location z, the tunneling enhanced carrier densities are pt (z) = p(z)(1 + T(E)exp{E/kT}) and nt (z) = n(z)(1 + T(E)exp{E/kT}). T(E) is the tunneling probability and E is the energy of the barrier. Hurkx and coworkers [48] introduced the tunneling enhancement factors n,p in order to write nt = n(1 + n ) and pt = p(1 + p ). Since re-emission from the traps is also enhanced by tunneling, the auxiliary functions in Eq. (2.67) become n∗t (z) = n∗ (1 + n ) and p∗t (z) = p∗ (1 + p ). Now, the recombination rate at a certain location is governed by the tunneling enhanced densities nt , n∗t , pt , and p∗t . Thus, we need to write the SRH equation (Eq. (2.67)) in the enhanced densities. Since, however, the carrier transport equations [Eqs. (2.9) and (2.10)] use the local densities n, n∗ , p, p∗ it is necessary to rewrite R(z) in the local densities. It is simple mathematics to show that this introduces the expressions γn = τn0,a (1 + n )−1 and γp = τp0,a (1 + p )−1 in the SRH Eq. (2.67). Although recombination enhancement by tunneling is an outcome of an increased density of carriers contributing to the recombination traffic at a certain location, it can formally be viewed as due to locally reduced carrier lifetimes. For a deep defect level it is n = p = [48]. For not a too large electric field (Fm < 9 × 105 V cm−1 ), Hurkx et al. gave an analytical expression for [48]. Here, however, we use the form developed by Rau

2.4 Diode Currents

[49] for parabolic band bending √ |F| qVb (z) E200 (z) = 2 3π exp F kT 3(kT)2

(2.89)

1

with F = (24 m∗ (kT)3 ) 2 /q¯h as the critical electric field and |F| as the electric field at the position of maximum recombination. As often, |F| is approximated by Fm = (2qNA,a Vb (z)/εa )1/2 .Vb (z) is the local band bending and E00 = (q¯h/2)(NA,a /m∗ εa )1/2 is the characteristic tunneling energy. As an example, can reduce the carrier lifetimes of electrons and holes by a factor of >10 if the electric field exceeds 2 × 105 V cm−1 (see Figure 2.23). For an asymmetric heterojunction the establishment of such a high electric field requires that the doping density in the absorber exceeds a value of about NA = 2 × 1016 cm−3 . The use of Eq. (2.89) is significant since it can be implemented in device simulation (for instance, see program PC1D [50]). An important input parameter for tunneling enhanced recombination is the tunneling effective mass. Hurkx gave m∗t = 0.25 × mE . A lower limit may be m∗t = 0.1 × mE which has been used in Figure 2.23. The simulated influence of tunneling on the open circuit voltage of an absorber/window type solar cell is discussed in Section 3.5. The local band bending Vb (z) = Vbi (z) − V(z) can be expressed by the relation of carrier densities exp{−qVb (z)/kT} = p(z)/p0 where p0 is the hole concentration in the absorber bulk. Thus, by adapting the abbreviation = E00 2 /3(kT)2 first introduced by Rau et al. in Ref. [47] it is convenient to write √ |F| (zm ) = 2 3π F

p p0

105

(2.90)

1019

104

Γ

1017

102 101

1016 100 10−1

1015 2

4

6

8 ×105

Electric Field F [V cm−1]

NA,a [cm−3]

1018 103

Figure 2.23 Tunneling enhancement factor and absorber doping NA,a for an asymmetric heterojunction (ND,w NA,a ). The enhancement factors are according to Eq. (7) in Ref. [48] (dashed line) or according to Eq. (2.89) (dotted line). The effective tunneling mass was ∗ assumed to be mt = 0.1 me . Other device parameters are εa = 10 and Eg = 1.2 eV.

57

58


where zm is the position of maximum recombination. For a discrete trap at Ed = Ei we can directly make use of Eq. (2.73) and for qV kT it is Rm = p/2γp = p(1+)/2τp0,a If tunneling dominates, that is, 1, we may write √ 3π |F| p0 p Rm = τp0,a F p

(2.91)

(2.92)

The maximum recombination rate for tunneling enhancement is located at (1 + )τn0,a p = (1 − )τp0,a n. Thus, we can again use Eq. (2.71) in order to read out the diode quality factor from the exponent of the carrier density. The exponent of p is 2/A = 1 − and accordingly 1/A = (1 − )/2. Tunneling enhanced recombination obviously can explain diode quality factors larger than 2. In the case of vanishing tunneling energy, that is, → 0, the diode factor becomes A = 2 as for SCR recombination by a deep defect level without tunneling enhancement. The current density in the case of tunneling enhanced recombination via a deep defect level can be written in the standard form Eq. (2.47) with 1/A = (1 − )/2 and J00 as listed in Table 2.2. The function (T/T∗ , 1 + ) is given in Eq. (2.87) and has been plotted in Ref. [47] for different values of . Theoretically, there exists the limit where it is ≈ 1. There, the current density due to tunneling enhanced recombination has the same magnitude as the current density due to thermally activated recombination described by Eq. (2.78). If we denote the former as Jdiode and the latter as Jdiode , the current density in a semi-log plot has the slope dln(Jdiode + Jdiode )/dV = q/AkT with 1/A = (1 + /2)/2. The A factor is smaller than 2. As can be seen in Figure 2.23, it is ≈ 1 for NA,a ≈ 1016 cm−3 . Tunneling enhanced recombination can also occur via an exponential defect distribution. The determination of the diode quality factor follows the same route as for a discrete level. In fact, Eq. (2.85) is already the appropriate expression for the maximum recombination rate. In the limit 1 we can insert /τp0,a for 1/γp and ∗+1− with Eq. (2.90) we obtain the dependency R ∼ PT/T . Thus by comparison with Eq. (2.71), the reverse diode quality factor for tunneling enhanced recombination via an exponential defect distribution reads T 1 E200 1 + ∗ (2.93) = 1− A 2 T 3(kT)2 The compilation of the inverse diode quality factors in Table 2.1 reveals the additive character of the different components to 1/A. Note again, however, that Eq. (2.93) is only valid for > 1. In the limit E00 → 0 (i.E., < 1) and with T∗ > T, Eq. (2.93) describes the inverse A factor for SCR recombination via an exponential defect distribution without tunneling enhancement. In the limit T∗ → ∞, Eq. (2.93) stands for tunneling enhanced recombination via midgap states. Finally, in the limit T∗ → ∞ and E00 → 0 Eq. (2.93) reduces to the A factor of SRH recombination via a deep trap.

Table 2.2 Collection of reference current densities J00 for defect related recombination in different regions of an n-type window/p-type absorber heterojunction with εw ND,w > εa NA,a and Eg,w Eg,a . According to the text, the abbreviation = E00 2 /3(kT)2 has been used. For the function (ϑ, ξ) see Eq. (2.87).


Defect

Without tunneling

SCR

Ec,a − Ed ≈ Eg,a /2

kT Fm

Nc,a Nv,a τn0 τp0

Nc,a Nv,a τn0 τp0

With tunneling

1/A 1/A

π 2

τ2 p0 Nc,a Nv,a

Nd (E)

kT Fm

QNR

Ec,a − Ed ≈ Eg,a /2

D N N q n Nc,a v,a dη(z) dz z=−w A,a a

IF

Ec,a − Ed ≈ Eg,i /2

qSp0 NA,a

Nd IF(E)

kT F

T/2T∗

T ,1 T∗

√ 2 3π

Nc,w Nv,a 1/A ND,w NA,a 1/A N N qSp0 NA,a N c,w Nv,a D,w A,a T T*

c,min Nc,w

Nc,a Nv,a τn0 τp0

1/A

1/A √ p Nc,a Nv,a 2 3π τ 0 τ τ p0 n0 p0 T × *,1+

(0, 1 + ) T/2T* τ2 p0 Nc,a Nv,a

×

–

ND,w Nc,min

× T/T*

∗ ν(Ec ) νT/T −1 ν(Ev ) 1+ν dν

N √ Nc,w Nv,a 1/A qSp0 NA,a 2 3π FFm Nc,min ND,w NA,a c,w 1/A √ N N × qSp0 NA,a 2 3π FFm N c,w Nv,a D,w A,a N N T/T* ∗ ν(Ec ) νT/T −1 D,w × T* Nc,min ν(Ev ) N 1+ν dν c,w T

c,min

2.4 Diode Currents

×

T

p0 τp0

kT F

N

59

60


The reference current density in the case of tunneling enhanced recombination via an exponential defect distribution has been derived by Rau [49]. It is listed in Table 2.2. In Figure 2.22, the saturation current densities for tunneling enhanced recombination via a deep trap (full lines) and via an exponential defect distribution (dashed lines) is plotted as a function of NA,a . Note that for all curves in Figure 2.22 the volume density of defect centers is identical. For the deep level and defect distribution, J0 for tunneling enhanced recombination increases for NA,a > 1017 cm−3 . (Below NA,a ≈ 1017 cm−3 , in the case of tunneling J0 is constant due to the missing factor Fm .) At the position of NA,a ≈ 1016 cm−3 tunneling and thermally activated recombination induce about the same current density. Approximately at this level of absorber doping, the tunneling enhancement factor reaches = 1. Here, the total saturation current density formally has to be written from the sum of conventional and tunneling enhanced recombination. Figure 2.22 confirms that tunneling enhanced recombination may well limit the performance of certain chalcogenide solar cells. 2.4.5.2 QNR Recombination For a long carrier lifetime, the main part of the injected carriers pass the SCR without recombination and enter the QNR (see Figure 2.19a). There they are minority carriers and are subject to diffusion and recombination via defect states (process 2 in Figure 2.16). Again, we can discriminate between recombination via deep and flat defects as well as via an exponential defect distribution. The principal band diagram in the QNR of a p-type absorber under forward bias is shown in Figure 2.20b. The QNR extends up to the back contact. In principle, the diode current due to QNR recombination can be calculated from Eq. (2.57). But, in order to discriminate QNR recombination from back surface recombination we consider the special case of a p-type absorber where it holds da Ln . Then, the minority carriers do not reach the back surface of the absorber (the back contact) and back surface recombination can be neglected. Instead of solving Eq. (2.57), we can make use of the temperature dependence of carrier densities according to the formalism in Eq. (2.52). Then, we can read out 1/A immediately. The diode current due to recombination in the QNR can – similar as in the SCR – be calculated from the continuity equation for electrons Eq. (2.13) in the integral form

Jdiode,QNR (−wa ) = q

−wa

−da

(Rn − G0 )dz + J( − da )

(2.94)

In general, application of Eq. (2.94) is inconvenient since we have to calculate Jdiode,QNR (−da ) separately. However, in the special case of da Ln it holds Jdiode,QNR (−da ) ≈ 0. In addition, if we consider the case of large forward bias (qV kT) only, we can neglect the generation rate G0 in Eq. (2.94) and let U = Rn − G0 ≈ Rn .

2.4 Diode Currents

For the recombination rate we employ the SRH equation [Eq. (2.67); without generation term] U=

τp0,a (n +

n∗ )

np

+ τn0,a p + p∗

(2.95)

where τn0,a and τp0,a are the bulk lifetimes. A deep defect Ed2 or a flat defect Ed1 may be present in the absorber. The flat defect Ed1 shall be above the electron demarcation level at Eg − En . Thus, Ed1 is located beyond the region of efficient recombination. For the deep defect Ed2 it holds p p∗ , n, n∗ , and p ≈ NA,a . For the flat defect Ed1 it is n∗ p, p∗ , n. For the deep defect the recombination rate can be approximated by U=

np τn0,a NA,a

(2.96)

In the case of a flat defect, the denominator has to be replaced by τp0,a n∗ . The voltage dependence for a flat and deep defect is identical. A diode quality factor of 1 is obtained in both cases as the exponent of the product np equals 1. Since n∗ may be larger than NA,a (the defect is beyond the demarcation levels), the recombination rate for a flat defect will be smaller than for a deep defect. We continue to discuss the deep defect recombination. The product np is not constant in the QNR but depends on the coordinate z. Since diffusion and recombination governs the decay of the minority carrier density, we can again make use of the collection function ηc (z) which will be discussed in depth in Section 2.5.4. It holds np = n2i ηc (z)exp{qV/kT}

(2.97)

simply because p is constant in the p-type QNR and n obeys the reciprocity relation Eq. (2.53). Then, Eq. (2.96) becomes U(z) =

n2i ηc (z) qV exp τn0,a NA,a kT

(2.98)

In case of a flat defect, we only have to replace the denominator by τp0,a n∗ . The saturation current density is calculated after J0 = q

n2i τn0,a NA,a

−wa −da

ηc (z)dz

in the case of a deep defect and J0 = q τ

(2.99) n2i p0,a n

∗

−wa −da

ηc (z)dz in case of a flat defect.

Here, we remind that these solutions are only valid in the limit da Ln , that is, without back surface recombination. In view of the identical diode quality factors for deep and flat defects in the QNR, it is not surprising that also for a defect distribution it results that A = 1. The physical reason for this diode factor is that for small and medium voltage bias only the quasi Fermi level of the minority carriers (EFn in the case of a p-type absorber) varies with voltage. With respect to any defect

61

62


level this variation is proportional to exp{qV/kT}. The recombination rate for an exponential defect distribution is given by Ec −E Ec exp − ∗ np kT dE U= kT∗ Ev τ p + τ N exp − Ec −E n0,a p0,a c kT

(2.100)

Again the case of T∗ > T is the interesting one. With the substitution ν = τp0,a Nc exp − EckT−E the integration over the bandgap can be executed. The τn0,a NA,a activation energy of J0 is Eg . J00 is listed in Table 2.2. These solutions of Eq. (2.98) are not valid for large applied voltage bias. Consider the case of a deep defect with Ed = Ei where the applied voltage is such that n = p. This means that p does no longer equal NA,a but also depends on the voltage. Now, the quasi Fermi levels move symmetrical with respect to the defect level and Eq. np p = . Comparison with Eq. (2.73) (2.95) becomes U ≈ Rn = τn0,a +τp0,a n

τn0,a +τp0,a

suggests that for such a large voltage bias, the diode quality factor reaches the value of 2. In practice this regime may be shielded by series resistances which may limit the current density at higher voltage. So far, we have analyzed the special case of QNR recombination in the limit da Ln where the current density at the back contact approaches zero. For many thin film solar cells, it will be da ≈ Ln and we cannot neglect J(−da ) in Eq. (2.94). Therefore, it is necessary to apply the general Eq. (2.57) where the derivative of ηc (z) at the position z = −wa is used. If Ln is known and, in addition, the back surface recombination velocity of electrons, Sn0,bc , is known, we can use Eq. (2.152) dηc (z) . The saturation current density for defect related in order to calculate dz z=−wa recombination in the QNR reads Dn n2i dηc (z) (2.101) J0 = q NA,a dz z=−wa Any details about the type of defect (flat, deep, continuous) in the QNR are included in the function ηc (z). In the case of a continuous defect distribution, the collection function will become temperature dependent. 2.4.5.3 Back Surface Recombination As a limiting scenario of the QNR recombination we may ask for the diode current in the case where back contact recombination is the dominant process, that is, if the bulk lifetime is very large and the interface recombination velocity is very small. Then the minority carrier electrons diffuse to the back contact and recombine with the abundant holes. We use Eq. (2.101) together with Eq. (2.153). In the limiting case that Ln,a da − wa we can approximate sinh([da − wa )/Ln,a ] ≈ (da − wa )/Ln,a and cosh[(da − wa )/Ln,a ] ≈ 1 and arrive at

Dn,a Dn,a (da − wa ) + Sno,bc L2n,a dηc (z)

≈ 2 (2.102) dz z=−wa Ln,a Sno,bc (da − wa ) + Dn,a

2.4 Diode Currents

which for Sn0,bc (da − wa ) Dn,a and Sn0,bc L2n,a Dn,a (da − wa ) reduces to dηc (z) n.a . Thus, J in the case of dominant back surface recombina≈ dD−w 0 ( a a) dz z=−wa tion scales with the inverse thickness of the absorber. An example for the ηc (z) curve may be found in Figure 2.33b (Ln = 10 µm). 2.4.5.4 Interface Recombination The absorber/emitter interface may exhibit a large number of interface states. Under circumstances these interface states form the main recombination channel of the device. This is particularly the case if the interface bandgap is smaller than the absorber bandgap. What we mean by interface bandgap is exemplified in Figure 2.24. Note that generally we focus on simple absorber/window heterostructures. We refer to absorber/buffer/window type structures where appropriate. In Figure 2.24a, a type I interface is depicted where Ec is positive (type I). Here, the interface bandgap equals the absorber bandgap. Recombination at the interface takes place between electrons from the absorber conduction band and holes from the absorber valence band (we do not depict the interface states). This process is indicated by the arrow in Figure 2.24a. In contrast Figure 2.24b shows a type II interface where Ec is negative. Here, we can define the interface bandgap as Eg,IF = Ec,w − Ev,a = Eg,a + Ec . There are two possible recombination paths for this type of interface. The ‘straight’ recombination with electrons and holes from the absorber. But in addition, cross-recombination where electrons from the emitter conduction band recombine with holes from the absorber valence band. Cross-recombination is depicted by the arrow in Figure 2.24b. Following the SRH recombination model, a type II interface with its reduced (interface) bandgap may well be a high recombination zone. The general form of the SRH equation in Eq. (2.67) is unchanged for the case of interface recombination. However, the net recombination rate R in Eq. (2.67) is given in units of cm−2 s−1 . The diode current density due to interface Ec Ev

(a)

Ep,a z= 0

0

z

Ep,a z =0

absorber (b)

Figure 2.24 Energy band diagrams at the interface between a p-type absorber and an n-type window (without buffer) for (a) Ec > 0 and (b) Ec < 0. The interface bandgap is given by the minimum of the conduction band and the maximum of the

EF

emitter

valence band. In (a) it is Eg,IF = Ec,a − Ev,a and in (b) it is Eg,IF = Ec,w − Ev,a . The arrow in (a) denotes the ‘straight’ recombination path and in (b) the cross-recombination channel. Note that the quantity Ep,az=0 is smaller in (b).

63

64


recombination is given by Jdiode,IF = qR

(2.103)

For interface recombination via defect states without tunneling, we identify γn = Sn0 −1 and γp = Sp0 −1 in Eq. (2.67). Sn0 and Sp0 denote the nominal interface recombination velocities.6) In the case of a discrete interface state of (area) density Nd,IF with defined capture cross sections σn,p , the recombination velocities become Sn,p = Nd,IF σn,p vn,p where vn,p are the thermal velocities of electrons and holes. Similar as for SCR recombination, interface recombination may be enhanced by tunneling. In that case, it is γn = [(1 + )Sn0 ]−1 and γp = [(1 + )Sp0 ]−1 . The interface recombination velocity grows directly with the interface defect density. However, it has an upper limit defined by the thermal velocity. There is no lower limit of Sn0,p0 . But one may ask for an equivalence value of Sn0,p0 , assuming an interface defect density equivalent to the three-dimensional bulk defect density Nd . Let bulk and surface RCs have the same capture cross sections. Then, the equivalent surface recombination velocities may be given by Sn0,p0 =τn0,p0 −1 Nd −1/3

(2.104)

For Nd = 1013 cm−3 and τn0,p0 = 10−8 s this results in Sn0,p0 ≈ 5 × 103 cm s−1 . If one assumes that the area defect density at the interface is not lower than the (area) density in the bulk, one may consider this value as the lower limit for Sn0,p0 . The same holds for the grain boundary recombination velocities Sn0,gb , Sp0,gb and the back contact recombination velocities Sn0,bc , Sp0,bc . In order to calculate the diode current for interface recombination, a number of cases have to be discriminated. Figure 2.25 shows five levels of parameters which can be combined. The first level is the conduction band offset which can be negative (including zero) and positive. The second level distinguishes if both quasi Fermi levels at the interface, En,az=0 and Ep,az=0 , can change with electric bias (left) or if one of the quasi Fermi levels is fixed (right case). In other words, if Fermi level pinning is absent or present (see Section 2.3.4). These cases depend on the interface charge density. If the Fermi energies are not pinned, we further have to discriminate three cases: nonsymmetric heterojunctions with εw ND,w > εa NA,a or εw ND,w < εa NA,a and a symmetric heterojunction with εw ND,w = εa NA,a . In level 4, we discriminate the type of RCs at the interface: These may be discrete or energetically distributed. Finally, recombination at the interface may take place with negligible influence of tunneling ( 1) or by tunneling enhancement ( 1). This manifold of cases 6) Note that the actual interface recombina-

tion velocity depends on the occupation of the interface states Nd,IF . This occupancy in turn depends on the densities of electrons and holes. The nominal interface recombination velocity Sn0 only gives an upper limit of recombination of electrons

assuming that all interface states NIF with capture cross section σn are empty of electrons. Accordingly, Sp0 gives an upper limit of recombination velocity of holes assuming that all interface states with capture cross section σp are empty of holes.

2.4 Diode Currents

Level 1

Level 2

Level 3

Level 4

∆Ec ≤0

I

∆Ec >0

NA,I ≈ND,w /ww ND,I ≈ NA,a/wa

NA,a / wa

εwND,W >εaNA,a

III

εaNA,a = εwND,w

Ed

εaNA,a > εwND,w

Dd(E) II

Level 5

Γ 1

Figure 2.25 Case discrimination for interface recombination at a heterojunction between a n-type window and a p-type absorber. The first level discriminates the conduction band offsets. The second level distinguishes if the charge in interface states is sufficient to pin one of the quasi Fermi energies or if this is not the case. Level 3

Γ 1

refers to different doping ratios with the cases (1–3) denoted in the text. Different defect distributions are differentiated in level 3. Finally, the recombination may ( 1) or may not ( 1) be enhanced by tunneling. Three groups of cases are discussed within this book.

we want to treat in groups I–III as indicated in Figure 2.25. Groups I and III are for the different band line ups but only treat the discrete interface state without tunneling. Group II extends the case of a negative Ec for an interface defect distribution and tunneling enhancement but will only consider an asymmetric heterojuction without Fermi level pinning. Group I: ∆Ec ≤ 0, Interface State at Ed Let us begin with the case Ec ≤ 0 (Figure 2.24b) and the other parameters of group I in Figure 2.25. We assume that deep defects at the interface of an absorber/window heterojunction are present. In that case the auxiliary functions n∗ z=0 and p∗ z=0 can be neglected and Eq. (2.67) becomes for qV 3kT nwz=0 paz=0 R≈U= (2.105) Sp nwz=0 +Sn paz=0

Cross recombination is already taken into account by use of nwz=0 . Only the case of Ec = 0 is ambiguous since then the max(nwz=0 , naz=0 ) has to be used. In the

65

66


case of a flat electron quasi Fermi energy across the interface, that is, EFn = 0 (see Figure 2.29), this maximum depends on the highest density of states, that is, max(Nc,w , Nc,a ). We first assume that the density of charged interface states is small, that is, no Fermi level pinning occurs in the meaning of level 2 of Figure 2.25. The voltage dependence of the interface electron and hole concentrations can be written as q (Vbi − V) nwz=0 = ND,w exp − θ kT q (Vbi − V) (2.106) paz=0 = NA,a exp − (1 − θ) kT Vbi is the built-in voltage which is given by qVbi = Eg,a − En,w − Ep,a + Ec . In Eq. (2.106), we introduced the variable θ = εa NA,a /(εa NA,a +εw ND,w )

(2.107)

which can be rewritten as 1 − θ = εw ND,w /(εa NA,a + εw ND,w ). Equation (2.106) assumes that the quasi Fermi levels are flat from the injecting contacts up the point of recombination, which is the interface. In other words, EFn and EFp shall be close to zero. This assumption is not trivial. As pointed out by Pauwels and Vanhoutte [51] for a high interface recombination velocity of S > 105 cm s−1 and a mobility of 0 and εw ND,w = εa NA,a of an absorber/window heterojunction exceeds Eg,a and is given by Eg,a + Ec . Cases between εw ND,w = εa NA,a and εw ND,w εa NA,a are difficult to treat analytically and are best analyzed by device simulation. The last case of group III in Figure 2.25 is εa NA,a εw ND,w . This is rather a pathological case where interface recombination is so large that the condition EFn = 0 at the interface (for definition of EFn see Figure 2.29) is not fulfilled. Tables 2.1 and 2.2 list the diode quality factors and reference current densities for the most relevant recombination paths and most relevant parameters of absorber/window heterojunction solar cells. Together with the activation energies in Table 2.3, these values define the saturation current density which is a measurable quantity. In Section 2.5.7, we show how to extract the diode parameters A and J0 from the experiment and how to interpret these in view of the recombination path. In Section 3.4 we show the influence of interface recombination on the Voc of a model device. The interface bandgap may be larger than the absorber bulk bandgap due to bandgap gradients or an additional thin layer between emitter and absorber. In that case the activation energy for the saturation current would correspond to the interface bandgap (case without Fermi level pinning) or to the pinning position with respect to the actual band edge at the interface (case with Fermi level pinning). A widening of the absorber bandgap at the interface can thus strongly reduce the interface recombination (see design rules in Section 3.10) Table 2.3 Collection of activation energies for defect related recombination in different regions of an absorber/window heterojunction.


Condition

Activation energy (Ea )

SCR QNR IF

– – No FLP FLP

Eg,a Eg,a Eg,i φb p

2.4 Diode Currents

Grain boundaries in a polycrystalline absorber film can as well be considered as interfaces. The SRH recombination rate can be calculated after Eq. (2.105) where Sn0,p0,gb is the grain boundary recombination velocity Sn0,p0 = Ngb σn,p vn,p . In Section 3.7, the effect of horizontal and vertical grain boundaries on the device performance is discussed. 2.4.6 Parallel Processes

In practical heterostructure solar cells, all recombination paths formerly discussed are active in parallel. Normally, one recombination mechanism will be dominating. However, the dominating mechanism may be changing under changing electrical bias. Consider N recombination mechanisms. The total diode current density is N ( qV (2.125) Jdiode = J0i exp Ai kT i=1 If all processes exhibit the same diode quality factor, the recombination path with the highest J0 dominates the diode current at any bias. From Table 2.1 we know, however, that the A factors can vary between 1 and 2 and beyond this range. In view of the exponential dependence in Eq. (2.125), this is a large variation. 2.4.6.1 SCR and QNR Recombination For typical heterojunction solar cells which are not dominated by interface recombination, SCR recombination, and QNR recombination take place in parallel. We first ask which recombination mechanism is the dominant one as a function of the carrier lifetime. We simply consider absorber bulk recombination via a single deep defect. The reference current densities J00,QNR and J00,SCR , given in Table 2.2, both depend on the carrier lifetime. If we assume equal density of states and equal lifetimes for electrons and holes, J00,SCR has a 1/τn0,p0 dependence. J00,QNR , in addition is influenced by ηc (z) which also includes the carrier lifetime. Consider the limiting case that the absorber thickness is much larger than the diffusion length. Then Eq. (2.109) can be used for J00,QNR and the integral in this equation be0 comes −da ηc (z)dz = Ln,p . (For recombination at vertical grain boundaries, Ln,p,eff ' has to be used.) With Ln,p = Dn,p τn0,p0 the reference current becomes J00,QNR ∼ √ 1/ τn0,p0 . Thus, the slope of J00,SCR is larger than that of J00,QNR . This is apparent in Figure 2.27b. In Figure 2.27a, Eg,a /q − Voc is calculated in the limit of SCR and QNR recombination. We find a certain lifetime which marks the transition from SCR limited to QNR limited recombination. In the example of Figure 2.27 this border lifetime is at τn0,p0 ≈ 10−9 s. Note that, close to this border lifetime, recombination in the SCR and QNR is about equally efficient. The border lifetime is largely independent of the absorber bandgap, the effective density of states and the absorber doping. Figure 2.19 shows the accumulated recombination rate for the devices with τn0,p0 = 10−8 and τn0,p0 = 10−11 s. For absorber layers with da < Ln,a , the correct expression for J00,QNR has to be used. Then, the integral in Eq. (2.152) strongly depends on the back surface

73


L [µm] 0.1 0.2 0.5 1.0 2.0

Eg,a /q - Voc [V]

0.2

SCR

0.3 0.4 0.5

QNR

0.6

J00 [mA cm−2]

0.7 (a) 0.8

109

QNR

107 105

SCR

3 (b) 10

(dVoc /dT) T=RT [mV/K]

74

2.4 2.2 2.0 1.8 1.6 1.4 10−11

(c)

10−10

10−9

10−8

10−7

τn,p [S]

Figure 2.27 (a) Calculated values of Voc for recombination in the QNR (dashed line) and in the SCR (solid line) as a function of the carrier lifetime. Device parameters as in Table 8.1. No buffer layer and no interface recombination have been included. The line of extended dashes has been calculated by a two-diode model. Input parameter for all

calculated Voc are simulated values of Jsc . Data points () give simulated Voc values for comparison. (b) Lifetime dependence of J00 for SCR ad QNR recombination. (c) (dVoc /dT)T=RT calculated at room temperature for SCR (solid line) and QNR (dashed line) recombination.

recombination velocities Sn0,bc and Sp0,bc . Imagine that the back surface would be perfectly passivated, that is, Sn0,bc = Sp0,bc = 0. Than ηc (z) would be constant in the 0 QNR and −da ηc (z)dz = da . The reference current density J00,QNR would depend on da and the remaining dependence on carrier lifetime is 1/τn0,p0 . This would shift the border lifetime to slightly higher values and would increase the Voc in the regime of QNR recombination. QNR recombination is advantageous over SCR recombination in view of its temperature dependence. This becomes apparent in Figure 2.27c. The reason is the smaller A factor of QNR recombination.

Jrec [mA cm−2]

2.4 Diode Currents

105 100 SCR QNR IF

10−5

0.0

0.2

0.4

0.6

0.8

1.0

Voltage [V] Figure 2.28 Diode current densities for different recombination paths of an absorber/window solar cell. Device parameters as in default Table 8.1. Solid line: SCR recombination calculated after Eq. (2.79). Dotted line: QNR recombination calculated after Eq. (2.99) with ηc (z) = 1. Dashed line: IF recombination calculated after Eq. (2.109).

The dominance of SCR or QNR recombination also depends on the voltage bias. We selected a device with τn0,p0 in the absorber and plotted the different diode currents as a function of electrical bias (Figure 2.28). Due to the diode quality factor A = 2, the Jdiode,SCR has the smallest voltage dependence. The voltage dependence of Jdiode,QNR is larger since QNR recombination type exhibits A = 1. The latter is also true for IF recombination (for θ → 0 – see Eq. (2.107)) which is given in addition in Figure 2.28. The example in Figure 2.28 shows (the well known result) that SCR recombination dominates at small bias and QNR recombination limits the total current for large forward bias. The total current density will be the maximum curve in Figure 2.28. In the example, the curve will change slope at 0.6–0.7 V electric bias. As the device parameters underlying Figure 2.28 are those of the default device, we note that the Voc of this device is limited by QNR recombination. 2.4.6.2 SCR and IF Recombination For the parameters selected in Figure 2.28, IF recombination does not dominate at any bias. However, we know from Eq. (2.116) that IF recombination critically depends on Ep,az=0 . In fact, J0 for IF recombination increases by 1 order of magnitude for a decrease of Ep,az=0 by 60 meV. Ep,az=0 decreases either by a conduction band cliff, by a too small doping ratio or by Fermi level pinning (FLP). With a conduction band cliff of Ec = −0.4 eV, IF recombination would be dominating in the complete voltage interval in Figure 2.28. For Ec = −0.25 eV, the total diode current after Eq. (2.125) is dominated by IF recombination for V > 0.5 V. As an effect, the open circuit voltage is limited by IF recombination. This aspect is discussed in Section 3.2. The limitation of the Voc from the different recombination mechanisms is discussed in Section 2.6.2.3.

75

76


2.4.7 Barriers for Diode Current

So far, we have assumed that the SCR is the only barrier for current transport in the device. There may be, however, other barriers which limit the diode current. Energy barriers can form at the interfaces of a device or, to a limited extent, within a layer exhibiting bandgap gradients. Figure 2.29 depicts two relevant examples for interface barriers. Figure 2.29b shows an electron barrier φb n which is due to a negative buffer/window conduction band discontinuity and band bending within the buffer layer. It holds φb n = En,az=0 + Ec b,a − EFn where En,az=0 is given by Eq. (2.35) and EFn is the drop of the electron Fermi level at the interface. According to Figure 2.6c, also negative charge at the buffer/window interface can lead to an electron barrier φb n . To some extent the electron barrier can be voltage dependent. In the example of Figure 2.29b we have the voltage dependent part En,az=0 − EFn and the voltage independent part +Ec b,a . Due to the voltage dependent part, the electron barrier can be considered as a nonlinear electronic element which is in series with the main diode. In Figure 2.29a, we depict an example for a barrier at the absorber back contact. The back contact represents a Schottky diode of opposite polarity as the main diode (heterojunction). It is a hole barrier φb p which can impede the injection of holes into the p-type absorber. Thus, it can block the diode current. In the following, we will assume that the main diode and the back contact diode are independent and the depletion regions do not overlap. Otherwise one speaks of a reach-through diode [54, 55]. Charge transport over both barriers in Figure 2.29 may be described by thermionic emission (TE) theory. Quantum mechanical effects at the energy barrier such as reflection or tunneling of carriers, are not considered. In the TE theory, the saturation current density is given by Jb0 = A∗ T2 exp{−φb /kT} ≈ 1/4qvn,p Nc,v exp{−φb /kT}

(2.126)

where A∗ is the effective Richardson constant and Nc,v is the density of states in the energy band of the majority carrier [56].

Φbn = En,b z=0

En,a z=0 ∆EFN

Φbp E (a) z =-da

(b)

z

z=0

Figure 2.29 Barriers for the diode current. (a) Barrier for hole injection φb p . (b) Barrier for electrons φb n depicted at small forward bias.

2.4 Diode Currents

For the electron barrier in Figure 2.29b, the total current across the barrier is given by the current density of electrons from the window to the absorber minus the current density from the absorber to the buffer [57]. −En,bz=0 Eb,a 1 c + En,az=0 JTE = qvn,p Nc,w exp − Nc,a exp − (2.127) 4 kT kT (The index TE refers to thermionic emission.) With En,bz=0 = En,az=0 + Eb,a c + EFn 9) we arrive at −En,az=0 EFn Nc,w exp + −1 (2.128) JTE = J0,TE exp Nc,a kT kT with J0,TE

1 Ec b,a = qvn,p Nc,a exp − 4 kT

(2.129)

EFn increases with electric forward bias. In contrast, En,az=0 decreases with bias provided that there is no FLP. The size of EFn depends on the value of Ec b,a . In our default device (Ec b,a = 0)EFn is below kT. This renders the bracket in Eq. (2.128) very small. However, J0,TE is large as long as Ec b,a + En,az=0 is not too large. Therefore, it is well possible to pass a diode current even in the presence of a limited conduction band barrier. If Ec b,a , however, is large, the total diode current becomes limited by the electron barrier. In that case, we may observe a very high Voc and a very low Jsc (see example in Figure 3.5). The barrier φb n in Figure 2.29b may be illumination-dependent as shown in the example of Figure 2.13. An illumination dependent barrier height φb n can lead to cross over of the dark and light JV curves. An electron barrier which is not (or hardly) voltage dependent will lead to a complete blocking of the forward current (leakage effects not considered). In that case, we will find a roll over of the illuminated JV curve. In the case of the blocking absorber back contact in Figure 2.29a, the total current density depends on the voltage drop at the back contact Vb by Jb = −Jb0 (exp{−qVb /kT} − 1). (The minus signs in front of Jb and Vb reflect the opposite polarity of the back contact junction.) The externally applied voltage, V, distributes over the main and back surface junction. It is V = Vm + Vb , where we use the subscript m for the main junction. The current density through the main junction and the back contact barrier must be equal, that is J = J0 (exp{qVm /AkT} − 1) = Jb0 (1 − exp{−qVb /kT})

(2.130)

At the moment, we neglect the effects of series and shunt resistances. As long as it is J0 (exp{qVm /AkT} − 1) Jb0 , the main voltage drop is at the main junction. Then, the flow of the diode current is not impeded by the back surface barrier. If the diode current, in contrast, approaches Jb0 , the applied voltage starts to drop at the 9) In Figure 2.29, EFn is positive since

dEFn /dz > 0.

77

78


back contact barrier. This leads to current saturation and a JV curve with a roll over anomaly (see Section 7.1.1). Note, that if the device acts as a photovoltaic generator the back contact barrier is forward biased. Thus a moderate back contact barrier does not impede the flow of photo current at short circuit [58]. This is exemplarily quantified in Section 3.8.

2.4.8 Bias Dependence

In Section 2.3.6 we discussed how deep states can modify the equilibrium band diagram due to their supplementary charge. In Section 2.4.5, we saw how deep states can become hole or electron traps dependent on their capture cross sections and on the position of the quasi Fermi levels. The decisive quantities are the demarcation levels for deep states. Thus, here it remains to mention some possible effects of a band diagram changed by supplementary charge due to voltage or light bias. We may place positive and negative charge in the absorber layer, at the buffer/absorber interface or in the buffer layer. Due to the normally much higher doping, we leave out the window layer. If under voltage or light bias negative charge is added to the SCR of a p-type absorber this can lead to:

1) An increased electric field, Fm , at the position of τn0,a −1 n = τp0,a −1 p. As a result the zone of effective recombination πkT/qFm in the absorber SCR is reduced. This leads to a lower diode current (see also Table 2.2). 2) Tunneling enhancement of SCR recombination if Fm becomes so large that it approaches F (see Section 2.4.5.1). 3) A change of the effective doping ratio between buffer and absorber (or emitter and absorber). As a result En,az=0 may be increased with the effect of an increased interface recombination.

Certainly, all effects can occur simultaneously. The net effect on the solar cell device will be observable as a gain or loss in the Voc . An increased negative charge at the absorber/emitter interface can lead to the opposite effects as described above. Similar can be the effects of negative charge in the buffer layer or at the buffer/window interface. If the diode saturation current is voltage dependent, we observe a strong violation of the superposition principle even if we correct the voltage dependence of the photo current. An example is the cross over of light and dark JV curves [59]. If charging and discharging of the deep states exhibits very large time constants, the recombination processes can become dependent on the experimental history in terms of minutes or hours [27, 60].

2.4 Diode Currents

2.4.9 Non-Homogeneities

So far, we have calculated diode currents in heterostructures formed out of homogeneous and isotropic layers. However, within the three-dimensional absorber there may be distinct zones of particular high defect densities: grain boundaries and regions of high strain, accumulated impurities, or stoichiometry deviations. These regions can induce a very high recombination rate provided they are located close to the point of τn0,a −1 n = τp0,a −1 p. As a result a large diode current flows independently of the defect density in the absorber host. A grain boundary is a good example which can stand also for other high recombination zones. Grain boundary recombination is discussed in Section 3.7. Here, we only note that limited regions of high defect density can strongly reduce the Voc while the Jsc can be less affected. In the same sense, the absorber/emitter interface may be considered as a high recombination zone. Besides the possibility of distinct regions of high defect densities, the defect density may (systematically) fluctuate on a lateral scale. Equation (2.48) shows that the saturation current density depends on J00 . According to Table 2.2 the reference current J00 for all considered recombination processes is proportional to τn0,p0 −1 or sn0,p0 , thus is proportional to the bulk or interface defect ' density. Furthermore, J00 directly or indirectly (for SCR recombination Fm ∼ NA,a ) depends on the doping density in the absorber. Both the defect density and the doping level can fluctuate on a lateral scale. Let us assume that, as a result, the reference current density varies by orders of magnitude and obeys a log normal distribution PLn according to dPLn (J00 ) =

2 − lnJ00 − lnJ00 dJ00 exp √ 2σn J00 σn 2π

(2.131)

where σn is the dimensionless standard deviation [61]. Then, J0 will obey the same log normal distribution and will be influenced by regions of high defect density. In order to understand the impact of a fluctuating J00 let us investigate the J00 −related term which limits the Voc of a device. This is −AkT/qln(J00 /Jsc ). It is valid for all recombination mechanisms discussed in this book (see Section 2.6.2.3). If we describe J00 by the log normal distribution, the term is modified to −AkT/qln (J00 /Jsc ). For a log normal distribution the median J00 is larger than the position of the distribution’s function maximum. Thus, small areas with large J00 limit the Voc much stronger than large areas with small J00 . Intuitively we do expect such a behavior. In the other case that J00 is only Gaussian distributed, the impact on Voc will be much smaller. Besides high defect densities, also a reduced activation energy for the recombination process can be the origin of a high recombination zone. Considering Eq. (2.48), it is clear that a variation of Ea has a super linear influence on the diode current. For bulk recombination in the absorber, the activation energy equals Eg . If Eg is locally reduced, a locally higher saturation current may be the result. An example for distinct regions of reduced bandgap are inclusions of a secondary

79

80


phase. Other reasons for bandgap fluctuations are variations in stoichiometry, alloy composition and strain. In the case of interface recombination, the activation energy is the interface bandgap (no Fermi level pinning) or the hole barrier (Fermi level pinning). Both quantities can fluctuate on a micro- and mesoscopic scale. Rau and Werner [33, 62] have proposed a general approach to describe the influence of a fluctuating activation energy. They assume Ea to obey a Gauss-normal distribution of

2 − E a − Ea dEa (2.132) exp dPG (Ea ) = √ 2σEa σE 2π a

Here, Ea is the mean of the distribution and σEa is the standard deviation (with the dimension of an energy). Insertion in Eq. (2.48) gives the saturation current density for a distributed Ea : ∞ Ea dPG (Ea ) J0 = J00 exp − (2.133) AkT −∞ 10)

Performing the integration with aid of tabulated integrals we find σ2 Ea Ea − J0 = J00 exp − AkT 2(AkT)2

(2.134)

Thus, the activation energy for the diode current decreases with increasing σEa . The cell recombination process may be described by an effective activation energy Ea,eff = Ea −

σ2 Ea 2AkT

(2.135)

Obviously, Ea,eff is smaller than Ea . The impact of Ea,eff on the Voc can be understood by considering the expression Voc = Ea,eff /q − AkT/q ln(J00 /Jsc ) which is rationalized in Section 2.6.2.3. We see that a reduced Ea,eff directly decreases Voc . Equation (2.134) is valid for any type of recombination which is thermally activated. Rau et al. showed that the effect of fluctuating J0 on the cell performance can be limited by a small series resistance [61]. We note in passing that if a cell with fluctuating activation energy is analyzed according to Eq. (2.48) instead of according to Eq. (2.134), a temperature dependent diode quality factor may be diagnosed (see Section 7.3.4).

2.5 Light Generated Currents

In this section, we discuss generation and recombination of charge carriers by photon absorption. The section’s logic follows from the principle of superposition

dPG (Ea ) = 1, integration is performed between −∞ and ∞;

10) In order to fulfill

although negative bandgaps have no physical meaning.

2.5 Light Generated Currents Jgenmax

Joptloss

Jgen

Jelloss(V)

Figure 2.30 Schematic presentation of photo currents in a solar cell. The generation current Jgen max is the ideal current density obtained in the absence of recombination and optical losses. The part Jgen can be collected in the solar cell structure while internal recombination amounts to Jel . The external current is Jph (V).

Jph(V)

of diode and photo currents. A failure of the superposition principle, often observed in chalcogenide solar cells, will be alleviated by allowing the photo current to be voltage dependent. Figure 2.30 gives an overview of the currents treated in this section. The generation current Jgen is lower than the maximum generation current Jgen max due to an optical loss of Jopt loss . The photo current is lower than the generation current due to an electronic loss of Jel loss . In this book we calculate Jph (V) as a negative quantity. One can easily verify by aid of Figure 2.15 that the flux of photogenerated electrons is from left to right and thus – in this case – represents a negative Jsc . 2.5.1 Generation Currents

The maximum generation current is obtained if we integrate over the solar photon spectrum and neglect all loss mechanisms. Indeed, it is instructive to look at the maximum possible generation current, Jgen max , as a function of the bandgap of the absorbing layers. The solar cell may consist of an emitter and absorber of bandgaps Eg,e and Eg,a , respectively. It may be homogeneous in the direction x, y thus defining a one-dimensional problem. A photon flux density djγ (E)/dE per unit area and unit energy [63] shall be impinging on the solar cell. The energy of the photons is given by ¯hω. Figure 2.31a shows the AM1.5G and AM0 equivalent photon flux densities as a function of photon energy. Alternatively, we can present the spectra in values of the energy flux density, djE , where it holds that djγ (E) = djE (E)/E

(2.136)

We write the energy flux density of the AM1.5G spectrum as jE 100 indicating the energy flux of 100 mW cm−2 . In the ideal case, only the absorber bandgap limits the absorbed flux density. Thus, we construct an ideal absorbance Aid (E) which is 0 for E < Eg,a and 1 for E ≥ Eg,a . This implies that the absorber is sufficiently thick to exclude optical transmittance, that is, T(E) = 0 for E ≥ Eg,a . Further it assumes that the reflectance R(E) is zero for Eg,a < E < ∞. The integral over the spectral photon flux density for the absorber/emitter combination reads

81

dJγ /dhν [s−1 m−2 eV−1]


4×1021 3 2 1 0 0

0

−2

−20

(b)

−40 −60

Jsc [mA cm−2]

(a)

Jsc [mA cm−2]

82

−4

Jsc,a

Eg,e

−6

Eg,a

Jsc,e

−80

−8 −10

−100 1.0

2.0 E [eV]

3.0

Figure 2.31 (a) Spectral photon flux density as a function of energy for AM1.5G (solid line) and AM0 (dashed line). The spectra have been recalculated after tabulated spectra of dje (λ)/dλ and are subject to a small

Jmax gen Eg,a = −q

or Jmax gen λg,a = −q

∞

Aid (E)

∞

Aid (λ) 0

numerical error. (b) Integrated photon flux densities for AM1.5G (solid line) and AM0 (dashed line) representing the maximum short circuit current density achievable for an absorber with bandgap Eg = E.

djγ (E) dE

0

4.0

djγ (λ) dλ

dE = −q

∞

djγ (E)

Eg,a

dE

λg,a

djγ (λ)

dλ = −q 0

dλ

dE dλ

(2.137)

In the first and second part of Eq. (2.137) we expressed Jgen max by aid of energy and wavelength dependent functions, respectively.11) Numerical integration of djγ (E)/dE gives the maximum generation current density as a function of the absorber bandgap (Figure 2.31b). It is clear that Jgen max (Eg,a ) increases with decreasing Eg,a . Jgen max (Eg,a ) is larger for the AM0 spectrum than for the AM1.5G spectrum due to the higher photon flux density. 11) It shall be remembered that djγ (λ) dλ

djγ (E) dE =

− hc2 E

where the minus sign indicates the different slopes of the spectra.


Figure 2.31b gives an upper bound for an ideal solar cell. The figure can also be used to read out the share of maximum generation current achievable from the absorber and emitter separately. Optical losses reduce Jgen max by the quantity Jopt loss . Optical loss may have several origins. We start with an ambiguous case which, in principle, could also be considered as an electronic loss. The window layers of a current chalcogenide solar cell exhibit low electronic quality and do not add to the photo current. Thus, all electron–hole pairs generated in the window layer are lost.12) We can write the photo current loss due to absorption in the window layer as

Jloss,w Eg,w = −q opt

∞

djγ (E)

Eg,w

dE

dE

(2.138)

We can directly use Figure 2.31b in order to read out the optical loss due to window absorption. For a window bandgap of about Eg,w = 3.5 eV and an assumed AM1.5G spectrum, Jopt loss,w amounts to 0.5 mA cm−2 . For the AM0 spectrum this loss is 1.4 mA cm−2 . With the assumption of a complete loss in the energy range of the window absorption, we concentrate on the remaining energy interval Eg,a < E < Eg,w . In this interval, photonic loss is caused by reflectance R(E) = 0, transmittance T(E) = 0, and by parasitic absorption in the window layer Apar (E) = 0. These losses can be quantified as = −q Jloss,O opt

Eg,w Eg,a

O(E)

djγ (E) dE

dE

(2.139)

where O(E) stands for R(E), T(E), or Apar (E). Jopt loss,T is due to photons in the interval Eg,a < E < Eg,w which pass the absorber and reach the back contact where they are not reflected but absorbed. This loss can be minimized by a sufficient absorber thickness or by a back contact reflector. Jopt loss,R is due to the total optical reflectance of the heterostructure. It can be minimized by matching of the indices of refraction of the different layers or by adding interference layers. Jopt loss,A is due to absorption of the window layer in the interval Eg,a < E < Eg,w . A typical example of Jopt loss,A is free carrier absorption in the highly doped window layer. The functions R(E), T(E), and Apar (E) can either be measured for the completed cell or can be generated from the optical constants of the constituting layers and by application of an appropriate matrix formalism [64]. The generation current which finally arises from the generation of electron–hole pairs in the photoactive layers amounts to Jgen = Jgen max − Jopt loss,w − Jopt loss,T − Jopt loss,R − Jopt loss,A . If we define the absorbance of the photoactive layers of the cell as A(E) = 1 − T(E) − R(E) − 12) Looking from the absorber side, Jopt loss,w in

fact is an optical loss. Certainly, the loss in the window layer could also be expressed by a zero collection function. Later, for

illustration purposes we will indeed set the collection function to zero in the range of the window layer.

83

84


Apar (E), we can also write the generation current as

Jgen = −q

Eg,w Eg,a

A(E)

djγ (E) dE

dE = −q

λg,a

λg,w

A(λ)

djγ (λ) dλ

dλ

(2.140)

with A(λ) = 1 − T(λ) − R(λ) − Apar (λ). We note that in a heterostructure consisting of i photoactive layers the total absorbance is A(λ) = i Ai (λ). The absorbance in the mth layer depends on the optical properties of all other layers. 2.5.2 Generation Function

While the absorbance only allows to calculate the density of photons per unit of time absorbed in the complete heterostructure or within the complete thickness of a single layer of a heterostructure, we are interested in a function which describes the position dependent generation rate. This function is the generation function G(z). It results from the absorption of all wavelengths at a given position z. In Table 2.4, we see that the generation current can be given as an integral over the position dependent generation function for an absorber/emitter heterostructure. The integration limits obey the following arguments: If the emitter consists of window and buffer, integration is performed over the buffer and absorber thickness.13) If the emitter only consists of a window layer, the higher integration limit is zero, thus corresponding to the surface of the absorber. As explained in 2.5.1, we leave out the window layer since the loss in this layer was already quoted as an optical loss. The relation between the absorbance in a particular layer and the generation function is mediated by the spectral generation rate dG(λ, z) (see auxiliary functions in Table 2.4). Note that this rate is not normalized but depends on the incoming photon flux. One can introduce the normalized generation rate Gn (λ, z) = dG(λ, z)/djγ (λ)

(2.141)

For an absorber/window heterojunction, the spectral generation rate in the wavelength range λg,w < λ < λg,a can be approximated by dG(λ, z) = djγ (λ)T(λ)(1 − R(λ))αa (λ)exp{αa (λ)z}

(2.142)

where R(λ) should be obtained from optical measurement, αa (λ) is the absorption coefficient of the absorber and T(λ) is the transmittance of the window layer. Equation (2.142) expresses the well known fact that light which arrives at the absorber generates charge carriers with an α(λ)exp{α(λ)z} depth dependence.14) For 13) As usual in this book, the zero position

shall be at the absorber/emitter interface. This is why we integrate from −da to db . 14) Note that here the propagation direction of photons is anti-parallel to the z-axis.

Therefore, the generation rate decreases toward negative z. This is why the reader may miss the minus sign in the exponential function in Eq. (2.142).

2.5 Light Generated Currents Table 2.4 Wavelength and position dependent formulations of the current densities introduced in this section. The integral boundaries are de , db , and −da for the thicknesses of window, buffer, and absorber layer, as well as λg,a and λg,w for the band

Wavelength dependent Jmax gen

−q

λg,a

djγ (λ) dλ

0

dλ

edge equivalent wavelengths of absorber and window. The auxiliary functions allow the transition between wavelength and position dependent descriptions. The functions are explained in the text.

Position dependent

–

Auxiliary functions

– A(λ) =

Jgen

−q

λg,w

Jph (V)

−q

λg,a

A(λ)

λg,a

λg,w

A(λ)

djγ (λ) dλ

djγ (λ)

× IQE(λ, V)dλ

dλ

dλ

−q

−db

−q

da

G(z)dz

G(z) =

da −db λg,a

λg,w

Gn (λ, z) = da

−db

G(z)ηc (z,V)dz

dG(λ,z) dλ dz dλ djγ (λ) dG(λ,z) dλ dλ

dG(λ,z) djγ (λ)

EQE(λ,V) = IQE(λ,V)A(λ) EQE(λ,V) = da dG(λ,z) ηc (z,V)dz −db djγ (λ)

an absorber/buffer/window heterostructure the approximation in the wavelength range λg,b < λ < λg,a reads dG(λ, z) = djγ (λ)T(λ)(1 − R(λ))exp{−αb (λ)db }αa (λ)exp{αa (λ)z}

(2.143)

where αb (λ) and db are the absorption coefficient and thickness of the buffer layer. These formulas neglect the effect of multiple reflection and interference. More precisely, dG(λ, z) is calculated by dividing each layer of the heterostructure in a large number of sublayers. Then the absorbance in each sublayer is calculated. Integration of a sublayer absorbance multiplied by the spectral photon flux density gives the spectral generation rate. This calculation is best performed by aid of an optical matrix calculation program. Knowing dG(λ, z), we can calculate the local generation rate of charge carriers excited by a spectral photon flux density djγ (λ) in the device from λg,a dG(λ, z) G(z) = dλ (2.144) dλ λg,w Figure 2.32 gives an optical generation function which is exemplary for a solar cell based on a strongly absorbing chalcogenide layer. The majority of charge carriers are generated within the first 500 nm of the absorber. The decline of G(z) is not purely exponential due to the integration over various dG(λ, z) functions with different α(λ). Knowing the generation function is essential for the simulation of

85


16× 1018 G(z) [cm−3 s−1]

86

-da

0 db

Figure 2.32 Optical generation function of the default absorber/buffer/window solar cell. Due to the smallest bandgap of the absorber with thickness da , most of the electron–hole pairs are generated therein. Note that the decline of the generation rate in the absorber is supra-exponential.

dw

12 8 4 0

−1500

−1000

−500 z [nm]

0

500

quantum efficiency (QE) measurements and for seeking the collection function (see below). By use of the functions – generation function, spectral generation function or normalized spectral generation function – the generation current can be written as Jgen (V) = −q = −q

db −da

G(z)dz = −q

λg,a

λg,w

djγ (λ) dλ

db

db λg,a

G(λ,z)dλdz

−da λg,w

Gn (λ, z)dzdλ

(2.145)

−da

An overview of the functions applied gives Table 2.4 2.5.3 Photo Current

Not all photons absorbed in the photoactive layers contribute to the photo current. Charge carriers which undergo recombination are responsible for the loss term Jel loss . This loss current can be voltage dependent. Table 2.4 shows that for the calculation of the photo current density from position dependent functions we introduced the collection function ηc (z, V). Before we proceed to discuss this collection function, let us briefly describe the connection between the collection function and the recombination rate. The photo current density in an illuminated heterojunction can be calculated by integration of the continuity equations (Eqs. (2.13) and (2.14)) for electrons and holes. We consider a thin film solar cell as depicted in Figure 2.2 where the junction normal is in the z direction and which is of the absorber/emitter type. Photons impinge at the front surface of the cell at z = dE . Integration of the steady state continuity equation yields de Jph (V) = −q (G(z) − U(z, V))dz (2.146) −da


In order to allow a voltage dependent photo current, the recombination rate is written as U(z, V) while the generation rate G(z) is assumed to be voltage independent. For zero applied voltage, Jph (0) equals the short circuit current, Jsc . The local recombination rate U(z, V) depends on the carrier density and therefore on the generation rate. Therefore, it is not a unique function of the device but changes with changing excitation. However, since there is the charge collecting junction in the solar cell we are able to measure a collection current. The collection current which is induced by any excitation source is given by the product of the generation function and the collection probability. It is de G(z)ηc (z, V)dz (2.147) Jph (V) = −q −da

The position dependent collection function ηc (z, V) is a unique function of the device. It is independent of the generation rate (nonlinear effects excluded) and valid for the particular device geometry. The collection function of the absorber only depends on the position of the collecting junction. There is no simple connection between U(z,V) and ηc (z, V). When the collection function is known, the complementary function (1 − ηc (z, V)) can be used to calculate the electronic loss term Jel loss (V) by a similar equation like Eq. (2.147). Certainly, Jel loss (V) can also be obtained from Jgen − Jph (V). 2.5.4 Collection Function

The internal collection function is the probability for a generated charge carrier to be collected at the contacts [65] and thereby to contribute to Jph (V). ηc (z, V) is valid for photonic excitation as well as for impact excitation by electrons or other particles. Moreover, ηc (z, V) is valid for carrier injection and limits the diode current in the diffusion limit (see Section 2.4). In a device as shown in Figure 2.4 the sought collection function describes recombination in the emitter as well as absorber including their interfaces and depletion regions. Note, however, that the photo current loss due to absorption in the window layer was already being accounted for as optical loss. In the following, we discuss the collection function in the QNR with or without grading regions, in the SCR and in the buffer layer. 2.5.4.1 Absorber Quasi Neutral Region Charge carriers generated within the QNR of the absorber (wa < −z < da ) either diffuse to the SCR at position −wa (see Figure 2.33) or recombine. Their collection function is position dependent and can be calculated using the reciprocity theorem for charge collection [66].15) The reciprocity theorem states that the local collection 15) The reciprocity theorem can also be derived

from the principle of detailed balance [65]. It is independent of the carrier statistics (Boltzmann or Fermi–Dirac) and also valid for nonlinear recombination.

87

88


function in the QNR is identical to the normalized excess minority carrier concentration of a voltage-biased device in the dark [65]. As an equation, this identity reads ηc (z) =n(z)/n(−wa )

(2.148)

−wA is the location of the carrier injecting SCR edge being identical to the charge-collecting junction. The advantage of using the reciprocity theorem is that this function is obtained from an homogeneous differential equation [67]. If ηc also depends on x and y in a three-dimensional device, the quantity ηc (r) equals n(r)/n(r ). For a one-dimensional device an analytical expression for ηc (z) in the QNR of the absorber can be derived. For a p-type absorber in the limit n p (low injection or medium excitation) as well as p∗ p (trap level at mid-gap), the net recombination rate in Eq. (2.67) is given by U − G0 = (n(z)p0 − n0 p0 )/τn0,a p0 = n(z)/τn0,a . Thus, the net recombination rate is linear in n(z).16) The differential equation for n(z) reads Dn

d2 n(z) n(z) − = −G(z) dz2 τn0,a

(2.149)

where Dn is the diffusion constant for electrons and τn0,a is the lifetime of the electron in the absorber. The reciprocity theorem relies on the linearity and symmetry of the differential operator defining this continuity equation. According to the reciprocity theorem, the differential equation for ηc (z) is homogeneous and reads Dn

d2 ηc (z,V) ηc (z,V) − =0 dz2 τn0,a

(2.150)

The boundary conditions are ηc (−wa ) = 1,

dηc (z) Sn0,bc = ηc (z) dz z=−da Dn

(2.151)

where da is the absorber thickness and −wa is the position of the SCR edge. With the ansatz ηc (z) = C1 exp{−z/Ln,a } + C2 exp{z/Ln,a }, Eq. (2.150) can be solved leading to S a sinh −z−d − Dn0,bc Ln,a n,a ηc (z,V) = S n0,bc 1 a (V) a (V) + Ln,a sinh da −w cosh da −w Dn,a Ln,a Ln,a 1 cosh Ln,a

−z−da Ln,a

16) This is very important since it allows to

solve the continuity equation for n in the QNR of an absorber. Furthermore, it allows to define a diffusion length as the

(2.152)

logarithmic decrement of the collection probability.


1.2 1.0 ηc

0.8

Ln,a = 10, 1, 0.5, 0.1 µm

Sp0,bc = 102 cm s−1

0.6 0.4

107 cm s−1

0.2 -da -(wa+Ln,a) (a)

-wa0 z [µm]

Figure 2.33 Collection function in the absorber as a function of the position z. Diffusion constant of electrons Dn,a = 1 cm2 s−1 . The charge collecting junction is located at z = −wa . In (a), the back surface recombination velocity is varied at

de

-wa0

-da (b)

de

z [µm]

fixed diffusion length of Ln,a = 1 µm. Note that the curves of Sn0,bc = 106 cm s−1 and Sn0,bc = 107 cm s−1 are almost identical. In (b), the diffusion length of electrons is varied at fixed Sn0,bc = 106 cm s−1 .

Here, Ln is the bulk diffusion length for electrons which is calculated from L2n,a = Dn,a τn0,a . Figure 2.33a shows an example of ηc (z) for different values of the back surface recombination velocity, Sn0,bc . The parameters da , Dn,a , and Ln,a were selected in order to mimic a realistic absorber layer (see Figure 2.33). For the cases Sn0,bc = 106 and 107 cm s−1 , ηc is very small near the back contact. For Sn0,bc = Dn,a /Ln,a (Sn0,bc = 104 cm s−1 in our example), ηc exponentially declines and the collection function is similar as in a semi-infinite absorber. If Sn0,bc is smaller than Dn,a /Ln,a , the back surface is passivated. This can be due to a minority carrier mirror by some kind of back surface gradient (see Sections 3.8 and 3.10). As explained in Section 2.4, the diode current in the limit of diffusion is calculated by use of the derivative of ηc (z) at the position z = −wa . From Eq. (2.152) we find dηc (z) = dz z=−wa

Sn0,bc 1 sinh(y) + Dn,a cosh(y) Ln,a Ln,a 2 ;y Sn0,bc 1 sinh(y) + Ln,a cosh(y) Dn,a

=

da − wa Ln,a

(2.153)

The collection function ηc (z) can be experimentally determined from a QE measurement or an electron beam induced current (EBIC) experiment. Knowing ηc (z), one can extract the parameters Ln,a and Sn0,bc by curve fitting [68]. The example in Figure 2.33 shows that both Ln,a and Sn0,bc influence the shape of the collection function. The influence of Sn0,bc becomes larger for larger Ln,a as is shown in Figure 2.33b. Only for Ln,a being much smaller than the absorber thickness, the effect of Sn0,bc may be neglected. In a thin film polycrystalline cell where z is the direction of the heterojunction, the collection function will vary with x and y due the effect of grain boundaries. We can use the reciprocity theorem in Eq. (2.148) to obtain plots of ηc (z) either from three-dimensional device simulators or from analytical approximates. Dugas solved the equation for the excess minority carrier density for a columnar grain

89


1.0 0.8 Ln,eff [µm]

90

0.6 0.4 0.2 102

g = 0.3, 1, 3 µm 103

104

105

106

107

−1

Sgb [cm s ] Figure 2.34 Effective diffusion length under the influence of grain boundary recombination calculated after Eq. (2.154) for Ln,a = 1 µm and Dn,a = 1 cm2 s−1 . Approximation of square shaped grains without back surface recombination. With increasing grain size g, the effect of Sgb on Ln,eff becomes reduced.

on a square shaped base by a Fourier-like decomposition of the excess minority carrier density [69, 70]. Examples for the collection function in the presence of grain boundary recombination can be found in Ref. [65]. Brendel and Rau derived an approximate expression for an effective diffusion length for the case that the absorber is sufficiently thick to neglect back surface recombination Ln,eff = (Ln,a −2 +2Sgb /Dn,a g)−1/2

(2.154)

Here, g denotes the grain size. Plots of Ln,eff versus Sgb in Figure 2.34 show that the effective diffusion length is sensitive to grain boundary recombination if the grain size is in the range of the bulk diffusion length Ln,a . 2.5.4.2 QNR with Graded Bandgap The collection of electrons being minority carriers in the QNR of a p-type absorber may be supported by a gradient in the electron affinity χ, that is, −dχ/dz > 0. In this case, the expression for the current density in Eq. (2.9) includes the second and third term and the one-dimensional diffusion equation (Eq. (2.149)) has to be modified to

Dn,a

d2 χ n(z) d2 n(z) µn,a dχ dn(z) µn,a − n(z) 2 − − = −G(z) (2.155) 2 dz q dz dz q dz τn,a

If we only allow a constant gradient of χ, the third term in Eq. (2.155) can be neglected. We can apply the generalized reciprocity theorem [71] which allows solving the homogeneous equation for ηc instead of solving Eq. (2.155): d2 ηc (z,V) ηc (z,V) 1 dχ dηc (z,V) − − =0 dz2 kT dz dz Dn,a τn,a

(2.156)


1

8 6

ηc

4 2

-(Ln,a / kT)× (dχ / dz) = 0,1,2,5,10

0.1

8 6

-da

-wa0

de

z [µm] Figure 2.35 Collection function in the absorber with Ln,a = 1 µm and Sn0,bc = 107 cm s−1 . The back surface gradient of the electron affinity (given in units of 1/Ln,a ) is varied.

The boundary conditions are identical with Eq. (2.151), we find the solution ηc = S

n0,bc Dn,a

S −k2 exp{k1 (wa + z)}− Dn0,bc − k1 exp{k2 (wa +z)}exp{(k1 − k2 )(wa − da )} n,a S S n0,bc n0,bc − k − k − exp{(k1 − k2 )(wa − da )} 2 1 Dn,a Dn,a (2.157) )

with k1,2 =

1 dχ 2kT dz

±

dχ

2

+L12 dz n,a The effect of a back surface gradient is to support carrier collection from the bulk of the absorber. This is particularly effectful for devices with small Ln,a (see Section 3.10). Figure 2.35 shows ηc (z) calculated after Eq. (2.157) for different effective force fields induced by a gradient in χ. The value of −(Ln,a /kT)(dχ/dz) = 10 corresponds to Ec = 0.4 eV for an absorber with thickness da = 1.5 µm. In Figure 2.35 we see the positive effect on the collection function and we may expect an increased Jsc . However, due to the bandgap gradient the generation function has to be modified since the bandgap gradient induces an effective bandgap enhancement. This is a negative effect on Jsc . The trade off between both effects is discussed in Section 3.10. 1 (2kT)2

2.5.4.3 QNR with Back Surface Field If the absorber bulk includes a back surface gradient in the form of, for example, a back surface field (gradient of the electric potential), the collection function in the remaining QNR and in the gradient region will have to be modified. Equation (2.152) points out the necessary changes in the QNR: in the collection function, we have to replace the absorber thickness da by an effective thickness da − wbsg where wbsg is the width of the back surface gradient region (BSGR). Further, we set Sbsg to a small value, provided the back surface gradient is large enough to establish a sufficient minority carrier mirror. This defines the collection function

91

92


at the position da − wbsg , that is, at the beginning of the BSGR

Sbsg da − wbsg − wa da − wbsg − wa −1 ηc da − wbsg = Lp,a sinh + cosh Dp,a Lp,a Lp,a (2.158) If the back surface gradient is nearly constant in the BSGR, the collection function will take on a constant value as given by Eq. (2.158). Close to the back contact, ηc (z) will drop again due to back surface recombination of electrons. 2.5.4.4 Absorber Space Charge Region Due to the small density of recombination partners, recombination of photo generated carriers in the SCR can often be neglected. In this case the collection function in the SCR is close to unity. For a high defect density and thus a small carrier lifetime, this approximation is not valid and recombination in the SCR cannot be neglected. The exact differential equations for electrons and holes in the SCR are derived from Eqs. (2.9−2.14) and read

d2 n(z) dn dF + qµn,a n = R(z) − G(z) + qµn,a F 2 dz dz dz dp dF d2 p(z) − qµp,a p = R(z) − G(z) Dp,a − qµp,a F 2 dz dz dz

Dn,a

(2.159)

These equations require numerical solutions as they are nonlinear in the case of SRH recombination. For photonic excitation Reichman calculated the wavelength dependent external collection efficiency [72]. If the drift term and the gradient of J can be neglected, one can use the Hecht equation [73]. However, we will use a different approach and will make use of Eq. (2.147). Using device simulation and a -function like generation profile, we can derive η(z) in the SCR for a generation density similar as under AM1.5 condition (see details below). We note that the validity of Eq. (2.147) does not imply that the reciprocity theorem is valid in the SCR. We further note that due to the nonlinearity of Eq. (2.159), a simulated η(z) is only valid for the selected generation density (AM1.5 equivalent). Figure 2.46a shows simulated functions of η(z) (AM1.5) for different absorber lifetimes where the interface recombination velocity is fixed to Sn0,p0 = 105 cm s−1 . The width of the SCR, that is, the position −wa has been extracted from device simulation but is in good agreement with the value obtained after Eq. (2.36). The figure shows that for a lifetime of about 10 ns, the assumption of η(z) = 1 in the SCR approximately is correct. For decreasing lifetime values, first there is a deviation from unity at the edge of the SCR (−wa ) and finally also at the absorber/buffer interface (z = 0). The curve with τn0,p0 = 10 ps is much smaller than unity in the complete SCR. Thus, the photo current generated in the SCR, which is calculated by folding η(z) with the generation function according to Eq. (2.147), is much smaller than in the case of τn0,p0 = 10 ns. This shows up as an external quantum efficiency (EQE) curve which is much below 100% in the complete wavelength range (see Section 2.5.5).


