Clean Electricity from Photovoltaics

Series on Photoconversion of Solar Energy — Vol. 1

CLEAN ELECTRICITY FROM PHOTOVOLTAICS

/

/

Imperial College Press

CLEAN ELECTRICITY FROM PHOTOVOLTAICS

Series on Photoconversion of Solar Energy — Vol, 1

CLEAN ELECTRICITY FROM PHOTOVOLTAICS

Editors

Mary D. Archer Imperial College, UK

Robert Hill University of Northumbria, UK

Imperial College Press

Published by Imperial College Press 57 Shelton Street Covent Garden London WC2H 9HE Distributed by World Scientific Publishing Co. Pte. Ltd. P O Box 128, Farrer Road, Singapore 912805 USA office: Suite IB, 1060 Main Street, River Edge, NJ 07661 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

Index prepared by Indexing Specialists, Hove, BN3 2DJ, UK

CLEAN ELECTRICITY FROM PHOTOVOLTAICS Series on Photoconversion of Solar Energy — Vol. 1 Copyright © 2001 by Imperial College Press All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.

ISBN

1-86094-161-3

Printed in Singapore.

This volume is dedicated with the affection and respect of its authors

Robert Hill 24 June 1937 — 26 November 1999

CONTENTS About the authors

xm

Preface

xxiii

1 The past and present M. D. Archer 1.1 1.2 1.3 1.4 1.5

1

Milestones in photovoltaic technology Evolution of the PV market Overview of photo voltaic cell operation Other junction types Sources of further information

4 11 14 24 28

2 Device physics of silicon solar cells J. O. Schumacher and W. Wettling 2.1 Introduction 2.2 Semiconductor device equations 2.3 Thep-n junction model of Shockley 2.4 Real diode characteristics 2.5 Numerical solar cell modelling 2.6 Concluding remarks

33

3 Principles of cell design J. Poortmans, J. Nijs and R. Mertens

91

3.1 3.2 3.3 3.4 3.5 3.6 3.7

Introduction Main cell types Optical design of cells Surface recombination losses and their reduction Bulk recombination losses and their reduction Design and fabrication of the metal contacts Conclusions

4 Crystalline silicon solar cells M. A. Green

33 35 37 55 67 86

91 93 99 108 121 133 140 149

4.1 Overview 4.2 Silicon cell development

149 151

Vll

viii

Contents

4.3 4.4 4.5 4.6 4.7 4.8

Substrate production Cell processing Cell costs Opportunities for improvement Silicon-supported thin films Summary

164 173 178 180 185 189

5 Amorphous silicon solar cells C. R. Wronski and D. E. Carlson

199

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11

Introduction Background Amorphous silicon-based materials Growth and microstructure Solar cells Solar cell structures PV modules Manufacturing costs Long-term reliability Environmental issues Challenges for the future

6 Cadmium telluride solar cells D. Bonnet 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9 6.10

Introduction Early work The potential of the base material Diodes and cells Cell production Module production Industrial status—achievements and projections Economic aspects Health and environmental aspects Conclusions

7 Cu(In,Ga)Se2 solar cells U.RauandH. W. Schock 7.1 Introduction

199 201 202 209 211 221 225 231 232 235 236 245 245 246 246 249 251 262 264 267 268 269 277 277

IX

7.2 7.3 7.4 7.5 7.6

Material properties Cell and module technology Device physics Wide-gap chalcopyrites Conclusions

8 Super-high efficiency III-V tandem and multijunction cells M Yamaguchi 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8

Introduction Principles of super-high efficiency multijunction solar cells Candidate materials for multijunction cells and their present status Epitaxial technologies for growing III-V compound cells Monolithic vs. multi-terminal connection modes Cell interconnection Possible applications of multijunction cells Predictions

9 Organic photovoltaic devices J. J. M. Halls andR. H. Friend 9.1 Introduction 9.2 Background—early work on photoresponsive organic semiconductors 9.3 Conjugated molecules: a new class of semiconductors 9.4 Basic organic photovoltaic cells 9.5 Photogeneration and charge transport in organic PV cells 9.6 The characteristics of organic photovoltaic cells 9.7 Heterojunction photovoltaic cells 9.8 Dispersed heterojunction photovoltaic cells 9.9 Diffuse interface photovoltaic cells 9.10 Towards future applications 9.11 Conclusions 10 Quantum well solar cells J. Nelson 10.1 Introduction 10.2 Device design, materials and technology 10.3 Physics of QWs

279 286 306 325 332 347 347 349 355 363 364 365 368 369 377 377 383 384 390 398 405 413 421 428 429 432 447 447 448 451

Contents

X

10.4 10.5 10.6 10.7

Performance characteristics of QWSCs Limits to efficiency Applications Conclusions

462 472 474 476

11 Thermophotovoltaic generation of electricity T. J. Coutts 11.1 Introduction 11.2 Radiators 11.3 Optical control elements 11.4 Device modelling 11.5 Potentially suitable materials 11.6 System modelling 11.7 Summary

481

12 Concentrator cells and systems A. Luque 12.1 Introduction 12.2 Concentrator solar cells 12.3 Tracking concentrators 12.4 Performance and cost considerations 12.5 Conclusion: under what circumstances is concentration worthwhile?

529

13 Cells and systems for space applications C. M. Hardingham 13.1 13.2 13.3 13.4 13.5 13.6 13.7 13.8

Space systems The space environment History of solar arrays in space Market trends and drivers in satellite power requirements Satellite solar arrays Space solar cell technology New approaches for satellite solar arrays Long-term directions

481 487 490 497 506 512 518

529 531 556 570 574

585 585 588 592 593 596 599 604 605

Contents 14 Storage of electrical energy R. M. Dell 14.1 Introduction 14.2 Electricity storage options 14.3 Kinetic energy storage 14.4 Hydrogen energy storage 14.5 Storage batteries 14.6 Super- and ultra-capacitors (electrochemical capacitors) 14.7 Conclusions 15 Photovoltaic modules, systems and applications N. M. Pearsall andR. Hill 15.1 15.2 15.3 15.4 15.5 15.6

Introduction Photovoltaic modules The photovoltaic array The photovoltaic system Costs of PV components and systems Conclusions

16 The photovoltaic business: manufacturers and markets B. McNelis 16.1 Introduction 16.2 Origins and structure of the industry 16.3 Growth in PV production 16.4 Manufacturers 16.5 Markets 16.6 Future market growth 16.7 International financing and new initiatives 16.8 Concluding remarks 17 The economics of photovoltaic technologies D. Anderson 17.1 17.2 17.3 17.4

Introduction Economics of PV applications The policy framework Conclusions

XI

609 609 610 614 618 633 662 663 671 671 672 683 688 704 710 713 713 715 716 718 726 732 734 736 741 741 742 754

xii

Contents

18 The outlook for PV in the 21st century E. H. Lysen andB. Yordi 18.1 18.2 18.3 18.4 18.5 18.6 18.7 18.8 18.9

I II HI IV

771

The changing outlook for PV PV and world energy supply PV can play an impressive local role The ultimate PV system Market development Barriers to the introduction of PV Costs International co-operation The future of PV

771 773 774 779 781 784 786 787 788

Appendices Fundamental Constants Useful Quantities and Conversion Factors List of Symbols Acronyms and Abbreviations

791 792 793 797

Index

799

ABOUT THE AUTHORS Dennis Anderson is a Professorial Research Fellow and Director of the Centre for Energy Policy and Technology in the T. H. Huxley School of Imperial College, London. At the time of writing his chapter, he was a Fellow of the UK Economic and Social Science Research Council (Global Environment Change Programme), undertaking research on innovation and the environment. He has previously held posts as the Energy and Industry Adviser of the World Bank, Chief Economist of Shell, and as an engineer in the electricity generating industry. He has published widely on the subjects of energy, economic growth and development. Mary Archer read chemistry at Oxford University and received her PhD on heterogeneous catalysis from Imperial College, London in 1968. Her interest in solar energy was sparked by attendance at the 1972 International Solar Energy Society in Paris, following which she founded the UK Section of ISIS, of which she is currently President. Her research at The Royal Institution, London (1972-1976), and Cambridge University (1976-1986) has centred on photoelectrochemical methods of solar energy conversion. Since leaving full-time academia in 1986, she has served on a number of energy policymaking bodies, including the UK Department of Energy's Renewable Energy Advisory Group (1991-92), the Department of Trade & Industry's Energy Advisory Panel (1993-98) and the Steering Committee of the Scolar Programme for Photo voltaics in the UK. She is a visiting professor in the Centre for Energy Policy and Technology at Imperial College, a Fellow of the Royal Society of Chemistry, and President of the National Energy Foundation, which promotes energy efficiency and the renewables. Dieter Bonnet was born in Stuttgart, Germany in 1937 and obtained his PhD on photoelectric properties of organic materials at Frankfurt University in 1963. In 1965, he joined Battelle Institute in Frankfurt, and in 1968 started work on thin-film solar cells based on II-VI compounds, including CdTe. In 1970, he made the world's first CdTe/CdS thinfilm solar cell in the presently known configuration. In June 1972—over 25 years ago—this cell had an AMO efficiency of 6%. In 1990, he resumed work on CdTe thinfilm cells, and in 1992 initiated the EUROCAD CdTe thin-film solar cell project, which is funded by the EU's Joule programme. Ten partners, among them three industrial companies, have since collaborated very successfully under this programme to develop CdTe cell technology. In 1993, after Battelle Frankfurt terminated business, Dieter Bonnet co-founded ANTEC GmbH, and he is presently leading efforts to set up a 10 MWp/year production plant using ANTEC's proprietary thin-film technology. Xlll

XIV

About the Authors

David Carlson is Chief Scientist of BP Solarex. He received his BS in physics from Rensselaer Polytechnic Institute, New York in 1963, and his PhD in physics from Rutgers University in 1968. After serving in the US Army for two years, he joined RCA Laboratories in 1970, where he invented the amorphous silicon solar cell in 1974 and became Group Head of Photovoltaic Device Research in 1977. In 1983, he joined Solarex Corporation (now BP Solar) as Director of Research of the Thin-Film Division, becoming General Manager in 1987. He was promoted to Vice-President in 1988, and to Chief Scientist in 1999. He received the Ross Coffin Purdy Award in 1975, the Walton Clark Medal in 1986, the IEEE William R. Cherry Award in 1988, and the ISES/University of Delaware Karl W. Boer Medal in 1995. He was co-recipient (with Christopher Wronski) of the 1984 IEEE Morris N. Liebmann Award. He is a Fellow of the IEEE and a member of the American Physical Society, the American Vacuum Society, the Materials Research Society and Sigma Xi. He has published more than 110 technical papers and holds 25 US patents. Timothy Courts was born in Newcastle upon Tyne, UK and gained his bachelor's and doctoral degrees in 1965 and 1969. He has worked on many topics, including charge transfer in thin copper films, discontinuous, continuous and cermet thin films, and surface scattering in thin metal films. He has been involved in solar cell research since about 1970. He joined the US National Renewable Energy Laboratory (NREL), where he is now a Research Fellow, in 1984. He helped to develop ITO/InP cells for space application, and InP/InGaAs cells with a record efficiency of 31.8%. He has had a keen interest in thermophotovoltaics (TPV) since 1992, and initiated TPV research and chaired four conferences on the topic at NREL. He is currently interested in CdTe cells and novel transparent conducting oxide (TCO) electrodes. Recently, his work in TCOs has broadened to include new materials and characterisation techniques. He was awarded the John A. Thornton Memorial Award by the American Vacuum Society in 1999. He has published over 170 papers, written one book and edited ten others. Ronald Dell is a chemist, educated at the University of Bristol, UK After several years in the US working on chemisorption and catalysis and two years in the Royal Naval Scientific Service, he joined the UK Atomic Energy Authority in 1959 and remained there until he retired in 1994. At Harwell he spent almost 20 years working in solidstate chemistry, especially of the actinide elements. In 1978, he switched to become head of the Applied Electrochemistry Department with particular interests in power sources and the use of electrochemical techniques to solve environmental problems. He is the author of nearly 100 scientific papers and reports and co-author of the book Batteries for Electric Vehicles (Research Studies Press, Baldock, Herts, UK, 1998).

About the Authors

xv

Richard Friend is the Cavendish Professor of Physics at the University of Cambridge. He has pioneered the study of organic polymers as semiconductors, and demonstrated that these materials can be used in wide range of semiconductor devices, including light-emitting diodes, transistors and photocells. He has been very active in the process of technology transfer of this research to development for products. He was one of the founders of Cambridge Display Technology (CDT), which is developing light-emitting diodes and other optoelectronic devices based on organic semiconductors, and he currently serves as Director and Chief Scientist of CDT. Martin Green is a Scientia Professor at the University of New South Wales, Sydney, the Director of the University's Photovoltaics Special Research Centre, and the Research Director of Pacific Solar Pty. Ltd., established to commercialise the University's silicon thin-film solar cell technology. He was born in Brisbane and educated at the University of Queensland and then McMaster University, Canada. His contributions to photovoltaics include the improvement of silicon solar cell performance by over 50% in the past 15 years. Major international awards include the IEEE William R. Cherry Award in 1990, the IEEE J. J. Ebers Award in 1995 and the 1999 Australia Prize, shared with his colleague and former student, Stuart Wenham, for "outstanding achievements in energy science and technology". He is a Fellow of the Australian Academy of Science, the Australian Academy of Technological Sciences and Engineering and the Institute of Electrical and Electronic Engineers. He is the author of four books on solar cells, several book chapters and numerous reports and papers in the area of semiconductor properties, microelectronics and solar cells. Jonathan Halls was born in Lincoln in 1972. After reading physics at Cambridge University, he began research for a PhD under the supervision of Professor Richard Friend in the Optoelectronics Group of the Cavendish Laboratory in Cambridge. His main field of research was that of photovoltaic cells based on conjugated polymers, and he investigated a number of approaches to increase their efficiency. In doing so, he pioneered a technique in which electron- and hole-accepting polymers are blended together, yielding a high surface area of active interface at which charge separation is efficient. This work resulted in a publication in Nature and the filing of a patent. In 1997, he began postdoctoral research in the same research group, during which time he has worked on organic light-emitting diodes, and is currently continuing to work with organic photovoltaic cells.

XVI

About the Authors

Chris Hardingham was born in Essex in 1963. Following a physics degree at Cambridge University, he worked at EEV (now Marconi Applied Technologies) on semiconductor process development for GaAs and related materials. He was awarded his PhD by Imperial College, London in 1998, for research into the use of electron beam techniques for semiconductor materials analysis. Following responsibilities for solar cell R&D, and solar cell engineering and project management, he moved to his present position of solar cell product manager at Marconi Applied Technologies in 1999. His interests include III-V materials for solar cells and other applications, and device and subsystems engineering for use in space. He holds several patents and patent applications in the field of III-V space solar cells, and has presented and written many papers in the field for technical conferences and peer-reviewed journals. Robert Hill (1937-1999) took his first degree in physics at Imperial College, London, and a PhD in solid-state luminescence. He worked in photovoltaics from 1971, originally on the science and technology of thin-film cells. He then widened his interests to include the economic and environmental aspects of production and applications, PV in developing countries and on buildings, and the policy aspects of PV dissemination. He founded the Newcastle Photovoltaics Applications Centre in 1984, and was its director until his retirement in 1998. In January 1999, he was appointed director of the Renewable Energy Agency for the North East (of the UK), funded by Government Office North East, with a remit to increase the use of renewable energy sources and promote the development of industrial capabilities in these technologies. He was a founder member of the British Photovoltaics Association and its chairman for the year 1999-2000. Antonio Luque obtained his Doctor of Engineering degree from the Polytechnic University of Madrid in 1967. In 1969, he joined the university staff and founded its Semiconductor Laboratory. In 1979, this centre became the Institute of Solar Energy that he leads at present. In 1981, he founded the company Isofoton to manufacture the bifacial cells he had invented, and he chaired its board until 1990. Professor Luque has written some 200 papers and registered some 12 patents, of which four are in exploitation. He has obtained 12 scientific awards, among which are the Spanish National Prize for Technology in 1989, the Becquerel Prize awarded by the European Commission for PV in 1992 and the Rey Jaime I Prize for the protection of the environment in 1999. He has been a member of the Spanish Academy of Engineering since 1995, and a member of the Advisory Council for Science and Technology, which advises the Spanish Prime Minister, since 1996.

About the Authors

xvn

Erik Lysen has been managing director of the Utrecht Centre for Energy Research since mid-1998. He received his master's degree in electrical engineering from Eindhoven University of Technology in 1972. In the seventies, he worked on wind power projects in developing countries, first as head of the CWD Wind Energy Group at the University of Groningen, and later at Eindhoven University of Technology. As senior project engineer for DHV Consultants, Amersfoort, and later as an independent consultant, he carried out energy projects for a number of clients such as the World Bank. From 1992 until 1998, he was Head of New Developments for the Netherlands Agency for Energy and the Environment (Novem). He has chaired the Executive Committee of the IEA Photovoltaic Power Systems Programme (IEA-PVPS) since 1998. He is a member of the Energy and Environment Steering Committee of the World Bank, and the Advisory Boards of the Solar Investment Fund of Triodos Bank and the PV Global Approval Program (PV-GAP). Bernard McNelis is managing director of IT Power, Eversley, UK, an international renewable energy research and consulting firm which he co-founded 20 years ago. After research in battery electrochemistry, he joined Solar Power Corporation in 1973. He moved on to work on solar buildings and large-scale solar thermodynamic power generation. He is one of the longest serving members of the British renewable energy industry, with more than 25 years experience of renewable energy technologies—photovoltaics, solar-thermal, solar-thermodynamic, wind and biomass. He has been an active member of the International Solar Energy Society since 1974, serving as chairman of UK-ISES in the period 1993-1996, director of ISES 1993-99, and Vice-President 1995-1997. He is currently chairman of the British Photovoltaic Association (P V-UK) and of the British Standards Institution PV Committee. He is also a member of the International Electrotechnical Commission PV Standards Committee (TC/82) and British representative for a number of International Energy Agency (IEA) PV activities. He led the IEA Photovoltaic Power Systems project on co-operation with developing countries. He has published more than 100 papers and contributed to five books on solar technology. Robert Mertens received his PhD from the Katholieke Uni versiteit of Leuven, Belgium in 1972 and was a visiting scientist at the University of Florida in 1973. On his return to Belgium in 1974, he became a senior research associate of the National Foundation for Scientific Research of Belgium. In 1984, he joined the Inter-University Microelectronics Centre (IMEC) in Leuven as Vice-President, later becoming senior Vice-President responsible for research on materials, components and packaging, including research on micro-systems, photovoltaics and solid-state sensors. Since 1984,

XV111

About the Authors

he has also served as a professor at the University of Leuven, where he teaches courses on electronic devices and the technology of electronic systems. In 1995, he was elected a Fellow of the IEEE for his "contributions to heavily doped semiconductors, bipolar transistors and silicon solar cells". Jenny Nelson is an EPSRC Advanced Research Fellow in the Department of Physics, Imperial College, London. She has been involved in photovoltaics research for over ten years, focussing on the theory, characterisation and optimisation of novel multi-bandgap and heterojunction photovoltaic devices. With Professor Keith Barnham, she was a pioneer of the quantum well solar cell, and more recently has extended her research to dye-sensitised photovoltaic systems. Her work has been supported by the Engineering and Physical Sciences Research Council and the Greenpeace Environmental Trust. Johan Nijs took his MS in electronic engineering, his PhD in applied sciences, and his MBA from the Katholieke Universiteit of Leuven (K.U. Leuven), Belgium in 1977, 1982 and 1994 respectively. In 1977, after a trainee period of two months at Philips, he joined the Electronics, Systems, Automation and Technology (ES AT) laboratory of K.U. Leuven, working on the fabrication of silicon solar cells. In 1982-83, he worked on amorphous silicon technology as a postdoctoral visiting scientist at the IBM Thomas J. Watson Research Center, Yorktown Heights, New York. In 1984, he joined the InterUniversity Micro-Electronics Centre (IMEC) in Leuven as head of the Silicon Materials Group, working on solar cells, bipolar transistors, low-temperature silicon epitaxy and polysilicon thin-film transistors on glass. He is currently Director of the Photovoltaics Department at IMEC, which undertakes long-term research on photovoltaic materials, concepts and technologies, industrial crystalline silicon cell fabrication technologies and photovoltaic systems integration. In 1990, he was appointed part-time assistant professor at K.U. Leuven. He has authored or co-authored more than 200 papers, and is the inventor or co-inventor on 10 patents or patent applications. Nicola Pearsall is Director of the Newcastle Photovoltaics Applications Centre at the University of Northumbria, having taken over on the retirement of Professor Robert Hill in the summer of 1998. She holds a degree in physics from the University of Manchester Institute of Science and Technology and obtained her PhD from Cranfield Institute of Technology for research on indium phosphide cells for satellite applications. She has been involved in research in photovoltaics for over 20 years, and has worked on the development of devices for space and terrestrial applications, testing methods, system design and performance analysis. Much of her current work is in the area of buildingintegrated photovoltaics.