The edge of the SCR defines the boundary condition for calculation of η(z) in the QNR. We see in Figure 2.46a that for small lifetime η(−wa ) = 1 is not fulfilled. This must be considered in the calculation of the collection function in the QNR, for example, by using a scaling factor. A further impact on carrier collection in the SCR has the absorber/emitter interface. We consider the band diagram under short circuit as given in Figure 2.39a. In the p-type absorber layer with inverted surface, holes are minority carriers near the heterojunction. They are driven by the electric field toward the edge of the depletion region where they become majority carriers but also by their concentration gradient toward the heterojunction where they recombine. In Figure 2.39a, we see Jp having a small negative peak directly at the absorber/buffer interface. This negative peak would increase upon decreased absorber doping and increased interface recombination velocity. Since recombination of holes at the interface is a loss mechanism, the collection function bends downwards close to the interface as is depicted in Figure 2.36b. The larger Sn0,p0 , the smaller is η(z) close to the interface at z = 0. The effect on the photo current Jph (V) is strong as the generation rate close to z = 0 is high (see Figure 2.32). 1.0 τ n0/p0,a =10 ns 0.8 ηc (z)

1 ns 0.6 0.4 0.2

τn0/p0,a =10 ns 100 ps 10 ps τn0/p0,a =100 ps

(a) 0.0 (c) 1.0

-wa(0 V)

0

z

ηc (z)

0.8 0.6

Sn0,p0 =103, 105, 107 cm s−1

0.4 0.2 0.0

(b)

0

-wa z

Figure 2.36 (a) Collection function of an absorber/buffer/window heterostructure in the SCR where the carrier lifetime is the parameter and the interface recombination velocity at the absorber/buffer interface is fixed at Sn0,p0 = 105 cm s−1 . (b) Collection function in the SCR and in the buffer layer where the interface recombination velocity at the absorber/buffer interface is the parameter

and the lifetime is fixed at τn0/p0,a = 10 ns. (c) Collection function at fixed Sn0,p0 = 105 cm s−1 and at τn0/p0,a = 10 ns, 100 ps for zero forward bias (full line) and 0.5 V forward bias (dotted line). All other device parameters as in the default Table 8.1. Data from device simulation as described in Section 2.5.4.

93

94


Naturally, the electric field under forward bias also has an effect on η(z). The trivial effect is that in forward bias the SCR becomes smaller. This is seen in Figure 2.36c for the example of τn0/p0,a = 10 ns. For shorter lifetimes, however, also the maximum of η(z) becomes reduced. This shows up in the JV curve of the cell as a small fill factor (FF). 2.5.4.5 Buffer Layer In a heterostructure of n-type buffer/window/p-type absorber, holes are minority carriers in the buffer layer. The collection function in the buffer layer is described by the collection probability of these holes. Principally, they can undergo bulk recombination as well as interface recombination. Since buffer layers normally are very thin and may also be depleted from majority carriers, interface recombination may well dominate over bulk recombination. Rothwarf proposed a simple formula which relates the interface recombination velocity and the drift velocity of minority carriers at the interface of a heterojunction [74]. Originally, this formula has been developed for interface recombination in a non-inverted absorber. In Figure 2.37 we depict an absorber/buffer/window heterostructure with interface states. A current density Jgen is due to minority carrier holes generated in the buffer. The fraction Jel loss of this current is lost via interface recombination while Jph can pass the interface. A collection probability ηc can be defined according to

ηc (V) =

Jph (V) Jph (V) + Jloss el (V)

(2.160)

The external current density can be calculated from TE theory according to Jph (V) = qµp,b pb (V)Fz=0 (V). The loss current Jel loss is obtained from Eq. (2.105) with RIF from Eq. (2.67) and reads Jel loss = qSp0 pa (V). Thus, ηc (V) can be written as µp,b Fz=0 (V) ηc (V) = (2.161) µp,b Fz=0 (V) + Sp0 ηc (V) is a constant within the buffer layer. A model which takes into account also a position independent bulk recombination probability has been presented Ec

Ev

Jelloss

Jph

E

Jgen 0 db Position z

Figure 2.37 Schematic band diagram of an absorber/buffer/window heterostructure containing interface states under illumination in short circuit condition. Jph denotes the photo current due to holes generated in the buffer layer or interface near zone of the absorber, Jel loss describes the current loss due to interface recombination.


in Ref. [75]. Figure 2.36b gives examples of the collection function in the buffer (0 < z < db ) for an absorber/buffer/window heterostructure. This being our default device exhibits interface recombination both at the absorber/buffer interface as well as at the buffer/window interface. Thus, Eq. (2.161) certainly is by far to simple. Nevertheless, we find that ηc (V) in Figure 2.36b has a small plateau in the buffer and only bends down if approaching the buffer/window interface. The value of µp,b Fz=0 in this device is 1.1 × 106 cm s−1 . Accordingly we find that, for Sn0,p0 = 107 cm s−1 , ηc (V) is very small in the buffer and, for Sn0,p0 = 105 cm s−1 , ηc (V) has moved toward larger values. If the buffer layer thickness can be varied in a wide range without changing the buffer layer properties, Eq. (2.161) can be a means to estimate the interface recombination velocity. 2.5.4.6 Simulating the Collection Function The collection function can be simulated using a device simulation program and with aid of Eq. (2.147). We apply a specific generation function of the type G(z) = G0 + (z − z ), where the delta function in practice is a narrow distribution at the point z and G0 is a constant generation rate similar as under AM1.5 illumination. Introducing the specific generation function into Eq. (2.147) yields

Jph (V) = −q

db −da

G0 ηc (z,V)dz − q

db

−da

δ(z − z )ηc (z,V)dz

(2.162)

Identifying Jph (V) with the integral over G0 ηc (z, V) Eq. (2.162) allows to calculate ηc (z , V) = Jph (V) − Jph (V). The currents Jph (V) and Jph (V) result from two separate simulations. Figure 2.38 shows the simulated collection function of the default device of absorber/buffer/window type at zero voltage bias. The function ηc (z) falls off to zero in the first nanometers of the window layer due to small lifetime therein and the finite interface recombination velocity at the buffer/window interface.

1.0

ηc (z)

0.8 0.6 0.4 0.2 0.0

-da

-wa

0 db

db +dw

z

Figure 2.38 Simulated collection function of an absorber/buffer/window device with parameters as in default Table 8.1.

95

96


2.5.5 Quantum Efficiency and Charge Collection Efficiency

For the case of photonic excitation, we want to derive a correlation between the collection function ηc (z, V) and EQE(λ, V). Both quantities can be voltage dependent. By use of the normalized generation function Eq. (2.147) reads λg,a djγ (λ) db n Jph (V) = −q G (λ,z)ηc (z,V)dzdλ (2.163) dλ −da λg,w In contrast, the QE is defined as the change in the charge carrier flux density −dJph (λ, V)/q per change of photon flux density at a given wavelength. EQE(λ, V) = −dJph (λ, V)/qdjγ (λ)

(2.164)

Equation (2.164) describes the wavelength dependent response of the solar cell: how many charge carriers are driven toward the external circuit for a certain number of incident photons. The definition in Eq. (2.164) can be rewritten as λg,a djγ (λ) EQE(λ, V)dλ (2.165) Jph (V) = −q dλ λg,w The QE can be expressed as a convolution of the normalized generation rate and the collection function d b Gn (λ,z)ηc (z,V)dz (2.166) EQE(λ,V)= −da

We remind that the EQE and internal quantum efficiency are related via the absorbance, that is EQE(λ, V) = IQE(λ, V)A(λ)

(2.167)

If the excitation of charge carriers in a solar cell is accomplished by electron bombardment, we speak of EBICs. Excitation by an electron beam is done, for example, for testing purposes. A charge collection efficiency can be defined as de Gn (E,z)ηc (z,V)dz (2.168) (E,V) = −da

n

where G (E, z) is the spatial dependent generation rate for a given electron beam energy (Eq. (2.141)). We see that again the property of the device to collect charge carriers is described by ηc (z, V). The QE in Eq. (2.166) and the charge collection efficiency in Eq. (2.168) can be considered as the Laplace transform of the collection function. If the generation functions are known and EQE(λ) or (E) have been measured at a certain bias voltage V , an inverse Laplace transformation will render the collection function η(z, V ). This has been demonstrated for optical [76] and electron beam excitation [77]. The concept of the collection function is particularly useful if both QE measurement and EBIC measurement have been performed on the same sample [78].


2.5.6 Barriers for Photo Current

Similar as a barrier for the diode current (see Section 2.4.7), the flow of photo current may be impeded by transport barriers. A back contact barrier does not block the photo current because this barrier diode is put under forward bias by the main diode. However, at the front contact there may be a barrier for electron transport imposed by a band discontinuity, a bandgap gradient or a specific doping profile. In the case of a conduction band discontinuity, the band diagram may be similar as in Figure 2.29b, however, the drop in EFn will be negative, that is, EFn < 0. The photo current generated in the absorber layer has to surmount the energy barrier Ec b,a . Replacing φb n in Eq. (2.126) by En,az=0 + Ec b,a and using naz=0 = Nc,a exp{−En,az=0 /kT}, it follows J0,TE = 1/4qνn Nc,a exp{−(Ec b,a + En,az=0 )/kT} = 1/4qνn naz=0 exp{−Ec b,a /kT} (2.169) as the limiting current density [57]. Basically, exp{−Ec b,a /kT} is a Boltzmann term for the TE over the conduction band spike. As long as the generation current17) is much smaller than the limiting current given by Eq. (2.169), the barrier will have only negligible influence on the photo current. For a given energy barrier, the free parameter of Eq. (2.169) is the carrier density naz=0 . The photo current can pass the absorber/buffer interface exhibiting a certain energy barrier Ec b,a in case that naz=0 is sufficiently high. This means for a p-type absorber that the electrons must be majority carriers at the interface. (For design rules, see Section 3.3.) If the generation current Jgen reaches the limiting current J0,TE , carrier transport is impeded and minority carriers accumulate at the absorber surface. This can be seen in Figure 2.39b. Due to the accumulation of electrons at the interface, the concentration gradient in the absorber is negative at all positions in the absorber bulk. This changes the collection function in the absorber bulk and leads to higher recombination in the bulk. We see that the electron related current density in Figure 2.39b is negative in the QNR of the absorber. A barrier for the photo current may also be established by a gradient of the bandgap at the front of an absorber. If the bandgap gradient is realized by lifting the conduction band edge close to the interface as in Figure 2.14c, the case is very similar as for a conduction band spike: only electrons with sufficient kinetic energy can pass this barrier. If the bandgap gradient realized by lowering the valence band edge of the absorber toward the interface (cf. Figure 2.14c), a region of dEFn /dz < 0 will form as described in Section 2.3.7. This will also reduce the photo current although we may not relate it to a photo current barrier. 17) For this argument we take the generation

current as a place holder for a ‘‘possible photo current.’’ What we mean is the photo

current which could flow without a photo current barrier.

97

98


-qϕ

1

-qϕ

Energy E [eV]

0 −3

-qϕ

-qϕ0

EFn

Ec

−4 −5

Ev EFp

−6

J [mA cm−2]

n,p [cm−3]

−7 1018 1016 1014 1012 1010 20 10 0 −10 −20 −30

p n p n

Jn Jn Jtot Jn

−1.5 −1.0 −0.5 (a)

p n

Jtot

Jp Jtot 0.0

Jp

Jp 0.5

−1.5 −1.0 −0.5 (b)

0.0

Position z [µm]

Figure 2.39 Solution curves for absorber/buffer/window heterostructures (default device from Table 8.1) under AM1.5 illumination. (a) Without photo current barrier and (b) with a photo current barrier of 0.5 eV due a conduction band offset at the

0.5

−1.5 −1.0 −0.5

0.0

0.5

(c)

buffer/absorber interface; and (c) with a p+ layer in the absorber front. Note the different scales of the current density. The potential energy without illumination, −qϕ0 , is given for comparison.

Finally, we treat the case that the absorber exhibits a p+ region next to the heterointerface [79]. In Figure 2.39c the large negative charge in the p+ layer leads to a large potential drop in the buffer and a small potential maximum in the absorber. If we introduce additional donor states at the buffer/p+ absorber interface, we can avoid the potential drop in the buffer – the principal barrier effect, however, would be similar. The p+ layer can be viewed as a front surface barrier for minority carriers. Such a barrier impedes the flow of electrons to the collecting interface. Electron current is only positive in the very interface near region but becomes negative for larger z (see Figure 2.39c). A solar cell with such kind of barrier has a high Voc – due to the large barrier for electron injection – but a small Jsc due to high recombination. The JV curve under illumination exhibits a kink which marks the transition from full photo current under reverse or moderate forward bias to a barrier reduced photo current for more positive bias. (In the example in Figure 2.39c, the kink is found at reverse bias.) If the kink is found under forward bias it leads to a low fill factor of the cell.


2.5.7 Voltage Dependence of Photo Current

In the preceding sections, we have allowed the collection function and the QE to be voltage dependent. A voltage dependent collection function will cause the photo current to be voltage dependent. This is particular of importance in the fourth quadrant of the JV curve, that is, under forward bias. A reduction of the photo current in the forward direction decreases the FF. Up to now, we encountered several mechanisms as possible origins for a voltage dependent photo current. These are variations with voltage of the SCR width, interface recombination, and photo current barrier. 2.5.7.1 Width of SCR A typical collection function of the absorber is close to unity in the SCR and declines in the QNR (see Figure 2.36). The collection width of photo induced carriers, Lcoll , may be approximated by wa + Ln,a , where wa is the SCR width and Ln,a the electron diffusion length. Due to an applied forward voltage, wa decreases [see Eq. (2.36)]. Generally, the photo current increases with reverse bias and decreases with forward bias. If the diffusion length is very small, then Lcoll ≈ wa and the influence of the applied voltage is large. Vice versa, if Ln,a wa , the changing SCR width has almost no influence on Lcoll and the voltage dependence of Jph is small. Thus, knowing the SCR width at zero bias and knowing the collection length, one can estimate the contribution of the SCR width on the photo current. A voltage dependent width of the SCR is diagnosed by voltage dependent QE measurements. For the absorber/buffer/window heterostructure Eq. (2.166) can be written as d b EQE(λ,V) = ηc (z)αb (λ)exp[ − αb (λ)z]dz T(λ)[1 − R(λ)] 0 0 −wa + αa (λ)exp[ − αa (λ)z]dz + ηc (z)αa (λ)exp[ − αa (λ)z]dz (2.170) −wa

−da

where the integrals give the contributions of the buffer layer, the absorber SCR (assuming ηc = 1) and the absorber QNR, respectively. We select a wavelength for which absorption in the buffer is about zero and assume Ln,a < da . In that case, the collection function after Eq. (2.152) can be approximated by ηc (z) = exp{(z + wa )/Ln,a } and Eq. (2.170) reduces to exp{ − αa (λ) wa (V)} EQE(λ,V) = T(λ)(1 − R(λ)) 1 − (2.171) 1 + αa (λ)Ln,a This is the well known Gärtner formula [80]. We consider the wavelength dependence of the EQE as a result of voltage dependent SCR variation. For short wavelengths with αa (λ)wa (V) 1, the EQE after Eq. (2.171) approaches unity and thus does not depend on the voltage. For longer wavelength, the variation of EQE with voltage increases. Thus, measuring the EQE at different voltages and

99


EQE/EQE (0V)

100

1.2

−0.5 V

1.0

+0.5 V

0.8 400

600

800

1000

Wavelength [nm] Figure 2.40 Simulated ratios of the external quantum efficiency (EQE) as a function of wavelength of an absorber/window heterostructure with Ln,a = 10−10 s in the absorber. Other parameters as in the default Table 8.1. Solid line: ratio at reverse bias, that is, EQE(−0.5 V)/EQE(0 V). Dashed line: ratio at forward bias, that is,

EQE(+0.5 V)/EQE(0 V). Charge carriers generated by deeply penetrating photons (long wavelength) exhibit a relatively higher collection probability at reverse bias due to the extended SCR width. Vice versa, at positive bias charge carriers generated by red photons suffer relatively more from the reduced SCR width.

normalizing to the EQE at zero voltage gives wavelength dependent functions as demonstrated in Figure 2.40. 2.5.7.2 Interface Recombination We consider an absorber/buffer/window heterostructure. The example collection function in Figure 2.36 is influenced by interface recombination in two ways. Holes photogenerated in the inversion region of the absorber and holes photogenerated in the buffer layer will partially undergo interface recombination. With applied forward bias, the electric field at the junction decreases and interface recombination increases. Thus, the photo current decreases with forward bias and increases with reverse bias. A good example of a voltage dependent collection function due to interface recombination is a p+ /n absorber/window heterostructure as the Cux S/CdS solar cell. Here, we find a large En,az=0 and high interface recombination velocity Sn0,p0 but it is Ec a,w = 0. In this cell there is competition between interface recombination and current sweep through the junction region [74]. If at higher forward bias, the electric field is reduced, interface recombination increases. 2.5.7.3 Photo Current Barriers A photo current barrier can lead to a voltage dependent photo current. We consider a conduction band barrier such as the one depicted in Figure 2.39b. In this example, Ec b,a is so large (0.5 eV) that the generation current exceeds J0,TE in Eq. (2.169) already at zero voltage. J0,TE was limited by a too small electron concentration naz=0 for the large value of Ec b,a . If the band diagram is such that Ec b,a is slightly smaller, the concentration naz=0 at the interface may be sufficient to allow photo current transport at zero bias. But the electron concentration at the interface decreases with forward bias. Consequently, the photo current may become voltage dependent.

2.6 Device Analysis and Parameters

All mechanisms described above have in common that the photo current increases with reverse bias and no voltage for which the photo current saturates can be given. Thus, we cannot define a unique reverse bias voltage where, for all types of cells, the photo current show a maximum. In order to define a voltage dependent external collection efficiency η(V) it is suggested to set η(0) = 1. Thus, we have Jph (V) = Jsc η(V)

(2.172)

Comparison with Eq. (2.147) reveals that the relation between the external collection efficiency η(V) and the collection function ηc (z, V) is given by d b 1 ηc (z,V)G(z)dz (2.173) η(V) = −Jsc −da


In this section, we discuss some aspects of device analysis which are most relevant for heterojunction solar cells. The goal is to determine the basic device parameters from a limited number of electric measurements. Therefore, it is instructive to consider the equivalent electric circuits of heterostructure solar cells at the beginning of this section. 2.6.1 Equivalent Circuits

The thin film solar cell and module can be described by equivalent circuits of electric components. A circuit model shall reflect the response of a solar cell upon voltage and light stimulation. It is useful to discriminate between DC and AC stimulation. While DC stimulation includes the normal solar cell operation, AC stimulation is merely for the purpose of device analysis. First, we concentrate on single heterojunction solar cells. Module aspects are discussed in the last part of this section. 2.6.1.1 DC Equivalent Circuit An ideal heterojunction solar cell which obeys the principle of superposition is described by only two electric components: An ideal current source which delivers the photo current and a diode. The current source is ideal in the sense that the current delivered does not depend on voltage. The real solar cell deviates from the ideal cell in several respects. First, the photo current may depend on voltage (see Section 2.5.7). In order to preserve the superposition principle, we allow the current delivered by the current source to depend on voltage in a nonlinear (but known) fashion. In Figure 2.41a this is expressed as Jph (V). Also the diode may be not ideal – in the sense that parallel processes are carrying the recombination current. In Figure 2.41a this would

101

102

2 Thin Film Heterostructures Vm Jph(V)

I

III

Vb

GJ

DBC D

GBC

Rs J

Gsh

SCLC

Gsh

Rs

xcell SCLC

V

IV

x

(a)

y

(c)

I RBC

RBC

GPI

Cell n

Interconnect

(b)

II

Rw,BC

RTCO

(d)

CBC

CJ

II Cell n +1

II

Figure 2.41 (a) Equivalent circuit for DC bias conditions comprising a network of a voltage dependent current source, diode, shunt resistance, series resistance, and space charge current limiter. (b) Equivalent circuit of a heterostructure solar cell under AC operation where the main diode and a possible diode at the back contact are simply replaced by conductances and capacitances. Note that the main diode may need to be described by a much more complicated equivalent circuit as depicted in

(e)

RBC

Figure 2.44. (c) Schematic plan view on part of two stripes of a solar module showing the three scribe lines [interconnect scheme (a) in Figure 2.43]. Note that the image is not drawn to scale. The dashed rectangle shows one module element. (d) Module element with distributed series resistances of window and back contact, Rw and RBC . Series integration introduces the interconnect elements Rw,Bc and GP1 . (e) Parallel circuit of module elements in direction y contacted with the series resistor elements RBC .

mean that we have to replace the diode by a parallel combination of two or more diodes. The DC equivalent circuit of a non-ideal solar cell in Figure 2.41a is completed by a shunt conductance Gsh and one or more series elements. The shunt conductance is assumed to be ohmic. Possible series elements are depicted in the box in Figure 2.41a. An ohmic series resistance is a lumped quantity including ohmic losses in the QNR or at the contacts. Often also the sheet conductances of window and back contact are included in the series resistance. Here, we take a different point of view in that we consider distributed series effects due to sheet conductance as part of the module circuit. Therefore, Figure 2.41a only represents a small solar cell being the building block of a large network (the module). We call this building block, labeled as I in Figure 2.41, an elementary diode. Another series element may be the effect of current limitation by space charge limited current (SCLC). The SCLC can be expressed as JSCLC = kVB m where k and m are constants which can depend on the trap density in the SCR [81]. Thus,


the SCLC is nonlinear in the voltage. The SCLC element can dominate the JV curve of a device at high forward bias [82]. Finally, a series element can be a back contact diode which has the opposite polarity as the main diodes. Certainly, two or more series elements may occur in parallel and/or in series and may dominate at different voltage ranges. If a voltage V is applied to the circuit in Figure 2.41a, the part Vb drops at the element in series. It holds V = Vb + Vm where Vm is the voltage at the main junction. The total current is ( qVm N J= − 1 + Vm Gsh + Jph (Vm ) J0i exp (2.174) Ai kT i=1 where N is the number of parallel recombination paths (see Section 2.4.6). For practical purposes, often only one diode is considered. In case of an ohmic series resistance, the voltage drop at the series element is simply Vb = Rs J. As a result of the series element, Eq. (2.174) becomes implicit. Extraction of the diode parameters from experimental JV curves is discussed in Section 2.6.2. The case of a back contact diode has been discussed in Section 2.4.7. 2.6.1.2 AC Equivalent Circuit The AC equivalent circuit in Figure 2.41b is derived from the DC model. It does not include the current source (because of its infinite internal resistance) and describes the shunting diode as a combination of one capacitance and one conductance. In the case of a simple pn junction, the conductance stands for the dominant recombination path of the diode current. It can be expressed by the saturation current density as qJ0 qV dJ = exp (2.175) GJ (V) = dV AkT AkT

A similar expression as Eq. (2.175) can be formulated for the conductance of the back contact diode. The capacitance stands for charging and discharging of shallow states at the edges of the SCR. For a heterostructure solar cell, the AC equivalent circuit may be more complicated than the one shown in Figure 2.41b. We give examples of absorber/window and absorber/buffer/window heterostructures in Figure 2.42. A simple absorber/window heterojunction without interface states (Figure 2.42a) may be described by two conductances and capacitors each in parallel. Since recombination is limited to the absorber side, the conductance Gw expresses the conductance of the window depletion layer [83] while Ga is given by Eq. (2.175). In the case of an n+ p heterojunction, Gw and Cw can be neglected. Also, the absorber/buffer/window heterostructure (Figure 2.42e) can be described by a parallel circuit. Thus, measuring the admittance Y(ω) of the cases a,e allows to identify Re(Y) with G and Im(Y) with ωC, where G and C are the total conductance and capacitance, respectively. Charging and discharging of an inversion layer (Figure 2.42b) will add a further capacitance to the equivalent circuit. Its value depends on the reduced chemical

103

104


Gw

Ga

Ww G−1 w =

(a)

Ca

Cw

Ga

Gw

0

Ga =

CINV (b)

Cw

Ca

CINV =

Ca

Gd

(d)

q kT

Ca =

Ga (c)

Cd =

Cd

Ga

Gw GIF

Ca

CIF

σn,w (z)−1 dz

qJ0

exp

AkT

qVa AkT

2NCkT εIF exp

− En,a

z=0

2kT

ε0εa

Wa

q2 Ndf(Ed)(1 − f(Ed)), ω kT

ω0

CIF = qNIF , ω ω0

Cw db

Ga

Gb

G−1 b =

Ca

Cb

Cb =

σn,b (z)−1 dz 0

(e)

Ga

Gb

(f) Ca

Wb

Gw

GIFa,b

GIFb,w

CIFa,b

CIFb,w

Cb

ε0εb

Cw Gb

Ga

GIFa,b CIFa,b

Ca (g) Gd

Cd

Cb

Figure 2.42 Band diagrams and corresponding equivalent circuits for heterostructures exhibiting (a) voltage drop in window and absorber, (b) an inversion layer in the absorber, (c) a singular deep trap level in the absorber, (d) interface states,

(e) a buffer layer, (f) interface states at the buffer/window and buffer/absorber interfaces, and (g) a combination of bulk and interface defects. The right column lists some mathematical descriptions for conductances and capacitors.


potential of the electrons at the interface, En,az=0 (see Figure 2.42) [83]. An inversion layer in the absorber can be the outcome of a type I band alignment as depicted in Figures 2.42b. Even for high window doping, large band bending on the window side will impede charge transport to the interface. Thus, the conductance Gw is needed in the equivalent circuit. A discrete trap level at given energy Ed in the SCR of the absorber can be modeled by introducing the series connection of a capacitance Cd and conductance Gd . In the example of Figure 2.42c, we assume an n+ p heterojunction. The elements Cd and Gd are connected in parallel to the absorber conductance. For a discrete trap level, charging and discharging takes place at the location of intersecting trap level and hole quasi Fermi level. The capacitance Cd will depend on frequency and temperature. An equivalent circuit for a trap density energetically distributed in the absorber bandgap can be found in Ref. [84]. Interface states (d,f) can add a further capacitance to the equivalent circuit of a heterojunction. The kinetic limitation of their charging and discharging is accounted for by the conductance GIF . The capacitance simply depends on the energetic density of interface states if the interface state density varies slowly with energy [85]. Inversion layer, trap level, interface states, as well as a back contact diode will render an Im(Y) which is frequency dependent. Thus the equivalent circuit describing the response of the solar cell to an AC voltage stimulation will be frequency dependent. This is the basis of admittance spectroscopy. 2.6.1.3 Module Equivalent Circuit In a thin film module the elementary diodes of Figure 2.41a are interconnected. Thus, the module is characterized by a further DC equivalent circuit model. This model includes distributed series resistances and the integrated series connection. An AC equivalent module may be constructed in a similar fashion as the DC module by interconnecting the building blocks III in Figure 2.41b. Possible schemes for the integrated series connection are depicted in Figure 2.43. All schemes introduce a series resistance Rw,Bc . In addition, schemes Figure 2.43a,b introduce the shunt conductance GP1 into the module equivalent circuit. Due to its most widely use, we concentrate on scheme (a). Figure 2.41c depicts a schematic close-up view of a module where the current is flowing in the x direction. The three lines denote the scribe lines P1–P3 (from left to right). Each module cell has a width, xcell , and a length, ycell . A dashed rectangle in Figure 2.41c marks a single element of a module cell. We call this element (which is labeled as II in Figure 2.41) a cell element. The equivalent circuit of the cell element II is depicted in Figure 2.41d. We see that besides the elementary diodes I, the cell element shows distributed series resistances of the window and back contact. The influence of Rw is larger than that of RBC due to the normally lower conductivity of the window layer. Rw is one of the parameters defining the optimum length of the cell. Design rules for the cell width are presented in Section 3.12. The cell element also includes the contact resistance and shunt conductance of the interconnection scheme.

105

106


Window

P1

P2

P3 Absorber Backcontact

GP1

Rw/BC

GP1

(a)

Rw/BC

(b) RTCO

Rw/BC Filler (c)

(d) Contact strip

Substrate

Grid

(e) Figure 2.43 Possible schemes for the series connection of thin film solar modules. The monolithical interconnects (a–c) are realized by different scribe and filling lines. (a) Separated scribes P1–P3, (b) overlapping P1 and P2, (c) P1 scribe with isolation filling. All interconnect schemes will exhibit contact

resistance Rw,bc . In addition, a shunt conductance GP1 can result within the P1 scribe line of schemes (a) and (b). The shingle type interconnection scheme in (d) is suitable for conductive substrates. In (e) the classical wafer interconnect has been realized using contact strips.

In the y direction, we can also think of a rectangle defining a different cell element (rectangle IV in Figure 2.41c). This cell element IV includes elementary diodes connected by distributed series resistance similar as in Figure 2.41d but without interconnection. Thus in a module, each elementary diode is connected to other diodes in two dimensions by virtue of series resistances in the window and back contact. The diodes form a diode array which may exhibit a fluctuation of electronic properties in two dimensions [86] (the influence of weak diodes on the module performance is discussed in Section 2.4.9). If there is no series interconnection, the array is cylinder symmetric. A weak elementary diode would be decoupled from other diodes in all directions by the relatively low conductance of the window layer. Due to the series interconnection, however, the module is not cylinder symmetric. Figure 2.41e gives an equivalent circuit of a series of cell elements II which are connected in the direction y by distributed series resistances of the back contact. Due to the normally higher conductance of the back contact, the influence of a locally weak diode on the diode array is larger than if there is no series interconnection.


2.6.2 Current–Voltage Analysis

Including series resistances and shunt conductances, the total current as a function of voltage under illumination is given by Jlight (V) = J0 (exp{q(V-Jlight Rs )/AkT}-1) + (V-Jlight Rs )Gsh + Jsc η(V) (2.176) Here, J0 is the saturation current density for the recombination mechanism under illumination. η(V) is the external collection efficiency. 2.6.2.1 External Collection Efficiency The external collection efficiency is an important function. Without knowing this function, it is not possible to apply the superposition principle which is so important for device analysis. The external collection efficiency is defined in Eq. (2.172). Initially we allow the external collection efficiency to be both voltage and illumination dependent. In the end, we hope that the external collection efficiency is only dependent on voltage at least on the relevant illumination level (100 mW cm−2 ). The photo current Jph (V, jE ) as a function of voltage and photon flux is a construct which renders the superposition principle applicable. A necessary condition for the following procedure is that the short circuit current density is linear in jE . Fortunately, this is fulfilled for most chalcogenide thin film solar cells in the range jE 10 − jE 100 (remember that jE x denotes an energy flux of density x mW cm−2 ). Thus, Jph (V, jE ) = Jsc η(V, jE ) where Jsc is given by Eq. (2.165). η(V, jE ) is the external collection efficiency. Possible reasons for the external collection efficiency being voltage dependent are discussed in Section 2.5.7. Jsc is easily measured and it remains the problem of how to determine η(V, jE ). We show how to solve this problem under the prerequisite that η(V) does not depend on jE . (If there is a barrier for the photo current this assumption may not be valid; see Section 2.5.6.) Consider two JV curves, Jlight (V, jE n ) and Jlight (V, jE m ), measured at two different energy fluxes, jE n and jE m [87]. By use of Eq. (2.179) the difference current reads

Jlight (V, jE n ) − Jlight (V, jE m ) = [Jdiode (V, jE n ) − Jdiode (V, jE m )] + [Jsc n η(V) − Jsc m η(V)]

(2.177)

and a tentative external collection efficiency may be calculated from ηnm (V) =

Jlight (V, jnE )−Jlight (V, jm E) Jnsc −Jm sc

(2.178)

The mathematical form of Eq. (2.178) provides that η(V = 0) = 1.18) Certainly, ηnm (V) only is the ‘true’ external collection efficiency if Jdiode (V, jE n ) = Jdiode (V, jE m ). 18) In [88], η = 1 has been set at the maxi-

mum of Jlight (V, jnE ) − Jlight (V, jm E ). For some chalcogenide cells which show a reverse

bias shunt under illumination, the definition η(0) = 1 is more useful.

107

108


Otherwise, ηnm (V) reflects both the variation of the voltage dependence of the collection efficiency and the voltage dependence of the difference Jdiode (V, jE n ) − Jdiode (V, jE m ). By measuring at energy fluxes, jE m,n,p,q , one may find an illumination range where plots of ηnm (V) and ηqp (V) are congruent. In this illumination range it can be assumed that ηnm (V) = ηqp (V) = η(V) and therefore Jdiode (V, jE n ) = Jdiode (V, jE m ). Possibly, this is at a high illumination level (close to AM1.5G) where the recombination mechanism has switched from dark to illuminated. It may be worthwhile to determine the collection function with jE 100−x , j100 , and jE 100+x where E x is small. The determination of η(V) is unaffected by a shunt conductance as long as the shunt is illumination independent. If η(V) is not independent on jE , that is, η = η(V, jE ), we will not likely find a combination of energy fluxes jE m,n,q,p where ηmn (V) and ηpq (V) are congruent. 2.6.2.2 Diode Parameters Current versus voltage measurements at different temperatures are the primary means to elucidate recombination processes and the energy band diagram of a solar cell. In Section 2.4 it was shown that all diode currents discussed in this book can be described by the three diode parameters: A, Ea , and J00 :

• The size of the diode quality factor, A (see Table 2.1), gives information about the location of dominant recombination. • The activation energy, Ea , should approximately equal the minimum bandgap in the absorber including its interfaces unless the Fermi energy is pinned at the absorber/emitter interface (Table 2.3). • The reference current density, J00 , is related to the carrier lifetimes (Table 2.2). J00 and Ea are combined in the saturation current density J0 . Further device parameters which are critical for the device performance are series resistance and shunt conductance. In this section, we show how to extract the diode parameters from experiment and how to avoid possible artifacts and pitfalls. In principle, one is interested in the cell parameters which are derived from an illuminated JV curve. This is because only parameters governing the device under AM1.5G illumination are relevant for the solar cell operation. The total current as a function of voltage measured under AM1.5G illumination and at a fixed temperature can be written as the sum Jlight (V, jE 100 ) = Jdiode (V, jE 100 ) + Jph (V, jE 100 ) + (V − Jlight Rs )Gsh

(2.179)

−2 where j100 E is the solar (or solar simulated) energy flux density of 100 mW cm 100 100 with an AM1.5G spectral characteristic [89]. Jdiode (V, jE ) and Jph (V, jE ) are the diode and photo currents as a function of voltage. Jdiode (V, j100 ) contains the wanted E information about the recombination mechanism. For the sake of generality, we allow Jdiode (V) = Jdiode (V, j100 ). If Jdiode is not illumination dependent, that is, the E recombination mechanisms in the dark and under illumination are identical, evaluation of the dark current Jdark = Jdiode (V) + (V − Jdark Rs )Gsh is preferable.

abs(J)

100

(a)

1.0 h(V)

10−1

h(V) 0.8

10−2

Jdiode, Jlight - Jsch(V)

Jdiode, Jlight - Jsch(V)

0.6 10−3 0.4

10−4

0.2

10−5 10−6

109

−0.5

0.0

−0.5

0.5

V [V]

Figure 2.44 Simulated JV curves for absorber/window heterostructures with carrier lifetimes (a) τn0,p0,a = 10−7 s and (b) τn0,p0,a = 10−10 s. Dark and AM1.5G illuminated JV curves (––) as well as a Jsc (Voc ) plot (open circles, left axis). The data point

(b)

0.0

0.5

0.0

V [V]

for jE 100 is given as a full circle. Calculated external collection efficiency η(V) (right axis). A series resistance of Rs = 1 cm2 and a shunt conductance of Gsh = 1 mS cm−2 have been included. Other device parameters as in the default Table 8.1.

With the knowledge of η(V), we can replace Jph (V, j100 ) in Eq. (2.179) by Jsc η(V), E which leads to Jdiode (V, jE 100 ) = Jlight (V, jE 100 ) − Jsc η(V) − (V − Jlight Rs )Gsh

(2.180)

Now, we can extract the recombination parameters by curve fitting. That this is not an easy task is illustrated in Figure 2.44. Here, we plotted simulated abs(J) versus V curves under illumination and in the dark for two devices with different absorber carrier lifetimes. The simulated Jlight have been corrected by −Jsc η(V). Both devices are simulated with a 1 mS cm−2 shunt conductance and 1 cm2 series resistance. For the smaller lifetime of τn0,a = 100 ps, the drop of η(V) is more pronounced. Obviously, for both devices the curves (Jlight + Jsc η(V)) and Jdark are subject to the series resistance Rs . There is almost no linear part of the ln[Jlight + Jsc η(V)] and lnJdark curves which can be used to extract A and J0 . From Figure 2.44 it is apparent that a correction of the (Jlight + Jsc η(V)) or Jdark data is necessary. So far, Rs and Gsh are unknown in Eq. (2.176). The first step is to determine the shunt conductance in the dark and under illumination from dη(V) dJ dJ Gsh,D = dark and Gsh,L = L − Jsc (2.181) dV dV dV V=0

V=0

V=0

Note that dη(V)/dV is negative. Also Jsc is a negative quantity. If the term dη(V) is omitted, the shunt conductance under illumination comes out Jsc dV V=0 larger than in the dark. This is an artifact. That Gsh is independent of illumination can be verified by comparing the two values of Gsh after Eq. (2.181). The next step is to determine Rs . Several techniques to determine the series resistance have been developed [90, 91]. Exemplarily, we follow the procedure

η(V)

abs(Jscη) [A cm−2]


110


proposed by Werner [92]. A corrected current density is calculated according to Jlight corr = Jlight (1 + Rs Gsh ) − Gsh V − Jsc η(V). In first order, we assume Rs = 0. A small signal conductance G = dJlight corr /dV is calculated from the experimental data. After some algebra, we obtain Rs G q 1 − G (2.182) = Jcorr AkT 1 + Rs Gsh L Thus, a plot of G/Jlight corr versus G gives a straight line with abscissa intercept (1 + Rs Gsh )/Rs and ordinate intercept q/AkT. Then, a second order approximation calculates Jlight corr including Rs . This procedure also applies to the dark JV curve. The impact of series resistances can completely be avoided by measuring pairs of Jsc and Voc at different jE . For V = Voc , Eq. (2.176) can be rewritten as −Jsc η(Voc ) = J0 exp{qVoc /AkT} + Voc Gsh

(2.183)

The simulated and corrected data [−Jsc η(Voc )] in Figure 2.44 closely follow the (Jlight − Jsc η(V)) curve except for the series resistance effect. The function ln(−Jsc η(V)) versus Voc can be evaluated by curve fitting where J0 , A, and Gsh are fit parameters. Graphically, the saturation current density is obtained from the extrapolation of the ln[−Jsc η(Voc )] curve at * Voc toward Voc = 0, that is, according to d ln[ − Jsc (Voc )η(Voc )] lnJ0 = ln[ − Jsc (Vˆ oc )η(Vˆ oc )] − Vˆ oc dVoc ˆ oc V = ln[ − Jsc (Vˆ oc )η(Vˆ oc )] −

qVˆ oc AkT

(2.184)

Vˆ oc lies on the linear part of the ln(−Jsc η(V)) curve. The other possibility is to calculate a corrected current density according to −Jsc corr = −Jsc η(V) − Voc Gsh and prepare a plot ln(−Jsc corr ) versus Voc . This plot yields a straight line with slope corr dln −Jsc q = AkT and with ordinate intercept lnJ0 .19) Figure 2.45a shows A factors dVoc determined in the dark and from Jsc (Voc ) plots for the two devices in Figure 2.44. Device 1 with τn0,p0 = 100 ps is limited by SCR recombination and accordingly it has A ≈ 1.9. We remind readers that, for SCR recombination, the deviation of the A factor from the value of 2 (derived in Section 2.4.5.1) is due to the electric field varying with applied voltage. Device 2 with τn0,p0,a = 100 ns shows a noticeable deviation between the diode quality factors from the dark JV curve and the Jsc (Voc ) plot. Due to the changing slope of the lnJdiode (V) curve, this curve is not easy to model (although it has been simulated and was not subject of experimental errors). Before we proceed to evaluate J00 and Ea , we investigate the error in the determination of the diode quality factor if the voltage dependence of the external collection 19) If the recombination mechanism changes

at high voltage bias (see, e.g., Figure 2.29), this change may not be reflected in the due to the limited range Jsc η(Voc ) curve

of Vˆ oc jE100 . It is recommended to use

solar simulated energy fluxes larger than j100 E in order to determine the slope at Vˆ oc jE100 . This slope yields the diode parameters which are decisive for the solar cell operation.

2.0 1.8

100 ps

1.6 1.4 1.2

100 ns

−20

100 ps

−30 100 ns

−40 −50 −60

1.0 200

(a)

In(J0 /A cm−2)

Diode quality factor A


240

280

320

T [K]

20 (b)

30 40 50 (AkT)−1 [eV]−1

60

Figure 2.45 (a) Simulated diode quality factors as a function of temperature for dark JV curves Jdiode (V) (––) and −Jsc (Voc ) curves (❡) for devices as in Figure 2.48. (b) Calculated values of ln(J0 ) versus (AkT−1 obtained by fitting simulated Jdiode (V) (––) and -Jsc (Voc ) (❡) curves.

efficiency is not taken into account. Let A and A be the diode quality factors obtained from the slopes of ln(Jlight + Jsc )(V) and ln[−Jsc (Voc )] curves determined at the positions Vˆ and Vˆ oc , respectively. In other words, A and A are from illuminated JV and −Jsc (Voc ) curves not corrected by η(V). It is not difficult to show [93] that these tentative factors A and A need to be corrected. It is AkT d ˆ ln η (V) and A = A η V 1− q dV ˆ V AkT d ln η (V) A = A 1 − (2.185) q dV Vˆ oc Typically, it holds that A > A > A ; and A deviates more strongly from A than A . This is the reason why evaluation of noncorrected Jlight + Jsc gives misleading results but a −Jsc (Voc ) evaluation without η(V) correction sometimes is sufficient. An exception is a cell with a back contact barrier where A becomes much smaller than A (see Section 6.1.6). With the knowledge of J0 and A determined at different temperatures it is possible to determine J00 and Ea by aid of Eq. (2.47). For the sake of generality, we assume that the diode quality factor could be temperature dependent. Table 2.1 reveals that this can be the case for recombination via a defect distribution or by tunneling enhanced recombination. Rewriting Eq. (2.47) yields ln(J0 /J00 ) = Ea /(A(T)kT)

(2.186)

All reference currents in Table 2.2 are only weakly temperature dependent. That is, none of the expressions for J00 does exhibit an exponential temperature dependence. Thus, a plot of ln J0 versus (A(T)kT)−1 20) approximately has the slope of the activation 20) Note that the correction of the kT scale by

A(T) appears necessary. A plot of AlnJ0 versus (kT)−1 is not appropriate for the derivation of Ea in case that A is temperature d dependent. Then, it is −1 A lnJ00 = 0. d (kT)

111

112


energy Ea and the ordinate intercept ln (J00 ). In Figure 2.45b we see that regardless of the carrier lifetime, all curves of ln (J0 ) exhibit the same slope corresponding to the bandgap of the absorber. This is expected since Ec is selected to zero and Fermi level pinning is excluded. Finally, we may summarize what we consider to be an appropriate procedure how to determine the device parameters of chalcogenide solar cells particularly in the case where η(V) shows a noticeable voltage dependence: 1) Determine the external collection efficiency η(V) from appropriate pairs of Jlight n (V), Jlight m (V). 2) Measure −Jsc (Voc ) at different jE and calculate −Jsc (Voc )η(Voc ) with η(Voc ) from step (1). 3) Determine Gsh from −Jsc (Voc )η(Voc ) or from Jdiode (V). 4) Obtain the light values of J0 and A from a fit of ln[−Jsc η(Voc )]. 5) Determine Rs from Jlight (V)η(V) or from Jdiode (V). 6) Obtain the dark values of J0 and A from a fit of Jdiode (V) 7) Determine A(T), J0 (T), Voc (T) from steps 1–6 performed at different temperatures. 8) Obtain Ea and J00 from a fit of lnJ0 versus [A(T)kT]−1 . It is instructive to finally compare the dark and illuminated values of Ea , A, and J00 . If these values greatly differ, the recombination mechanism may be illumination dependent. 2.6.2.3 Open Circuit Voltage By definition, the open circuit voltage Voc is the voltage the cell develops at zero load. Figure 2.46b,c shows different heterostructures under open circuit condition. Although, the partial current densities of electrons and holes can be large, the total current density Jtot (z) = Jn (z) + Jp (z) is zero everywhere in the device. The value of Voc is given by the difference between the quasi Fermi energy of holes at the absorber contact and the quasi Fermi energy of electrons at the emitter contact. Thus, it is

Voc = 1/q[EFn (de ) − EFp (−da )]

(2.187)

Within the absorber, the splitting of the quasi Fermi energies may be even larger than at the contacts. This can be recognized for the device in Figure 2.46c which exhibits the highest splitting of the quasi Fermi energies within the absorber. This maximum splitting of EFn and EFp cannot be translated into a measurable Voc . The electron quasi Fermi energy drops off toward the absorber/buffer interface due to high interface recombination. The splitting of the quasi Fermi energies can be sensed by evaluation of the spectral luminescence using Planck’s generalized law [94]. For not too high illumination (excess majority carrier concentrations being small compared to the dark concentrations), Voc can alternatively be expressed as L (dϕ/dz − dϕ0 /dz)dz (2.188) Voc = 0


Ev

1

(a)

1015 1010

Ev

1 0

p

n

Jtot

0.4

Jp

0.2

Jn

0.0 −1.5 −1.0 −0.5

J [mA cm−2] n,p [cm−3]


0

2

EFp

EFn

0.0

Position z [µm]

1015 n 1010 Jn 20 10 Jtot 0 Jp −10 −20 −1.5 −1.0 −0.5 0.0

(b)

Figure 2.46 (a) Default device at +0.6 V forward bias. (b) Default device under AM1.5G illumination at Voc . Device parameters as in Table 8.1. (c) Device with Ea,b c = −0.3 eV under AM1.5G illumination

3 2

EFp Ev

1

Position z [µm]

(c)

p 1015 n 1010 20 J Jtot p 10 0 −10 Jn −20 −1.5 −1.0 −0.5 0.0 Position z [µm]

at Voc . Other device parameters of (c) as in default Table 8.1. The steps in Jn,p at the interface are due to the dominating interface recombination limiting the Voc of the device in (c).

where ϕ and ϕ0 are the electric potentials under illumination and in the dark. Fonash [22] showed that, by use of Eqs. (2.5−2.12) and with the assumption of position independent mobilities, Eq. (2.188) can be rewritten as L dEg 1 L σn dEc σ 1 L σp dEc F0 dz − dz − − dz Voc = − σ q 0 σ dz q 0 σ dz dz 0 L L σp dlnNv σn dlnNc kT dσn 1 dσp kT − dz + − dz − q 0 σ dz σ dz q 0 σ dz dz (2.189) where F0 is the electric field under equilibrium, σn = qµn n and σp = qµp p are the excess conductivities at Voc . Equation (2.189) tells that there are different contributions to the open circuit voltage of a device. The first term on the right hand side of Eq. (2.189) is the contribution of the electrostatic potential. The second to fourth terms are the so-called effective force terms. These are contributions to the open circuit voltage due to variations of conduction band energy, energy gap, and density of states within the device. The fifth term is a diffusion term and is denoted as the Dember voltage [22].21) 21) The term ‘‘Dember voltage’’ may be mis-

leading since here it does not require

Ec

0 p


2

3

4

Ec

EFn Energy E [eV]

Energy E [eV]

EFp

3

4

Ec

EFn

Energy E [eV]

4

113

differences in the electron and hole mobilities.


-qϕ

1 Energy E [eV]

0

EFn

-qϕ0

3.0 EFp

2.0 1.0

F, F0 [V cm−1]

0.0 0 −2

F

F0

−4 −6 × 104

∆σn /σ

1.0 ∆σn,p / σ

∆σp /σ 0.5 0.0 105

d∆σp /dz

+ 100 ↑ 100 ↓

−

± d∆σn,p /dz [S/cm]

114

105 −0.8

d∆σn/dz −0.4 0.0 Position z [µm]

Figure 2.47 Default heterostructure of the absorber/buffer/window type under open circuit condition. Plots of potential energy, band edges, electric field, excess conductivities, and carrier gradients have been

0.4

obtained from device simulation. Device parameters as in the default Table 8.1. Note that positive and negative values of dσp /dz and dσn /dz are printed upwards and downwards on a log scale.

In order to evaluate the integrals in Eq. (2.189), it is necessary to know the electric field in the dark and the excess conductivities. However, σn and σp depend on the voltage which has been developed due to illumination, that is, σn (Voc ) and σp (Voc ). Thus Eq. (2.189) is an implicit function and its use is limited for contemplation rather than for prediction of the open circuit voltage. In Figure 2.47, relevant quantities for Eq. (2.189) are plotted for a typical heterostructure solar cell. Since this heterostructure exhibits step-like variations of Eg , one can initially


assume that the Voc has a considerable contribution of the third term in Eq. (2.189). However, this is not the case. Due to the overlap of σn,p and F0 , the first term in Eq. (2.189) is the dominant one. Also the Dember term has a noticeable size due to the large values of dσn and dσp . The effective force field (the third term) amounts to only some millielectronvolts, since dEg /dz (which is in the range of 108 eV cm−1 ) is limited to the heterointerfaces. Extended bandgap gradients which may be implemented in the absorber to promote carrier collection, also contribute very little to the open circuit voltage. The reason is that either these gradients of the band edges are small compared to the gradient of the electric potential (for instance in the case of a front surface gradient) or they do not coincide with high values of σn,p (for instance in the case of a back surface gradient). A second (and more useful) approach to describe the open circuit voltage developed by a solar cell under illumination employs the principle of superposition (see comment in Section 2.4) which is formulated in Eq. (2.176). Under open circuit condition, where J is zero, it is Voc =

Ea AkT −Jsc η(Voc ) Voc Gsh − + ln q q J00 J00

(2.190)

A shunt conductance of 10−3 S cm−2 is 102 times smaller than the quotient Jsc /Voc of a typical cell. Thus, under one-sun illumination, the second term in the bracket can safely be neglected, giving Ea AkT −Jsc η(Voc ) Voc = + ln q q J00

(2.191)

with the derivative Ak J00 dVoc = − ln dT q −Jsc η(Voc )

(2.192)

Here, it is assumed that J00 and −Jsc η(V) are not temperature dependent in the temperature range of interest. Furthermore, the derivative dEg /dT is neglected. As discussed in Section 2.4, the activation energy, Ea , of the diode current depends on the dominant recombination process. Equation (2.191) shows that Ea can be obtained from a plot of Voc versus T. The intercept with the ordinate (T = 0 K) of a tangent line to the Voc versus T curve is interpreted as Ea /q according to Ea /q = Voc

T=0

= VocT=T0 −T0

dVoc = T0 dT T

(2.193)

where T0 can be around RT. Application of Eq. (2.193) is straightforward as long as the diode quality factor and the reference current density are temperature

115

116


independent.22) In case of a temperature dependent diode quality factor, Eq. (2.193) reads

VocT=0 =

Ea dA kT0 2 -Jsc η(V) − ln q q J00 dT T=T0

(2.194)

Even if A = A(T) can be determined from analysis of the JV curve, evaluation of Eq. (2.194) is difficult since it requires knowledge of J00 . Thus, in the case of a temperature dependent A factor, evaluation of J0 (T) is recommended (see Section 2.6.1). The temperature dependence of the open circuit voltage as described by Eq. (2.192) is very important for the actual module efficiency under operating conditions. (dVoc /dT)T=RT (where RT = room temperature) has been calculated and plotted in Figure 2.27c. The full line describes the regime of SCR recombination while the dashed line of QNR recombination. (dVoc /dT) decreases with increasing carrier lifetime. Also of importance under operating conditions is the dependence of Voc on the illumination intensity. Again, this can be described by Eq. (2.191). Figure 2.48 shows examples of calculated and simulated values of Voc for QNR and SCR recombination. It is the A factor which causes a faster decline of Voc for SCR recombination. Note that no change in the solar spectrum due to lower sun inclination has been taken into account. A shunt conductance smaller than 10−4 S cm−2 (Rp > 10 kcm2 ) has no noticeable influence on the Voc (PI ) curves in the range from 100 to 1mWcm−2 . Also, a shunt conductance of 10−3 S cm−2 appears to be tolerable as long as the solar device is designed for energy harvest in an outdoor application. For indoor use, however, the data in Figure 2.48 need a shunt conductance smaller than 10−4 S cm−2 . It is noted that the illumination dependence of Voc and Jsc can be used to determine the saturation current density and diode quality factor of a device (see Section 2.6.1). 2.6.2.4 Fill Factor The cell delivers the maximum output power at a voltage Vmpp and a current density Jmpp (MPP =maximum power point). The fill factor describes the relationship between Voc , Jsc , and the power at the MPP

FF = Vmpp Jmpp /Voc Jsc 22) Device simulation programs may use tem-

perature dependent material parameters such as τn,0 (T), τp0,a (T), Nv,a (T), or Nc,a (T). According to semiconductor theory, Nv,a and Nc,a vary with temperature as Nv,a = N0v,a (T/T0 )3 and Nc,a = N0c,a (T/T0 )3 . In a

(2.195) device simulation program this power law may be referenced to T0 = 300 K. It increases the slope of the Voc (T) curve by about 3kT0 in the case of SCR and QNR recombination (Burgelman, M., personal communication).


QNR

0.7

Voc [V]

0.6

SCR

0.5 0.4 Rp

0.3 2

0.1

4

6

2

4

6

1

2

4

6

10

100

−2

PI [mW cm ] parameter used is the parallel resistance of Rp (k cm2 ) = 1, 10, and [∞]. Calculation for a AM1.5G spectrum of variable integral intensity. Experimentally this corresponds to the use of gray filters.

Figure 2.48 Calculated values of Voc for QNR (τn0,p0 = 10−8 s) and SCR recombination (τn0,p0 = 10−10 s) limited devices at 300 K as a function of illumination intensity according to Eq. (2.190). The

For any current–voltage relationship, the following equation holds dJ/dVmpp = −Jmpp /Vmpp

(2.196)

Equation (2.196) describes a specific property of the MPP. This property can be used to calculate Vmpp for a cell which obeys the superposition principle J = J0 (exp{qV/AkT} − 1) + Jph (V)

(2.197)

Here A is the diode quality factor, J0 is the saturation current density, and Jph (V) is the voltage dependent photo current density. With Eq. (2.196) we obtain qVmpp dJph (V) Jmpp q = J0 (2.198) − exp + Vmpp AkT AkT dV V=Vmpp

with Jmpp = J0 (exp{qVmpp /AkT}) + Jph (Vmpp ) Equation (2.198) can be rewritten as qV mpp −Jph (Vmpp ) − J0 exp AkT Vmpp = qV dJ qJ0 ph mpp exp + dV AkT AkT

(2.199)

(2.200)

V=Vmpp

Due to the implicit character of Eq. (2.200), it must be solved numerically in order to determine Vmpp . A requisite for the calculation is the knowledge of Jph (V) which finally is related, according to Eq. (2.165), to a voltage dependent QE(λ, V). (Jph is a negative quantity.) For the ideal case that dJph /dV = 0 and Jph = Jsc , Eq. (2.200) reduces to q(Voc − Vmpp ) AkT Vmpp = exp −1 (2.201) q AkT

117


1.0 0.9

Eg,a =1.8 eV Eg,a =0.8 eV

0.8 FF

118

A=1.0 A=1.1 A=2.0

0.7 0.6 0.5 0.4 0.0

0.5

1.0

1.5

2.0

Voc [V] Figure 2.49 Calculated fill factor values for A = 1.0, 1.1, and 2.0 according to Eq. (2.202). The markers give fill factors as obtained by device simulation for a the absorber/window default device but with varying absorber bandgap energy (Eg,a = 0.8–1.8 eV).

Equation (2.200) can be approximated by Vmpp = Voc − AkT/q ln(qVoc /AkT + 1) [19, 95]. Accordingly, Jmpp can be calculated after Eq. (2.199). The result for the fill factor in this ideal case reads qVoc qVoc AkT − ln AkT + 1 (2.202) FF = qVoc AkT + 1 In Figure 2.49, the fill factor as a function of Voc is plotted as obtained from Eq. (2.202) for diode factors of A = 1.0, 1.1, 2.0. The fill factor increases with increasing Voc . Highest fill factors are obtained for A = 1. The accuracy of the approximation in Eq. (2.202) is tested by comparison to simulated data of a absorber/window heterojunction (simulation parameters are given in the figure caption). In the simulation, the bandgap, Eg,a , of the absorber is varied from 0.8 to 1.8 eV. In this bandgap range, Voc linearly increases with Eg,a . All simulated JV curves exhibit a diode factor A of 1.1. It is apparent that the simulated data indeed resemble the calculated curve for a diode factor A = 1.1. The FF(Voc ) dependence as depicted in Figure 2.49 is taken into account if the solar energy conversion efficiency is calculated as a function of the optimum bandgap. Apparently, high bandgap materials are favored from the fill factor point of view. However, this effect cannot overcompensate the decline of the photo current density at higher bandgaps. Looking at Eq. (2.200), we may ask for the influence of a voltage dependent QE on the fill factor. There is, however, no simple relation between dJph (V)/dV and FF since Jph is nonlinearly dependent on the voltage, that is, dJph (V)/dV = f (V). An instructive example is the influence of a voltage dependent SCR. In the SCR the collection function for charge carriers ηc is unity, that is, all generated charge carriers are collected (consider the graphs in Figure 2.33 as representative for the nonbiased case). If the applied voltage bias changes, the width of the SCR changes.


In the ideal case, the width of the SCR is proportional to (V)1/2 . Since the collection function in the SCR is unity but declines beyond SCR, a variation of the SCR changes the collection function in a nonsimple fashion. As a result, the derivative dJph (V)/dV depends on the voltage. 2.6.3 Capacitance–Voltage Analysis

Measurement of capacitance as a function of voltage is a simple and widely used technique [96]. The capacitance is derived from the imaginary part of the measured admittance Y, that is, C = Im(Y)/ω. Identifying this capacitance with the SCR capacitance directly assumes that the correct equivalent circuit is a capacitance and a conductance in parallel. However, as revealed by Figure 2.42, the equivalent circuit can be more complicated due to capacitance contributions from deep defects, interface defects and additional resistance elements. We have to select conditions for the C–V measurement which allow to simplify the equivalent circuit and to facilitate the interpretation of the capacitance. The geometrical capacitance of the depletion region of an absorber/buffer/ window heterostructure is given by 1 C= w a/ + wb/ + ww/ εa εb εw

(2.203)

In the specific case of a highly doped window layer (ww ≈ 0), a fully depleted buffer layer (wb = db ), and without deep bulk states in the absorber, the quantity wa is given by Eq. (2.36). We want to make use of this Eq. (2.36) in order to show how to obtain the built-in potential Vbi and the doping density NA from a C–V measurement. For the specific case, Eq. (2.203) becomes 1 db κIF −1 db 2 2 q2 ND,b db 2 = + + 2 [q(Vbi − V)] + + vIF z2 C εb κIF εb κIF q NA,a εa κIF 2εb (2.204) In the Mott–Schottky approach 1/C2 is plotted versus V. Thus, we have to consider the expression 2(Vbi − V) db 2 1 εb ND,b κIF − 1 2 2νIF z2 1 = + 2 + + + 2 C2 qNA,a εa κIF εb κIF κIF 2 εa NA,a q NA,a εa κIF q2 ND,b d2b db κIF − 1 2 db 2 +2 + 2 q (Vbi − V) + + νIF z2 εb κIF εb κIF q NA,a εa κIF 2εb (2.205) The first term on the right hand side of Eq. (2.205) resembles the C–V dependence of a simple n+ p heterojunction where the SCR only extends into the p-type absorber. The second term accounts for the geometrical capacitance contribution of the buffer layer. The third term represents the influence of interface charge. In the case of a fixed charge density at the buffer/absorber interface, it is κIF = 1

119

120


1/C2

NA,a

db, NIF+

Vbi

V

Figure 2.50 Schematic 1/C2 versus V plot illustrating the effect of increasing absorber doping concentration, NA,a , and a potential drop at the buffer layer or positive interface charge, N+ IF . The case of negative interface charge is not depicted.

(see Section 2.3.2), and the fourth term on the right hand side of Eq. (2.205) vanishes. Then, 1/C2 becomes linear in V which allows to extract NA,a from the slope of the 1/C2 versus V curve according to + −1  1 2 d C 2   NA,a = − (2.206) qεa dV Obviously, a large slope of 1/C2 indicates a small value of NA,a and vice versa (see example in Figure 2.50). If we allow the (shallow) doping density NA,a to vary in the SCR, Eq. (2.206) can be used to determine the doping profile NA,a (z) by evaluating the slope of 1/C2 at different voltages. This gives NA,a (V), which is translated to NA,a [wa (V)] by aid of Eq. (2.203) and the measured relation C(V). Equation (2.205) is also valid for a continuous charge density at the buffer/absorber interface. Such a density dNIF /dE makes the coefficient κIF = 1 + νIF and leads to a nonlinear and curved 1/C2 versus V plot. The deviation from linearity is understandable since upon applied voltage bias, the interface charge density changes, leading to a varying potential distribution in the different layers. We return to the case of κIF = 1 (fixed interface charge) where the built-in potential Vbi can be inferred from db 2 d 1+ νIF z2 εb ND,b + 1+ Vbi + 2 − C2 εb dV εa NA,a q + 1+ 2 C+ + V = V 1 C2 = 0 = (2.207) d 1 2 − dV C From the extrapolation of 1/C2 to 1/C2 = 0, Vbi can be obtained provided that the parameters of the buffer layer and the interface charge are known. A buffer layer as well as positive interface charge leads to an extrapolated voltage value which is larger than the built-in potential (see Figure 2.50). Accordingly, a negative interface charge leads to a smaller voltage value. For our default device with a 50 nm buffer


Ec Ed,A Ev -wa

-zd

0

Position z Figure 2.51 Schematic illustration of the effect of a deep acceptor trap in the bandgap of an absorber with energy Ed . The states of density Nd,A add to the negative space charge in the range zd < −z < 0. This space charge is not considered in Eq. (2.36) and thus distorts the measured NA,a . As a

second influence, charging and discharging of the deep states at position −zd leads to an additional admittance contribution and thus to the measured capacitance. The latter effect may be circumvented by freezing out the response of deep states.

layer, the second term in Eq. (2.207) amounts to below 5% of Vbi and therefore is in the range of the accuracy limit of the C–V method. Note, however, that due to the parabolic dependence on db the second term rises fast with increasing db . If Vbi has been obtained from a C–V measurement, the band offsets can be derived provided that the dopant concentrations in the different layers are known. The basis for this procedure are the Eq. (2.22) for the absorber/window and Eq. (2.33) for the absorber/buffer/window heterostructures. In the last step, we have to account for the influence of deep states on the C–V measurement. Consider an absorber with deep acceptor states as shown in Figure 2.51, which is part of a pin+ structure (absorber/buffer/window). As described in Section 2.3.6, the charge in the deep traps will influence the potential distribution in the device and, in particular, the SCR width of the absorber. This is the static part of the influence of deep traps. In addition, charging and discharging of the traps with the frequency of the measurement, ωAC , can add to the measured capacitance. This is the dynamic part of the influence of deep traps. As explained in the next section, the defect response is maximal at the position where it is Ed = Ep,a (z). This is at the location −z = zd . Three quantities are important to discriminate different measurement regimes: (i) the emission rate of the deep traps ω0 = 2/τe where τe is the emission time constant, (ii) the frequency of the AC voltage ωAC , and (iii) the sweep rate of the DC bias voltage ωDC . In a C–V measurement, we normally have ωAC > ωDC . In addition, we must always guarantee ωAC < τ−1 d = σa /εa where τd is the dielectric relaxation time and σa is the conductivity. Otherwise, the measured capacitance corresponds to the sample geometry. The different regimes for a C–V measurement are: 1) ωAC > ωDC > ω0 . The deep trap can follow neither the AC frequency of the capacitance measurement nor the loading/disloading of the DC bias change. In that case, the slope of the 1/C2 curve gives the shallow absorber doping density NA,a [97].

121

122


2) ωAC > ω0 > ωDC . Due to the high AC frequency, the dynamic response of the deep state does not add to the capacitance and can be neglected. From the slope of the 1/C2 curve we obtain a larger doping concentration with value NA,a + Nd,A zd /wa provided that both state densities are homogeneous ) in the Ed −Ep,a zd SCR. This is the static influence of deep states. If we replace wa = 1 − q(V −V) bi

[98] and measure the 1/C2 profile at different temperatures, NA,a and Nd,A can be obtained by curve fitting [99]. 3) ωT > ωAC > ωDC . If the AC frequency is low enough to charge/discharge the deep states at the position −zd , evaluation of the 1/C2 curve after Eq. (2.206) gives the sum of shallow doping concentration and deep state density, that is, NA,a + Nd,A . The task now is to find the appropriate measurement conditions for one of the cases described above. This can be accomplished by aid of admittance spectroscopy. 2.6.4 Admittance Spectroscopy

Spectroscopy of the device admittance is a technique for the measurement of doping density, deep defect levels, and other electronic properties. By measuring the small-signal AC admittance of the junction under different conditions, for example, with frequency and temperature as parameters, defect states in the SCR [100] and at the interface [85] can be analyzed. First, we consider a single defect state in the SCR of the absorber. We refer to the Gaussian acceptor defect in the absorber of the default device (Table 8.1) with parameters Nd = 1014 cm−3 , Ed = 0.4 eV, and σd = 0.05 eV. The defect can communicate with the valence band by the transactions illustrated in Figure 2.17. As the state is in the lower part of the band gap, communication with the conduction band can be neglected. Charging and discharging of the state will give a certain admittance contribution. As for the C–V measurement, we interpret the imaginary part of the admittance in terms of a capacitance. In the capacitance are lumped the contributions of the SCR and of the defect. Figure 2.42c reveals that the capacitance Cd of a defect state in the absorber SCR is in parallel with the SCR capacitance, Ca . We note, however, that if there is a large series resistance, the equivalent circuit has to be extended [101] and the capacitance has to be extracted by mathematical manipulations. Charging and discharging of the defect state is governed by the rate expressions in Eq. (2.65). At the position where the Fermi level crosses the defect state, rates II and IV of Eq. (2.65) will be equal and defect charging and discharging will be most efficient. In Figure 2.51, we have marked the position where it holds τc,p = τe,p = τp0 by −zd . For small enough frequency, the defect occupation is governed by the corresponding quasi Fermi level and can be charged and discharged with frequency ω. However, for too large a frequency charging and discharging of the defect can no longer follow the AC frequency. The border between the low frequency and high frequency regime is denoted as ω0 .

28.8 CSCR ω0

28.7 103

104

105

Frequency ω

(a)

Nd [cm−3 eV−1]

CSCR + Cd

-dC/dInw [a.u.]

Capacitance C [nF cm−2]


3 × 1014

Figure 2.52 (a) Simulated capacity of the default absorber/buffer/window device from Table 8.1 as a function of frequency and the capacity derivative. For frequency ω < ω0 , charging and discharging of the defect state

1 0 0.0

106 [s−1]

2

(b)

0.5

1.0

E [eV]

represents a capacity contribution, Cd , which adds to the capacity of the SCR, CSCR . (b) Evaluation of the defect density Nd according to Eq. (2.214).