About the Authors

xix

Jozef Poortmans received his degree in electronic engineering from the Katholieke Universiteit of Leuven, Belgium, in 1985, and then joined the new Inter-University Microelectronic Centre (IMEC) in Leuven, working on laser recrystallisation of polysilicon and amorphous silicon for solar cells and thin-film transistors. In 1993, he received his PhD for a study of strained Si/Ge layers. He then joined the Photovoltaics Group (later Department) of IMEC, where he is currently in charge of the Advanced Solar Cells Group. This group has three main activities: low-thermal-budget processes (rapid thermal processing and plasma deposition), the fabrication of thin-film crystalline Si solar cells on Si and foreign substrates, and organic solar cells. He has authored or co-authored more than 140 papers, as well as two book chapters on the properties of Si/Ge alloys and heterojunction bipolar transistors. Uwe Rau received his PhD in physics in 1991 from the University of Tubingen, Germany, for his work on temporal and spatial structure formation in the lowtemperature electronic transport of bulk semiconductors. From 1991 to 1994, he worked at the Max Planck-Institut fiir Festkorperforschung, Stuttgart on Schottky contacts, semiconductor heteroj unctions and silicon solar cells. From 1994 to 1997, he worked at the University of Bayreuth, Germany, on electrical characterisation and simulation of Si and CuInSe2 solar cells. In 1997, he joined the Institut fiir Physikalische Elektronik at the University of Stuttgart, where he became leader of the Device Analysis Group. His research interests centre on transport phenomena, especially electrical transport in solar cell heteroj unction devices and interface and bulk defects in semiconductors. He has authored or co-authored more than 100 scientific publications. Hans-Werner Schock leads the compound semiconductor thin-film group of the Institute of Physical Electronics at the University of Stuttgart, Germany. He received his diploma in electrical engineering in 1974, and doctoral degree in electrical engineering in 1986, from the University's Faculty of Electrical Engineering. Since the early 1970s, he has worked on the development of polycrystalline II-VI and I—III—VI2 compound semiconductor thin-film solar cells, from basic investigations to the transfer to pilot fabrication. He also developed chalcogenide compound phosphors for tnin-film electroluminescence. Since 1986, he has co-ordinated the research on chalcopyrite-based solar cells in the European photovoltaic programme. He is the author or co-author of more than 250 contributions in books, scientific journals and conference proceedings.

XX

About the Authors

Jiirgen Schumacher studied physics in Frankfurt/Main and Freiburg in Germany. He is currently working toward completion of his PhD on the simulation and characterisation of novel and high-efficiency solar cell devices at the Fraunhofer Institute for Solar Energy Systems in Freiburg. As part of his studies, he worked as a visiting scientist at the University of New South Wales, Sydney, Australia in the Photovoltaics Special Research Centre headed by Professor Martin Green. Wolfram Wettling is head of the Department of Solar Cells Materials and Technology of the Fraunhofer Institute for Solar Energy Systems (ISE) in Freiburg, Germany, which is the largest institute devoted to solar energy R&D in Europe. He also teaches semiconductor physics at the University of Freiburg. After studying physics in Freiburg and Karlsruhe and a post-doctoral year at the Technical University of Copenhagen, he joined the Fraunhofer Institute for Applied Solid State Physics in 1970, working in various fields of solid-state physics such as magnetism, magneto-optics, light scattering, electron-phonon and magnon-phonon interaction, laser development and III-V semiconductors. He has also worked as a visiting scientist at the Hebrew University, Jerusalem and Colorado State University, Fort Collins. In 1988, he joined the Fraunhofer ISE and since then has been involved in the development of highly efficient crystalline silicon and III-V solar cells. He is the author or co-author of about 150 papers, half of them in the field of photovoltaics. Christopher Wronski is Leonhard Professor of Microelectronic Materials and Devices and co-director of the Center for Thin Film Devices at Pennsylvania State University. He received his BS in physics from Imperial College, London in 1960, and his PhD from London University in 1963. From 1963 to 1967, he worked at 3M Research Laboratories. In 1967, he joined the RCA David Sarnoff Research Laboratory, where he collaborated with David Carlson in making the first amorphous silicon solar cells in 1974. His collaboration with David Staebler led to the discovery in 1976 of the reversible lightinduced changes in amorphous silicon known as the Staebler-Wronski effect. Professor Wronski initiated a number of research programmes on amorphous silicon cells at RCA, and later at Exxon Corporate Research Laboratories, which he joined in 1978. At Exxon he was a member of the team that pioneered the development of optical enhancement for amorphous silicon cells. He was also active in studies on multi-layered amorphous superlattices for application to solar cells and photoreceptors. In 1984, he was corecipient (with David Carlson) of the IEEE Morris N. Liebmann Award. He has over 250 publications and ten US patents, and is a Fellow of the IEEE and the American Physical Society.

About the Authors

xxi

Masafumi Yamaguchi is a professor at the Toyota Technological Institute, Nagoya, Japan. He received his BS and PhD degrees from Hokkaido University in 1968 and 1978 respectively. In 1968, he joined the NTT Electrical Communications Laboratories in Tokyo, working on radiation damage in Si and III-V compounds, ZnSe blue-lightemitting diodes and III-V solar cells. In 1983, he discovered the superior radiation resistance of InP, and in 1987 his group developed high-efficiency InP, GaAs-on-Si and AlGaAs/GaAs tandem cells. As chairman of NEDO's Super High-Efficiency Solar Cell Committee, he has contributed to the attainment of very high efficiency InGaP/GaAs dual-junction cells. His research interests include high-efficiency multijunction, concentrator, polycrystalline and thin-film Si cells, radiation damage to solar cells and materials and new carbon-based materials for photovoltaics. He is the chairman of the Photovoltaic Power Generation Technologies Research Committee of the Institute of Electrical Engineers of Japan, and will serve as the Programme Chairman of the Third World Conference on Photovoltaic Energy Conversion, to be held in Osaka in 2003. He received the Vacuum Science Paper Award in 1981, and the Irving Weinberg Award for contributions to space photovoltaics in 1997. Beatriz Yordi has been responsible for the PV sector of the European Commission's Directorate-General for Energy and Transport since October 1994. She was born in La Coruna, Spain and took her Bachelor's Degree in physics at the University of Santiago de Compostela in 1987. Following a year of research in the Department of Optics and Materials Structure at the University of Santiago, she joined the Research Centre for Energy, Environment and Technology (Ciemat) in Madrid, working in the Institutes of Energy Studies and Renewable Energy. From 1991 to 1994, she served as Chief Engineer for the Toledo 1 MW photovoltaic plant, a project with several technical innovations (novel PV cells and a novel tracking system) that was co-funded by the European Commission, the Spanish and German governments and three European utilities.

PREFACE And there the unregulated sun Slopes down to rest when day is done And wakes a vague, unpunctual star ... Rupert Brooke, The Old Vicarage, Grantchester, May 1912.

Since the dawn of history, man has been fascinated by the Sun, the provider of the light and warmth that sustains life on Earth. In pre-industrial times, our major sources of energy—wood, wind and water power—derived from solar energy. The subsequent discovery and massive exploitation of fossil fuels laid down in the Earth's crust by early aeons of photosynthetic activity have conditioned the developed world to be dependent on convenient, readily available energy. But we are living on our energy capital. The Earth's reserves of coal, oil and gas are finite and likely to become resource-depleted in the course of this century. A sense of living on borrowed time was therefore appropriate even before concerns about global climate change, sustainability and energy security combined to raise interest in renewable energy to its current encouraging level. This book is the first in a series of four multi-authorial works on the photoconversion of solar energy. It was created from my long-held conviction that, despite slow starts and setbacks, solar energy—broadly defined to encompass other renewable energy forms that derive from solar—will become the Earth's major energy source within this century. The Sun is a source of both radiant heat and light, and techniques for using solar energy correspondingly divide into thermal methods (solar power towers, water heaters and so on) and photoconversion (sometimes called direct) methods. Photoconversion is the subject of this book series. A photoconverter is a device that converts sunlight (or any other source of light) into a useful form of energy, usually electrical power or a chemical fuel, in a process that relies, not on a raised temperature, but on the selective excitation of molecules or electrons in a light-absorbing material and their subsequent de-excitation in a way that produces energy in a useful form. Volume I covers the most developed of the man photoconversion devices, photovoltaic (PV) cells, which are solid-state semiconductor devices that produce electrical power on illumination. Volume II will cover the natural photoconversion system of photosynthesis, the potential of biomass as an energy source and the global carbon budget. Volume III will explore the less developed but exciting possibilities of synthesising artificial 'molecule-based' photoelectrochemical or photochemical photoconverters. Finally, Volume IV will draw together the common themes of photoconversion and provide some background material.

XXlll

XXIV

Preface

The series is intended mainly for senior undergraduates, graduate students and scientists and technologists working on solar photoconversion. Chapters 1-12 of this book deal with PV cell design, device physics and the main cell types—crystalline and amorphous silicon, cadmium telluride and copper indium diselenide—as well as more advanced or less developed options such as quantum-well and thermophotovoltaic cells. These chapters are mainly technical, requiring sound knowledge of physics, chemistry or materials science for ready understanding. Chapters 13-18 deal with PV systems, manufacturers, markets and economics and are accessible without specialist knowledge. A multi-authorial work owes its very existence to its authors, and my wholehearted thanks must go to the twenty-five distinguished individuals, all recognised authorities in their own fields, who have contributed to this book and patiently answered my queries during the editing stage. I have also been helped by discussions about PV with many friends and colleagues, and visits to installations throughout the world: I have been up Swiss mountains, onto Japanese rooftops and into the Arizona desert, and thoroughly enjoyed every minute. I am most grateful to those who have read and commented on various parts of this book or provided specialist information in advance of publication: Dennis Anderson, Jeffrey and William Archer, Stephen Feldberg, Martin Green, Eric Lysen, Larry Kazmerski, Bernard McNelis and Nicola Pearsall. I also warmly thank Alexandra Anghel, Barrie Clark, Stuart Honan and my PA Jane Williams for editorial assistance, and Ellen Haigh and John Navas of IC Press and Alan Pui of World Scientific Press for guiding the book to publication. For me the sad part of writing this preface is that I must do so in the first person, for my co-editor Professor Robert Hill died suddenly on 26 November 1999. Bob was the most knowledgeable champion of photovoltaics in the UK, and his premature death has deprived the British PV community of its cornerstone. He had drafted his chapter with Nicky Pearsall some months before he died, and the flow of emails delivering his astute editorial comments on other chapters continued until the day before his death. Bob believed unshakeably in the future of PV. Although he knew that system costs will have to fall by another factor of 2-3 if PV is to become cost-competitive in major new grid-accessible markets, there are good grounds for believing this is possible. PV technology is still young, and significant further economies of scale from larger manufacturing facilities, as well as further advances in the fundamental science, can confidently be expected. The world's first-generation televisions and mobile telephones were at least as uncommon and expensive as PV is now.

The Old Vicarage, Grantchester December 2000

Mary Archer

CHAPTER 1

THE PAST AND PRESENT MARY D. ARCHER Centre for Energy Policy and Technology, Imperial College of Science, Technology and Medicine, London SW7 2AZ, U.K. mdal2@cam. ac. uk

Time present and time past Are both perhaps present in time future. T. S. Eliot Burnt Norton, Four Quartets, 1935-1942.

Photovoltaic (PV) cells generate electric power when illuminated by sunlight or artificial light. They are by far the most highly developed of the man-made photoconversion devices. Born of the space age in the 1950s, their earliest terrestrial applications emerged in the 1970s and they are now poised for significant market expansion in the new millennium. PV technology is elegant and benign, with a number of striking advantages over conventional methods of electricity generation. First and foremost, solar energy is the world's major renewable energy resource. PV power can be generated from the Sun anywhere—in temperate or tropical locations, in urban or rural environments, in distributed or grid-feeding mode—where the insolation is adequate. As a fuel-free distributed resource, PV could in the long run make a major contribution to national energy security and carbon dioxide abatement. In the UK, for example, each kWp of PV installed avoids the emission of about 1 tonne C0 2 per year. PV is uniquely scalable, the only energy source that can supply power on a scale of milliwatts to megawatts from an easily replicated modular technology with excellent economies of scale in manufacture. A typical crystalline silicon PV cell generates about 1.5 peak watts1 (Wp) of DC power, a typical PV module about 50 Wp, and the world's largest multimodule arrays (for example, the 3.3 MWe plant at Serre, Italy) generate upward of a megawatt apiece.

' The power output of a PV cell or module is rated in peak watts (Wp), meaning the power output at 25 C under standard AMI.5 solar radiation of global irradiance 1 kW m"2. To convert from peak watt output to 24-hour average power output in a sunny location, divide by ~5.

1

2

M. D. Archer

PV cells are made of thin semiconductor wafers or films. They contain small amounts only of (usually non-toxic) materials and, when manufactured in volume, have modest embedded energy. They possess no moving parts, generate no emissions, require no cooling water and are silent in operation. PV systems are reliable, easy to use and longlived if properly maintained (most commercial modules have lifetime guarantees of 25 years, though some balance-of-system components, notably storage batteries, are less reliable and long-lived than this). Carefully designed, PV arrays are not visually intrusive, and can indeed add architectural merit to the aesthetic of a built structure. PV really has only three drawbacks. First is the intermittence and seasonality of sunlight. As President Gerald Ford is alleged to have remarked, "Solar energy isn't going to happen overnight." The lack of inexpensive and efficient methods of storing electrical energy, and the poor match between the solar and electrical demand peaks in many locations and applications, are stumbling blocks for PV. For small stand-alone applications, battery storage, unsatisfactory as it is, is the only practical storage option. This can be avoided in grid-connected applications where surplus power can be sold to the grid; where there are many distributed or embedded PV generators spread over a geographic region, this has the additional benefit of'integrating out' the fluctuations in local PV contributions. For PV to contribute to global electricity supply on a very large scale, cost-effective means of intercontinental transmission of electrical power (or perhaps of a chemical vector, such as hydrogen, derived from electrical power) would need to be developed. Another characteristic of solar energy that is sometimes perceived as a difficulty is its low power density. The solar power received at Earth's surface, averaged over day and night, winter and summer, varies from about 100 W m 2 in temperate locations to about 300 W m"2 in sunbelt regions. All solar technologies therefore require substantial areas to be covered by solar converters, or by optical concentrators coupled to solar converters, for substantial amounts of power to be generated.2 Taking the UK as an example, the south of England receives insolation of roughly 1 TWh per square kilometre per year, so an area of-2,500 km2 would need to be covered with 15% efficient PV modules to generate the UK's present electricity consumption of-350 TWh/y. The most elegant and cost-effective method of deploying such area-intensive technology is on the surfaces of built structures, rather than as free-standing arrays. This is the more attractive if the PV facade replaces, and avoids the cost of, conventional cladding.

2

Hydroelectric power is, however, considerably more area-intensive than solar power (Anderson and Ahmed, 1993).

The Past and Present

3

This brings us to the second difficulty with PV—its cost. Manufacture of most cell types is an intricate operation, requiring careful control of semiconductor growth and purity and many processing steps. PV systems are expensive, although module costs have fallen substantially—about five-fold in the last twenty years—as the market has grown. In 1999, the PV modules market was worth $665m, and the total value of the business—systems, installation and so forth—was about $2billion (SU, 2000). Current module manufacturing costs are 3-4/Wp, and balance-of-system (BOS) costs can raise the total system cost to 6$/Wp if no battery storage is needed, and 8-10$/Wp if storage is needed. A capital cost of 6$/Wp translates to a PV electricity cost of ~60e7kWh in lowinsolation areas such as western Europe, and ~250/kWh in southern Europe, the USA and much of the developing world.3 These high costs for PV-generated electricity are often compared unfavourably with typical retail prices of -10-150/kWh for grid electricity, and do indeed make PV seem expensive in locations with immediate access to the grid, particularly where (as is often the case) distribution costs are subsidised. But reinforcing or extending the grid to supply increased or new demands is also expensive. The fairer question is under what circumstances the life-cycle costs of supplying a given load by reinforcing or extending the grid would exceed those of installing a stand-alone PV system to supply the same demand. In grid-connected locations, the cost of strengthening the grid to meet increased peak demands is usually concealed by cross subsidy, but can be 15-300/kWh or even more. Provision of peak electricity from a PV substation can therefore become cost-competitive where there is good coincidence between the demand peak and the solar peak. As for grid extension, it is generally cheaper to electrify an isolated village-sized community by PV than extend the grid by 5 km or more to reach it. Access to the grid is in any case not an option for 2 billion or so people (40% of world population) in the developing world. Their conventional small-power options—batteries and diesel generators—compare even less favourably with PV. The current life-cycle costs of PV systems (even with battery storage included) are only about one-tenth to one-half those of secondary batteries, and less than those of diesel generators for loads of under ~30 kWh/day. The third difficulty for PV is one faced by many emergent technologies—ignorance. It is often said that familiarity breeds contempt, but unfamiliarity breeds it too, together with scepticism over manufacturers' claims, veiled or unveiled hostility from established 3

The unit cost of PV electricity depends not only on the capital cost and lifetime of the system components, but also on the local insolation and the cost of borrowing money to finance the system. Energy costs and prices vary widely within and between countries. The costs and assertions in this section are baldly stated, but derive from the detailed costings and assessments of Chapters 15 and 17.