As shown in Figure 2.52a, C(ω) represents a step-like function. The characteristic frequency ω0 can be determined by a maximum in the plot −dC(ω)/dlnω versus ω. For Nd NA,a it is [102] Cd (ω) ∝

ω20

ω20 + ω2

(2.208)

Furthermore, the derivative dC/dlnω is a measure of the defect density. ω0 is related to the emission time constant τp0 according to [102] ω0 = 2/τp0 = 2ν0 exp{−Ed /kT}

(2.209)

where ν0 is the attempt to escape frequency, ν0 = Nv,a vp σp . σp is the (hole) capture cross section of the defect. With the temperature dependences of vp ∝ T1/2 and Nv ∝ T3/2 we can write ν0 = σp vp Nv,a = ξ0 T2

(2.210)

which yields ω0 = 2ξ0 T2 exp{−Ed /kT}

(2.211)

With the definition of the effective density of states and vp = (3kT/mp )3/2 , the emission factor ξ0 for emission in the valence band comes out as ξ0 =

√ √ A∗ m p k2 6π4π 3 σp = 6π σp h q

(2.212)

with A∗ , the effective Richardson constant. Thus, from an Arrhenius plot of ω0 /T2 versus T−1 , the defect energy Ed can be determined. This is shown in Figure 2.53d for the bulk defect (). We remind that in this work the defect energy Ed is related to the valence band. Thus, in the case of a defect in the lower part of the band gap, the activation energy of the ω0 /T2 versus T−1 plot directly gives Ed . We note, however, that in principle an admittance experiment gives the same response for a

123

124

3.5 3.0

En,az=0

EF Ev

2.5 2.0 (a)

200 K

28

27

0.0

Position z [µm]

0.2

0

1 2 × 1010 −2

(b) NIF [cm

−1

eV ]

Ea=Ed 10−1

T 100 K

103 (c)

100

10−2 26

−0.4 −0.2

101

29 ω0 /T2 [s−1 K−2]

Capacitance density [nF cm−2]

Energy E [eV]

4.0 Ed

Ea=En,a z=0

300 K

Capacitance density [nF cm−2]

32

Ec

104

105

106

Frequency ω [Hz]

(d)

T

−1

−1

[K ]

+V 28 −V 26

24 103

4 6 × 10−3

2

200 K

30

(e)

105

Frequency ω [Hz]

Figure 2.53 Figure 2.53(a) Band diagram of the absorber/buffer/window default solar cell but with a constant density of donor states at the absorber/buffer interface. Ed marks the Gaussian-broadened acceptor-type defect state in the absorber band gap. (b) Interface state density NIF . (c) Simulated capacitance density spectra for different temperatures. The capacitance spectra exhibit a frequency dispersion due to the capacitance contribution of the interface states () and of the bulk state (◦). Markers give the characteristic frequencies. (d) Arrhenius plot of ω0 . The slope Ea corresponds to the intersection of EF with the interface state density, that is, to En,az=0 , and to the defect energy Ed . (e) Capacitance spectra simulated at 200 K, but with different bias voltage. The shift of the inflection point ω0 upon eletric bias reflects the variation of the electron quasi Fermi level at the interface.


4.5


HT and an ET. The activation energy of ω0 is simply the energetic distance between the band edge of the carrier involved and the location of the defect level in the band gap. Discrimination between ETs and HTs is possible by deep-level transient spectroscopy. The characteristic frequency, ω0 , defines the border between a low frequency capacity CLF and a high frequency capacity CHF . In Figure 2.52a it is CLF = Cd + CSCR and CHF = CSCR . The difference is the capacity of the defect response. Besides from a deep defect state, a capacitance step can also arise from an energy barrier at a contact [103], from the dielectric relaxation in a bulk layer, and from interface states (see below). Fermi level pinning resulting from high concentrations of deep-level defects can distort the measured activation energy and the apparent capture cross section [104]. Walter et al. [102] generalized the admittance theory of Losee [100] to a continuous defect densities. Now, the Fermi level intersects the defect density in an extended region which depends on the type of band bending. For a p-i-n diode where the built-in voltage drops linearly in the i layer of thickness w, the defect density can be calculated after [102] Nd (Ed ) =

dC ω Vbi 2

wa qVbi − Ep,a − Ed dω kT

(2.213)

In contrast, one finds for a pn+ diode with parabolic band bending in the p layer 3/2

Nd (Ed ) = −

2Vbi dC ω ,

dω kT wa q qVbi − Ep,a − Ed

(2.214)

Thus, evaluating the −dC/dlnω curve in Figure 2.52a yields the defect density as a function of energy. In Figure 2.52b, the ordinate results from Eq. (2.214) [102] while the abscissa is obtained from E = ln(ω/2ν0 ). Here, ν0 includes the unknown quantity ξ0 which, however, can be fitted by bringing Nd (E) curves of different temperatures into congruency. This is shown in Figure 2.52b where a plot of Nd as a function of E gives a defect band of width 2σ = 0.1 eV and with a maximum at 0.4 eV – in accordance with the Gaussian defect in the default device (Table 8.1). Also interface states can be charged and discharged by an AC stimulation and thence can contribute to the cell’s capacitance. A prerequisite is that the Fermi energy intersects the interface state density. The equivalent circuit of a device with interface states is depicted in Figure 2.42d. The equivalent circuit of a device with both deep bulk defect and interface defect states is a combination of Figure 2.42c and d. In the fortunate case where the bulk defects are deeper than the interface defects, the equivalent circuit is again reduced to Figure 2.42d. In the example of Figure 2.53, we add a continuity of donor type interface states to our default absorber/buffer/window heterostructure. At sufficiently low frequency, electrons are captured and emitted by interface states located at the Fermi energy and thus contribute to the cell’s capacitance. If the AC frequency exceeds the emission rate, τ−1 e,n of the interface states at En,az=0 , the interface states can no longer follow.

125

126


The limiting frequency again is interpreted using Eq. (2.211) where Ed equals the quantity En,az=0 . From an Arrhenius plot we obtain Ea = En,az=0 , as shown in Figure 2.42e. In case that the Fermi level is not pinned, the capacitance step CLF − CHF is related to the interface state density – in the simplest case it is CLF − CHF = CIF = qNIF . However, in the case where the Fermi level is pinned the capacitance step has a geometrical meaning, as explained by Niemegeers et al. [79] and cited in the following. In the low frequency range, electrons are supplied from the window through the buffer to the buffer/absorber interface. The AC potential builds up between the buffer/absorber interface and the edge of the absorber SCR. The low frequency capacitance density is given by CLF =

εa wa

(2.215)

If the interface states can no longer follow the applied AC signal (ω > ω0 ), the AC potential builds up between the edge of the emitter SCR (for high window doping this is close to the buffer/window interface) and the edge of the absorber SCR. That is 1 (2.216) CHF = w a/ + we/ εe εa where the index e refers to the emitter (window + buffer, or only window). The step in the capacitance frequency curves thus is related to the space charge width we 1 1 we ≈ εe − (2.217) CHF CLF In case of a highly doped window, a depleted buffer, and large NIF the quantity we about equals the buffer thickness. Measurement of En,az=0 by admittance spectroscopy can be obscured by the dielectric relaxation effect in the least conductive layer of the heterostructure. In the example in Figure 2.53 this can be the buffer layer since it may be depleted from carriers. In this case, ω0 is determined by the inverse of the dielectric relaxation time τd of the electrons in the buffer layer [79] ω0 = ωd = 1/τd = qµn n/εb . With n = Nc,b exp{−(En,b )} it is obvious that ω0 is still temperature activated and we may detect frequency dispersion of the capacitance. An Arrhenius plot of ω0 yields the quantity En,b for the buffer layer. In the case where the band offset Ec a,b is known, we can calculate the difference between conduction band and Fermi energy in the absorber from En,az=0 = En,b − Ec a,b . The parameters in Figure 2.53 are chosen, however, to make sure that the buffer is sufficiently conductive and ωd is very large. Bulk defects are continuous in space but may show a narrow energy distribution. In contrast, interface state densities are discrete in space and (often) continuous in energy. How can we discriminate between an homogeneous bulk defect and a confined interface defect? If the bias voltage on the sample is changed, the


bulk defect state crosses the Fermi energy only at a different position within the SCR – but the limiting frequency remains the same. However, due to applied bias, the interface state density is intersected by the quasi Fermi energy at a different energy level – provided that the Fermi level is not pinned. Thus, the limiting frequency of a device with a nonpinned Fermi level changes. This is exemplified in Figure 2.53f. In conclusion, if the limiting frequency ω0 is bias independent, the frequency dispersion of the capacitance may be due to a bulk defect or due to interface states plus Fermi level pinning.

127

129

3 Design Rules for Heterostructure Solar Cells and Modules The solar conversion efficiency of a heterostructure cell is determined by the material parameters of the single layers as well as by device parameters. By device parameters we mean band alignment, doping ratios, film thicknesses, and other parameters. These device parameters have in common that they determine the layout of the band diagram. In this chapter, we conduct design studies based on calculated or simulated solar cell parameters. We consider a related sequence of examples which then give rules for the heterostructure design. It is not so much the absolute number of a solar cell parameter which is important, but how the cell performance depends on this parameter in general. The specific band diagram of a heterostructure influences all four recombination paths of Figure 2.46. The Fermi level position at the interface affects interface recombination. The size of the electric field influences space charge region (SCR) recombination. Bandgap gradients play a role in quasi neutral region (QNR) recombination. Electric field and film thickness determine the impact of back surface recombination. An extreme example is recombination path (1) which, on the one hand, can strongly deteriorate the cell performance but, on the other hand, can largely be deactivated by a proper design of the band diagram – even in the presence of interface states. In order to minimize recombination and maximize performance, specific rules for the design of the band diagram have to be regarded. Here, we discuss these rules based on the different degrees of freedom for the design of a heterostructure band diagram. Table 3.1 depicts the study logic which (with exceptions) is followed in the next sections. Starting with a simple absorber/window heterostructure, the design study step by step adds further device parameters. From the choice of a buffer layer thickness onward, the device is of the absorber/buffer/window type. Note that the final outline of the cell is not identical to the default device in all design aspects.

3.1 Absorber Bandgap

The absorber bandgap determines the maximum short circuit current density of the heterostructure (see Figure 2.31b). Decreasing the bandgap increases the Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

130

3 Design Rules for Heterostructure Solar Cells and Modules Table 3.1 List of design studies and derived design rules. Settings denotes device parameters which have been set in accordance with the output of a design study.

Design parameter

Main impact

Rule

Setting

Absorber bandgap

Generation current, maximum splitting of EFn − EFp Interface recombination

1 eV < Eg,a < 1.6 eV

1.2 eV

Ec ≥ 0

Ec = 0 eV Sn0,p0,IF = 105 cm s−1 ND,w = 1018 cm−3 No FLP

Band alignment Emitter doping Fermi level pinning (FLP) Absorber doping

Absorber thickness Grain boundaries (GB) Back contact barrier Buffer thickness

Interface recombination, optical loss Interface recombination

ND,w NA,a

Collection length, recombination region, interface recombination Recombination rate, collection length Effective diffusion length, saturation current

1015 < NA,a < 1017 cm−3

Impedance, recombination rate Photo current, optical reflection Interface recombination

Front surface gradient Back surface gradient Volume recombination, back surface recombination

NA,a = 1016 cm−3

da = 1.5 µm Sn0,p0,bc = 107 cm s−1 Sn0,p0,gb ≤ 103 cm No GBs s−1 no horizontal GBs p p ϕb < 0.3 eV ϕb = Ep,az=0 db = minimal Eg, IF > Eg, a , dmin minimal dmin = f (da )

db = 50 nm ND,b = 1015 cm−3 dmin,front = 10 nm, Ev,front = 0.2eV dmin,back = 0.5 µm, Ec,back = 0.3eV

photo current; but, Voc decreases, see Eq. (2.191). It is clear that a plot of solar cell efficiency versus absorber bandgap exhibits a maximum. But, is this maximum dependent on the electronic parameters of the absorber or is it a unique function? As examples, we selected two limiting scenarios (Figure 3.1). The first scenario only includes radiative recombination (Shockley–Queisser limit) and the second uses recombination in the SCR. Both scenarios have in common that optical absorption in the emitter is neglected. The Voc of the devices have been calculated by use of Eq. (2.191) with J00 for radiative recombination only (see Eq. 7 in Ref. [33]) and for SCR recombination (Table 2.3). The fill factor (FF) has been calculated after Eq. (2.202). Jsc has been obtained from Eq. (2.137) for both scenarios. Thus, even for SCR recombination the Jsc is assumed to be ideal. Such a cell requires excellent light trapping.1) 1) However, we found that if we calculate

Jsc (Eg ) by considering the effect of a limited collection probability in the absorber, the

shape of the Jsc (Eg ) curve in Figure 2.7b is largely unaffected.

3.2 Band Alignment 35 SQL

η [%]

30 25

SCR

20 15 10 0.8

1.0

1.2

1.4

1.6

1.8

2.0

Eg [eV]

Figure 3.1 Calculated maximum solar cell efficiencies for AM1.5G (solid lines) and AM0 (dashed lines) in the limit of radiative recombination only (Shockley–Queisser limit, SQL) and for recombination in the SCR with

absorber lifetime τn0,p0 = 10−10 s. For both cases, the full Jsc (Eg ) according to Figure 2.31b was assumed. Calculation without reflection loss and without absorption in the emitter layers.

In the Shockley–Queisser limit, maxima of η(Eg ) at 1.15 and 1.35 eV can be discerned. Thus, optimum efficiency can be expected for an absorber bandgap of 1.4 eV. (The ripples in the η(Eg ) curve stem from the absorption bands in the AM1.5G spectrum.) The curve for the AM0 spectrum has a single maximum at 1.3 eV. The picture, however, changes if the absorber is in the limit of SCR recombination. Now, the maximum of η(Eg ) is shifted to larger values of Eg . The reason is the larger loss in Voc due to the term −q−1 AkTln(J00 /Jsc ) in Eq. (2.191). with decreasing electric field Fm . As Fm in According to Table 2.3, J00 increases √ the absorber is proportional to Vbi [combine Eqs. (2.36) and (2.80)], J00 for SCR recombination is larger for lower bandgap absorbers. Thus for absorbers with poor electronic quality, the optimum bandgap is higher than for high quality material. In particular, for an absorber bandgap below 1 eV, it is very important to have high quality material. This picture does not change if we in addition let the Jsc be reduced in the SCR scenario. Since the optimum bandgap of an absorber depends on the electronic quality, we may only state the well known rule: Rule 1: The absorber of a single junction heterostructure should have a bandgap of 1 eV < Eg,a < 1.6 eV. We note again that this rule does not take into account optical and electronic losses in the emitter. If the actual wavelength range which contributes to Jgen becomes smaller, the efficiency maximum is shifted to smaller band gap values. 3.2 Band Alignment

The band alignment of a heterostructure is determined by the offsets of valence and conduction band at the interface. These offsets are approximately fixed for a

131

3 Design Rules for Heterostructure Solar Cells and Modules

Ec Energy E [eV]

132

(a)

EF

Ev

(b)

(c)

(d)

Position z Figure 3.2 Schematic examples of band alignments which are critical for the function of a heterostructure solar cell. In (a), the band alignment in an absorber/window may be varied. In the absorber/buffer/window heterostructure, the band offset at the

absorber/buffer interface (b), at both interfaces buffer/window and buffer/absorber (c), and at the buffer/window interface (d) may be varied. Example values for the Voc are given in Figure 3.5.

given combination of heteropartners. Only dipole layers, in principle, may change the band offsets. The dipole concept of interface engineering [20], however, has not yet been realized in chalcogenide solar cells. Thus, the established degree of freedom for interface engineering is the choice of the heteropartners. In order to emulate this choice in the following section, we consider devices with layers of fixed bandgap but different band offsets; and we ask: (i) which solar cell parameter is mostly affected by the band alignment, and (ii) is there a rule for the design of the band alignment? A survey of the band alignments to be discussed is given in Figure 3.2. First, we discuss the band alignment at the interface of an absorber/window heterojunction. It is assumed that the window layer is highly n-type doped with ND,w = 1018 cm−3 . The absorber is moderately doped p-type with NA,a = 1016 cm−3 . The interface between window and absorber contains interface states. Without restriction in generality, we assume these states to be located in the middle of the interface bandgap, that is, at (Ec,min + Ev,a )/2 where Ec,min is the minimum of window and absorber conduction band edge. As explained in Section 2.4.5.4, we further assume that cross recombination at the interface can take place. Electrons from the lowest lying conduction band edge can recombine with holes from the highest lying valence band edge. We vary the interface quality by varying the nominal interface recombination velocity S = Sn0,p0 = NIF σn,p vn,p . The interface states are of neutral character and do not influence the band diagram.2) 2) Defect states are called ‘‘neutral’’ if their

state density is small and they do not noticeable influence the band diagram. Their charge can be neglected if compared to the charge density in the SCR. They may

nevertheless induce carrier recombination if their capture cross sections are large. Here, we use ‘‘neutral’’ defect states in order to decouple interface recombination and Fermi level pinning.

3.2 Band Alignment

In Figure 3.3 it can be seen that for a nominal interface recombination velocity of Sn0,p0 = 103 cm s−1 , efficiencies of above 20% are obtained for conduction band offsets in the range −0.2 eV < Ec w,a < 0.4 eV. Thereby, all solar cell parameters are largely independent of Ec . Only for Ec w,a > 0.4 eV, Jsc and FF strongly decline. (We come back to this point below.) The tolerance of the solar cell parameters against variation of Ec gradually vanishes with increasing interface recombination. However, even for very high nominal interface recombination velocity of S = 107 cm s−1 , there is a range of conduction band offsets (0.1 eV < 0.8

Voc [eV ]

0.7 0.6 0.5

Sn0,p0

0.4 0.3

Jsc [mA cm−2]

36 35 34 33 84

FF [%]

80 76 72 24

η [%]

20 16 12 8

−0.4

−0.2

0.0

0.2

0.4

∆Ecw,a [eV ] Figure 3.3 Simulated solar cell parameters as a function of Ec w,a for an absorber/window heterostructure. Other device parameters according to default Table 8.1. Nominal interface recombination velocities

are Sn0,p0 = 103 , 104 , . . . , 107 cm s−1 (direction of arrows) realized by interface defects energetically located in the middle of the interface bandgap.

133

134


Ec w,a < 0.4 eV) where Voc is independent of Sn0,p0 and only the values of Jsc are slightly reduced. In this range of Ec w,a , the diode current obviously is small and photo generated electrons can pass the highly defective absorber/window interface without significant recombination. This surprising result can be explained by regarding the carrier concentrations, as depicted in Figure 3.4b and by the aid of Eq. (2.116). Interface states communicate with the valence and conduction bands of both the window and the absorber layer. Thus, the kinetic of occupation of the interface states is related to the concentration of carriers in four bands. Figure 3.4b shows that the electron concentration at the surface for all values of Ec is much larger than the hole concentration in the absorber. (We omit the hole concentration at the window surface, since this concentration is very small.) For positive Ec , it is the electron concentration at the absorber surface which dominates the occupation of interface defects. Thus, the interface defects are largely occupied by electrons being the majority carriers at the interface. Interface recombination is governed by the concentration of free holes at the absorber surface. In the case of Ec w,a > 0, the hole concentration is small both for zero bias (Figure 3.4b, dashed line) and for forward bias (full line). This is the reason why the diode current is small (high Voc ) and the photo current is high (high Jsc ). The saturation current density J0 for interface recombination is expressed by Eq. (2.116). According to this equation, J0 becomes small for a large value of Ep,az=0 (or a small value of En,az=0 ). This is also shown in Figure 2.26. Ep,az=0 corresponds to p the hole barrier φb . A requisite for Ep,az=0 being large is that Ec is not negative, that is, Ec ≥ 0. Then, the saturation current, J0 , due to interface recombination can be negligible even for a large value of Sp0 . If the J0 for interface recombination is small, the device will be limited by bulk recombination. Obviously, interface recombination can be switched off by a proper band alignment provided that the doping ratio ND,w /NA,a is high. This allows: Rule 2: The band alignment in a heterojunction solar cell should be such as to allow Ep,az=0 ≈ Eg,a . Rule 2 basically demands that the absorber is in inversion. Inversion means that the minority carriers in the bulk of the absorber are majority carriers at the site of possible recombination – here at the interface.3) Rule 2 is valid for an emitter consisting of a single window layer as well as for an emitter consisting of a buffer/window combination. For a negative Ec , Voc declines proportional to −Ec . The onset of this decline depends on Sn0,p0 . The onset marks the point where interface recombination starts to dominate over bulk recombination. The decline can be explained by the decreased value of Ep,az=0 (Figure 3.4a). Physically, it is due to the increased 3) Another

formulation is that the pn-junction (the location of n = p) should be located within the absorber layer. It can be considered as a buried junction. This term however is difficult to quantify. Even for slightly higher emitter doping,

the absorber would host the pn-junction at zero bias. For a certain forward bias, however, the location of n = p would move to the interface and the cell would become interface limited.

3.2 Band Alignment

Ep,az = 0 [eV]

1.2 1.0 0.8 0.6 (a) 1020

n,p [cm−3 ]

1018 nw

1016

z= 0

14

naz = 0

10

1012

paz = 0

10

10

108

−0.4

−0.2

0.2

0.0

0.4

∆Ecw,a [eV ]

(b)

Ep,az = 0 [eV ] 1.20

1.10

1.00

0.90

0.80

0.70

0.4

0.5

0.8

Voc [ V ]

0.7 0.6 0.5 0.4 Sn0,p0 0.3 0.0 (c)

0.1

0.2

0.3

En,a z= 0 [eV ]

Figure 3.4 (a) Simulated values of Ep,az=0 as a function of the conduction band offset Ec w,a . Absorber/window type heterojunction solar cell as in Figure 3.2a. (b) Carrier densities at the two sides of the absorber/window interface under AM1.5G illumination at zero bias (dashed lines) and at 0.7 V forward bias (solid lines). Hole concentration

at the absorber side, pa , electron concentration at the absorber side, na , and at the window side, nw , with interface recombination velocity Sn0,p0 = 10 cm s−1 . (c) Voc values from Figure 3.3 plotted as a function of En,az=0 and Ep,az=0 for different values of the nominal interface recombination velocity 103 = Sn0,p0 , . . . , 107 cm s−1 .

135

136


hole concentration at the absorber surface at forward bias (Figure 3.4b). For highest interface recombination velocity, the decline of Voc already sets in for Eg,a − Ep,az=0 > 0.05 eV. In Section 3.10 we show that a cell with negative conduction band offset can be ‘cured’ by the introduction of a bandgap gradient at the front of the absorber. Thereby, the hole barrier b p or Ep,az=0 is increased. Next, we want to address the question why, for all Ec w,a < 0.4 eV, the influence of interface recombination on the Jsc is small. This can be understood by inspecting the hole concentration at zero bias under AM1.5G illumination. This concentration is much smaller than the electron concentration. Thus, photo generated electrons representing the photo current in the cell can pass the interface without significant recombination – even for Sn0,p0 = 107 cm s−1 . Note that Jsc becomes strongly affected by interface recombination if the doping ratio ND,w /NA,a is smaller. This point is addressed in Section 3.4. For Ec w,a > 0.4 eV, Jsc and FF markedly decline, defining a cell efficiency below 10%. The reason is the strong spike in the conduction band which forms a barrier for the photo current (see Section 2.5.6). This current barrier is intuitively understood and has often been invoked as an argument against any kind of spike in the conduction band. We see, however, that current flow is not impeded for Ec w,a < 0.4 eV where the full Jsc can pass the interface. The decisive quantities, as has been derived in Section 2.5.6, are the Boltzmann term for the thermionic emission over the conduction band spike, exp{−Ec w,a /kT}, and the absorber electron concentration at the surface, naz=0 . Below Ec w,a = 0.4 eV the electron concentration is sufficient and the spike in the conduction band does not impede flow of photo generated electrons. For a too large value of Ec w,a , the electron concentration (naz=0 ) is not sufficient to overcompensate the small Boltzmann term. In the example of Figure 3.4, this is the case for Ec w,a > 0.4 eV. For the absorber/window heterostructure we can summarize: (i) a positive conduction band offset is required for large Sn0,p0 and (ii) the short circuit current is large as long as the absorber is in inversion and the conduction band spike is not too large. In principle, both mechanisms are also valid for an absorber/buffer/window heterojunction with varying alignment at the buffer/absorber interface according to Figure 3.2b. Nevertheless curve b in Figure 3.5 reveals that for the absorber/buffer/window heterostructure, the achieved open circuit voltage is smaller than for an absorber/window heterojunction (curve a). The reason is that here we assumed a low buffer doping. Therefore, the buffer acts like a dielectric and induces a further potential drop as is sketched in Figure 3.2b. Only for very large Ec b,a , Voc approaches the value of the absorber/window heterojunction. Case c in Figures 3.2 and 3.5 varies the conduction band offset buffer/window as well as buffer/absorber by letting the sum equal zero, that is, χ = Ec w,b + Ec b,a = 0. The drop of Voc for Ec b,a is similar as for case b but there is a strong increase of Voc for Ec b,a > 0.3 eV. This is due to a barrier for the diode current. We regret that such cells exhibit a low Jsc due to the concomitant large barrier for the photo current. Finally we consider the band alignment at the buffer/window interface while keeping Ec b,a = 0 eV. The dependence of Voc for case d in Figure 3.5 is similar as

3.3 Emitter Doping and Doping Ratio

1.4 1.2 c Voc [V ]

1.0 0.8

b

a

0.6 0.4

d

0.2 0.0

−0.4

−0.2

0.0

0.2

0.4

∆Ec [eV] Figure 3.5 Simulated values for Voc as a function of the conduction band offset for the variations as sketched in Figure 3.2 Other device parameters as in default Table 8.1.

for case b. The strong increase of Voc for Ec w,b < −0.3 eV is due to the barrier for the diode current. It comes alongside with a kink in the JV curve and low values of Jsc and FF. For the following sections, we choose Ec = 0 at the absorber surface.


The emitter doping or, to be more precise, the absorber/emitter doping ratio also has a strong impact on interface recombination. It turns out that a high doping ratio helps to reduce interface recombination. We again consider the junction between an n-type window and a p-type absorber as depicted in Figure 3.2a. The conduction band offset is zero, that is, Ec w,a = 0. There is no interface charge and the absorber doping is NA,a = 1016 cm−3 . By varying the window doping ND,w between 1015 and 1018 cm−3 , we can change Ep,az=0 according to Eq. (2.24).4) We assume energetically discrete interface states located in the middle of the interface bandgap. As in Section 3.1, we vary the influence of the interface by varying the nominal interface recombination velocity Sn0,p0 = NIF σn,p vn,p . Simulated values for Ep,az=0 as a function of the doping ratio ND,w /NA,a are given in Figure 3.6b. For ND,w /NA,a = 0.1 we find the Fermi level at the interface close to the valence band edge of the absorber, that is, Ep,az=0 < Eg,a /2. Most of the built-in potential drops in the window layer. In contrast, for ND,w /NA,a = 100 the 4) Note that according to Eq. (2.22), a change

in the window doping concentration with a fixed absorber concentration, slightly

changes the built-in potential. For the presented numerical example, however, this will remain a minor effect.

137

138


Ep,az =0 [eV]

Ep,az =0, Ep,a z =0 [eV]

0.4 0.6 0.8 1.0 1.1

Sn0,p0

30 20

0 80

0.8 En,az=0

Ep,a

z=0

0.4 0.0 0.01

0.1

1 ND,w / NA,a

10

100

1016

70 60 50

1012 108 104 100

20

η [%]

1.2

(b)

10

na,pa [cm−3]

FF [%]

Jsc [mA cm −2]

Voc [ V ]

0.35 0.8 0.7 0.6 0.5 0.4 0.3

paz= 0

naz= 0 na0z= 0 0.4

0.8

1.0

Ep,az=0 [eV]

(c)

15

0.6

0

pa z= 0

10 5 0 0.01

(a)

0.1

1 ND,w / NA,a

10

100

Figure 3.6 (a) Simulated solar cell parameters as a function of the doping ratio ND,w /NA,a of window and absorber layer. Window/absorber heterostructure as depicted in Figure 3.2a with conduction band offset Ec w,a = 0 eV. Other device parameters as in default Table 8.1. Nominal interface recombination velocities Sn0,p0 = 103 , 104 , . . . , 107 cm s−1 (direction of arrows) realized by a single interface defect

energetically located in the middle of the absorber bandgap at the interface. (b) Electron and hole reduced chemical potentials En,az=0 , Ep,az=0 as a function of the doping ratio ND,w /NA,a simulated at equilibrium. (c) Electron and hole concentrations at the absorber side of the heterojunction under AM1.5G illumination at zero electric bias (na o z=0 , pa 0 z=0 ) and at forward bias of 0.7 V (naz=0 , pa ). z=0

potential drop is mostly in the absorber layer since Ep,az=0 > Eg,a /2. The absorber is in inversion and the hole barrier is large. With an interface recombination velocity of 103 cm s−1 , the cell efficiencies are well above 18%, irrespective of the value of Ep,az=0 (see Figure 3.6a). The devices are mainly limited by bulk recombination. Whether the larger portion of the band bending is found in the window layer (Ep,az=0 < 0.6 eV) or in the absorber layer (Ep,az=0 > 0.6 eV), approximately has no influence on Voc , Jsc , and


FF.5) With increasing interface recombination velocity the device performance starts to become sensitive to the distribution of the band bending and thus to Ep,az=0 . The Voc first exhibits a minimum at about Ep,az=0 = 0.6 eV. With increasing Sn0,p0 , this minimum becomes deeper and directly translates into a minimum of the solar cell efficiency. From Figure 3.6c the reason for the minimum in Voc becomes apparent: for Ep,az=0 = Eg /2, the carrier densities directly at the interface become approximately equal at forward bias. According to Eq. (2.67), this is the condition for a maximal recombination rate provided that Sn0 ≈ Sp0 .6) In the case of Ep,az=0 = Eg,a /2, one of the carriers becomes a minority carrier at the interface, recombination decreases and a higher Voc can be obtained. For not too large Sn0,p0 , the Voc is symmetric in Ep,az=0 . A different behavior is observed for Jsc . In the case of band bending exclusively in the absorber, that is, Ep,az=0 > Eg,a /2, Jsc is largely independent of Sn0,p0 . If, on the one hand, the band bending is mostly in the emitter, that is, Ep,az=0 < Eg,a /2, Jsc becomes very sensitive to Sn0,p0 . Thus, Jsc is non-symmetric with respect to Ep,az=0 = Eg,a /2 and, as a consequence, the efficiency η is also non-symmetric. The reason can be found in the interface recombination of electrons which are photo generated in the absorber layer. In the extreme case of Ep,az=0 < 0.4 eV, photo generated electrons arrive at the interface as minority carriers (naz=0 and paz=0 in Figure 3.6c). Since holes are abundant at zero bias, recombination can become efficient at high Sn0,p0 . At forward bias, on the other hand, the difference between naz=0 and paz=0 at Ep,az=0 = 0.4 eV again is very large. Now, minority carrier (electron) injection from the emitter into the base is difficult due to the large band bending in the emitter. This is why Voc starts to recover for Ep,az=0 < 0.4 eV. The small asymmetry of Voc with respect to Ep,az=0 = Eg,a /2 originates from the asymmetric Jsc .7) Figure 3.6 reveals that the concept of a solar cell where the electric field mostly extends in the absorber is the preferred one. Even for high nominal interface recombination velocity of Sn0,p0 = 107 cm s−1 , the cell efficiency can be large provided that Ep,az=0 ≈ Eg,a , that is, if the absorber is fully inverted. Then electrons arrive at the interface as majority carriers both as the diode current at forward bias and as the photo current at zero bias. In a fully inverted absorber, recombination 5) Certainly, this conclusion is only valid if the

7) A solar cell where the electric field mostly

carrier lifetime is high and the collection region is not limited to the SCR. Otherwise Jsc and FF would decline for higher Ep,az=0 and thus for a smaller SCR being the main collection region. 6) Not for any type of heterojunction the minimum in Voc is found at Ep,az=0 = Eg /2. The exact position of the minimum of Voc depends on the density of states. Further on, it depends on the ratio of the nominal interface recombination velocity. For Sn0 < Sp0 (Sn0 > Sp0 ), the position where Voc becomes minimal shifts to Ep,az=0 > Eg /2 (Ep,az=0 < Eg /2).

extends within the emitter is referred to as injection solar cell. Photo generation takes place in one semiconductor and carrier separation by the electric field takes place in the other semiconductor. The principle of an injection solar cell is realised in the Cu2 S/CdS heterojunction [105]. Requisites for high performance are the carrier collection by diffusion in the absorber layer (since there is no SCR) as well as an interface with very low interface recombination velocity.

139

140


of the electrons is limited by the availability of holes. Without Fermi level pinning, inversion of the absorber surface is accomplished by a high absorber/emitter doping ratio. This allows: Rule 3: In order to induce strong inversion of the absorber surface, the doping ratio between emitter and absorber should be high. Rule 3 is also valid for an absorber/buffer/window heterostructure. It also applies to heterostructures with Ec = 0. In fact, Rule 3 has already been regarded in the discussion of Section 3.1. If there is a barrier for the photo current, that is, Ec w,a or Ec b,a > 0, the requirement of Ep,az=0 ≈ Eg,a is even more important (see Section 2.5.6) since it also provides a high carrier density for thermionic emission. In case that there is no Fermi level pinning and Ec w,a = 0 (absorber/window type), the activation energy of J0 would correspond to the absorber bandgap. Only the diode quality factor would depend on the doping ratio. It would change from A ≈ 2 for Ep,az=0 ≈ Eg,a to A ≈ 1 for Ep,az=0 = Eg,a /2 according to Eq. (2.111). Principally, the same holds for the absorber/buffer/window heterostructure where A is given by Eq. (2.113). Complying with Section 3.12, the emitter should also be highly doped because of its function as the front electrode of the solar cell or module. With a required sheet conductance of 5 /sqr. and typical mobilities in the range of 40 cm2 V−1 s−1 , carrier concentrations exceeding 1020 cm−3 are necessary. In the following, we assume that the window layer doping is ND,w = 1020 cm−3 .

3.4 Fermi Level Pinning

So far, we have assumed that the position of the Fermi level at the interface is completely determined by the doping ratio of absorber and emitter. In Section 2.3.4, however, it is shown that the Fermi level position at the interface can likewise be determined by interface charge. If the interface charge is high enough, the doping ratio between window and absorber has no influence on Ep,az=0 . This situation is denoted as Fermi level pinning. Here, for clarity, we assume two types of interface states at the absorber/window interface. Neutral states which cause recombination of carriers and charged states which induce Fermi level pinning. It may be argued that Fermi level pinning is not a solar cell design instrument since the interface density is not in the hand of the experimentalist. However, Fermi level pinning has large ramifications and may become controllable in the future. The principle dependence of the solar cell parameters on Ep,az=0 or En,az=0 in the case of Fermi level pinning is the same as depicted in Figure 3.6. The smaller Ep,az=0 , the smaller is Voc in case of high Sn0,p0 . A numerical example is given in Figure 3.7 where we calculated Voc after Eq. (2.191). We see that if the Fermi level position at the absorber surface is fixed by acceptor-like interface states, the Voc can become lower than if it is fixed by the absorber/window doping ratio. This is so, although the hole barrier is identical.

3.4 Fermi Level Pinning

Ep,az=0 [eV] 1.20 0.7

1.10

1.00

1.90

1.80

0.2 0.3 En,az=0 [eV]

0.4

Voc [V ]

0.6 0.5 0.4

∆Ec=0, NIF =0

0.3

∆Ec=0, NIF−>0 ∆Ec 1017 cm−3 due to tunneling enhanced recombination and Auger recombination. In addition, there is a general trend of decreasing η with increasing NA,a for large bandgap absorbers. This is due to the relative importance of effects (1) and (3) in the effect listing at the top of this paragraph. The 60 mV per decade gain in Voc due to increasing NA,a is relatively less 8) In accordance with Section 2.4.4 we se-

lected maximum values for the Auger coefficients. 9) The tunneling mass is still a quantity of debate. Hurkx gives a value of 0.1 × me for silicon [48]. Since the tunneling electrons

and holes have energies much lower than the band edge energies, the tunneling mass is smaller than the band edge mass [106]. Values as low as 0.01 × me for silicon can be found in the literature [107].

145


1.4

Voc [ V ]

1.2

Eg =1.72 eV 1.51 eV

1.0 1.32 eV 0.8 0.6

1.15 eV 1.02 eV

Jsc [mA cm−2]

40

30

20

0.9

FF

0.8 0.7 0.6 0.5 22 20 η [%]

146

18 16 14 1014

1015 1016 NA,a [cm−3]

1017

Figure 3.9 Simulated values of Voc , Jsc , FF, and η for an absorber/window heterojunction with ND,w = 5 × 1019 cm−3 without interface recombination as a function of the absorber bandgap Eg . Case with tunneling enhanced recombination plus Auger recombination.

The conduction band offset is Ec w,a = 0 eV. Other device parameters according to default Table 8.1. The absorber bandgaps are Eg = 1.02, 1.15, 1.32, 1.51, 1.72 eV in direction of arrow.

3.6 Absorber Thickness

important for larger bandgaps. Yet, the Jsc loss due to decreasing SCR is relatively more important for larger bandgaps. As was discussed in Section 2.3.6, deep defects can become charged within the depletion region and thus can alter the band diagram. Consider, for instance, acceptor like deep states in the absorber. Since they become ionized in the depletion region of the p-type absorber, they add to the space charge and increase electric field and band bending. Thus deep acceptor states must be counted in addition while applying Rule 4. If these deep defect induce a strong lattice relaxation, they may be persistently charged by electrons and thus induce metastable effects (see Section 6.2.4). For the subsequent design studies (see Table 3.1), we set the absorber doping to NA,a = 1016 cm−3 .


The absorber thickness can influence the cell parameters in various ways. Upon reduced thickness: 1) 2) 3) 4)

The collection length can reduce thus deteriorating Jsc . The impact of back contact (BC) recombination can increase. The width of the recombination zone can be reduced, thus increasing Voc . In the case of a highly reflecting BC the local generation rate may be increased, thus increasing Voc .

Fabricating a chalcogenide solar cell with a minimum thickness of the absorber layer has two potential advantages: it can reduce production costs (lower material consumption, higher throughput) and it can decrease recombination losses. In a thoughts experiment, we start with a thick absorber layer which develops the full solar performance defined by its bulk and interface properties. By gradually reducing the absorber thickness, we may find the minimum thickness where first, the photo current generation is still high, and second, the minority carrier recombination at the BC is still negligible. An overview of optical and electronic processes influencing the photo current generation in a chalcogenide solar cell is given in Figure 3.10. What is the optical influence of the BC? Photons can either be absorbed within the BC layer or they can be reflected. The reflected photons again are either absorbed during the second light pass through the absorber layer or escape the device by transmission through the emitter (multiple reflectance is another possibility). Thus, reflection and absorption describe the optical influence of the BC. Naturally, they depend on the reflectance and absorbance of the BC layer. Reflection and absorption can as a first approach be considered as independent from reflection and absorption in the emitter. Below, we differentiate 0 and 100% reflection (transmission through the BC is neglected). The electronic influence of the BC can be expressed by the BC recombination velocity. This recombination velocity influences both the diode current and the

147

Window

Buffer

Absorber


Back contact

148

Back contact reflection Emitter reflection

AM 1.5G

Parasitic absorption Recombination

Photo current

Figure 3.10 Loss mechanisms reducing the photo current in an absorber/buffer/window solar cell (after [34]).

photo current. In case that there is no band bending at the BC we have to deal with the recombination velocity of the minority carriers, here electrons. We differentiate low and high back surface recombination velocities, that is, Sn0,bc = 103 cm s−1 and Sn0,bc = 107 cm s−1 . In order to demonstrate the influence and interrelation of optical and electronic BC properties, we show simulated solar cell parameters as a function of the absorber thickness.10) For Sn0,bc = 103 cm s−1 and RBC = 100% (Figure 3.11a), Voc indeed increases with decreasing da except for the case τn0,p0 = 1 µs. In the case of τn0,p0 = 100 ps, the increase in Voc is found at very low da where the absorber is completely depleted and band bending is defined by the inverse film thickness. As explained in Section 3.5, SCR recombination decreases and thus Jdiode,SCR decreases with steeper bands and thus with reduced film thickness. In the case of τn0,p0 = 10 ns and Sn0,bc = 103 cm s−1 , recombination in the QNR is dominant and becomes reduced with reduced absorber thickness. This is so in despite of the reducing photo current. Finally, in the case of τn0,p0 = 1 µs, the diode current is limited by BC recombination for all thicknesses [see Eq. (2.152)]. Here, Voc decreases with decreasing da due to the reduction in Jph . The short circuit current density, is found to decrease for da < 1 µm. This can be rationalized by the average absorption length of band edge photons of 1/α. Thus, for the absorption coefficient α0 = 104 cm−1 typical for several chalcogenide 10) Note that the simulated cells in Figure 3.11

exhibit a highly inverted absorber due to the high doping ratio. The nominal recombination velocity at the absorber/window

interface is Sn0,p0 = 105 cm s−1 . With Ec = 0 eV, interface recombination is not the dominant process.


149

Voc [V]

0.8 0.7 Sn0,bc =103 cm/s

Sn0,bc =103 cm/s

RBC =100%

RBC =0%

Sn0,bc = 107 cm/s

0.6

Sn0,bc = 107 cm/s

Jsc [mA cm−2]

40 35 30 25 20

RBC = 0%

RBC = 100%

FF [%]

15 85 80 75 70 65

η [%]

25 20 15 10 0.1 (a)

2

4 6

1

da [µm]

2

0.1 (b)

2

4 6

1

da [µm]

Figure 3.11 Solar cell parameters for an absorber/window heterojunction. For the material parameters see default Table 8.1. The carrier lifetime as the family parameter is τn0,p0 = 100 ps (solid line), 10 ns

2

0.1 (c)

2

4 6

1 da [µm]

2

0.1 (d)

2

4 6

1

da [µm]

(dashed line), and 1000 ns (dotted line). (a) Sn0,bc = 103 cm s−1 , RBC = 100%, (b) Sn0,bc = 103 cm s−1 , RBC = 0%, (c) Sn0,bc = 107 cm s−1 , RBC = 100%, (d) Sn0,bc = 107 cm s−1 , RBC = 0%.

absorbers (see Table 4.5), a limiting absorber thickness of 1 µm may be identified. Such a value has been encountered in several publications [110–112]. Nevertheless, we refrain from formulating a design rule for the limiting absorber thickness since, according to Figure 3.11, it strongly depends on Sn0,bc and RBC . A drastic example is a cell with Sn0,bc = 107 cm s−1 and τn0,p0 = 10 ns in Figure 3.11c,d where the efficiency already declines for da < 3 µm, irrespective of the BC reflectance. For the design discussion in the following sections, we continue to anticipate an absorber thickness of 1.5 µm. The example in Figure 3.11 reveals that a reduced absorber thickness must be compensated by an improved BC reflectance and a reduced BC recombination. At very low absorber thickness, it is advised to apply light trapping structures at the BC (for an overview see Ref. [65]). Recombination may be reduced by a band edge gradient (see Section 3.10) or a back surface field (see Section 3.10).

2

150


3.7 Grain Boundaries

In Section 2.4.5.4 we treat the absorber surface, that is, the absorber/emitter interface as a special zone where the (volume averaged) recombination rate can be higher than in the absorber bulk. It is similar with the absorber BC. In a polycrystalline absorber also grain boundaries (GBs) can exhibit a recombination rate higher than in the bulk. In addition, these GBs may be charged and induce band bending. GBs are 2-dim recombination zones which can be oriented more parallel or more perpendicular to the main junction [see markers (a) and (b) in Figure 1.1]. They are characterized by a nominal GB recombination velocity, Vn,p = Ngb σn,p Sn0,p0,gb , much like the absorber/emitter interface. For given values of Sn0,gb and Sp0,gb the recombination rate can be calculated after Eq. (2.67), provided that the carrier concentrations at the GB are known. Green developed upper bounds for GB effects assuming the carrier densities n,p in the presence of a GB to be identical to the case without GB [113]. This overestimates GB effects since GB recombination reduces the carrier concentrations with respect to the GB-free case. Using 3-dim device simulation, Gloeckler et al. was able to quantify the impact of vertical and horizontal GBs on solar cell performance [114, 115]. Much is learned from an horizontal 2-dim GB which is located at a distance zgb from the main junction, that is, parallel to the junction between emitter and absorber. Initially, the GB may not be charged (neutral GB) and it may exhibit Sn0,gb = Sp0,gb = Sgb due to equal capture cross sections for electrons and holes. Note that we consider the recombination states, which define Sgb , to be independent from charged states which induce band bending. We are interested in the position zgb of maximal impact on the device performance. Figure 3.12a shows contour plots of normalized solar cell parameters for a neutral GB. For a given Sgb , the parameters Voc , FF, and η show a distinct minimum if the GB is located at zgb ≈ 90 nm. Figure 3.13b reveals that this position coincides with the location of n ∼ p at forward bias (V = 0.65 V). Precisely spoken, a horizontal GB has the strongest impact if it is located at the position Sn0,gb na = Sp0,gb pa . Up to Sgb = 105 cm s−1 , a neutral GB at the edge of the SCR reduces the efficiency by maximal 5%. Located deep in the QNR, the influence of a horizontal GB is even lower. Note that beyond the position of the horizontal GB – deeper in the absorber – the splitting of the quasi Fermi levels is largely unaffected. It may be surprising that Jsc is only marginally affected by a horizontal GB located in the SCR. The reason is that at zero bias, the carrier densities at the location of Sn0,gb na = Sp0,gb pa are small. This can be seen in Figure 3.13a. Only for very high Sgb , Jsc tends to be reduced. Then, the minimum of Jsc also is for a GB at the location of Sn0,gb na = Sp0,gb pa . If the GB is located in the QNR, the loss of Jsc can be modeled by an effective lifetime obtained from 1/τeff ,n = 1/τn0 + Sn0 /da (and accordingly for 1/τeff ,p ).


107 0.75

0.9

0 .6 5

0.85

106

5

105

0.95

104

Voc

103 107

85

0.85

0.65

0.

85

0.95

0.

106

0.95 105 Jsc

103 107

0.8

5

0.55

106

0.95

Sgb [cm/s]

104

0.95

105

0.85

0.95

104

FF

103

103 (a)

η 10−6

10−5 zgb [cm]

0.65

0.9

0.95

0.95

5

0.75

0.85

0.8

5

104

5 0.6

105

3 0.

5

7 0.

106

0.3

5

107

5

10−4

Figure 3.12 Normalized solar cell parameters Voc , Jsc , FF, and η (from top to bottom) from device simulation for (a) neutral and (b) charged grain boundary located at zgb in the absorber layer. The positive charge density is 2 × 1011 q cm−2 . Material parameters

10−6 (b)

10−5 zgb [cm]

10−4

as in default Table 8.1. The grain boundary was realized by an equivalent layer of thickness dgb = 2 nm with Nd = NIF /dgb recombination centers. Capture cross sections of 10−12 to 10−8 cm2 mimic grain boundary recombination velocities of 103 –107 cm s−1 .

151


Energy E [eV]

4

3 EFn

EFp

2

n,p [cm−3]

1015

p n

1010 GB rec. 1022

Rn [cm−3 s−1]

152

1020 1018 1016 −0.4

(a)

0.0 −0.2 z [µm]

Figure 3.13 Band edge energies and quasi Fermi energies, carrier densities, as well as electron recombination rate, R, as a function of depth in an absorber/window heterostructure. Material parameters as in default Table 8.1 with Sgb = 105 cm s−1 . The neutral grain

−0.4

0.2 (b)

0.0 −0.2 z [µm]

0.2

boundary at dgb = 90 nm was realized, as described in the caption of Figure 3.12. (a) V = 0 V and (b) V = 0.65 V roughly corresponding to the simulated Voc of this device. The position of dgb is characterized by a peak in R.

Thus, the effective lifetime model mimics the GB states evenly distributed within the absorber layer. The model is a good approximation for Sn0,p0 up to 105 cm s−1 [114]. Similar as for interfaces, the GB states may be charged. Negative charge can be provided by acceptor states close to Ev . They would induce upward band bending and an electron-repulsive barrier (Figure 3.14a). In a p-type absorber, GB acceptor states of very high density would be needed since, due to a small Ep,az=zgb , their occupation would be low. Thus, a negatively charged GB is unlikely in p-type absorbers. A positively charged GB (more likely in a p-type absorber) leads to an hole-repulsive and electron-attractive barrier (Figure 3.14b). It would be due to high lying charged donor states. Since the positively charged GB attracts electrons, we expect it to enhance GB recombination [113]. In Figure 3.12b we fix the area charge density of the GB to 2 × 1011 q cm−2 and vary the location of the GB. In the QNR this area charge induces a band bending


Ec

Ep,a

z=zgb

Ep,a

EF

−qVbigb

Ev qVbi

(a)

(b)

gb

(c)

qVbigb +∆Ev (d)

Figure 3.14 Sketch of the band diagram perpendicular to a grain boundary showing (a) upward band bending due to negative GB charge, (b) downward band bending due to positive charge, (c) a layer with increased bandgap realized by a valence band offset, and (d) combined accumulation and depletion zones [116]. gb

of qVbi = 0.11 eV. We see that the principle dependence of the cell parameters on zgb is not altered: maximum impact is on Voc and FF. The minimum in η is shifted to slightly larger zgb . This is due to the extended SCR in the presence of positive charge. Figure 3.12 reveals that the impact of a charged GB for a given Sgb is larger than for a neutral GB. Note, however, that the GB band bending induced by the fixed charge is much smaller in the SCR than in the QNR due to lower carrier concentration. Calculating the solar cell parameters in Figure 3.12 assumed that the horizontal GB is exactly parallel to the main junction. Certainly, this is an artificial assumption since in reality ‘horizontal’ GBs may show some inclination with respect to the junction plane. Nevertheless, we may draw the conclusion: Rule 5: Horizontal GBs in the SCR should be avoided. As seen above, a GB located outside of the SCR has a minor impact on the device performance as long as Sgb is not too large. Rule 5 may favor substrate-type solar cells over superstrate-type solar cells: In substrate-type solar cells, absorber material from the initial growth phase exhibiting a high GB density is at the BC. Thus, substrate-type solar cells may exhibit less GBs in the SCR. We note that if low Voc and FF but high Jsc are diagnosed for a particular thin film solar cell, a high density of horizontal GBs could be the reason. This is similar to interface recombination. Recombination at horizontal GBs in the SCR will lead to a diode quality factor close to 2. The saturation current is activated by the GB bandgap, normally, Eg,a . Thus, device analysis does not give a hint for GB recombination, besides the discrepancy of high Jsc and low Voc (see Section 7.2.2). The impact of vertical GBs can now be understood based on the knowledge about horizontal GBs. If the vertical GB runs through the complete absorber, recombination is highest in the range of Sn0,gb na = Sp0,gb pa , thus where the vertical GB passes the SCR. Furthermore, Voc and FF is more affected than Jsc . This is

153


Voc , Jsc , FF, η (norm.)

1.0

A

154

Jsc 0.9

FF

0.8

Voc

0.7 0.6

η

0.5 2.0

A

1.5 1.0

102

103

104

105

106

107

Sgb [cm s−1] Figure 3.15 Neutral grain boundary recombination on columnar grain boundaries. Data from 2-dim device simulation. The grain size (distance between two grain boundaries) is 1 µm. All parameters are normalized to a cell without grain boundaries with absorber electron lifetime of 2 ns. After Gloeckler et al. [115].

visualized in Figure 3.15, obtained by 2-dim device simulation with 1 µm grain size [115]. The efficiency loss at high Sgb is mainly due to Voc and FF. Certainly, for a given Sgb the reduction in η for a vertical GB in Figure 3.15 is less than for a horizontal GB located in the SCR (Figure 3.12) simply due to geometrical reasons. The example of Figure 3.15 reveals that, for 1 µm grain size, high solar cell performance can only be achieved if the recombination velocity of a vertical GB does not exceed Sgb = 103 cm s−1 with τn0,p0 = 0.2 ns. This may be formulated as: Rule 6: The effective GB recombination velocity for 1 µm absorber grain size shall not exceed the bulk equivalent velocity [see Eq. (2.104)]. We use the term effective Sgb since band bending and band offsets at the GB can modify the nominal GB velocity. We admit, however, that Sgb eff is not well defined. For comparison, some GBs in multicrystalline Si have recombination velocities Sn0,p0 above 106 cm s−1 [117]. If the vertical GB in a p-type absorber is positively charged (hole-repulsive band bending), the GB attracts electrons and recombination increases – exactly like in a horizontal GB. At large band bending, the GB inverts and the pn junction extends from the absorber/emitter interface along the grain surfaces deep into the absorber layer. This can cause shunting of the cell by an n-type channel (see Figure 3.16) [115]. Strong inversion, in contrast, can reduce GB recombination at zero bias due to hole depletion. For an inverted GB, the recombination rate at zero bias reads Rgb = pa Sp0,gb . Since pa is low, a large hole-repulsive GB exhibits a large Jsc . Due


0.0 (a) 0.5

p~n

n >> p

1.0

z [µm]

1.5 p >> n

V=0 0.0 (b) 0.5

p~n 4 1.0

0

n>p

−4 1.5

p >> n V = 0.5 V 0.2

0.4

0.6

0.8

x [µm] Figure 3.16 Log(n/p) along a vertical GB by JE100 with spectrum equivalent AM1.5. from 2-dim device simulation: (a) at V = 0 V Sgb = 105 cm s−1 . Absorber doping NA.a = and (b) V = 0.5 V forward bias. GB lo2 × 1016 cm−3 , Eg,a = 1.15 eV. Gray scale cated at x = 0.5 µm and absorber/emitter represents log(pa /na ). From Ref. [115] with junction located at z = 0 µm. Illumination permission.

to the collecting properties of the n-type channel, Jsc can even exceed the GB free case. However, at forward bias the inversion is reduced due to partial occupation of the (high lying) donor states. The hole concentration increases up to the condition pa ∼ na . Thus, a strong hole-repulsive GB alone is not appropriate to deactivate GB recombination. A means to stabilize a hole-repulsive barrier at the GB is to reduce the valence band edge at the GB. As a consequence the majority carrier concentration is reduced by the factor exp{Ev /kT}. This has been studied by Gloeckler et al. [115] and Taretto et al. [118] using 2-dim device simulation. A lowered valence band edge may be combined with a positive band bending. A sketch of a GB cross section with band bending and valence band offset is shown in Figure 3.14c. Provided that the sum of Ev + qVbi gb is large enough, recombination at a vertical GB can be largely suppressed. The result is a solar cell which is hardly affected by vertical

155

156


GBs. Figure 3.14d gives the GB band diagram which combines an electron barrier with hole depletion [116].

3.8 Back Contact Barrier

At the back surface of an absorber, a contact barrier may form due to Fermi level pinning or due to the particular band line-up with the BC band structure (see Figure 2.26b). Thus in principle, we are concerned with a Schottky barrier of barrier height φb p where the index p denotes a barrier for holes. In the equivalent circuit, a BC barrier can be described by a BC diode (see Section 2.6.1) as long as the BC barrier does not interfere with the main diode. That is, as long as the depletion regions of the two diodes do not overlap. The BC diode is of opposite polarity than the main diode. The photo current flows through the back diode as a forward majority carrier current. In order to admit this current flow, the BC diode needs to be forward biased. This bias voltage is supplied by the main diode which thereby itself becomes forward biased. The bias voltage is Vm = −Vb where m and b denote the main and back diode, respectively. Figure 3.17a shows a heterojunction with BC diode at short circuit. The diode equation for the BC diode yields −Vb = kT/q ln(1 + Jph (Vb )/Jb0 )

(3.3)

where Jb0 is the saturation current of the back diode. Thus, −Vb increases with decreasing Jb0 . If we assume thermionic emission limited transport over the BC barrier, we can use Eq. (2.126) to see that −Vb increases with increasing φb p : the larger the barrier height, the larger is the voltage drop on main and back diode under short circuit. In Figure 3.17a, Vb equals the difference in the electric potentials ϕ − ϕ0 taken in the center of the absorber where ϕ0 stands for the non-illuminated case. Using device simulation, we investigate the principle influences of a BC barrier on cell parameters. In Figure 3.18, we differentiate between different absorber thicknesses. First, we see an increase of Jsc with increasing da . This is an independent effect of the absorber thickness which is only due to an increased charge collection. However we find that for our default device, the effect of φb p on Jsc is negligible: Although the main cell is forward biased under short circuit (see Figure 3.17a), the effect of voltage dependent current collection is too small to depress Jsc . This is different for the other solar cell parameters. The FF starts to degrade for φb p > 0.3, independent from the absorber thickness. Now, this is indeed the effect of voltage dependent charge collection. And as the forward bias of the main junction is similar for all thicknesses in Figure 3.18, the FF loss is independent of da . Considering the influence of φb p on Voc , we have to discriminate between different absorber thicknesses. For thick absorbers, Voc is hardly influenced even for φb p = 0.4 eV. The reason is that the main and BC diode can be considered as two independent devices: At Voc , no current is passing through the combination of two anti-parallel diodes. No current can be blocked by the BC

3.8 Back Contact Barrier

−qϕ0 1

1

− qj

Ec

−4 −5

Ev EFp

−7 1018

1016 n

1012 1010 Jn

−20

Jp

Jtot

−30 −1.5

1014 1012

20 J [mA cm−2 ]

J [mA cm−2 ]

−10

1016

1010

0

(a)

−5

−7

1014

−qϕ0

−4

−6

p

−qϕ

−3

−6

1018 n,p [cm−3]

Energy E [eV]

−3

0 EFn

n,p [cm−3]

Energy E [eV]

0

−1.0

−0.5

0.0

Position z [µm]

0

Jtot

−10 −20 −1.5

0.5

Jn

10

−1.0

(b)

JP −0.5

0.0

0.5

Position z [µm]

Figure 3.17 Solution curves for an absorber/window heterojunction with back contact barrier (a) under short circuit condition and (b) at open circuit (AM1.5). Device parameters as in default Table 8.1 but with −qφb = 0.5 eV as back contact barrier height. qϕ0 gives the potential energy for the non-illuminated case.

diode and the recombination of light generated carriers is only governed by the main diode. For our default device (see Table 8.1) we have Ln,a = 1.13 µm and wa = 0.33 µm. Thus the criterion is approximately da /(Ln,a + wa ) ≥ 3. For thinner cells, φb p = 0.3 eV already leads to some loss in Voc . A larger BC barrier height further decreases the cell performance. Comparing Figure 3.17b with Figure 2.19b (no BC barrier), the slope of the electron quasi Fermi level toward the BC is much stronger. The BC barrier leads to enhanced minority carrier transport to the BC and thus to higher recombination both in the bulk and at the BC [54, 55]. In other words, the main diode and the BC diode are not independent. A BC barrier together with minority carrier recombination at the BC can lead to a cross over between dark and light JV curves already for moderate barrier heights

157


0.4

4

0.73

4

0.7

0.73

8

0. 73 2

0.3 0.2

36.6

0.0 0.3

78.8

78.4

78.2

78

77.6

77.8 79

36.8

Jsc [mA cm−2]

35 .8

0.2 0.1

Voc [ V ]

36.4

36

35.4

0.

36.2

0.0 0.3

6

73

0.1

φbp [eV]

78.6

79. 2

0.2 0.1

21 .6

21.2

21.4

20

0.0 0.3

FF [%]

0.1 0.0

.8

0.2 η [%]

20

158

2

4

6 da [µm]

8

Figure 3.18 Simulated solar cell parameters of an absorber/window heterojunction with back contact barrier of barrier height φb p . Contour plot showing the influence φb p and the absorber thickness da . Other material parameters as in Table 8.1.

[119]. Large BC barriers may be diagnosed by a roll over in the JV curve or, which is equivalent, by a saturation of the current density under forward bias. If the forward current approaches the reverse saturation current of the back diode, Jb0 , any further increase in V drops at the back diode, that is, V = Vb . Thus Vm remains constant and the current saturates. If the barrier height φb p is small, the rollover effect is only observed at low temperature. Can we formulate a general rule for the maximum tolerable barrier height at the BC? Figure 3.18 suggests that a barrier height of Eg,b > Eg,a . Thus, absorption of photons with wavelengths between λEg,w and λEg,a in the buffer layer may take place. Depending of the collection function of the buffer layer this may induce a collection loss. The collection loss can be calculated as a current loss by use of Eq. (2.147) if we replace ηc (z) by [1 − ηc (z)]. Let us assume that the collection function in the n-type buffer layer is small, that is, photo generated holes recombine in the buffer volume or at the buffer/absorber interface. Then, the collection (recombination) loss scales with the buffer layer thickness. There may be a buffer thickness up to which the optical gain compensates the collection loss. This thickness will depend on the specific optical parameters of window, buffer, and absorbers. Thus, a general rule cannot state a particular buffer layer thickness.

159


1.0

∆Jsc [mA cm−2]

160

opt

∆Jsc

0.0 ∆Jsc −1.0 opt

∆Jsc −2.0 0

20

40

60

80

100

CdS Thickness [nm] Figure 3.20 Simulated values of the optical gain Jsc opt , the collection loss Jsc col and the total variance Jsc = Jsc opt + Jsc col as a function of the thickness of a buffer layer in a CuIn0.7 Ga0.3 Se2 /CdS/ZnO heterostructure (after [34]). Refractive indices and other device parameters are similar as in Table 8.1.

It may be instructive to consider the example in Figure 3.20 as developed by Orgassa [34]. For the material parameters chosen, a buffer layer of up to 40 nm hardly induces a total optical loss. About this buffer layer thickness is indeed realized in typical chalcogenide solar cells. Certainly, it is a requisite that the thickness is appropriate to serve other functions of the buffer layer, namely as a surface passivation and a diffusion barrier. In Section 3.3, we develop Rule 3, which states that the emitter should be much more highly doped than the absorber. What is the meaning of this rule if the emitter consists of a combination of layers? In principle, Rule 3 applies to the layer in direct contact to the absorber layer, which is the buffer layer. Thus, the n-type buffer layer should be doped higher than the p-type absorber layer. However, this may not be feasible due to experimental limitations. In addition, there is evidence that a thin resistive buffer layer improves the performance of cells with lateral fluctuations of the diode current [61]. We come back to this point. For the rest of this section let us assume that doping in the buffer layer is low and that the buffer rather acts as a dielectric layer. Then, as explained in Section 2.3.3, there is a potential drop within the buffer layer, much as sketched in Figure 3.2c. This potential drop leads to a decrease in Ep,az=0 (see Section 2.3.3). For the case of NA,a = 1016 cm−3 we can read out a decrease of about 0.1 eV in Ep,az=0 from Figure 2.7c for db = 40 nm. Figure 3.6a indicates that such a decrease in Ep,az=0 combined with a high nominal interface recombination velocity can induce a noticeable decrease of Voc , even for a high absorber/window doping ratio. This consideration underlines that the thickness of a buffer layer with low doping is critical for the device performance as long as there is no Fermi level pinning. In the case of Fermi level pinning close

3.9 Buffer Thickness

to En,az=0 = 0 by donor-like states, the potential drop in the buffer will have no influence. However, even without Fermi level pinning there exists the possibility to compensate the buffer layer potential drop by a positive sum of the conduction band offsets, that is, χ = Ec w,b + Ec b,a > 0. All of this is expressed by Eq. (2.35). A further possibility to compensate for the potential drop in the buffer layer is to widen the front bandgap of the absorber. This is the subject of Section 3.10. As mentioned above, a lowly doped buffer layer can cure the impact of mesoscopic fluctuations of the diode current. These fluctuations may well be present at least in industrially fabricated cells [86] (see also Section 2.4.9). Rau et al. [61] showed that adding a local series resistance to the cell limits the influence of spatially non-homogeneous diode currents. Series resistances up to 1 cm2 are appropriate depending on the level of non-homogeneity. The buffer layer can act as a local series resistance of the cell. Consider, for instance, a layer of 40 nm thickness, carrier mobility of 1 cm2 V−1 s−1 and ND,b = 1015 cm−3 . With a calculated series resistance of 50 m cm2 , this layer is appropriate to partly decouple regions of different diode current. Within our study logic (see Table 3.1) we now continue to investigate heterostructures of the absorber/buffer/window type. The buffer layer in the simulated or calculated examples shall have a thickness of 50 nm and a doping concentration of ND,b = 1015 cm−3 . Thus according to Figure 2.7, the buffer acts like a dielectric and there is a potential drop on the buffer. The energy band diagram of this cell is depicted in Figure 3.21a. Ep,az=0 of this cell does not approach Eg,a and accordingly the cell’s Voc is not optimal. A solution for this problem is outlined in Section 3.10.

Ec

Ec

Ev

Ev

−wa (a)

−wa

0db Position z

(b)

Figure 3.21 (a) Energy band diagram of an absorber/buffer/window cell with potential drop in the buffer layer and without front surface gradient. This cell is the starting point for the following design studies

0db Position z and will be further modified. (b) Similar cell as in (a), but with Ep,az=0 ≈ Eg,a due to Fermi level pinning. With a bulk lifetime of τn0 = 10 ns, the efficiency of cell (b) will be limited by bulk recombination.

161


3.10 Front Surface Gradient

A front surface bandgap gradient aims at increasing the open circuit voltage of a solar cell while retaining the short circuit current. If the bandgap at the location of highest recombination is increased, the diode current will decrease and the open circuit voltage will rise. First, we discuss the case of optimizing a thin film cell the Voc of which is limited by IF recombination. According to Figure 3.4, IF recombination increases for Ep,az=0 Eg,a . The absence of absorber inversion with Ep,az=0 Eg,a can be a consequence of the potential drop across the lowly doped buffer layer [see Eq. (2.35)] or of an unfavorable FLP position. In the example of Figure 3.22a, the cell has En,az=0 = 0.33 eV or Ep,az=0 = 0.87 eV due to the potential drop in the buffer layer. The band diagram of this cell is illustrated in Figure 3.21a. The cell efficiency is limited to 17.6% by interface recombination (Voc = 0.66 V, Jsc = 35.2 mA cm−2 , FF = 75.9%) as here we put

(a) 0.9 0.85

Voc [ V ]

0.2

0.8 0.75

0.1 0.0 0.3

0.7

(b)

0.2

26

28

−2

Jsc [mA cm ]

30

0.1

32

35

72

68 70

0.1

FF [%]

(d)

0.2

η [%]

0.1

20

0.0 5

16 17

19

6 7 89

18

0.3

64

78

0.2 0.0

66

0.3

76

(c)

34

79

0.0

74

0.3

−∆Ev,front [eV ]

162

18

2

3

4 5 6 7 89

0.1

1 dmin [µm]

Figure 3.22 Simulated solar cell parameters Voc , Jsc , FF, and η (from a–d) for a linear front bandgap gradient in the absorber as a function of Ev,front and dmin (for definition of dmin see Figure 2.14). Device parameters according to default Table 8.1 but with Sn0,p0 = 107 cm s−1 . The bottom line

cell has a potential drop in the buffer layer and therefore exhibits En,az=0 = 0.33 eV. This cell is mainly limited by interface recombination with the following parameters Voc = 0.657 V, Jsc = 35.2 mA cm−2 , FF = 75.9%, η = 17.6%.