4

M. D. Archer

suppliers and inappropriate regulatory and market structures. Even if consumers are aware of the potential benefits of PV, they can seldom buy 'plug and play' systems off the shelf, and are understandably reluctant to purchase non-standard components for one-off systems. Thus PV faces a dilemma. It is the second fastest growing energy technology in the world, but it is unfamiliar and—in the eyes of many—untested. In 1999 the global PV market grew by 31.5% {PVNews, February 2000), a growth rate exceeded only by wind power, which grew by 35% (IEA, 1999). Were a 30% growth rate to be maintained, PV would meet 1% of projected global electricity demand in 2018, and 10% in 2028. However, such a high growth rate is achievable only because and while PV is growing from a tiny base. In the USA, for example, PV currently provides less than 0.005% of total electricity consumption (KPMG, 1999). Worldwide, about 200 MWp of PV capacity was installed in 1999, and cumulative installed PV capacity is only just over 1 GWp. On average, this supplies -0.2 GWe of PV-generated power, which is only a tiny proportion of the world's current electrical consumption of ~3000 GWe. Although PV is in a virtuous cycle where costs decline as markets expand, its future growth will not be driven by market forces alone at anything like a 30% growth rate. Public policies have played an important role in the development of the industry to date. In Chapter 17, Dennis Anderson argues that further subsidy or tax incentives for PV will be economically efficient and politically justifiable so long as cost curves are declining, the level of prospective use is large and the environmental advantages are demonstrable.

1.1 Milestones in photovoltaic technology The discovery of photovoltaism is commonly, if inaccurately,4 ascribed to Becquerel (1839), who observed that photocurrents were produced on illuminating platinum electrodes coated with silver chloride or silver bromide and immersed in aqueous solution. The observation by Smith (1873) of photoconductivity in solid selenium led to the discovery of the photovoltaic effect in a purely solid-state device by Adams and Day (1877), who observed photovoltages in a selenium rod to which platinum contacts had been sealed, which they (incorrectly) ascribed to light-induced recrystallisation of the selenium. The first practical photovoltaic device—a light meter consisting of a thin layer 4

Becquerel's observation was strictly speaking a photoelectrochemical effect, but its basis—the rectifying junction formed between two dissimilar electric conductors—is the same as that of the photovoltaic effect in purely solid-state devices.

The Past and Present

5

of selenium sandwiched between an iron base plate and a semi-transparent gold top layer made by Fritts (1883)—was promoted by the German industrialist Werner von Siemens as demonstrating "for the first time, the direct conversion of the energy of light into electrical energy" (Siemens, 1885). Photometers based on selenium photocells were commercialised in Germany in the 1930s and are still in use. The selenium photocell is an example of a barrier layer cell, so called because it contains an electrical barrier that is highly resistive to current flow in one direction—a rectifying junction, in modern parlance. Two further barrier layer cells, the thallous sulphide cell (Case, 1920) and the copper oxide cell (Grondahl and Geiger,1927), were developed during the 1920s, but all had solar conversion efficiencies well below 1%. The book by Lange (1938) gives an account of these early devices. The electrical barrier of barrier layer cells was originally thought to lodge in an interfacial foreign layer of high resistivity such as an oxide, but Schottky (1938), and independently Davydov (1939) and Mott (1939), showed that a third phase was not necessarily involved. Rather, metal | semiconductor junctions could in themselves be rectifying by virtue of the space-charge layer created in the semiconductor by charge redistribution when contact was made with a metal of different work function. Metal | semiconductor devices make inefficient solar converters because their dark currents are relatively large and this diminishes the photovoltaic response. Semiconductor!semiconductor junctions a r e better in this regard. The father of the modern photovoltaic cell is Russell Ohl, a metallurgist at Bell Telephone Laboratories in New Jersey, who observed that crystallisation of a melt of commercial 'high purity' silicon produced a "well-defined barrier having a high degree of photovoltaic response" (Ohl, 1941). This barrier was in fact a p-n junction formed from the unequal distribution of impurities as the Si crystal grew from the melt. From this discovery, after a delay occasioned by World War II, grew the seminal work of Chapin et al. (1954) on the diffused p-n junction in single-crystal silicon and Bell Lab's successful drive to develop photovoltaic devices suitable for use in the infant space industry. The first p-n junctions to be reported, however, were the germanium homojunction of Lark-Horovitz's group at Purdue University (Benzer, 1946, 1947) and the quasi-homojunction formed by pressing together a wafer of lead-enriched lead sulphide with one of sulphur-enriched lead sulphide (Sosnowski et al., 1947). The modern era of silicon photovoltaics is described by Martin Green in Chapter 4, and Fig. 1.1 shows the evolution of silicon cell efficiency. Silicon (Si) is the material with which the electronics industry feels most at home, and Si single-crystal and

6

M. D. Archer

32 28

NREL Multijunction concentrators T 3-junction (2-termina! monolithic) A 2-junction (2-terminal monolithic)

24 20 E

16 -

ARCO

1975

D shar

D" P

Georgia UNSW

Solarex

12 " 8

„

Q-Georgia Tech

Westinghouse

Crystalline Si cells • Single-crystal • Multicrystalline • Thin Si

1980

AstroPower

1985

1995

2000

Year Figure 1.1

Best research cell efficiencies for single-crystal, multicrystalline and thin c-Si cells, and for

multijunction (III—V) concentrator cells. Source: Kazmerski (2000).

multicrystalline homojunction cells dominate the PV market, between them holding -80% of 1999 sales. In the past, the silicon needed by the cell manufacturing industry all came from the 10 ohm cmp-type waste material discarded by the electronics industry, which can provide sufficient good-quality feedstock silicon to make up to about 200 MWp/y of Si solar cells. The PV market is now expanding past this level, so new entrants in the field must seek new sources of silicon feedstock. Despite their longevity, reliability and environmental compatibility, crystalline silicon cells remain relatively complex and heavy devices with significant materials and fabrication costs. One drawback of Si is its relatively poor light absorption, which means that unsophisticated cells must be at least 250 pm thick to absorb all the active wavelengths in sunlight with reasonable efficiency. Surface texturisation of cells to produce light-trapping geometries allows Si cells to be made much thinner (less than 80 //m) and still perform excellently, but it is impossible to use conventional cell fabrication technology to cut such thin wafers from crystal boules. There are various ways of growing thin crystalline Si films directly, but in the past these have led to cells of only modest performance. However, the advanced silicon ribbon and film deposition

The Past and Present

7

technologies, described in Chapter 4, now promise thin Si devices of useful efficiency. Fig. 1.1 shows recent advances in thin c-Si (crystalline silicon) cell efficiency). From the 1970s, when terrestrial applications of crystalline silicon technology began to emerge, there has been a parallel effort to develop semiconductors other than Si in order to make thin-film (polycrystalline) devices of lower cost and better light-absorbing properties. The original motive for investigating thin-film cells was not, however, lower cost but their better power-to-weight ratio for space applications. The first thin-film PV device was the cuprous sulphide/cadmium sulphide (p-Cu2S/«-CdS) heterojunction, made in single-crystal form by Reynolds ef al. (1954), and in thin-film form by Carlson (1956) at the Clevite Research Center, Cleveland, Ohio. The thin-film cell excited much interest because of the simplicity of its manufacture and low intrinsic costs. Clevite Corporation mounted a major development effort on thin-film CdS technology in 1964, and several others followed suit. However, in spite of some promising results, reviewed by Hill and Meakin (1985), these cells suffered from poor stability arising from the high diffusivity of copper, and there were also serious problems in making ohmic contacts to Cu2S. Cadmium sulphide lives on, however, as the window layer of the cadmium telluride and copper indium diselenide cells, despite problems with the use of the toxic metal cadmium in what is intended as an environmentally benign product.5 The Japanese had effectively already delivered the coup de grace to Cu2S/CdS technology by the early 1980s, by commercialising small amorphous hydrogenated silicon (a-Si:H) PV panels of modest but sufficient efficiency to power small consumer goods such as watches and calculators, thus providing PV with an assured market of ~1 MW/y and the cash flow to drive further R&D. Amorphous silicon of good quality (with sufficiently few mid-gap states to be dopable either n- orp-type) had been made by Spear and Le Comber (1975) in Dundee. Independently, David Carlson and Chris Wronski, then both at RCA, made several square centimetre n-i-p andp-i-n cells of-2% efficiency (Carlson and Wronski, 1976), and smaller area MIS cells of 5.5% efficiency. The n-i-p and p-i-n cells were to be the forerunner of modern a-Si:H photovoltaic technology. The Staebler-Wronski effect, which is the -10-20% diminution of efficiency that occurs on the first prolonged exposure of a cell to light, was discovered soon afterwards, in 1977. Puzzling and unwelcome as this was, ways to mitigate its impact by using thin cells (in which this volume recombination effect is diminished) in multijunction, light-trapping structures have been successfully developed, as Wronski and Carlson describe in Chapter 5.

5

CdS also lives on in the paintings of impressionists such as Monet, whose favourite yellow pigment it was.

8

M. D. Archer

While there is still a market for single-junction a-Si:H modules of modest (4-6%) stabilised efficiency in consumer applications where the cost per watt delivered is more important than the watts per unit area, they are being supplanted by dual- and triplejunction devices of much better performance. Figure 1.2 shows the evolution of a-Si:H module efficiency and Fig. 1.3 that of research-cell efficiency. The initial efficiency of the best laboratory triple-junction cells is now -15%, their stabilised efficiency is -12%, and the stabilised efficiency of commercial dual- and triple-junction modules is -10%. Amorphous Si technology has the potential for further cost reduction with the current scale-up of manufacturing facilities, and now seems poised to break into the power market.

14

12

-

i I stabilised efficiency 4

__


6%) p-n GaAs device was the monocrystalline cell of Jenny et al. (1956). The /?-AlGaAs/w-GaAs heterojunction cell was reported by Alferov et al. (1971), and the p- AlGaAs/p-GaAs/«-GaAs heteroface cell, which quickly achieved an AMI efficiency of 15.3%, by Woodall and Hovel (1972). From then on, the story of GaAs for space applications is taken up by Chris Hardingham in Chapter 13, and its use in conjunction with other III-V semiconductors in high-efficiency tandem cells is described by Masafumi Yamaguchi in Chapter 8 (Fig. 1.3 shows some recent efficiency records). Organic semiconductors have in the past been plagued by high resistivity and poor reproducibility, leading to very disappointing efficiencies of

(c)

(b)

(.-i)

Figure 1.8 Darkp-n homojunction cell in the dark (a) at equilibrium; (b) under forward biasF,; (c) under reverse bias Vj, showing the generation and recombination currents as dotted lines and the Fermi levels as red dashed lines.

Figure 1.8 shows how the band bending is affected and a current is caused to flow when a bias voltage Vj is applied across the cell in the dark. At equilibrium (Fig. 1.8a), no net current9 flows through any part of the cell. However, small, balanced tluxes of electrons in the conduction band and holes in the valence band pass each way across the junction. These are referred to as generation and recombination currents. The {thermal) generation currents ih and ie shown in Fig. 1.8a come from the minority carriers (electrons in the p side and holes in the n side) generated throughout the device, albeit at a minuscule rate, by thermal excitation. Those minority carriers that reach the junction without recombining are swept across it in opposite directions by the strong electric field. The recombination currents i£nc and i°rec also shown in Fig. 1.8a come from majority carriers (holes in the/? side and electrons in the n side) that flow 'up' the bandbending barrier (this is energetically unfavourable, but entropically favourable because the carriers move from a region of high to low concentration). At equilibrium, the generation and recombination currents in each band exactly balance each other. The sum of the hole and electron thermal generation currents is called the saturation current density /'„ of the junction. o

h.Ren

e.gen

h.rec

e.rev

'All the currents given the symbol i in Figs. 1.8-1.10 are strictly speaking current densities.

(1.2)

19

The Past and Present

When a forward10 bias voltage Vj is applied across the junction of the dark cell, the barrier height is reduced to q Vb = q( Vb° - Vj), as shown in Fig. 1.8b. This does not affect the generation currents, but it strongly increases the recombination currents. The net current across the junction, which is the difference between the recombination current and the generation current, is called the dark current or junction current zj. >j(Vj) = ih,rec(Vj)

+ i

e,rec(Vj)-kgen-ie,gen = W P

+

K.JVj)

~ ' \rec ~ ' °e,rec 0

-3)

When a reverse bias (Vj < 0) is applied, the barrier height is increased to qVb =

=

")

1 For simplicity we do not here account for a voltage drop due to the series resistance of a solar cell; Vla denotes the portion of the applied voltage that appears across the junction.

43

Device Physics of Silicon Solar Cells The external electron and hole current densities are given by

(2.35) l

l

h

(2.36)

l

h,rec

h,gen

The net current density is the sum of electron and hole currents (2.37)

» = K + ** Using eqs. 2.30 to 2.37 the net current density is therefore given by

Wja)

= W„)(2.95)

The total current density is found by adding the diffusive minority carrier flow at the edges of the depletion regions as in Section 2.3.4. Including the change in current density qgW arising from the generation of electron-hole pairs in the depletion region of width W=Wn+ Wp, we obtain 'total = ie(-Wp)

+

h(Wn)-tj = 0, - <jtj is the potential difference along ltj. A thorough derivation of the discretised electron and hole continuity equations, eqs. 2.4 and 2.5, is given in a thesis by Heiser (1991). The discretised continuity equations are F

"

3

"ST 1 »S[njB^-nMj)]

F," = -Z^-uSlpjBW-pM^

+ V,(r,-gl) = 0 + V^-g,) = 0

(2.140) (2.141)

Device Physics of Silicon Solar Cells

71

where B is the Bernoulli function B{x) =

±—-

(2.142)

(expjc) - 1

and Uy is the mobility, which is assumed to be constant along the box edge perpendicular to / y . To solve the discretised differential equations 2.139-2.141 numerically, the physical entities have to be scaled. For example, carrier concentrations are scaled by the intrinsic carrier concentration, the electrostatic potential 0 is scaled by the thermal voltage Vfl,, and the electric field is scaled by V^/Lt,, where LD is the Debye length (eq. 2.18). This scaling is essential for the numerical calculation because the potential typically varies by one or two orders of magnitude whereas the carrier densities vary over ten to twenty orders of magnitude. For the N nodes of the discretisation mesh, we obtain 3N partial differential equations from eqs. 2.139-2.141 with the solution variables 0 , n and p. These differential equations can be abbreviated as F*(,n,p) = 0 F,*W,n,p) = 0 p

Ft Q,n,p)

(2.143)

=0

These equations can be solved by the Newton method. Given the nonlinear system of equations 2.143 written as F(z) = 0

(2.144)

the Newton procedure iteratively computes a new solution zf+1 = zf + 8skz?

(2.145)

from the old one z*. The update Sskzk is found as the solution of the equation

E^YT 2

5s

"zj = ~mk)

(2-146)

To achieve numerical convergence of the Newton iteration, a damping factor 0

.....A

0.0 1

C

CD

-0.4 h -0.8

\

-1.2 10' 9

i

-i

r

-•

1

•

i

.

•

i

i

•

P

-

i

15

i "

1(b)

10" 10

-

fi,

10" 109 107 10" 10' 10' 0.04

-r i—r i i i i i I-MJ

carrier density [cm

•

n

10' 3

0.03

dark

illuminated

Electron density Hole density

Electron density Hole density

i

i

.

0.01

i

t

(0)

•T" 0.02
<de

Figure 4.7 A family of four related high efficiency solar cell structures: (a) the passivated emitter and rear cell (PERC cell); (b) the passivated emitter, rear locally diffused cell (PERL cell) which took efficiency above 24% in the early 1990s; (c) the passivated emitter, rear totally diffused cell (PERT cell); and (d) the passivated emitter, rear floating junction cell (PERFcell). Source: Green and Hansen (1998).

162

M. A. Green

The final structure shown in Fig. 4.7d, the PERF cell (passivated emitter rear floating junction) offers perhaps the best long-term potential for high performance. This structure has produced the highest open-circuit voltage silicon cells to date with open-circuit voltage up to 720 mV demonstrated under standard test conditions (Wenham et ai, 1994), together with efficiencies above 23%. One feature of these more recent cell designs is the very effective trapping of light within the cell. By depositing metal over the entire rear surface of the cell but ensuring it is displaced from the silicon substrate by an intervening layer of oxide, very high rear reflectance for light striking this rear reflector structure from within the cell is obtained. When combined with appropriate geometrical structure on the front surface of the cell, weakly absorbed light that is reflected from this rear surface can be trapped quite effectively within the cell after total internal reflection from this front surface. This greatly extends the response of the cell to infrared wavelengths. Cells that convert such wavelengths with an efficiency approaching 50% have been demonstrated (Green et ai, 1992).Although oxide passivation remains the most effective technique for passivating cell surfaces yet demonstrated in experimental devices, recent work using wider band-gap amorphous or microcrystalline silicon layers has also produced encouraging results, as has work with specially deposited silicon nitride. A cell using an amorphous silicon emitter layer has been reported to give good performance (Tanaka et ai, 1993), although fundamentally limited to being less efficient than an oxide-passivated cell due to the poor electronic quality and the strong absorption in the amorphous silicon material. Similarly, cells with rear passivation using microcrystalline silicon layers instead of thermal oxide have produced quite encouraging results (Okomoto et ai, 1997) although they are also inherently not capable of matching the optical performance of oxide passivated devices due to parasitic absorption in the microcrystalline layer. Excellent surface passivation properties have also been reported for silicon nitride deposited by a remote plasma approach (Aberle et ai, 1997). How will cell design evolve in the future? Some insight is provided by Fig. 4.8, which shows the calculated intrinsic energy conversion efficiency bounds on single-junction silicon solar cells, with and without 'lambertian' light trapping. In 'lambertian' light trapping schemes, the light direction within the cell is randomised (Green, 1995) allowing path-length enhancements to be quite readily calculated (about 50 in idealised situations). The best laboratory cells have demonstrated close to 85% of the achievable efficiency, according to this figure. In the best experimental devices, performance losses of the order of 5% arise from less than ideal values of each of the short-circuit current, open-circuit voltage and fill factor parameters. The short-circuit current losses are most easily identified and reduced. These come from metal finger

Crystalline Silicon Solar Cells

163

coverage of the top surface, top-surface reflection loss, and less than perfect light trapping in the experimental cells. The voltage loss arises from finite surface and bulk recombination in excess of the lower limit imposed by intrinsic Auger recombination processes (Green, 1984). The fill factor loss comes not only from ohmic series resistance loss within the cell, but also from the same factors producing the opencircuit voltage loss. To eliminate the latter, parasitic recombination must be sufficiently reduced so that the dominant recombination component at the cell's maximum power point is Auger recombination. This is a more challenging requirement than the corresponding criterion at open-circuit voltage (Green, 1984). 3J

'•"'Tambertian

^^-——

/

3D -

/planar, cne pass

1

10 7

n \j

i

r Tvrnr|

10

1 — | | ) ITlfl

T~ I I I M i l l

100

1CO0

'

n - m n

10C00

Thickness (urn)

Figure 4.8 Limiting efficiency of a silicon solar cell as a function of cell thickness with and without lambertian light trapping (global AM1.5 spectrum, 100 mW cm""2, 25 C). Source: Green (1995).

As opposed to the case of laboratory devices, most manufacturers of commercial cells would be very pleased to be producing consistently cells of half the limiting efficiency of Fig. 4.8. Some of the difference between laboratory and commercial cell performance is due to poorer quality of silicon substrate material. A large component, however, is due to limits imposed by the present screen-printing process predominantly used for commercial cell fabrication. The penalty for the processing simplicity offered by this approach is a compromised cell design, since a heavily doped emitter layer appears unavoidable. Improved designs such as the buried-contact cell offer the seemingly contradictory advantages of both higher cell performance and lower overall manufacturing costs.

164

M. A. Green

It seems that eventually it should be feasible to produce low-cost commercial silicon cells of efficiency above 20% with such improved cell designs by paying attention to the passivation of both front and rear surfaces, by thinning the cells to reduce bulk recombination and by modifying the crystal growth processes to produce low-cost silicon customised for photovoltaics, particularly in its ability to withstand high-temperature processing without loss of electronic quality. An interesting result highlighted by Fig. 4.8 is the way that light trapping allows high performance, in principle, from silicon cells that are only 1 \im thick. This provides a justification for expecting very high performance, eventually, from the thin, supported silicon cells discussed in Section 4.7. To approach the limiting performance, the demands on bulk quality become less severe as the cell becomes thinner (Green, 1995). However, those upon light trapping and surface passivation become more severe. Various approaches have been suggested which have the potential, in principle, for exceeding even the efficiency limits of Fig. 4.8. These include the use of tandem cells, the use of high-energy photons to create more than one electron-hole pair (Werner et al, 1994), or the use of sub-band-gap photons in schemes such as incorporation of regions of lower band gap (Healy and Green, 1992), multiple quantum wells (Barnham and Duggan, 1990) or mid-gap impurity levels (Wolf, 1960). Experimentally, the tandem cell approach appears the most likely to have impact in the long term, once the problems with lattice-matching a top cell to silicon with a suitable band gap are overcome. Promising results have been demonstrated with aSi/c-Si tandem cells, although a better performing top cell will be required in the longer term to retain a performance advantage over that offered by the rear cell alone.

4.3 Substrate production 4.3.1 Standard process Not only is silicon solar cell technology capable of benefiting directly from the economies of scale of the silicon microelectronics industry, but also it is capable of using scrap material from this industry because the requirements for material quality in photovoltaics are less demanding than in the more general microelectronics field. Until the photovoltaic industry requires a larger volume of silicon than the microelectronics industry, it will be difficult for approaches customised for photovoltaics to compete with the costs of reject silicon from microelectronics. Accordingly, given the present size relativities, most silicon cells are made from

Crystalline Silicon Solar Cells

165

standard silicon source material originally intended for microelectronics. Over the last ten years, the size relativities have not changed enormously, since both industries have been steadily growing. Explosive growth in the photovoltaics industry, such as that stimulated by urban residential rooftop applications of photovoltaics in 1997, will increasingly upset this delicate balance. For microelectronics, the starting point for producing the requisite high quality "semiconductor grade" silicon is a lower grade of silicon known as "metallurgical grade", produced by the reduction in an arc furnace of quartzite by carbon, the latter generally in the form of wood chips. This metallurgical grade silicon is of about 98% purity and is produced in large quantities for the steel and aluminium industries. A relatively small quantity is refined for microelectronics by conversion to a volatile intermediary that can be purified by fractional distillation. The purified intermediate compound is then decomposed to re-extract the silicon in a highly purified form. Generally the metallurgical grade silicon is converted by hydrochloric acid to trichlorosilane which is then purified to 99.9999999% (nine "nines") purity by fractional distillation. Silicon is then extracted from the trichlorosilane by reducing the latter by hydrogen at high temperature. In this process electrically heated silicon rods are exposed to a trichlorosilane/hydrogen mixture which reacts on the surface of the rods, depositing silicon onto them and hence building up their cross section. These rods grow with a fine-grain polycrystalline silicon microstructure. After the rod diameter has increased to the required size, the process is stopped and the rods mechanically broken into smaller chunks, which maintain "nine-nines" purity. These chunks then become the starting point for the growth of ingots of good crystalline quality. As previously mentioned, crystalline ingots are generally grown by the Czochralski process. In this process, the purified silicon chunks are melted in a quartz crucible along with small pieces of silicon heavily doped with boron. This produces a boron-doped melt into which a seed crystal is dipped and slowly withdrawn (Fig. 4.9a). For high quality crystal growth, good temperature uniformity and slow and steady growth are required. Typically ingots are grown to about 10-15 cm in diameter and 1-2 metres in length, weighing 50-100 kg. The crystallographic orientation of the seed is transferred to the grown crystal. Generally, for photovoltaics, the crystal is grown with a preferred orientation so that the wafers which are sliced from the crystal perpendicular to the growth axis have surfaces parallel to {100} crystallographic planes. Prior to slicing these ingots into wafers, the ingots are generally subject to a centreless grinding operation to remove the slight fluctuations in diameter along the length of the ingot that occur during crystal growth.

M. A. Green

Figure 4.9

(a) Czochralski (CZ) growth; (b) squared-off CZ ingot. Source: Green and Hansen (1998).

Alternatively, the ingots can be "squared-off by sawing off large sections parallel to the growth axis (Fig. 4.9b), giving "quasi-square" wafers after wafering. The large pieces of silicon sawn off in this approach are then generally recycled by re-melting as feedstock for the CZ growth.

Crystalline Silicon Solar Cells

Figure 4.10

167

(a) Inner diameter wafer sawing; (b) continuous wire sawing. After Dietl et al. (1981).

The technique traditionally used in microelectronics for sawing wafers from ingots has been based on the use of inner diameter saws. In this technique, shown Fig. 4.10a, thin metal sheet blades are given dimensional solidity by being held in tension around their outer perimeter. The cutting surface is a diamond-impregnated region surrounding a hole within the tensioned metal sheet. This technique gives excellent dimensional tolerance, although there are limitations arising from the thickness of the silicon wafers that is possible to produce while still maintaining high yield. Other limitations arise from the wastage of silicon as "kerf loss during cutting. Generally, about 10-15 wafers per centimetre of ingot length are achieved by this process. An alternative technique increasingly being used in photovoltaics is based on wire sawing (Fig. 4.10b). In this case, tensioned wire is used to guide an abrasive slurry through the ingot. Advantages are thinner wafers and less surface damage for these wafers as well as lower kerf or cutting loss, allowing the sawing of over 20 wafers per centimetre.

168

M. A. Green

In the 1970s and early 1980s, several other options for preparing silicon feedstock were investigated as part of a large US government PV program encouraged by the Carter administration (Christensen, 1985). A great diversity of alternative routes to producing pure silicon were investigated. These ranged from those involving radically different approaches to those exploring only minor changes from the sequence outlined above, such as the use of different compounds of silicon as the intermediate during the purification process. One such process, based on the use of silane as the intermediate (Christensen, 1985), is now being used commercially, although the product is used exclusively for the microelectronics industry. Parallel development has been conducted outside this program, notably in Germany (Aulich, 1996) and Japan. In Japan, Kawasaki Steel have been investigating an alternative route to preparing cheap silicon feedstock from metallurgical grade precursors and were scheduled to begin pilot production with this sequence in 1998 (Sakaguchi et al., 1997). To produce ingots from the pure silicon feedstock, a modification of the CZ process which produces "tricrystalline" silicon has also been used for photovoltaics (Endros et al, 1997). Wafers cut from the crystals have a different {111} equivalent orientation for each third of their surface, with the differently orientated regions separated by a twinning plane. Claimed advantages of higher growth rates and greater mechanical strength are probably not large enough to offset disadvantages of not being able to chemically texture such wafers and the poor electronic quality near the twinning planes. Another alternative to the standard Czochralski process for producing crystalline ingots is the floatzone (FZ) process. Although some studies have predicted superior economics when compared with the Czochralski process for cell production due to the elimination of consumables such as quartz crucibles, the FZ process, as commercially implemented, is capable of accepting feedstocks only in the form of high quality cylindrical rods. This makes it unsuitable for using low-cost off-grade material. However, the casting and directional solidification processes used to produce multicrystalline silicon are generally extremely tolerant of poor quality feedstock material. These techniques will be discussed in more detail in the following section.

4.3.2 Multicrystalline silicon ingots In 1998, about 30% of the world's photovoltaic production was based on multicrystalline silicon wafers. Several companies have developed commercial processes for producing the precursor multicrystalline silicon ingots (Ferrazza, 1996). Advantages over the Czochralski process are lower capital costs, higher throughput and a higher tolerance to poor feedstock quality.

Crystalline Silicon Solar Cells

169

(b)

Figure 4.11 (a) Directional solidification of silicon within a mould; (b) sawing of large ingot into smaller sub-sections. Source: Green and Hansen (1998). The technique involves controllably solidifying molten silicon in a suitable

container to give silicon ingots with large columnar grains generally growing from the bottom of the crucible upwards (Fig. 4.1 la). Pioneers with this approach for modern photovoltaics in the mid-1970s were Wacker Chemitronic of Germany (Authier, 1978) and Solarex of the USA (Lindmayer, 1976). In the 1980s, other manufacturers including Eurosolare/Crystallox, Kyocera, Bayer, Crystal Systems and Sumitomo Sitex had developed processes capable of producing good quality multicrystalline material. Techniques differ between these manufacturers in the choice of crucible material, the method of loading the crucible with silicon and the method for controlling the cooling of the melt. A good summary can be found elsewhere

170

M. A. Green

(Ferrazza, 1996). The size of a nominally rectilinear ingot can be very large, up to 60 cm x 60 cm x 20 cm, and these ingots can weigh several hundred kilograms (Khattak and Schmid, 1997). The large ingots are sawn into smaller sections as shown in Fig. 4.11b, eventually to give wafers generally 10-15 cm along the sides. These smaller sections can be sawn by the standard inner-diameter or continuous wire sawing processes. The resulting multicrystalline wafers are capable of producing cells of about 80% of the performance of a monocrystalline cell fabricated on a CZ wafer. However, because of the higher packing density possible due to their square or rectangular geometry, this performance difference is largely masked at the module level with multicrystalline module performance lying in the range demonstrated by modules made from monocrystalline cells. An interesting variation on this approach is the continuous casting process such as developed by Sumitomo Sitex. In this case, electromagnetic fields are used to constrain the molten silicon to produce essentially a continuous ingot of multicrystalline silicon (Sarti et al., 1997).

4.3.3 Sheet and ribbon silicon Although there is the potential for substantial cost reductions in both the cost of preparing the silicon feedstock and in forming crystalline or multicrystalline ingots from it, one unavoidable cost with the silicon wafer approach is the cost of sawing the ingot into wafers. Several studies have suggested that the lower bound on this cost may be something of the order of US$0.20/watt (Christensen, 1985; Bruton et al, 1997). This has provided the rationale for investigating approaches that produce silicon directly in the form of self-supporting sheets without the need for sawing (Bergin, 1980; Shulz and Sirtl, 1984). Commercially, the most advanced sheet or ribbon approach is based on the edgedefined film-fed growth (EFG) technique of Fig. 4.12. As originally developed in the early 1970s, this involved the pulling of a thin sheet of silicon ribbon from a strip of molten silicon formed by capillary action at the top of a graphite dye (Fig. 4.12a). Substantially higher throughput was obtained with the more symmetrical configuration shown in Fig. 4.12b, where the ribbon is pulled in the form of a hollow nonagon. Individual wafers are then cut from the sides of the nonagon, normally by laser scribing wafers from each of the sides. The material produced is multicrystalline with elongated grains and of a similar quality to the standard directionally solidified multicrystalline material. Commercial cells made from EFG material have been available sporadically since the early 1980s with a large 25 MW/yr facility recently announced.

171

Crystalline Silicon Solar Cells

iyciystalline ribbon

molten silicon macs by capilary action

carbcnde

A

A

/^L ;en silicon

(a)

(b)

Figure 4.12 (a) Edge-defined, film-fed growth (EFG) method; (b) growth of a nonagonal ribbon of silicon using the EFG method. Source: Green and Hansen (1998).

172

M. A. Green

An even older ribbon growth process is the dendritic web approach of Fig. 4.13 first described by Westinghouse in the 1960s. In this approach, close thermal control is used to cause two dendrites spaced several centimetres from each other to solidify first during the growth step. When these are drawn from the melt, a thin sheet of molten silicon is trapped between them. This quickly solidifies to form a ribbon. After a substantial research and demonstration program by Westinghouse in the 1970s and early 1980s, this approach is now under development by Ebara Solar (Narashima et ai, 1997). A somewhat related approach is the string ribbon approach. In this case, the molten silicon is trapped between two graphite strings that are drawn from the melt. This relaxes the requirement on thermal control, compared with the previous dendritic web approach. The string ribbon approach is under development by Evergreen Solar (Janoch etal, 1997'; Wallace etai, 1997). Another interesting approach that was developed in the 1980s relied on direct casting of silicon wafers using a centrifugal casting approach to overcome surface tension problems within the closely spaced faces of a horizontally aligned graphite mould (Maeda and Hide, 1987). Despite initially promising results, this approach appears to be no longer under active development. A compact but thorough review of most of the above ribbon processes including references is given elsewhere (Shulz and Sirtl, 1984). A thorough bibliography of work prior to 1980 has also been published (Bergin, 1980). Somewhat related to the above ribbon approaches are other sheet approaches which produce silicon films on substrates from which they are subsequently detached. The most developed version of this technology is the VEST technology developed by Mitsubishi (Hamamoto et ai, 1997). In this approach, a potentially reusable silicon substrate is oxidised, then vias are etched in the oxide and then a silicon film is deposited on top of the oxide. This film is subsequently laser-recrystallised. The thickness of this seeding layer is then increased by the subsequent high-temperature epitaxial growth of the silicon layer. After reaching a target thickness of 50-80 |im, the film is detached from the substrate which is then potentially reusable. Promising efficiencies over 16% have been obtained from this approach for substrates that are only 60-70 |xm in thickness, but are still self-supporting. Other researchers have suggested similar techniques to produce even thinner films. Some have suggested the use of a silicon wafer treated to produce a layer of porous silicon along the surface as the source of the crystallographic template for the subsequent growth of a silicon layer, which is then detached (Wenham and Green, 1995; Brendel, 1997). If this detached layer is too thin to be self-supporting, it could be transferred to structurally strong components such as the glass layer in a structural superstrate design. Another

Crystalline Silicon Solar Cells

173

variant involves forming vias through the oxide to a {100) orientated substrate and the subsequent use of liquid phase epitaxy to grow a mesh of silicon on the substrate which again is detached after processing (Weber et al., 1997). In this case, layers of about 70 |Xm thickness are envisaged, although cell processing on the unusual geometries that result would pose obvious challenges. silicon dendrites or carbon string

Figure 4.13 Schematic illustrating either the dendritic web growth process or the string ribbon approach. Source: Green and Hansen (1998).

4.4 Cell processing 4.4.1 Standard process In the previous section, standard and non-standard ways of forming the silicon substrate were described. The major commercial substrates are those formed by the wafering of monocrystalline and multicrystalline ingots, with a much smaller quantity of EFG ribbon substrates produced commercially. Since monocrystalline silicon wafers are the norm, processing for these will be described first with the processing of other types treated as variations on the monocrystalline processing approach.

174

M. A. Green

At present, no photovoltaic manufacturer prepares polysilicon source material. Manufacturers generally purchase off-specification material from the microelectronics industry or, alternatively, bypass the crystal growth step by purchasing silicon wafers. Processing starts by chemically cleaning the starting wafers and etching their surfaces, generally in a sodium hydroxide etchant, to remove saw damage from the wafers. For monocrystalline wafers, the next step is crystallographic texturing, again using sodium hydroxide but in a more dilute solution. The composition and temperature of this solution determines the texturing quality (King and Buck, 1991), including the size of the pyramidal features resulting from the texturing and the percentage of wafer surface area successfully covered by such features. Texturing is a demanding step in cell processing and the quality of texturing varies enormously between different manufacturers. Cell performance, however, is not critically dependent on texturing quality. The next major stage of processing is the diffusion of the cell junction. This is generally achieved by spraying or spinning a compound containing phosphorus onto the cell surface, followed by heating at high temperature to allow phosphorus dopant atoms to seep into the cell surface by thermal diffusion. Typically, the depth of diffusion is less than 1 jum. The same thermal diffusion process is widely used in microelectronics but processing for photovoltaics generally involves cruder equipment and techniques, since the aim is to produce cells at the lowest possible cost without unduly sacrificing cell performance. Although the diffusion is required over only one surface of the wafer and processing techniques are generally chosen to encourage such a result, phosphorus invariably seeps into both wafer surfaces to some extent. To break the connection between the phosphorus diffused into front and rear surfaces, an 'edge junction isolation' step is required to remove the thin phosphorus layer around the edge of the wafer. This isolation is often achieved by 'coin stacking' the wafers so that only their edges are exposed and then placing the stack in a plasma etcher to remove a small section of silicon from the wafer edge, hence breaking the conductive link between front and rear surfaces. The screen printing of metal contacts onto the front and rear surfaces completes cell processing. Silver paste consisting of a suspension of fine particles of silver and glass frit in an organic medium together with appropriate binders (Hoernstra et al, 1997) is squeezed through a patterned screening mesh onto the cell surface. After application, the paste is dried at low temperature and then fired at a higher temperature to drive off the remaining organics and to allow the silver regions to coalesce. The glass frit is important in promoting adhesion to the silicon substrate. Often pastes are doped with phosphorus to help prevent the screened contact from penetrating the thin phosphorus skin that it is intended to contact.

Crystalline Silicon Solar Cells

175

The paste for the top surface is printed in a characteristic finger pattern to minimise the resistive losses in the cell while allowing as much light as possible into it. Sometimes the rear contact is patterned, not to allow light into the cell, but merely to reduce the amount of paste required and hence reduce the cost of this processing step. Sometimes small quantities of aluminium are added to the paste used on the rear surface to give a heavily doped p-type 'back-surface field' region underlying the rear metal contact or, alternatively, separate screening and firing of an Al paste over the entire rear surface is used to more fully optimise this feature (Nijs et ah, 1996). This screen-printing method for applying the metal contact was borrowed in the early 1970s from the hybrid microelectronics industry (Ralph, 1975). This ensured the ready availability of both screen-printing equipment and the paste-firing furnaces suited to this application. Labour and equipment costs associated with this step tend to be very low. However, the pastes themselves can be expensive and an even larger cost penalty is paid for the simplicity of this approach by the forfeiture of the inherently available power output from the silicon wafer, as discussed later. A quarter wave antireflection coating can be applied to the cell at this stage. Generally, titanium dioxide is used as the antireflection coating material due to the simplicity of depositing this compound and its almost ideal refractive index for this application. Some manufacturers deposit the antireflection coating before the metal paste-firing step and fire the paste through this coating. The cells are then ready for testing under a solar simulator. Cells are usually graded based on their short-circuit current or current at a nominal operating voltage, e.g., 450 mV. Generally, cells are sorted into 5% performance bins. This sorting is required to reduce the amount of mismatch within the completed module. To a large extent, the output current of the module is determined by that of the worst cell in the module, resulting in large power losses within mismatched modules. Even worse, low output cells can become reverse-biased under some modes of module operation and destroy the module by localised over-heating. Very similar processing is applied to multicrystalline silicon wafers. In this case, most of the grains will have incorrect orientation for effective texturing by anisotropic etching, although such texturing is sometimes used for the relatively small benefit that can be obtained. However, a quarter-wave interference antireflection coating has been mandatory for good performance from multicrystalline materials. One disadvantage of anisotropic texturing of these materials is that different grains etch at different rates, giving a very uneven surface due to steps at grain boundaries. This can pose hazards for continuity of the subsequently screened metal lines. Accordingly, some manufacturers prefer to etch multicrystalline silicon with an isotropic etch to maintain a smooth surface.

176

M. A. Green

The rippled surface that is a natural consequence of the EFG ribbon growth process poses similar continuity hazards for screen-printed metallisation. To accommodate this rough surface, a novel technique has been developed whereby the metal paste is squeezed through an orifice and then drops to the cell surface, much the same as squeezing toothpaste from its tube onto a toothbrush.