3.10 Front Surface Gradient

again Sn0,p0 = 107 cm s−1 and thus limit the cell by interface recombination. Now, we want to introduce a front surface gradient (FSG) of the type d in Figure 2.14 and to derive a rule for dmin and Ev . Solar cells of the absorber/buffer/window type with different combinations of Ec and dmin are simulated and the JV curves analyzed. Figure 3.22 shows contour plots of simulated values of Voc , Jsc , FF, and η as a function of dmin and Ev . We find that the efficiency of a cell which is mainly limited by interface recombination can be largely improved by a FSG. With −Ev,front = 0.2 eV and dmin = 50 nm, the bottom line cell can be improved from 17.6 to 20.4% efficiency. The improvement is mostly due to an enhanced Voc . For larger dmin , however, the cell efficiency even declines below the bottom line value. Obviously this gives: Rule 8: In order to fully or partially deactivate interface recombination, the bandgap of the p-type absorber should be widened by lowering the valence band edge (Ev,front < 0). The region of Ev,front < 0 should be narrow (some tunneling lengths). The FSG increases Ep,az=0 or, which is the same, the hole barrier φb p at the buffer/absorber interface. We note that front gradients of the conduction band are not helpful in order to deactivate interface recombination (unless they completely change the band alignment). The reason is that a conduction band gradient does not alter the interface bandgap. The hole barrier will remain unchanged. For a p-type absorber, design concepts (e) and (g) of Figure 2.14 are not valuable. A cell which is limited by bulk recombination hardly benefits from a front surface bandgap gradient. To the contrary, it can suffer from loss in Jsc due to a positive slope of the quasi Fermi level dEFn /dz > 0. In the logic of this chapter, we realize the bulk limited cell by inducing strong Fermi level pinning at the buffer/absorber interface at a position of En,az=0 ≈ 0. In addition, we reduce the interface recombination velocity to Sn0,p0 = 103 cm s−1 . Thus, the cell with a bulk absorber lifetime of 10 ns has a Voc which is limited by bulk recombination and has solar cell parameters of Voc = 0.74 V, Jsc = 35.0 mA cm−2 , FF = 79.3%, and η = 20.6%. The energy band diagram of this cell is sketched in Figure 3.21b. Before we look at the impact of a FSG, we have to consider the impact of the effective bandgap. A bandgap gradient, modifies the effective bandgap of the absorber layer. According to Figure 3.1 a change of the bandgap is connected with a change of the maximum efficiency. If, for instance, the reference cell has a bandgap below the point of maximum efficiency in Figure 3.1, a bandgap widening will theoretically and (in the best case) experimentally increase the efficiency not only as a result of the bandgap gradient but also as a result of the η(Eg ) dependence. Thus, it is recommended to quantify the benefit of a graded absorber by comparing to a cell which has the same open circuit voltage but no bandgap gradient. Following Gloeckler [114], the normalization can be accomplished by X = X(E, dmin ) − X(const Eg , same Voc )

(3.4)

163

3 Design Rules for Heterostructure Solar Cells and Modules −2

∆Jsc

−1

0.2 0.1

∆FF

−4

−6

0

2 0

0.2

−2 ∆η

0.4

−3

.5 −2

0.3

−1

−1

0.0

−0.5

0.1

1

0.2

−8

0

0.0 0.3

−1.5

−0.5

−2

0.3

−∆Ev,front [eV ]

164

−1.5

0.1 0.0

0

5 6 7 89

2

3

4 5 6 7 89

0.1

1 dmin,front [µm]

Figure 3.23 Simulated normalized solar cell parameters Jsc (mA cm−2 ), FF (%), and η(%) for a linear front bandgap gradient in the absorber as a function of Ev,front and dmin (for definition of dmin see Figure 2.14).

Device parameters according to default Table 8.1. Strong inversion at the buffer/absorber interface realized by FLP at En,az=0 = 0 eV. Reduced interface recombination velocity of Sn0,p0 = 103 cm s−1 .

where X stands for a solar cell parameter Jsc , FF, and η. dX is the gain of a parameter relative to a cell with identical Voc . This normalization is necessary for cells limited by bulk recombination. Figure 3.23 gives normalized values of Jsc , FF, and η. For the bulk limited cell there is a small efficiency gain of 0.4% with the parameter combination EV,front = 0.15 eV and dmin ≈ 0.2 µm. This efficiency gain results from a combination of a loss in Jsc and enhancements of FF. The loss in Jsc has been explained in Section 2.3.7. It is due to the positive slope of the electron quasi Fermi level dEFn /dz > 0 in part of the absorber. The small gain in FF is due to a reduction of SCR recombination due to the larger bandgap in the SCR. For dmin > 0.4 µm, all FSGs result in an efficiency deterioration mostly due to a reduction in FF. As that part of the absorber which has dEFn /dz > 0 increases with voltage, current collection becomes voltage dependent – leading to a reduced FF. In summary, a FSG is hardly useful to optimize a cell which is limited by bulk recombination. This is even more true if we consider that a bandgap gradient can, under circumstances, induce electronic defects (see next section) – a fact that we neglected hitherto. Finally, we come back to our study logic. The small Ep,az=0 caused by the buffer layer introduced in Section 3.9 can be ‘cured’ by a FSG of −Ev,front = 0.2 eV and dmin,front = 10 nm. This reduces interface recombination and improves the solar cell parameters to Voc = 0.75 V, Jsc = 34.8 mA cm−2 , FF = 77.9%, and η = 20.36%. The energy band diagram of this cell is depicted in Figure 3.24.

3.11 Back Surface Gradients Figure 3.24 Sketch of an absorber/buffer/ window cell with potential drop in the buffer layer and with front surface gradient of dmin = 50 nm and −Ev,front = 0.2 eV.

Ec

Ev

−wa

dmin

0db

Position z

3.11 Back Surface Gradients

The motivation to implement a back surface gradient (BSG) is to reduce the photo current loss and to decrease the diode current. Back surface bandgap gradients have been studied, for example, in Refs. [120–123]. A BSG can optimize the cell in two respects: (i) It can minimize bulk recombination, and (ii) minimize back surface recombination. Whether bulk recombination or back surface recombination is the limiting process depends on the reduced absorber thickness da = (da − wa )/Ln,a . A sensitivity analysis shows that the limiting process is mainly bulk recombination for da > 1 and mainly back surface recombination for da < 1. In Section 2.3.7, we show that a BSG of the conduction band adds a current contribution, JCBM , and thereby increases the photo current. An alternative explanation is that the BSG increases the collection function (see Section 2.5.4.2). A BSG in the absorber can be realized by a chemical gradient (change in chemical composition) in a mixed crystal system. An example is the isopleth system CuInSe2 –CuGaSe2 where the bandgap widens with increasing [Ga]. If this chemical gradient leads to a gradient in the lattice constant, defects may be formed. The total defect density Nd may then depend on the gradient Ec,back /(da − dmin,back ). As a first approach we may assume a linear dependence, that is, Nd = Nd0 + Cc Ec,back /(da − dmin,back ) where Cc is a material parameter. Thus, the lifetime may become τ(dmin,back , Ec,back ) =

τ0 Nd0 Ec,back Nd0 + Cc da −dmin,back

(3.5)

In a lattice matched mixed crystal system, Cc is negligible. Thus for the beginning, we assume that the region of the absorber exhibiting a BSG has the same carrier lifetime as the ungraded region.

165

166


75

0.4

0.7

0.3

0.77

0.3

35.6

0.1

∆Ec,back [eV]

77

77.6

0.2 0.1

77.4

0.0

20.8

77.2 1 21.

0.4 2 0.3 1.2 0.2

−0

0.3

0

0.2

0.4

0.1 0.0 0.3 0

21.2

.6

0.6

0.2

21.1

0.1 20.5 5 6 78 9

2

3

4

5 6 7 89

0.1 (a)

.2

0.4

0.4

0.1 0.0

0

0.0 0.4

.4

77

0.3

0.5

0.1 35.6

0.0 0.4

1

0.2

0

0.2

0.4

1

.4

35

35.2

0.3

0.755 1

0.0 0.4

0

0.1

0.2

0.765 0.76

35.8

0.2

∆Ec,back [eV]

0.75

In the following, we discuss the impact of a BSG for two absorbers with thicknesses of da = 3.0 and 0.5 µm. Thus, we deviate from our default absorber thickness of 1.5 µm for a moment but come back to this thickness at the end of the section. The BSG is realized by a gradient of the conduction band in the range Ec = 0–0.5 eV. With the default electron lifetime of 10−8 s, the absorber has a diffusion length of 1.1 µm. Both cells are not limited by interface recombination as we introduced the FSG with dmin,front = 10 nm and −Ev,front = 0.2 eV, as shown in Figure 3.24. First, we consider the cell with the thick absorber (da = 3.0 µm). The SCR width wa equals 0.36 µm. With the parameters given above, this translates into a reduced absorber thickness of da = 2.4. The reference cell with a homogeneous bandgap of 1.2 eV, that is, without BSG, exhibits an efficiency of 20.5%. Thus, the reference cell is already highly efficient. The dominating limitation of the cell is carrier recombination in the absorber bulk. Therefore, in the first place we may improve the cell by reducing the bulk recombination. Figure 3.25 gives the results of simulated solar cell parameters as contour plots on an absolute scale (a) and after normalization (b) to a cell with identical Voc .

0.0

2

1 dmin,back [µm]

Figure 3.25 Simulated solar cell parameters Voc (V), Jsc (mA cm−2 ), FF (%), and η (%) (from top to bottom) for an absorber/buffer/window heterostructure with linear back surface bandgap gradient of varying Ec,back and dmin,back (for definition see Figure 2.14). Absorber thickness 3 µm.

0.2 5 6 7 89

2

0.4 3

4 5 6 7 89

0.1 (b)

2

1 dmin,back [µm]

Device parameters according to default Table 8.1 but with front surface gradient of dmin,front = 10 nm, Ev,front = 0.2 eV (see band diagram in Figure 3.24). (a) Absolute values. (b) Values normalized to a cell without bandgap gradient but with the identical Voc [see Eq. (3.4)].


With a BSG, the efficiency increases from 20.5 to 21.1%. This absolute efficiency increase is also found for the normalized parameters. It results from of a gain in all cell parameters. From the contour plot of the efficiency we see that optimally the bandgap gradient extends across the QNR of the absorber and induces a bandgap widening of 0.2 ± 0.1 eV. Thus, the absorber has a dmin,back which corresponds to wa . Why are BSGs with a larger dmin,back not as efficient? The answer is that the limiting recombination takes place in the bulk of the absorber. The larger dmin,back , the less photo generated carriers can benefit from the additional force provided by the bandgap gradient. Using slightly different cell parameters, Gloeckler derived the same rule for a BSG in a thick absorber [114]: Rule 9: For a cell limited by bulk recombination, a back surface bandgap gradient should extend through the QNR of the absorber. Certainly, this rule is based on the assumption that the gradient does not induce further defects and thereby reduce the carrier lifetime [124]. If the material parameter Cc is not negligible, the benefit of a back surface bandgap gradient becomes smaller and will eventually disappear. In Figure 3.26, we simulate this case using Cc = 1010 cm−2 eV−1 . We see that in principle Rule 9 remains valid. The benefit of the BSG, however, is reduced by the reduced carrier lifetime in the grading region. Next, we ask for a bandgap gradient in a thin absorber layer but therefore return to a perfectly lattice matched system (Cc = 0). An example has been calculated in Figure 3.27. The absorber thickness was reduced to 0.5 µm while all electronic parameters are identical as for the device calculated in Figure 3.25. Thus, the reduced thickness of the absorber is da = 0.12. With a back surface recombination velocity of 107 cm s−1 , this cell is limited by back surface recombination. The efficiency of the bottom line cell with a homogeneous bandgap of 1.2 eV is only 17.2%, that is,

4e - 09

6e - 09

0.33

0

0

8e-09

.2

−0

0.3 0.2

0

(b)

0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 5

09 2e-

∆Ec,back [eV]

(a)

6 7 89

2

3

4 5 6 7 89

0.1

2

1 dmin,back [µm]

Figure 3.26 Device as in Figure 3.25 but with gradient dependent carrier lifetime. (a) Carrier lifetime in the grading region for a gradient with Ec,back and dmin,back . Without bandgap gradient, the absorber lifetime is 10−8 s (default value). With increasing

Ec,back and dmin,back the gradient becomes larger and the carrier lifetime is reduced according to Eq. (3.5). (b) Efficiency normalized to a cell without bandgap gradient but with the identical Voc (see Section 3.10).

167

0.7 9

0.4

0.78


168

Voc [V]

0.3 0.2

0.77 0.76 0.74

0.1

0.75

0.0

Jsc [mA cm−2] 32.5

0.1

33

FF [%]

0.3 0.2 0.1

77.5 76.5

0.0 0.3

0.3 0.2 0.1

∆FF [%] 1

1.5

1

0.0

η [%]

0.15

0.2

19.5 18.5

19

0.20

0.25

0.30

dmin,back [µm]

∆η [%]

1.5

0.3 20

0.1

(a)

0.4

0.4 19

0.2 0.0 0.10

∆Jsc 2.5 [mA cm−2]

1

0.5

0.0 ∆Ec,back [eV]

0.4

2

0.2

0.0 77

∆Ec,back [eV]

0.1

1.5

0.3

32

0.2

0.4

1

0.4

0.5

0.3

5 31.

33.5

0.4

2

0.1 0.35

0.0 0.10

0.40

(b)

Figure 3.27 Similar device as in Figure 3.25 but with absorber thickness of 0.5 µm. Linear back surface bandgap gradient in the absorber with dmin,back = 0–0.5 eV and Ec,back = 0.1–1.4 µm. Back surface recombination velocity Sn0,bc = 107 cm s−1 .

1

0.5

0.15

0.20

0.25

0.30

0.35

0.40

dmin,back [µm]

Other device parameters according to default Table 8.1. (a) Graphs showing Voc (V), Jsc (mA cm−2 ), FF (%), and η(%) (top to bottom). (b) Graphs showing Jsc , FF, and η normalized to a cell without bandgap gradient but with the identical Voc (see text).

3.2% lower than the thick bottom line cell in Figure 3.25. In Figure 3.27, the BSG was varied in the ranges Ec,back = 0–0.5 eV and dmin,back = 0.1–0.4 µm. It can be seen that with a bandgap widening of 0.25 ± 0.1 eV and dmin,back = 0.4 eV, the absolute cell efficiency rises to simulated 20.2%. The efficiency gain is caused by a gain in Jsc and Voc . Thus, the optimum gradient, as in the case of the thick absorber, uses a bandgap widening of about 0.3 eV. But for the low thickness absorber which is limited by back surface recombination, the gradient does not extend through the absorber but is restricted to a range near the BC. The same conclusion can be drawn from plots of the relative cell parameters on the right hand side of Figure 3.27. In order to show the systematic for the optimum location of dmin,back in absorbers of different thicknesses, the value of dmin,back has been varied for different absorber thicknesses but constant Ec,back = 0.3 eV. Figure 3.28 gives the efficiency for absorbers with thicknesses between 0.9 and 2.0 µm (reduced thicknesses of 0.5, 0.8, 1.0, 1.2, 1.5 µm). The plot of the cell efficiencies points out that the best value of dmin,back (giving the highest efficiency) moves from the back side of the absorber to the front side of the absorber when the absorber thickness is increased.


Efficiency η [%]

wa 21.2 20.8 20.4 20.0 0.0

d'a = 0.5, 0.8, 1.0, 1.2, 1.5 0.5

1.0 dmin,back [µm]

1.5

Figure 3.28 Simulated solar cell efficiency for a linear back surface bandgap gradient of Ec,back = 0.3 eV in the absorber as a function of dmin,back (for definition of dmin,back , see Figure 2.14). Different reduced absorber

2.0

thicknesses. Device parameters according to default Table 8.1 with front surface gradient (dmin,front = 10 nm, Ev,front = 0.2 eV). For a typical band diagram see Figure 3.29.

Ec

EF Ev

−da

−dmin,back

0

Position z Figure 3.29 Energy band diagram of an absorber/buffer/window cell optimized with a front surface gradient to reduce interface recombination and with a back surface gradient to reduce volume and back contact

recombination. dmin,back = 0.5 µm was selected according to Figure 3.28 where it is da = 1.17. This cell represents the optimized device of this study. Other parameters as in Table 8.1.

Thus for cells being limited mainly by back surface recombination (da < 1) we can formulate: Rule 10: For a cell limited by back surface recombination, the back surface bandgap gradient should be restricted to the back surface near region of the absorber. Rule 10 remains valid if the gradient leads to a reduction in carrier lifetime in the grading region. The phenomenological reason for the inferior function of a BSG extending through the complete thin absorber with da < 1 is the lower short circuit current due to the higher bandgap in average. This higher bandgap is not

169


fully compensated by an increased Voc . If the BSG is restricted to the back surface near region, the effective field is very high near the BC, which acts as an electron mirror. The default absorber thickness of this study is da = 1.5 µm, which gives a reduced thickness of 1.17 µm (Ln,a = 1.1 µm). According to Figure 3.28, the optimum dmin,back is about 0.5 µm. Thus, the optimized cell has a double grading structure (as illustrated in Figure 3.29) which gives an efficiency of 20.4% with parameters Voc = 0.787 V, Jsc = 33.76 mA cm−2 , and FF = 76.8%. It may be instructive to look at the different contributions for this efficiency in more detail. In Figure 3.30 we see that without a bandgap gradient, the device performance is limited to about 17%. (For the band diagram of this cell, see Figure 3.21a.) Introduction of

Voc [ V ]

0.80 0.75 0.70

FF [%]

Jsc [mA cm−2]

0.65 34.8 34.4 34.0 33.6 77 76

21 η [%]

20 19 18

Figure 3.30 Simulated solar cell parameters for an absorber/buffer/window heterostructure with a thin absorber of da = 0.5 µm. Material parameters as in default Table 8.1. Case discrimination without bandgap gradi-

FSG / BSG

FSG

BSG

17 No Gradient

170

ents, with back surface bandgap gradient of dmin,back = 0.1 µm, Ec,back = 0.2 eV (BSG), with a front surface bandgap gradient of dmin,front = 0.01 µm, Ev,front = 0.2 eV (FSG) and with a combination of BSG and FSG.

3.12 Monolithic Series Interconnection

a back surface bandgap gradient (BSG) alone decreases the current but increases the FF. Introduction of a FSG alone improves the open circuit voltage due to the suppression of interface recombination. We also see a slight increase in FF. With the FSG, the cell is ‘cured’ from interface recombination. Adding to this cell the BSG leads to the highest efficiency and to the band diagram as in Figure 3.29. With the given absorber carrier lifetime this cell can only be improved by appropriate light trapping structures [65]. If indeed the absorber carrier lifetime could be increased we could reduce the bulk recombination rate. It is interesting that thereby, we could bring the cell back to the situation where its open circuit voltage is limited by interface recombination. As stated in Section 2.4.6, it is the combination of J00 , A, and Ea values which determines the dominant recombination process limiting the open circuit voltage. Similar to a back surface bandgap gradient is the function of a back surface field [125]. It can be realized by an absorber layer of increased doping concentration near the BC. However, it can also be realized by an accumulation layer at the absorber/BC interface. The result is an electric field which reduces the photo current loss and decreases the diode current. Both a p+ layer and an accumulation layer lead to an additional potential drop at forward bias. In order to calculate the benefit of a back surface field, we refer to Ref. [126]. We are interested in the relative benefit of a back surface field for different absorber thicknesses. The electric field is realized by an accumulation layer at the BC which is due to contact formation and is characterized by the barrier height φb − φb 0 where φb 0 is the barrier height under flatband condition. For φb − φb 0 < 0, an electron barrier has been formed at the BC. At the same time we can inspect the opposite case: a depletion layer with electron barrier φb − φb 0 > 0 (see Section 2.4.7). In Figure 3.31 we see that the benefit of a back surface field (φb − φb 0 < 0) is modest for reduced absorber thicknesses of da > 1. This complies with the finding of Figure 3.11 which shows that recombination at the BC hardly reduces the cell parameters for τn0,p0 = 10 ns and large absorber thickness. For da < 1, the positive effect of a back surface field becomes obvious. However, a hole barrier reduces the efficiency to 14.8%. If such a hole barrier cannot be avoided due to the band alignment at the absorber/BC interface, a back surface bandgap gradient is required in order to partly compensate for the efficiency loss.


Monolithic series interconnection is one of the cost relevant advantages of thin film technologies over wafer-based technologies. Frequently used interconnection schemes are shown in Figure 2.43. For all interconnection schemes, the cell length xcell , as defined in Figure 2.41c, has to be optimized. This optimization has a number of variables as given in Table 3.2. We see that there is always a trade-off between FF losses and Jsc losses.

171


φb –φb0 [eV ]

172

0.2 0.1 0.0 −0.1 −0.2 0.2 0.1 0.0 −0.1 −0.2 0.2 0.1 0.0 −0.1 −0.2 0.2 0.1 0.0 −0.1 −0.2

2 0.7 0.73

Voc [V] 0.74

0.745

0.75

0.76

32

Jsc [mA cm−2]

33 34

35

35.5

74

75

FF [%] 76 77

17

η [%]

18 19

2

20 20.8

3

20.5

4 5 6 7 89

2

1 da′ [µm] Figure 3.31 Contour plot of the solar cell parameters (from top to bottom) Voc , Jsc , FF, and η as a function of the absorber thickness and the back contact barrier φb . Device parameters as in default Table 8.1. Front surface gradient as discussed in

Section 3.10 and as shown in Figure 3.24 has served to deactivate interface recombination. A hole (electron) barrier is given for φb > 0 (φb < 0). Further parameters: RBC = 0 and Sn0,p0,bc = 107 cm s−1 .

Table 3.2 Parameters for the optimization of the monolithic series interconnection given as interrelated couples which define a trade-off between FF and Jsc losses. The interrelated parameters are in italics.

Couple number

FF loss

Jsc loss

1

Window layer sheet resistance, R , w Shunt conductance across P1, GP1 Contact resistance between window and back contact layer, Rw,BC Cell width, xcell

Window layer absorbance, 1 − Rw − Tw Width xP1 of the interconnection region Width xP2 of the interconnection region

2

3

Number of cells per module, Ncell = xmod /xcell

η [%]

0.96

14

(b) 2 10 Rw [Ω/sqr.]

4

6 8 100

(η-ηMod) /η [%]

6 8

16

24 8 22

7

20

6

18

5 20

(c) Figure 3.32 (a) Optical transmission of ZnO films (after [128]) (markers) together with a fit to the data according to the algebraic function T = T0 − (Rw,0 /Rw )m where T0 = 1, R0 = 0.28, and m = 0.753 has been used. (b) Cell efficiency η as a function of short circuit current density of simulated absorber/window heterostructures (simulation parameters see Table 8.1). Module efficiency

Xcell [mm]

0.88 4

(a)

20 18

0.92

ηMod,

Transmission, Tw


25 30 Jsc [mA cm−2]

35

ηMod simulated for an optimized cell width xcell using the program Design of M. Burgelman and A. Niemegeers [129]. Optimization parameters are Rw = 8 /sqr., Tw = 0.92, Rw,BC = 10−3 cm2 , and interconnect loss 400 µm. (c) Relative efficiency loss due to series interconnection and optimum cell width xcell .

The first trade-off concerns the window layer, which is characterized by an optical transparency Tw (λ) at a given sheet resistance, R,w : due to free carrier absorption and due to high impurity concentrations, the transparency reduces with lower sheet resistance. This effect is wavelength-dependent. Thus, in order to derive an empirical function Tw (R,w ), the optical transmission should be weighted by the normalized internal quantum efficiency of a cell with particular absorber bandgap [127]. Figure 3.32a gives an example of the transparency of experimental ZnO thin films for an absorber bandgap of 1.15 eV as a function of R,w . Increasing the window layer film thickness reduces R,w and thus reduces the loss in FF, but also increases 1 − Tw and thus the loss in Jsc . Trade-off 1 in Table 3.1 is interrelated with trade-off 3 since the cell length xcell plays a role. We note in passing that increasing the window layer thickness also increases the production costs. For the second trade-off, optimization of xcell depends on the contact resistance of window and BC Rw,BC as well as the shunt conductance GP1 (see Figure 2.41d). GP1 depends on the sheet conductance of the absorber and the width xP1 . Rw,BC depends on the specific contact resistance and the width xP2 . Thus, the free parameters are xP1 and xP2 . GP1 and Rw,BC decrease with increasing width of P1 and P2. The price to be paid is a higher area loss and therefore a lower Jsc . The third trade-off for the cell width concerns the relative area loss due to the interconnection and the distributed series resistance: if the cell length is too small,

173

174


the area loss is too large thus diminishing Jsc . If the cell length is too large, the distributed series resistance diminish the FF. An optimum of xcell occurs which depends on the Jsc of the elementary diode and thus on the absorber bandgap. If material cost aspects are taken into account, a cell width somewhat smaller than the optimum width may be selected. In order to compare different absorber bandgaps, we assume that sheet conductance and optical transparency of the window layer are fixed. This is a simplification that slightly discriminates large bandgap absorbers since they hardly suffer from free carrier absorption in the window layer. Nevertheless, some valid conclusions can be drawn from the simulation. We use the software package Design of M. Burgelman and A. Niemegeers [129] in order to obtain plots of module efficiency ηmod and xcell versus the absorber bandgap, Eg,a . Figure 3.32b reveals that the module efficiency is around 20% below the cell efficiency. This is due to area loss, resistance loss, and transmission loss. The efficiency loss due to monolithic series interconnection is a structural problem. Besides optimizing the parameters in Table 3.1, it is proposed to add a contact metal grid on top of the window layer of each interconnected cell [130]. This allows a reduction in the window thickness, thereby reducing optical losses at constant resistance losses. A metal grid, however, requires an additional (and not simple) process step. Figure 3.32c reveals that the optimum cell length increases with decreasing Jsc and thus with increasing absorber bandgap. Also, the relative loss in module efficiency is smaller for lower Jsc . In reality, this effect is even larger since transparency losses due to free carrier absorption are overestimated for the larger absorber band gap. Due to the large sum of input parameters for the monolithic interconnection, it is not possible to give a design rule. A rough guide may be the cell width as a function of bandgap, which has been plotted in Figure 3.32c. Table 3.2 already suggested that the efficiency loss by monolithic series interconnection is due to reductions in FF and Jsc . Indeed, the Voc,mod of the module is simply given by Ncell × Voc,cell . Depending on the application, a particular Voc,mod may be required for a given module size. This would introduce a constraint on Ncell and would require to re-assess window layer thickness, interconnection width and cell length.

175

4 Thin Film Material Properties Within this chapter we give an account of absorber, buffer, and window properties in as far as they are relevant for the function of chalcogenide cells. Since several properties of the respected materials are still not completely known, we emphasize that this account is a snapshot of the current understanding rather than a completed data base. For preparation aspects we refer to Chapter 5. Nevertheless, here we first want to rationalize why several chalcogenide films can be obtained through relatively easy film synthesis methods – a fact which makes chalcogenide films suitable for thin film photovoltaics and which is one of the reasons for their high solar cell efficiencies.

4.1 AII –BVI Absorbers

Compound semiconductors of the type AII –BVI form a large group of materials with a variety of bandgaps (Figure 4.1). Within this group, only CdTe has a bandgap appropriate for a single junction solar cell (1.0 eV < Eg < 1.5 eV). This is why the following section focuses on CdTe material properties and only touches other binary phases and solid solutions as far as they are relevant for CdTe-based solar cells. 4.1.1 Physico-Chemical Properties

CdTe crystallizes in the zincblende (ZB) structure with space group F43m (Figure 4.2a). It is the only compound in the Cd1−x Tex phase diagram (Figure 4.3a). If grown on hexagonal substrates, thin layers of CdTe may also be stabilized in the wurtzite structure [131]. CdTe exhibits a high formation enthalpy (heat of fusion) of about −100 kJ mol−1 (see Table 4.1), which makes it thermodynamically very stable. The CdTe melting temperature of 1365 K is much higher than that of Cd (594 K) and Te (723 K). CdTe sublimates and evaporates congruently via the equilibrium reaction CdTe ⇔ Cd + 1/2Te2 . The high thermodynamic stability together with its congruent evaporation allows CdTe to be formed by various preparation methods. Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

4 Thin Film Material Properties 2.8 CdS Band gap [eV]

176

ZnTe

2.4

2.0 CdSe

CdTe

1.6

1.2 5.6

5.8

6.0

6.2

6.4

6.6

Lattice parameter [Å]

Figure 4.1 Bandgap vs lattice constant of selected II–VI semiconductors of hexagonal (hexagons) and cubic (squares) crystal structure.

AIBIIICVI2

AIIBVI

(a)

AII BVI

(b)

AI

BIIIAI + 2VAI

AI2BIICIVDVI4

(c)

AI

(d)

AI

BIII

BII

BIII

CVI

CIV

VAI

DVI

BIIIAI CVI

Figure 4.2 Crystal structures of (a) zincblende (ZB), (b) chalcopyrite (CP), (c) kesterite, (d) chalcopyrite structure with defect complex of BIII AI –2VAI [134].

McCandless has discussed the thermochemical aspects of CdTe processing [132, 133]. Table 4.1 gives the chemical diffusion coefficients at 500 ◦ C for some impurity elements which are relevant for the solar cell process. These values are measured on monocrystalline CdTe where the effect of grain boundary diffusion is negligible. Among the impurities listed, Cu and Na have the highest reported diffusivity


Liquidus Solidus

1600 1365 K

1200 z

1000 800

z+w

w

778 K

600

(a)

1.6 1.5

1.4 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 x x x (b) (c)

Figure 4.3 Atmospheric pressure pseudobinary phase diagrams of (a) Cd1−x Tex (after Jianrong et al. [135]), (b) CdTe1−x Cl2x (after Tai et al. [136]), and (c) CdTe1−x Sx (after Ohata et al. [137]) (z – zinc blende, w – wurtzite). The calculated energy bandgap of CdTe1−x Sx is given as a dashed line on the right axis [138].

outperforming the diffusion of Cl and In. The Cu diffusivity appears to be further promoted by grain boundaries as the diffusivity is higher in thin films than in crystals [157]. S has the lowest diffusion constant. It was found, however, that under a CdCl2 : O2 : Ar vapor pressure, the diffusion coefficient of S is increased by 3 orders of magnitude [158]. The large chemical self-diffusion coefficient DCdTe guarantees the growth of homogeneous CdTe films. The pseudobinary phase diagram of CdTe1−x Sx (Figure 4.3c) is so important as chemical intermixing at the device CdTe/CdS interface has been observed. The stable crystallographic structure of Te-rich CdTe1−x Sx is the ZB structure and that of S-rich CdTe1−x Sx is the wurtzite structure. The phase diagram in Figure 4.3c shows a large miscibility gap between CdTe and CdS. This miscibility gap becomes smaller at high temperatures. At 673 K it ranges from x = 0.06 to 0.97 [138]. However in a metastable state, thin solid solution films with all values of x in the ZB structure can be formed [159]. Within each structure type, the lattice parameter of CdTe1−x Sx follows Vegard’s rule [138]. The pseudobinary phase diagram of CdTe1−x Sex (not shown) has been studied experimentally in the temperature range 1070–1570 K [160]. It exhibits a ZB phase field on the CdTe side and a wurtzite phase field of the CdSe side in accordance with the crystal structures of the pure binaries. At 1070 K, the border between ZB and wurtzite is at around x = 0.5. Between the two phase fields there is a narrow (3%) two-phase region at equilibrium. Other publications report on a more extended two-phase region where the minimum bandgap occurs for the ZB structure [161]. Theoretical calculations point out a positive mixing enthalpy at T = 0 [162] and a miscibility gap below 450 K which widens toward lower temperatures [163].

Eg [eV]

T [K]

1400

177

178

4 Thin Film Material Properties Table 4.1

CdTe physico-chemical and opto-electronic properties.

Property

Symbol

Value or range

References

Melting point Heat of fusion (300 K) Entropy (300 K) Space group Lattice constant Thermal expansion coefficient (300 K) Density Optical band gap (300 K) Temperature dependence of Eg,a Electron effective mass Hole effective mass

Tm H0f S0 – a α ρ Eg dEg /dT mn ∗ /m0 mlp ∗ /m0 mhp ∗ /m0 Nc Nv ni µn µp ε/ε0 ε∞ ε0 n Lα0 U0 τrad D at 773 K (cm2 s−1 )

1365 K −100 kJ mol−1 95(1) J K−1 mol−1 F43m 6.48 A˚

[139] [140] [140] – [141] [142] [141] [143–145] [145] [146] [147] – Calculated Calculated Calculated [147] [147] [147] [147] [147, 148] Calculated Eq. (2.60) Calculated [149] [150] [151] [152] [153] [154] [155] [156]

Effective density of states Intrinsic carrier concentration (300 K) Electron mobility Hole mobility Static dielectric constant Optical dielectric constant Refractive index (600 nm) Absorption length Radiative recombination rate Radiative lifetime Bulk diffusion coefficients

5.3 × 10−6 K−1 5.86 g cm−3 1.49(1) eV −4 × 10−4 eV/K 0.098 0.145 0.82 8 × 1017 cm−3 2 × 1019 cm−3 1.1 × 106 cm−3 1050 cm2 V−1 s−1 104 cm2 V−1 s−1 10 7.1–7.3 3.1 1.2 × 10−4 cm 28 cm−3 s−1 1 × 10−6 s CdTe: 3 × 10−7 Zn: 8 × 10−12 Cu: 5 × 10−9 S: 4 × 10−15 Cl: 8 × 10−11 In: 2 × 10−11 Na: ∼10−9 Au: 6 × 10−12

Alloying of CdTe and ZnTe can lead to an absorber layer with larger band gap which may be used in a top cell of a tandem device. CdTe and ZnTe both crystallize in the ZB structure. Thus, no separated phase fields are expected but it remains the question of a miscibility gap. Such a gap has been pointed out by Haluoi et al. [164] to extend up to around 570 K. Random ordered thin films, in contrast, have been prepared in the complete solid solution system [148, 165, 166]. The minimum temperature of miscibility thus may depend on the preparation method. Due to the important role of the chlorine treatment for CdTe solar cell, we are interested in the CdTe1−x Cl2x phase system. A typical chlorine treatment temperature is 670 K. The equilibrium phase diagram after Ref. [136] is shown in


Figure 4.3b. A eutectic point at x = 0.74 is depicted with a melting temperature of 778 K. In the presence of oxygen, the melting temperature may be even lower [167]. Presently, it is not clear if the chlorine treatment leads to a liquid phase recrystallization of the CdTe film at 670 K. 4.1.2 Lattice Dynamics

Raman spectroscopy is a powerful tool for material characterization. The vibrational spectrum of ZB crystals with point group T2d consists of two three-fold degenerate zone-center vibrational modes of symmetry F2 [168]. Of these, three are acoustic modes. The optic mode is split into transversal optic (TO) and longitudinal optic (LO) contributions. Thus for randomly oriented films, the Raman spectrum of CdTe exhibits two lines at around 143 cm−1 (TO) and 167 cm−1 (LO). In Raman backscattering geometry, the dipole selection rules are as follows: from 100 LO scattering is allowed but TO is forbidden, from 110 LO is forbidden but TO is allowed, and from 111 both LO and TO are allowed. Nevertheless, the forbidden phonon LO mode from 110 can be observed for resonant Raman scattering [169]. The phonon dispersion curve for CdTe has been calculated, for example, in Refs. [170, 171]. The LO mode exhibits of slope with dE/dk < 0 around the point. Hence, we expect that in the case of high defect density and a resulting phonon confinement, the A1 mode becomes asymmetric toward smaller wavenumbers – in accordance with the experiment [172]. An indication for crystal quality of CdTe material is the number of higher-order LO lines which can well be on the order of 5 [166]. A collection of Raman lines from secondary phases is given in Table 4.2. For single crystals of CdTe and its alloys, Raman spectroscopy often reveals Te precipitates by the occurrence of the Te A1 mode at around 122 cm−1 [169, 173]. The solid solution systems Cd1−x Znx Te [166, 174, 175], Cd1−x Mnx Te [174], CdTe1−x Sex [176], and CdTe1−x Sx [177] all exhibit a two-mode behavior. Table 4.2 Vibrational modes of CdTe and main vibrational modes of related compounds detected by Raman scattering.

Phase/references

Vibrational modes (cm−1 )

CdTe Te [173, 178, 179] CdTeO3 [180] TeO2 [181] CdO [182]

143 (TO), 167 (LO) 92, 127, 143 650a, 760a 148, 393, 648 270, 400, 480, 960

a

Broad band.

179

180

4 Thin Film Material Properties

4.1.3 Electronic Properties

The electronic band structure of ZB CdTe has been calculated by Chelikowsky and Cohen [183]. The valence band maximum at the point of CdTe is four-fold degenerate and splits into two bands at k = 0. The heavy hole band with an effective mass m∗ hp and the light hole band with an effective mass m∗ lp . At the point the two-fold degenerate spin-orbit split-off band is 0.96 eV below the valence band maximum. Thus, only the heavy and light hole bands need to be considered here. Table 4.1 gives the effective masses of electrons and holes for CdTe. In the limit of parabolic bands we make use of 2πm∗n,p kT (4.1) Nc,v = 2 h2 in order to calculate the effective density of states of conduction and valence band (Table 4.1). With these values, we calculated the intrinsic carrier concentration. For crystalline CdTe an electron mobility of 1050 cm2 V−1 s−1 and a hole mobility of 104 cm2 V−1 s−1 have been reported [147]. For polycrystalline thin films of device quality, hole mobilities of ∼5 for Cu-doped films and ∼50 cm2 V−1 s−1 for undoped films have been measured. These values are lower than the peak values listed in Ref. [147] – an observation that has been ascribed to carrier scattering at charged grain boundaries [184]. Thus, using 40 cm2 V−1 s−1 for the hole mobility and 320 cm2 V−1 s−1 for the electron mobility in device simulation appears reasonable [185]. In our general simulations of a chalcogenide device in Chapter 1 we will use even more conservative values (see default Table 8.1 ). 4.1.3.1 Practical Doping Limits CdTe is recognized as an amphoteric semiconductor which can be doped n-type and p-type. Electron concentrations and hole concentrations of up to 1018 cm−3 have been achieved by arsenic doping [186] and chlorine doping [187], respectively. In polycrystalline CdTe absorber layers of up to date solar cells, however, hole concentrations are only in the range of 1015 cm−3 [188]. Figure 4.4 gives calculated defect transition energies in the bandgap of CdTe [189]. The error of calculated transition energies is about 0.1 eV. We differentiate between intrinsic defects and impurity levels. Intrinsic defects in the form of interstitials, vacancies, and antisites can act as donors or acceptors provided that their transition energies are shallow. Figure 4.4 reveals that most of the intrinsic defects have large transition energies and therefore may lie between the demarcation levels for recombination (Section 2.3.6). The single ionized Cd vacancy is the shallowest acceptor energetically situated at Ev + 0.13 eV. We conclude that based on the calculated energies in Figure 4.4 intrinsic p-type doping requires a Cd-poor preparation. The hole concentration will be limited by the large defect transition energy of VCd (0/−) . In principle, impurities in CdTe can provide more shallow defect levels. The character of the substitutional impurities in Figure 4.4 is in accord with this rule of


Ec

1.5

CdTe

Cui

AlCd lnCd

Nai

(2+ /0)

(+/0)

GaCd TeCd

ClTe

(2+ /0)

TeCd

(2+ /+)

VTe

(2+ /0)

Tei

(0 /2−)

VCd VCd

(−/2−) VCd + Cui

1.0 Energy [eV ]

(+ /0)

Tei

(+/0)

0.5

0.0 Figure 4.4

Ev

(0/−)

VCd + ClTe

CuCd AuCd AgCd NaCd

(0/−)

Sb Te As Te P Te N Te

(0/−)

Theoretical values of electronic defect levels in CdTe. After Ref. [189].

thumb. Substitutional impurities with an electron excess are donors and with an electron deficit are acceptors. Shallow acceptors are NaCd as well as NTe and PTe . (+/0) Since, however, the defect transition energy of Nai is very similar to the one (0/−) of NaCd , acceptor-type NaCd defects may be compensated by donor-type Nai defects [189]. Chlorine treatment (Section 5.1.5) and back contact formation (Section 5.1.8) are important process steps for high efficiency devices and are known to change the hole concentration in the absorber. They can lead to a Te and the in-diffusion of O, Cl, and Cu atoms. Cu and Cl can form defect complexes with VCd provided that the material is Te-rich. Indeed, the VCd –ClTe defect complex has a relatively small activation energy (see Figure 4.4). Cu atoms can behave as amphoteric impurities, that is, Cu is a donor in the interstitial site (Cui ) or an acceptor when substituting Cd (CuCd ). The defect complex VCd –Cui is an acceptor, the complex CuCd –Cui mentioned in Ref. [190] would be neutral and could account for the low doping efficiency of Cu atoms. Nevertheless, Cu is believed to be a p-type dopant in CdTe [184, 191, 192]. While oxygen is not expected to form a shallow dopant, there are several reports on the doping effect of O2 used as ambient during CdTe deposition [11, 193]. The hole concentration with oxygen doping can reach 1015 cm−3 while without oxygen it is 1 order of magnitude lower. Since the oxygen concentration in the films is 1019 –1020 cm−3 [194] we may not speak of a typical doping effect. The role of

181

182


oxygen in CdTe is discussed in Section 5.1.6. It was reported that the p-type doping also increases with substrate temperature by creation of native defects [195]. 4.1.3.2 Defect Spectroscopy Defect levels in CdTe solar cells have been investigated by use of admittance spectroscopy (AS) and deep-level transient spectroscopy (DLTS). We concentrate on these methods as they are appropriate to determine absolute defect densities. Table 4.3 gives selected defect energies found in device quality CdTe absorbers in the solar cell. Note that more defects have been found in samples with non-optimum CdCl2 treatment [196], without nitric–phosphoric acid (NP) etching step [197], or in cells with a Cu treated back contact [198]. For cells which may exhibit a large back contact barrier (e.g., not etched cells), impedance data should be interpreted based on an equivalent circuit which includes the back contact depletion layer (see Ref. [199] and Section 2.6.4). A further influence factor on the measured defect density is the chemical environment of the CdCl2 treatment. In Ref. [200], different defects have been discriminated for cells being CdCl2 treated in air or in vacuum. The first observation is that their number is large although Table 4.3 presents only a selection of defects. Only one tentative assignment to a microscopic defect has been given. The second observation is that the defect densities appear to be rather small. Values exceeding 1014 cm−3 have been reported for cells which had no CdCl2 treatment [196]. Otherwise, the defect densities are in the range of 1013 cm−3 . Using AS, no defect density was found in Ref. [196] for a cell with good CdCl2 treatment. In order to estimate the impact of the found defects on the device performance, the capture cross sections for electrons and holes must be known. So far, however, these have not been measured with ample precision; instead the capture cross sections are rather fitting parameters. Burgelman et al. conclude that defect densities by DLTS are generally measured too small and introduced deep acceptors of 1015 cm−3 in order to simulate the measured JV curves [200]. A clear relation between measured defect densities and device performance has yet not been found. Thus, in order to simulate a state of the art CdTe solar cells one should rather use measured carrier lifetimes as input parameters for the simulation. 4.1.3.3 Minority Carrier Lifetime The lifetime of minority carriers in the bulk of the absorber can determine the open circuit voltage of the cell (as long as the device is not limited by interface recombination). In Section 2.4.6 it is shown that a lifetime of τ ≈ 10−9 s roughly marks the border between recombination in the SCR and recombination in the QNR as the limiting processes for the diode current. Therefore, knowledge about the lifetime is very important for a comprehensive understanding of a device. Minority carrier lifetimes have been investigated using photoluminescence (PL) decay measurements or photoconductivity transients. Specimens were either bare chalcogenide absorbers or finished solar cells. Often, the decay curve does not

Defect energies measured using admittance spectroscopy (AS) and deep-level transient spectroscopy (DLTS) in device quality CdTe solar cells. Hole or electron trap assignment as derived from DLTS measurement.

Table 4.3

AS

–

–

–

0.51 [197]

0.43 [197]

–

DLTS

0.76 [201]

0.74 [200]

0.72 [200]

0.5 [200]

0.44 [200]

0.42 [200]

Trap Treatment

h CdCl2 in O2

h CdCl2 in vacuum [200]

h CdCl2 in air [200]

h CdCl2 in vacuum [200]

e CdCl2 in air [200]

e CdCl2 in air [200]

Assignment

–

Density (cm−3 )

–

–

4 × 1012 [200]

–

5 × 1013 [200]

–

3 × 1011 [200]

–

6 × 1012 [200]

–

4 × 1012 [200]

– 0.27–0.35 [201, 202] e,h HgTe : Cu [202]

–

7 × 1012 [201]

0.18 [197] 0.185 [200] h CdCl2 in air [200] no NP etch [197] –

2 × 1012 [200]

0.13 [196, 198] 0.13 [200]

0.11 [200] h CdCl2 in air [200]

–

3 × 1012 [200] 4.1 AII –BVI Absorbers

h CdCl2 in air [200] Poor CdCl2 [196] V− Cd [198] + V− Cd − ClTe [198] 1 × 1011 [200] 2 × 1013 [196]

–

183

184


follow a simple mono-exponential time dependence [35, 36, 203, 204]. Instead, it needs to be described by at least two exponentials where the first exponential can fit the initial decay and the second the late decay. Under the best conditions, the PL decay would reflect bulk recombination and grain boundary recombination. Both quantities are aggregated in an effective lifetime, τeff , which limits the device performance. Unfortunately, other effects can reduce the measured lifetime such as surface recombination and charge carrier separation at internal electric fields. If surface recombination is large on the bare chalcogenide absorber but is much smaller in the solar cell, the lifetime of a bare absorber would be measured erroneously small. The same is the case for charge carrier separation in a depletion layer. Charge separation quenches the radiative recombination and the measured lifetime comes out too small [205]. Therefore, an assignment of the PL decay to the bulk carrier lifetime is not straightforward. Empirically, however, it was found that lifetime values calculated from the PL decay time constants are correlated with the device Voc both for chalcopyrite (CP) [206] as well as for CdTe based devices [36]. This may justify to label measured decay times as the limiting bulk minority carrier lifetimes. Table 4.4 gives measured PL decay times for CdTe absorbers. The initial PL decay of CdTe samples or solar cells was described by a mono-exponential decay function [36]. The derived lifetime values are at best 1 ns. They show a dependence on CdCl2 treatment, sulfur diffusion [207], oxygen concentration [36], and back contact deposition [208]. According to Figure 2.27, we expect up to date CdTe solar cells to be limited by SCR recombination.

Table 4.4 Photoluminescence decay times for various chalcogenide absorbers and solar cells. All decay times have been obtained from fits to mono- or bi-exponential functions. For the excitation conditions, see references.

Absorber Cu(In1−x Gax ) Se2 , x = 0.3 − 0.4 Cu(In1−x Gax ) Se2 , x = 0.3 − 0.4 CuInSe2 Cu(In1−x Gax ) Se2 , x = 0.25 − 0.35 Cu(In1−x Gax ) Se2 , x = 0.25 − 0.35 CdTe CdTe

Emitter – CdS/ZnO – – – SnO2 CdS/SnO2

Remark

τ1 (ns)

τ2 (ns)

Reference

Three-stage process Three-stage process Sequential process Three-stage process Three-stage process Non-activated Activated

0.7–3

5–15

[209]

2–10

10–50

[209]

Not evaluated 7–55

[206]

0.2–1

[35]

Not evaluated

250

–

[210]

0

∆cf 0 ∆y < 0 ∆z > 0

AICVI

AIBIIICVI2

III

B

VI

2C 3

BIIICVI

∆z < 0 AI2CVI

AI

∆x > 0

∆x < 0

Figure 4.9 Generic phase triangle of group I (AI ), group III (BIII ) and group VI (CVI ) atoms. Tie lines connect phases which – at a given temperature – are in equilibrium. For all Cu-chalcopyrites and Ag-chalcopyrites, the ternary phase AI BIII CVI 2 exists along the line

BIII

AI 2 CVI –BIII 2 CVI 3 . More ternary phases may be present. As an example, we depict an additional ternary phase between AI BIII CVI 2 and BIII 2 CVI 3 . In addition, more tie lines may be present in the specific phase triangle.

The CuInSe2 CP phase in Figure 4.10 is denoted as α. According to Table 4.8, this phase has a formation enthalpy at 0 K of −49.8 kJ mol−1 (−2.03 eV) which is in good agreement with calculated values (−2 eV [134] and −1.8 eV [265]). Other ternary phases along the In2 Se3 –Cu2 Se isopleth are CuIn3 Se5 (β), CuIn5 Se8 (χ), and δ. The β phase is a defect CP, built by ordered arrays of defect complexes 2VCu –InCu on the cation sublattice [134]. This phase can be derived from the CP CuInSe2 by having one 2VCu –InCu defect complex in every 20-atom Cu5 In5 Se10 unit cell [134]. Often, the β phase is referred to as ordered defect compound (ODC). A structure model of the ODC phase is still a matter of debate [258, 266–270]. Both, the α and the β structures exhibit an fcc Se sublattice. The existence range of the β-phase extends from a Cu content of 16.0 to 10.5 at%. The δ phase is the high temperature modification of both CuInSe2 and CuIn3 Se5 . It is of sphalerite type, that is, it does not show cation ordering. The third ternary phase along the isopleth of y = 0 is CuIn5 Se8 . Its stable structure is hexagonal while the tetragonal form is metastable [271]. For 7.5 at% < Cu < 10.5 at% CuIn5 Se8 and CuIn3 Se5 coexist in a two-phase region. As a metastable phase, CuInSe2 and other CPs can also occur in the CuAu ordering. The CuAu ordering P42m is most closely related to the chalcopyrite ordering as it also exhibits AI n BIII 4−n tetrahedra with n = 2. The formation energies of CP and CuAu differ only by 2 meV/atom. Other structures exhibiting n = 1 and 3 tetrahedra, CuPt ordering R3m or Y2-like ordering, are thermodynamically less stable. Early reports on their existence in CuInSe2 thin films [282] have not yet been confirmed. It shall be emphasized that all polytypes are only different with respect to

195


196

L

1000

1000 900

+

600 β

Temperature [°C]

Temperature [°C]

α α

δ

+ γ

α+ Cu2Se(HT)

400 β

300

α+ β

10

700

L +δ +α

600 500

15

20

25

at. % Cu

30

L4 +α InSe +α

300 200

35

L3 +α

+ Cu2Se(HT)

α+ Cu2Se(RT)

L3 +δ

L +δ

400

200

(a)

δ

800

700

100

L3

β+δ

800

500

L

δ

900

α 0

10

(b)

20

α +γ-CuSe

30

40

50

at. % Cu

Figure 4.10 Pseudobinary equilibrium phase diagrams of (a) In2 Se3 –Cu2 Se and (b) InSe-CuSe. After Refs. [260, 262, 263].

their cation sublattice but have an identical fcc chalcogen sublattice. The existence of the CuAu ordering in CuInSe2 and CuInS2 can be inferred from diffraction measurements [283] and from Raman spectroscopy [284]. The metastable phases of CuInSe2 are not included in the phase diagram of Figure 4.10. The end members of the phase cut in Figure 4.10b are In2 Se3 and Cu2 Se. In2 Se3 exists in four modifications depending on composition and temperature. The βand the γ-phase can co-exist between 470 and 1080 K. β-In2 Se3 is a rhombohedral layered structure with slight selenium deficiency. The γ-In2 Se3 phase is of the defect wurtzite-type [285]. In order to satisfy the octet rule for sp3 hybridization, one-third of cation sites in γ-In2 Se3 are vacant. Both structures, β- and γ-In2 Se3 , do not exhibit an fcc anion sublattice. Cu2 Se exists as a low temperature α-phase (stable below 396 K) and as a high temperature β-phase (stable up to 1403 K). The β-phase has the anion fcc sublattice and thus is closely related to the CP structure. It can be formed by introducing anti-sites and interstitials of the type CuIn and Cui into the cation sublattice of CuInSe2 . In equilibrium, the CuInSe2 α-phase has a small but measurable existence range. Along the In2 Se3 –Cu2 Se pseudobinary cut (Figure 4.10b), CuInSe2 is found as

4.2 AI –BIII –C2VI Absorbers

the single phase with a Cu content from 24.0 to 24.5 at% (0.94 < [Cu]/[In] < 0.97. Samples with a Cu content of 25 at% (i.e., with the nominal composition of CuInSe2 ) are not single phase but include Cu2 Se as a secondary phase. Samples with a Cu content of 25 at%. At 649 K there is a phase transition from CuSe → Cu2 Se + Seliq . The stabilization of the Se liquid (Sliq ) on the surface of a growing film will depend on the Se vapor pressure. Above 795 K, the α-phase is in equilibrium with a Cu-Se liquid denoted as L3 in Figure 4.10a. Thus in case of high selenium pressure, the vapor growth of Cu-rich CuInSe2 may take place in the presence of a Cu-Se liquid as proposed by Klenk et al. [288]. Such vapor–liquid–solid (VLS) growth mechanism is known to reduce defect densities due to high mobility of atoms in the liquid. It should be noted, however, that ion mobility under Cu-rich conditions can also be realized along the y = 0 isopleth: the high temperature phase β-Cu2 Se is an ambipolar ion conductor. The diffusion of In3+ in β-Cu2 Se at 650 ◦ C was estimated as 4 × 106 cm2 s−1 which is a value typically found for diffusion in liquids [289]. Thus, even the presence of the Cu2 Se secondary phase during CuInSe2 growth may lead to a VLS type of growth process and thus to a low defect density. The phase diagram of Cu-In-Se shows some peculiarities which are also found for other Cu- and Ag-chalcopyrites: 1) In the Cu-In-Se system, the homogeneity range of the AI BIII CVI 2 phase extends toward x < 0 (x defined in Eq. (4.5)). This is also the case for CuInS2

197

198


[290], CuGaSe2 [291], CuInTe2 [261], CuGaTe2 [292], and AgInTe2 [261]. The homogeneity range of CuGaSe2 is smaller than that of CuInSe2 [287]. The homogeneity range of CuAlSe2 may be on the Cu-rich side of the y = 0 isopleth (i.e., x > 0) [256]. 2) In the Cu–In–Se system, a ternary phase of the chemical formula AI BIII 3 CVI 5 is stable at room temperature. This phase is a result of ordering of defect complexes of the type 2VA –BA [134]. The AI BIII 3 CVI 5 phase is also known for the Cu–Ga–Se system. One earlier paper by Ganbarov et al. reports on CuIn3 S5 and CuIn3 Te5 [293]. 3) A ternary phase of the chemical formula AI BIII 5 CVI 8 exists in many systems such as Cu–Ga–Se, Cu–Ga–S, Cu–In–S, Cu–Ga–Te, Cu–In–Te, Ag–Ga–Te, Ag–In–Se (see Table 4.7). Its structure is either cubic, tetragonal, or hexagonal at room temperature. Since binary or ternary phases can be concomitant during or after the growth of CPs, crystallographic relations between the CP and its secondary phases are important. Table 4.7 lists some crystallographic data of relevant phases. It also gives relative volumes of some binary and ternary phases. Growth of Cu(In, Ga)CVI 2 under Cu-rich conditions (or as for CuInSe2 even under stoichiometric conditions) will inevitably lead to the formation of a Cu-CVI secondary phase which, at growth temperature, is Cu2−x CVI . The couple Cu(In, Ga)CVI 2 and Cu2−x CVI can show an epitactic or topotactic relationship [294]. Table 4.7 reveals that, for the system Cu2−x Se/CuInSe2 with x = 0.2, the volume ratio at room temperature is close to unity. Thus, the volume mismatch is small and we may expect a small strain within the layers. The volume ratio systematically increases in the order In → Ga → Al. It becomes even larger for x → 0 and at higher temperature due to the thermal expansion coefficient of Cu2 Se (2.5 × 10−5 K−1 ) [295] being larger than that of CuBIII Se2 . Growth of Cu(In, Ga)CVI 2 under Cu-poor conditions may lead to the formation of a secondary Cu1−x (In, Ga)1+x/3 CVI 2 phase. In Table 4.7 we see that also the CuIn3 Se5 /CuInSe2 volume ratio is more close to unity that the CuGa3 Se5 /CuGaSe2 volume ratio. With respect to the volume mismatch to its secondary phases, α–CuInSe2 is in a privileged state compared to other CuBIII Se2 -chalcopyrites. 4.2.1.2 Multinary Phases AI BIII CVI 2 semiconductors of the system (Cu, Ag)(Al, Ga, In)(S, Se, Te)2 exhibit several isopleths of complete miscibility. For solar cells, the miscibility of different ternaries is important because it allows tailoring a desired electronic bandgap in the multinary system. Figure 4.8 depicts a selection of ternary phases in the Eg versus (2a + c)/4 parameter space. Those ternary isopleths for which complete miscibility has been shown experimentally are depicted as dashed lines in Figure 4.8. It includes the isopleths between most of the Cu-chalcopyrites. However, there appears to be no complete miscibility between some Cu-chalcopyrites and Ag-chalcopyrites [296] (marked as dotted lines in Figure 4.8). The reason for the miscibility gap may be the large difference in the ion radii between Ag and Cu.


Of technological importance is the ternary phase NaInSe2 . No miscibility between CuInSe2 and NaInSe2 has been reported. The reasons are that NaInSe2 exhibits a layered structure in its stable state (space group R3m) and that the mixing enthalpy for the reaction xNaInSe2 + (1 − x) CuInSe2 ⇔ Nax Cu1−x InSe2 is positive [297]. Vegard’s law states that the lattice parameter of a solid solution depends linearly on the concentration of the substituting elements [298]. This law is approximately valid in the systems Cu(In1−x Gax )Se2 [299] and CuInSe2−2× S2x [300]. Deviations from Vegard’s law are found, for example, for Cu1−x Agx InSe2 and other Ag-containing systems. The approximate validity of Vegard’s law can be used to tailor pentanary CP compounds without tetragonal distortion [301] and therefore without crystal field splitting. Tetragonal AI BIII CVI 2 compounds can also form solid solutions with binary II VI D E compounds. (Here DII and EVI denote the group II and group VI elements of the binary.) One source of interest in these solid solutions is a possible partial replacement of In (and Cu) by a less expensive element. Another source of interest in AI BIII CVI 2 –DII EVI systems is the possible intermixing taking place at the interface between the ternary and the binary. The system CuInS2 + 2CdSe ⇔ CuInSe2 + 2CdS has been investigated by Olekseyuk et al. [302, 303]. Figure 4.11 shows the isothermal section at 620 K. There exists a solid solution range between CuInSe2 and the two binaries in the CP structure. The δ phase is of the sphalerite type – the high temperature modification of CuInSe2 . (Note that here we use phase symbols different from those in the work of Parasyuk et al. [302].) Thus, by the addition of DII EVI to AI BIII CVI 2 , the sphalerite crystal structure can be stabilized down to room temperature. Finally, at large amounts of DII EVI the hexagonal η-phase is formed. Large two-phase fields are found in between α, δ, and η. If the lattice constants of each structure are transformed into an effective cubic lattice constant, then a linear variation according to Vegard’s law is observed [304]. For VI VI all CuInCVI = S, Se, Te, the ZB structure is found for 2 –ZnC 2 systems with C large x and the CP structure for small x [305]. There is a miscibility gap (two-phase field) for intermediate x. Since all ZnCVI 2 crystallize in the cubic structure, no solid solution η phase is formed. CulnS2

2CdS

α+δ

η

δ+η α

CulnSe2

δ

2CdSe

Figure 4.11 Isothermal section of the reciprocal system CuInSe2 + 2CdS ⇔ CuInS2 + CdSe at 620 K. After Ref. [302]. Marked are the phase fields of α : chalcopyrite, δ : sphalerite, and η : wurtzite as well as the two-phase fields.

199

200


4.2.1.3 Diffusion Coefficients The knowledge about the chemical diffusion coefficients in monocrystalline CPs is far less comprehensive than for CdTe. A few more data are available for polycrystalline films. Since thin film data depend on grain boundary density and thus are specific for each investigated film, these data may be used with caution. Cu is assumed to be the most mobile species in Cu-chalcopyrites. Comparison between the self-diffusion constant of Se and Cu in CuInSe2 crystals (Table 4.8) supports this assumption. Cu ions were found to migrate under an electric field in the form of Cu+ [276]. As VCu are acceptors in CuInSe2 (see Section 4.2.3.1) and Cui are donors, migration of Cu can locally change the doping type. Electromigration of ions has been applied for the formation of pn junctions in bulk CuInSe2 [276] and thin films [306, 307]. The diffusion constant of indium is only known for CuInS2 at 920 K. There, it is in the range of 10−8 to 10−10 cm2 s−1 and decreases with increasing pS2 partial pressure. This decrease in DIn is in accordance with a model which predicts the migration of indium anti-site defects by means of vacancies on the copper sublattice [308]. Indium anti-sites and copper vacancies are assumed to be very abundant in CP semiconductors [134]. For the case of Se diffusion in CuInSe2 , a slight increase of DSe with pSe2 has been found [277]. A much higher value of the selenium self diffusion coefficient of DSe ∼ 5 × 10−6 cm2 s−1 was measured for polycrystalline films [309]. This large value was attributed to grain boundary diffusion. The sulfur diffusion constant in monocrystalline CuInSe2 is small (Table 4.8). Again, for polycrystalline thin films, the value of DS appears to be 3 orders of magnitude larger [310]. The highest reported diffusion coefficient in CuInSe2 is the one for Cd at 723 K. With an activation energy of around Ea = 1.2 eV, DCd exhibits an unusual small pre-exponential factor of 160 cm2 s−1 [280, 281]. This is particularly surprising since the diffusion constant of Zn (the smaller atom) was found to be much lower [279]. In polycrystalline films, the diffusion coefficient of Mo has been determined with an activation energy of 0.53 eV and a pre-exponential factor of 1.3 × 10−8 cm2 s−1 . At 773 K this amounts to DMo = 5 × 10−12 cm2 s−1 . Ga diffusion in CuInSe2 is essential if a Cu(In1−x Gax )Se2 absorber is to be formed from the ternaries CuInSe2 and CuGaSe2 . Due to the different reaction kinetics, CuGaSe2 is often found at the back contact. For high device performance, Ga diffusion toward the heterojunction is required. When discussing Ga diffusion in CuInSe2 , its dependence on the element Na should be regarded. Lundberg et al. found that the presence of Na leads to a decreased Ga (and In) diffusion [311] both for Cu-rich and Cu-poor films. The role of the Cu content [312] is secondary. For polycrystalline CuInSe2 , Marudachalam et al. report a value of DGa = 4 × 10−11 cm2 s−1 at 650 ◦ C [313]. A similar value (DIn = 1.5 × 10−11 cm2 s−1 ) was found for In diffusion in CuGaSe2 . Both results were measured for films on Na-containing glass (soda lime) without specifying the Na content in the substrate and film. Quantitative results available for epitaxial CuGaSe2 on GaAs may not

4.2 AI –BIII –C2VI Absorbers Table 4.8

CuInSe2 physico-chemical and opto-electronic properties.

Property

Symbol

Melting point Heat of fusion (300 K) Entropy (300 K) Space group Lattice constant

Tm H0f S0 – ˚ a (A) ˚ c (A)

Thermal expansion coeffient (300 K) Density Optical band gap at 300 K Temperature dependence of Eg,a Electron effective mass Hole effective mass Effective density of states Intrinsic carrier concentration Electron mobility Hole mobility Static dielectric constant Optical dielectric constant Refractive index (600 nm) Absorption length Radiative recombination rate Radiative lifetime Bulk diffusion coefficients

Value or range

1259 K −49.8 KJ mol−1 39.55 J K−1 mol−1 I42d 5.7815 11.6188 α 8.32 × 10−6 K−1 (a axis) 7.89 × 10−6 K−1 (c axis) ρ 5.75 g cm−3 1.02 eV Eg dEg /dT −2 × 10−4 eV K−1 ∗ mn /m0 0.09 mp ∗ /m0 0.092 (light hole) 0.71 (heavy hole) 7 × 1017 cm−3 Nc Nv 1.5 × 1019 cm−3 5 × 109 cm−3 ni µn 100 − 1000 cm2 V−1 cm−1 50 − 180 cm2 V−1 cm−1 µp ε/ε0 13.6(2.4) 8.1(1.4) ε∞ n 3.0 Lα0 1 × 10−4 cm 1.6 × 109 cm−3 s−1 U0 τrad 1.6 × 10−6 s D(cm2 s−1 ) Cu: 10−10 (673 K) Se: ∼10−12 (973 K) S: 10−16 (850 K) Zn: 10−13 (773 K) Cd: 6 × 10−7 (723 K)

References [261] – – – [272] [273] [272] [273] [273] [273] [273] Calculated Calculated Calculated [274] [274] [275] [275] [34] – [33] [33] [276] [277] [278] [279] [280, 281]

easily be compared with [314]. They clearly show increased Ga diffusion for Cu-rich films. 4.2.2 Lattice Dynamics

The vibrational spectrum of CP crystals with point group D12 2D consists of 24 zone-center vibrational modes [315]

1A1 + 2A2 + 3B1 + 4B2 + 7E

(4.6)

201

202


Of these, three are classified as acoustic modes (B2 + E) and the remaining 21 as optic modes. The E modes are doubly degenerated. Except for two silent modes (2A2 ), there are 22 Raman active modes (including the splitting of doubly degenerated modes) 1A1 + 3B1 + 3B2 (LO) + 3B2 (TO) + 6E(LO) + 6E(TO)

(4.7)

and 18 infrared active modes 3B2 (LO) + 3B2 (TO) + 6E(LO) + 6E(TO)

(4.8)

For non-resonant excitation, the A1 mode is the Raman mode of highest intensity (see Table 4.9 for A1 peak positions). It corresponds to a pure anti-phase vibration of the anions in the lattice while the cations remain at rest. A critical assessment of optical mode assignment can be found in Ref. [316]. Phonon dispersion curves have been calculated for CuInSe2 [317] and CuInS2 [318]. The calculations show that for the A1 mode it holds dE/dk > 0 around the point. Hence, we expect that in the case of high defect density and a resulting phonon confinement, the A1 mode becomes asymmetric toward higher wavenumbers – accompanied by a significant peak broadening [319]. This indeed has been observed and has been used for quality assessment of CP absorber layers [320, 321]. The A1 mode also shifts to higher wavenumbers for decreasing [Cu]/[BIII ] ratio [315]. This has been attributed to differences in the bond-stretching force constants between BIII –CVI 2 and Cu-CVI 2 bonds. In the case of CuInSe2 the shift is much smaller than the effect of peak broadening with decreasing [Cu]/[BIII ] ratio [322]. In the solid solution systems CuIn1−x Gax Se2 , the A1 mode shows a single mode behavior. The position of the A1 mode increases linearly with the Ga content from the position corresponding to CuInSe2 (173 cm−1 ) to that of CuGaSe2 (184 cm−1 ) Table 4.9

Compound CuAlS2 CuGaS2 CuInS2 CuAlSe2 CuGaSe2 CuInSe2 CuGaTe2 CuInTe2 AgGaS2 AgInS2 AgGaSe2 AgInSe2 AgGaTe2 AgInTe2

Wavenumbers of A1 Raman modes of different Cu- and Ag-chalcopyrites. Wavenumber (cm−1 )

References

315 310 290 187 184 173 135 125 295 265 175 172 130 122

[323] [324] [324] [325] [315] [326] [327] [328] [323] [329] [329] [329] [329] [330]

4.2 AI –BIII –C2VI Absorbers Table 4.10 Main vibrational modes from secondary phases detected by Raman scattering in AI -BIII -CVI 2 processed layers.

Phase/References

Vibrational modes (cm –1 )

CuAu − CuInS2 [283] CuAu − CuInSe2 [283] CuS [337, 338] CuSe, Cu2−x Se [332, 337, 339] Cu3 Se2 [339] ODC Cu-In-Se [332, 340]

305 183 19, 475 43, 263 49, 58 156 (CuIn2 Se3.5 ) 153 (CuIn3 Se5 ) 151 (CuIn5 Se8 ) 327, 341, 360 382, 408 169, 240 140 116, 173, 219

CuIn5 S8 [335, 341] MoS2 [335] MoSe2 [336] InS [338] InSe [342]

[315, 326]. In the system CuInSe1−y Sy , we find a two-mode behavior where the relative intensity of the S-derived and the Se-derived A1 modes depends on the value of y [331]. In addition, the A1 Se-Se band shifts to higher wavenumbers with increasing S content. Raman scattering is very well suited for detection of secondary phases in the layers. These include Cu-rich binary phases (as Cu-S and Cu-Se compounds) that are characteristic of layers grown under Cu-excess conditions [332–334]. On the other hand, growing of the layers under Cu-poor conditions favors the formation of Cu-poor ODCs, that have been observed mainly in Cu-poor CuInSe2 systems [332]. Additional phases detectable by Raman scattering include the spinel CuIn5 S8 phase from Cu-poor CuInS2 [335], MoS2 , and MoSe2 at the interface region between the absorber and the back Mo contact [335, 336], and In-S and In-Se binary compounds. Table 4.10 summarizes the main vibrational modes characteristic of some frequently observed secondary phases. 4.2.3 Electronic Properties

A ternary AI –BIII –CVI 2 semiconductor in the CP structure can be viewed as the isoelectric analog of a binary AII –BVI compound in the ZB structure. The tetragonal unit cell of the CP structure contains eight atoms – four times more than the cubic ZB structure. Thus, the volume of the CP Brillouin zone is four times smaller (Figure 4.12). CPs show smaller bandgaps than their binary analogs [343]. This so called bandgap anomaly is caused by chemical Eg chem and structural Eg struc contributions. Chemical, we call the p-d hybridization contribution Eg d . Due to the high lying d-states of the AI atom, hybridization of anion p-states, and cation

203

204


T∆ kz Γx

T∆ NL Γ TX

NΣ

TX

ky

kx ΓW

ΓW

Figure 4.12 Chalcopyrite (CP; bold lines) and zincblende (ZB; solid lines) Brillouin zone. The polyhedra represented by the dotted lines give the ZB reciprocal-space regions which fold into the chalcopyrite Brillouin zone. Symmetry points refer to the chalcopyrite structure while the subscripts give the ZB symmetry points.

d-states leads to an upward shift of the anion p states and thus to a reduced bandgap. Chemical, we also call the cation electronegativity contribution Eg CE . It is caused by the charge difference between the two cations. The structural contribution Eg struc of the bandgap anomaly is mainly caused by the anion displacement u. Only a small contribution comes from the tetragonal distortion η. The electronic band structure of several Cu-chalcopyrites has been calculated in Refs. [168, 169]. Table 4.11 shows experimental bandgap energies of CP and related semiconductors. Table 4.8 lists effective masses of electrons and holes in CuInSe2 and the effective densities of states derived by use of Eq. (4.1). The measured electron and hole mobilities are in good agreement with calculated values for CuInSe2 of mn = 0.089 × m0 and mp = 0.69 × m0 [265]. Electron and hole mobilities are in the same value range as for CdTe (Table 4.1); although the valence band minimum (VBM) of CuInSe2 has 50% contribution of relatively localized Cu 3d. In the following, we will concentrate on CuInSe2 and other technologically relevant CPs. Cu-BIII -O2 delafossites have indirect band gaps of 1.97 eV (CuAlO2 ), 0.95 eV (CuGaO2 ), and 0.41 eV (CuInO2 ), however much larger optical band gaps above 3 eV [344]. 4.2.3.1 Single Point Defects In a ternary semiconductor – due to the three different atoms – many types of native point defects (single defects) can potentially form. Their concentrations depend

Room temperature bandgaps of phases in ternary Cu and Ag phase diagrams.

Table 4.11

Eg (eV)

Compound

Eg (eV)

Compound

Eg (eV)

Compound

Eg (eV)

Compound

Eg (eV)

CuAlS2 CuGaS2 CuInS2 CuAlSe2 CuGaSe2 CuInSe2 CuAlTe2 CuGaTe2 CuInTe2 AgAlS2 AgGaS2 AgInS2 AgAlSe2 AgGaSe2 AgInSe2 AgAlTe2 AgGaTe2 AgInTe2

3.49 2.43 1.52 2.67 1.68 1.04 1.65 1.24 1.08 3.6 2.73 1.87 2.5 1.83 1.24 2.35 1.36 1.04

Al2 S3 Ga2 S3 β − In2 S3 Al2 Se3 Ga2 Se3 α − In2 Se3 Al2 Te3 Ga2 Te3 In2 Te3 S – – Se – – Te – –

4.1 2.7 2.2a 3.1 2.6 1.35 1.2 1.35 1.05 2.6 – – 1.85 – – 0.33 – –

Cu2−x S(x ≤ 0.005) Cu2−x S(x = 0.2) CuS Cu2 Se Cu2−x Se(x = 0.2) CuSe Cu2−x Te – – Ag2 S – – β − Ag2 Se – – Ag2 Te – –

1.2 [345] 1.55 [346] 2.2 1.0–1.1 [350] 1.4a [351] About 2.0 1.1 – – 0.9–1.0 – – 0.1 – – 0.03 – –

– – CuIn3 S5 – CuGa3 Se5 CuIn3 Se5 – CuGa3 Te5 CuIn3 Te5 – – – – – – – – –

– – 1.35a [347] – 1.85 [352] 1.23 [352] – 1.0a [355] 1.25 [356] – – – – – – – – –

– – CuIn5 S8 – CuGa5 Se8 CuIn5 Se8 – CuGa5 Te8 CuIn5 Te8 – – AgIn5 S8 – – AgIn5 Se8 – – AgIn5 Te8

– – 1.3a [348, 349] – 1.8 [353] 1.15 [354] – 1.0a [355] 1.22 [357] – – 1.7 [358] – – 1.27a [359] – – 1.1a [360]

a Indirect

band gap.


Compound

205


c

b

2− 0

∆Hf

206

a

0

0

VBM

CBM -zp

-zn

Reduced chemical potentials

Figure 4.13 Sketch of defect formation energies, Hf , as a function of the reduced chemical potentials ζn , ζp of three different defects. Solid dots denote values of ζn , ζp where transitions between charge states occur. The defect formation energy is constant for the neutral defect but varies with ζn , ζp for the charged defect. (a) Acceptor type defect, (b) donor type defect, and (c)

amphoteric defect where the neutral state is highly unstable. The energy ranking of the three defects is arbitrary. Also arbitrarily, the donor type defect was selected to demonstrated a negative defect formation energy where the defect would spontaneously form if the Fermi level is close to the conduction band.

on material composition and defect formation enthalpies. The defect formation enthalpy of a neutral defect depends on certain chemical potentials [134]. The reason is that in forming an atomic defect, the atom is transferred to or from a chemical reservoir. The chemical potential, for example, of the metal of atom i, is characterized by a chemical potential µi . The defect formation of a charged defect, in addition, depends on the chemical potentials of electron (ζn ) or hole (ζp ) – as the charge carrier is transferred from or to the carrier ensemble. Thus, defect formation enthalpies must be given in a multidimensional parameter space, that is, Hf = Hf (µi , ζn , ζp ). Figure 4.13 schematically depicts the dependence of the defect formation energy on the reduced chemical potentials of electrons and holes for an acceptor-type defect (Figure 4.13a) and a donor-type defect (Figure 4.13b). (The use of reduced chemical potentials of charge carriers implies position-independent band edges – see Figure 2.1 for definitions.) Figure 4.13 assumes that the chemical potentials of the atomic reservoirs are fixed. We see that for the acceptor-type defect (Figure 4.13a) in the (−) charged state, Hf decreases with increasing ζn , that is, with increasing Fermi level in the band gap, while Hf is constant in the neutral state. As an extreme, Hf can become negative at high electron concentrations. Zhang et al. [134] calculated CuInSe2 defect enthalpies as a function of composition and charge carrier chemical potentials. For the device relevant composition of CuInSe2 (Cu-poor), Domain et al. calculated the formation enthalpies of native cation point defects in the order [265] VCu < CuIn < VIn < InCu < Cui . The neutral Cu vacancy has the lowest formation enthalpy of about 0.1 eV. With rising Fermi energy in the bandgap, that is, with smaller En , the neutral defects VCu , CuIn , and


1.0 0.9 0.8

Energy [eV ]

0.7 0.6

InCu VSe Cui Vln Culn

Ec (+ /0) (2+ /+) (0/−) (+ /0)

(2− /3−) OSe (−/2−)

0.5 0.4

Vln

0.3 Cu ln 0.2 0.1 0.0

CdCu MgCu, ZnCu

(−/2−)

(+ /0) (V -V ) Se Cu

(−/2−) (+/0)*

(−/2−)

OSe

(0/−)

Naln

(−/2−)

Donor Acceptor

(0/−)

Vln

(0/−)

VSe VCu

(2+/0) (0/−)

Naln

(0/−) (VSe-VCu)

(+/−)

Mgln Cdln Znln

(0/−) (VSe-VCu)

(0/−)* Ev

0 4 8 12 Number of experiments

Figure 4.14 Theoretical values of electronic defect transition levels in CuInSe2 . After Refs. [134, 362, 363]. Asterisks mark transitions which only occur in a metastable state.

VIn will become negatively charged. In the charged state, however, the formation energies of these defects will further reduce (see Figure 4.13a). The result can be a negative formation energy of VCu as first calculated by Zhang et al. [134]. This means spontaneous formation of VCu − acceptors which would inhibit a further rise of the Fermi energy. Also in AgBIII Se2 compounds, the AI vacancy has the lowest formation enthalpy [361]. For the pure ternary CuInSe2 , Figure 4.14 gives calculated defect transition levels of native defects [134] (these data where obtained by LDA correction using fixed VBM) and impurity point defects [21]. The most shallow intrinsic acceptor level in CuInSe2 is formed by the VCu defect. As this defect also has a low formation enthalpy at Cu-poor composition, it appears to be responsible for p-type doping. The next deeper intrinsic acceptor level is formed by the In vacancy at E(0/−) = Ev + 0.17 eV. A high density of this defect would require a film composition which is In-poor. Substitutional group II dopants such as Cd, Zn, and Mg can form shallow acceptor states if they are on an In site. To prevent the concomitant formation of DII Cu donors (DII = Cd, Zn, Mg), one would have to select Cu-rich growth conditions. This, however, would at the same time reduce the density of assisting VCu acceptors. The BIII Cu 2+ antisite defect is a double donor which can cause n-type conductivity. However, the BIII Cu defect can transform, by capture of two electrons, into a DX-like center which essentially corresponds to a BIII i + -V-Cu defect pair [364]. In this configuration where the In atom has moved from the lattice position in direction of [226], the InCu defect forms the deep BIII DX level. The spontaneous

207

208


BIII Cu 2+ + 2e → BIII DX transformation takes place at a Fermi level position of Ev + 0.92 eV and Ev + 0.84 eV for CuInSe2 and CuGaSe2 , respectively [364]. Anion–cation antisite defects such as SeCu or CuSe are estimated to have high formation enthalpies (Hf > 5 eV) [275] – an estimation which is based on the Van Vechten model [365]. In contrast, the nature of the VSe defect is still a matter of debate. Earlier, it was believed that the VSe (2+/0) transition energy is close to the conduction band or, in other words, VSe can form a shallow double donor [275, 297] that can lead to n-type conductivity. More recent band structure calculations point out that the VSe (2+/0) transition in CuBIII Se2 (BIII = In, Ga) is close to the VBM [366]. Calculated transition energies for VSe (2+/0) and VSe (0/−) are given in Figure 4.14. The donor transition VSe (2+/0) is very far from the conduction band minimum (CBM) and the VSe (0/−) acceptor transition is very far from the VBM [362]. Thus, the isolated VSe defect shows amphoteric behavior where the usual order of acceptor and donor transitions is inverted. The reason for the unusual behavior of the VSe defect is the formation of BIII –BIII dimers in the VSe 0 configuration and BIII –BIII bond breakage in the VSe 2+ configuration, thus a strong lattice relaxation. Figure 4.15 shows the principle order of bonding and antibonding states of the VSe in CuInSe2 . For a CuInSe2 absorber with −ζp ∼ 0 eV, that is, for strongly p-type material, the VSe 2+ configuration has the lowest formation enthalpy and thus forms the equilibrium state. Population of the VSe 2+ state by an electron using light or voltage bias induces, after thermal activation, the VSe 2+ → VSe 0 transition. Due to the BIII − BIII dimer formation, the bonding state of the VSe 0 configuration shifts from within the conduction band into the valence band. Now, the bonding state of the VSe 0 configuration is below the VBM and will be occupied by a second electron. 0 , (V − VSe Se + VCu)

2+, (V + VSe Se + VCu)

CBM

ED

antibonding EA

VBM

bonding BIII–BIII dimer

no dimer

Figure 4.15 Schematic presentation of the bonding and antibonding states of BIII levels at the VSe site of CuInSe2 in the configuration VSe 0 (left) and VSe 2+ (right). After Ref. [362]. In the VSe 0 configuration, the bonding level can be occupied by two electrons donating two holes to the VBM. Occupation of the antibonding level leads to VSe − and VSe 2− .


Thus, a second hole is released to the VBM. As a result, the excitation of an electron (which was the initial effect) leads to p-type photoconductivity [366]. The VSe defect can be converted from an electron trap into a hole trap. The metastable VSe 0 configuration relaxes to the stable VSe 2+ by capture of two holes from the valence band. This relaxation is connected with an energy barrier and requires activation. It can be observed with a large time constant, especially at low temperature. It was argued by Lany and Zunger that the VSe density may be small [362]. Since the VCu density will exceed the VSe density (even in stoichiometric CuInSe2 ) and since furthermore the pair binding energy of the (VSe + VCu ) complex is negative (see below), most of the VSe defects are bound to VCu [362] or, in other words, VSe should occur mostly in the form of a (VSe + VCu ) defect complex. 4.2.3.2 Defect Complexes So far, information about defect complexes in CuInSe2 and related compounds comes from theoretical calculations rather than from experiment. The low formation enthalpy of several isolated native point defects of opposite charge signals the possibility of easy formation of charge-neutral defect complexes [134]. For CuInSe2 , the formation enthalpies of the isolated defect complexes (2VCu + InCu ), (CuIn + 2VCu ), (VCu + Cui ), (InCu + CuIn ), and (VSe + VCu ) have been calculated [134, 362]. Outstanding is the (2VCu + InCu ) complex which, for Cu-poor CuInSe2 , shows a negative [134] or zero [265] formation enthalpy. This defect complex does not produce significant deformation of the crystal [265]. The two VCu defects produce an expansion and the InCu defect a compression of the unit cell. Figure 4.2c shows the calculated minimum Madelung energy configuration of the (2VCu + InCu ) defect complex according to band structure calculation [134]. A minimum Madelung energy has also been found for the (2VCu + InCu ) complex in CuInS2 [367]. Formation of the (2VCu + InCu ) complex can explain the large existence range of the α phase at room temperature evident in Figure 4.10a. The electronic benefit is that deep levels of VCu and InCu are moved out of the bandgap region [368]. No transition level of the (2VCu + InCu ) complex shows up in the bandgap range in Figure 4.14. Thus, a large off-stoichiometry of CuInSe2 can be accommodated without forming deleterious amounts of electronic states in the bandgap. Similar as the InCu defect, the (2VCu + InCu ) complex can show DX behavior [364] (see above). As outlined above, calculated binding (not formation) energies of the (VSe + VCu ) complex in CuInSe2 and CuGaSe2 are negative [362] for all positions of the Fermi energy in the bandgap. Thus, the majority of VSe defects will be bound in (VSe + VCu ) complexes. Lany and Zunger [362] showed that the principle order of bonding and antibonding states in the (VSe + VCu )− and (VSe + VCu )+ configurations is similar as for the isolated VSe defect configurations (see Figure 4.15). The V0Se configuration having the BIII –BIII dimer corresponds to (VSe + VCu )− and the V2+ Se configuration showing BIII –BIII bond breakage is also found for (VSe + VCu )+ . Thus, the principle mechanism of metastability proposed by the strong shift of bonding states is applicable to the (VSe + VCu ) complex. The reactions involving a change from

209

210


donor to acceptor configuration are (VSe + VCu )+ + e− → (VSe + VCu )− + h+

(4.9)

with an activation energy of 0.1 eV for CuInSe2 and 0 eV for CuGaSe2 and (VSe + VCu )+ → (VSe + VCu )− + 2h+

(4.10)

with an activation energy of 0.73 eV for CuInSe2 and 0.92 eV for CuGaSe2 . The reverse reaction bringing the defect complex back to the acceptor configuration involves the capture of two holes (VSe + VCu )− + 2h+ → (VSe + VCu )+

(4.11)

with an activation energy of 0.35 eV for CuInSe2 and 0.28 eV for CuGaSe2 [362]. Figure 4.14 shows the transition energies of (VSe + VCu )− in CuInSe2 . Graph c in Figure 4.13 schematically shows the formation enthalpy and labels the stable states as a function of the Fermi energy. The topmost level would correspond to the (VSe + VCu )−/2− transition. It is derived from occupation of the antibonding state in the (VSe + VCu )− configuration. The (VSe + VCu )2− state is stable for a Fermi level close to the CBM, that is, for −ζn → 0. The (VSe + VCu )+/− transition is located in the lower part of the bandgap. Note that this transition energy does not cause an optical transition level. It rather means that if the Fermi energy is below this level, the (VSe + VCu )+ configuration is stable and the (VSe + VCu )− configuration is metastable. If the Fermi energy lies higher in the bandgap, the (VSe + VCu )− is stable and (VSe + VCu )+ is metastable. The shallow donor transition (VSe + VCu )+/0 and the shallow acceptor transition (VSe + VCu )0/− can only occur from a metastable state. This is why these transitions are marked with an asterisk in Figure 4.14. In summary, the (VSe + VCu ) complex can lead to neither equilibrium n-type nor equilibrium p-type conductivity. Similar as for the VSe defect, however, it can lead to metastable p-type conductivity [362]. Furthermore, it introduces an antibonding level in the (VSe + VCu )− configuration which is situated 0.85 eV above the VBM in CuInSe2 . 4.2.3.3 Defect Spectroscopy Defect densities in CP absorber layers have been determined by electrical methods such as AS, DLTS, drive level capacitance profiling (DLCP), and optical methods such as PL spectroscopy. We will see that the electrical and optical methods bring out different aspects of defects in the CP films. Electrical Spectroscopy Methods Evaluation of temperature dependent AS was mostly based on the work of Walter et al. [102]. In Section 2.6.4, it is shown that a defect density, which is constant in the z-direction, can be quantified in absolute numbers as a function of energy. Figure 4.16 gives the defect density as derived from AS for a CuInSe2 absorber [193, 194]. Two distinct defect maxima can be discerned which for the sample under investigation are located at about 0.15 and 0.3 eV. Using AS, these two peaks have been found for many samples of Cu(In, Ga)(Se, S)2 although at slightly varying energetic positions. The defect density at 0.3 eV in


Defect density (cm−3 /eV)

1018 CulnSe2 T= 80K–320K ν0 = 5 × 1011s−1

1017

1016

1015

0.1

0.2

0.3

0.4

0.5

Defect depth Ed (eV) Figure 4.16 Exemplary defect density of a CuInSe2 absorber layer as function of energy (after Ref. [374], with permission). The defect densities at about 180 and 300 meV are labeled N1 and N2 , respectively, in the text.

CuInSe2 has been baptized N2 defect. By use of DLTS measurements it has been identified as a majority carrier trap [369–372]. It was claimed that the N2 defect can act as an intrinsic reference level which is essentially constant in the system Cu(In1−x Gax )Se2 and increases with y in the system Cu(In, Ga)(Se1−y Sy )2 [373]. Using N2 as a reference level, a valence band offset of −0.23 eV have been deduced between CuInSe2 and CuInS2 and below 0.04 eV between CuInSe2 and CuGaSe2 which is largely in accordance with theoretical calculations (see Section 4.5). In CuInS2 devices, also a defect at about 0.3 eV has been observed by AS [192, 193] and has accordingly been labeled as N2 [373]. In the energy range of 0.22–0.35 eV DLTS shows up to four majority carrier traps where the trap at 0.35 eV has the highest density [375]. Thus, it appears that the N2 has an identical origin within the complete solid solution system of CuIn1−x Gax Se1−y Sy . An inverse relation between the N2 density and Voc is indicated in Refs. [376, 377]. In Ref. [378] the integral defect density derived from AS and DLCP measurements has been correlated with the Voc . It appears that the density of the N2 defect is a good marker for medium-quality CP films. Figure 4.17 shows that the N2 defect density has a minimum at the energy gap of about 1.15 eV which corresponds to a Ga content of ≈0.3 [379]. Since, x ≈ 0.3 in CuIn1−x Gax Se2 solar cells marks the stoichiometry of the highest efficient solar cells, there appeared to be a relation between the N2 defect density and the device performance. However, for films which give devices with efficiencies above 15%, the N2 defect is below the detection limit of AS [380, 381] which we estimate at 1014 cm−3 . Thus, the upper limit of deep defects in device quality CuIn1−x Gax Se2 solar cells may be 1014 cm−3 . The peak at 150 meV in Figure 4.16 has been labeled N1 . The origin of this N1 defect is still a matter of debate. DLTS shows at least one minority carrier trap at

211


1017 Defect concentration Nd [cm−3]

212

1-Stage 3-Stage 1016

1015

1.0

1.2

1.4

1.6

Band gap energy Eg [eV] Figure 4.17 Density of the N2 defect seen in Figure 4.16 as a function of the energy gap of CuIn1−x Gax Se2 absorber layers as determined by temperature dependent admittance spectroscopy (after Ref. [379], with

permission). The CuIn1−x Gax Se2 absorbers were grown by a single-layer process except for one film (three-stage process). The bandgap has been derived from external quantum efficiency measurements.

about Ec − 0.1 eV which may be identical to the N1 defect [371, 383]. Herberholz et al. find the energy position of N1 to vary with temperature annealing in air [384]. As a conclusion, they assign the activation energy of N1 to the Fermi level intersecting a homogeneous defect density at the CdS/CuIn1−x Gax Se2 interface. Thus, the energy of the N1 defect would correspond to the value of En,az=0 . Indeed, AS cannot discriminate between: (i) a defect density which is discrete in energy and homogeneous in space, and (ii) a defect density which is homogeneous in energy and discrete in space. In addition, a ‘defect step’ in AS may result from a barrier for the diode current such as a back surface diode [103]. An assignment of N1 to a volume defect has been deduced in Ref. [380]. In the solid solution system of CuIn1−x Gax Se1−y Sy the energy of the N1 defect level is independent of x [373] but dependent on y. In highly efficient CuIn1−x Gax Se2 based devices, the N1 defect is often the only defect which can be detected by AS [377]. The N1 defect density was found to be 1016 cm−3 being largely independent of x in CuIn1−x Gax Se2 [381]. If this defect is a bulk defect, it will act as a doping level and will not be an efficient recombination center. A further interpretation of the N1 defect is in assignment to a barrier at the back contact [382]. Using DLTS analysis, other defect levels have been identified as majority carrier traps at Ev + 0.94 eV [383] and Ev + 0.5 eV [370, 380]. However, no correlation with device performance was reported, probably because of their low densities. Electron spin resonance measurement shows Fe(III) impurities of 1017 cm−3 density in certain films [385]. Photoluminescence Spectroscopy Thin films of CuInSe2 , CuGaSe2 , and CuInS2 all show characteristic PL features depending on their stoichiometry deviation x. Exemplarily, Figure 4.18 gives PL spectra measured at 10 K of CuInSe2 films that


DA2 ∆X

Figure 4.18 Photoluminescence (PL) spectra of CuInSe2 films of different stoichiometry deviation measured at 10 K. Redrawn from Ref. [386].

DA1 exc

PL intensity [a.u.]

+ 0.18

+ 0.05

− 0.06

− 0.09

− 0.11 0.85

0.90

0.95

1.00

1.05

Energy [eV ]

where grown under Cu-excess (x > 0) and with Cu-deficiency (x < 0) [386]. (Note that the Cu-excess was removed by KCN etching.) Films grown under Cu-excess show excitonic luminescence peaks and two or more donor–acceptor pair (DAP) transitions. The excitonic emission may be composed of bound and free excitons [387]. The DAP transitions have been explained by Siebentritt et al. based on one donor (ionization energy 12 meV) and two acceptors (ionization energies 40 and 60 meV) [386]. These defects are too shallow to be detected by AS (see above). The DAP peaks show the usual blue shift by 1–2 meV per decade of increasing excitation power [388, 389]. For CuInS2 and CuGaSe2 , also a deep luminescence may be observed [390, 391] which has been used for quality assessment of thin film absorbers [392]. The total PL emission intensity of Cu-rich films is at least a factor of ten smaller than that of Cu-poor films [393, 394]. The PL spectrum of a Cu-poor thin film with x = −0.1 has the following characteristic: excitonic luminescence is missing and quasi-DAP transitions are observed. The quasi-DAP emissions are asymmetrically broadened (see Figure 4.18), have a very large blue shift (>10 meV per decade) upon increasing excitation power and show a red shift upon increasing temperature [386, 393, 395, 396]. Missing exciton emission as well as quasi-DAP emission may both be explained by potential fluctuations [397–399]: due to the large covalent bonding and the high dielectric constant, excitons in CP crystals are of the Wannier–Mott type with a radius much larger than the lattice spacing. The potential fluctuations lead to a dissociation of those excitons. In the picture of potential fluctuations, the excitation-induced blue shift of the quasi-DAP transitions is due to: (i) the increasing probability to populate defects outside the minima and maxima of the band edges, and (ii) the decreasing

213

214


amplitude of the fluctuating potentials by neutralization and screening of defect states. By evaluation of the asymmetric quasi-DAP peak, a fluctuation amplitude above 100 meV has been deduced [399]. The density of compensating donors and acceptors is in the range of 1018 cm−3 [398]. It is reported that by air-annealing (see Section 5.2.7) the PL spectrum of a Cu-deficient film changes drastically and becomes similar to that of a Cu-rich film [400]. 4.2.3.4 Practical Doping Limits The achievable doping level in the CP Cu(In1−x Gax )(Se1−y Sy )2 solid solution system depends on the parameters x and y, on the Cu/III element ratio, and on impurity atoms such as Na. Our approach is to start with the discussion of pure CuInSe2 and consider variations induced by adding Ga and S in a second step. The effect of Na doping is discussed in Section 5.2.4. Single crystals of CuInSe2 have been prepared as n-type and p-type if they are doped via intrinsic defects [401]. Migliorato et al. reported that a high Se pressure induces p-type conductivity while an In excess and a minimum Se pressure results in n-type conductivity [402]. The ability to make n-type and p-type CuInSe2 was exploited by the formation of p-n homojunctions [403, 404]. Also CuInS2 can be doped p-type and n-type by intrinsic defects [405, 406]. In contrast, only p-type conductivity can be obtained by intrinsic doping of CuGaSe2 crystals [407] and thin films [408]. Only one work reports on extrinsic n-type doping of CuGaSe2 using a combination of Ge and Zn dopants [409]. Theoretically, it has been predicted that in thermodynamic equilibrium CuGaSe2 cannot be doped n-type. The reason is that the bulk Fermi level is pinned by the spontaneous creation of compensating VCu acceptors upon rising EF [410]. Considering Cu-poor films (stoichiometry deviation x < 0 and with Na doping) in the solid solution system Cu(In1−x Gax )Se2 with x < 0.5, free hole densities are in the range of 1015 –1016 cm−3 [411] and 5 × 1014 − 1 × 1015 cm−3 [412]. But one has to differentiate the relaxed state after heating in the dark or the more relevant state after white (or red) light exposure (see Section 6.2.4). In the latter case, which resembles the operating state of the solar cell, the carrier density in the bulk can be above 1016 cm−3 [413]. There are theoretical [287] as well as experimental [274, 414] evidences that a limited Ga addition to CuInSe2 increases the free hole density. However, there are also contradictory results which show no effect [411] or even a reverse effect [415]. Cu-poor films of CuGaSe2 without Na doping appear to have a very low carrier concentration of p ≈ 1012 cm−3 . Na doping or the use of soda lime glass increases this concentration to 1015 cm−3 [416, 417]. For Cu-poor films of Cu(In1−x Gax )(Se1−y Sy )2 with Na doping there is a clear trend of reduced carrier concentration upon increasing y. An example is shown in Figure 4.19 which give data of Seebeck measurements [418]. Cu-poor films of CuInS2 have a very low hole concentration and conductivity [419, 420]. As revealed by Figure 4.19, adding Ga appears to reduce this problem. Also for Cu-poor CuInS2 films, Na doping is efficient [421, 422].


p [cm−3]

1017

Ga /(Ga+ ln) 0.5

1016

0.3

1015

0 1014

0.0

0.2

0.4

0.6

0.8

S/(S+Se)

Figure 4.19 Density of holes (p) in Cu(In1−x Gax )(Se1−y Sy )2 thin films with Na doping (from the glass) and Cu poor composition as determined by thermoelectric (Seebeck) measurements. Redrawn after Ref. [418].

Cu(In1−x Gax )(Se1−y Sy )2 thin films grown under Cu-excess exhibit a secondary phase which is a binary Cu chalcogenide. If this phase precipitates at the surface, it can be etched selectively by a cyanide solution [423, 424]. After etching, the films are close to the ideal composition (Cu/III ≈ 1). These films exhibit a hole concentration in the range of p ≈ 1017 cm−3 [416, 425]. The influence of Na doping is negligible [426]. 4.2.3.5 Carrier Mobility In high efficiency polycrystalline CP-based solar cells, the absorber’s micro-crystallites (grains) have the same dimension as the film thickness and can extend from the back contact up to the heterojunction. In the latter case, carriers pass no GB in the operation modus of the solar cell. Thus, we are interested in carrier mobility within the grains (bulk mobility) which may be measured in single crystals or epitaxial layers. Reported values of single crystal electron mobilities in CuInSe2 at room temperature are µn = 100–1000 cm2 V−1 s−1 [274, 275]. The bulk hole mobilities of CuInSe2 crystals are µp = 50–180 cm2 V−1 s−1 [274], that is, lower than the electron mobilities. Values around µp = 2 × 102 cm2 V−1 s−1 were also found in solid solution system of Cu-poor epitaxial Cu(In1−x Gax )Se2 [414] as well as for Cu-rich epitaxial CuGaSe2 [427]. Modeling reveals that in the latter material, nonpolar optical phonon scattering and acoustical phonon scattering dominate the mobility at room temperature while charged defect scattering only becomes effective at lower temperature [427]. No data on µn are available for polycrystalline films. Hole mobilities in polycrystalline films are probably subject to transport barriers at the grain boundaries [427]. Reported hole mobility values are in the range of 20 − 40 cm2 V−1 s−1 [412, 427, 428] and therefore about 1 order of magnitude lower as in the bulk. Typical mobility values which have been used for device simulation are µn = 50 − 100 cm2 V−1 s−1 and µp = 20 − 25 cm2 V−1 s−1 [79, 185].

215

216


4.2.3.6 Minority Carrier Lifetime Time-resolved photoluminescence (TRPL) measurements were applied to measure the lifetime in CP thin films and devices. PL decay of Cu-poor CuIn1−x Gax Se2 samples can exhibit up to two distinct slopes which give two distinct lifetimes. The initial (τ1 ) and a late decay lifetimes (τ2 ) of CP absorbers are given in Table 4.4. In the case of a single decay lifetime [429] this is referred to as τ1 in Table 4.4. Ohnesorge et al. were able to correlate the τ2 of bare CuIn1−x Gax Se2 absorbers with the Voc value of respective solar cells [206]. Values up to 50 ns for τ2 at room temperature have been reported in Refs. [206, 209]. The decay lifetime τ1 of CuIn0.7 Ga0.3 Se2 absorbers of outstanding material quality is in the range of 100 ns [429] with the largest value of 250 ns [429]. We note that this lifetime equals one-quarter of the radiative lifetime limit (see Table 4.8). However, the measured lifetime value strongly reduces upon air exposure and can only be measured within minutes after film preparation. Obviously, the material degrades with time and this may be the reason for the large distribution of τ1 in Table 4.4. The non-degraded lifetime value can be sustained by capping the CuIn0.7 Ga0.3 Se2 layer with a buffer layer, such as CdS [429]. It should be noted that in finished solar cells (with TCO), the apparent lifetime τ1 can again be smaller due to the charge separation effect. Differences in lifetime between Cu-rich and Cu-poor CuIn0.7 Ga0.3 Se2 absorbers were investigated in Ref. [430]. The lifetime of Cu-rich absorbers after growth and after device finishing was smaller by a factor of five. This agrees with the smaller PL intensity of Cu-rich films by a factor of ten (see Section 4.2.3.3). In Section 2.4.6, it was shown that a minority carrier lifetime well in excess of 10 ns leads to a diode current which, at forward bias corresponding to Voc , is limited by QNR recombination. As a result, the diode quality factor should be close to 1 (see Table 2.2), as is the case in high efficiency CuIn0.7 Ga0.3 Se2 absorbers (Figure 6.9). 4.2.4 Optical Properties

Optical functions of Cu-chalcopyrites have been investigated with reference to the electronic band structure in Ref. [431]. We are interested in the near bandgap region and in the absolute absorption coefficient. Since CP semiconductors exhibit the minimum bandgap at the Brillouin zone center, we look at the optical transitions at the point. Figure 4.5 schematically derives the energy states at the bandgap of CuInSe2 and CuGaSe2 from a hypothetical ZB structure which does not show spin-orbit splitting. The introduction of the tetragonal crystal field splits the degenerate 15 state into states 4 and 5 . This splitting is positive for CuInSe2 and negative for CuGaSe2 due to the different tetragonal distortions. Introduction of spin-orbit interaction finally leads to three topmost valence bands at the point. In Figure 4.5 the optical transitions are differentiated by their allowed symmetry. Since for cf > 0 the smallest gap transition 6 → 6 is allowed only for E ⊥ c (crystal axis c), care must be taken in determining the absorption edge from optical measurements [212].


4.2.4.1 Ternary Semiconductors The optical constants n, k of CP thin films and crystals have been determined by spectrophotometry (transmission/reflection) [34] and by spectroscopic ellipsometry [431, 432]. Kreuter et al. determined the anisotropic optical constants for slightly In-rich CuInSe2 [433]. Data of thin films from Paulson et al. [432] obtained by spectroscopic ellipsometry are in good agreement with data from Orgassa [34] obtained from spectrophotometry. Data from single crystals or polycrystalline bulk material as measured by Alonso et al. [431] give generally larger absorption coefficients. The reason may be the lower density of the thin films as compared to bulk material or, in other words, the presence of nano-voids in the films. The advantage of spectrophotometric data is that they are sensitive to sub-bandgap absorption. This is why we selected the data of Orgassa for Cu(In, Ga)Se2 in Figure 4.20. The n and k values of CuInSe2 as a function of energy from Ref. [34] are given in Figure 4.20b. Spectroscopic features of these curves have been analyzed in detail based on electronic structure calculations [212, 431]. At the bandgap energy, weak structures labeled E0 (A, B, C) can be observed which originate from energy transitions at the Brillouin zone center . According to Figure 4.5, structure E0 (C)

800

1200

λ [nm] 600

400 1.2

3.3 3.2

120 Eu [meV]

1.4

3.4

1.0

E0

3.1

A,B

(d)

800

λ [nm] 600

0.6

C

3.0

0.4

2.9

0.2

(a) 2.8

0.0

1200

40 0 0.0 0.5 1.0 x in Culn1-xGaxSe2

k

n

0.8

80

400

α [cm−1]

105 104

CulnSe2 CulnS2 CuGaSe2

103

x = 0.23, 0.51, 0.78

102 1.0 (b)

1.5

2.0

2.5

3.0

hν [eV]

Figure 4.20 Optical data for CuIn1−x Gax Se2 from spectrophotometry [34] and CuInS2 from ellipsometry [431]. (a) Optical constants n, k of CuInSe2 . (b) Absorption coefficients of CuInSe2 (), CuInS2 (), and CuGaSe2 (). Simulated curves after

3.5 (c)

1.0

1.5

2.0

2.5

3.0

3.5

hν [eV]

Eq. (4.2) using the semi-empirical parameters of Table 4.5. (c) Absorption coefficients of CuIn1−x Gax Se2 thin films with x = 0.23, 0.51, 0.78 (direction of arrow). (d) Urbach energy according to Eq. (4.12) in the system CuIn1−x Gax Se2 .

217

218


is the transition from the spin-orbit split-off band. The structures E0 (A) and E0 (B) cannot be resolved for CuInSe2 . At the fundamental bandgap, the absorption curves of CuInSe2 and CuInS2 rapidly increase by several orders of magnitude (Figure 4.20c). They can well be fitted by Eq. (4.2) up to about 3 eV photon energy with fit results given in Table 4.5. The bandgaps of thin films deviate from those of bulk samples by up to 0.05 eV. From the α0 values we calculate absorption lengths of Lα0 = 1.02 × 10−4 cm (CuInSe2 ), 0.96 × 10−4 cm (CuInS2 ), and 1.19 × 10−4 cm (CuGaSe2 ). Thus, all chalcogenide absorber films discussed in this work exhibit about the same absorption strength. Radiative lifetime values calculated after Eq. (2.60) are in the range of 10−6 s. Table 4.11 gives a compilation of room temperature bandgaps for phases in the ternary phase diagrams. The shape of the sub-bandgap tail may obey an exponential law of the Urbach type

hv − E2 α(hv) = α2 exp EU

(4.12)

where E2 is an energy close to the bandgap and EU is the Urbach energy. The Urbach energy is a measure of disorder in the material or an effect of high doping concentration (NA > 1018 cm−3 ). From the reasonable fit of the absorption data in Figure 4.20b we conclude that in CuInSe2 and in CuInS2 sub-bandgap absorption is small. EU for CuInSe2 and CuInS2 [434] equals kT at room temperature which suggests that the disorder is due to thermal vibration of lattice atoms. In Figure 4.20b, the near bandgap absorption of CuGaSe2 shows deviation from the fitted line. Meeder et al., however, report on Eu = 25 meV also for CuGaSe2 [435]. The size-dependent optical bandgaps of I-III-VI2 quantum dots can be found in Ref. [436]. 4.2.4.2 Multinary Semiconductors In Figure 4.10, we highlighted possible miscibility relations between different Cuand Ag-chalcopyrites. Most important and best investigated is the CuIn1−x Gax Se2 system [437]. Figure 4.20c gives absorption spectra for a series of CuIn1−x Gax Se2 thin films with different x (data from Orgassa [34]). With increasing x, the bandgap of the solid solution film increases. The optical bowing parameter b for this solid solution is ∼0.2 eV (see Table 4.12) which is smaller than the bandgap difference of 0.66 eV. Thus, the bandgap of any CuGax In1−x Se2 solid solution is larger than that of CuInSe2 . The absorption spectra in Figure 4.20c can well be fitted by the semi-empirical expression in Eq. (4.2). As can be seen in Table 4.5, the fit parameters of α0 are similar to the values of the pure ternaries. A deviation from the fitted curve can be found at the band edge indicating sub-bandgap absorption. The Eu data, determined by Orgassa for the series CuIn1−x Gax Se2 , are reprinted in Figure 4.20d. The Urbach energy of the solid solution films have a larger Eu than the pure ternaries. As the doping concentration of solid solution films is below 1018 cm−3 , we assume disorder as the source for sub-bandgap absorption.

4.2 AI –BIII –C2VI Absorbers Table 4.12 Optical bowing parameter b of the fundamental bandgaps of chalcopyrite solid solutions and their bandgap differences. For DII 2x (CuIn)1−x CVI 2 solid solutions, the bowing parameters are only valid for x 1.

CuIn(Sx Se1−x )2 CuGa(Sx Se1−x )2 CuGax In1−x Se2 CuGax In1−x S2 Agx Cu1−x InTe2 Agx Cu1−x InSe2 AgIn(Sex Te1−x )2 CuIn(Sex Te1−x )2 CuGa(Sex Te1−x )2 AgGa(Sex Te1−x )2 CuAlx In1−x Se2 Zn2x (CuIn)1−x S2 Zn2x (CuIn)1−x Se2 Zn2x (CuIn)1−x Te2 Cd2x (CuIn)1−x Se2

b (eV)

Eg A -Eg B (eV)

References

0.005 0.005 ∼0.2 0.2 0.24 0.3 0.36 0.44 0.45 0.46 0.62 1.88 1.6 0.78 0.34

0.5 0.75 0.66 0.91 0.03 0.22 0.2 0.01 0.44 0.47 1.65 2.18 1.7 1.38 0.65

[439] [439] [212, 432] [440] [441] [442] [443] [444] [441] [443] [445] [446] [446] [446] [304]

In contrast, Cohen et al. found similar Urbach energies for all x in CuIn1−x Gax Se2 using photocapacitance spectroscopy [438]. As is visible in Table 4.12, in the CuIn(Sex Te1−x )2 and Agx Cu1−x InSe2 solid solutions the bowing parameter is larger than the bandgap difference. This means that upon adding small amounts of Te or Ag into CuInSe2 the bandgap becomes smaller. In contrast, solid solutions of CuAlx In1−x Se2 and CuIn(Sx Se1−x )2 will always have a larger bandgap than CuInSe2 . Cohen reported on an increase in Urbach energy with x in the solid solution CuAlx In1−x Se2 [438]. In Section 4.2.1 we have already pointed out the possibility to form solid solutions between CPs and binary compounds. It is tempting to determine also for these solid solutions a bowing parameter according to Eq. (4.3). However due to the Cu-3d states, the electronic band structure of Cu-chalcopyrites is very different from the DII EVI band structure. Replacing small amounts of DII atoms by Cu0.5 BIII 0.5 lowers the bandgap significantly. For example, an amount of 1 mol% Cu0.5 In0.5 S in ZnS lowers the bandgap energy by nearly 1 eV [447]. However, for larger x VI in DII 2x (CuIn)1−x C2 solid solutions, phenomenological bowing parameters can be determined. Reported values are given in Table 4.12. All bowing parameters are small enough that the bandgap of any solid solution is larger than Eg of the VI CP end member. There is the possibility to grow DII 2x (CuIn)1−x C2 films in the sphalerite structure with bandgaps in the optimum range for single junction solar cells.

219

220


4.2.5 Surface Properties

In this section, we concentrate on surface composition and surface electronics of Cu-chalcopyrite thin films – two aspects which are highly relevant for the solar cell. Other surface aspects such as the influence of corrosion, oxidation, etching, and so on, have been reviewed in an earlier article [448]. Table 4.6 lists the charge neutrality levels of some AI –BIII –CVI 2 semiconductors which will be used to estimate interface band offsets. 4.2.5.1 Surface Composition In contrast to cleaved CP crystals [448], the surface composition of a polycrystalline thin film can strongly deviate from the bulk composition. This has been revealed by X-ray photoelectron spectroscopy (XPS) on as grown vacuum transferred thin films. Figure 4.21 gives the surface cation ratio, BIII /(BIII + Cu)surf for polycrystalline CuInSe2 [447–449], CuGaSe2 [452], and CuInS2 [453] thin films as a function of the bulk cation ratio as determined by EDS. The dashed lines denote the identity BIII (BIII + Cu)surf = BIII /(BIII + Cu)bulk . For all three CP films a general trend can be observed: a small BIII excess in the bulk (BIII /(BIII + Cu)bulk > 0.5) leads to a much larger BIII excess at the surface. For the Cu-poor CuInSe2 and CuInS2 with BIII /(BIII + Cu)bulk > 0.5, the surface cation ratio is stabilized at a value of BIII /(BIII + Cu)surf = 0.75. This value agrees with the cation ratio in CuIn3 CVI 5 . Thus, it was concluded that the surfaces of slightly Cu-poor CuInSe2 and CuInS2 films are covered by surface phases of CuIn3 Se5 and CuIn3 S5 , respectively [454, 455]. We note, however, that a CuIn3 S5 phase is not yet established on the Cu2 S–In2 S3 phase isopleth (see Section 4.2.1). For CuGaSe2 , Klenk found In/(In + Cu)surf = 0.83 which corresponds to a composition of CuGa5 Se8 [452]. In contrast, Contreras et al. [456] and Negami et al. [352] also assigned the surface phase of Cu-poor CuGaSe2 to CuGa3 Se5 . A different view on the Cu-depletion comes from hard XPS and band structure calculation. Mönig et al. [457] show that a very thin surface layer (0.6 nm) of Cu-poor films is completely depleted of Cu in accordance with surface reconstruction models of the metal terminated (112) surface [458, 459]. In the case of CuInS2 [455] and CuInSe2 (D. Schmid and M. Ruckh, 1994, personal communication), a surface composition of BIII /(BIII + Cu)surf = 0.75 was also found on the rear side of thin films, that is, at the Mo/CuInCVI 2 interface. Scheer et al. proposed that the segregation of a Cu-poor phase, such as CuIn3 CVI 5 , may occur on each single grain in a polycrystalline film and, accordingly, also holds for grain boundaries [455]. Separation of the phases CuInSe2 and CuIn3 Se5 is a result of the phase diagram in Figure 4.10a. Upon cooling, a Cu-poor film can enter the two-phase field α + β. The phase separation obviously leads to surface segregation of the β-phase. It may be further supported by surface band bending (see below). Next, we consider the off-stoichiometric films with BIII /(BIII + Cu)bulk < 0.5, that is, Cu-rich films, in Figure 4.21. For CuInSe2 and CuGaSe2 the surface is covered by


0.8

CulnSe2

Culn3Se5

CuGaSe2

CuGa5Se8 CuGa3Se5

0.6 0.4

BIII / (BIII +Cu)surf

0.2 0.8 0.6 0.4 0.2 Culn5S8 Culn3S5

0.8 0.6

CulnS2

0.4 0.2 0.3

0.4

0.5

0.6

B /(B +Cu)bulk III

III

Figure 4.21 Surface versus bulk cation ratio BIII /(BIII + Cu) for CuInSe2 [450], CuGaSe2 [452], and CuInS2 [453] thin films. The surface cation ratios were determined by X-ray photoelectron spectroscopy (XPS) and the

bulk values by energy dispersive X-ray fluorescence (EDS). Dashed lines mark the identity BIII /(BIII + Cu)surf = BIII /(BIII + Cu)bulk . Right axes give the cation ratios of secondary phases.

a Cu2−x Se phase whenever the bulk composition falls below BIII /(BIII + Cu)bulk = 0.5. The occurrence of Cu2−x Se can experimentally be discerned by its Cu Auger parameter of α = 1850.4 eV [450] being different from α = 1849.4 eV for CuInSe2 [460]. The situation is more complicated for CuInS2 . Using XPS measurements, the occurrence of CuS was deduced from the Cu(LVV) Auger line shape. However, we see in Figure 4.21 that the Cu-rich surface composition follows the bulk composition. This contradiction is explained by island formation of CuS on top of a Cu-poor layer [453]. Island formation of CuS on CuInS2 has later been confirmed by atomic force microscopy images [461]. For CuInS2 , it was shown that the CuS phase only covers the front surface of a Cu-rich CuInS2 film and not the back surface [455]. The CuS phase at the front surface can be completely removed by a short KCN etch. In contrast, in CuGaSe2 the phase Cu2−x Se has been found as inclusions in the bulk [462]. Klenk et al. showed that a surface Cu-chalcogenide phase during film growth increases the CP grain size [288]. As Cu-chalcogenides are mostly degenerate p-type semiconductors they need to be removed from the surface for good cell performance. This can be accomplished by: (i) a final BIII -rich deposition, (ii) by cyanide etching, or (iii) to

221


1.2

CulnSe2

Eg 0.8 0.4 0.0 1.5 Epz=0 [eV]

222

Eg CuGaSe2

1.0 0.5 0.0 1.2

Eg CulnS2

0.8 0.4 0.0

0.40

0.50

0.60

BIII/(BIII + Cu)bulk Figure 4.22 Equilibrium reduced chemical potential of holes (equivalent to EF − Ev ) as a function of the bulk cation ratio of CuInSe2 [463], CuGaSe2 [463], and CuInS2 [465] polycrystalline thin films. Epz=0 measured by ultraviolet photoelectron spectroscopy (UPS) on vacuum transferred polycrystalline thin films.

a small extent also by an alkaline solution as provided by the wet-chemical buffer deposition. 4.2.5.2 Surface Electronics The surface Fermi level position of polycrystalline thin films was the subject of many investigations. As for the surface composition, distinct differences with respect to the bulk Fermi level have been discovered. Figure 4.22 gives the surface Fermi level with reference to the valence band, that is, Epz=0 , for CuInSe2 [463], CuInS2 [464], and CuGaSe2 [463] as a function of BIII /(BIII + Cu)bulk . All data are from polycrystalline thin films which are p-type with NA,a > 1014 cm−3 due to intrinsic defects. Thus initially, we expect the Fermi level to be close to the valence band, that is, 0 < Epz=0 < 0.3 eV.


For Cu-rich films, we observe a value of Epz=0 ∼ 0 eV for all CP films investigated in Figure 4.22. This is a result of degenerate p-type conductivity of the segregated Cu-chalcogenides [466]. In the case of partial coverage of the surface with Cu–CVI as for CuInS2 , photoelectron yield from the Cu–CVI fixes the leading edge of a photoelectron spectrum at the Fermi energy. For Cu-poor films, the surface electronic is not so simple. First, the Fermi level of Cu-poor films of CuInSe2 is Epz=0 = 1.1 eV. Hence, Cu-poor films of CuInSe2 exhibit a type inversion at the interface. This type inversion was also inferred from spectroscopic scanning tunneling microscopy studies [467]. Second, the position of the Fermi level of Epz=0 = 1.1 eV exceeds the bandgap of CuInSe2 (Eg = 1.02 eV). In view of the surface cation ratio in Figure 4.21, it can be concluded that surface analysis of Cu-poor CuInSe2 films probes the surface of a CuIn3 Se5 overlayer. Indeed, the bandgap of the CuIn3 Se5 phase is 1.23 eV [352]. As grown, CuIn3 Se5 films are slightly n-type with very high resistivity [352]. Thus, the reason for the surface Fermi level position is not a space charge in the CuIn3 Se5 overlayer but rather a surface charge. Charged surface states may induce FLP and cause band bending within the absorber film. Schmid et al. showed that the surface Fermi level of Cu(In, Ga)Se2 solid solutions strongly depends on the Ga content [463]. For small amounts of Ga in the film, the surface Fermi level of Cu-poor films drops toward EF − EVBM = 0.8 eV. This value then is roughly constant for increasing Ga/(Ga + In) ratio. Indeed, Figure 4.22 shows that the Fermi level of Cu-poor CuGaSe2 films is at Epz=0 = 0.8 eV while the bandgap of CuGa3 Se5 is 1.8 eV [352]. Thus, the Fermi level is at midgap of the surface phase CuGa3 Se5 . For Cu-poor CuInS2 , the Fermi level is at Epz=0 = 1.3 eV displaying a type inversion at the surface. As mentioned above, the surface composition of Cu-poor films also point toward a CuIn3 S5 surface phase. Electronic surface type inversion and Cu-depletion is found for several Cu-poor CP thin films. What is the cause and what is the effect? Due to the low carrier concentration in the AI BIII 3 CVI 5 phases a pn junction between n–AI B3 III C2 VI and p–AI BIII C2 VI can be ruled out. Thus, the type inversion is not the effect of the Cu-depletion. A second possibility is that both phenomena are independent. However, adsorption experiments on clean surfaces of CP crystals indicated that Cu-depletion is the effect of the surface Fermi level position [468]. Klein et al. showed that upon increasing Epz=0 above ∼0.8 eV, the surface becomes partially VI depleted from Cu [469]. Thus, it may be that the thin film CuBIII 3 C5 at the surface of Cu-poor films is the effect of both FLP and thermodynamic phase separation. In the end, the CuBIII 3 CVI 5 segregation is stabilized by the ordering energy of 2VCu –BIII Cu defect complexes in the CuBIII 3 CVI 5 structure. For contact formation, the zero charge transfer barrier height (charge neutrality level or branch point energy) is important. Mönch has determined values with respect to the valence band maximum for CuGaSe2 (0.93 eV) and CuInSe2 (0.75 eV) which allow reasonable prediction of the barrier height of some CP Schottky barriers [228]. Thus, similar as for more simple estimations [470], the zero charge transfer barrier height of CuInSe2 was found to lie in the upper half of the bandgap.

223

224


4.2.6 Properties of Grain Boundaries

Also CP thin films contain various kinds of grain boundaries which can be classified by the coincidence site lattice notation (see explanation in Section 4.1.6). Experimentally observed CP GBs fall into two categories: near 3-type twin boundaries and random boundaries [471]. The lower image in Figure 4.23 shows an example of a CuInS2 layer in cross section where the image represents an electron backscatter diffraction (EBSD) map [472]. In such a map, 3 grain boundaries and random boundaries can be differentiated. Random grain boundaries (indicated by b and c in Figure 4.23) give the impression that the film consists of small crystallites of irregular shape. Within these so-called crystallites, 3-type GBs can be found. The 3-type GBs are indicated by a, d and e in Figure 4.23. The cathodoluminescence (CL) map in the upper image in Figure 4.23A indicates that many random GBs exhibit a reduced CL intensity, while this is not the case for the majority of 3-type GBs. The CL intensity arises from a radiative recombination process, which is in competition with non-radiative recombination. The reduced CL intensity at the random GBs points out the increased non-radiative recombination. We come back to this point below. Device simulation has shown that in CuInSe2 and Cu(In1−x Gax )Se2 with small x ≤ 0.3 GBs must be electrically benign [115, 118, 473]. The effective GB recombination velocity cannot exceed 103 cm s−1 in order to model high efficiency devices (η > 19%). Since the density of near 3-type and random GBs is in the same order of magnitude, this must be valid for both types of GBs – although perhaps for different reasons. A small recombination velocity can be obtained either by a small GB defect density, by a passivation of GBs, via a special band diagram, or by a combination of the former. The atomic structure of GBs in CP films so far has not been resolved. Constructing an analogy to CdTe [237], Yan et al. calculated the electronic structure of double-positioning twin boundaries in CuInSe2 . Due to large atomic relaxation, these twin boundaries do not create deep levels in the bandgap and therefore b

CL

c

e

d

a

EBSD

b c

a

d

e

Figure 4.23 Grain boundaries in CuInS2 absorber as depicted by cathodoluminescence (CL, upper image) and electron back scatter diffraction (EBSD, lower image). The cross section was prepared with a polished surface as described in Ref. [472]. 3 near GBs are indicated by a, d, and e. EBSD sample temperature 300 K. Cathodoluminescence acquisition conditions are 5 kV acceleration voltage, 6 K sample temperature and 820 nm CL detection wavelength. While random grain boundaries clearly show up as dark lines in cathodoluminescence, many 3 near GBs do not show cathodoluminescence contrast.


exhibit a small defect density. It was argued that this finding can be generalized to many GBs in CuInSe2 [474]. A special band diagram in the form of a valence band offset (cf. Figure 3.14c) was predicted by Persson and Zunger [475, 476]. The cation-terminated reflectional 3 twin boundary of CuInSe2 [and Cu(In1−x Gax )Se2 with small x] exhibits Cu vacancies and a therefore reduced density of states in the VBM. The reduced density of states in the VBM can be interpreted as a lowered valence band at the GB or, in other words, as a valence band offset between the grain and its GB. Cu depletion at CP surfaces is a common phenomenon (see Section 4.2.5). The analogous Cu depletion at GBs [475] has been suggested earlier [455]. Using device simulation Gloeckler et al. showed that indeed a valence band offset similar to Figure 3.14c can lead to GB passivation [115] (see also Section 3.7) provided the offset becomes Ev ≈ 0.3 eV. Experimental support for the Ev theory comes from Auger spectroscopy of cross sectional grain boundaries which shows a reduced Cu/(In + Ga) concentration at the vertical surface of grain boundaries. However, transmission electron microscopy analysis of near 3 type and random GBs so far has failed to display the Cu depletion at GBs [477, 478]. Other observations regarding GBs in CP thin films come from luminescence and Kelvin probe force microscopy (KPFM) maps. As shown in Figure 4.23, GBs can show up dark (reduced luminescence intensity) in CL [279, 280] or in scanning tunneling luminescence (STL) maps [479]. The reduced luminescence intensity can result from increased non-radiative recombination at GBs but also from a larger bandgap at the GB. Indeed Romero et al. showed that in the case of CuInSe2 if both electrons and holes are generated in the electron impact mode of the scanning tunneling microscope, STL shows no GB contrast. To the contrary if only electrons are injected in the tunneling mode, GBs show up as dark lines similar as for CuInS2 in Figure 4.23. A possible explanation is that the dark luminescence contrast at GBs is due to a hole barrier being the result of a bandgap widening. Also in CL, a bandgap widening would lead to a dark contrast of the GB due to the reduced electron beam-induced carrier generation. The CL intensity from GBs was found not to shift in energy upon increased pump power [396]. Furthermore, there seems to be a smaller GB CL contrast for [110/201] oriented films with Na doping [480]. KPFM shows bright contrast at GBs which can be interpreted either as a reduced work function at the GB site or as a depletion type GB (Figure 3.14b). The normal experimental configuration of KPFM is in top view. The variation in work function in CuInSe2 ranges from 150 meV [481] to 500 meV [482]. This can be due to a positive charge in the GB [483]. The GB contrast in KPFM may depend on the film texture [480] and the Na content [474]. There are indications that the typical Na concentration of 0.1 at% in a Cu(In, Ga)(Se, S)2 film is mostly accommodated at the grain boundaries (see Section 5.2.4). Device simulation [115] showed that a GB depletion layer can support carrier collection at grain boundaries [484]. However, this depletion layer deteriorates Voc unless there is in addition a valence band offset at the GB [115, 485].

225

226


4.3 Buffer Layers

The term buffer layer here refers only to the layer which forms the contact to the absorber layer. Any other layer which transmits light to the cell, is considered as part of the window layer. Table 4.13 gives the material properties of the most important buffer layers for chalcogenide solar cells. The properties of ZnO as a buffer layer can be taken from Table 4.14. Mixed buffer layers such as Zn(O, OH, S) can be considered as a phase mixture. For a comprehensive discussion of preparation and function of the buffer layer in a particular chalcogenide cell, see Sections 5.1.3 and 5.2.8 in Chapter 5.

4.4 Window Layers

The term window layer here refers to all transparent layers which transmit light into the buffer and absorber of the solar cell. The functions of the window layer are threefold: (i) to form a transparent low resistant contact, (ii) to form a high resistant element to screen shunts [61], and (iii) to form a diffusion barrier against chemical elements. The latter function mostly refers to the superstrate configuration. The chemical element can be a doping element from the window layer itself. Together, the three functions call for a bilayer structure of a low resistance window (LRW) and high resistance window (HRW). 4.4.1 Low Resistance Windows

The LRW layer shall combine high optical transparency in the wavelength region of the active cell elements and low sheet resistance. The optical transparency extends from the effective bandgap up to the plasma absorption frequency in the infrared. As both the sheet resistance and the optical transparency decrease with thickness, this calls for a compromise. A figure of merit has been defined [498] which is the ratio of electrical conductivity σ = (d × Rsq )−1 to the absorption coefficient in the optical range of the cell, α, 1 σ = . (4.13) [ln α Rsq (1 − R) − ln T] Table 4.14 presents fundamental data of window layers which are relevant for chalcogenide solar cells. We note that the figure of merit, σ/ρ, does not account for free carrier absorption in the infrared. Also the optical bandgap values in Table 4.14 depend on the carrier concentration due to Burstein–Moss shift. Typically, a LRW has a sheet resistance of 5–10 sq in the monolithically interconnected module (without grid) and 20–80 sq for a laboratory cell employing a grid. In general there is a trade-off between low sheet resistance and low infrared absorption: the higher the carrier concentration, the higher the infrared absorption, and the lower

Table 4.13

Material properties of selected buffer materials for chalcogenide solar cells.

Property

Symbol

CdS

Cd(OH)2

ZnS

Zn(OH)2 ZnSe

β − In2 S3

In(OH)3

α − In2 Se3

Space group Lattice constant

– ˚ a (A) ˚ b (A) ˚ c (A) αa (K−1 )

P63 mc 4.137 [147] – 6.716 4.3 × 10−6

P3m1 3.5 – 4.7 –

F43m 5.410 – – 6.71 × 10−6

P63 mc 3.19 – 4.71 –

F43m 5.669 – – 7.8 × 10−6

I41 /amd 7.62 – 32.33 11.4 × 10−6

P63/mmc 7.97 – – –

P63/mmc 4 [486] – 19.2 12 × 10−6

αc (K−1 ) ρ(g cm−3 ) Eg (eV)

2.8 × 10−6 4.82 2.42 (dir)

– – – 4.09 3.15 (dir) [487] 3.5 (dir)

– – –

– 5.26 2.72 (dir)

10.8 × 10−6 – – 4.9 – 5.48 [486] 2.2 (indir) 3.25 (dir) >4.2 [488] 1.35

dEg /dT(eV K−1 )

−4.8 × 10−4

–

−6.3 × 10−4

–

−5.6 × 10−4

–

–

mn ∗ /m0

0.15 [147]

–

0.2

–

0.137

–

–

–

mp ∗ /m0 (HH)

0.68

–

1.42

–

0.82

–

–

–

(LH) (SO) Nc (cm−3 )

0.15 0.24 1 × 1018

– – –

0.36 0.56 2 × 1018

– – –

0.15 0.24 1 × 1018

– – –

– – –

– – –

– – – –

4 × 1019 107 72 8.3

– – – –

2 × 1019 1500 [147] 355 [147] 8.9

– – – 13.5 [489]

– – – –

– – – 16.68 [490]

–

5.1

–

5.9

7 [491]

–

9.53 [490]

Thermal expansion coefficient (300 K) Density Optical band gap (300 K) Temperature dependence Eg,a Electron effective mass Hole effective mass (300 K)

2 × 1019 Nv (cm−3 ) µn (cm2 V−1 s−1 ) 390 [147] µp (cm2 V−1 s−1 ) 50 [147] ε/ε0 10.21(E||c) 8.99 (E⊥c) ε∞ /ε0 5.38 (E||c) 5.31 (E⊥c)

227

Electron mobility Hole mobility Static dielectric constant Optical dielectric constant

4.4 Window Layers

Effective density of states

−7.5 × 10−4

Table 4.14

Material properties of selected window materials for chalcogenide solar cells. In2 O3

SnO2

Cd2 SnO4

ZnO

Zn2 SnO4

Space group Lattice parameters

– ˚ a (A) ˚ b (A) ˚ c (A)

Thermal expansion coefficient (300 K)

αa (K−1 )

Ia3 10.12 10.12 10.12 6.7 × 10−6

P42 /mnm 4.74 – 3.19 3.7 × 10−6

Fd3m [492] 9.143 [492] – – 3.35 × 10−6

P63 mc 3.250 – 5.207 4.31 × 10−6

Fd3m 8.657 – – –

αc (K−1 ) ρ(g cm−3 ) Eg (eV)

– 7.12 2.7 (indir) 3.7 (dir)

4 × 10−6 7 3.6 (dir)

– – 3 [493]

2.49 × 10−6 5.68 3.4 (dir)

– – 3.35 [494]

mn ∗ /m0 Nc (cm−3 )

0.3 [495] 4 × 1018

0.23 (||c) 0.3 (⊥c) 2.7 − 4 × 1018

–

0.23 2.7 × 1018

0.16–0.26 [494] 1.6 − 3.3 × 1018

ε/ε0 ε∞ /ε0

∼9 4 [496]

9.6 (E||c) 13.5 (E⊥c) 4.17 (E||c) 3.78 (E⊥c)

– –

8.7 (E||c) 7.8 (E⊥c) 3.7 (E||c) 3.7 (E⊥c)

– –

– ρ( cm) µn (cm2 V−1 s−1 ) (cm−3 ) −1

Sn 2.0 × 10−4 [497] 32 [497] 9.8 × 1020 4.1 [498]b

F 4.0 × 10−4 [497] 38 [497] 2.4 × 1021 3.1 [498]b

Al 8 × 10−4 19.2 4 × 1020 5.1b

– – – – –

ρ( cm)

10−2 [499]

10−2 [497] 1 [500]

Density Optical band gap at 300 K Electron effective mass Effective density of states (calculateda) Static dielectric constant Optical dielectric constant Low resistance layers Dopant Resistivity Mobility N σ/α High resistance layers Resistivity

a For

parabolic conduction band from Nc = 2

b Film c Film

2πmn∗ kT h2

3/2

3/2 ; = 2.5 × 1019 m∗n /m0 cm−3 .

grown by chemical vapor deposition. grown by sputtering, transmission of 98% evaluated at 650 nm.

– 2.0 × 10−4 [497] 29 [497] 1.2 × 1021 4.9 [493]c –

> 104 [501, 502]

10−2 [494]


Symbol

228

Property

4.4 Window Layers

the transparency. A way out could be a higher mobility which would allow to use lower doped films with less free carriers. However, following Ellmer, there appears to be a mobility limit for highly doped LRW oxides (n > 5 × 1020 cm−3 ) in the range of around 40 cm2 (V−1 s−1 ) [503]. This is why a resistivity of around 1 × 10−4 cm appears to be a practical lower limit [504]. SnO2 is the dominant tin bearing mineral occurring in sedimentary deposits [505]. Thus, SnO2 is highly stable in natural environments. SnO2 : F on glass is commercially produced by chemical vapor deposition – a fast and cost effective method. Typically, the low resistance SnO2 : F is deposited on a SiO2 barrier to prevent diffusion of sodium from the glass. SnO2 : F exhibits about the highest temperature, mechanical and chemical stability of all LRWs. However, its σ/α value is the lowest in Table 4.14. SnO2 has a direct bandgap of about 3.5 eV at room temperature. In2 O3 has been doped with Sn, F, and Ge [506]. The established method for producing In2 O3 : Sn(ITO) is DC sputtering in an Ar : O2 mixture. Targets are from In2 O3 (90%) and SnO2 (10%), a typical substrate temperature is 250 ◦ C. In2 O3 : Sn films are used in large amounts in the flat-panel industry. Sputtering, although being more costly than chemical vapor deposition, is compatible with the production of superstrate and substrate type of cells due to the relatively low substrate temperature. We see in Table 4.14 that In2 O3 : Sn films have a very low resistivity and a higher σ/α value than SnO2 : F. For sputtered In2 O3 : Sn films, σ/α well exceeds 5. In2 O3 : Sn appears to have an indirect bandgap at around 2.4 eV [496]. Cadmium stannate, Cd2 SnO4 , has long been investigated [507, 508] but came into major focus due to its application in the champion CdTe device of 2001 [8]. Cd2 SnO4 films are grown by co-sputtering from SnO2 and CdO targets or by sputtering from a hot-pressed SnO2 : CdO target. CdO and CdSnO3 are possible secondary phases [508]. Room temperature sputtering in Ar atmosphere results in amorphous films which need to be annealed at 500–600 ◦ C in order to reach a high σ/α value [493]. The optimum Cd/Sn ratio is slightly larger than 2 [493]. Due to an indirect bandgap at around 2.2 eV [509, 510], the film transparency in the 300–500 nm range is not as high as for other window layers [493]. ZnO films are doped with Al, Ga, In, and B in the case of sputtering and F in the case of chemical vapor deposition [511]. Sputtering of low resistance ZnO can take place at a low substrate temperature such as 100 ◦ C. Often, the substrate is not heated but the steady state substrate temperature reaches 100 ◦ C. Higher substrate temperatures lead to better values of σ/α. Doped films can be deposited by radio frequency, midfrequency, and DC sputtering. Reactive sputtering from metallic targets still suffers from poor reproducibility. ZnO is relatively stable in hydrogen plasma (only relevant for a-Si solar cells) but is easily etched in acid solution and its lateral conductivity degrades in moisture. The degradation is much stronger on rough substrates than on flat substrates [512].

229

230


4.4.2 High Resistance Windows

The HRW layer can be placed between the LRW and the buffer layer. In some works, the HRW is referred to as a second buffer layer. All type of films in Table 4.14, including Zn2 SnO4 , can be used as HRW layers. According to Ref. [61] the optimum resistivity in order for the HRW to act as a high resistant shunt screening element is 105 cm for a film of 100 nm thickness. We see in Table 4.14 that only the sputtered thin films of ZnO provides such high resistivity. In2 O3 , SnO2 , and Zn2 SnO4 have much lower resistivity if prepared as polycrystalline thin films.