4.4.2 Limitations of the screen-printing approach There are four main limitations arising from the screen-printing approach to applying the front contact which cause the simplicity in processing to be at the expense of cell performance. As noted above, performance can be reduced well below that inherently achievable. One limitation is that the phosphorus diffusion has to be heavier than desirable purely from the point of view of cell performance, to allow reliable low resistance contact between the screen-printed metal and the diffusion. Typically, sheet resistivities of this diffusion less than 60 ohms/square are required (Green, 1995; De Clercq et al., 1997). Such diffusions generally reduce the quality of the silicon in the region near the cell surface where blue wavelengths in sunlight are strongly absorbed. A screen-printed cell does not therefore respond well to blue wavelengths in sunlight, wasting at least 10% of the possible current output through this deficiency. The remaining three limitations relate to the geometry and conductivity of the metal lines it is possible to produce by the standard screen-printing process. Since the paste thickness shrinks to about one-third of its original thickness during firing (silver constitutes only 25-30% by volume of the original paste, with up to 5% glass frit), it is very difficult to achieve metal lines with high aspect ratio (height/width). High aspect ratios are the key to designing metal grids which result in low overall losses (Serreze, 1978). The nature of the screening meshes that have sufficient ruggedness for use in commercial production means it is very difficult to achieve fine lines using screen printing in production. Typically, 150 \im is the minimum width that can be cost-effectively achieved. This limitation means that there will generally be high shading losses in screen-printed cells due to the large percentage (10-15%) coverage of the front surface by the metal. Additionally, the relatively poor conductivity of the fired silver paste—about 2 times lower than that of pure silver for large features such as busbars but up to 6 times lower for finer features such as fingers (de Moor et al, 1997)—fundamentally limits ability to optimise metal contact design, in much the same way as does the low aspect ratio previously discussed. Recent work describes improved laboratory cell performance based on experimental screens formed by cutting patterns in thin metal sheets using a laser (Nijs et al.,

Crystalline Silicon Solar Cells

177

1996). This approach is reported to allow reduced linewidths, although the authors may be overly optimistic about the potential in a production setting (de Moor et al., 1997). With a standard sequence under laboratory conditions, cell efficiency is limited to less than 15% even with these improved linewidths (Nijs et al, 1996). In a more complex sequence involving inherently costly steps such as the deposition of a plasma nitride antireflection coating and a complicated rear Al screening and removal step, efficiency approaching 17% has been confirmed. Some caution is required in accepting the authors' enthusiasm as to how transferable these sequences may be to a production setting. The same authors (Nijs et al, 1996) also analyse cost and performance relative to established commercial sequences such as the buried-contact sequence described below but these analyses appear to be skewed by overly optimistic assumptions about screen-printing metallisation parameters.

4.4.3 Buried-contact solar cells As mentioned in Section 4.2, buried-contact cells were developed as a way of incorporating some of the efficiency improvements demonstrated in the mid-1980s into low-cost commercial cell production sequences. This aim has been successfully realised with recent independent costing studies showing that the buried-contact cell not only produces the highest commercial silicon cell efficiency, but also the lowest cost approach for fabricating commercial silicon cells, of any of those at any reasonable state of development (Bruton et al, 1997). The processing of buried-contact cells begins similarly to that outlined for screenprinted cells. In the commercially most successful buried-contact sequence (Jordan and Nagle, 1994), the incoming wafers are cleaned and textured as with conventional wafers and then diffused. A silicon nitride antireflection layer is grown by chemical vapour deposition over the entire top surface of the cell. Grooves are next formed in this surface through the antireflection coating and prior diffusion. A standard neodynium YAG laser readily produces grooves of about 20 /xm width. The depth depends on the laser power, but desirably lies in the range 20-60 /m\. After etching to clean the grooves, a second diffusion, which is restricted by the nitride to the regions that have been laser-grooved, is performed. Aluminium is then evaporated onto the rear of the wafer and sintered. Electrolessly plated nickel followed by similarly applied copper and silver is then deposited. Again, the insulating nitride restricts the plating to the grooved areas and to the rear of the wafer which has already been metallised by aluminium.

178

M. A. Green

When applied to the same commercial silicon wafers as used in the screen-printing process, cell efficiencies in the 17-18% range are obtained. Figure 4.14 compares the performance of a screen-printed and a buried-contact cell fabricated on the same quality starting material (BP Solar, 1991). A performance advantage of 20-30% is demonstrated by the buried-contact approach, largely as a result of improved shortcircuit current density but with other significant contributions coming from improved fill factor and open-circuit voltage. In addition to this performance advantage under standard test conditions, field studies have shown that buried-contact cells give up to 15% more energy per rated watt as a result of an even larger performance margin at low light intensities and under the bluer light associated with cloudy conditions (Mason etal., 1997). 4

V

^ \

tn

I,. c

I

o

1 •]

^ ^ ^ » Buried Conlacl _ _ _ . Screen-Printed

\ \ \ \ \ \ \ \ \ \ \ \ \

0 . 0.0

0.2

0.4

0.6

Voltage, Volts

Figure 4.14 Output characteristics of buried-contact cells compared with screen-printed cells. After BP Solar (1991).

4.5 Cell costs There have been many studies of the costs of the different stages of silicon cell production using different basic assumptions, particularly in relation to the production volume assumed in the study and the cost of polysilicon source material. Probably the most recent and most authoritative is one conducted under the auspices of the European Union Photovoltaic Program (Bruton et al., 1997). This study involved representatives of seven major European photovoltaic manufacturers and research laboratories, and is valuable for the breadth of representation and the diversity of approaches explored. Since the groups involved are known for their strong views on the virtues of the different sequences studied, the study also involves an undoubtedly

179

Crystalline Silicon Solar Cells

hard-won political and technical consensus of those intimately involved with these issues. The key assumptions of the study were manufacturing volume of 500 MWp of solar cells per annum and the availability of silicon source material at US$25 per kg. A number of different technologies were compared. Important comparisons were between EFG ribbon, multicrystalline and crystalline wafer technologies, between screen-printed, buried-contact, metal-insulator-semiconductor and PERL cell processing sequences, in various combinations of wafers and processing, and between two different module encapsulation approaches. However, the results from all possible combinations were not studied (or, at least, not published), but only the seven selected combinations shown in Table 4.1. Table 4.1 Summary of published results of a European Commission study of manufacturing costs for 500 MWp per year factory ID

Wafer8

Process

Cell efficiency* study (present)

#1 #2 #3 #4 #5 #6 #7

DS CZ CZ CZ CZ CZ EFG

SP SP LGBC MIS/A MIS/B PERL SP

15% (12.6-14.8%) 16% (13.9-15.6%) 18% (16.5-17.5%) 17% (N/A) 17% (12.2%) 20% (N/A) 14.4% (12%)

Estimated cost (ECU/Wp)' 0.91 1.25 1.15 1.28 1.34 1.78 0.71

Key variable Wafer Wafer/process Process Process/module Module Process Wafer

°DS: directional solidification; CZ: Czochralski growth; EFG: edge-defined film-fed growth; SP: screenprinted; LGBC: laser grooved, buried-contact; MIS/A: metal-insulator-semiconductor; MIS/B: as for MIS/A but with resin-fill packaging; PERL: passivated emitter, rear locally diffused (less appropriate acronym LBSF used in study). 'The cell efficiencies assumed in the study in some cases differ appreciably from present average production values, deduced by the present author from manufacturers' data sheets or the results from large field installations. C1ECU « US$1.2. Source: Bruton etal., 1997.

Several key results can be deduced from this Table. When comparing screenprinted cells on ribbon (EFG), multicrystalline (DS) and monocrystalline (CZ) wafers, the ribbon produces the lowest cost of 0.71 ECU/WP followed by the multicrystalline wafers at ECU0.91/Wp and the monocrystalline wafers at ECU1.25/Wp. The advantage of the ribbon stems almost entirely from the fact that it does not need to be sawn, as previously mentioned.

180

M. A. Green

Comparing between the different processing approaches on single crystal wafers, the cheapest is the buried-contact at 1.15 ECU/Wp, followed by the screen-printed at 1.25 ECUAVp, followed by the metal-insulator-semiconductor at 1.28 ECUAVp, followed by the PERL at 1.78 ECUAVp. The buried-contact achieves its cost advantage over the screen-printing approach by virtue of the increased efficiency giving more power per unit processing area. Such advantages would transfer to the less expensive substrate approaches studied, suggesting the best possible combination of wafer, process and moduling approach would result in manufacturing costs well below 0.70 ECUAVp. In the module area, the standard laminated module approach is calculated to be slightly cheaper than an alternative resin-fill approach. Compared to the predictions of this study, present manufacturers fabricate screen-printed monocrystalline and multicrystalline cells and buried-contact monocrystalline cells in roughly 10-20 MWp per year production capacities with large-volume selling prices of modules in 1999 of about US$4AVp (a similar amount in ECUAVp). Present encapsulated cell efficiencies, deduced by the present author mainly from manufacturers' data sheets or from recent field performance, are also shown in Table 4.1, indicating the various levels of extrapolation in cell performance assumed for the different cell technologies in the study.

4.6 Opportunities for improvement 4.6.1 Commercial cells The large differential between the efficiencies of a typical screen-printed commercial cell (15%) and the best laboratory silicon cell (24%) shows the enormous potential for further efficiency improvement in commercial devices. Part of this potential has been recently realised with the commercialisation of the buried-contact cell with cell efficiencies in the 17-18% range obtained in production. There remains scope for a further substantial performance improvement. One reason for the difference between laboratory and the best commercial cells is the difference between the CZ wafers used in commercial production and the FZ wafers used for the best laboratory cells. CZ grown wafers are invariably contaminated with oxygen and carbon during growth to a much higher level than FZ wafers, due to use of quartz crucibles and graphite heaters in the CZ process. These impurities give rise to a much more subtle dependence on processing conditions, in the CZ material, of an important silicon material property for producing high performance cells, the minority carrier diffusion length. For example, applying the

Crystalline Silicon Solar Cells

181

high temperature processing associated with PERL type sequences to CZ silicon gives a large spread in results depending on the supplier of CZ material and hence most probably on the oxygen and carbon content (Wettling et al, 1996). Additionally, the quality of CZ material as compared with FZ falls off quite rapidly as the boron content is increased, possibly because boron/oxygen complexes form within the material. This reduces flexibility in cell design since it eliminates the possibility of using low resistivity CZ substrates. In microelectronics, high oxygen content resulting from the CZ process is regarded as an asset. Oxygen increases the mechanical strength of the wafers as well as allowing gettering of surface regions, where the operational microelectronic devices are confined, by the precipitation of oxygen defects away from the wafer surfaces. A simple option for improving the suitability of CZ material for photovoltaics may be merely to change the crucible material used in the CZ process. For example, experiments have been conducted with silicon nitride coated crucibles as a way of reducing oxygen content within the material while increasing that of nitrogen (Shimura, 1989). However, relatively little exploration of such possibilities has been undertaken, probably because the benefits would not, in any case, be seen with the standard screen-printing approach. More sophisticated cell processing sequences would be required (such as offered by the buried-contact approach) to obtain the full benefits from such improved temperature tolerance. As opposed to the case of CZ silicon, much experimentation has been conducted with directionally solidified multicrystalline silicon. It may well be that multicrystalline silicon eventually exceeds the standard CZ material in its performance potential for photovoltaics due to the eventually better high temperature tolerance of the material. For example, the recent demonstration of 19.8% efficiency upon multicrystalline silicon (Zhao et al, 1998) puts the performance of this material right in the middle of the performance range observed with a similar sequence using CZ wafers (Wettling et al, 1996). The trend towards thinner cells that arises primarily from efforts to reduce the costs of the silicon wafer may actually help to improve the cell efficiency. Thin wafers give the opportunity for back-surface fields or other rear-surface passivation approaches to be used to improve cell performance, primarily through increased voltage output. Again, the buried-contact processing sequence would be capable of realising such potential performance advantages due to its high open-circuit potential, which is largely untapped in wafers of standard thickness. Bifacial cell designs offer another way of effectively improving cell efficiency. Recent studies suggest that module output can be improved by approximately 20% in standard open back configuration without any special effort if use can be made of light scattered onto the rear of the module (Chieng and Green, 1993). However, as cells are

182

M. A. Green

now being increasingly used in the residential market where rear illumination of the module is unlikely, only part of the market would benefit from this improvement. Japanese groups in particular are showing increasing interest in amorphous and microcrystalline silicon-based surface passivations as demonstrated by the 'HIT' cell structure (Tanaka, et al., 1993; Sawada et al., 1994), as shown in Fig. 4.15. This cell structure has demonstrated an open-circuit voltage capability similar to that of the buried-contact approach. However, it is inherently incapable of giving a similar current output due to light absorption in the 'transparent' conducting oxide layer required to give lateral conductivity to the amorphous silicon emitter as well as the less than 100% collection efficiency from the latter region.

Figure 4.15 HIT (Heterojunction with Inlrinsic Thin Layer) cell on textured crystalline silicon substrate. After Green and Hansen (1998).

Given better feedstock material or cells below 150 u.m in thickness, improved rearsurface passivation approaches such as demonstrated by the PERC and PERL cells of Fig. 4.7 could become appropriate. A promising start has been made with the double sided buried-contact cell which applies high quality oxide passivation to both top and rear surfaces (Green, 1995). Other options may be rear passivation layers based on amorphous, microcrystalline or polycrystalline silicon (Okamoto et al., 1997), or on specially deposited silicon nitride (Aberle et al., 1997).

Crystalline Silicon Solar Cells

183

4.6.2 Laboratory cells For laboratory cells, an appropriate reference point for performance is the AM 1.5 detailed-balance efficiency limit of 33% for material of the band gap of silicon. However, it has been shown that another intrinsic process, Auger recombination, provides a more severe fundamental limit for silicon than the radiative recombination processes assumed in the detailed-balance limit (Green, 1984; Tiedje et al., 1984). Unlike the detailed-balance limit, the Auger limit for a silicon cell is dependent on the cell thickness, as shown in Fig. 4.8 (Green, 1995). This difference arises because the detailed-balance calculation includes photon recycling which makes nett recombination rates independent of cell volume. With lambertian light trapping, the optimum cell performance in the Auger limit is 29% for a cell of about 80 |im thickness. Such a cell would have an open-circuit voltage of about 760 mV, higher than the highest value ever demonstrated for silicon of 720 mV. The voltages of these best performing experimental devices demonstrating 720 mV were limited by surface recombination rather than bulk recombination. Figure 4.16 shows the results of efficiency calculations with various amounts of surface recombination added, characterised in terms of the open-circuit voltage limit that this recombination would impose if it were the only recombination process in the cell. Increasing surface recombination reduces the value of the obtainable efficiency as well as pushing the optimum cell thickness to larger values.

1

10

100 Thickness (um)

1000

10000

Figure 4.16 Limiting efficiency of silicon cell with lambertian light trapping as a function of surface recombination velocity, characterised in terms of the voltage limit imposed by this recombination. Source: Green (1999).

184

M. A. Green

Figure 4.16 makes it clear that, to improve silicon cell efficiency much beyond 26%, improved surface passivation (both surfaces) is essential beyond the 720 mV capability presently demonstrated. If this improved quality cannot be achieved, an alternative possibility is to maintain the same quality of surface passivation as presently demonstrated and reduce the effective threshold energy of the photovoltaic process within the bulk regions of the cell. Techniques such as alloying sections of the cell with germanium to reduce its band gap (Healy and Green, 1992) or doping with a photoactive impurity to give impurity photovoltaic effects in the bulk region (Keevers and Green, 1994) have been suggested and shown, in some cases, to have theoretical advantages. However, no experimental performance advantage has been demonstrated by either technique to date. A well-proven approach for improving solar cell efficiency is the use of the tandem cell structure. Efforts to produce tandem cells with silicon have not yet given good results due to the inability to find a suitable wide band-gap partner that is latticematched to silicon (Corkish, 1991). For low quality cells, amorphous silicon/ polycrystalline silicon tandems have given improved results over either cell type alone (Yamamoto et al, 1997; Shah et al, 1997). As the quality of the polycrystalline lower cell in this combination improves, however, it is doubted that this situation will continue (see Section 4.7). A higher performance top cell will eventually be required, which could be provided by a crystalline compound cell if difficulties with lattice mismatch to the silicon substrate can be overcome. Fuller use of the available photon energy by incorporating efficient impact ionisation process has been suggested as a way of boosting cell performance by generating more than one electron-hole pair from one high energy photon (Kolodinski et al, 1993; Werner et al, 1994). However, such processes are quite weak in silicon with increases in current density limited to less than 0.1mA cm-2 (Green, 1987). Manipulating the details of the band gap of silicon, for example by alloying with germanium; may improve prospects. However, since the high-energy photons of most interest for this process are absorbed very close to the surface of silicon, such approaches may interfere with the ability to obtain well-passivated surfaces. Limited experimental work with shallow germanium implants has not given any nett performance benefit (Keevers et al, 1996). More advanced concepts such as the multiple quantum wells discussed in Ch. 10 might also be appropriate (Barnham and Duggan, 1990). A recent area of interest has been the use of ZnS/Si multiple quantum wells. Not only is there a good lattice match between ZnS and Si, but recent studies suggest is may be easier to obtain direct bandgap-like properties from such multiple quantum wells than the Si/Ge alternative which has been the focus of most past study. There is still some question as to whether or not

Crystalline Silicon Solar Cells

185

a multiple quantum well device offers a performance advantage above the detailedbalance limit (Araujo et al., 1994), but this approach almost certainly offers performance advantage above the Auger limit for silicon, in principle.

4.7 Silicon-supported thin films There has long been an interest in transferring the strengths demonstrated by crystalline silicon wafer technology to cells based on silicon thin films. Historically, work can be divided into two phases: (i) that before the 1980s when the benefits of light trapping were not fully appreciated; and (ii) that after the mid-1980s where light trapping has been regarded as an essential feature of any silicon thin-film cell design. The early work laboured under what is now known to be a misconception that quite thick layers (>20 |xm) of silicon would be required to give reasonable performance due to silicon's poor absorption characteristics arising from its indirect band gap (see Fig. 4.8). However, since light trapping can increase the effective optical thickness of a silicon cell by 10-50 times, this means that layers of only 1 jjm or so thickness are still inherently capable of producing similar performance to much thicker layers. Approaches to producing supported silicon films can be divided into hightemperature and low-temperature strategies depending on whether or not the substrate is heated to high temperature during the silicon deposition or subsequent processing.

4.7.1 High-temperature supported films One of the earliest silicon supported film approaches was the 'silicon-on-ceramic' approach (Christensen, 1985) whereby a ribbon of ceramic material was dipped into a molten silicon bath or pulled across the surface of a silicon melt so that one side was coated with silicon. This produced silicon of modest quality and the approach suffered from difficulties in making rear contact to the cells, since the ceramics used were insulating. This approach was discontinued in the early 1980s. Early work by Ting and Shirley Chu involved the deposition of silicon onto a range of foreign substrates by high temperature chemical vapour deposition (Chu, 1977). Operational cells were obtained using a number of substrate materials. The best results were obtained by depositing the silicon layers on multicrystalline silicon substrates prepared from metallurgical grade silicon. Given the previous studies that have shown that sawing of wafers represents one of the major costs in any wafer-type approach, the overall economics of such an approach using a wafer substrate are questionable, regardless of

186

M. A. Green

the quality of this substrate. Other early work involved the deposition of silicon onto ceramic substrates by high temperature CVD and the subsequent increase in crystal size by melting and directional solidification (Minagawa et al., 1976). In the post-1980 era, efforts in silicon supported film were revitalised by the US company AstroPower (Barnett et al, 1985). In this company's approach, silicon is deposited onto a conducting ceramic substrate by a technique which has not been disclosed but which has been reported to involve the deposition of silicon from solution in molten metal as one step (Barnett et al, 1989). The resulting ceramicbased wafers can then be processed in the same way as a multicrystalline silicon wafer. Efficiency up to 16.6% has been confirmed for films that are apparently in the 50-100 [im thickness range (Bai et al, 1997). Efforts are underway to include more effective light trapping into these devices and to produce an integrated module upon insulating ceramic (Ford et al., 1997). More recently, promising laboratory results have also been obtained by a German collaborative effort using much thinner films. These films were formed by first depositing and recrystallising a thin silicon layer upon a silicon carbide coated graphite substrate followed by the deposition of an epitaxial layer of silicon of about 30 ixm thickness upon this recrystallised layer. Cell efficiency above 11 % has been confirmed for a cell of this 30 ^m thickness (Ltidemann et al, 1997).

4.7.2 Low-temperature approaches One of the first papers addressing silicon photovoltaic thin films described the deposition of silicon by low temperature chemical vapour deposition onto an aluminium substrate (Fang et al., 1974). A surprisingly large grain size was obtained, attributed to eutectic reaction with the aluminium. In more recent times, laser crystallisation has been used in the active matrix liquid crystal display industry to produce relatively small-grain polycrystalline silicon films from amorphous silicon precursors, generally deposited by low-pressure chemical vapour deposition. Grain sizes are typically less than a micron or so, so that these films would probably not be suitable for photovoltaics. Also, thicknesses for the active matrix display industry tend to be only about 100 nm, which would be too thin for photovoltaic application. From 1989, a group at Sanyo explored the use of low-temperature solid-phase crystallisation of amorphous silicon as a technique for producing thin-film polycrystalline silicon cells. Good results have been obtained with 9.2% (unconfirmed) efficiency reported in 1995 (Baba et al., 1995). These cells were approximately 1 cm2 in area deposited onto a textured metallic substrate and heated at

Crystalline Silicon Solar Cells

187

approximately 600 C for many hours to enable the crystallisation of the originally amorphous films. After crystallisation, the HIT structure developed by Sanyo is used to complete the cell processing at low temperature. Two groups have reported good results using silicon thin films deposited directly in microcrystalline form onto glass substrates. The University of Neuchatel has reported unconfirmed efficiencies of about 7% for 3 /zm thick microcrystalline cells deposited at 500 C (Shah et al., 1997). The cell has a p-i-n structure with the intrinsic region comprising most of the device thickness. The cell is designed for this intrinsic region to be depleted during normal device operation to create a high electric field to aid carrier collection, as with a standard amorphous silicon cell. Finally, Kaneka Corporation (Yamamoto et al., 1997) has reported efficiencies over 10% with a similar device structure, shown in Fig. 4.17. Nearly the same efficiency was obtained when the total device thickness was varied over the 1.5-3.5 /urn range. Both the above groups have reported even higher efficiencies when amorphous silicon cells are used in a tandem configuration on top of the microcrystalline device. Given the relatively small amount of effort so far dedicated to this area, these results are extremely encouraging, and show the enormous potential of such low temperature approaches.

/ poly-S[

Figure 4.17 Structure of 9.4% efficient thin-film microcrystalline solar cell developed by Kaneka. Cell thickness is typically 1-3 //m. After Yamamoto el al. (1997).

188

M. A. Green

4.7.3 Multilayer solar cells The parallel multilayer solar cell shown in Fig. 4.18 provides an effective cell design when dealing with low quality silicon such as obtained by low temperature approaches (Green and Wenham, 1994). This cell differs from a tandem cell because all junctions are connected in parallel. The active region of any cell is the depletion region straddling the metallurgical junction and the region within a diffusion length on either side of this depletion region. The challenge in producing a thin supported silicon device of high performance is therefore in having material quality sufficiently good for this active volume to be wide enough to result in a large amount of current collection. In the microcrystailine work previously reported, attempts have been made to enlarge this active region by expanding the junction region by making this region as lightly doped as possible. The multilayer approach provides an alternative (or complementary) way of achieving the same result. By having multiple p-n junctions dispersed throughout the material, it is possible to make the whole volume of material electronically active regardless of material quality. The approach is particularly appropriate when the parallel layers are very heavily doped allowing unique thin-film cell and module designs where the lateral conductivities of the doped layers are sufficiently high to allow lateral current flow without appreciable resistance loss. This removes the need for transparent conducting oxides used in other thin-film technologies to provide this lateral conductance.

Figure 4.18 Parallel multilayer cell schematic. The red and while layers correspond to different doping polarities. Each layer is thinner than the minority carrier collection distance. Source: Green and Hansen (1998).

Crystalline Silicon Solar Cells

189

Figure 4.19 Fabrication of UNSW multilayer cells: 1. Glass superstratc; 2. Multilayer deposition; 3. First polarity groove; 4. Second polarity groove; 5. Metallisation. The cell is designed to be illuminated from the glass side (the underside in this schematic) although bifacial operation is feasible. Opposite polarity regions in adjacent cells are connected, providing automatic series connection within the module. Source; Green and Hansen (1998).

At UNSW, the parallel muitijunction approach has been combined with the buriedcontact approach to produce the device fabrication sequence shown in Fig. 4.19. After deposition of a multilayer stack on a low temperature substrate such as glass, laser grooving and groove doping of one polarity is applied to connect all the layers of this polarity together in parallel. A second laser grooving and doping step involving the other polarity follows. By aligning to the first step, series interconnection of the cells is also achieved in a very elegant process. Pacific Solar in Sydney is working on commercialising this process, although few details have yet been published (Pacific Solar, 1997).

4.8 Summary Although crystalline silicon devices have dominated the commercial marketplace for photovoltaics for more than two decades, there still remains scope for considerable improvement in both the performance and cost of these cells. Recent studies suggest that manufacturing costs well below US$1 AVp are obtainable in manufacturing volumes of 500 MWp per year, without major changes in present processing sequences. This suggests module costs will steadily decrease from present values of about US$4/Wp as manufacturing volumes continue to increase. The average energy

190

M. A. Green

conversion efficiency of product sold should also increase due largely to the increasing market share of high efficiency approaches, particularly the buried-contact cell approach. The trend towards thinner silicon wafers to decrease wafer cost is compatible with ongoing increases in cell efficiency provided cell structures as effective as the buried-contact approach are adopted and improved methods are demonstrated in production for passivating the rear surface of the cell. Substrates with more consistent high temperature performance than standard CZ grown silicon may be required to allow the full performance potential of the silicon wafer approach to be obtained. Particularly promising progress has been made in this area with multicrystalline silicon over recent years. This enormous potential for both performance and cost reduction will make these bulk silicon approaches an increasingly challenging target for the thin-film approaches currently under development. In this context, excellent recent progress has been made with supported silicon film. Films processed at low temperature on substrates such as glass have made exceptional gains in the laboratory over the last 2-3 years and offer great promise for stable low cost thin-film cell technology for the future. Innovative cell design such as the parallel multilayer cell will allow the particular properties of thin-film silicon to be used to their full advantage.

References Aberle A. G., Lauinger T. and Hezel R. (1997), 'Remote PECVD silicon—a key technology for the crystalline silicon PV industry of the 21st. century?', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 684-689. Araujo G. L., Marti A., Ragay F. W. and Wolter J. H. (1994), 'Efficiency of multiple quantum well solar cells', Proc. 12th. European Photovoltaic Solar Energy Conf., Amsterdam, H. S. Stephens & Associates, Bedford, 1481-1484. Aulich H. A. (1996), 'Review of solar-grade silicon research programs carried out in Germany during the 80s', 9th. Sunshine Workshop on Crystalline Silicon Solar Cells, unpublished. Authier B. (1978), 'Polycrystalline silicon with columnar structure', Festkorperprobleme XVIII, F. Vieweg & Son, Wiesbaden. Baba T., Shima M , Matsuyama T., Tsuge S., Wakisaka K. and Tsuda S. (1995), '9.2% efficiency thin-film polycrystalline solar cell by a novel solid phase crystallisation method', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 1708-1711.

Crystalline Silicon Solar Cells

191

Bai Y., Ford D. H., Rand J. A., Hall R. B. and Barnett A. M. (1997), '16.6% efficient Silicon-Film™ polycrystalline silicon solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 35-38. Barnett A. M., Hall R. B., Fardig D. A. and Culik J. S. (1985), 'Silicon-film solar cells on steel substrates', Conf. Record 18th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 1094-1099. Barnett A. M., Dumian F. A., Ford D. H., Kendall C. L., Rand J. A., Rock M. L. and Hall R. B. (1989), 'Thin silicon-film solar cells on ceramic substrates', Proc. 4th. Int. Photovoltaic Solar Energy Conf, Sydney, Inst. Radio and Electronics Engineers, 151-158. Barnett A. M., Ford D. H., Checchi J. C , Culik S., Hall R. B., Jackson E. L., Kendall C. L. and Rand J. A. (1997), 'Very-large-area Silicon-Film™ solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 999-1002. Barnham K. W. J. and Duggan G. (1990), 'A new approach to high-efficiency multiband-gap solar cells', J. Appl. Phys. 67, 3490-3493. Bergin D. O. (1980), 'Shaped crystal growth—a selected bibliography', J. Crystal Growth 50, 381-396. BP Solar (1991), Data Sheet, BP Saturn Solar Cells. Brandhorst H. W. and Bernatowicz D. T. (1980), 'Space solar cells—high efficiency and radiation damage', Conf. Record 14th. IEEE Photovoltaic Specialists Conf, San Diego, IEEE Press, Piscataway, 667-671. Brendel R. (1997), 'A novel process for ultrathin monocrystalline silicon solar cells on glass', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 1354-1357. Bruton T. M., Luthardt G., Rasch K.-D., Roy K., Dorrity I. A., Garrard B., Teale L., Alonso J., Ugalde U., Declerq K., Nijs J., Szlufcik J., Rauber A., Wettling W. and Vallera A. (1997), 'A study of the manufacture at 500 MWp p.a. of crystalline silicon photovoltaic modules', Proc. 14th. European Photovoltaic Solar Energy Barcelona, H. S. Stephens & Associates, Bedford, 11-16. Chapin D. M., Fuller C. S. and Pearson G. O. (1954), 'A new silicon p-n junction photocell for converting solar radiation into electrical power', J. Appl. Phys. 25, 676-677. Chieng C. and Green M. A. (1993), 'Computer simulation of enhanced output from bifacial photovoltaic modules', Prog, in Photovoltaics 1, 293-299. Christensen E. (1985, ed.), Flat Plate Solar Array Project, Final Report, Jet Propulsion Laboratory.

192

M. A. Green

Chu T. L. (1977), 'Silicon films on foreign substrates for solar cells', J. Crystal Growth 39, 45-60. Corkish R. (1991), 'Some candidate materials for lattice-matched liquid phase epitaxial growth on silicon', Solar Cells 31, 537-548. De Clercq K., Frisson L., Szlufcik J., Nijs J. and Mertens R. (1997), 'Design for manufacturability of a silicon solar cell process based on screen printing by means of manufacturing science techniques', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 135-138. Dietl J., Helmreich D. and Sirtl E. (1981), "Solar' silicon', in Silicon, Vol. 5 of Crystal Growth Properties and Applications, pp. 43-107. Endros A. L., Einzinger R. and Martinelli, G. (1997), Tri-silicon: a novel substrate for thin wafer solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 112-114. Fang P. H., Ephrath L. and Nowak W. B. (1974), Appl. Phys. Lett. 25, 583. Ferrazza F. (1996), 'Growth and post-growth processes of multicrystalline silicon for photovoltaic use', in Polycrystalline Semiconductors IV—Physics, Chemistry and Technology, Pizzini S., Strunk H. P. and Werner J. H., eds., Vol. 51-52, Solid State Phenomena, Scitech Publ., Zug, Switzerland. Fischer D., Keppner H., Kroll U., Torres P., Meier J., Platz R., Dubail S., Selvan J. A. A., Vaucher N. P., Ziegler Y., Tscharner R., Hof C , Beck N., Goetz M., Pernet P., Goerlitzer M , Wyrsch N., Vuille J., Cuperus J. and Shah A. (1997), 'Recent progress of the "Micromorph" tandem solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 2347-2350. Fisher H. and Pschunder W. (1976), 'Low cost solar cells based on large area unconventional silicon', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 86-82. Ford D. H., Rand J. A., Barnett A. M., DellDonne E. J., Ingram A. E. and Hall R. B. (1997), 'Development of light-trapped, interconnected, silicon-film modules', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 631-634. Godlewski M. P., Baraona C. R. and Brandhorst H. W. (1973), 'Low-high junction theory applied to solar cells', Conf. Record 10th. IEEE Photovoltaic Specialists Conf, Palo Alto, IEEE Press, Piscataway, 40-49. Green M. A. (1984), 'Limits on the open-circuit voltage and efficiency of silicon solar cells imposed by intrinsic Auger processes', IEEE Trans. Electron Devices, ED31,671-678.

Crystalline Silicon Solar Cells

193

Green M. A. (1987), High Efficiency Silicon Solar Cells, Trans Tech Publications, Aedermannsdorf, Switzerland. Green M. A., Zhao J., Wang A. and Wenham S. R. (1992), '45% efficient silicon photovoltaic cell under monochromatic light', IEEE Electron Devices Lett. 31, 317-318. Green M. A. and Wenham S. R. (1994), 'Novel parallel multijunction solar cell', Appl. Phys. Lett. 65, 2907-2909. Green M. A. (1995), Silicon Solar Cells: Advanced Principles and Practice, Bridge Printery, Sydney. Green M.A. and Hansen J. (1998), Catalogue of Photovoltaic Drawings, Photovoltaics Special Research Centre, University of New South Wales, Sydney. Green M. A. (1999), 'Limiting efficiency of bulk and thin-film silicon solar cells in the presence of surface recombination', Prog, in Photovoltaics 7, 327-330. Guttler G. and Queisser H. J. (1970), 'Impurity photovoltaic effect in silicon', Energy Conversion 10, 51-55. Hamamoto S., Morikawa H., Naomoto H., Kawama Y., Takami A., Arimoto S., Ishihara T. and Namba K. (1997), 'Development of the VEST cell process for low cost fabrication', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 2328-2332. Haynos J., Allison J., Arndt R. and Meulenberg A. (1974), 'The COMSAT nonreflective silicon solar cell: a second generation improved cell', Int. Conf. on Photovoltaic Power Generation, Hamburg, 487. Healy S. A. and Green M. A. (1992), 'Efficiency enhancements in c-Si solar cells by the incorporation of a region alloyed with germanium', Solar Energy Mat. Solar Cells 28, 273-284. Hoernstra J., Weeber A. W., de Moor H. H. C. and Sinke W. C. (1997), 'The importance of paste rheology in improving fine line, thick film screen printing of front side metallization', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 823-826. Janoch R., Wallace R. and Hanoka J. I. (1997), 'Commercialization of silicon sheet via the string ribbon crystal growth technique', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 95-98. Jordan D. and Nagle J. P. (1994), 'New generation of high-efficiency solar cells: development, processing and marketing', Prog, in Photovoltaics 2, 171-176. Keevers M. J. and Green M. A. (1994), 'Efficiency improvements of silicon solar cells by the impurity photovoltaic effect', J. Appl. Phys. 75,4022^*033.

194

M. A. Green

Keevers M. J., Zhang G. C , Saris F. W., Zhao J. and Green M. A. (1995), 'Screening of optical dopants in silicon solar cells for improved infrared response', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 1215-1218. Khattak C. P. and Schmid F. (1997), 'Growth and characterization of 200 kg multicrystalline silicon ingots by HEM', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 111-114. King D. L. and Buck E. M. (1991), 'Experimental optimization of an anisotropic etching process for random texturization of silicon solar cells', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, SOSSOS. Kingsbury E. F. and Ohl R. S. (1952), 'Photoelectric properties of ionically bombarded silicon', Bell Syst. Tech. J. 31, 802-815. Knobloch J., Glunz S. W., Biro D., Warta W., Schaffer E. and Wettling W. (1996) 'Solar cells with efficiencies above 21% processed from Czochralski grown silicon', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 405^108. Kolodinski S., Werner J. H., Wittchen T. and Queisser H. J. (1993), 'Quantum efficiencies exceeding unity due to impact ionization in silicon solar cells', Appl. Phys. Lett. 63, 2405-2407. Lew Yan Voon L. C., Ram-Mohan L. R. and Soref R. A. (1997), 'Electronic and optical properties of (001) Si/ZnS heterostructures', Appl. Phys. Lett. 70, 1837— 1839. Lindmayer J. and Allison J. (1973), 'The violet cell: an improved silicon solar cell', COMSAT Tech. Rev. 3, 1-22. Lindmayer J. (1976), 'Semi-crystalline silicon solar cells', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 82-85. Liidermann, R., Schaefer S., Schiile C. and Hebling C. (1997), 'Dry processing of mcsilicon thin-film solar cells on foreign substrates leading to 11% efficiency', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 159-162. Mason N. B., Bruton T. M. and Heasman K. C. (1997), 'Factors which maximise the kWh/kWp performance of PV installations in northern Europe', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 2021-2024. Maeda Y. and Hide I. (1987), 'Solar cells using spinning Si polycrystals', Tech. Digest, 3rd. Int. Photovoltaic Sci. Eng. Conf, 87-92.

Crystalline Silicon Solar Cells

195

Minagawa S., Saitoh T., Warbisako T., Nakamura N., Itoh H. and Tokuyama T. (1976), 'Fabrication and characterization of solar cells using dendritic silicon thin films grown on alumina ceramic', Conf. Record 12th. IEEE Photovoltaic Specialists Conf., Baton Rouge, IEEE Press, Piscataway, 77-81. de Moor H. H. C , Hoornstra J., Weeber A. W., Burgers A. R. and Sinke W. C. (1997), 'Printing high and fine metal lines using stencils', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 404^07. Narashima S., Crotty G., Krygowski T., Rohatgi A. and Meier D. L. (1997), 'Backsurface field and emitter passivation effects in the record high efficiency n-type dendritic web silicon solar cell', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 235-238. Nijs J. Demesmaeker E., Szlufcik J., Poortmans J., Frisson L., de Clercq K., Ghannam M., Mertens R. and van Overstraeten R. (1996), 'Recent improvements in the screen-printing technology and comparison with the buried-contact technology by 2D-simulation', Solar Energy Mat. Solar Cells 41-42, 101-118. Ohl R. S. (1941), 'Light sensitive electric device', US Patent 2,402,622 (27 March), 'Light-sensitive electric device including silicon', US Patent 2,443,542 (27 May). Okamoto S., Nishida M., Shindo T., Komatsu Y., Yasue S., Kaneiwa M. and Nanmori T. (1997), '23.5% efficient silicon solar cell with rear micro contacts of c-Si/mcSi:H heterostructure', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 255-258. Pacific Solar Pty. Ltd. (1997), Annual Review, Sydney. Ralph E. L. (1975), 'Recent advancements in low cost solar cell processing', Conf. Record 11th. IEEE Photovoltaic Specialists Conf, Scottsdale, IEEE Press, Piscataway, 315-316. Riordan M. and Hoddeson L. (1997), Crystal Fire: The Birth of the Information Age, Norton, New York. Rohatgi A., Narasimhi S., Kamra S., Doshi P., Khattak C. P., Emery K. and Field H. J. (1996), 'Record high 18.6% efficient solar cell on HEM multicrystalline material', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 741-744. Sakaguchi Y., Yuge N., Nakamura N., Baba H., Hanazawa K., Abe M. and Kato Y. (1997), 'Purification of metallic grade silicon up to solar grade by NEDO melt purification process', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 157-160.