4.5 Interfaces

The line-up of energy bands is of paramount importance for heterostructure solar cells. An unfavorable band line-up can be responsible for interface recombination (Section 2.4.5.4) and energy barriers (Sections 2.4.7 and 2.5.6). In this section, we discuss semiconductor/semiconductor interfaces such as absorber/buffer and buffer/window, in other words the front contact interfaces. Back contact issues are presented in Section 5.1.8 for CdTe and Section 5.2.2 for AI –BIII –CVI 2 based cells. The band line-up can be measured or can be calculated from band structure calculation theory. A third possibility is the calculation according to linear models. Experimentally, band offsets are determined by measuring the valence band offset Ev using photoemission spectroscopy. The most common method is external photoemission. After determination of Ev , Eq. (2.17) is applied to calculate the conduction band offset Ec . Conduction band discontinuities can directly be measured by internal photoemission yield spectroscopy [513]. The band line-up between semiconductors A and B can be calculated by first principles, self-consistent electronic structure theory. In the past this has been accomplished for many semiconductor contacts including some chalcogenide heterojunctions. A compilation of theoretical data is given in Table 8.2. For a long time there also existed simple linear models to predict heterojunction band offsets from intrinsic semiconductor properties. The interface-induced gap states (IFIGSs) model [534] has already been applied to chalcogenide semiconductors [228]. We make use of this model in order to discuss possible influence parameters for the line-up of energy bands at heterojunctions. The IFIGS model predicts the alignment of semiconductor specific energy levels across the interface. These levels are called branch point energies or charge-neutrality levels. Without additional dipoles, the model is linear, transitive, and independent of interface orientation. Hence, in the first order we have EV = Eb A − Eb B = Eb

(4.14)

4.5 Interfaces

where Eb A and Eb B are the branch point energies given with respect to the valence band maxima of semiconductors A and B. (Using our sign convention for Ev , see Section 2.3.1, semiconductor A should have the larger bandgap.) The branch point energies or charge neutrality levels used here are taken from Mönch [228] and are given in Table 4.6. Equation (4.14) has to be modified if there are sources of dipoles: 1) A difference in electronegativity, XA − XB , between semiconductors A and B modifies the band offset by Sx (XA − XB ) where Sx depends on the density of states around the charge neutrality level. For compound semiconductors XA,B is given by the geometric mean of the atomic electronegativities constituting the semiconductor. 2) Electrical double layers with charge Qid separated by di can contribute to a potential step at the interface by (q/εi ε0 ) Qid di where q, εi , and ε0 have the usual meaning. 3) Equation (4.14) also has to be modified if one or two of the heterojunction partners are strained. Thus the full equation for EV reads EV = Eb A − Eb B + Sx (XA − XB ) + (q/εi ε0 )Qid di + Estrain v

(4.15)

As explained by Mönch, the influence of different electronegativities for heterojunction band offsets is small. By use of Eq. (5.65) in Ref. [228], we can estimate the coefficient Sx for CuInSe2 for semiconductor junctions to be Sx = 0.3. In Table 8.2 we have printed calculated values of Sx (XA − XB ) for CuInSe2 heterojunctions. We see that indeed these values are small and can practically be neglected. Electric double layers at an interface can be the outcome of a particular interface geometry or interface orientation. First principles calculations have been accomplished in order to investigate the influence of interface orientation, intermixing, and abruptness for isovalent systems, for example, II–VI/II–VI heterojunctions, and heterovalent systems, for example, II–VI/III–V heterojunctions. At heterovalent interfaces, intermixing was predicted to determine the band line-up. Band offsets at isovalent interfaces are in first order independent of orientation, intermixing, and interface abruptness [535]. The latter we also expect for I–III–VI2 /II–III interfaces since, on average, a rigid polar interface has no interface charge [534]. Table 8.2 contains data for the valence band offset between absorber and contact layers for crystalline and polycrystalline substrates. For the crystalline substrates, cleavage was used for substrate preparation. Hence the orientation of the substrates was [011]clp in case of CuInSe2 /CdS and CuInS2 /CdS. For the thin film measurements, however, the orientation is not defined. We can only speculate that, due to the high stability of the Cu-chalcopyrite (112)clp surface, mainly (112)clp /(111)cub interfaces in the case of ZnSe and ZnS, and (112)clp /(0001)hex interfaces in the case of CdS may have formed. The first-principles calculations used (112)clp oriented interfaces [536]. Since we expect similar band offsets for different interface orientations, data for single crystalline and polycrystalline chalcogenide

231

232


substrates and their overlayers should be rather similar. This is indeed the case for CuInSe2 but not for CuInS2 /CdS, where we find a difference of about 0.6 eV. This difference cannot be due to strain effects although the latter influence is large. Interface strain can influence the band line-up between semiconductors [537]. Interface strain is expected for lattice-mismatched systems but may also be derived from non-stoichiometry in the interface lattice plane of semiconductor A and/or B. It was theoretically found that the strain effect on the band line-up mainly is due to bulk band structure effects of the semiconductors [537]. Strain due to lattice mismatch is typically accommodated within the overlayer and not within the substrate. Typical maximum thickness of a strained overlayer due to pseudomorphic ˚ Above this overlayer thickness, strain may be relaxed by growth is about 50 A. dislocation formation. Since 50 A˚ is about the information depth of photoemission experiments and, therefore, experimental Ev values are determined for thin overlayers, the experimental data may be influenced by strain – a strain which is not present in the actual device. Some of the II–VI overlayers may be strained. In Table 8.2, we give the theoretical mismatch a0 /a0 for the heterostructures under consideration. We estimated the strain effect on EV by considering only the hydrostatic component e11 = (a0 /a0 ) − 1. The influence of valence band splitting and shear components is neglected. Thus the estimated strain effect is a lower limit. Furthermore, we consider only the strain effect on the valence band maximum of the overlayer and neglect the influence on the charge neutrality level. The value of the substrate valence band maximum is unaffected. Then the change of the band line-up is Ev strain = −av Tr(e)

(4.16)

where av is the band-edge deformation potential and Tr(e) = 3 × e11 [538]. Values of the deformation potentials for ZnS (av = 2.31) and ZnSe (av = 1.65) can be found in Ref. [538]. For CdS, we assume av = 2 eV. Wei and Zunger have calculated the strain effect using first principles calculations [514] of the systems CuInSe2 /CdS and CuInSe2 /ZnSe. The estimated and calculated data as given in Table 8.2 are in good agreement. We find that the influence of strain on some of the heterostructures can be considerable, especially for CuInSe2 /ZnS, CuInS2 /ZnS, and CuGaSe2 /ZnSe. There, pseudomorphic growth of the II–VI semiconductor would lead to a noticeable decrease of EV . In Figure 4.24, we compare experimental, theoretical, and modeled data for Ev . In Figure 4.24a, we discriminate between data obtained by ab initio calculation of a particular heterojunction and data derived from the transitivity rule. For CP substrates, we select those experiments where the substrate surface composition is known. This is mostly the case for CP crystal substrates but to a minor extent for polycrystalline films. The reason is that many as-grown CP film surfaces are non-stoichiometric which may influence the band line-up. The agreement between experimental and theoretical data in Figure 4.24a is fair. The largest deviation of −0.27 eV is found for the system CuIn3 Se5 /ZnS. The lattice mismatch of CuIn3 Se5 /ZnS is large. It is not clear, however, if the deviation in Ev,exp is due to strain in the ZnS overlayer. Also, the theoretical and modeled data in Figure 4.24c

4.5 Interfaces

CulnSe2

CulnS2

CuGaSe2

CdTe

2.0

233

Culn3Se5

2.0 ZnO

CdS

1.0

CdS

CdS

CdS

ZnSe CdS

0.5

CdS

ZnS CdS

ZnTe

CdS

ZnSe

ZnTe

∆Ev,theor [eV]

∆E v,exp

1.5

CdS

CdSZnS ZnS CdS ZnSe 1.0 CulnS2 CdS CdSe ZnSe ZnS 0.5 ZnSe

1.5

CdS

CdTe

0.0

0.0

ZnTe

CulnTe2 ZnSe

0.0 0.5 1.0 1.5 2.0 (a)

∆Ev,theor [eV ]

0.0 0.5 1.0 1.5 2.0 (b)

Figure 4.24 Valence band offsets for some chalcogenide heterojunctions. (a) Experimental values of Ev versus theoretical values obtained by ab initio calculations (solid symbols) and application of transitivity rule (open symbols). The dashed line marks the identity and provides a visual guide. Values for overlayers on crystalline substrates are

∆EB [eV]

0.0 0.5 1.0 1.5 2.0 (c)

∆EB [eV]

labeled as X-Sub. (b) Experimental values of Ev as a function of modeled differences in the branch point energy after Eq. (4.14). (c) Theoretical values of Ev from ab initio calculation (solid symbols) and from transitivity rule (open symbols) as a function of differences in branch point energy after Eq. (4.14).

Table 4.15 Experimental Fermi level positions and their method of detection for different chalcogenide heterostructures. FLP marks cases where Fermi level pinning has experimentally been found.

Layer 1

Layer 2

Cu(In0.76 Ga0.24 )3 Se5 a CdTe

CdS

Band gap at side of layer 1 (eV)

Ep,az=0 at side of layer 1 (eV)

Method

1.33

1.24 (FLP)

CdS

Non-activated 1.49

1.02 (FLP)

CdTeb

CdS

Activated

Graded

No FLP

CdS

SnO2

2.4

2 eV (FLP)

Admittance spectroscopy Photoelectron spectroscopy Photoelectron spectroscopy –

a The

Remark

–

–

Reference

[79] [226] [226] [533]

phase Cu(In0.76 Ga0.24 )3 Se5 is the surface phase of the Cu(In0.76 Ga0.24 )Se2 absorber semiconductor. Its band gap is about 0.18 eV larger than the absorber’s band gap of 1.15 eV [456]. Thus in direct contact to CdS is not the Cu(In0.76 Ga0.24 )Se2 absorber but its surface phase with larger band gap. b Activation of the CdS/CdTe heterostructure by CdCl2 treatment in ultrahigh vacuum. The treatment includes CdCl2 vapor exposure at 420 ◦ C for 20 min in a closed space arrangement.

Eg,a [eV]


Eg,IF

234

1.5

1.0 0.0

0.5

1.0

x in Culn1-xGaxSe2 Figure 4.25 Interface bandgap Eg,IF at the CuIn1−x Gax Se2 /CdS interface calculated after data from Ref. [539] as a function of x. Data are obtained by direct and inverse photoemission spectroscopy. Calculation

assumes for all x a CdS bandgap of 2.4 eV. Dashed line indicates the absorber bandgap calculated after Eq. (4.3) with b = 0.2 and Eg A − Eg B = 0.66 according to Table 4.12.

are in good agreement. Deviations to be discerned for the CuInS2 contacts, may be due to an non-appropriate branch point of CuInS2 . We are of the opinion that the IFIGS approach is suitable for chalcogenide interfaces. Therefore, in Figure 4.24b we plot experimental versus modeled data for Ev . Especially for the CdTe substrate, the agreement is impressive. We use the data of Ev in order to develop device models for some chalcogenide standard devices. Table 4.15 gives experimentally detected Fermi level positions and methods of detection for different chalcogenide heterostructures. Using photoelectron spectroscopy, FLP has been deduced for the CdTe/CdS interface before CdCl2 activation and for the CdS/SnO2 interface. An indication that FLP is present at the Cu(In0.76 Ga0.24 )Se2 /CdS interface is given by AS [79] but requires further confirmation. The interface bandgap in the system Cu(Inx , Ga1−x )Se2 /CdS has been determined using a combination of direct and inverse photoemission spectroscopy [539]. In Figure 4.25, it is apparent that for x < 0.5 the interface bandgap exceeds the absorber bandgap.

235

5 Thin Film Technology In this chapter we give an overview of the technological aspects of CdTe and chalcopyrite solar module fabrication. Each component or process step is described and the mandatory requirements are listed. So-called magic treatment steps or magic elements, which good cells need, are specified as far as possible; and we try to give an explanation or at least a speculation of their role. It is the aim of this chapter to provide the necessary information on how to produce highly efficient devices for both technologies.

5.1 CdTe Cells and Modules

Today, the CdTe solar cell is a heterostructure comprising several differently doped thin layers. Originally, it was a simple CdS/CdTe heterojunction [10] where CdS served as the doped window layer. Later, CdS became a thin buffer layer and the function of the transparent front contact was taken over mainly by SnO2 : F, In2 O3 : Sn and other transparent conducting oxides. Due to reasons which are discussed below, the CdTe solar cell most often is of the superstrate type (Figure 1.6). This is why this section is ordered according to the superstrate device fabrication steps. 5.1.1 Substrates

Requirements on the substrate are: (i) low cost, (ii) high stability (thermal, mechanical, chemical), (iii) good isolation (electrical, chemical), (iv) small thermal expansion mismatch to thin films, (v) low weight, and, perhaps, (vi) flexibility and (vii) transparency. No substrate fulfills all requirements. For the superstrate configuration transparency is important. Remarkably, glasses of very different thermal expansion coefficients were used for superstrate CdTe cells such as soda lime glass [540], borosilicate glass [541], and aluminosilicate glass (see Table 5.1). FeOx , being an additive to the standard soda lime glass, induces optical absorption which accounts for ∼1 mA cm−2 Jsc loss compared to borosilicate glass [541]. The loss can be reduced by use of low iron soda lime glass (white glass). Also the low softening Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

236

5 Thin Film Technology Table 5.1

Properties of substrates for chalcogenide solar cells.

Substrate

Root mean square roughness (nm)

Thermal expansion coefficient (300 K; 10−6 K−1 )

Softening temperature ( ◦ C)

Comment

Float glass Borosilicate glass Aluminosilicate glass Kapton E Upilex S

0.5 [543] 0.5

9.0 3.0–4.0

550 820

–

∼5.0

900

Contains ∼15% Na2 O Contains ∼4% Na2 O, ∼13% B2 O3 Contains ∼36% Al2 O3

– –

Ti sheet Steel sheet Cr steel Mo

– 20 [544] 20–40 [544] –

16 (320–570 K) 12.0–24.0 (290–670 K) 8.6–9.7 (290–770 K) 13.0 11.0 4.8–5.9 (290–870 K)

– –

Polyimide foil Polyimide foil

– – – –

– – Contains ∼16 at% Cr –

temperature of soda lime glass is a problem for large substrates (see Table 5.1), especially in bottom-up thin film deposition. The current world record CdTe solar cell has been prepared on borosilicate glass (Corning 7059) of 1 mm thickness [8]. Also, polymer substrates allow us to realize the superstrate configuration. By the use of 10 µm Upilex foils, more than 11% efficiency has been realized [542]. For a substrate-type CdTe cell, metal and polymer foils can be employed. As metal substrate, molybdenum has been used. This metal does not form a too large electric barrier [545]. The barrier can be further reduced by applying a nitrogen-doped ZnTe interfacial layer (for back contacts see Section 5.1.8). A problem of metal substrates is the high surface roughness, which depends on the production process. Adhesion of the thin films may become poor at dimples and cavities. Particles, loosely attached to the metal surface, can cause shunts if they detach during the solar cell process. Polishing and brushing of the metal foil can reduce these problems. 5.1.2 Window Layers for CdTe Cells

The commonly used superstrate configuration of the CdTe solar cell has three implications for the window layer:

1) The window layer can – in principle – be deposited at high substrate temperature, allowing for a higher quality material with a higher carrier mobility.


2) As the window is deposited on glass with typically small roughness, it is less prone to degradation in moisture than a rough low-resistance window (LRW) for chalcopyrite cells (see Section 6.2.5). 3) Finally, the window layer has to withstand the subsequent high-temperature processes. Most high efficiency CdTe solar cells have a bilayer window. The LRW and high resistance window (HRW) can employ the same oxide or different oxides. As only optical transparency up to 900 nm is required, the LRW can have a very high carrier concentration. Free carrier absorption is not a limitation. LRWs for CdTe cells are SnO2 :F, In2 O3 :Sn, Cd2 SnO4 (see Table 4.14), as well as CdIn2 O4 [546]. ZnO : Al was successfully applied only for sputtered CdTe cells [547]. Without a HRW, SnO2 : F was reported to yield improved performance, while the In2 O3 : Sn/CdS contact can exhibit a barrier for the diode current [548]. In the champion CdTe cell of 2001, the HRW Zn2 SnO4 has been used [8], together with Cd2 SnO4 as the LRW [549]. Interdiffusion between CdS and Zn2 SnO4 partly consumes the CdS layer, leading to a higher short circuit current [550]. Other HRWs used are In2 O3 and SnO2 [551]. We emphasize that neither of them has an as-grown resistivity in the range of 104 cm, which would be required to serve as a good shunt screening element [61, 86]. We note, however, that the resistivity of the HRW after CdCl2 treatment is not known. Application of a HRW mostly increases Voc , although a positive Jsc effect has also been reported [551]. Ferekides reports that the CdS thickness is crucial even in the presence of a HRW [497]. The HRW appear to increase the red light collection [497]. Finally, we mention that the cell efficiency can increase with increased window roughness (haze) [552] provided that the HRW and CdS thickness is sufficient. Otherwise, rough window layers can cause local shunting [553]. 5.1.3 Buffer Layers for CdTe Cells

Today, almost all efficient CdTe solar cells have a CdS buffer layer. According to Section 3.9, the buffer layer may increase the efficiency by reducing reflection losses but may also decrease the efficiency due to light absorption and carrier recombination. As the dielectric constants, ε/ε0 , and hence the refractive indices of CdTe, CdS, and SnO2 all are similar, the optical gain is small. What remains is the detrimental current loss at high buffer layer thickness. It is commonly believed that charge carriers generated in the CdS buffer recombine and are not collected. Possible reasons are small carrier lifetime in the CdS layer or interface recombination. The CdS layer can be deposited by all methods that are listed in Table 5.2 for CdTe deposition. Often, from conceptual aspects, the same deposition method as for the CdTe layer is used. An additional method for CdS growth is chemical bath deposition (CBD). A recipe for CdS CBD is given in Table 5.3. The finished CdS layer may be annealed in H2 prior to CdTe deposition [554]. In combination with close space sublimation (CSS) of CdTe, only chemical bath-deposited CdS [8, 12] and chemical vapor-deposited CdS [555] give the highest efficiencies. CBD CdS has

237

238

Characteristics of CdTe deposition processes which have been used to make efficient solar cells.

Ambient Pressure (Pa) Substrate temperature (K) Deposition rate (µm min−1 ) Typical film thickness (µm) Source for Cl Source for O Champion cell efficiency (%) Champion module efficiency (%)

Physical vapor deposition

Close space sublimation

Vapor transport deposition

Sputter deposition

Electro deposition

MOCVD

Spray pyrolysis

Vacuum 10−4 720

N2 , Ar, He 103 750–900

N2 , Ar, He 103 –104 900

Ar 10−2 650

Solution 106 350

Inert gas 106 500–700

Air, inert gas Air 106 106 800 300

0.01–0.5

1–5

0.1–10

0.1

0.01–0.1

0.01–0.1

1

–

4

4

3

2

2

2

10

12

– Ambient 10 [558]

CdCl2 O2 16.5 [8]

CdCl2 O2 14.1 [559]

– Ambient 14.0 [547]

Solution Solution 12.7 [560]

DMCl, tBuCl Solution Precursor S 13.3 [561] 14.7 [562]

7 [564]

11 [565]

–

–

11 [566]

–

10.5 [562]

Screen printing

Paste – 12.8 [563] 11 [566]

5 Thin Film Technology

Table 5.2

5.1 CdTe Cells and Modules Table 5.3

Constituents for chemical bath deposition of CdS.

Chemical

Molarity (M)

Function

Cadmium acetate (Cd(CH3 COO)2 ) Ammonium acetate (NH4 CH3 COO) Ammonium hydroxide (NH4 OH) Thiourea (CS(NH2 )2 )

0.5–2.0 × 10−3 8–12 × 10−3 0.2–0.5 1–4 × 10−3

Cadmium source Buffer Complexing agent Sulfur source

After Ref. [554].

smaller grains and thus provides a more homogeneous coverage of the window layer when compared to CSS CdS. Furthermore, CBD CdS has a lower absorption in the photon range above 2.4 eV and a slightly larger bandgap [556]. As it contains oxygen, the CBD CdS film is less prone to sulfur diffusion and thus can be made thinner with lower risk of pinhole formation. In order to use CSS CdS [556] a very high oxygen partial pressure in the ambient is necessary [557]. This also leads to smaller grains and thus to a better coverage of the window layer. During the CdCl2 treatment (see Section 5.1.5), diffusion of sulfur into the CdTe layer occurs [158, 567]. Thereby, the CdS layer is partly consumed preferentially at grain boundaries where sulfur diffusion is enhanced (see Section 4.1.6). If the CdS layer thickness is not sufficient, the CdS film can – at least on a local scale – be fully consumed. This leads to a local contact between absorber and window which is considered as a shunt. (In view of interface recombination as explained in Section 2.4.5.4, we may rather think of a high recombination site at the interface due to an unfavorable position of Ep,az=0 .) The minimum thickness of CdS prepared by CBD is around 100 nm, which is the as-deposited thickness [554, 568]. Sulfur diffusion can be reduced by the supply of oxygen (see Section 5.1.5) during CdTe deposition and by heat treatment of the CdS/CdTe stack prior to the CdCl2 treatment [569]. The measurable effects of a too low CdS thickness are reduced values of open circuit voltage and fill factor [497]. If we assume a crystallographic relationship between {111} planes of CdTe and {0001} planes of CdS, we calculate a lattice mismatch of 10.8% (see Table 8.2). This could result in a very high interface defect density. It was argued that the CdS/CdTe intermixing due to sulfur diffusion may reduce the defect density. Recent lifetime measurements, however, point in another direction. Metzger et al. found that CdS deposition onto a CdTe layer increases the lifetime of minority carriers in CdTe – even without S diffusion (due to missing CdCl2 treatment) [207]. Thus, we consider the CdS buffer as a passivation layer for CdTe surface states. Assuming a rigid band edge model, we calculate a conduction band offset of −0.1 to 0.0 eV from the valence band offset data in Table 8.2. Thus, the CdS buffer layer provides a favorable band lineup to CdTe in accordance with the rule outlined in Section3.2.

239

240


In order to reduce optical absorption, alloy films of Cd1−x Znx S have been applied by chemical surface deposition [568] or metal organic chemical vapor deposition (MOCVD) [561]. In fact, the highest efficient MOCVD-deposited CdTe solar cell in Table 5.2 had a Cd1−x Znx S with x = 0.1. Widening the bandgap of the buffer layer certainly is a viable goal for the future. 5.1.4 CdTe Absorber Layer

In Section 4.1.1, we listed three reasons why CdTe is rather forgiving in terms of film growth. It is the only compound in the Cd-Te phase diagram (Figure 4.3). It has a high negative formation enthalpy. The vapor pressures of Cd and Te2 are substantially higher than that of CdTe. There is a fourth reason which may be equally important: every CdTe film, for high efficiency solar cells, is annealed in the presence of CdCl2 at around 450 ◦ C. Therefore deficiencies, which lie in the simplicity of the deposition process, are partly annealed out during the CdCl2 treatment. The fact that CdTe can be annealed without decomposition represents the fourth reason. Table 5.2 shows that CdTe films, which deliver more than 12% efficiency in the solar cell, can be grown by a variety of methods. Nevertheless, devices exceeding 15% efficiency so far have only been deposited by CSS [8, 12, 555, 570]. CdTe can be evaporated from a single source where the binary source material is sublimed. The vapor consists (mainly) of Cd and Te2 species which react at the substrate to form a layer of CdTe. (For Cd1−x ZnTe or CdTe1−x Sx deposition, one can simply use a combination of two binary sources.) Within the error margin of all measurement techniques, CdTe films grown from one binary source are stoichiometric (although the higher vapor pressure of Cd proposes the formation of VCd ; Figure 5.1). For substrate temperatures above 720 K, re-evaporation from the substrate occurs. Thus, at low chamber background pressure, the net deposition rate decreases with TSub . This is the reason why physical vapor deposition (PVD) is limited in the substrate temperature. In order to avoid re-evaporation, a substantial background pressure is necessary. This implies that the source–substrate distance should be small. The requirement of a small source-substrate distance is fulfilled for both CSS and vapor transport deposition (VTD) – the two methods of highest substrate temperature in Table 5.2. The principle of CSS can be seen in Figure 5.2. The space between source and substrate is in the range 2–20 mm. Transport of species from the source to the substrate is controlled by diffusion. In order of increasing substrate temperature, the temperature difference between source and substrate is from 50 K [571] to 200 K [572]. The deposition rate can be in the micrometers per minute range. Due to the large grain size, there appears to be a minimum film thickness of about 4 µm (cf. Table 5.2). Smaller film thicknesses lead to pinholes. A background pressure of around 103 Pa is formed by inert gas (see Table 5.2) and (as an option) O2 of around 102 Pa. The source material is CdTe in the form of a granulate which is filled in open line sources. Typical purity is 5 N [573]. Alternatively, a melted CdTe


T [K] 1000 700 500 105

103

400

Cd Te2

Figure 5.1 Vapor–solid Psat versus inverse temperature T−1 diagram for CdO, CdS, TeO2 , CdTe, CdCl2 , Te2 , and Cd. After Ref. [140].

CdCl2

Psat [Pa]

CdTe 101

TeO2 CdS

10−1

CdO

10−3

10−5

0.5

1.0

1.5 1/ T

2.0 2.5×10−3

[K−1 ]

block has been used [570]. The sources are electrically heated. The long dimension of the source slightly exceeds the substrate width. Its short dimension influences the maximum substrate speed during inline production, which can be around 1 m min−1 . The CSS process can also been used for CdS and CdCl2 deposition. Due to the granulate form of the source material, the deposition configuration is bottom up. Thus, the substrates must be hanging over the source and must be transported by rollers with contact to the substrate edges. This configuration limits the maximum substrate temperature to about 750 K being well below the softening point of the float glass. Commercial CdTe producers tend to use low-cost (white) glass with low softening temperature. Highest efficiency laboratory solar cells, however, have been grown on borosilicate glass (Corning 7059) at around 900 K [8, 12]. To date, the lower substrate temperature of CSS in production poses one constrain of the commercial CdTe efficiency [574]. Nevertheless, above 10% efficient modules should be feasible even with soda lime glass. As-deposited CSS films exhibit grains sizes in the micrometer range, with no preferred orientation of the grains. The CSS process had been used to demonstrate the first CdTe cell efficiency above 10% [11]. It is commercially employed by Antec, Germany [575]. VTD is another method which can combine high substrate temperature with high deposition rate. The principle design of a VTD evaporator is shown in Figure 5.2. The source material is continuously injected in the form of CdTe powder into a cylindrical and permeable membrane. The membrane is heated and in there, the CdTe powder is evaporated. A carrier gas transports the vapor through the membrane to the volume of the heated second cylinder. The gas components can be similar as for CSS (see Table 5.2). Downward to the substrate, the second

241

242


Physical vapor deposition

Close space sublimation

Vapor transport deposition

Porous frit

CdTe

Figure 5.2 Schematics of physical vapor based deposition methods. For differences in deposition characteristics see Table 5.2.

cylinder has a distributor slit through which the carrier gas transports the vapor to the substrate by convective transport. At the substrate, the gas reacts to form CdTe. Similar to CSS, very high deposition rates can be obtained at high substrate temperature. Due to the top-down geometry, the glass can be heated up to slightly above the softening point. This, however, requires a re-strengthening of the glass by a quenching step after deposition [576]. The VTD process has been treated theoretically [577, 578] but is still a matter of research. Best cell efficiencies are above 14% [559] and commercialization on a GW scale has been established by First Solar, Inc., USA. Also spray pyrolysis at a substrate temperature of around 800 K can deliver CdTe films at a rate of 1 µm min−1 . The process takes place at atmospheric pressure and therefore is attractive due to low equipment costs. A solution containing a Cd precursor such as CdCl2 and a Te precursor, such as TeO2 or an organic Te source, are sprayed onto the heated substrate. The solvent evaporates leaving a CdTe : Cl film on the substrate. Grain sizes can be in the micrometer range with random orientation. On the cell level, an efficiency of 14.7% has been achieved by Golden Photon Inc., USA [562]. Also, modules exceeding 10% efficiency have been demonstrated [562]. An all-sputtered CdTe solar cell of 14% efficiency has recently been demonstrated. Remarkably, this cell has a ZnO:Al LRW [547]. CdTe sputtering takes place at low substrate temperature of around 650 K [579]. Typically, this method has a relatively small deposition rate of around 0.1 µm min−1 – a drawback which may be compensated by extended sputtering zones. Both electrodeposition and screen printing of CdTe had been used for commercial products by BP Solar and Matsushita Batteries, respectively. Both methods deposit a CdTe film at room temperature (RT) with subsequent annealing. Cl is added in the solution or in the printing slurry. A clear disadvantage of the screen printing method is the high film thickness of more than 10 µm (see Table 5.2). For electrodeposition and screen printing, the efficiencies on the module level clearly exceeded 10% (Table 5.2). MOCVD gave more than 13% efficient cells, however, it is not clear if this method is apt for large-scale volume production [561].


5.1.5 Activation by Chlorine Treatment

The presence of chlorine in CdTe films is a prerequisite requisite for high efficiency solar cells. Chlorine can be incorporated as a dopant during film growth (in situ) or by post-deposition treatment. Possible chlorine sources are liquids such as CdCl2 :CH3 OH or CdCl2 :H2 O. The corresponding deposition process is dipping or dripping followed by a drying step which leaves a CdCl2 film on the surface. In gaseous form, the chlorine sources can be CdCl2 , HCl, or Cl2 vapor. The chlorine treatment – whether as an in situ or a post-deposition process – takes place at elevated temperature [580]. Examples for in situ chlorine treatment are spray deposition of CdTe with intermittent CdCl2 : CH3 OH spray [581] or CdTe/CdCl2 co-sublimation [582], both on heated substrates (TSub > 400 ◦ C). Thus, the chlorine treatment takes place simultaneously to the CdTe deposition. In situ supply of chlorine has the advantage that no removal of chlorine residuals is necessary. However, in situ supply requires control of the chlorine flux. A typical example for a post-deposition treatment is CdCl2 deposition on as-grown CdTe film, for example, by CSS, followed by heating in air or vacuum and finally washing off the CdCl2 residuals. CdCl2 thickness depends on the CdTe layer thickness and ranges from tens of nanometers [246] up to 200 nm [579]. Alternatively, CdCl2 is deposited on the heated CdTe. Then, prevention of residue formation can be accomplished by maintaining a sufficient substrate temperature during or after treatment. The post-deposition method offers more flexibility. But, it requires minute control of the substrate temperature [201] during CdCl2 treatment. Moreover, the optimum treatment temperature seems to depend on the CdTe film quality and thus varies with varying deposition temperature. Typical chlorine concentrations in CdTe are 1019 cm−3 [583]. The chlorine concentration in the CdS layer can be higher [583, 584]. The structural and electronic effects of the chlorine treatment are rather similar – regardless of whether it is an in situ or post-deposition step. In the following, we want to concentrate on those effects which are relevant for the solar cell performance. Structural effects, which are induced or promoted by the chlorine treatment, are: 1)

Increase in grain size [133, 583, 585]. If the as-grown CdTe grain size is small, then the chlorine treatment induces inter-grain re-crystallization. This leads to larger grains and thus to fewer high- grain boundaries. Even for large grained films, inter-grain re-crystallization is relevant for the seed layer of deposition, that is, for the film fraction near to the substrate. In a superstrate type of device, this is the CdTe fraction next to the CdS interface [583] where the carrier concentration n equals p and the recombination rate is high (see Section 2.4.5.1). Inter-grain re-crystallization in the presence of chlorine can lead to a removal of preferred [110] CdTe orientation and a randomization of the orientational relationship to the CdS buffer layer [242, 585, 586].

243

244


2) Reduction in stacking fault density [242]. This effect is referred to as intra-grain re-crystallization [133]. It basically reduces the density of 3 grain boundaries which are very abundant in CdTe. The intra-grain re-crystallization is also more pronounced in films grown at low substrate temperature (>400 ◦ C) [242]. 3) CdTe/CdS interdiffusion. Sulfur diffusion from the CdS to the CdTe layer is enhanced by the CdCl2 treatment [158, 207]. Sulfur diffusion leads to the formation of CdTe1−x Sx by partial consumption of the CdS buffer layer (see also Section 4.1.1). Thereby, the CdS thickness is reduced or, in extreme cases, completely consumed. The diffusion process proceeds via Fickian bulk and grain–boundary diffusion and has a Arhennius-type temperature dependence [158]. Also Te diffusion into CdS takes place [133, 587]. As stated above, films grown at high substrate temperature hardly show the structural CdCl2 effects (1) and (2) [583, 588]. Nevertheless, even those films give an increased device performance upon CdCl2 treatment. There are electronic effects of the chlorine treatment which appear to be partly independent of the structural effects. 1) Increase in conductivity or p-type doping [582, 589, 590]. The conductivity effect shows up in a decreased series resistance of the device. It may be related to a modified GB potential as chlorine is mostly accommodated at the grain boundaries (see Section 4.1.6). In addition, there are indications for an increased density of deep acceptors [591] and a more homogeneous acceptor distribution [592]. A clear assignment of the deep acceptor level to a particular defect currently is not possible. Theoretical data collected in Figure 4.4 show that ClTe forms a deep donor while the defect complex VCd − ClTe is a shallow acceptor. 2) Increase in carrier lifetime. The lifetime determined by time-resolved photoluminescence typically increases from 200 ps to 2 ns [8] upon CdCl2 treatment. In accordance with such a lifetime increase, an increase in diffusion length by a factor of 3 has been observed [246]. The mechanism of lifetime increase is not clear. Attempts to find a decrease in the deep defect density were not successful [589]. Reports on a decrease in grain boundary recombination velocity are contradictory [207, 593]. 3) Basically all solar cell parameters Jsc , Voc , FF are improving under chlorine treatment (see example in Figure 5.3). Untreated cells can exhibit a poor blue response in the quantum efficiency curve indicating a very extended space charge region (see Section 7.4.3). The chlorine treatment substantially improves the current collection. This was interpreted as due to the increased acceptor density and thus increased electric field strength [591]. In principle, both the increased carrier concentration and the increased lifetime can contribute to the increase in Voc . As the CdTe solar cell appears to be limited by space charge region recombination (see Section 6.1.2), an improved structural quality [effects (1) and (2)], especially close to the CdS interface, is beneficial. 4) According to Table 4.15 and the work of Jägermann et al. [226], the CdCl2 treatment can unpin the Fermi level at the CdTe/CdS interface.


25

η [%]

0.6

TSub = 625 °C

20 15 10 5

0.4 Yes/Yes

No/Yes Yes/No Cu/CdCl2 Treatment

(a)

No/No

0 400 450 500 550 600 650 Substrate temperature [°C] (b)

Jsc [mA cm−2]

FF

0.8

Voc [V],

5) The post-deposition chlorine treatment is a critical process step with a process window of about 20 K. Treatment time depends on the CdTe thickness. An overtreatment can lead to reduced Voc [594] and finally to delamination [8]. CdCl2 can introduce impurities such as Cu, Sb, Bi, As, Al, and Si which may influence the doping of the CdTe layer [576]. The influence of oxygen during the chlorine treatment is covered in the next section.

Figure 5.3 (a) Solar cell parameters of CdTe cells as a function of the Cu and CdCl2 treatment for a CdTe film grown by close space sublimation at 625 ◦ C. (b) Cell parameters for devices with Cu/CdCl2 treatment and with CdTe absorbers grown by different groups but at different substrate temperatures. After Ref. [595].

5.1.6 Influence of Oxygen

Next to chlorine, oxygen is an important contaminant in high efficiency CdTe devices [596]. In 1982, the Eastman Kodak group was the first to report a CdTe thin film solar cell with efficiency exceeding 10%. The presence of oxygen in the CSS chamber allowed this milestone to be reached [11]. Typically all solar cell parameters are higher if oxygen is incorporated in the CdTe film. Oxygen incorporation can take place either during growth or during CdCl2 treatment. Oxygen sources are the process gas, the CdS layer, and the TCO [556]. To give a number, the oxygen partial pressure during CSS growth of CdTe shall be in the range of 133 Pa (1 Torr) [36, 596, 597]. The resultant oxygen concentration in the CdTe film is 1019 –1020 cm−3 [194, 583, 598]. The best solar cell so far which was made without intentional oxygen incorporation showed an efficiency of 12.8% under AM 1.5 [599]. However, all high efficiency cells (η > 15%) are fabricated with oxygen incorporation. To date it is not clear which effect of oxygen in CdTe thin films is responsible for the superior device performance. 1) There is an indication that oxygen reduces the eutectic temperature in the system CdTe-CdCl2 [167] (see also Section 4.1.1). Thus, participation of oxygen during the chloride treatment (see Section 5.1.5) may lead to a liquid phase recrystallization.

245

246


2)

On the one hand, oxygen addition during CdTe deposition increases the nucleation rate and therefore reduces the grain size [599]. This leads to better coverage of the CdS layer and to a reduction in pinhole density. On the other hand the smaller grain size may increase grain boundary effects on the device performance. The increased nucleation rate due to oxygen addition is also found for CdS films deposited by CSS [541]. For the CdS, it is particularly important as pinholes in the buffer layer lead to shunting. 3) Sulfur diffusion via grain boundaries is reduced [244]. Therefore, the risk is reduced that a thin CdS layer is perforated near a CdTe grain boundary and a micro contact (shunt) between CdTe and TCO is formed. 4) Oxygen present in the CdCl2 treatment was found to increase the hole concentration by 1 order of magnitude [193, 596] or to decrease the film resistivity [600]. Nollet et al. argued that the increased hole concentration may be the main effect of oxygen incorporation [193]. The space charge region recombination rate is reduced upon higher doping (see Section 2.4.5.1). 5) The charge carrier lifetime measured by time-resolved photoluminescence increases with oxygen partial pressure present during CSS growth of CdTe [36]. Certainly, also this effect can reduce the recombination rate in the space charge region. Secondary phases which may form due to oxygen exposure of the films are CdO and CdTeO3 [599]. These phases may be detected by Raman spectroscopy (see Table 4.2). During growth, O2 admixture to the ambient can substantially change the sublimation rate due to oxygen adsorption on the CdTe source material [599]. This can cause instability of process parameters. Furthermore, it was observed that, depending on the oxygen partial pressure during CSS, different surface features (particles) develop on the growing CdTe film [599]. These particles can induce shunting. 5.1.7 Influence of Copper

So far, the highest efficient CdTe solar cells used a Cu-containing back contact [541, 550]. The positive effect of Cu is that the back contact resistance is reduced [601, 602] and the fill factor increased [603]. Other effects on device performance are an increased Voc [604–606] sometimes combined with a decreased Jsc [606]. Typical cell parameters can be seen in Figure 5.3a. Besides the beneficial effects of Cu, it is reported that Cu reduces the cell stability. Under forward bias stress, the degradation of cell parameters Voc , and fill factor increases with increasing Cu content in the cell [607]. It is further observed that the crossover between dark and light JV curve increases with Cu concentration [608]. Depending on the type of back contact, there are different sources for Cu incorporation in CdTe. If a graphite paste is used, Cu is included in the HgTe : Cu powder which is added to the paste [571]. A typical concentration in the paste is 3 at% Cu [605]. Also a CuTe : HgTe powder has been used as an additive to the graphite paste [550]. Another approach is to deposit a ZnTe:Cu layer with around


6 at% Cu by sputtering [604]. For research purposes, Cu can directly be evaporated on the etched CdTe layer [608]. The appropriate Cu thickness is around 3 nm for 3 µm CdTe [547]. For each process, there is an optimum Cu concentration in CdTe which has to be adjusted empirically [609]. The problem in assigning clear effects of Cu in CdTe is that Cu enters the device – even without intentional doping. Typical sources for Cu contamination are the graphite paste itself [601] or the CdCl2 treatment [604]. Cu diffusion from the back contact proceeds via grain boundaries [610]. Cu diffusion into the CdTe thin film appears to be promoted by the etching step [610] which often precedes the back contact formation (see Figure 5.4). In Section 4.1.6, we stated that most of the Cu atoms in polycrystalline CdTe are located at grain boundaries [610, 611]. Exceptions may be twin boundaries and stacking faults where no Cu was found [611]. Near the back contact, the Cu concentration is highest [605] and leads to highly conductive grain boundaries [610]. If the back side of CdTe is highly Te enriched (due to etching), Cux Te can form [612]. However, forward bias stress removes Cu atoms from the back contact region [603]. Close to the CdS junction, the Cu atom concentration in CdTe is 1017 –1018 cm−3 and does not change upon device stress [605]. A very high concentration of Cu is accommodated in the CdS layer [605], probably also at grain boundaries. This Cu content in CdS increases drastically with forward bias stress at elevated temperature [605]. In the following, we give a survey of physical effects of Cu in the CdTe solar cell which may be responsible for both the increased cell performance and the possibly accelerated degradation: 1) Reduction in charge carrier lifetime. Using time-resolved photoluminescence it was found that Cu addition decreases the apparent lifetime by half an order of magnitude [606]. One also finds an increased defect density [202, 606] which is in agreement with the lifetime measurement. Yet, there may be a small process window for increased carrier lifetime upon Cu doping [208]. 2) Increased p-type doping. This effect appears to be in the same order of magnitude as the chlorine and oxygen effect on conductivity [606], that is, an increase in hole concentration by 1 order of magnitude. 3) Reduction in back contact barrier. The result is an increased forward current [602] and better fill factor. Currently, it is not clear if this effect is merely an outcome of effect (2) or if it is due to a change in barrier height of the back contact barrier. 4) Compensation doping of CdS. As Cu atoms from the back contact appear to accumulate in the CdS layer – an effect which is promoted by forward bias stress – they may compensation dope the CdS layer. In consequence, the barrier for the diode current is increased and the crossover effect becomes visible [603]. A usual effect is a strong apparent quantum efficiency under forward bias (Section 7.4.6). Obviously, there are beneficial and detrimental effects of Cu doping. A CdTe cell with more than 12% efficiency has been prepared without intentional Cu addition [571], which stands as proof that high Cu concentrations are not always required.

247

248


(2)

(1)

(11)

(12)

(10)

(4)–(6)

(3)

(9)

(8)

(7)

(16) (14) (15)

(13)

Lamination foil tailoring

Cover glass washing

Figure 5.4 Schematic of a commercial production line for CdTe thin film modules. Numbers refer to process steps listed in Table 5.4. Table 5.4 Process sequence for monolithically integrated CdTe-based modules. Sketches are not to scale.

Step number Process step

Remark

1 2 3

Glass washing TCO sputtering P1 scribing

Detergent-free, deionized water – Laser scribing

4 5 6

Buffer deposition Inline process Absorber deposition High temperature process Activation One or more process steps

7

P2 scribing

Laser or mechanical scribing

8 9

Etching Back contact

Spray or dip etching Deposition

10

P3 scribing

Laser or mechanical scribing

11

Edge insulation

Sand blasting or laser ablation

12

Contacting

Adhesive tape

13 14 15 16

Electrical pretesting Lamination Contact box Testing

Isolation test Polymer foil – Flash solar simulator

Sketch – – P1

– – – P2

– –

P1 P2 P3

Contact

– – – –


5.1.8 Back Contact

For several reasons, the formation of a back contact to p-CdTe with low barrier height is not so trivial. The charge neutrality level of CdTe lies in the upper half of the bandgap (Table 4.6). This leads to large barrier heights at the non-reacted CdTe/metal interface [613, 614]. The CdTe layer is difficult to dope p-type which impedes tunneling junction formation. The back contact has to be formed at relatively low temperature in order to avoid deterioration of the main junction. 5.1.8.1 Surface Modification Today’s CdTe back contacts are often formed in two steps: (i) surface modification, and (ii) contact formation. During surface modification, foreign phases (mostly oxides) are removed from the CdTe surface and a Te-rich surface is formed. The most common process variant is wet etching in oxidative solutions. One such solution is nitric phosphoric (NP) acid, with typical composition 70% phosphoric acid (H3 PO4 ), 1% nitric acid (HNO3 ), 29% de-ionized water. The etching solution can be stored for many months [615]. Etching times are 30–60 s [568, 571]. Also spray NP etching has been employed. The NP solution strongly etches the surface and grain boundaries [616] and is therefore critical for thin CdTe films. It removes oxides and chlorides as well as cadmium [617]. The CdTe surface after NP etching has been found to be stable against strong oxidation for ∼1 h in ambient air [615]. In order to slow down the etching process, a more diluted NP solution or alternatively a nitric acetic acid may be employed [618]. Bromine methanol is an alternative etching solution (BrMeOH etch). Typical concentration is 0.1–1.0% Br2 in MeOH [616], with etching times of 5–20 s [568]. The BrMeOH solution is prepared soon before application [615] and therefore is considered less practical in production [618]. Compared to the NP etch, the BrMeOH etch leads to less surface smoothening and less grain boundary widening [619]. The BrMeOH etched films soon oxidize in ambient air [615]. Both etching solutions, (NP and BrMeOH) oxidize surface Te2− ions to Te0 and dissolve Cd2+ ions. They produce a Te layer on the CdTe absorber layer. In the case of the NP etch, the Te layer can have thicknesses of 30–100 nm [617]. With BrMeOH etching, the Te film thickness is only 7–8 A˚ [615]. Evaporated Te films have RT p-type conductivity of 1–10 S cm−1 [620] with p ≈ 1017 cm−3 and µp ≈ 150 cm2 V−1 s−1 [621] or can even be degenerated with p ≈ 1018 cm−3 with µp ≈ 2 cm2 V−1 s−1 [622]. The bandgap of Te is 0.33 eV at RT [142]. The valence band offset at the CdTe/Te interface is reported to be Ev = 0.2–0.6 eV, with strong dependence on film thickness [533, 617]. Grain boundary Te-enrichment is considered to be problematic in terms of shunting. In order to match vacuum processing, a dry surface modification has been developed. Using Te + H2 in the gas phase CdTeO3 and CdO oxides are removed and a Te layer is deposited [623, 624]. Another variant of dry surface modification is ion beam etching which only removes surface contaminants but does not lead to

249

250


Te enrichment. Using ion beam etching and subsequent ZnTe : Cu/Ni contacting, more than 12% efficient solar cells have been prepared [625]. In some cases, surface modification is completely left out and the back contact is directly formed. Examples are with doped graphite pastes as direct contact [626], with Cu halide as primary contact [627], and by application of a ZnTe : Cu/Ti contact [208]. 5.1.8.2 Primary and Secondary Contacts As direct p-CdTe/metal contacts give large barrier heights [613], an interface layer (primary contact) is introduced. The primary contact shall be a semiconductor with small valence band offset to CdTe in order to allow efficient hole transport. It is highly doped in order to form a tunneling junction to the secondary contact. And finally, it is grown at low substrate temperature (≤300 ◦ C). The simplest approach would be to use the Te layer which formed during surface modification. The reported performance values of a Te/Ni back contact, however, are poor [618, 623]. Thus, the primary contact layers added to the structure are mostly telluride compounds. An often used primary contact to CdTe is a Cux Te phase. Cux Te can be formed by depositing elemental Cu [612] or the compound Cux Te [628] on the etched (and Te rich) surface. It is surmised that phases of Cux Te also form by in-diffusion of Cu [629] or Cux Te [630] from a Cu-containing secondary contact (Cu-doped graphite paste). Cu2 Te has a bandgap around 1.1 eV [631] and can be highly doped p-type (≈1021 cm−3 ) [632]. Although the ideal Cux Te/CdTe interface has a high Ev = 0.8 eV [633], Cux Te-based back contacts systematically show high device efficiency if compared to other back contacts [8, 634]. The barrier height has been measured to 0.3 eV [619] in obvious contradiction to the found Ev . By using different deposited Cu thicknesses, Cux Te phases with different x can be formed [635]. Whether device performance is superior with a Cu1.4 Te/CuTe phase mixture [612] or with the pure Cu2 Te phase [624] is still not clear. Cu-related oxides formed during device fabrication also play a role and can induce a rollover anomaly [636]. As Cux Te is highly doped, Cux Te forms a tunneling junction to the contacting metal. Thus, many metals such as Al, Ni, Cr, ITO, Mo, graphite, and so on, have been used as secondary contacts. In another variant of a Cu-containing back contact, a Cu halide is deposited by vapor deposition [627]. For the effect of Cu in CdTe, see Section 5.1.7. Cu-containing back contacts give the highest efficiency but are often less stable (see Section 6.1.7) A Cu-free back contact is formed by sputtering Sb2 Te3 as the primary contact [637]. When combined with Mo as the secondary contact, good stability has been demonstrated [634]. Typical film thicknesses are 300 nm Sb2 Te2 and 300 nm Mo [570]. Sb2 Te3 is a small bandgap semiconductor (Eg ≈ 0.2 eV, [638]) with a hole concentration of around 5 × 1019 cm−3 [639]. Substrate temperatures for the sputter process are 150–300 ◦ C where higher values give better conductivity and stability [639]. The barrier height of a Sb2 Te2 /Mo back contact was measured as Bp = 0.35 eV [619]. Although this barrier height induces already a noticeable


rollover anomaly of the JV curve (see Section 6.1.4), efficiencies exceeding 15% have been reported [570]. ZnTe fulfills the three necessary requirements for a primary contact. It has a small valence band offset to CdTe of Ev = −0.1 eV [640, 641]. It can be highly doped p-type by nitrogen doping [640, 642] (p ≈ 2 × 1018 cm−3 ) or copper [643] (p = 1019 –1020 cm−3 ). Thus, in spite of the large bandgap of ZnTe (Eg = 2.3 eV), low resistance (LR) contacts to the secondary materials Au and Ni can be prepared [642, 643]. And finally, doped ZnTe can be prepared by evaporation or sputtering at not too high substrate temperature (300 ◦ C, [208]). ZnTe : Cu film thicknesses range from 40 nm [618] to 0.5 µm. Cu diffusion into CdTe, if taking place on the right level, is observed to increase CdTe p-type doping and to improve the carrier lifetime [208]. Crossover of dark and light JV curves for ZnTe : Cu is observed (see Section 6.1.4). Solar cell efficiencies exceeding 13% have been obtained with ZnTe : Cu [644] and around 10% with ZnTe : N [645]. After an initial cell degradation under stress, the cells with a ZnTe : N back contact show stable performance [642]. Other Cu-free back contacts comprise the transparent In2 O3 : Sn [646] and graphite paste doped with Ni2 P [647]. In both cases, no primary contact has been applied (besides the inherently present Te layer after etching). 5.1.9 Module Fabrication and Life Cycle Analysis

A process sequence for the fabrication of monolithically integrated CdTe modules is listed in Table 5.4. Due to the superstrate configuration, the TCO layer is deposited first. TCO covered glass, including a diffusion barrier, may be a purchase part which then requires washing before processing. According to the process sequence in Figure 5.4, the monolithical interconnection scheme is as shown in Figure 2.43a. Also use of the scheme shown in Figure 2.43c has been reported [648]. The latter requires an additional filling step of the P1 trench. Buffer deposition, absorber deposition, and activation may be accomplished in one vacuum chamber as shown in Figure 5.4. The productivity of CdTe module manufacturing is typically high at all fabrication steps. In particular, CdTe vapor deposition can be accomplished at very high rates (cf. Section 5.1.4) allowing a line throughput of around 1 module min−1 . The CdTe module is protected by a glass back sheet which is laminated by use of an interlayer foil. Lamination foils for CdTe and Cu(In,Ga)(S,Se)2 modules are mostly from ethylene vinyl acetate (EVA) or polyvinyl butyral (PVB). Other lamination foils are from, for example, thermoplastic polyurethane (TPU), ionomer (ION), or thermoplastic silicone (TSI). EVA shows the second lowest water vapor transmission rate (after ION) but still the highest transparency degradation in ultraviolet light [649]. If the module is not framed, special attention has to be paid to the mechanical intactness of the glass edges, as micro-cracks can lead to glass breakage. Edge seaming is a solution to this problem. CdTe module production implies handling and processing of hazardous chemical substances such as CdS, CdTe, and CdCl2 powders or related liquids. Among these,

251

252


due to its solubility in water, CdCl2 may be the most critical. Elemental cadmium is a lung carcinogen that also can damage kidneys and bones. Very limited data exist on the toxicology of the compound CdTe [650] – therefore the hazard persented by elemental Cd is the basis for the assessment of the environmental, health, and safety (EHS) risks of all Cd compounds [651, 652]. EHS risks are connected with Cd mining, Cd compound production, module production, life cycle accidents, recycling, and waste disposal. According to Fthenakis et al., the highest Cd emission occurs during Cd purification and CdTe feedstock production [652]. However, the total emission is only 23 mg Cd per 1 GWh of produced energy [652]. During module production, special safety measures have to be taken in order to prevent human incorporation of Cd-containing gases and dusts. Most critical appears to be the overhaul of the CdTe deposition chamber where particles and dusts may emerge [653]. More than one decade of experience exists about the exposure of employees in a CdTe module production. Monitoring tests for urine cadmium, blood cadmium, and β2 microglobulin excretion have been evaluated. Urine cadmium is primarily indicative of long-term cadmium exposure, blood cadmium is primarily indicative of recent exposure, and β2 microglobulin is a secondary indicator of cadmium exposure. None of these biological markers were found to be increased for long-term employees [653]. Obviously, safety measures during CdTe module production can be effective. Normal operation of a CdTe module does not emit a measurable quantity of Cd. The vapor pressure of CdTe at operation temperature is too low and the films are laminated between two diffusion barriers (normally glass). A critical situation is a fire accident. Experimentally, the effect of flame temperatures of 760–1100 ◦ C on a glass–glass laminated CdTe module has been assessed [654]. Under these conditions, 0.4–0.6% of the Cd content was released through the open perimeter of the module before the two sheets of glass fused [652]. Higher flame temperatures seldom occur but probably will lead to higher emissions. Damaged modules can be disposed of in landfills as they pass leaching criteria for non-hazardous waste [651]. At the end of lifetime, the CdTe module can be recycled [655, 656] where the expected emissions are negligible compared to other process steps in the CdTe life cycle [652].

5.2 Cu(In,Ga)(S,Se)2 Cells and Modules

Substantial progress in module technologies related to area and efficiency is a prerequisite for the success of Cu(In,Ga)(S,Se)2 (CIGS)-based modules on the PV market. Production lines mainly rely on the fabrication of modules on glass substrates with cell interconnects by laser scribing and mechanical scribing. However, roll to roll manufacturing on flexible foil substrates is a further option on the way to cost reduction. Typical modules have sizes of 0.7–1.2 m2 with efficiencies of up to 13%. Smaller prototypes reach 16% efficiencies.


5.2.1 Substrates

Most Cu(In,Ga)(S,Se)2 solar cells are made in the substrate configuration for which the substrate in principle could be opaque. Nevertheless, the most common substrate is soda lime glass. It has about the same thermal expansion coefficient as Cu(In,Ga)(S,Se)2 (compare Tables 5.1 and 4.8). However, its limited thermal stability allows only process temperatures around 500 ◦ C for large substrates (see Table 5.1). Above this temperature, the glass starts to warp which can aggravate the alignment of scribe lines for the interconnect. Soda lime glass with higher thermal stability but with a similar expansion coefficient is under development. Soda lime glass can act as a sodium source for the absorber layer (see Section 5.2.4) but this can lead to local oversupply with Na. Therefore, glass coated with a diffusion barrier [657] (SiNx or SiOx ) or low Na glass can be used. Soda lime glass produced in a float process has a Sn-side and a Sn-free side. Normally, the thin films are deposited on the Sn-free side of the glass. The typical glass thickness is 2–3 mm but also 1 mm glass has been used [658]. Glass edges may be seamed in order to prevent glass breakage upon handling. Borosilicate glass has a higher softening temperature but a much lower thermal expansion coefficient which can cause delamination problems. The high weight and missing flexibility of glass substrates have stimulated research for alternative substrates such as polymers and metals (Table 5.1). Using polyimide, cell efficiencies around 15% have been realized [659]1) . Limiting factor is the low maximum substrate temperature of around 420 ◦ C [660]. The limit is set by the large thermal expansion of the polyimide (see Table 5.1) rather than by polyimide decomposition which takes place above 500 ◦ C. In the case of monolithic integration on polymer substrates, also laser patterning of P2 and P3 has to be established [661]. Steel substrates are attractive due to their low cost but they need an electrical barrier to allow for monolithic integration [662] and a chemical barrier to prevent Fe diffusion [544]. Single cell efficiencies above 17% have been realized [663]. Ti substrates have been considered for space applications due to their low weight [664]. 5.2.2 Back Contacts

Mo is the most widely used contact material for the base electrode in AI –BIII –CVI 2 thin film solar cells. It was chosen because of its comparatively low-cost, high melting point of about 2700 ◦ C and its expected low diffusivity in the semiconductor films. Polycrystalline Mo films are usually sputtered at RT or electron beam evaporated onto glass substrates at 300–400 ◦ C [665]. The resistivity of sputtered 1) An efficiency of 17.3% has been achieved

by the group at EMPA/Switzerland.

253

254


Mo increases with oxygen incorporated from the residual gas atmosphere [666] and with increasing tensile stress in the film [667]. Tensile (compressive) stress is the result of high (low) sputter gas pressure. Films under tensile stress typically better adhere to the substrate. This is why a double layer of Mo films deposited at high and low gas pressure can easily fulfill the back contact specification – good adherence and low resistivity [667]. Typical Mo resisitivity for modules is 10 µ cm−1 and the sheet resistance is 200 m m−2 . Mo reactivity with selenium or sulfur is relatively small [111]. Nevertheless, at high substrate temperature (≈500 ◦ C) a MoSe2 film [336, 668, 669] or a MoS2 film [455] with thickness up to 100 nm forms during and/or before growth of the chalcopyrite film [670]. Thus, MoX2 (X = Se, S) can form at the Mo/chalcopyrite interface and/or on top of the Mo layer before chalcopyrite formation by exposure to chalcogen vapor. MoX2 with X = Se, S are layered semiconductors with an indirect bandgap of 1.06–1.16 eV [671, 672] and 1.17–1.35 eV [671, 673], respectively. One work reports on a MoSe2 band gap of 1.41 eV which shows up as a peak in the differentiated external quantum efficiency (EQE) [674]. If the MoX2 film were oriented such that the c axis, showing weak van der Waals bonding, is perpendicular to the substrate, then this layer could impede the adhesion of the chalcopyrite absorber. Fortunately, the c axis often is parallel to the substrate as revealed by transmission electron microscopy [668, 670]. The parallel orientation appears to be preferred for MoX2 growth at the Mo/chalcopyrite interface while the perpendicular orientation is preferred for the surface controlled Mo chalcogenization [668, 670, 675]. Both processes may contribute to the MoX2 film thickness while the growth rate of c axis parallel is higher [675]. The transmission electron microscope picture in Figure 5.5 gives an example of a mixed oriented MoSe2 film on a rough Mo layer. In the parallel c axis configuration, diffusion of metal ions into the van der Waals planes of MoSe2 could lead to chemical intermixing at the MoSe2 /AI -BIII -CVI 2 interface and could induce the observed ohmic contact characteristic [676, 677]. Indeed, Ga diffusion into the Mo layer is reported in Ref. [678] and diffusion of Cu, In, and Se ZSW-splnS 951797(pre) 21-10-05

CIGS

MoSe2

Mo

10 nm

Figure 5.5 Transmission electron micrograph of a Cu(In,Ga)Se2 /MoSe2 /Mo interface exhibiting differently oriented domains of MoSe2 . Provided by the courtesy of D. Abou-Ras.


in Ref. [679]. The thickness of the MoSe2 layer increases with increasing substrate temperature [670]. It is smaller for larger Ga content [680] and becomes below the detection limit in the absence of Na in the CIGS layer [681]. It also decreases with increasing O content in the Mo layer [657]. Despite the differences in MoSe2 properties, the barrier height at the Mo/MoSe2 /Cu(In1−x Gax )Se2 interface is small and the contact can be considered as ohmic [676]. In humid air, the Mo back contact can oxidize. This is particularly critical for the Mo in the P3 isolation scribe [682, 683] where a decrease in conductivity can lead to reduction of the module fill factor. Completely oxidized Mo films are optically transparent. The optical reflection of Mo in the wavelength range of 500–1000 nm is only 60%. Orgassa et al. [111] and Malmström et al. [684] tested alternative back contacts with the aim of improving the back contact reflectivity, which is important for thinner absorbers. While Cr,V, Ti, and Mn must be ruled out due to strong reactivity with selenium, the use of Ta and Nb leads to good device quality. Unlike Mo and W, however, Ta and Nb back contacts needed a Ga back surface gradient for high Voc . It was argued that these metals do not form the benign electronic self passivation which is provided by the Mo- and W-containing back contact – possibly due to inferior properties of the selenides formed. Also in the case of a ZrN back reflector better results are obtained with a Ga back surface gradient [684]. For more information on metal contacts to AI -BIII -CVI 2 semiconductors, the reader is referred to Ref. [448]. 5.2.3 Cu(In,Ga)(S,Se)2 Absorber Layers

The preparation of a Cu(In,Ga)(S,Se)2 -based solar cell continues with the formation of the absorber material on the Mo-coated substrate. We may discriminate two process types (see Figure 5.6): (i) co-evaporation where a chalcogenide film is formed already during deposition, and (ii) deposition reaction where a precursor film reacts with the chalcogen in a second step. Both process types (although being fundamentally different) lead to similar material as long as the following requirements are fulfilled: (i) the substrate temperature should reach around 500 ◦ C, (ii) the overall composition should be Cu-poor (there are exceptions), (iii) an oversupply of chalcogen during the process should be provided, (iv) the Ga/(Ga + In) ratio should be 0.2–0.3 (see Section 5.2.5), and (v) sodium doping should be foreseen at some stage of the process (see Section 5.2.4). Optimally, a film of morphology as shown in the secondary electron micrograph of Figure 1.3 is obtained. Although being mandatory, the Cu-poor composition of photovoltaic-grade Cu(In,Ga)(S,Se)2 films is not reflected in this chemical formula. For instance, we speak of a CuIn1−x Gax (S,Se)2 absorber although it is Cu/(Ga + In) < 1 (or, equivalently, x < 0). The allowed stoichiometry deviation x (Eq. (4.5)) is astonishingly large, yielding a wide process window with respect to composition. Devices with efficiencies above 14% are obtained from absorbers with x between

255

256


Deposition Reaction

Co-Evaporation

Se,S H2Se, H2S, Mo-Se (dr1)

(dr 2) Cu ln GaSe S

ln Cu:Ga

(dr3)

N2, Ar

Cu(ln,Ga)(S,Se)2

S, Se ln Cu:Ga

Molybdenum Substrate

Figure 5.6 Film formation processes based on deposition reaction and co-evaporation. Due to the non-transparent contact, the cell is in the substrate configuration. Typical Cu(In,Ga)(S,Se)2 film thickness is 2–3 µm.

−0.07 and −0.44 (or, equivalently, with 0.92 > Cu/(In + Ga > 0.56) if the sample contains Na [685]. However, highly efficient cells have −0.2 < x < −0.1. Cu-rich Cu(In,Ga)(S,Se)2 films show the segregation of a binary phase (Cu2x Se or Cux S) preferentially at the surface of the absorber film. The metallic nature of these phases does not allow the formation of efficient solar cells. However, the importance of the Cu-rich composition is given by its role during film growth. Cu-rich films have grain sizes in excess of 1 µm, whereas Cu-poor films have much smaller grains [686]. The larger grain size has been explained by the presence of a binary Cu-chalcogenide phase which segregates during growth of a Cu-rich Cu(In,Ga)(S,Se)2 film [288]. According to the model shown in Figure 5.7, Cux Se forms a quasi-liquid surface layer and leads to a liquid–solid type of growth. The binary Cu chalcogenide phase mediates the growth of the chalcopyrite crystallites,

In, Ga, Se CuxSe Single phase Cu(ln, Ga)Se2

CIGSe

(a)

(b)

Figure 5.7 Model of Cu(In,Ga)Se2 growth in the presence of Cux Se. (a) Liquid–solid growth takes place due the presence of Cux Se. The phase Cu2x Se acts as a quasi liquid with high Cu mobility. (b) Fully crystallized slightly Cu-poor Cu(In,Ga)Se2 film which is of single phase.

5.2 Cu(In,Ga)(S,Se)2 Cells and Modules Table 5.5

Survey and rough assessment of different CIGS technologies. Process schemes, as labeled in Figure 5.6

Typical process time (min) Process Up-scaling capability (m2 )

dr1

dr2

dr3

co

2 Inline >4

60 Batch 4

10 Inline 5 throughout all evaporation periods. In the two-stage process, the Cu-rich growth phase is realized by an excess of Cu during the first stage. In the second stage, the film is converted to Cu-poor composition. The substrate temperature is high in both stages. In the three-stage process, the sequence is In + Ga + Se deposition, Cu + Se deposition, and again In + Ga + Se deposition [4]. The substrate temperature in the first stage can be 300 ◦ C. Only for a small time period, the film is Cu-rich. In this period, where Cu is diffusing into the BIII x Sey film prepared in stage 1, also the substrate temperature should be >500 ◦ C (or even 600 ◦ C). In the Cu-rich period and shortly before, grain growth takes place

257


Heater / Substrate AAS Source

AAS Sensor

EEIS Sensor Mass Spectrometer

Evaporation sources Figure 5.8 Co-evaporation set-up as used for the preparation of laboratory-scale solar cells and mini-modules, including various methods for process control. The control methods are atom absorption spectroscopy (AAS), electron impact emission spectroscopy (EIES), and mass spectrometry.

2-stage Rate

258

3-stage

Se Cu ln Ga

Cu-rich growth Time or position Figure 5.9 Element fluxes for two- and three-stage co-evaporation processes. For a stationary substrate, the flux rates are given as a function of time. For a moving substrate, the flux rates are given as a function

of position. The heights of element fluxes are not to scale. Horizontal bars give the time or position where the film composition is Cu-rich.

by recrystallization. The two-stage and three-stage processes differ in the final film morphology. As nucleation in the two-stage process takes place under Cu-excess (with Cux Se as a second phase), the films show large grains and large surface roughness. The Cu-rich period in the three-stage process, however, leads to a recrystallization of small grained material (from the first stage) and thus produces large grains with small surface roughness. Currently, the highest laboratory cell efficiencies are achieved by the three-stage method [7]. The processes sketched in Figure 5.9 take place as a function of time if stationary substrates are coated in the laboratory. Or they take place as a function of position


Raw plates

Atmosphere

Conditioning

Line sources Coated plates

Heater Load lock

259

Vaccum Deposition

Atmosphere

Figure 5.10 Inline process with line sources for coating of large-area substrates in top-down evaporation geometry.

if moving substrates are coated in an inline system. For inline co-evaporation of large-area Cu(In,Ga)Se2 absorber films, the substrates are continuously moving through a deposition zone. The substrate width can be >1 m. Most critical is the evaporation of the metals. Two coating schemes are possible: (i) bottom up, and (ii) top down. For both configurations, point evaporation sources or line sources can be applied. Line sources have a multifold of orifices which lead to a largely homogeneous evaporation profile. However, also two separate point sources positioned at each end of the substrate width can provide homogeneous coverage. For top-down deposition, the vapor coming from either a point source or a line source is deflected downwards by heated shields. The advantage of the bottom-up configuration is that no debris from the sources can fall onto the film surface. The advantage of the top-down geometry is that the substrate is suspended on carriers allowing higher substrate temperatures. This configuration is depicted in Figure 5.10. Both deposition types, however, have proven reliable in production. Also for both types of deposition, the Se supply can be from shower-type evaporators where a hot tube having multiple orifices guides the Se vapor to the deposition zone. One advantage of the evaporation method is that material deposition and film formation are performed during one process step. A disadvantage is that evaporation requires the online control of elemental fluxes or film composition. A feedback control based on either electron impact emission spectrometry (EEIS), quadruple mass analysis (QMA), or atomic absorption spectrometry (AAS) allows the control of flux rates of each individual evaporation source. X-ray fluorescence spectrometry is an inline capable method to control the film composition. Furthermore, laser light scattering which uses morphological and optical changes during film growth has been suggested for inline application. Laser light scattering has been proven reliable for stationary substrates [692] and appears to be promising for moving substrates. A survey of control methods can be found in Ref. [693]. For a stationary substrate, the composition of the growing film can also be monitored based on the thermal emissivity of the film [694]. At the transition between In-rich and Cu-rich composition the thermal emission changes, resulting in a change of substrate

260


temperature at constant heating power or a change of heating power to maintain a constant substrate temperature. A further limitation of the co-evaporation method in production may be its throughput. Cycle times below 10 min require evaporation sources which allow high growth rates. A comparison of cycle times is given in Table 5.5. 5.2.3.2 Deposition Reaction The complexity of the equipment for co-evaporation processes can be circumvented by separation of deposition and reaction into two different processing steps. A stack of precursor layers is deposited first. These layers can be metal layers, alloys, or complete compounds typically deposited without substrate heating. Second, the precursor stack is transformed in a thermal reaction step into the semiconductor film. The precursor layers can be put down by several deposition methods. Approved large-area techniques such as sputtering or electrodeposition are applicable. The Cu/(Ga + In) and the Ga/(Ga + In) ratio of the final Cu(In,Ga)(S,Se)2 film are mostly given by the thicknesses of the precursor layers and thus can be sufficiently controlled, for example, by X-ray fluorescence spectrometry. Only indium and selenium in the form of In2 Se may (to a small extent) be lost in the reaction step. There are different variants of chalcogen supply for the reaction step. In the reaction scheme dr1 (see Figure 5.6), elemental selenium, or sulfur vapor is supplied. In the reaction scheme dr2, compound gases are the chalcogen source. The structural and electronic properties of the semiconductor film depend on the controlled reaction of the precursor film with the chalcogen. This reaction is not easy to control because it is driven by reaction and diffusion processes. Significant progress in understanding the film formation has been made by in situ studies of the reaction kinetics during film formation by in situ X-ray diffraction [695, 696]. In the following, we describe selenization from H2 Se gas and annealing of stacked elemental layers (process dr3) in more detail. Selenization from H2 Se Gas First films were prepared by selenization of metal precursors in H2 Se or H2 S [697–699]. The same method was used to fabricate the first large-area modules [597]. A modification of this process is the basis for Cu(In,Ga)(S,Se)2 solar cell production on a GW scale [5]. This process is schematically drawn in Figure 5.6 (dr2). A stack of Cu : Ga alloy and elemental In is sputter-deposited on the Mo-coated glass substrate. Selenization takes place under H2 Se/Ar atmosphere, to form the selenide film. Typical selenization time is hours as can be seen in Table 5.5. Device performance is improved if sulfur is added as H2 S atmosphere toward the end of the process. The resulting surface layer of Cu(In,Ga)(S,Se)2 reduces recombination and improves open circuit voltage (see Section 5.2.6). Annealing of Stacked Elemental Layers If the chalcogen component is part of the precursor stack (dr3), we may speak of a stacked elemental layer [700]. The selenium layer can be deposited by evaporation [701] and provides a small oversupply of Se.