196

M. A. Green

Sard D., Durand F., Choudhury A., Marfaing Y., Einhaus R. and Luque A. (1997), 'Electromagnetic cold crucible continuous casting for multicrystalline silicon solar cells', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 849-852. Sawada T., Terada N., Tsuge S., Baba T., Takahama T., Wakisaka K., Tsuda S. and Nakano S. (1994) 'High-efficiency a-Si/c-Si heterojunction solar cell', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1219-1225. Schulz M. and Sirtl E. (1984), 'Silicon sheet', in Landolt-Bornstein: Numerical Data and Functional Relationships in Science and Technology, Springer-Verlag, Berlin, Vol. 17c, Section 6.1.2.5.3, pp. 52-54 and pp. 442-444. SerrezeH. B. (1978), 'Optimizing solar cell performance by simultaneous consideration of grid pattern design and interconnect configurations', Conf. Record 13th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 609-614. Shah A., Meier J., Torres P., Kroll U., Fischer D., Beck N., Wyrsch N. and Keppner H. (1997), 'Recent progress on microcrystalline solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 569-574. Shimura F. (1989), Semiconductor Silicon Crystal Technology, Academic Press, New York. Tanaka M., Taguchi M., Takahama T., Sawada T., Kuroda S., Matsuyama T., Tsuda S., Takeoka A., Nakano S., Hanfusa H. and Kuwano Y. (1993), 'Development of a new heteroj unction structure (ACJ-HIT) and its application to polycrystalline silicon solar cells', Prog, in Photovoltaics 1, 85-92. Tiedje T., Yablonovitch E., Cody G.D. and Brooks B. G. (1984), 'Limiting efficiency of silicon solar cells', IEEE Trans. Electron Devices ED-31, 711-716. Verlinden P. J., Sinton R. A., Wickham K., Crane R. A. and Swanson R. M. (1997), 'Backside-contact silicon solar cells with improved efficiency for the '96 world solar challenge', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S. Stephens & Associates, Bedford, 96-99. Wallace R. L., Hanoka J. I., Narasimha S., Kamra S. and Rohatgi A. (1997), 'Thin silicon string ribbon for high efficiency polycrystalline solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, Piscataway, 99-102. Weber K. J., Catchpole K., Stocks M. and Blakers A. W. (1997), 'Lift-off of silicon epitaxial layers for solar cell applications', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 107-110.

Crystalline Silicon Solar Cells

197

Wenham S. R., Robinson R., Dai X., Zhao J., Wang A., Tang, Y. H., Ebong A., Honsberg C. B. and Green M. A. (1994), 'Rear-surface effects in high efficiency silicon solar cells', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1278-1282. Wenham S. R. and Green M. A. (1995), 'Thin film silicon formation using porous silicon as a contamination barrier', Australian Provisional Patent Application PN6061, 19 October 1995. Werner J. H., Brendel R. and Queisser H. J. (1994), 'New upper efficiency limits for semiconductor solar cells', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1742-1745. Wolf M. (1960), 'Limitations and possibilities for improvement of photovoltaic solar energy converters', Proc. Inst. Radio Engineers 48, 1246-1263. Wolf M. (1976), 'Historical development of solar cells', in Solar Cells, Backus C. E., ed., IEEE Press, Piscataway. Yamamoto K., Yoshimi M., Suzuki T., Okamoto Y., Tawada Y. and Nakajima A. (1997), 'Thin film poly-Si solar cell with "Star Structure" on glass substrate fabricated at low temperature', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 575-580. Zhao J., Wang A., Altermatt P. and Green M. A. (1995), '24% efficient silicon solar cells with double layer antireflection coatings and reduced resistance loss', Appl. Phys. Lett. 66, 3636-3638.

CHAPTER 5

AMORPHOUS SILICON SOLAR CELLS CHRISTOPHER R. WRONSKI Centerfor Thin Film Devices, The Pennsylvania State University, University Park, PA 16802 Crwece @ engr.psu. edu and DAVID E. CARLSON BP Solarex, 3601 Lagrange Parkway, Toano, VA 23168 [email protected]

One advantage of being disorderly is that one is constantly making exciting discoveries. A. A. Milne, Winnie-The-Pooh, 1926.

5.1 Introduction In the last few years significant progress has been made in improving the efficiencies of amorphous silicon (a-Si)-based solar cells and in the scale of production of a-Si PV modules, which is currently about 30 peak megawatts (MWP) per year. These advances, together with the already established large-scale mass production of liquid crystal displays based on hydrogenated amorphous silicon (a-Si:H) thin-film transistors, indicate a coming of age for amorphous silicon technology. The first small-area amorphous silicon solar cells were fabricated with initial efficiencies of 12% (Carlson and Wronski, 1976). Today one square foot panels are being fabricated with stabilised efficiencies greater than 10% and modules are commercially produced with areas up to 12 square feet (Wronski, 1996; Guha, 1996; Forrest, 1997). The progress in a-Si solar cell technology is due to concurrent advances in the areas of new materials, novel cell designs and the large-area deposition techniques suitable for mass production. This progress has predominantly been a consequence of the lack of long-range order, such as is present in crystalline silicon (c-Si), which on one hand drastically changes the photovoltaic properties of the a-Si materials but on the other offers great flexibility in the manufacture of different solar cell structures as well as large-area monolithic modules.

199

200

C. R. Wronski and D. E. Carlson

Several outstanding features of amorphous silicon-based technology have allowed continuous advances to be made in the performance and manufacture of a-Si solar cells. Cells can be fabricated using a-Si alloyed with hydrogen, germanium and carbon to form semiconductors with band gaps between about 1.3 and 2.0 eV (Dawson et ah, 1992). These band gaps allow the fabrication of not only singlejunction, but also tandem and triple-junction, stacked cells designed in such a way as to maximise the absorption of the solar spectrum (Yang and Guha, 1992; Yang et al., 1998). These a-Si alloys are excellent candidates for thin-film solar cells since, unlike crystalline silicon, they have optical absorption coefficients similar to those found in direct band-gap semiconductors (Collins and Vedam, 1995). In addition, unlike any other amorphous semiconductors, these a-Si-based materials can be doped both p-and n-type (Spear and LeComber, 1975), which allows high-quality junctions and nohmic contacts to be used in solar cell fabrication (Carlson and Wronski, 1976; Wronski, et al., 1976). Last but not least, the fabrication processes of these cells are run at temperatures less than 300 C, which allows uniform thin films and cells to be reproducibly deposited over large areas. Because the long-range order present in c-Si is absent in a-Si there are significant differences between the semiconductor properties of the two materials. The properties that are important in solar cells and their contributions to cell performance are reviewed in Section 5.3. These include the properties of intrinsic and doped materials as well as the reversible light-induced changes that occur under sunlight. Despite the amorphous nature of the materials, their photovoltaic properties do depend on their microstructure. This is determined by growth kinetics and hence by the various fabrication conditions discussed in Section 5.4. The principal mechanisms that determine the operation of efficient a-Si-based solar cells, and which are different from those in crystalline Si cells, are briefly described in Section 5.5. Next, a review is presented of how these cells have been optimised by incorporating different a-Si:H alloy materials into various cell structures and multijunction stacked cells. Section 5.6 describes the different cell structures utilised by various organisations and discusses the characteristics and performance of single-junction, tandem and triple-junction cell structures. Section 5.7 discusses the development and commercialisation of a-Si:H modules fabricated on glass, stainless steel and plastic substrates. Issues related to the manufacturing costs of such modules are presented in Section 5.8 and their long-term reliability is discussed in Section 5.9. Environmental issues and the challenges for the future are reviewed in Sections 5.10 and 5.11 respectively.

Amorphous Silicon Solar Cells

201

5.2 Background Amorphous silicon deposited from silane was first investigated by Chittik et al. (1969). Although subsequent work was carried out on this material, its potential as a useful optoelectronic material was not recognised until 1976. In that year, Carlson and Wronski (1976) reported the first results on 2% efficient a-Si solar cells and shortly afterwards on ones with 5% efficiency (Carlson et al., 1976). This sparked worldwide interest in a-Si and a-Si-based solar cells, and led to numerous fundamental studies on the materials and improving the performance of a-Si solar cells. Once the importance of hydrogen in these materials had been recognised (Brodsky et al., 1977), a wide range of deposition techniques and conditions were employed in attempts to improve the properties of hydrogenated amorphous silicon (a-Si:H) for solar cells. These included not only the intrinsic optoelectronic and photovoltaic properties of the material, but also the equally important, large changes that occur upon exposure to sunlight and which are perfectly reversible on annealing at ~150C for a few hours (Staebler and Wronski, 1977). These light-induced changes, known as the StaeblerWronski effect (SWE), manifest themselves in both thin-film materials and solar cells. The early discovery of the SWE had an enormous effect on the development of a-Si solar cells and their technology by having a major impact on cell design. Engineering approaches were developed for minimising the effects of the SWE on the degraded steady-state (i.e. stabilised) cell efficiencies by making the cells as thin as possible (Hanak and Korsun, 1982). Reduction in cell thickness, and the corresponding lower absorption of sunlight in the cell, leads to lower short-circuit currents (i^), which reduces the possible power conversion efficiency. This problem was greatly reduced with the development of efficient optical enhancement (Yablonovitch and Cody, 1982; Deckman et al., 1984) obtained by introducing textured, rather than smooth, optical reflectors. Such optical enhancement, which was first successfully applied to a-Si-based solar cells, is now extensively used in all types of thin-film solar cells. Another key element in the development of efficient a-Si solar cells with thinner absorber layers was the introduction of amorphous silicon-germanium alloys, which have band gaps significantly lower than those of a-Si:H. This allowed not only single- but also tandem and triple-junction cells to be fabricated (Yang and Guha, 1992; Yang et al., 1984). With successful engineering, both the initial and the degraded steady-state efficiencies of these cells have been steadily improved.

202

C. R. Wronski and D. E Carlson

5.3 Amorphous silicon based materials 5.3. J Introduction The materials used in amorphous silicon-based solar cells are prepared by plasmaenhanced chemical vapour deposition (PECVD), which relies on the decomposition of gases containing silane (S1H4). These materials are in fact silicon-hydrogen alloys, typically containing 5-20 at.% hydrogen. The key function of the hydrogen in these materials is to passivate the broken Si bonds that are introduced by the absence of the long-range order present in c-Si. The differences in the structure between c-Si and hydrogenated amorphous silicon (a-Si:H), and the passivation of dangling bonds by hydrogen, is illustrated in Fig. 5.1. The diagram on the left shows the tetrahedrally bonded crystal structure that extends throughout the lattice of c-Si. On the right is a schematic diagram of the a-Si:H network, which has a Si-Si nearest neighbour configuration similar that of c-Si but no long-range order, and consequently a large number of dangling (broken) bonds. Luckily, as indicated in the figure, the vast majority of these bonds are passivated by the hydrogen in the a-Si:H-based materials.

Figure 5.1 The tetrahedrally bonded crystal structure of c-Si is shown on the left. The absence of longrange order and passivation of Si dangling bonds by hydrogen is illustrated on the right.

This passivation greatly reduces the density of the -10 cm" dangling-bond defects present in unhydrogenated a-Si. The incorporation of hydrogen also leads to materials with significantly wider band gaps than c-Si, but which, because of their disorder, have at the same time much higher optical absorption. The atomic hydrogen present during the growth of the materials also plays a very important role in determining the growth kinetics and the resulting microstructure. With high dilution

Amorphous Silicon Solar Cells

203

of the feedstock silane gas with hydrogen it is even possible (and sometimes advantageous) to obtain microcrystalline Si:H (/JC-Si:H) alloys (Hirose, 1984; Tanaka and Matsuda, 1987; Collins and Fujiwara, 1997). Hydrogenated amorphous silicon can also be grown with germanium and carbon to produce a-SiGe:H and a-SiC:H alloys, materials, which are extremely useful as components of a-Si:H-based solar cells. However, because the properties of a-Si-based materials depend on the growth and deposition conditions, unlike c-Si there is no unique form of a-Si:H. While this allows a wide range of materials to be developed for solar cell applications, it has on the other hand made it difficult to obtain consistent material parameters, since studies are generally carried out on materials grown under different conditions.

5.3.2 Band gaps and optical absorption The disorder in a-Si:H-based materials transforms the nature of the optical absorption associated with the indirect band gap of c-Si to that of direct band-gap semiconductors (Collins and Vedam, 1995). Incorporation of hydrogen into the a-Si:H network not only removes defects, and defect states in the forbidden gap, but also widens the gap. Hence when the hydrogen content in a-Si:H is increased from about 5% to 20%, the band gap increases from -1.6 eV to -1.8 eV (Zanzucchi et al., 1977). The band gaps of a-SiGe:H and a-SiC:H alloys depend on the concentrations of the Ge and C as well as that of hydrogen. The alloys used in solar cells have Ge up to about 60 at.% and C up to about 20 at.% with band gaps which are down to about 1.3 eV for the a-SiGe:H materials and up to about 2.0 eV for the a-SiC:H materials (Dawson et al, 1992; Lu et al, 1994; Ganguly and Matsuda, 1996). By incorporating different amounts of hydrogen into a-Si:H, it is possible to change not only its band gap but also the density of states in the gap; materials prepared with 5-15 at.% of hydrogen generally have densities of dangling-bond states on the order of 1015 to 1016 cm"3. The optical absorption of such a-Si:H materials is shown in Fig. 5.2 for three RF PECVD films prepared under similar conditions but at different substrate temperatures Ts, with the corresponding optical gaps Ug (Tauc et al., 1966) also shown in the figure. The changes in the gaps are clearly reflected in the systematic horizontal shifts of the absorption spectra, where regions of a greater than 10 cm" correspond to optical transitions between the valence and conduction bands. The regions of a between about 103 and 10 cm-1, which are exponential in nature, arise from the absorption in valence-band tail states, created by the disorder in these amorphous materials (Roxlo et al., 1983). The densities of the valence-band tail states are significantly higher than those of the conduction-band tails (Tiedje, 1984), even in

204

C. R. Wronski and D. E. Carlson

I 1.0

1.2 1.4 Energy/eV

1.6

Figure 5.2 Optical absorption a versus photon energy for a-Si:H films deposited by RF PECVD at three substrate temperatures, T„ showing also the resulting band gaps Ug. After Wronski (1996).

the recently developed materials with a highly ordered network (Koh et al., 1998). The shoulders in the absorption spectra at photon energies less than -1.4 eV, with values of a below -10 cm"1, are due to optical transitions originating from defect states near and around the middle of the gap (Jackson et al., 1983; Wronski et al., 1997). The energy band diagram and the distribution of gap states in a-Si:H are shown schematically in Fig. 5.3. Unlike c-Si, the a-Si materials have a continuous distribution of localised states in the gap through which electrons and holes cannot move freely. Near the conduction and valence band edges are the two sets of band tail states, whose densities decrease exponentially from the main conduction and valence band edges. There are also deep-lying gap states whose densities, generally represented by gaussian distributions, consist of: neutral dangling-bond states (D°) in the middle of the gap; negatively charged defect states (D~) below the middle of the gap; and positively charged defect states (D+) above the middle of the gap (Branz and Silver, 1990; Powell and Dean, 1993; Jiao et al, 1996a). These deep-lying states are very important in determining the collection of photogenerated carriers in a-Si solar cells. The absorption that is useful in creating free carriers in solar cells is at values of a greater than about 103 cm-1, which corresponds to photons with energies greater than the band gap of the material. Even though a wide range of band gaps is available from

Amorphous Silicon Solar Cells

1

205

Conduction Band

/ ""CTTTair

I

=^H

z>°* ^ > D°

^ = = *!!>-0-9.

illumination

\

O

dark conductivity

50

100

ISO

2O0

250

Time/min

Figure 5.4 Dark conductivity and photoconductivity of an a-Si:H thin film shown as a function of the illumination time with sunlight. After Staebler and Wronski (1980).

responsible for their creation. There is general consensus, however, that the hydrogen that plays a key role in eliminating dangling-bond defects in a-Si:H alloys also plays a key role in their light-induced creation (Lee et al., 1996; Carlson and Rajan, 1998). For a long time, the widely held view was that the only defect states produced by light were associated with the neutral dangling bond, D°. However, there is now extensive evidence indicating the importance of microstructure, other than that associated purely with hydrogen, and showing that light-induced changes in the charged defect states are just as important as, if not more important than, those in the D° states (Wronski et al., 1997; Lu et al, 1999). Significant progress has been made over the years, not only in improving the initial (state A) properties of a-Si-based materials, but also in reducing the SWE. This has been achieved by optimising growth conditions to improve the microstructure of the materials through incorporation of hydrogen into the network. As a result it is possible to obtain solar cells with not only higher initial efficiencies but, more importantly, better performance after they reach degraded steady state under illumination with 1 Sun. In addition, these materials and their solar cells require much shorter times to reach the degraded steady state in sunlight, under 100 hours as compared to thousands of hours in the past (Yang and Chen, 1994; Lee et al., 1996). This makes fundamental studies of the SWE, as well as those on solar cell improvements, more amenable to detailed investigations.

Amorphous Silicon Solar Cells

209

5.4 Growth and microstructure The a-Si:H-based materials used in solar cells are usually deposited by PECVD at substrate temperatures from about 100 to 300 C. The decomposition of the feedstock gases may be carried out in a variety of reactor geometries, and with plasmas generated over an extremely wide frequency range, including DC; RF (13.56 MHz); VHF (60-100 MHz); and microwave (2.45 GHz) (Uchida, 1984; Watanabe et al., 1986; Tanaka and Matsuda, 1987; Shah et al, 1996; Collins and Fujiwara, 1997). However, in all cases the growth process can be considered to occur in three stages. The first stage is the dissociation of S1H4 into a partly ionised reactive mixture. Next, while the mixture is transported to the surface of the growing film, there are continuous chemical reactions between the different species. The species arriving at the surface are adsorbed on the growing film, where they can react with both the film itself and the radicals in the gas phase. The resulting by-products (mainly hydrogen and unreacted silane radicals) desorb from or are etched off the surface by the reactive species arriving at it. The main precursor in the growth is the SiH3 radical, but other neutral species, such as Si, SiH and SiH2, also reach the growing surface and have a pronounced effect on the structural, optoelectronic and photovoltaic properties of the materials. Hydrogen coverage of the growing surface is desirable since it is a critical factor for surface mobility of the precursor species. The high mobility allows the radicals to find more stable sites for forming a dense random network, leading to superior material. At a given substrate temperature, therefore, there is a trade off between any increase in surface diffusion and the desorption of hydrogen that leaves behind unpassivated dangling bonds. The quality of amorphous silicon-based films is determined by deposition parameters such as the substrate temperature, the pressure, the flow rate of the source gases, the plasma frequency, the power and the electrode spacing. As we have just noted, the substrate temperature is a critical parameter, and since it controls hydrogen incorporation it can be used to tailor the band gap of the materials. While decreasing the substrate temperature can increase the optical band gaps, the accompanying changes in the growth processes must be taken into account. These include both the changes in microstructure arising from the lower diffusivity of species on the surface, and an undesirable tendency to incorporate polyhydrides such as SiH2 and S1H3. The optimum substrate temperature range of -180-250 C maximises the surface mobility of the surface radicals while at the same time allowing adequate hydrogen surface coverage for passivating Si dangling bonds (Tanaka and Matsuda, 1987). The properties of the materials also depend on the pressure of the source gases. At low pressures, the growing surface can suffer severe ion bombardment. At high

210

C. R. Wronski and D. E. Carlson

pressures, on the other hand, because of the increased collision frequency between electrons and radicals in the plasma, there is a tendency to create powder in the gas phase, which introduces dihydride and polyhydride complexes into the deposited material. The flow rate of the source gases is an important deposition parameter since it determines the residence time of the different molecules in the plasma and hence affects the growth kinetics. The frequency used also affects the nature of the plasmas, and in particular the ion bombardment intensity, which becomes significantly lower at VHF and microwave frequencies. The nature of the plasmas and growth processes also changes with the introduction of the alloy-forming gases, GeH4 and CH4, the ntype dopant PH3, and the p-type dopants B2H6 or trimethyl boron. However, in all the types of depositions using PECVD, hydrogen plays a key role in reducing defects and improving the quality of the a-Si materials. 10000

1000 •
•;

0.5

^r/' J*/

1.0

Voltage/V

Figure 5.8 The dark forward current-voltage characteristics of two p-i-n solar cells that are identical except that one has a hydrogen-treated (improved) pli interface region and the other does not. After Wronski et al. (1997).