The stacked elemental layer is annealed at temperatures above 500 ◦ C in either an inert or a Se atmosphere. The control of film formation improved by rapid thermal processing (RTP) [687]. The highest efficiencies are obtained if the RTP of the Se-containing stack is performed in an S-containing atmosphere (either elemental S or H2 S gas). The RT process can also be performed in inert gas [702] to prevent Se desorption. It is a fast process with minute process time which is inline capable (see Table 5.5). Due to the high Se pressure in the RTP containment, MoSe2 formation can be fast at high temperature. However, a high temperature is needed for Ga diffusion from back to front. On the laboratory scale, the efficiencies of cells made by these preparation routes are smaller by about 3% (absolute), as compared with the record values obtained by co-evaporation. However, on the module level, co-evaporated and sequentially prepared absorbers have about the same efficiency and reach about 16% on 30 × 30 cm2 [5]. This discrepancy might originate from the easier control of parameter windows for the sequential process (e.g., composition over large substrate surface). Sequential processes need two or even three stages for absorber completion, often in separate equipment. These additional processing steps may counterbalance the advantage of easier element deposition by sputtering. Also the control over composition and growth during co-evaporation is not equally possible for the selenization process. Fortunately, the distribution of the elements within the film grown during the selenization process turns out to be close to what one thinks to be an optimum, especially if the process includes the sulfurization stage. Since the formation of CuInSe2 is much faster than that of CuGaSe2 , and because film growth starts from the top, Ga is concentrated toward the back surface of the film. An increasing Ga content implies an increase in band-gap energy. This introduces a back-surface gradient, improving carrier collection. In turn, S from the sulfurization step is found preferentially toward the front surface of the film, where it reduces recombination losses and also increases the absorber band gap in the space-charge region close to the metallurgical heterojunction. Annealing of Compound Precursors There have been various attempts to deposit the precursor layer already in the form of a compound (not depicted in Figure 5.6). These include electrodeposition [703–705] and particle deposition [706, 707]. Electrodeposition of all elements simultaneously (including selenium) requires the minute adjustment of concentrations in the solution and the application of complexing agents. For all processes, annealing at high temperature must be the second step. (As the precursor film is not in its thermodynamic stable state, we speak of the reaction step.) The annealing of the precursor compound is a recrystallization process that competes with the decomposition of the compound. Hence, optimization of the reaction step is difficult. Cells with good efficiencies were obtained by a hybride process combining electrodeposition of a Cu-rich CuInSe2 film and subsequent conditioning by a vacuum evaporation step of In(Se) [708]. Again, annealing of the films at elevated temperature in a suitable atmosphere is important.

261

262


5.2.3.3 Sputtering Since sputtering is known as a very powerful method for coating of large areas, several attempts for the direct synthesis of compound films where made. Early attempts demonstrated the principle feasibility of the method [679] for selenides and for sulfides [709]. A successful transfer to a roll to roll process on flexible substrates shows the big potential of this method. 5.2.3.4 Epitaxy, Chemical Vapor Deposition, and Vapor Transport Processes Besides selenization and co-evaporation, other deposition methods have been studied, either to obtain films with very high quality or to reduce cost of film deposition on large areas. Molecular beam epitaxy (MBE) [710, 711] or MOCVD [712] have revealed interesting features for fundamental studies, such as phase segregation and defect formation, but are not appropriate for large-area production. Furthermore, none of these approaches led to devices with efficiencies comparable to polycrystalline thin film based structures. Chemical vapor transport of Cu2 Se and Ga2 Se3 was successfully applied to grow CuGaSe2 thin films [713]. In spray pyrolysis, films are synthesized by spraying a solution of reactive species onto substrates at temperatures high enough to evaporate solvents. Compounds are synthesized by reaction of the components on the substrate. The potential of this non-vacuum method is quite high, although it has not yet led to devices with high performance [714, 715]. 5.2.4 Influence of Sodium

For today’s AI -BIII -CVI 2 solar cells, doping of the absorber layer with sodium is a major requisite [716]. This is true if the absorber layer has a Cu-poor composition as is the case for all selenide based absorbers. It is not true if the absorber layer has a Cu-rich composition (after growth) – as for a specific type of CuInS2 solar cells [717]. A typical Na concentration in a Cu(In,Ga)Se2 absorber, prone to yield high efficiency solar cells, is around 0.1 at% [718, 719]. For Cu-poor CuInS2 films the concentration should be even higher (∼1.0 at%) [420, 422]. Upon appropriate Na doping, mainly Voc and FF of the solar cell increase [422, 685, 716, 720, 721]. In addition, there are reports on structural and electronic consequences of the Na doping. Relating these consequences to the gain in Voc and FF is the basis of several models for the effect of Na in AI -BIII -CVI 2 materials. The source of Na for a AI -BIII -CVI 2 film is either the substrate or a Na-containing precursor layer (NaF, Na2 S, Na2 Se, etc.). Soda lime glass contains up to 15% NaO2 . It is the standard glass for a AI -BIII -CVI 2 solar cell (also because of the good match in expansion coefficient which is 9 × 10−6 K−1 ). In order to dope the film, Na has to diffuse out of the glass and through the Mo back contact. The diffusion channels appear to be Mo grain boundaries or Mo oxide phases present in grain boundaries [722]. For both types of Na supply (soda lime glass or Na precursor), it is clear that the absorber growth takes place in the presence of Na. It has been reported that Na influences grain size and preferential orientation, however, these


reports are contradictory. In fact, the morphology influence depends on the type of absorber formation. A clear increase in grain size is found for films grown with a bi-layer type of co-evaporation process (Cu-rich/Cu-poor) [718, 723] where the grain size increases from ∼0.3 to 1 µm upon Na addition. Also, for sequential Cu(In,Ga)Se2 growth, grain size enhancement is observed [720]. For a process which inherently yields large grains, such as the three-stage co-evaporation process (see Section 5.2.3.1), the opposite behavior has been reported – a decreased grain size due to Na doping [724]. Nevertheless, also for three-stage co-evaporated films, Na doping is indispensable [724]. An effect which is common for all types of chalcopyrite film formation, is the increased conductivity due to Na doping. Typically, a 0.1 at% Na incorporation increases the conductivity by 1 order of magnitude [685, 725] as compared to Na-free films. In the case of CuGaSe2 even a factor of 103 was reported [416]. Hall measurements [726], capacitance–voltage profiles [719, 727, 728] and drive-level capacitance measurements [721] assign this conductivity increase at least partially to an increased net hole concentration. According to Erslev et al., Na-doped films are also more susceptible to persistent light-induced conductivity enhancement [729] (see Section 6.1.5). Thus in the light-soaked state, Na-doped films indeed show a larger hole concentration than Na-free films. The larger hole density alone would suffice to explain a certain Voc increase, if the solar cell devices are limited by SCR or QNR recombination (see Section 3.5). As shown by Rudmann et al. the conductivity increase can also be achieved if Na is incorporated in the film by diffusion after the deposition process [730]. Thereby, the film morphology is unaffected. This is an important result as it proves that the effect of increased carrier concentration due to Na doping is (at least partially) an electronic effect which is not (completely) mediated by structural changes. The increased hole concentration due to Na doping may have its origin: (i) in the grain interior where Na acts as an acceptor or where it passivates donors, and/or (ii) at the grain boundaries where it passivates donor states. Before we come to models proposed for both scenarios, let us ask for the experimental evidence about the Na distribution. Thereby, we bear in mind that 0.1 at% Na corresponds to a Na concentration of 2 × 1019 cm−3 . The net doping concentration in thin films, however, is around p = 1016 cm−3 – thus below the detection limit of the experiment mentioned in the following. Using SIMS mapping Bodegard et al. found that in a device quality thin film, Na is located mainly at GBs [718]. This was confirmed by Niles et al. by Auger electron spectroscopy. In the latter study, also O was found localized at GBs [731]. So far TEM analysis gave no evidence for Na at grain boundaries [478]. In an attempt to discriminate a sodium bulk from a sodium grain boundary effect, Na was diffused into stoichiometric CuInSe2 crystals. However, the bulk Na concentration was found to be below the detection limit and, instead, Na formed precipitates at the surface [732]. A similar result was obtained by Heske et al. who could detect a large but fixed amount of Na at the CIGS/CdS interface [733]. Concluding these results, we can state that the high concentration of Na which is optimum for cell efficiency is not soluble in the chalcopyrite crystal. Most of the Na is at grain boundaries.

263

264


This complies with theoretical considerations of Wei et al. who deduced that the alloy Nax Cu1−x InSe2 has a positive mixing enthalpy and large amounts of Na will lead to formation of precipitates [297]. The experimentally observed precipitate of Na-doped Cu(In,Ga)Se2 thin films is Na2 Sex (x = 1, 2, 3, 4, 6) [734] and that of Na-doped CuInS2 films is NaInS2 [735]. Knowing that a large amount of Na is found in grain boundaries, the grain boundary passivation model of Cahen [736] (see Section 4.2.6) could explain the effect of Na on the hole density [737, 738]. After this model, Na acts as a catalyst for the oxidation of In dangling bonds. Thus, the Na effect (via oxygen) is to reduce the density of compensating donors located at grain boundaries by oxygenation. As a consequence, the net hole concentration in the films rises. Experimental support for this model may come from the concomitant detection of Na and O [420, 685, 731]. Let us come back to a possible Na concentration in the range of 1016 cm−3 , which may be in the grain interior and which still may be responsible for the conductivity enhancement. The first-principles calculations of Wei et al. have shown that the NaIn acceptor level is not very shallow (see Figure 4.14) and thus is ruled out as the decisive defect [297]. NaCu on the one hand does not introduce a gap level and therefore should be electronically inactive in CuInSe2 . On the other hand, the formation of Na(InCu) is exothermic, and therefore removal of InCu donor states could be the main influence of Na doping [297]. The same mechanism is proposed for CuInS2 [739]. A conclusion, whether Na mainly acts at the grain boundary or in the grain interior, cannot be given yet. Finally, we give a brief summary of other effects of Na in AI -BIII -CVI 2 absorber layers: 1) Na diffusion (via grain boundaries) appears to be maximal for stoichiometric films [685]. 2) More Na is typically found in films with smaller grain size [719, 740]. 3) Na reduces the diffusion of metals in the growing film [660, 724]. This can lead to an unwanted Ga gradient in the film. 4) In humid air, the presence of Na doping leads to accelerated surface oxidation in the form of Se oxides [734]. In addition, the formation of elemental Se is observed [734]. The latter effect may be critical for device fabrication and stability. 5) Na diffusion from the absorber seems to be promoted by exposure to air [740]. 6) Na (in the form of Na2 Sex ) can be largely removed from the surface of a film by a H2 O rinse [723], but it is not completely removed in a buffer CBD [733]. 7) Na appears to support the growth of MoSe2 at the back contact [681, 740]. 8) There are indications that Na promotes the incorporation of the chalcogen component into the growing film [728, 741, 742]. In sequential AI -BIII -CVI 2 formation this can influence the reaction path [421, 742] and can delay the formation of the chalcopyrite phase. This can be advantageous in terms of larger grain size. 9) Na widens the existence range of the chalcopyrite α phase toward the group III atom side [413].


5.2.5 Influence of Gallium

Many experiments showed that Ga addition to CuInSe2 [381, 743, 744] as well as to CuInS2 [745] increases the solar cell efficiency. The main improvement due to Ga addition is in Voc and FF. Today an optimum Ga content in CuIn1−x Gax Se2 appears at x ∼ 0.3 [7]. During coevaporation, gallium can be incorporated easily using a separate effusion cell. At constant Ga flux, this yields homogeneous Ga depth profiles. However, front surface gradients as well as back surface gradients can be produced by a varied Ga flux rate [110, 121]. Locally increasing the Ga content results in a locally increased bandgap by mainly shifting the energetic conduction band position [515]. During CuIn1−x Gax Se2 and CuIn1−x Gax S2 formation in a deposition reaction process, Ga tends to accumulate near the back contact [743, 746]. This phase separation was explained by the kinetically delayed CuGaSe2 [747] or CuGaS2 [748] formation. Annealing in H2 S above 500 ◦ C reduces the Ga gradient [749]. A Ga gradient toward the back contact acts as a back surface field which helps to improve all cell parameters [121], especially for thin absorbers [110]. This is in accordance with the rule in Section 3.11. What is the influence of Ga on the absorber properties besides a bandgap gradient? At first place Ga increases the bandgap providing a better match to the solar spectrum. However, Figure 6.8 reveals that the efficiency gain by Ga addition is larger than the increase of the theoretical η curve. As discussed in Section 4.2.3.4, the conductivity in the system CuIn1−x Gax Se2 does not clearly depend on x. Ga widens the existence range of the α phase and, consequently, the tolerance to deviations from the Cu/(Cu + Ga + In) ratio [413]. At x ∼ 0.25, the grain size of CuIn1−x Gax Se2 films prepared by the three-stage process has a maximum [750]. Furthermore, x ∼ 0.25 marks the position where the c/a ratio of the elementary cell equals 2 [750]. Thus, we could anticipate that Ga addition, up to some point, reduces the defect concentration. Indeed, in earlier films a clear minimum of the electronic bulk defect density [379, 751] was found at x = 0.3. However, the respective defect state N2 (see Section 4.2.3.3) in more recent cells is below the detection limit of admittance spectroscopy in the complete CuIn1−x Gax Se2 solid solution system [381] and a relation between defect density and cell performance is currently not obvious. For pure CuGaSe2 , a deep defect is found in admittance spectroscopy [381]. Theory predicts that CuGaSe2 spontaneously forms VCu defects, being acceptors, whenever a band bending is induced by heterojunction formation [21]. Thus, a high recombination zone in the space charge region may result, which may lead to very small carrier lifetimes (cf. Section 6.2.2). 5.2.6 Influence of Sulfur

In the system CuInSe2 – 2 y S2 y the bandgap can be tuned from 1.02 to 1.52 eV [752]. In the case of Cu-poor CuInSe2 – 2 y S2 y films, the Se/S ratio in the film depends

265


50

540

40 520 30 500

20 10

8

10

12

14

16

18

Voc [ V]

Carrier lifetime τ [ns]

266

480

y= S/(Se + S) Figure 5.11 Carrier lifetime measured by time-resolved photoluminescence () and open circuit voltage (•) as a function of the sulfur content in the absorber. Redrawn after Ref. [753].

on the supply of the chalcogen component while under Cu-rich growth conditions, the incorporation of sulfur is strongly preferred [752]. The element sulfur is very important for chalcopyrite formation in a deposition reaction process. Many CuIn1−x Gax Se2 solar cells where the absorber is grown by reaction of metallic precursors in a chalcogen gas use a CuInSe2 – 2 y S2 y or CuIn1−x Gax Se2 – 2 y S2 y surface layer [749, 753, 754]. Sulfur addition to CuInSe2 leads to an energetic reduction of the valence band maximum and an enhancement of the conduction band minimum according to theory [515] and experiment [373]. Thus, sulfur could widen the interface bandgap and thereby depress interface recombination. In addition, it has been shown that sulfur leads to an increased carrier lifetime in the absorber (see Figure 5.11). The effect on the cell parameters is mostly an increase in Voc as shown in the example of Figure 5.11 with little influence on Jsc [687]. This was explained by a passivation of recombination centers [755]. A too high sulfur concentration in the Cu-poor absorber leads to a reduction in carrier concentration [756] (see Section 4.2.3.4). Therefore, today efficient CuInS2 solar cells use Cu-rich grown absorbers that have been chemically etched in order to remove Cux S binary phase segregations [424, 717]. We note that these films grown Cu-rich have a high carrier concentration of around 1017 cm−3 . They show, however, limited Voc values due to a reduced activation energy of the diode current (see Section 6.2.2). 5.2.7 Influence of Oxygen

Oxygenation2) of complete solar cells of the type CdS/CuInSe2 (early type) for some minutes at 200 ◦ C in dry air was an important post-deposition treatment 2) Here we use the terminus ‘oxygenation’

(instead of ‘oxidation’) because we want to refer to the incorporation of oxygen rather

than to the specific type of charge transfer. For a detailed discussion see Ref. [757].


S + + + + −

−

−

+ + + + −

+ +

−

−

−

+ + + + −

+ −

− +

− +

−

− +

−

−

In

+ +

S

S

+

+ −

S

+

+

Figure 5.12 Defect chemical reaction sequence of O2 at the surface of a chalcopyrite semiconductor (here CuInS2 ). After Ref. [736]. A surface In atom forms two unsaturated bonds due to a missing chalcogen atom. These defects form donor states which are ionized by donating electrons to the conduction band and compensate ionized

S

In

S,O

S

acceptor states. A surface charge results which is screened by depletion zones and thus leads to surface band bending. Chemically adsorbed oxygen passivates the donor states. As a consequence, the compensation is reduced, the concentration of free hole is increased, and the band bending is reduced.

and led to increases in Voc and FF [758, 759]. The effect was explained by the Cahen–Noufi model presented in Figure 5.12. According to this model, non-oxygenated chalcopyrite absorbers would have chalcogen vacancies, VX (X = Se, S), at surfaces and grain boundaries. These vacancies should be positively charged and should induce band bending, up to type inversion at the surface or grain boundary (Figure 5.12). The grain interior could have become depleted from free carriers. Oxygen could passivate the VX donor states at grain boundaries and at the surface. It should reduce the band bending and increase the net doping density in the absorber. As an experimental proof, after oxygenation an increased dark conductivity had been measured which was partly reversible upon reductive treatments [759]. According to first principles calculations, OSe is a deep acceptor (see Figure 4.14) which may become more shallow at the interface [297]. In principle, type inversion at the interface is beneficial for suppressing interface recombination (see Section 3.4). While the as-grown CuInSe2 surface indeed is type inverted (see Figure 4.22), oxygenation was shown to reduce (but not to remove) this type inversion [26]. Therefore, oxygenation of the absorber surface should be detrimental. However, it was argued that wet chemical CdS deposition removes the oxides at the surface, thereby restoring the type inversion of the absorber surface [738]. Thus, oxygenation would reduce grain boundary barriers but not the interface hole barrier. The formation of Cu–BIII –O2 delafossite phases [760] has not been observed. Since the introduction of Ga into the Cu(In,Ga)Se2 absorber and since the employment of Na doping, the oxygenation treatment is no longer applied. (As

267

268


an exception oxygenation of the absorber/buffer preproduct for highly efficient Cu(In,Ga)Se2 cells is reported in Ref. [7].) It is argued that, with Na doping, oxygenation already takes place during air exposure of the absorber at RT [737, 738]. Hence, the typical Na-doped chalcopyrite absorber film, when processed to solar cells, is in the oxygenated state. Further oxygenation by a heat treatment can be critical and can even reduce the efficiency [26]. For strong oxygenation of the absorber in humid air a kink forms in the light JV curve, and the dark and light JV curves cross [761]. Braunger assigned the kink to the segregation of elemental selenium and the formation of an electronic barrier [762]. A survey on the influence of dry and wet oxidation treatments of chalcopyrite semiconductors can be found in Ref. [448]. 5.2.8 Buffer Layers of CIGS

The introduction of a thin CdS buffer layer into the CuIn1−x Gax Se2 (CIGSe) solar cell was an important innovation to exceed 12% efficiency for chalcopyrite thin film cells in 1987 [6]. Since then, the basic layout of the CIGSe solar cell has not changed very much. CBD of CdS still gives the best performing and most reliable solar cells. Thus, we start by descriptions of the CdS deposition process. 5.2.8.1 Chemical Bath Deposited CdS CdS films are deposited by CBD from ammonia solutions using thiourea as the sulfur source and cadmium–ammonia complex ions as cadmium precursors. Other sulfur sources and Cd salts are less common [763]. Aqueous stock solutions are prepared using a Cd salt [Cd(CH3 COO)2 = Cd(AcO)2 , 0.1 M], thiourea [CS(NH2 )2 , 0.2 M], and ammonia (2.44 M) [764]. Chemical purities are 4 N for the Cd salt and 2 N for thiourea. The latter may be further cleaned by recrystallization. Deposition of a standard CdS buffer for a CuIn0.7 Ga0.3 Se2 solar cell may use concentrations of 1.3 mM Cd(AcO)2 , 0.12 M thiourea, and 1.0 M ammonia. The pH of the solution typically is 11, and the final bath temperature 60 ◦ C . After about 5 min, a film of 50 nm thickness has grown. The ammonium salt given in Table 5.3, which buffers the pH of the solution, is not commonly used for CuIn0.7 Ga0.3 Se2 cells. The general chemical reaction of the process is given by − Cd (NH3 )2+ 4 + SC (NH2 )2 + 2OH → CdS + CN2 H2 + 4NH3 + 2H2 O (5.1)

Ammonia supplies the ligand for the metal ions and controls (via the pH) the hydrolysis of thiourea. The actual deposition rate can depend on the thiourea contaminations. Thus, its concentration may have to be adjusted for each batch of thiourea. The actual reaction scheme behind Eq. (5.1) is still a matter of debate. In particular, it has been postulated that Cd(OH)2 species adsorbed at the CIGS absorber surface form intermediate products [764] which facilitate the heterogeneous growth mode [765]. Here, heterogeneous growth means that films grow in an ion by ion mode on the substrate. This growth mode is in competition

60

60

40

40

20

20

0

0

2 4 6 Deposition time [min]

8

Thickness [nm]

Temperature [°C]


0

Figure 5.13 Growth of CdS thin buffer layer in a chemical bath. Solution temperature and film thickness after [769].

with the homogeneous growth where CdS particles form in solution and precipitate at the surface. In the so-called ‘cold start’ recipe, the substrates are immersed into a beaker containing the freshly prepared chemical solution which is stirred. Placing the beaker into a hot water bath (60 ◦ C) constitutes the starting point of the process. An example of the evolution of the solution temperature and the film thickness over time is given in Figure 5.13. During the incubation phase (0–0.5 min) of the process, CdS deposition is negligible but surface contaminations such as Na2 O/NaOH, Na2 CO3 , and Ga2 O3 are cleaned from the surface of the absorber. Thus, metal ions from the absorber enter the chemical solution. In the case of Cu-rich absorbers, which have not been etched in cyanide, this can change the entire chemical deposition process. In Cu-poor films, also surface depletion of Cu has been observed [762] which may, however, rather reflect diffusion of Cu into the absorber bulk. After 30 s in the chemical bath adsorbed Cd ions have formed a film of Cd(Se,OH) [766]. Cd ions may have been diffused into the absorber layer [767]. The same modified CuIn1−x Gax Se2 can be obtained by a partial electrolyte treatment where the sulfur source has been left out. With this treatment alone and window deposition by sputtering, working solar cells have been obtained [768]. The growth phase starts with the heterogeneous growth of CdS. The substrate (absorber) is completely covered by CdS after approximately 1 min deposition time [770] which corresponds to a film thickness of some nanometers. Growth proceeds with a rate of around 0.3 nm s−1 . At the end of the process, the saturation phase is reached where the solution has been depleted from non-reacted Cd species. Before the saturation phase is reached, a strong homogeneous growth sets in which leads to rougher and more porous films [771] due to adsorption of CdS particles. Thus, in order to minimize particle adsorption and to reduce film roughness, the deposition is terminated before the saturation phase. Films are generally rinsed in deionized water and dried in nitrogen. The beaker is cleaned in diluted HCl and can be used for the next deposition. The CBD process can be automated either with

269

270


parallel processing of many substrates or in a quasi inline system. For production, liquid waste management is an issue [772, 773]. Nevertheless, the CBD process is well suited for large-scale production and requires less investment than a vacuum process. Finished CdS films show [Cd]/[S] > 1 and contain considerable amounts of carbon, oxygen, nitrogen, and hydrogen. The reported oxygen concentration in CBD-CdS is between 1% [774] and 10% [775]. Oxygen is incorporated as Cd(OH)2 , CdCO3 , and H2 O. The first two phases accommodate part of the Cd surplus. In addition, Cd(Se,OH) and diffused Cd may account for [Cd]/[S] > 1. From the fact that the hydrogen concentration of around 10% exceeds the concentration of other impurities by a factor of 2, it was concluded that the CdS film contains atomic hydrogen [774]. Also In ions appear to diffuse into the CdS film [776]. The crystal structure of CBD–CdS is mostly hexagonal but films also contain the cubic phase [777]. Hetero-epitaxy between CdS and CuIn1−x Gax Se2 has been observed on a local scale [778, 779]. In the bulk of the CdS film, the grain size is in the range 20–30 nm. The minimum CdS thickness for high efficiency cells is around 30 nm [769]. According to experience, an absorber layer covered by CdS forms a stable pre-product and can be stored for days or even weeks. The CdS overlayer electronically passivates the absorber surface and preserves a high carrier lifetime in the absorber, as measurable by photoluminescence decay measurements [429]. Surface photovoltage measurements support the passivating role of a CdS overlayer [523]. 5.2.8.2 Alternative Buffer Layers The development of a buffer layer which is Cd-free was and still is a major research topic [780]. The CdS buffer layer accounts for an around 2 mA cm−2 loss in Jsc [781] which could be avoided using a higher bandgap material. A summary of results for alternative buffer layers is given in Table 5.6. It shows that Zn- and In-based buffer layers were among the most investigated. The bandgap of In2 S3 is indirect (see Table 4.13) which is beneficial in terms of current collection but which appears problematic in terms of interface recombination [782]. Nevertheless, efficiencies exceeding 16% have been realized. The best performing Cd-free buffer layer is the ZnS-based Zn(OH,O,S) with a champion efficiency of 18.6%. This buffer layer – here grown with a different recipe (see Ref. [780]) – is already used in production [783]. It shows stronger light soaking metastability than a CdS buffer layer [784] but nevertheless allows to pass the IEC 61646 test norm [785]. Alternative buffer layers have been deposited by dry processes such as PVD, CVD, and sputtering. If buffer layers such as Inx Sey and ZnInx Sey could be deposited directly after the absorber layer without breaking the vacuum, cost advantages over the CBD process could be realized. For buffer layers deposited in a separate vacuum chamber this cost argument no longer applies. This brings us back to non-vacuum deposition methods such as spray pyrolysis [809]. Table 5.6 also denotes cases in which the HRW layer has been left out. This is particularly interesting as one complete process step can be avoided. The direct

5.2 Cu(In,Ga)(S,Se)2 Cells and Modules Table 5.6 Summary of Cd-free CuIn1−x Gax Se2 solar cells where the CdS buffer layer has been replaced by other buffer layers.

Buffer

Deposition method

HRW

Pre- or posttreatment

ZnS-based

CBD ILGAR ALCVD CBD MOCVD ALPVD PVD CBD ILGAR ED ALCVD MOCVD Sputtering ALCVD Sputtering CBD ALCVD MOCVD Spray ILGAR Sputtering PVD PVD

+ + + + + – – – + – – – – + – + + + + + + + –

– – – – – Post: air anneal Post: air anneal Post: air anneal – – Pre: NH4 OH – – – Post: air anneal Post: air anneal – Post: air anneal Post: air anneal – Post: air anneal Post: air anneal Post: air anneal

ZnSe-based

ZnInSex ZnO-based

ZnMgO In2 S3 -based

Inx Sey -based

Efficiency

Light soaking effect reported

References

18.6 14.1 16.0 15.7 12.6 11.6 15.3 14.3 15.0 11.4 14.6 13.4 15.0 16.2 16.2 15.7 16.4 12.3 12.4 15.3 12.2 14.8 13.0

+ – – – – + – – + – – – – + – + – – – – – – –

[786, 787] [788] [789] [790–792] [793] [794] [795, 796] [797] [798] [799] [800, 801] [802] [803] [804] [805] [806] [807] [808] [809] [788] [810] [811] [812]

ALCVD, atomic layer chemical vapor deposition; ALPVD, atomic layer physical vapor deposition; CBD, chemical bath deposition; ED, electrodeposition; HRW, high resistance window; ILGAR, ion layer gas reaction; MOCVD, metal organic chemical vapor deposition; PVD, physical vapor deposition.

deposition of ZnO onto CuIn1−x Gax Se2 (x ≈ 0.3) is particularly interesting. With the exception of ion layer gas reaction (ILGAR) all ZnO buffers in Table 5.6 replaced the HRW layer. In the case of sputtered ZnO, the term ‘buffer-free device’ has been employed. According to Table 8.2 and Ref. [813], the direct band offset between the CuIn1−x Gax Se2 surface (with OVC layer) and ZnO is unfavorable due to a negative conduction band offset which can cause interface recombination (see Section 2.4.5.4). Nevertheless, efficiencies of up to 15% have been realized. The band offset between CuIn1−x Gax Se2 (x ≈0.3) and the larger bandgap material ZnMgO should be more favorable [814]; and indeed, efficiencies over 16% are reported in Table 5.6 for ZnMgO buffer layers. The problem of light-soaking or air-annealing sensitivity, however, remains.

271

272


5.2.9 Window Layers of CIGS

The window of a CuIn1−x Gax Se2 solar cell consists of two layers (see also Section 4.4), the HRW layer which is deposited onto the buffer layer and the LRW. A typical combination, which we describe in detail, is the high resistance (HR) ZnO and the LR n-type ZnO : Al. For general information on ZnO windows, see Section 4.4. The HR ZnO is rf (radio frequency, 13.6 MHz) sputtered from a ceramic target in Ar with 15%), the roll over effect does not limit FF and Voc . The second important anomaly, even reported for CdTe cells with η > 15%, is the cross over between dark and light JV curve [833]. The principle appearance and its possible causations are outlined in Section 7.1.2. As the cross over anomaly comes along with an apparent quantum efficiency gain for wavelengths λ = 350–500 nm (see Section 7.4.6), it has been related to an electron barrier at the absorber/emitter heterojunction [835, 841]. Upon blue light exposure, the electron barrier is reduced and the diode current is increased [834]. Thus, the higher diode current under (blue light) illumination appears to be the cause for the cross over effect. The example in Section 2.3.6 explains how deep acceptors in the buffer layer can modify the electron barrier. Different than in Figure 2.13, the electron barrier may also be related to charged states at the buffer/window interface (see Section 6.1.6). A third, less understood anomaly of a CdTe solar cell is the increase in shunt conductance under illumination with respect to the dark measurement [833]. 6.1.5 Transient Effects and Metastability

CdTe cells and modules do show metastability effects being special cases of device anomalies. Most relevant for the application is the increase in Voc after starting the illumination (light soaking metastability). This increase typically is in the range of 2–6% [842, 843]. The Voc increase is reversible upon storage in darkness where the storage time is important (memory effect). The Voc increase has to be taken into account during stability testing by indoor measurements [843]. Typically, a 30 min light soak should precede the indoor measurement at standard test condition. Then, high-quality modules can be measured indoors by voltage sweeps above 1 ms [844]. The Voc increase under illumination may be partly compensated by the stress induced Voc reduction as discussed in Section 6.1.7. Another form of metastability are current transients [845]. Upon electrical forward bias in the dark, the diode current exhibits a transient behavior with current growth as shown in Figure 6.3b or with current decay [846, 847]. During JV analysis this

281

282

6 Photovoltaic Properties of Standard Devices

2500

I (µA)

2000 800 1500 I (µA)

1000 500 −0.2 (a)

−0.3 V

+0.3 V

−0.3 V

600

−20 I (nA)

400 200

0.2

0.4

0.6

Volts (V)

0.8

0

−4 nA

(b) −200

50 −4 nA

25 Time (s)

Figure 6.3 (a) Dark JV curves for a CdTe cell at 90 ◦ C with arrows indicating the sweep polarity. (b) Current transient of the cell upon voltage switching between −0.3 V and +0.3 V. Cell size 0.21 cm2 . From Ref. [848] with permission.

current transient will exemplify as hysteresis effect shown in Figure 6.3a. Current transients have a dynamics below 5% for highly efficient cells and can reach up to 50% for cells with inferior performance [847]. Transient effects increase upon device stressing, that is, the JV curve hysteresis seen in Figure 6.3a becomes more pronounced in degraded cells and modules [843]. The physical reason for the current metastability is not known. Proposed mechanisms involve deep defects which are charged or discharged under forward bias and change the potential drop in the CdTe absorber [847]. Thereby, either the diode current is increased (growth transient) or decreased (decay transient). A dark current decrease would comply with the Voc increase described above. We have discussed the effect of charged defect states in Section 2.4.8. Transient effects, which arise from charged defects, are reversible and reproducible. However, metastabilities in CdTe cells have been reported which show up as erratic current jumps in the JV curve. These have been described as nonlinear shunt paths possibly due to metallic inclusions [848]. 6.1.6 Device Model

Experience shows that CdTe devices with more than 15% efficiency can be produced using different combinations of low resistance and high resistance windows. The high resistance window may even be left out (see Section 5.1.2). Thus, we may simplify matter and discuss a baseline device model similar to the one developed by Gloeckler and Sites [185]. The emitter shall include only one window layer (TCO) and the buffer (CdS). Only as a variant, we will also comment on an emitter type which contains a SnO2 /CdS interface. The baseline band diagram under equilibrium is shown in Figure 6.4a. Parameters used for

6.1 CdTe Device Properties

5

SnO2

CdS

Energy [eV]

4 3 CdTe

2

CdTe

1 0 −4 (a)

CdS TCO

TCO −2

−0.2

0.0

Position z [µm]

Figure 6.4 (a) Band diagram for a TCO/CdS/CdTe solar cell in the dark as obtained by device simulation. The zero of energy is at the valence band maximum of the TCO layer. Defect and doping parameters used in the simulation are given in Table 8.3. (b) ITO/SnO2 /CdS/CdTe solar cell simulated in the dark (solid line) and under

−0.2

0.2 (b)

0.0

0.2

Position z [µm]

AM1.5 illumination (dashed lines). Due to acceptor states at the SnO2 /CdS interface (see Table 4.16), the CdS layer is fully depleted in the dark. Photogenerated holes fill the compensating acceptors in the CdS layer under illumination and the inversion of the absorber increases.

the simulation are given in Table 8.3. The simulated solar cell has Voc = 0.877 V, Jsc = 24.8 mA cm−2 , FF = 73.7%, and η = 16.0%. The critical parameters of the device model are as follows: • Absorber. Champion devices often use CdTe layers close to 10 µm thick [8]. In the device model, we selected only 4 µm CdTe in order to follow the technological trend. Together with the low doping and the back contact barrier, this results in a vanishing quasi neutral region (QNR). The CdTe layer is lowly doped with 2 × 1014 cm−3 , being somewhat lower than the value derived from IQE simulation (see Section 6.1.3). In other works, nonhomogenenous doping levels were employed [849]. The lifetime of electrons in Table 8.3 is fixed by donor type defects of 5 × 1013 cm−3 and an electron cross-section of σn = 10−12 cm−2 . This reproduces the experimental lifetime value of 2 ns (see Section 4.1.3). The lifetime of holes is one order of magnitude larger with σp = 10−13 cm−2 . Thus, the hole cross-section is smaller than in the limit of Coulomb attraction but larger than the geometrical cross-section. The low doping level leads to a wide zone of high SCR recombination and thus to a high J0 (see Section 4.1.3). Together with the lifetime of electrons the wide zone of high SCR recombination limits Voc . The same Voc can be simulated by a smaller lifetime and a higher doping concentration (smaller recombination zone). However, this would result in a red loss in the IQE curve. • Back contact. We assume a high back contact recombination velocity for electrons and a back contact barrier height of 0.4 eV for holes (cf. Section 5.1.8). This leads

283


η(V) 0 30 20 10 0 −10

Jlight -Jdiode

20

Jlight

0 Jdiode −20 0.0

0.4

0.8

1.2 0.0

Voltage [V] (a)

Jlight -Jdiode [mA cm−2]

η(V)

1

Current density [mA cm−2]

284

0.4

0.8

1.2

Voltage [V] (b)

Figure 6.5 Simulated J(V) and η(V) curves of devices in Figure 6.4a,b. Device parameters as in Table 8.3.

to a linear and thus non-exponential J(V) characteristic in the first quadrant of the JV curve but not to complete saturation (roll over) as shown in Figure 6.5a. The roll over has been treated in Section 6.1.4. Further, the back contact results in a bias-dependent external collection efficiency η(V). In consequence, the simulated diode quality factor after the Jsc (Voc ) method is A = 1.3. Correcting after Eq. (2.185), the diode quality factor comes out as A ≈ 2.0. • CdS layer. The CdS layer shall be compensated with a donor concentration of ND,b = 1.1 × 1017 cm−3 and an acceptor level at midgap with Nd = 1017 cm−3 . The compensating acceptors have asymmetric capture cross-sections and thus can be partially neutralized by photogenerated holes. Hence, in the dark the buffer contains less positive charge and shows less band bending than under illumination. We note that the CdS doping level and degree of compensation is not known from experiment and the respective entries in Table 8.3 are speculative. However, the assumed compensation mechanism allows to explain the apparent quantum efficiency effects discussed in Section 6.1.4 for particular devices. We further note that in the cell of Figure 6.4a due to the small thickness of the CdS, the compensation mechanism has negligible influence. As a result, for the cell in Figure 6.4a there is no cross over of dark and light JV curve up to 1.2 V (Figure 6.5). • CdTe/CdS interface. We assume a high interface recombination velocity of Sn0 = 106 cm s−1 by introduction of donor-like defects at this interface. The low doping level of CdTe provides that in Figure 6.4a, where there is no FLP at any of the emitter interfaces, the CdTe surface is fully inverted and Ep,az=0 is almost equal to Eg,a . Thus, interface recombination is suppressed in the dark and under illumination by a large value of Ep,az=0 . Due to the low absorber doping, the interface donors with density 1011 cm−2 do not influence the potential distribution between absorber and emitter.

6.1 CdTe Device Properties

• SnO2 /CdS interface. A particular device structure can include a SnO2 /CdS interface. At this interface, the Fermi level was found to be pinned at Ec − 0.4 eV (Table 4.16) by a high density of acceptor states [533]. This can lead to a band diagram as depicted in Figure 6.4b. In the dark, the low net donor concentration in the buffer does not allow to compensate for the negative charge at the SnO2 /CdS interface. Thus, Ep,az=0 becomes large, as discussed in Section 2.3.4. As a result, the device can become interface recombination limited in the dark. Depending on the value of Ep,az=0 in the dark, the diode quality factor can be below 2.0. Under illumination (dashed lines in Figure 6.4b), the compensating acceptors in CdS are partly neutralized due to the asymmetric capture cross-sections. The net positive charge in the buffer increases and there is a band bending in the buffer. This renders a smaller Ep,az=0 and a diminution of interface recombination as was proposed in Ref. [834]. The effect can be seen in Figure 6.5b. The change from interface recombination to SCR recombination under illumination is revealed by an increase in the difference current Jlight − Jdiode (positive apparent quantum efficiency) at around V = 0.8 V. Photons in the wavelength range of 400–500 nm trigger this effect. The modulated barrier further leads to a negative apparent quantum efficiency for V > 1 V. Again blue photons trigger this anomaly which is the cross over anomaly as described in Section 6.1.4.

6.1.7 Stability

Performance ratio [%]

The stability of module performance is a sensible subject. Typically, the manufacturer guarantees 20 years of module operation with maximal 20% performance loss. This translates into a maximum allowable efficiency degradation of 1% year−1 . Accelerated aging is the tool to extrapolate the module stability for field deployment. CdTe modules have successfully passed the accelerating tests under the test protocol of the International Electrotechnical Commission (IEC 61646). This includes, among others, the harsh conditions of the damp/heat test (85% humidity, 85 ◦ C, 1000 h, darkness). Nevertheless, it is worthwhile to ask for the stability of CdTe modules in the field and under additional stress conditions. Outdoor tests showed that stable performance of CdTe modules with degradation less than 1% year−1 is possible [553, 850, 851]. Figure 6.6 shows the performance ratio of a CdTe system over a period of four years. However, as not all modules were found to be of similar stability [852], it was suspected that the stability depends on 100 80 60 40 20 0

2004

2005

2006

2007

Figure 6.6 Performance ratio of a CdTe system as a function of time. Source: R. Gegenwarth, personal communication.

285

286


the fabrication process. Damp/heat stability does not seem to be problematic – even non-encapsulated mini-modules have passed the damp/heat test [852]. However, device degradation has been reported for illumination at elevated temperature and under electrical bias. Two principle electrical bias conditions have to be differentiated: under open circuit or forward bias, CdTe cells can show a reduction of Voc and FF together with induction of a roll over anomaly (Section 7.1.1) [623, 853]. Under short circuit or reverse bias stress, degradation is less [607] but still there can be a reduction of Voc and FF together with enhancement of the cross over anomaly (Section 7.1.2) [623, 853]. The degradation effects are stronger for Cu-containing cells [607, 853]. The roll over anomaly can be removed by re-contacting the cell [840] and accordingly has been attributed to the back contact. The reduction in Voc and FF has been assigned to CdTe bulk degradation [840] which also may partly be reversible [854]. Although the Cu content plays a decisive role for degradation [607], also other fabrication parameters such as the NP etch, CdTe + CdS thicknesses, and O2 in CdCl2 process have a strong influence [855]. Under optimized preparation conditions and at moderate heat stress (65 ◦ C), stability of modules and cells has been reported even for Cu-containing cells [627, 843, 850, 856]. Cahen and coworkers showed that oxygen during high temperature testing is detrimental for the cell. They proved that Cu-containing cells can show stable performance with oxygen exclusion [857]. This finding would explain why modules are more stable in the field than sometimes to be extrapolated from heat stress testing of bare cells [854].

6.2 AI –BIII –CVI 2 Device Properties

Among the various AI –BIII –CVI 2 based solar cells, currently we find the most efficient cells in the Cu(Inx Ga1−x )Se2 solid solution at x = 0.3. Thus, we mainly focus this section on Cu(In0.3 Ga0.7 )Se2 devices and explain their function as a solar cell. We summarize the knowledge about the band diagram and discuss sources of metastability. Certainly, the large miscibility in the chalcopyrite system allows cells to be prepared also from Cu(Inx Ga1−x )(Sy Se1−y ) or from other pentenary solid solutions. These are treated where appropriate. 6.2.1 Solar Cell Parameters

While the maximum theoretical solar cell efficiency is expected for cells with an absorber bandgap of around 1.4 eV (see Figure 3.1), the best AI –BIII -CVI 2 cells today have a bandgap of around 1.15 eV. Figure 6.7 shows that at this bandgap, both the fill factor and the Voc are highest. For Eg,a > 1.15 eV, Voc falls off the dashed line of proportionality between Voc and Eg,a . The band gap increase does not lead to the expected Voc increase. In contrast, the Jsc of most cells reported in Figure 6.7 follows the expected trend. The difference between experimental and

6.2 AI –BIII –C2VI Device Properties

30

η [%]

25 20 15 10 5 1.0

Ag(In,Ga)Se2 CuGaSe2

0.9 Voc [V]

Cu(In,Ga)S2 0.8 0.7

Cu(In,Ga)Se2

Cu(In,Ga)Se2 Cu(In,Al)Se2

Cu(In,Al)Se2

0.6 0.5

CuInS2

CuInSe2

Fill factor [%]

90 80 70 60

Jsc [mA cm−2]

50 40 30 20 10

1.0

1.2 1.4 1.6 Bandgap energy [eV]

1.8

Figure 6.7 Experimental solar cell parameters of I III VI AI –BIII –CVI 2 -based solar cells as a function of the A –B –C2 bandgap. Dotted line indicates the Shockley–Queisser limit, dashed line represents a guide to the eye assuming a Voc proportional to the bandgap but normalized to the best CuIn1−x Gax Se2 cell with x = 0.3.

287


theoretical Jsc is mainly due to optical losses. Only for the CuGaSe2 cell there are strong electronic losses in addition to a reduction in the Jsc , as will be shown below. We emphasize that the trends in Figure 6.7 reflect the technological status rather than fundamental principles – although some fundamental trends have been formulated in Section 5.2.5. 6.2.2 Diode Currents

If we compare the diode currents of highly efficient CuIn1−x Gax Se2 cells in the dark and under illumination, no indication for a change in the recombination mechanism from dark to illumination is obtained. Highly efficient cells are well behaved where a plot of log(J + Jsc ) as a function of V–JRs exhibits a constant slope within two current decades [7]. Comparing different efficient cells shows the expected trend of an improved performance due to a reduced diode quality factor [7, 858]. The diode quality factor is lowest for CuIn1−x Gax Se2 cells with low Ga content (x ≤ 0.3), as can be seen in Figure 6.8a. Also up to this Ga content, the saturation current density in Figure 6.8b shows the expected behavior: J0 decreases with increasing bandgap. All cells of Figure 6.8 are grown Cu-poor with x (Eq. (4.5)) between −0.1 and −0.2 [7]. The activation energy of the saturation current under illumination for Cu-poor CuIn1−x Gax Se2 cells is plotted in Figure 6.9. It can be seen that for all Cu-poor cells, the activation energy approximately equals the bandgap. Let us first consider cells with x ∼ 0.3 which give the highest efficiency. From Section 2.4.5.4, we know that an activation energy equal to Eg,a does not exclude interface recombination since, without FLP, Ea for interface recombination 10−8 2.2 2.0 1.8 1.6 1.4

10−10 10−11

1.2 1.0

(a)

10−9 J0 [A cm−2]

Diode quality factor A

288

10−12

0.0 0.2 0.4 0.6 0.8 1.0 x in Culn1-x BIIIxSe2

(b)

Figure 6.8 (a) Diode quality factor of CuIn1−x BIII x Se2 cells as a function of x: (+) CuIn1−x Gax Se2 (data from Ref. [859]), (∗ ) CuIn1−x Gax Se2 (data from Ref. [7, 858]), (#) CuIn1−x Alx Se2 (data from Ref. [860]).

0.0 0.2 0.4 0.6 0.8 1.0 x in Culn1-x BIIIxSe2

(b) Saturation current density J0 as a function of x in CuIn1−x Gax Se2 . Data from Ref. [7, 858]. The dashed lines in (a) and (b) only indicate interpreted trends.


Activation Energy Ea [eV]

1.8 1.6

Cu-poor x=0 x = 0.25 , , , y=0

1.4 x ~ 0.25 1.2 Cu-rich

1.0

x=0 0.8

1.0

1.2 1.4 1.6 Bandgap Eg,a [eV]

1.8

Figure 6.9 Activation energy of the saturation current Jo as a function of the bandgap in CuIn1−x Gax Se2−y Sy cells prepared from Cu-poor and Cu-rich absorbers and with a CdS/ZnO emitter. The data from Ref. [859]

•

() and Ref. [861] ( ) were obtained from Jsc (Voc ) plots at varying temperatures. The data from Ref. [862] (♦) were obtained from J(V) analysis under illumination.

(IFR) equals the interface bandgap Eg,IF . But what is the interface bandgap? In Figure 4.25 we see that the interface bandgap of CuIn1−x Gax Se2 /CdS at the position x ∼ 0.3 exceeds the absorber bandgap. Thus, if interface recombination limits the devices with x ∼ 0.3, the activation energy of J0 should rather reflect a position of FLP. Indeed as the A factor is close to unity, FLP (which requires A = 1) is a possibility. In that case Ea = Ep,az=0 would be similar to Eg,a . We use Eq. (2.116) and calculate Sp0 = 2 × 107 cm/s as the electron interface recombination velocity which is necessary in order to give a J0 of 2 × 10−12 A cm−2 . This value is very high and already at the upper limit of Sp0 = vp (see Section 2.4.5.4). We therefore dismiss the possibility of IFR for highly efficient cells with x ∼ 0.3 and conclude that cells with an A factor close to 1.0 are limited by recombination in the QNR including the back contact. In that case, Ea corresponds to the bulk band gap. With A = 1.13 and Eg,a = 1.15 eV we calculate J00 = 3 × 105 A cm−2 from Eq. (2.48). Using the expression for J00 from Table 2.3 with Dn,a = 0.5 cm2 s−1 d and NA,a = 1016 cm−3 we arrive at ηc (z) z=−wa = 0.26 µm−1 (and the reciprocal dz ≈3.7 µm). This is about the slope of the ηc (z) curve for CuInSe2 in Figure 6.10. For CuIn1−x Gax Se2 /CdS cells with x > 0.3, we see in Figure 6.8b that the expected trend of logarithmically decreasing J0 with increasing x is violated. Instead J0 increases with x and the currently best CuGaSe2 cell has J0 = 10−9 A cm−2 [417]. Also the A factor increases with x and reaches A ≈ 2 for CuGaSe2 . However, Ea (for Cu-poor cells) continues to equal Eg,a (Figure 6.9). If we regard Figure 4.25, where it is Eg,a − Eg,IF ≈ 0.3 for CuIn1−x Gax Se2 /CdS devices with x > 0.5, interface recombination for the Cu-poor CuGaSe2 cell must be ruled out because otherwise Ea would equal Eg,IF . That interface recombination with Eg,IF < Eg,a

289

290


does not dominate the recombination is somewhat surprising. In order to stay below the observed J0 with Eg,IF = 1.4 eV and A = 2.0, the interface recombination velocity must be below Sn0,p0 = 2 × 103 cm s−1 . However, with Ea = Eg,a and Ea > Eg,IF we have to assume that the CuGaSe2 is limited by recombination in the bulk and, in particular, in the SCR. With Eg,a = 1.67 eV and A = 2.0, the reference current density J00 = 2 × 105 A cm−2 is calculated from Eq. (2.48). We use Eq. (2.79) with Fm = 105 V cm−1 and τn0 = τp0 = τn0,p0 and estimate τn0,p0 = 5 × 10−12 s as the carrier lifetime necessary to explain the large J0 . As the A factor is temperature-dependent, recombination may be enhanced by tunneling [864] in which case a somewhat larger lifetime can produce the observed saturation current density. In any case, the lifetime of current CuGaSe2 layers must be many orders of magnitude smaller than that of CuIn0.7 Ga0.3 Se2 layers. Indeed the defect density was found to be much larger (Section 4.2.3.3) and the diffusion length much smaller (Section 6.2.3). If, by any means, the lifetime of CuGaSe2 could be increased or the interface recombination Sn0,p0 increased, we would encounter interface recombination at the CuGaSe2 /CdS interface and a reduced activation energy of the diode current. Perhaps this is already seen in Figure 6.9 where one data point for Cu-poor CuGaSe2 represents a measurement of Ea = 1.19 eV [862]. In Figure 6.9 there is the group of solar cells which have been grown Cu-rich and which all have Ea < Eg,a . Thus, their diode current is limited by recombination at a position where it holds Eg,a = Ea or at a position where the Fermi level is pinned at Ea (see Table 2.4). The cells from CuInSe2−y Sy absorbers (Figure 6.9, ) have the end member CuInS2 with Eg,a = 1.52 eV which has been investigated in more detail. In CuInS2 solar cells, the diode quality factor is A ≈ 1.5 which excludes FLP. Under illumination, the activation energy of J0 is about 0.3 eV below Eg,a . This in accordance with a measured negative conduction band offset at the CuInS2 /CdS interface [865, 866]. Thus, it appears that the Cu-rich CuInS2 cell in Figure 6.9 exhibits interface recombination as the dominant recombination mechanism. However, in the darkness one finds Ea = Eg,a [867] which has not been explained so far. About the other cells with label Cu-rich in Figure 6.90 much less is known. Equally, interface recombination may be surmised as the reason for the bandgap independent activation energy. Solar cells based on CuIn0.7 Ga0.3 Se2−y Sy (Figure 6.9, ) absorbers appear to have an increased interface bandgap according to Figure 6.9. This gives way to larger open circuit voltages for those Cu-rich cells as has been reported in Ref. [868].

•

6.2.3 Collection Function

The collection function, ηc (z), in CuIn1−x Gax Se2−2y S2y solar cells has been investigated by EBIC experiments [68, 415, 869, 870]. The following results have been obtained: the diffusion length in CuIn1−x Gax Se2 cells with low Ga content (x < 0.3) well exceeds 1.5 µm [415, 870, 871]. For highly efficient CuInSe2 based cells, carrier collection up to the back contact is reported [869] which may reflect passivation of the back contact or a back surface field. For the CuInSe2 based cell in

6.2 AI –BIII –C2VI Device Properties Figure 6.10 Collection functions of (a) CuInSe2 , (b) CuGaSe2 , and (c) CuInS2 − based cells as a function of depth. All cells have a ZnO window and CdS buffer. Redrawn from Ref. [863]. Cells are not identical to Figure 6.11.

ηc (z) [a.u.]

(a)

(c) (b) −3.0

−2.0

−1.0

0.0

Position z [µm] 100 EQE [%]

80 60 (b) (c)

40

(a)

20 0

400

600 800 1000 Wavelength [nm]

1200

Figure 6.11 External quantum efficiency curves of (a) Mo/CuIn0.7 Ga0.3 Se2 /CdS/ZnO solar cell (reproduced after Ref. [7]), (b) Mo/CuGaSe2 /CdS/ZnO solar cell (reproduced after Ref. [417]), and (c) Mo/CuInS2 /CdS/ZnO solar cell (reproduced after Ref. [877]).

d Figure 6.10 we find ηc (z) z=−wa ≈ (6.5 µm)−1 . As expected, the diffusion length dz is largely independent of the sample temperature between 123 and 373 K [872]. The space charge region typically is below 0.5 µm [872]. A back surface gradient by the Ga depth profile increases the collection function at high depths as indicated by the long wavelength EQE [873]. We note in passing that determination of the collection function in graded bandgap absorbers is difficult. EBIC experiments have indicated that the collection function is improved by a Ga gradient [874]. With Ln,a ≈ da ≈ 3 µm and anticipated diffusion constant of Dn,a ≈ 0.5 cm/s, we calculate the carrier lifetime τn0 ≈ 180 ns which is in good agreement with measured values [429]. CuIn1−x Gax Se2 cells with x ≥ 0.75 have a substantially decreased diffusion length with values below 0.2 µm. Thus, the wide bandgap CuGaSe2 cells to date have poor charge collection properties. In contrast, wide bandgap CuInS2 cells from Cu-rich absorbers also show Ln,a > 1 µm [863] but with much smaller lifetimes (Section 4.2.3.6). For Cu-rich absorber we would anticipate a larger diffusion constant than for Cu-poor. The collection of charge carriers in CuIn0.7 Ga0.3 Se2 is largely homogeneous on a lateral scale and grain boundary recombination effects must be small [78, 869, 875]. This is a prerequisite to achieve a high external quantum efficiency in a wide wavelength interval. Indeed, the external quantum efficiency of the Mo/CuIn0.7 Ga0.3 Se2 /CdS/ZnO cell (Figure 6.11, curve a) reaches 96.5% at 824 nm

291

292


and stays at high values up to long wavelengths. However, the influence of free carrier absorption at long wavelength cannot be neglected [876] which makes the simulation of quantum efficiency curves more involved. Curve c in Figure 6.11 reveals that, among the wide bandgap chalcopyrites, CuInS2 exhibits superior collection properties. 6.2.4 Transient Effects and Metastability

Transient effects in CuInSe2 based devices have been reported for a long time. For early heterojunction solar cells of the type CuInSe2 /CdS, Ruberto and Rothwarf reported on a transient increase in Voc with duration of white light illumination [60]. More bias conditions have since then been found to induce transient effects. In the following, we will differentiate red light effect, forward bias effect, blue light effect, and reverse bias effect. In a sense, also the state produced by particle radiation is a metastable one. Table 6.2 gives an overview of the transient effects in CuIn1−x Gax Se2 absorbers and solar cells under different bias conditions. But first, let us discuss the relaxed state. 6.2.4.1 Relaxed State The starting point of any transient experiment is a CuIn1−x Gax Se2 solar cell or absorber layer which has been stored in the dark at elevated temperature (330 K) for many hours (typically overnight). This brings the cell into the relaxed state. Inherently, the relaxed state of the cell is difficult to analyze by an illuminated JV curve because illumination will change the status of the cell. Likewise, it is Table 6.2

Summary of observed transient effects in Cu(In, Ga)Se2 films and solar cells.

Primary bias condition

Secondary bias condition

Primary parameter

Derived parameter

References

Red light

T ≈ 300 K

C σ Nd

NA,a − ND,a NA,a − ND,a –

[878] [27] [380]

Blue light

Voc , T ≈ 300 K

FF

–

[59]

White light

Open circuit

Voc , FF, C, σ

–

[60, 878, 879]

Forward bias

Dark, T ≈ 300 K

Voc C Nd,p

– NA,a − ND,a –

[60] [880] [369, 881]

Reverse bias

Dark, T ≈ 300 K

C FF Nd (0.35 eV) –

NA,a − ND,a – – Ed (N1 )

[880] [882] [883] [884]


difficult to test the relaxed state by capacitance–voltage (C-V) measurements due to the influence of electrical bias. Experimentally, we may approximate the relaxed state by measuring a JV curve under red light illumination at low temperature (Eg,b (buffer layer bandgap) [59]. Low-temperature blue light illumination has a smaller effect [885]. Obviously, temperature activation is required. From the photon energy dependence it was concluded by Eisgruber et al., that the effect takes place in the buffer layer or is mediated by the buffer layer [59]. Blue light illumination completely removes the kink in the JV curve of a relaxed cell. It gradually reduces the kink of a preceedingly reverse biased cell, although here the effect is not complete. For Eg,b = 2.4 eV, the photon flux density with hν > Eg,b in a standard AM1.5 spectrum is about Jγ = 4 × 1016 cm−2 . This flux is sufficient to induce the blue light effect. The improvement in FF by blue light is persistent for some minutes at room temperature [59]. The blue light effect can be resembled by irradiation of the cell’s interface region by electrons, for example, in an electron microscope. Kniese et al. report about a barrier for the EBIC in the relaxed state of the device. This barrier is persistently removed upon irradiating the heterojunction interface of the ZnO/CdS/ CuIn1−x Gax Se2 solar cell with electrons [893]. 6.2.4.6 White Light Effect Many light soaking experiments have been performed with the standard AM1.5 spectrum because this is what the module will see in the field. But phenomenologically white light induces a combination of blue and red light effects (Table 6.2). Voc and FF of the device will increase. Persistent photoconductivity is observed on the CuIn1−x Gax Se2 absorber films. White light bias leads to an overall capacitance increase, in that respect the red light effect prevails. In general, the white light effect is more pronounced for Cd-free buffer layers. A point which is not sufficiently understood so far [894]. White light illumination sets the cell into the high performance state. A model of this state in the form of a band diagram can be seen in Figure 6.12. No barriers for photo or diode currents are active in this state. 6.2.4.7 Reverse Bias Effect Typically, a voltage bias of −1 V applied for 1 h at 300 K puts a sample in the reverse bias state. In that state we will observe a decrease in fill factor. In extreme cases, a hysteresis curve can be formed by incremental and decremental voltage sweeps. It will be observed that the FF is higher if the JV curve is measured from positive to negative bias. The reverse bias effect is temperature activated and becomes ineffective below 300 K [884]. The characteristic of a poor FF we have already mentioned for the relaxed state of a cell. One may ask if the reverse bias treatment only brings the sample back to the relaxed state or into a new metastable state. The latter appears to be the case as revealed by C-V measurements. The capacitance of a reverse biased cell is larger than that of the relaxed state [892]. Space charge density profiles show an increased net acceptor density in the CuIn1−x Gax Se2 bulk as well as toward the interface [880, 882, 892]. Reverse bias DLTS shows a defect level at 0.35 eV in the reverse biased sample. Annealing at 340–360 K brings the cell back into the


relaxed state. Thermally stimulated capacitance measurements show that this takes place with an activation energy of 0.3 eV [882]. Note that in an earlier publication, this activation energy has been interpreted as the emission energy of a deep trap [894]. The reverse bias effect can be a problem for solar modules which are partially shaded. The shaded cells of a series integrated module will be under reverse bias. Fortunately, mild annealing (340 K) together with or followed by white light exposure brings the cells into the high performance state. The reverse bias metastability is more expressed for chemically deposited buffer layers [894]. In addition, it is more expressed for buffer layers other than CdS [895]. However, it can be persistently reduced by air annealing [894]. 6.2.4.8 Models for Metastability Igalson and Schock first proposed that metastability in Cu(In, Ga)Se2 has its origin in one or more defects which induce strong lattice relaxation upon occupation [369]. Later, Lany and Zunger [362] argued that the red light, the blue light and the reverse bias effect can be explained by virtue of metastability of the (VSe –VCu ) complex (cf. Section 4.2.3.2). We will summarize the findings of Lany and Zunger by use of Figure 6.12.

• Deep in the absorber, the majority of (VSe –VCu ) complexes – due to a low lying Fermi level – is in the stable donor configuration (VSe –VCu )+ . Generation of electrons by red light photons (red light effect) or electron injection (forward bias effect) induces the reaction in Eq. (4.9). A share of the available defect complexes transforms into (VSe –VCu )− and thereby releases free holes. Due to a small energy barrier this transformation is temperature activated [880]. The (VSe –VCu )− configuration is metastable because the Fermi level is below the transition energy (−/+) in Figure 4.14 (see also graph c in the schematic picture ε(−/2−) ε(+/−) d(2−/−)

E

d(−/+) z

Figure 6.12 Schematic representation of the absorber bandgap as a function of distance from the heterointerface (from Ref. [362] with permission). The dashed lines give the (+/−) and (−/2−) transition energies of the (VSe − VCu ) complex (cf. Figure 4.14). In the spatial interval (0, d2−/− ), the complex is stable in the −2 state. In the interval (d2−/− , d−/+ ) the −1 state, that is, the

acceptor configuration, is stable while the +1 state is metastable. The metastable +1 state can be produced by thermal activation together with capture of two holes according to Eq. (4.11). Beyond the position d−/+ , the complex is stable in the donor configuration +1. This state can be transformed into the metastable −1 state by electron capture.

297

298


of Figure 4.13). The complex can return to the stable state via reaction Eq. (4.11) by thermal activation and simultaneous hole capture. • Close to the heterointerface, the acceptor configuration of the (VSe –VCu ) complex is stable, that is, we find a large density of (VSe –VCu )− . Very close to the interface even the (VSe –VCu )2− state can be populated increasing the space charge in a narrow range of the absorber. This high negative charge forms a p+ layer and leads to a barrier for the photo current [885], as is exemplarily explained in Figure 2.39c. Upon reverse bias, the region of high negative charge further extends thereby increasing the barrier for the photo current (reverse bias effect). In addition, defect complexes in the depletion region are brought into the acceptor configuration in which they remain metastably after bias removal. To the contrary, blue light provides holes – generated either in the buffer or in the absorber interface layer – for the transformation of (VSe –VCu )− into (VSe –VCu )+ (blue light effect). Close to the interface, the (VSe –VCu )+ configuration is the metastable one. The reduced negative charge upon blue light illumination removes the photo current barrier. Another explanation of the reverse bias effect invokes the influence of Cu migration [286]. Theoretical calculations [410] as well as experiments [469] have shown that for small En the formation enthalpy of the VCu can become negative, that is, vacancies of Cu spontaneously form. Thus, we have the reaction CuCu → VCu − + Cui + . Under the influence of the reverse bias electric field Cui + ions move away from the junction, leaving behind VCu − acceptors. Either these vacancies or their function in the (VSe –VCu ) complex may contribute to the increased negative charge near the interface upon reverse bias treatment. Another explanation for the reverse bias effect use the interpretation that the activation energy of the N1 defect displays the value of En,az=0 . Experimentally, it was found that the activation energy of N1 differs in the various states [885]. Thus, a different potential distribution may be an additional source of metastable changes of the cell under reverse bias. 6.2.4.9 Implications for Module Testing For testing Cu(In1−x Gax )Se2 modules indoors, device metastability effects have to be regarded. As we can expect, light soaking was found to be necessary in order to close the gap between the higher values of Voc and FF measured outdoors and the indoor measurement. Useful soaking parameters are 20 min at 40 ◦ C under 800 W/m2 [896]. As measurement under standard test conditions requires cooling the module to 25 ◦ C after light soaking, we expect that during this 15–30 min cooling time in darkness there is a gradual return to the relaxed state of the device. Although better than without light soaking, even the light-soaked values of Voc and FF are still inferior to outdoor data [896] – a problem which can hardly be avoided. After light soaking, the use of a pulsed solar simulator (flasher) appears to pose no problems provided the time of the voltage sweep is not below 1 ms [896]. We conclude that the metastabilities of Cu(In1−x Gax )Se2 modules do not exclude measuring their efficiency with an accuracy of about 2% – provided that


Energy [eV]

4 3 2 1 0 −3

−2

−1 −0.2 Position z [µm]

0.2

Figure 6.13 Model band diagram of a Cu(In0.7 Ga0.3 )Se2 solar cell with parameters from Table 8.4. The layers are (from left to right) graded CIGSe absorber (2 µm), ungraded CIGSe absorber (1 µm), OVC (30 nm), CdS (50 nm), HR-ZnO (50 nm), LR-ZnO (200 nm). The zero of energy is deliberately set to the Ev,ZnO .

light soaking is employed. Light soaking today is part of the IEC 61646 test protocol. 6.2.5 Device Model

A high-efficiency Cu(Inx Ga1−x )Se2 solar cell has a Mo/MoSe2 back contact and a CdS/ZnO front contact. In order to make a model band diagram, some data required – such as doping concentrations and dielectric constants – are known. Some others can be inferred from physical arguments. For the Cu(In1−x Gax )Se2 layer we assume the light-soaked state where the doping density is NA,a = 1016 cm−3 . We refrain from programming special features in order to reflect the transient effects listed in the former section. In Table 8.4 we give the baseline parameters of a Cu(In0.7 Ga0.3 )Se2 solar cell where the In and Ga composition is only valid in the first one-third of the absorber. These parameters shall be representative for the metastable state induced by white light illumination (see Section 6.2.4). The baseline parameters define a 20% efficient cell with Voc = 0.718 V, Jsc = 35.6 mA cm−2 , and FF = 81.4% as obtained by device simulation. The simulated model band diagram in equilibrium can be seen in Figure 6.13. What are the critical parameters for this model band diagram? • Optical losses. Ref. [34] showed that, for an absorber thickness of da = 3 µm, optical reflection and absorption of and within the emitter layers n-ZnO, i-ZnO, and CdS is a major factor limiting the short circuit current density. Optical loss at the back side of such a thick absorber can be neglected. Thus, the critical parameters in Table 8.4 for optical losses are front contact reflection, front contact shading, ZnO thickness and CdS thickness. They mainly determine Jsc . The low window reflection of 2% assumed in Table 8.4 can be realized

299

300


by an antireflection coating with 105 nm MgF2 [34]. Three percent shading can be realized by photolithographic patterning of the front grid. Free carrier absorption in the transparent conducting oxide (here ZnO) depends on the (necessary) doping density (see Figure 5.14). The loss can be limited by a minimum ZnO thickness. The optical loss in the CdS layer also depends on the thickness. However, the negative effect of recombination in the CdS layer is partly outweighed by the positive effect of optical matching between ZnO and Cu(In0.7 Ga0.3 )Se2 [781]. What remains a free parameter is the recombination in CdS. • Recombination in CdS. Engelhardt et al. [75] and Orgassa [34] showed that the charge carrier collection probability in the CdS layer is between 30 and 40%. Thus, some photons absorbed in the CdS layer contribute to the photo current. In the model of blue light metastability (see Section 6.2.4), holes generated in the buffer layer compensate negative charge in an absorber defect layer close to the interface. In Table 8.4 an acceptor defect density of 1.5 × 1016 cm−3 in CdS provides a 30% collection probability. • Absorber. The simulated efficiency critically depends on the electron lifetime of the Cu(In0.7 Ga0.3 )Se2 layer. With an assumed defect density of 5 × 1012 cm−3 , capture cross-section of 10−13 cm−2 and electron thermal velocity of 107 cm/s, an electron lifetime of the Cu(In0.7 Ga0.3 )Se2 layer of 100 ns is implemented. This lifetime value is in accordance with experimental values discussed in Section 6.1.3. It is valid in the first one-third of a graded absorber or in the complete thickness of an absorber without grading. An electron lifetime of 100 ns and an electron mobility of 50 cm2 (Vs)−1 define a diffusion length of 3.6 µm which complies with the collection function discussed in Section 6.2.3. The simulated JV curve exhibits a saturation current density of 10−12 A cm−2 which very good agrees with experimental data of nearly 20% efficient cells [816, 858]. The simulated diode quality factor at room temperature is A = 1.1 being also in good agreement with the experiment [858]. The simulated A factor shows a slight increase toward lower temperature. Therefore, the activation energy of J0 was determined by aid of Eq. (2.186). For the simulated cell we find Ea = 1.22 eV, a value which corresponds to the absorber bandgap of 1.15 eV plus the expected extra term of 3 kT0 (see footnote 22 in Chapter 2). The positive impact of decreasing J0 and A on the device performance was experimentally verified in Ref. [858]. It is important to note that with an electron lifetime of 100 ns, the cell Voc is mainly limited by recombination in the QNR of the absorber provided the buffer/absorber interface is inactive (we come to this point below). • Back surface gradient. With a diffusion length of >2 µm, the cell performance depends on recombination both in the QNR and at the back contact. Therefore, a compromise between Rules 9 and 10 would apply. High-efficiency Cu(In1−x Gax )Se2 solar cells exhibit a Ga gradient within about the last two-thirds of the absorber [803]. We added a conduction band gradient that extends two-thirds of the absorber thickness. This is similar to reported gradients in Refs. [110, 121, 803]. The x parameter in Cu(In1−x Gax )Se2 increases from 0.2 to 0.4, which according to Table 4.12 produces a bandgap widening from 1.15


•

•

•

• •

to 1.34 eV. This bandgap gradient is realized by an increase in Ga content in the absorber. The optical bowing parameter of 0.2 (see Table 4.12) is neglected for the sake of simplicity. No band bending at the back surface is assumed. The nominal back surface recombination velocity is Sn = 104 cm s−1 . With Cc of 2 × 1011 cm−2 eV−1 , we put a value which compensates the positive impact of the band gap gradient. It renders an electron lifetime of 5 × 10−9 s in the graded region of the absorber. If we remove the absorber back surface gradient, we largely arrive at the same solar cell parameters. ODC layer. The bandgap of the absorber front surface is taken as 1.4 eV. It is widened with respect to the bulk bandgap by a lowering of the valence band maximum. A bandgap widening at the surface has been reported in Refs. [522, 897] and is our interpretation of the interface band gap depicted in Figure 4.25. An interface band gap widening may also be caused by S diffusion in the Cu(In0.7 Ga0.3 )Se2 layer. The ODC layer still lacks some experimental evidence, for example, by TEM analysis. The ODC layer or any other source for a lowering of the valence band maximum relaxes the constraint in the value of Ep,az=0 for the device performance. Band offsets. Small conduction band offsets between ZnO and CdS (Eb,w = c = 0.1 eV) and the absorber have been −0.1 eV) as well as between CdS (Ea,b c introduced. These values are a compromise between different experimental and theoretical data. Niemegeers et al. [79] gave upper limits for the band offsets i-ZnO/CdS and CdS/ODC from admittance spectroscopy which are −0.18 and +0.11 eV, respectively. Since together both band offsets cancel out, that is, χ = 0, their impact on Ep,az=0 is null. We note, however, that even conduction band offsets of +0.1 eV (ZnO/CdS) and −0.1 eV (CdS/ODC, that is, 0.2 eV deeper band edges of CdS) as found in Refs. [522, 532] would not alter the device performance and would not make interface recombination the dominant mechanism because of the lower VBM of the ODC. CdS/ODC interface. The cell in Table 8.4 exhibits two types of interface defects at the CdS/ODC interface. (i) A defect in the middle of the interface bandgap of the CdS/ODC interface provides recombination centers with density Nd2,IF . Lying between the demarcation levels, these defects define a nominal interface recombination velocity of Sn0 = 105 cm/s, a value which is two orders of magnitude below the thermal velocity. No information about defect densities in the interface bandgap are available but due to the large Ep,az=0 (see next point) there is a large tolerance for Sn0 . (ii) High-lying donor states pin the electron quasi Fermi level at the interface and lead to a large value of Ep,az=0 . According to the Cahen–Noufi model (see Section 5.2.7) the donor states may arise from chalcogen vacancies. We note that due to the ODC layer, the cell shows similar solar cell parameters without FLP-even in this case the value of Ep,az=0 is large enough (for Sn0 = 105 cm/s). i-ZnO/CdS interface. This interface shall only contain acceptor states of such low density that no additional band bending occurs [777]. Doping densities. The ZnO is formed from an undoped layer and a ZnO : Al or ZnO : Ga doped layer. The doping level of ZnO : Al or ZnO : Ga is around

301

302


5 × 1020 cm−3 but in Table 8.4 was set to 1018 cm−3 as simulation programms use Boltzmann approximation. That of the undoped layer of typical thickness 50 nm is not exactly known. The measured conductivity of that layer is σ < 10−4 ( cm)−1 . The typical doping level of Cu(Inx , Ga1−x )Se2 is 1016 cm−3 which corresponds to the illuminated state (see Section 6.2.6). • Back contact. Little is known about the specific electronic properties of a MoSe2 /Cu(In1−x Gax )Se2 back contact. We neglected the MoSe2 layer in our device model and used flat band conditions.

6.2.6 Stability

First, we discuss the stability of the chalcopyrite based heterostructure against dry heat stress. The majority of chalcopyrite solar cells are stable in 85 ◦ C dry heat [898, 899]. Only CuGaSe2 cells [898] as well as Cu(In, Ga)(Se, S)2 mini-modules with chemical vapor deposited ZnO [900] exhibited dry heat degradation. Remarkably, CuIn1−x Gax Se2 cells with x = 0.3 can withstand 300 ◦ C in vacuum for 30 min [901]. Above this temperature, diffusion of Cd or Zn from the buffer to the absorber layer takes place [902] reducing the carrier concentration in the absorber [901]. This result proves that chemical instability is not an issue for chalcopyrite solar cells [903]. Under illumination, chalcopyrite cells and modules show an initially increase in efficiency [904]. The main effect is an increase in Voc (light soaking effect as described in Section 6.2.4) by several millivolts. Vice versa, we must state that chalcopyrite devices degrade in darkness. Although fully reversible under illumination, this dark degradation is a problem for testing and energy rating [905]. The light soaking effect is minimal for high-performance cells with a CdS buffer layer. It is considered a problem for Zn(OH,S) buffer layers [785]. Today, the IEC test protocol allows light soaking prior to the performance measurement. Also remarkable for chalcopyrite devices is the performance stability against fluxes of high-energy particles as prevalent in space orbits. Space testing [906] as well as ground earth simulations [907, 908] proved the radiation hardness of Cu(In, Ga)Se2 and CuInS2 cells being superior to GaAs or Si cells. Although particle bombardment induces an increase in the defect density N2 of Cu(In, Ga)Se2 (see Section 4.2.3.3), these defects will partly anneal out already at operation temperature in space (∼350 K) [908, 909]. The second important aspect is the damp-heat stability. Non-encapsulated devices typically degrade by fill factor and, to a smaller degree, by Voc [898, 904, 910, 911]. A 1000 h stress test at 85 ◦ C and 85% relative humidity typically leads to more than 50% degradation [912]. For CuIn1−x Gax Se2 cells, degradation was found to be minimal for x = 0.4 [898]. The fill factor degradation is due to an increased series resistance [913]. This can be caused by increases in ZnO lateral resistivity, in ZnO/Mo contact resistance and in Mo sheet resistance. Using specific test structures [914] it was found that


the increase in ZnO lateral resistivity is stronger for absorber layers with higher roughness [915, 916]. Rougher absorbers lead to different grain boundaries in ZnO which impede the carrier transport [512]. The increase in ZnO/Mo contact resistance is more severe if the high-resistance ZnO is in contact with Mo instead of the low-resistance ZnO [914]. According to Section 5.2.10, performing the P2 scribe after deposition of the high-resistance window would be advantageous although less practical. The increase in Mo sheet resistance is simply due to Mo corrosion [682, 916]. Degradation of Voc has been explained by an increased defect density and increased width of the space charge region [917]. In particular, the density of the N2 defect (see Section 4.2.3.3) detected by DLTS and admittance spectroscopy increases after damp heat stress. By application of appropriate encapsulation and sealing (see Section 5.2.10) several Cu(In, Ga)Se2 and CuInS2 based modules have passed the IEC 61646 test. It should be noted, however, that the damp/heat testing conditions, although leading to very high water concentration in the first 10 cm from the edges, leaves the center of a 60 × 120 cm2 module almost dry. Therefore it is important to ask for the experience in the field. Cu(In, Ga)Se2 modules and systems have been monitored using outdoor installations (Table 6.3). Below 0.2%/year degradation has been observed for early Siemens Solar Industries (SSI) modules over a period of more than 15 years. This is an important finding as it proves that there is no inherent degradation mechanism. Later SSI modules exhibit higher degradation. Wuerth Solar modules show degradation of not more than around 0.5%/year – although the monitoring period is shorter. Detailed analysis of the module JV curves revealed Table 6.3

Results of outdoor performance monitoring of Cu(In,Ga)Se2 modules.

Year of Monitoring Initial deployment period efficiency (years) (%)

Degradation Manufacturer Monitoring (% year−1 )

References

1992 1993 1995 1999 2001 2002 2003

16.2 15.2 13.2 9.2 6 5.7 5

7.95 0.16 ± 0.05 8.69 0.36 ± 0.05 9.75 1.05 ± 0.05 11.6 2.29 ± 0.05 Not reported 0.42 ± 0.13 10.9 0.64 ± 0.05 9.2 0.0 ± 0.05

Module, SRC Module, SRC Module, SRC Module, SRC Module, PR Module, SRC Module, outdoorc

[913] [913] [913] [913] [918] [913] –

2004

6

10.3

System, PR

[919, 920]

SSIa SSI SSI SSI Not reported NREL (B)b Wuerth Solar –0.08 ± 0.13 Wuerth Solar

PR = performance ratio; SRC = standard reporting conditions. SSI, Siemens Solar Industries. b Manufacturer not reported. c 25 ◦ C at 1000 W insolation. a

303

304


that the primary loss factor of the SSI modules is the increase in series resistance which leads to deterioration of the FF. In contrast, the Voc and Isc data are mostly stable over time. The type B module of the NREL test bed (see Table 6.3) shows an increase in shunt conductance and almost no increase in series resistance. The only complete solar system reported in Table 6.3 did not show degradation in performance ratio over a four year period.

305

7 Appendix A: Frequently Observed Anomalies During solar cell development, one may encounter device anomalies which are caused by non-optimum material properties or non-optimum cell design. The following collection of anomalies and explanatory models is by no means complete but rather lists the effects addressed in this book.

7.1 JV Curves 7.1.1 Roll Over Effect

J V

The JV curve (dark or light) shows current saturation at an electric forward bias. In general, the saturation is due to a barrier for the diode current. 1) Back contact barrier according to Figure 2.29. The forward current of the main diode is limited by the saturation current of the back contact diode. Upon increasing electric forward bias, the forward current saturates, exhibiting a roll over shape. The roll over effect shows up in dark and light JV curves. If the saturation current of the light current is larger, there is in addition a cross over anomaly [119]. 2) Acceptor states at buffer/window interface. If the electron barrier is fixed due to FLP at the buffer/window interface and ϕb n is large enough (see band diagram in Figure 2.6c), the diode current is totally blocked. Only photo current can pass the barrier. The effect is a roll over anomaly of the light JV curve where the saturation current is close to zero. 3) Positive conduction band offset at the buffer/window interface. Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices. Roland Scheer and Hans-Werner Schock Copyright  2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-31459-1

306

7 Appendix A: Frequently Observed Anomalies

7.1.2 Cross Over

J

V

Dark and light JV curves cross at a large electric forward bias. 1) There is an electron barrier φb n in the conduction band (see Figure 2.39b). Upon photodoping of the buffer layer, the potential drop over the buffer and thus the electron barrier is reduced. Then, the diode current is larger under illumination [59]. 2) An electron barrier can also form due to a high density of acceptors in the absorber’s near interface region (Figure 2.39c). The large negative charge in the acceptor states leads to a large potential drop over the buffer layer. This forms the electron barrier φb n in the dark. Under illumination, the absorber acceptors are filled with holes which are photogenerated in the buffer layer. As a result, the acceptor charge is reduced and so is the electron barrier φb n [79]. 3) If the cross over comes along with a roll over anomaly, the cross over may be due to a minority carrier recombination at a back contact junction [119]. Under illumination, the diode current is larger than in the dark. 7.1.3 Kink in Light JV Curve

J

V

A kink type anomaly in the light JV curve occurs in the third or fourth quadrant of the JV curve. It is an extreme form of a voltage-dependent photo current. All models given below are based on a barrier for the photo current. 1)

Due to a positive conduction band offset Ec b,a in the system absorber/ buffer/window (see band diagram in Figure 2.39b). In the case of low interface recombination velocity at the absorber/buffer interface electrons accumulate, leading to an inversion of dEFn /dz and a flow of electrons back to the absorber bulk. In the absorber bulk, the electrons recombine. This can take place for Sn0 = 103 cm/s at the absorber/buffer interface. In the case of higher interface

7.1 JV Curves

recombination velocity, the photo current recombines at the interface for intermediate forward bias. This anomaly can occur jointly with the anomaly given in Section 7.4.2. 2) Due to a thin p+ layer at the absorber front surface. This leads to a band diagram as illustrated in Figure 2.39c. Depending on the doping of the p+ layer, the kink occurs in the third quadrant (higher doping) or in the fourth quadrant (lower doping). The reason for the photo current diminution is a sign reversal of dEFn /dz deeper in the absorber. This anomaly can come along with an IQE showing a low red response (see Section 7.4.4). 3) Conduction band gradient at the absorber surface. It also produces a barrier for the photo current similar to (1). The current passing the barrier by thermionic emission depends on the electron concentration at the barrier which reduces upon forward bias.

7.1.4 Violation of Shifting Approximation

J

V Models

The shifting approximation states that the light JV curve is obtained by shifting the dark JV curve by Jsc (see Section 2.4.1). The photo current and thus the collection function η(V) is voltage-dependent. In Section 2.5.7 we list reasons for a voltage-dependent collection function. If the voltage dependence of the collection function is taken into account, the superposition principle can be conserved. 2) Due to an illumination-dependent shunt path in a polycrystalline absorber layer there may be secondary phases connecting the back contact with the front contact. If these secondary phases are photoconductive, the shunt path may drain a higher current under illumination. 3) The diode current under illumination deviates from the diode current in the dark at low voltages. Reasons for this anomaly may be a saturation of the SRH recombination rate [921], a large series resistance changing the boundary condition for QNR recombination [31], a change from low to high injection condition under illumination or a change from interface recombination to SCR recombination due to a light modulated potential barrier [834]. 1)

307

308


7.2 Solar Cell Parameters 7.2.1 Reduced Jsc but High Voc

With reduced Jsc we mean the current density under short circuit (see Figure 2.31b). However, Voc roughly equals the rule of thumb Voc = Eg,a /q − 0.5 V. 1) Barrier for the photo current. If a kink in the light JV curve (see Section 7.1.3) lies at negative voltage bias, the short circuit current is small. A much higher photo current is observed at negative voltage bias. 2) High recombination of photogenerated minority carriers at the interface. If bulk minority carriers remain minorities at the interface, majority carriers are abundant at the interface. An example is a p+ -absorber/n-window heterostructure. The collection function as given in Eq. (2.160) can be small.

7.2.2 Reduced Voc but High Jsc

A high Jsc means that the current density under short circuit approaches the expected current after Figure 2.31b. However, Voc is considerably smaller than after the rule of thumb Voc = Eg,a /q − 0.5 V. 1) High defect zone of limited extension near the position n = p. An example is a planar grain boundary located near n = p (see Section 3.7). Due to the limited extension of the zone, recombination of the photo current is small at zero voltage bias. Thus, there is a high Jsc . The diode current at forward bias, however, can be large and can reduce Voc . Columnar grain boundaries also provide high recombination at n = p (see again Section 3.7). At the grain interior, their impact on photo current recombination is low if the distance of grain boundaries is large compared to the absorption length. This case should show the J0 activation of the absorber bandgap unless the Fermi level is pinned at the grain boundary. Furthermore, this case should show a Fermi level splitting in the absorber bulk which is larger than qVoc . 2) A special case of (1) is interface recombination. The high defect zone is directly at the interface. The potential distribution decides whether at the interface the condition n = p is fulfilled (see Section 3.3). Recombination, however, also increases if the bandgap is reduced (see also Section 3.2). This case also should come along with a Fermi level splitting which is much larger than Voc . 3) Back contact barrier (see Section 3.8). This comes along with a roll over of the dark and light JV curves. 4) Nonhomogeneity of absorber bandgap (see Section 2.4.9) and thus fluctuation of the activation energy of the dominant recombination process. This also mostly reduces Voc and not Jsc . See (3) of Section 7.3.4.

7.3 Diode Parameters

7.2.3 High Jsc but Low FF

A high Jsc means that the current density under short circuit approaches the expected current after Figure 2.31b. Anomalies listed in Section 7.1.4 result in a high Jsc and a low FF. Anomalies listed in Section 7.1.3 have the same effect provided that the kink is in the fourth quadrant. The kink may exhibit small curvature due to lateral variation of the barrier for the photo current. 2) A large diode quality factor of the recombination mechanism predominant around the maximum power point results in a low FF. 1)

7.3 Diode Parameters 7.3.1 Diode Parameter A > 2

Diode parameter A exceeds the limiting value for SRH recombination of 2.0 in a certain voltage range. This is due to a recombination mechanism which gradually saturates with increasing electric forward bias. 1)

Tunneling enhanced surface or SCR recombination at large intrinsic fields (see Table 2.2). 2) Recombination via coupled defect levels at weak intrinsic fields. Schenk and Krumbein showed that for diodes ideality factors exceeding 2.0 can result from a rapid charge transfer between donors and acceptors of high density where at least one of these levels is shallow [922]. 3) Donor–acceptor pair recombination where saturation is due to the limited number of recombination partners [923]. 4) Fluctuation of the activation energy of the dominant recombination process. See (3) of Section 7.3.4. 7.3.2 Activation Energy Ea < Eg,a

The activation energy of the diode current evaluated after Eq. (2.48) is smaller than the bandgap of the absorber, Eg,a . 1) Due to Ec < 0, recombination takes place at the interface. The activation energy equals Eg,IF = Eg,a − Ec where Eg,IF is the interface bandgap. 2) Fermi level pinning at the absorber/emitter interface. The activation energy is Ea = φb p = Ep,az=0 .

309

310


3) Band bending in the absorber is very steep. The barrier may be partly traversed by tunneling. (It is not clear how to describe the temperature dependence and thus how to extract the activation energy.) 4) Due to nonhomogeneity of the absorber bandgap (see Section 2.4.9). 7.3.3 Diode Quality Factor Illumination Dependent

1) In the case of interface recombination, the A factor depends on the ratio of the built-in voltages in emitter and absorber (see Section 2.4.5.4). Interface states which may become charged under illumination can change the band diagram. Thus, the A factor is different in the dark and under illumination. See also Ref. [924] (illumination with red light). 2) In extreme cases of photodoping, the dominant recombination mechanism may change. In the example of Figure 6.5b, the value of Ep,az=0 increases under illumination. Thus, interface recombination with A ≈ 2(Ep,az=0 is close to Eg,IF /2) may change to bulk recombination. If the latter has A ≈ 1 (QNR recombination), then the diode quality factor is strongly illumination-dependent. 7.3.4 Diode Quality Factor Temperature-Dependent

A temperature-dependent diode quality factor may occur in the complete investigated temperature interval or only at lower temperature. 1) As can be seen in Table 2.2, the diode quality factor becomes temperaturedependent for SCR recombination on a defect distribution Dd (E). 2) Tunneling enhanced recombination either in the SCR or at the interface can be the reason for A(T) (Table 2.2). 3) Fluctuation of the activation energy of the dominant recombination process. In particular for interface recombination with Fermi level pinning the activation energy (φb p = Ep,az=0 ) may be strongly fluctuating. Such a cell should be analyzed based on Eq. (2.134). If such a cell is analyzed based on Eq. (2.48) instead, a temperature-dependent diode quality factor A (T) will be diagnosed. Equating the arguments of Eqs. (2.48) and (2.134) and performing simple algebra delivers: −1 σE2 a Ea 1 − A (T) = (7.1) Ea A 2Ea A2 kT where A is the true diode quality factor which is constant and determined by the type of the recombination process. E a is the mean activation energy which may not be known. A increases with decreasing temperature. The curvature of A is less than for tunneling enhanced recombination.

7.4 Quantum Efficiency

7.4 Quantum Efficiency 7.4.1 High Jsc but Low EQE

With low EQE we mean a quantum efficiency which if integrated using Eq. (2.165) is lower than the measured Jsc . Often the quantum efficiency curve shows a small maximum value. 1) Barrier for the photo current which is large under low light intensity or monochromatic illumination but becomes lowered by photodoping of the buffer at AM1.5 illumination [215]. 2) Large number of micro shunts. If the cell is irradiated only on a limited area (as in QE measurement), the non-illuminated part acts as a shunting load. Seen from the active solar cell, this load is in parallel with the input resistance of the QE current amplifier. Although the total shunt resistance of the cell (measured as a macroscopic quantity by JV analysis) is high, the local shunt resistance seen from the active solar cell part may be small. Thus, the current is drained throughout the shunting load. Reducing the cell area to the area of illumination may increase the QE. 7.4.2 Low Jsc but High EQE

With low Jsc we mean a short circuit current which is smaller than the one calculated from the integrated EQE using Eq. (2.165). 1) Barrier for the photo current. Due to a limited thermionic emission current after Eq. (2.165), the small current density of a quantum efficiency measurement can pass the barrier. In contrast the high current density under AM1.5 illumination can not pass the barrier because −Jsc (jE 100 ) is larger than J0,TE . 7.4.3 Low Blue Response in IQE

IQE

λ

The IQE curve is inclined toward small wavelengths. This is due to small values of the collection function close to the interface absorber/emitter.

311

312


1) Due to hole recombination. If the location with p = n is deep in the absorber, holes are minority carriers in the region close to the interface to the emitter. The holes can recombine either at the interface (where they are minorities) or recombine while drifting toward the back of the absorber (see also Section 2.5.4.4). 7.4.4 Low Red Response in IQE

IQE

λ

The IQE curve is inclined toward large wavelengths. This is due to a small collection function in the bulk of the absorber. 1)

Small lifetime of those carriers which are minorities in the absorber bulk. The absorber doping is such as to provide the location of n = p close to the junction. Otherwise, a low blue response and a low red response occur simultaneously and the maximum IQE is small. 2) High minority carrier lifetime but thin absorber with high recombination velocity at the back contact. 3) Due to p+ layer at the absorber surface. This leads to a kink in the JV curve (see Section 7.1.3) and – for a large barrier – to a reduced Jsc . In the IQE, one finds quantum collection in the small wavelength range. Charge carriers generated deeper in the absorber experience the positive slope of En,a (see also Figure 2.39c). 7.4.5 Quantum Efficiency Low at All Wavelengths

The internal quantum efficiency curve at zero bias is small at all wavelengths. 1) A high series resistance or a back contact barrier puts the main junction under forward bias. As a result, the width of the space charge region is reduced or interface recombination sets in. 2) Due to low doping, holes (as minority carriers generated by the blue photons near the heterojunction) and electrons (as minority carriers generated by the red photons close to the back surface) have to drift a long paths toward their majority type regions. If the carrier lifetime is small, the total IQE is small. 3) A kink in the light JV curve is found at zero bias (see Section 7.1.3) and there is a barrier for the photo current.

7.5 Transient Effects

4) The local shunt density is high [see (2) in Section 7.4.1]. 7.4.6 Apparent Quantum Efficiency

1

EQE

λ

The quantum efficiency measured under voltage bias may become larger than 1.0 in a certain wavelength interval. We speak of an apparent quantum efficiency (AQE). If measured using a lock-in technique, a value larger than 1.0 can occur for negative and positive QE. 1)

2)

Negative AQE. If there is a cross over effect which is induced by light of a certain wavelength interval, the photoconductivity of the element absorbing in this wavelength interval may be the reason for the cross over. Examples are shown in Figures 2.12 and 2.13 for photoconductive absorber and buffer, respectively. In both cases, the reduction of the conduction band barrier under illumination leads to an increased forward current under illumination and thus to a cross over effect as well as a negative AQE at high forward bias. In the case of a photoconductive buffer (Figure 2.13), the AQE increase occurs for photon absorption in the buffer. Positive AQE. If the dominant recombination mechanism changes under illumination within the wavelength interval, then a positive AQE may be detected at forward bias. An example is Figure 6.5b where the cell changes from interface domination to bulk domination upon illumination with light absorbed in the buffer layer.

7.5 Transient Effects 7.5.1 Voc Time-Dependent with dVoc /dt > 0

After switching on the AM1.5 illumination, Voc rises with time. This is due to a reduction in carrier recombination. 1)

The SCR shrinks due an increased absorber doping. Consequently, the cumulated recombination rate becomes reduced (see Section 3.5). If dopant formation in the absorber is temperature-activated, the time constant for this process depends on the activation energy and can be large.

313

314


2) Recombination is reduced due to increased absorber doping. The built-in voltage increases and thus the recombination decreases. 3) IF recombination decreases due to reduced absorber doping. Similar to (1), this can be due to a temperature-activated compensating dopant formation in the absorber. The consequence is a time-dependent band diagram (see also Section 3.3) with time-dependent decrease of potential drop in the emitter. 7.5.2 Voc Time-Dependent with dVoc /dt < 0

After switching on the AM1.5 illumination, Voc falls with time. This is due to an enhancement in carrier recombination resulting from the inverse effects of Section 7.5.1.

315

8 Appendix B: Tables

Table 8.1 Default values of physical quantities for window, buffer, and absorber semiconductors. General properties T (K) Vn (cm s−1 ) Vp (cm s−1 ) Contact properties Sno (cm s−1 ) Spo (cm s−1 ) ϕb (eV) Reflectivity Transport model

300 107 107 Front 107 107 0 0 Thermal emission

Layer properties d (µm) Eg (eV) ε Nc (cm−3 ) Nv (cm−3 ) µn (cm2 V−1 s−1 ) µp (cm2 V−1 s−1 ) ND/A (cm−3 ) Nd (cm−3 )(Gauss) Ed (eV) σd (full width half maximum) σn (cm s−1 ) σp (cm s−1 )

Window 0.5 3.40a 9 4 × 1018 9 × 1018 50 20 ND : 1018 A : 1016 1.7 0.05

Buffer 0.05 2.40a 10 2 × 1018 2 × 1019 50 20 ND : 1.1 × 1016 A : 1016 1.2 0.05

Absorber 1.5 1.20a 12 2 × 1018 2 × 1018 50 20 NA : 1016 D : 1014 0.4 0.05

10−15 10−12

10−15 10−12

10−13 10−15

Interface properties Ec (eV) σn (cm s−1 ) σp (cm s−1 ) ND,IF (cm−2 ) Ed,IF Transport model

Window/buffer 0 10−12 10−15 1010 Eg,b /2 Thermal emission

Window/absorber 0 10−12 10−15 109 Eg,a /2 Thermal emission

Buffer/absorber 0 10−12 10−15 109 Eg,a /2 Thermal emission

Back 104 107 0 0 Thermal emission

a Parameters of optical absorption according to Eq. (4.2): α0 = 104 cm−1 , α1 = 103 cm−1 , B1 = 0.4 eV, E1 = Eg .


316

Experimental ∆Ev for thin film substrate

Layer 1

Layer 2

Mismatch factor asub /aovl − 1(%)

CuInS2

ZnS

2.3

–

0.16

–

ZnSe

–2.3

–

–0.1

–

CdS

–5.0

–

–0.3

ZnS

7.3

–

–0.46

ZnSe ZnTe

2.6 –

–0.1 [515] –

–0.1 –

0.7 [515] –

– –

0.7 [515] –

CdS CdSe

–0.6 –4.2

–0.04 [515] –

0.05 –

0.83 [519] –

– –

1.07 [515] –

– –

– 0.28 [514]

CuInSe2

CdTe –10.6 CuInS2 4.7

Strain induced Strain induced shift d∆Ev shift d∆Ev from ab initio after Eq. (4.16) calculation

– –

– –

Experimental ∆Ev for crystalline substrate

0.6 [517] –

– –

–

Theoretical ∆Ev from ab initio calculation –

0.7 [516]

–

1.2 [518]

–

–

–

Calculated ∆Ev from transitivity rule

Calculated ∆Ev after Eq. (4.15)

0.95 [162, 514, 515] 0.42 [162, 514, 515] 0.79 [162, 514, 515] 1.23 [162, 514, 515] – –0.03 [162, 514, 515] – 0.65 [162, 514, 515] 0.08 –

0.58 0.01 0.46 1.3 0.73 – 1.18 0.78 0.37 0.72

(continued overleaf)


Heterocontacts between chalcogenide substrates and chalcogenide overlayers. The lattice mismatch was calculated according to Tables 4.7, 4.13 and 4.14, where for the ternary semiconductors a mean lattice constant of a/2 + c/4 was used. Sx (XA − XB ) was calculated using Miedema values of electronegativity as taken from Ref. [228], where the anion electronegativity was neglected, and an S factor of 0.3 eV per Miedema unit was introduced. Strain contribution to Ev was from ab initio calculation [513], strain contribution to Ev according to Eq. (4.16), experimental Ev for crystalline substrates, experimental Ev for polycrystalline substrates, theoretical Ev from ab initio calculations, theoretical Ev from transitivity rule, modeled Ev after Eq. (4.14), modeled Ev after Eq. (4.15) without interface dipole and strain effect. Table 8.2

Table 8.2

Layer 1

CuGaSe2

CuIn3 Se5

(Continued) Layer 2

Mismatch factor asub /aovl − 1(%)

CuInTe2

–6.2




–

–

CuGaSe2 3.5

–

–

–

–

CuAlSe2

–

–

–

–

CuIn3 Se5 1.3

–

–

–

–

ZnS

–3.7

–

0.24

–

–

ZnSe

–0.9

–

0.04

–

–

CdS

–3.9

–

0.23

CdS

–1.8

–

0.1

–

0.86 [453]

ZnS

– – 6

– – –

– – –0.42

– – –

0.8 [521] 0.96 [522] 0.62 [453]

– –1.7

– –

– 0.1

– –

2.3 [523] 0.88 [524]

–

–

–

–

0.98 [524]

ZnO Cu(In0.7 Ga0.3 )3 CdS Se5 CuGa3 Se5 CdS

–

0.93 [520]

–


0.5 [162, 514, – 515] 0.04 [162, 514, – 515] 0.26 [162, 514, – 515] 0.34 [162, 514, – 515] – 1.2 [162, 514, 515] – 0.68 [162, 514, 515] – 1.03 [162, 514, 515] – 0.73 [162, 514, 515] – – – – – 0.89 [162, 514, 515] – – – – –

–

0.2 0.18 0.5 – 1.12 0.55 1.0 – – – – – – –

(continued overleaf)

317

–



–

Theoretical ∆Ev from ab initio calculation

318

(Continued)

Layer 1

Layer 2 Mismatch factor asub /aovl − 1(%)




Theoretical ∆Ev from ab initio calculation



CdTe

ZnS ZnSe ZnTe CdSzb CdSwz CdSe ZnO TiO2 SnO2 In2 O3 SiO2 V2 O3 Te ZnO

– – – –0.19 – – – – – – – – – – – –

– – – – – – – – – – 4.7 [528] – – – – –

– – 0.1 0.94 [525] – – 1.9 [226] 2.6 [527] 2.2 [226] 2.1 [226] – 1.5 [226] 0.5 [529] 1.2 [530] 0.96 [531] 1.2 [532]

– – – 0.99 [162] 0.78 [526] 0.57 [162] – – – – – – – – – –

1.15 [514] 0.62 [514] –0.29 [514] – – – – – – – – – – – – –

0.93 0.36 –0.12 0.81 0.81 0.41 1.92 – – – – – – 1.11 – –

CdS

SnO2 a Calculated

20 14.7 6.2 11.4 10.8a – – – – – – – – – – –

√ after dsub /dolv − 1 with dwz = a and dzb = a/ 2.

– – – – – – – – – – – – – – – –


Table 8.2

8 Appendix B: Tables Table 8.3 Defect and doping parameters for the simulation of a TCO/SnO2 /CdS/CdTe/BC solar cell in Figure 6.5.

General properties Vn (cm s−1 ) Vp (cm s−1 ) Rs ( cm2 ) Rp ( cm2 )

107 107 1 5000

Contact properties Sn0 (cm s−1 ) Sp0 (cm s−1 ) φb (eV) Reflectivity Transport model

Front 107 107 0.1 0.06 Thermal emission

Layer properties d (µm) Eg (eV) ε/ε0 Nc (cm−3 ) Nv (cm−3 ) µn [cm2 (Vs)−1 ] µp [cm2 (Vs)−1 ] ND/A (cm−3 ) Nd (cm−3 ) Ed (eV) σn (cm s−1 ) σp (cm s−1 )

TCO 0.5 3.60 9 4 × 1018 2 × 1019 100 25 ND : 1018 A : 1016 Mid-gap 10−15 10−12

Interface properties Ec (eV) σn (cm s−1 ) σp (cm s−1 ) Nd1,IF (cm−2 ) Ed1,IF Transport model

TCO/CdS 0.0 – – – – –

Data partly extracted from Refs. [833, 834].

Back 107 107 0.4 0.8 Thermal emission CdS 0.025 2.40 10 1 × 1018 2 × 1019 100 25 ND : 1.1 × 1017 A : 1017 Mid-gap 10−17 10−12

CdTe 4 1.49 10 8 × 1017 5 × 1018 320 40 NA : 2 × 1014 D : 5 × 1013 Mid-gap 10−12 10−13

CdS/CdTe 0.0 10−12 10−13 D : 1 × 1011 Mid-gap Thermal emission

319

320


Semiconductor and device parameters for the band diagram model of a CuIn0.7 Ga0.3 Se2 based solar cell.

Table 8.4

General properties Vn (cm s−1 ) Vp (cm s−1 ) Rs ( cm2 ) Rp ( cm2 ) Contact properties Sno (cm s−1 ) Spo (cm s−1 ) φb (eV) Reflectivity Absorption Transport model Layer properties d (µm) Eg (eV) dmin,back (µm) Ec ,back (eV) Cc (cm−2 eV−1 ) ε/ε0 Nc (cm−3 ) Nv cm−3 ) µn [cm2 (Vs)−1 ] µp [cm2 (Vs)−1 ] ND/A (cm−3 ) Nd (cm−3 ) Ed (eV) σn (cm s−1 ) σp (cm s−1 ) Interface properties Ec (eV) Nd1,IF (cm−2 ) Ed1,IF eV σn (cm s−1 ) σp (cm s−1 ) Nd2,IF (cm−2 ) Ed2,IF eV σn (cm s−1 ) σp (cm s−1 ) Transport model

107 107 0.22 5000 Front

Back

107 107 0 0.02 0.03 Thermal emission

104 107 0 0.6 0.007 Thermal emission Culn0.7 Ga0.3 Se2 3 1.15 1 0.19 2 × 1011 13.6 7 × 1017 1.5 × 1019 20 20 NA : 1016 D : 5 × 1012 Mid-gap 10−13 10−15 OVC/ CuIn0.7 Ga0.3 Se2 0 – – – – – – – – –

n-ZnO

i-ZnO

CdS

OVC

0.2 3.40 – – – 9 4 × 1018 9 × 1018 20 20 ND : 1018 A : 1016 Mid-gap 10−15 10−12 ZnO/CdS

0.05 3.30 – – – 9 4 × 1018 9 × 1018 20 20 ND : 1.01 × 1018 A : 1018 Mid-gap 10−15 10−12 –

0.05 2.40 – – – 10 4 × 1018 9 × 1018 20 20 ND : 4 × 1015 A : 3 × 1015 Mid-gap 10−15 10−12 –

0.03 1.35 – – – 13.6 7 × 1017 1.5 × 1019 20 20 NA : 1.1 × 1016 D : 1016 Mid-gap 10−15 10−12 CdS/OVC

−0.1 – – – – – – – – –

– – – – – – – – – –

– – – – – – – – – –

0.1 D : 1013 Ev,b + 2.15 10−12 10−15 A : 1010 Mid-gap 10−15 10−12 Thermal emission

321

References 1. Shay, J.L., Wagner, S., and Kasper,

2.

3.

4.

5.

6.

7.

8.

9.

H.M. (1975) Efficient CuInSe2 /CdS solar cells. Appl. Phys. Lett., 27(2), 89. Kazmerski, L.L., White, F.R., and Morgan, G.K. (1976) Thin film CuInSe2 /CdS heterojunction solar cells. Appl. Phys. Lett., 29(4), 268. Mickelsen, R.A. et al. (1984) Polycrystalline thin-film CuInSe2 /CdZnS solar cells. IEEE Trans. Electron. Devices, ED-31(5), 542. Gabor, A.M. et al. (1994) High efficiency CuInx Ga1-x Se2 solar cells made from (Inx Ga1-x )2 Se3 precursor films. Appl. Phys. Lett., 65(2), 198. Sugimoto, H. et al. (2009) Impact of Cu(InGa)(SeS)2 absorber quality and circuit uniformity on improved efficiency; application of photoluminescence and electroluminescence techniques. Proceedings of the 24th European Photovoltaic Solar Energy Conference, p. 2465. Mitchell, K. et al. (1987) High efficiency thin film tandem PV modules. Proceedings of the 19th IEEE Photovoltaic Specialist Conference, p. 13. Repins, I. et al. (2008) 19.9%-efficient ZnO/CdS/ CuInGaSe2 solar cell with 81.2% fill factor. Prog. Photovoltaics: Res. Appl., 16, 235. Wu, X. et al. (2001) 16.5%-efficiency CdS/CdTe polycrystalline thin film solar cell. Proceedings of the17th European Photovoltaic Solar Energy Conference, Stephens & Associates, Munich, p. 995. Cusano, D.A. (1963) CdTe solar cells and photovoltaic heterojunctions in

10.

11.

12.

13.

14.

15.

16.

17.

II–VI compounds. Solid State Electron., 6, 217. Bonnet, D. and Rabenhorst, H. (1972) New results on the development of a thin-film p-CdTe-n-CdS heterojunction solar cell. Proceedings of the 9th IEEE Photovoltaic Specialist Conference 1972, p. 129. Tyan, Y.S. and Perez-Albuerne, E.A. (1982) Efficient thin film CdS/CdTe solar cells. Proceedings of the 16th IEEE Photovoltaic Specialist Conference, IEEE, San Diego, p. 794. Britt, J. and Ferekides, C. (1993) Thin-film CdS/CdTe solar cell with 15.8% efficiency. Appl. Phys. Lett., 62(22), 2851. Zweibel, K. (2007) The Terawatt Challenge for Thin Film PV, in Thin Film Solar Cells (ed. J. Poortmans), John Wiley & Sons, Ltd, Chichester, pp. 427–462. Keshner, M.S. and Arya, R. (2004) Study of Potential Cost Reductions Resulting from Super-Large-Scale Manufacturing of PV Modules, NREL. Fthenakis, V.M., Kim, H.C., and Alsema, E. (2008) Emissions from photovoltaic life cycles. Environ. Sci. Technol., 42(6), 2168. Anderson, B.A. (2000) Materials availability for large-scale thin-film photovoltaics. Prog. Photovoltaics: Res. Appl., 8, 61. Schubert, B.-A. et al. (2008) An economic approach to evaluate the range of coverage of indium and its impact on indium based thin-film solar cells – recent results of Cu2 ZnSnS4 (CZTS) based solar cells. Proceedings of the


322

References

18.

19. 20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

23rd European Photovoltaic Solar Energy Conference, p. 3788. Tauc, J. (1962) Photo and Thermoelectric Effects in Semiconductors, Pergamon Press, Oxford. Wuerfel, P. (2004) Physics of Solar Cells, Wiley-VCH Verlag GmbH, Berlin. Franciosi, A. and Van de Walle, C.G. (1996) Heterojunction band offset engineering. Surf. Sci. Rep., 25, 1–140. Persson, C. et al. (2005) n-type doping of CuInSe2 and CuGaSe2 . Phys. Rev. B, 72, 035211. Fonash, S.J. (1981) Solar cell device physics, in Energy Science and Engineering (ed. J. Denton), Academic Press, New York, p. 45. Sze, S.M. (1981) Physics of Semiconductor Devices, Wiley-Interscience, New York. Donelly, J.P. and Milnes, A.G. (1967) The capacity of p-n heterojunctions including the effects of interface states. IEEE Trans. Electron. Devices, ED-14, 63. Mönch, W. (1995) Semiconductor Surfaces and Interfaces, 2nd edn, Vol. 26, Springer Series in Surface Sciences (ed. G. Ertl), Springer, Berlin, Heidelberg. Rau, U. et al. (1999) Oxygenation and air-annealing effects on the electronic properties of Cu(In,Ga)Se2 films and devices. J. Appl. Phys., 86(1), 497. Meyer, T. et al. (1999) A model for the open circuit voltage relaxation in Cu(In,Ga)Se2 heterojunction solar cells. Eur. Phys. J. Appl. Phys., 8, 43. Shiau, J.-J., Fahrenbruch, A.L., and Bube, R.H. (1986) Interpretation of capacitance versus voltage measurements in the presence of a high density of deep levels. J. Appl. Phys, 59, 2879. Blakemore, J.S. (1962) Semiconductor statistics, in Semiconductors (ed. H.K. Henisch), Pergamon Press, Oxford. Lang, D.V. (1974) Deep-level transient spectroscopy: a new method to characterize traps in semiconductors. J. Appl. Phys, 45(7), 3023. Lindholm, F.A., Fossum, J.G., and Burgess, E.L. (1979) Application of the superposition principle to solar-cell analysis. IEEE Trans. Electron. Devices, 26(3), 165.

32. Shockley, W. and Queisser, H.J. (1961)

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

Detailed balance limit of efficiency of p-n junction solar cells. J. Appl. Phys., 32(3), 510. Werner, J.H., Mattheis, J., and Rau, U. (2005) Efficiency limitations of polycrystalline thin film solar cells: case of Cu(In,Ga)Se2 . Thin Solid Films, 480/481, 399–409. Orgassa, K. (2004) Coherent Optical Analysis of the ZnO/CdS/Cu(In,Ga)Se2 Thin Film Solar Cell, Shaker Verlag, Aachen. Shimakawa, S. et al. (2006) Characterisation of Cu(In,Ga)Se2 thin films by time-resolved photoluminescence. Phys. Status Solidi A, 203(11), 2630. Metzger, W.K. et al. (2003) Time-resolved photoluminescence studies of CdTe Solar cells. J. Appl. Phys., 94(5), 3549. Altermatt, P.P. et al. (1996) Spatially resoved analysis and minimization of resistive losses in high-efficiency Si solar cells. Prog. Photovoltaics, 4, 399. Bube, R.H. (1992) Photoelectronic Properties of Semiconductors, University Press, Cambridge. Dziewior, J. and Schmid, W. (1977) Auger coefficients for highly doped and highly excited silicon. Appl. Phys. Lett., 31(5), 346. Benz, G. and Conradt, R. (1977) Auger recombination in GaAs and GaSb. Phys. Rev. B, 16(2), 843. Kerr, M.J., Cuevas, A., and Campbell, P. (2003) Limiting efficiency of crystalline silicon solar cells due to Coulomb-enhanced Auger recombination. Prog. Photovoltaics: Res. Appl., 11, 97. Pankove, J.I. (1971) Optical Processes in Semiconductors, Dover Publications, New York. Shockley, W. and Read, W.T. (1952) Statistics of the recombinations of holes and electrons. Phys. Rev., 87(5), 835. Hall, R.N. (1952) Electron-hole recombination in germanium. Phys. Rev., 87, 387. Sah, C.T., Noyce, R.N., and Shockley, W. (1957) Carrier generation and recombination in p-n junctions and p-n

References

46.

47.

48.

49.

50.

51.

52.

53.

54.

55.

56.

junction characteristics. Proceedings of the IRE 45, IRE, p. 1228. Walter, T., Herberholz, R., and Schock, H.W. (1996) Distribution of defects in polycrystalline chalcopyrite thin films. Solid State Phenomena, 51-52, 309–316. Rau, U. et al. (2000) Electronic loss mechanisms in chalcopyrite based heterojunction solar cells. Thin Solid Films, 361/362, 298–302. Hurkx, G.A.M., Klaassen, D.B.M., and Knuvers, M.P.G. (1992) A new recombination model for device simulation including tunneling. IEEE Trans. Electron. Devices, 39(2), 331. Rau, U. (1999) Tunneling-enhanced recombination in Cu(In,Ga)Se2 heterojunction solar cells. Appl. Phys. Lett., 74(1), 111. Clugston, D. and Basore, P. (1997) PC1D version 5: 32-bit solar cell modeling on personal computers, Conference Record, 26th IEEE Photovoltaic Specialists Conference, IEEE, Anaheim, CA, p. 207. Pauwels, H. and Vanhoutte, G. (1978) The influence of interface states and energy barriers on the efficiency of heterojunction solar cells. J. Phys. D, 11, 649. Scheer, R. (2009) Activation energy of heterojunction diode currents in the limit of interface recombination. J. Appl. Phys., 105, 104505. Wilhelm, H., Schock, H.W., and Scheer, R. (2010) Interface recombination in heterojunction solar cells: influence of buffer layer thickness. J. Appl. Phys., submitted. Roussillon, Y. et al. (2004) Back contact and reach-through diode effects in thin-film photovoltaics. J. Appl. Phys., 96, 7283. Pan, J., Gloeckler, M., and Sites, J.R. (2006) Hole current impedance and electron current enhancement by back-contact barriers in CdTe thin films solar cells. J. Appl. Phys., 100, 124505. Rhoderick, E.H., Frensley, W.R., and Shaw, M.P. (1993) Properties of junctions and barriers, in Handbook on Semiconductors (ed. T.S. Moss), Elsevier Science Publishers, Amsterdam.

57. Niemegeers, A., Burgelman, M., and

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

68.

Vos, A.D. (1995) On the CdS/CuInSe2 conduction band discontinuity. Appl. Phys. Lett., 67(6), 843. Demtsu, S.H. and Sites, J.R. (2006) Effect of back-contact barrier on thin-film CdTe solar cells. Thin Solid Films, 510, 320–324. Eisgruber, I.L. et al. (1998) Blue-photon modification of nonstandard diode barrier in CuInSe2 solar cells. Sol. Energy Mater. Sol. Cells, 53, 367–377. Ruberto, M.N. and Rothwarf, A. (1987) Time-dependent open circuit voltage in CuInSe2 /CdS solar cells: theory and experiment. J. Appl. Phys., 61(9), 4662. Rau, U., Grabitz, P.O., and Werner, J.H. (2004) Resistive limitations to spatially inhomogeneous electronic losses in solar cells. Appl. Phys. Lett., 85(24), 6010. Rau, U. and Werner, J.H. (2004) Radiative efficiency limits of solar cells with lateral band-gap fluctuations. Appl. Phys. Lett., 84(19), 3735. Hulstrom, R., Bird, R., and Riordan, C. (1985) Spectral solar irradiance data sets for selected terrestrial conditions. Solar Cells, 15, 365. Harbecke, B. (1986) Coherent and incoherent reflection and transmission of multilayer structrures. Appl. Phys. B, 39, 165. Brendel, R. (2003) Thin-Film Crystalline Silicon Solar Cells, Wiley-VCH Verlag GmbH, Weinheim. Donolato, C. (1985) A reciprocity theorem for charge collection. Appl. Phys. Lett., 46(3), 270. Green, M.A. (1997) Generalized relationship between dark carrier distribution and photocarrier collection in solar cells. J. Appl. Phys., 81(1), 268–271. Scheer, R. (1995) Electron-beam induced current profiles for thin film heterojunction analysis, Polycrystalline Semiconductors IV – Physics, Chemistry and Technology, Trans Tech Publications, p. 527.

323

324

References 69. Dugas, J. (1994) 3D modelling of a

70.

71.

72.

73.

74.

75.

76.

77.

78.

79.

reverse cell made with improved multicrystalline silicon wafers. Sol. Energy Mater. Sol. Cells, 32, 71–88. Dugas, J. (1996) Modelling of Thin Polycrystalline Silicon Solar Cells on Low Temperature Substrates. Proceedings of the 4th international Conference on Polycrystalline Semiconduction, Scitec Publications, Zürich-Vetinov, p. 521. Donolato, C. (1989) An alternative proof of the generalized reciprocity theorem for charge collection. J. Appl. Phys., 66(9), 4525. Reichman, J. (1980) Collection efficiency of low-mobility solar cells. Appl. Phys. Lett., 38(4), 251. Hecht, K. (1932) Zum Mechanismus des lichtelektrischen Primärstromes in isolierenden Kristallen. Zeitschrift Phys., 77, 235. Rothwarf, A., Phillips, J., and Wyeth, N.C. (1978) Junction field and recombination phenomena in the CdS/Cu2 S Solar cell: Theory and experiment. Proceedings of the 13th IEEE Photovoltaic Specialists Conference, IEEE, New York, p. 399. Engelhardt, F. et al. (1999) Cu(In,Ga)Se2 solar cells with a ZnSe buffer layer: interface characterisation by quantum efficiency measurements. Prog. Photovoltaics: Res. Appl., 7, 423. Tuominen, E. et al. (1997) A method extracting solar cell parameters from spectral response by inverse Laplace transform. Phys. Scr., T69, 306. Konovalov, I. and Breitenstein, O. (2001) An iterative algorithm for determining depth profiles of collection probability by electron-beam-induced current. Semicond. Sci. Technol., 16, 26. Scheer, R. (1999) Qualitative and quantitative analysis of thin film heterostructures by EBIC in Polycrystalline Semiconductors V (eds J.H. Werner, H.-P. Strunk, and H.W. Schock), Scitech Publications, Uettikon am See, pp. 57–68. Niemegeers, A. et al. (1998) Model for electronic transport in Cu(In,Ga)Se2

80.

81.

82.

83.

84.

85.

86.

87.

88.

89.

solar cells. Prog. Photovolatics: Res. Appl., 6, 407–421. Gärtner, W.W. (1959) Depletion-layer photoeffects in semiconductors. Phys. Rev., 116(1), 84. Lampert, M.A. (1964) Volumecontrolled current injection in insulators. Rep. Prog. Phys., 27, 329. Pallares, J. et al. (2006) A compact equivalent circuit for the dark current-voltage characteristics of nonideal solar cells. J. Appl. Phys., 100, 084513. Niemegeers, A., Gillis, S., and Burgelman, M. (1998) Interpretation of capacitance spectra in the special case of novel thin film CdTe/CdS and CIGS/CdS solar cell device structures, in Proceedings of the 2nd World Photovoltaic Solar Energy Conference (ed. H. Ossenbrink), Stephens, Vienna, p. 1071. Viktorovitch, P. and Moddel, G. (1980) Interpretation of the conductance and capacitance frequency dependence of hydrogenated amorphous silicon Schottky barrier diodes. J. Appl. Phys., 51(9), 4847. Nicollian, E.H. and Goetzberger, A. (1965) MOS conductance technique for measuring surface state parameters. Appl. Phys. Lett., 7(8), 216. Karpov, V.G., Compaan, A.D., and Shvydka, D. (2004) Random diode arrays and mesoscale physics of large-area semiconductor devices. Phys. Rev. B, 69, 045325. Hegedus, S.S. (1997) Current-voltage analysis of a-Si and a-SiGe solar cells including voltage-dependent photo current collection. Prog. Photovoltaics: Res. Appl., 5, 151. Phillips, J.E., Titus, J., and Hofmann, D. (1997) Determining the voltage dependance of the light generated current in CuInSe2 based solar cells using I-V measurements mode as different light intensities. Proceedings of the 26th IEEE Photovoltaic Specialists Conference, IEEE, Anaheim, CA, p. 463. Luque, A. and Hegedus, S. (eds) (2003) Handbook of Photovoltaic Science and Engineering, John Wiley & Sons, Ltd, Chichester.

References 90. Mialhe, P., Khoury, A., and Charles,

91.

92.

93.

94.

95.

96.

97.

98.

99.

100.

101.

102.

J.P. (1984) A review of techniques to determine the series resistance of solar cells. Phys. Status Solidi A, 83, 403. Sites, J.R. and Mauk, P.H. (1989) Diode quality factor determination for thin-film solar cells. Solar Cells, 27, 411. Werner, J.H. (1988) Schottky barrier and pn-junction I/V plots – small signal evaluation. Appl. Phys. A, 47, 291–300. Eron, M. and Rothwarf, A. (1984) Effects of a voltage-dependent light-generated current on solar cell measurements: CuInSe2 /Cd(Zn)S. Appl. Phys. Lett., 44(1), 131. Bauer, G.H. et al. (2005) Quasi-Fermi level splitting and identification of recombination losses from room temperature luminescence in Cu(In1-x Gax )Se2 thin films versus optical band gap. Thin Solid Films, 480/481, 410–414. de Vos, A. (1983) The fill factor of a solar cell from a mathematical point of view. Solar Cells, 8, 283. Palmer, D.W. (1999) Characterisation of semiconductor heterostructures by capacitance methods. Microelectron. J., 30, 665–672. Zhota, Y. (1973) Frequency dependence of C and dV/dC ˆ -2 of Schottky barriers containing deep impurities. Solid State Electron., 16, 1029. Blood, P. and Orton, J.W. (1992) The Electrical Characterization of Semiconductors: Majority Carriers and Electron States, Academic Press Limited, London. Reislöhner, U. et al. (2007) Doping profiles in CdTe/CdS thin film solar cells. Thin Solid Films, 515, 6175. Losee, D.L. (1975) Admittance spectroscopy of impurity levels in Schottky barriers. J. Appl. Phys., 46(5), 2204. Kneisel, J. et al. (2000) Admittance spectroscopy of efficient CuInS2 thin film solar cells. J. Appl. Phys., 88(9), 5474. Walter, T. et al. (1996) Determination of defect distributions from admittance measurements and application to Cu(In,Ga)Se2 based heterojunctions. J. Phys., 80(8), 4411.

103. Burgelman, M. and Nollet, P. (2005)

104.

105.

106.

107.

108.

109.

110.

111.

112.

113.

114.

Admittance spectroscopy of thin film solar cells. Solid State Ionics, 176, 2171. Seymour, F.H., Kaydanov, V., and Ohno, T.R. (2006) Simulated admittance spectroscopy measurements of high concentration deep level defects in CdTe thin-film solar cells. J. Appl. Phys., 100, 33710. Böer, K.W. (1976) Photovoltaic effect in CdS-Cu2 S heterojunctions. Phys. Rev. B, 13(12), 5373. Lassnig, R. (1987) Tunneling through semiconductor heterojunction barriers. Solid State Commun., 61(9), 577. Yukadinovic, M. et al. (2004) Transport in tunneling recombination junctions: a combined computer simulation study. J. Appl. Phys., 96(12), 7289. Lu, Z.H., Hanna, M.C., and Majerfeld, A. (1994) Determination of band gap narrowing and hole density for heavily C-doped GaAs by photoluminescence spectroscopy. Appl. Phys. Lett., 64(1), 88. Schenk, A. (1998) Finite-temperature full random-phase approximation model of band gap narrowing for silicon device simulation. J. Appl. Phys., 84(7), 3684. Lundberg, O. et al. (2003) Influence of the Cu(In,Ga)Se2 thickness and Ga grading on solar cell performance. Prog. Photovoltaics: Res. Appl., 11(2), 77. Orgassa, K., Schock, H.W., and Werner, J.H. (2003) Alternative back contact materials for thin film Cu(In,Ga)Se2 solar cells. Thin Solid Films, 431/432, 387. Gupta, A., Parikh, V., and Compaan, A.D. (2006) High efficiency ultra-thin sputtered CdTe solar cells. Sol. Energy Mater. Sol. Cells, 90, 2263. Green, M.A. (1996) Bounds upon grain boundary effects in minority carrier semiconductor devices: a rigorous ‘‘perturbation’’ approach with application to silicon solar cells. J. Appl. Phys., 80(3), 1515. Goeckler, M. (2005) Device Physics of Cu(InGa)Se2 Thin-Film Solar Cells, Colorado State University, Fort Collins, CO.

325

326

References 115. Gloeckler, M., Sites, J.R., and Metzger,

116.

117.

118.

119.

120.

121.

122.

123.

124.

125.

126.

127.

W.K. (2005) Grain-boundary recombination in Cu(In,Ga)Se2 solar cells. J. Appl. Phys., 98, 113704. Woods, L.M. et al. (1998) Electrical characterisation of CdTe grain-boundary properties from as processed CdTe/CdS solar cells, in 2nd World Conference on Photovoltaic Solar Energy Conversion (eds J. Schmid and H. Ossenbrink), Stephens, Vienna, p. 1043. Taretto, K., Rau, U., and Werner, J.H. (2003) Method to extract diffusion length from solar cell parameters-application to polycrystalline silicon . J. Appl. Phys., 93(9), 5447. Taretto, K., Rau, U., and Werner, J.H. (2005) Numerical simulation of grain boundary effects in Cu(In,Ga)Se2 thin-film solar cells. Thin Solid Films, 480/481, 8. Niemegeers, A. and Burgelman, M. (1997) Effects of the Au/CdTe back contact on IV and CV characteristics af Au/CdTe/CdS/TCO solar cells. J. Appl. Phys., 81(6), 2881. Topic, M., Smole, F., and Furlan, J. (1996) Band-gap engineering in CdS/Cu(In,Ga)Se2 solar cells. J. Appl. Phys., 79(11), 8537. Dullweber, T. et al. (2001) Back surface band gap gradings in Cu(In,Ga)Se2 solar cells. Thin Solid Films, 387, 11. Gloeckler, M. and Sites, J.R. (2005) Band-gap grading in Cu(In,Ga)Se2 solar cells. J. Phys. Chem. Solid, 66(11), 1891. Song, J. et al. (2004) Device modeling and simulation of the performance of Cu(In1-x ,Gax )Se2 solar cells. Solid State Electron., 48, 73. Green, M.A. (2009) Do built-in fields improve solar cell performance. Prog. Photovoltaics: Res. Appl., 17, 57. Sah, C.T. (1982) Thickness dependence of solar cell performance. Solid State Electron., 25(9), 960. Roos, Ov. (1978) A simple theory of back surface field (BSF) solar cells. Journal of Applied Physics, 49(6), 3503. Wiedeman, S. et al. (1996) Optimization of Cu(In,Ga)Se2 photovoltaic modules,

128.

129.

130.

131.

132.

133.

134.

135.

136.

137.

138.

Proceedings of the 13th European Photovoltaic Solar Energy Conference, Stephens and Associates, p. 2059. Wennerberg, J., Kessler, J. and Stolt, L. (2001) Design of gridded Cu(In,Ga)Se2 thin film PV modules. Sol. Energy Mater. Sol. Cells, 67, 59. Burgelman, M. and Niemegeers, A. (1997) Calculation of CIS and CdTe module efficiencies. Sol. Energy Mater. Sol. Cells, 51, 129–143. Wennerberg, J., Kessler, J., and Stolt, L. (2001) Design of grided Cu(In,Ga)Se2 thin film PV modules. Sol. Energy Mater. Sol. Cells, 67, 59. Wei, S.H. and Zhang, S.B. (2000) Structure stability and carrier localization in CdX (X=S,Se,Te) semiconductors. Phys. Rev. B, 62(11), 6944. McCandless, B. (2001) Thermochemical and kinetic aspects of cadmium telluride solar cell processing, in II-VI Compound Semiconductor Photovoltaic Materials (eds R. Birkmire et al.), MRS, Warrendale, PA, p. H.1.6. McCandless, B.E. and Sites, J.R. (2003) Cadmium telluride solar cells, in Handbook of Photovoltaic Science and Engineering (eds A. Luque and S. Hegedus), John Wiley & Sons, Ltd, Chichester, pp. 617–662. Zhang, S.B. et al. (1998) Defect physics of the CuInSe2 chalcopyrite semiconductor. Phys. Rev. B, 57(16), 9642. Jianrong, Y. et al. (1995) Thermodynamic and phase diagram assessment of the Cd-Te and Hg-Te systems. CALPHAD, 19, 399. Tai, H. and Hori, S. (1976) Equilibrium phase diagrams of the CdTe-CdCl2 and CdTe-CdBr2 systems. J. Jpn Inst. Met., 40, 732. Ohata, K., Saraie, J., and Tanaka, T. (1973) Phase diagram of the CdS-CdTe pseudobinary system. Jpn J. Appl. Phys., 12(8), 1198. Jensen, D.G., McCandless, B.E., and Birkmire, R.W. (1996) Thin-film cadmium sulfide/cadmium telluride alloys, in Thin Films for Photovoltaic and Related Device Applications (eds D. Ginley et al.), Materials Research Society, Pittsburgh, p. 325.

References 139. Aven, M. and Prener, J. (1967) Physics

140.

141.

142.

143.

144.

145.

146.

147.

148.

149.

150.

and Chemistry of II-VI Compounds, John Wiley & Sons, Inc., New York, pp. 211–212. Knacke, O., Kubaschewski, O., and Hesselmann, K. (1991) Thermochemical Properties of Inorganic Substances, 2nd edn, Springer-Verlag, New York. Pamplin, B.R. and Berger, L.I. (2003) Handbook of chemistry and physics, in CRC Handbook of Chemistry and Physics (ed. D.R. Lide), CRC Press, Boca Raton, p. 57. Landolt, R. and Börnstein, B. (1989) in Numerical Data and Functional Relationships in Science and Technology (eds A. Goldmann and E.E. Koch), Springer, Berlin. Mathew, X. and Enriquez, J.P. (2000) Temperature dependence of the optical transitions in CdTe thin film – investigation with photoresponse spectra. Sol. Energy Mater. Sol. Cells, 63, 347. Mitchell, K., Fahrenbruch, A.L., and Bube, R.H. (1977) Photovoltaic determination of optical-absorption coefficient in CdTe. J. Appl. Phys., 48, 829. Rakhshani, A.E. (1997) Electrodeposited CdTe – optical properties. J. Appl. Phys., 81(12), 7988. Imanaka, Y. et al. (2006) Cyclotron resonance in II–VI semiconductors at THz region. Phys. Status Solidi A, 243(4), 939. Adachi, A. (2005) Properties of group IV, III-V, and II-VI semiconductors, in Materials for Electronic and Optoelectronic Applications (eds P. Capper, S. Kasap and A. Willoughby), John Wiley & Sons, Ltd, Chichester, p. 27. Paulson, P.D., McCandless, B.E., and Birkmire, R.W. (2004) Optical properties of Cd1-x Znx Te Te films in a device structure using variable angle spectroscopic ellipsometry. J. Appl. Phys., 95(6), 3010. Grill, R. et al. (2002) Chemical self-diffusion in CdTe. Phys. Status Solidi A, 229, 161. Dzhafarov, T.D. and Ongul, F. (2005) Modification of CdTe thin films by Zn reactive diffusion. J. Phys. D, 38, 3754.

151. Dzhafarov, T.D. et al. (2005) Diffu-

152.

153.

154.

155.

156.

157.

158.

159.

160.

161.

162.

163.

sion and influence of Cu on properties of CdTe thin films and CdTe/CdS cells. Sol. Energy Mater. Sol. Cells, 85, 371. Lane, D.W. et al. (1999) Sulphur diffusion in CdTe and the phase diagram of the CdS-CdTe pseudo-binary alloy. J. Cryst. Growth, 197, 743. Malzbender, J. et al. (1996) Studies on the diffusion of the halogens into CdTe. Semiconduct. Sci. Technol., 11, 741. Watson, E. and Shaw, D. (1983) The solubility and diffusivity of In in CdTe. J. Phys. C, 16, 515. Ivanov, Y.M., Pavlova, G.S., and Kanunova, E.L. (1988) Diffusion of indium and sodium in chlorine-doped CdTe. Inorg. Mater., 24, 1681. Teramoto, I. and Takayanagi, S. (1962) Behavior of gold in cadmium telluride crystals. J. Phys. Soc. Jpn, 17(7), 1137. Chou, H.C. et al. (1996) Approach toward high efficiency CdTe/CdS heterojunction solar cells. Mater. Chem. Phys., 43, 178. McCandless, B., Engelmann, M.G., and Birkmire, R.W. (2001) Interdiffusion of CdS/CdTe thin films: modeling X-ray diffraction line profiles. J. Appl. Phys., 89(2), 988. Wei, K. et al. (1999) Optical properties of CdTe1-x Sx (0<x

Chalcogenide Photovoltaics

Chalcogenide Photovoltaics: Physics, Technologies, and Thin Film Devices

Applied Photovoltaics

Semiconducting Chalcogenide Glass II: Properties of Chalcogenide Glasses

Concentrator photovoltaics

Applied Photovoltaics

Organic Photovoltaics

Photovoltaics And Architecture

Next generation photovoltaics

Photovoltaics and Architecture

Building Integrated Photovoltaics

Nanotechnology for Photovoltaics

Clean Electricity from Photovoltaics

Photovoltaics and Architecture

Clean Electricity from Photovoltaics

Practical Handbook of Photovoltaics

Nanotechnology for Photovoltaics

Photovoltaics and Architecture

Organic Photovoltaics: Concepts and Realization

High-Efficient Low-Cost Photovoltaics

Organic Photovoltaics. Concepts and Realization

Semiconducting Chalcogenide Glass III, Volume 80: Applications of Chalcogenide Glasses (Semiconductors and Semimetals)

Semiconducting Chalcogenide. Glass I. Volume 78

Concentrator Photovoltaics (Springer Series in Optical Sciences)

Semiconducting Chalcogenide Glass II, Volume 79: Properties of Chalcogenide Glasses (Semiconductors and Semimetals)

Amorphous Chalcogenide Semiconductors and Related Materials

Practical Handbook of Photovoltaics: Fundamentals and Applications

Practical Handbook of Photovoltaics: Fundamentals and Applications

Energy Management in Buildings Using Photovoltaics

Optical Nonlinearities in Chalcogenide Glasses and Their Applications

High-efficient low-cost photovoltaics: recent developments

Chalcogenide Photovoltaics