The mechanisms determining the operation of a-Si solar cells clearly depend strongly on the light-induced defects associated with the SWE. The introduction of defects after prolonged exposure to sunlight reduces the free-carrier lifetimes and increases the space charge, which leads to a redistribution of the electric fields across the /-layers. This leads to changes in the quantum efficiencies as a function of wavelength, but fortunately these result in only a small decrease in /sc in high-quality

Amorphous Silicon Solar Cells

215

solar cells, particularly when the Mayer is less than -300 nm thick. The defects generated in the bulk of the Mayers do have a large effect on the fill factors, and their light-induced changes are the major contributor to the loss in cell efficiencies. Lightinduced defects in the pli interface regions have the most pronounced effect on lowering the open-circuit voltages, but with good pli interfaces high open-circuit voltages that do not degrade in sunlight are obtained. In summary, the limitations on the operation of a-Si solar cells imposed by the low mobilities of the free carriers are compensated by the high optical absorption and low space-charge densities of the Mayers, which allow the electric fields to sweep out the carriers before they recombine. Both the recombination lifetimes and the electric fields depend strongly on the densities and types of the defect states. Thus in large part they determine the thicknesses of cells in which the photogenerated carriers can be sufficiently well collected to give high values of fill factors.

5.5.2

Optimisation of solar cells

From the beginning, the developers of a-Si-based solar cells sought improvements in the performance, not only of the initial cells, but also of the degraded, steady-state cells. At first the effort focussed on single-junction a-Si:H cells, but this quickly expanded to include optimisation of tandem and multijunction cells. Particular attention was paid to the degraded steady-state efficiencies obtained after prolonged exposure to AM 1.5 sunlight. The development proceeded along several tracks, which included improvements in materials and solar cell structures as well as engineering approaches for minimising the effects of the SWE. It also relied heavily on the flexibility that a-Si alloys offer in terms of band gaps and their ability to generate high open-circuit voltages with p-type a-SiC:H and /ic-Si:H materials. Improvements in V^ were obtained with better p-type contacts and improved pli interface regions. Improvements in ix were obtained by using lower band-gap materials to increase absorption in the intrinsic layers. For single-junction solar cells, the a-Si:H band gaps giving the highest efficiencies are around 1.7 eV. With these band gaps, open-circuit voltages of 0.9 V can be obtained while at the same time a 1 /jm thick Mayer absorbs a fraction of AM 1.5 sunlight that is sufficient to generate a short circuit current of -18 mAcnf 2 . The challenge, however, is to maximise ix by increasing the thickness of the Mayers while at the same time retaining the collection of carriers at a level necessary for high values of fill factors. Improved materials with low defect densities allow the thicknesses of Mayers to be extended while still having high carrier collection efficiencies, but thus far these thicknesses are still significantly

216

C. R. Wronski and D. E. Carlson

less than 1 /im. The acceptable thickness of high-performance cells is further limited by the SWE, since the light-induced defects degrade the carrier transport and further limit the carrier collection efficiencies. A major breakthrough in achieving thin, high-efficiency cells was achieved with the development of optical enhancement based on textured substrates and reflectors (Yablonovitch and Cody, 1982; Deckman et al., 1984). The optical enhancement arises from the large-angle scattering caused by the surface texture, which produces multiple internal reflections that allow weakly absorbed light to undergo many passes through the Mayer. This greatly increases the already high optical absorptivity at longer wavelengths, so that significantly higher quantum efficiencies can be obtained 1.0

1

•

.

.9 .8

FlatAg

! • '

\ \

s f .6 8 -

Flat Cr - ^

\ ^ '•. \ \ \

Detached" reflector

•.

VC

B .5 j> t> .4 •o

* £

3

Tuned reflector

-

\

*.

\ \

• Theory detached reflector

.1

\

\

'••

* Theory uned reflector i

.4

.5

i

i

.6

.7

>i .8

Wavelength/^ m

Figure 5.9 Experimental collection efficiency as a function of wavelength for an a-Si:H p-i-n solar cell with four different reflectors: flat Cr, flat Ag, textured tuned or textured detached. The results are normalised to the peak quantum efficiency at 0.55 /an. Shown as symbols are the theoretical results for a tuned and a detached reflector. After Deckman etal. (1984).

at these wavelengths without any increase in the cell thickness. Figure 5.9 shows the improvement that can be obtained by depositing an n(a-Si:H)-i(a-Si:H)-/?(a-SiC:H) cell on a textured, rather than flat, metal reflector. When the flat Cr or Ag reflector is replaced with a textured detached or tuned reflector, the quantum efficiency at 0.7 /an is improved from -10 to 20% of the peak value (observed at 0.55 fan) to -0.5 of the maximum for the tuned reflector, and -0.6 for the detached reflector. The tuned reflectors consist of metal evaporated on appropriately textured glass onto which the n-i-p cell is deposited. The detached reflectors have the same metal combination but with the addition of a conductive oxide (Sn0 2 , ZnO) several thousands of Angstroms thick for 'detaching' the n-i-p cell from the metal. Using textured substrates it has been possible to obtain short-circuit currents of about 18 mA cm" with a-Si:H of Ug -1.7 eV and Mayers much thinner than 1 /an.

Amorphous Silicon Solar Cells

217

150


200 V), the tin oxide is scribed into 143 separate strips running along the short dimension of the plate that define the individual cells. The plates are washed before being loaded into a multi-chamber deposition system that coats the plates with all the semiconductor layers shown in Fig. 5.15, including the ZnO layer. Each chamber of this system can hold four plates at a time in a vertical orientation. Seven of the twelve chambers are used to deposit amorphous and microcrystalline Si:H alloys by DC plasma-enhanced chemical vapour deposition (PECVD), one chamber is used to deposit ZnO by low-pressure CVD, and the other chambers act as buffers to minimise cross contamination. After exiting the vertical deposition system, the a-Si and ZnO layers are scribed with another Nd-YAG laser. These scribes are located several tens of microns to one side of the tin oxide scribes (see Fig. 5.18) and are performed at a lower power density so as to not scribe the tin oxide layer itself. The plates are then coated with about 300 nm of aluminium by magnetron sputtering before performing another laser scribe that removes the a-Si, the ZnO and the Al in a region to one side of the earlier scribes. This last scribe (the metal laser scribe) completes the series connection as shown in Fig. 5.18, since the front contact of each strip cell is connected in series to

228

C. R. WronskiandD.

E. Carlson

silicon dioxide tin oxide

back contact

Figure 5.18 Schematic of the BP Solarex series-connected tandem device structure. The series connection is made by the back («) contact of one cell to the tin oxide (p) contact of the adjacent cell. The indentation is a result of sequential scribing and represents the thickness of the cell that was removed to form (isolated) adjacent cells.

the back contact of the next adjacent cell by the aluminium deposited in the amorphous silicon scribe. Next, a final laser scribe is made at a relatively high power around the perimeter of the module to ensure electrical isolation. The modules are then cleaned in an ultrasonic bath to remove all debris before passing to a bed-of-nails station that applies a reverse bias to electrically cure cells that are excessively leaky due to small shorts (Nostrand and Hanak, 1979). After electrical curing, the performance of the modules is tested using a solar simulator. The modules are then completed by encapsulating a back plate of float glass to the front plate with ethyl vinyl acetate (EVA), attaching lead wires and mounting the module in a frame. In some applications, a frame is not needed or is supplied by the customer. The completed modules then undergo a final power test before being shipped to customers. This manufacturing process is capable of producing tandem modules with relatively high yields, as shown in Fig. 5.19 for a run of 4 ft2 modules made in a pilot manufacturing mode. In this pilot run of more than 160 modules, the average initial conversion efficiency was 8.9% with only a few modules falling outside the control limits of 8.1% and 9.8%. (Single-junction modules were also produced with high electrical yields (-95%) over a period of about 5 years at the Solarex facility in Newtown, PA before being phased out in 1998.)

Amorphous Silicon Solar Cells

229

I Figure 5.19

100 Run number A run chart showing the initial efficiency of 4 ft2 tandem modules vs. run number.

5.7.3 Modules on metal substrates Table 5.4 Manufacturing steps for USSC triple-junction modules on steel substrates Process step Process step description no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14

The roll of stainless steel foil is washed, rinsed and dried. Al and ZnO are sequentially deposited by magnetron sputtering. a-Si and ^c-Si alloy layers are deposited by RF PECVD. ITO is deposited by magnetron sputtering. The roll is cut into slabs. Slabs are processed to define cell size. Slabs are passivated to remove shunts. Slabs are tested to determine device quality. Conductive pads and grid wires are applied. Slabs' are cut into predetermined cell sizes. Cells are interconnected. Cell block is laminated. Modules are framed and junction boxes are added. Modules undergo highpot and performance tests.

230

C. R. Wronski and D. E. Carlson

Table 5.4 shows the various steps in the USSC manufacturing process for producing a-Si triple-junction modules on stainless steel foil (Guha et al, 1999). A roll of stainless steel foil is first washed in a wash machine, where it is also rinsed in deionised water and dried in an infrared oven. The roll is then loaded into a magnetron sputtering machine where aluminium and zinc oxide are sequentially deposited to form a highly reflective back contact. In the next step, the roll is loaded into a ninechamber, in-line PECVD system where the a-Si and pic-Si alloy layers are deposited continuously on the moving foil. The roll is then loaded into another magnetron sputtering machine where a layer of indium tin oxide (ITO) is deposited. This ITO layer acts as both a top electrical contact and an antireflection coating. The roll of foil is then moved to a semi-automated module assembly area where the foil is cut into 9.4" x 14" slabs. The slabs are processed to remove shunts and to define the specific cells before attaching electrodes and cutting the cells. The individual cells are then interconnected and laminated into a module. The module is completed by adding a frame and a junction box. The modules are subjected to a high-voltage test, and the output power is measured under simulated sunlight before shipping to customers. USSC reports that the stabilised aperture-area efficiency for these products is about 7.5% (Guha et al, 1999).

5.7.4

Modules on plastic substrates

The various steps in the manufacturing process used by Iowa Thin Film Technologies for producing a-Si tandem modules on plastic substrates are listed in Table 5.5 (Braymen et al, 1999). As a substrate material, Iowa Thin Film uses a polyimide plastic film that can tolerate temperatures of about 250 C and does not outgas significantly in a vacuum. They deposit an Al film as a back contact layer and then a same band-gap, a-Si tandem structure. A Nd-YAG laser is used to scribe through both the Al and the a-Si layers to define the individual cell segments. They then use screen-printing to place an insulating ink in the laser scribe and also a second strip of insulating ink in a region to one side and parallel to the laser scribe. Next, ZnO is sputtered onto the substrate, and then a conductive (Ag) ink is placed in the region over the laser scribe (which was previously filled with the insulating ink). A laser is then used to scribe the ZnO in the region above the second insulating ink strip. The series interconnection is completed by using a laser to weld or fuse the Ag ink to the bottom Al contact of the adjacent cell.

Amorphous Silicon Solar Cells

231

Table 5.5 Manufacturing steps for tandem modules made by Iowa Thin Film Technologies on plastic substrates Process Process step description step no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14

The roll of polyimide is washed. Al is sputtered deposited on the polyimide. a-Si layers are deposited by RF PECVD. Both the Al and the a-Si layers are scribed by a laser. Insulating ink is screen-printed in strips in the interconnect region. ZnO is deposited by sputtering. Ag ink grid lines are screen-printed in the interconnect region. The ZnO is laser scribed. A laser is used to weld the Ag ink to the Al bottom contact. Bus bars are attached to the modules. Modules are laminated. Modules are cut from the roll. Modules are framed and wired. Modules undergo performance tests.

Bus bars are attached to the modules while they are still on the roll. The modules are laminated to a Tedlar sheet using a roll-based laminator and then cut into individual modules before being framed, wired and tested. Amorphous silicon modules made on plastic substrates are lightweight and flexible and are used in a number of consumer applications.

5.8

Manufacturing costs

In past years, a number of organisations have estimated that the total cost of manufacturing a-Si PV modules should be less than $1/WP once the production volume of the plants reaches about 10 MWp per year (Carlson, 1989). More recently, Woodcock et al. (1997) have analysed the manufacturing costs for the three leading thin-film PV technologies—a-Si, copper indium diselenide (CIS), and cadmium telluride (CdTe)—at production volumes of 60 MWP /yr. As shown in Fig. 5.20, the projected manufacturing costs are significantly less than $1/Wp in each case. It is interesting to note that at a production volume of 60 MWp/yr, the materials costs are more than half of the total manufacturing costs in each case. As improvements are

232

C. R. Wronski and D. E. Carlson

Figure 5.20 Projected manufacturing costs for a-Si, CIS and CdTe thin film PV modules made in plants with 60 MWp /year capacity. Source: Woodcock el al. (1997).

made in thin-film PV manufacturing technology and as the plants become even larger, the labour costs and the fixed costs should decline even further so that materials costs will become the major factor limiting further cost reductions. For a-Si tandem modules, the major cost elements are currently the framing, the encapsulation and the tin oxide-coated glass. The framing and encapsulation alone account for about 37% of the total manufacturing cost for 8 ft2 tandem modules, while the tin oxide coated glass substrate and back glass plate constitute ~23% of the total cost. The semiconductor feedstock materials constitute much smaller percentages, with germane accounting for about 13% and silane only about 2% of the total cost. At present the utilisation of feedstock gases in commercial PECVD reactors is poor, with no more than a few percent of the silicon or germanium winding up in the a-Si alloy films. If the utilisation could be improved to about 70% or better, then the cost for the semiconductor materials would be less than $0.02/Wp.

5.9

Long-term reliability

All commercial PV modules are periodically subjected to a series of accelerated environmental tests to assure long-term reliability. BP Solarex a-Si tandem PV modules are subjected to the series of tests shown in Fig. 5.21. Some of the more critical tests are the wet hi-pot test, the thermal cycle test, the temperature humidity test and the light-soak test.

233

Amorphous Silicon Solar Cells I 8 modules [

Visual inspection

T

Performance @ STC

T

r

' GmtroL module

VVL'

11 JHU mst

K~

7^

Annealing

current

3Z

M^

Light soak

200 Thermal cycles

Damp~heat

I 2 modules Outdoor exposure

UVtest

^

3Z

Annealing

^_

50 thermal cycles

Hot spot endurance

JZ

10 Humidity freeze cycles 1 module

I 1 inodule

Robustness, of terminations

\

Mechanical load

Twist test

ir^x

Wet leakage

3Z

Wet hi-pot test

31

Light soaking

Figure 5.21

The environmental test sequence used for BP Solarex thin-film PV modules.

Since some modules will be used in high-voltage, grid-connected applications, they must be able to pass the wet hi-pot test. In some grid-connected systems, the array voltage can be greater than 1000 V so modules must exhibit a leakage current of less than 50 /lA while wet and with an applied voltage greater than twice the array voltage plus 1500 V. Because of this stringent requirement, BP Solarex tandem modules are encapsulated in EVA between two sheets of glass. This type of module passes all the accelerated environmental tests shown in Fig. 5.21. Two critical tests that check the integrity of the module against moisture penetration are the thermal cycle test and the humidity freeze test. The former involves 50 cycles between -40 C and 90 C, and the latter 10 cycles between -40 C and 85 C at 85% humidity.

234

C. R. Wronski and D. E. Carlson

600

700

800

Wavelength (nm) Figure 5.22

Spectral response of a tandem cell before and after 20 hours at 180 C.

The BP Solarex tandem structure is quite stable with respect to thermal degradation. Unlike earlier a-Si solar cells with an Al rear contact, which exhibited degradation effects at temperatures of -130C (Willing et al., 1987), there are no significant thermal degradation effects associated with the Sn0 2 or ZnO contacts used in the present structure. However, the tandem cells will start to show some degradation when heated for prolonged times at temperatures on the order of 180 C. Figure 5.22 shows the change in spectral response of a tandem cell on heating for 20 hours at 180 C in the dark. As observed in earlier work (Carlson and Rajan, 1995), the a-Si front junction exhibits a loss in quantum efficiency, mainly in the shortwavelength regime, due to hydrogen motion. The a-SiGe back junction shows a loss in quantum efficiency that is more evenly distributed over its response spectrum. After 20 hours at 180 C, the efficiency decreased by only 2.2%. The thermal degradation increases dramatically with further increases in temperature. After 150 minutes at 220 C, the conversion efficiency typically falls about 15% with about half the loss due to degradation in the fill factor. Since the activation energy for thermal degradation in the dark is -1.7 eV (Carlson and Rajan, 1995), this degradation is negligible under normal operating conditions. Light soaking of a-Si modules is another critical test since all a-Si devices exhibit light-induced degradation arising from the Staebler-Wronski effect (Staebler and Wronski, 1977). This degradation can be enhanced by contaminants in the Mayers or at interfaces, and some recent work (Carlson and Ganguly, 2000) shows that irreversible light-induced degradation can be caused by trace amounts of boron in the Mayer. BP Solarex tandem modules typically exhibit about 13-17% degradation on

235

Amorphous Silicon Solar Cells 4

3

&

2

1

0 0

5

10

15

2D

Figure 5.23 A histogram showing the degradation experienced by twelve 4 ft2 tandem modules after exposure to 600 hours of simulated sunlight.

light soaking, but this can increase to 20-25% if contaminants are present. Figure 5.23 shows the distribution of the degradation suffered by twelve BP Solarex a-Si tandem modules (4 ft2) subjected to simulated sunlight for 600 hours. On average, these modules degraded by only about 12.9%. In general, light-induced degradation of a-Si PV modules saturates or reaches a steady state after about 10 to 10 hours of exposure to sunlight, depending on the deposition conditions. Saturation occurs in ~102 hours for a-Si single-junction modules grown in discharge atmospheres that contain silane heavily diluted in hydrogen. Once the modules reach steady state, the conversion efficiency then exhibits normal seasonal variations due to changes in the average ambient temperature and seasonal changes in the solar spectrum.

S.10

Environmental issues

Photovoltaic solar energy is viewed by many as an ideal way to produce power from a virtually inexhaustible energy source without noise or pollution. However, there are environmental issues that must be addressed to assure a trouble-free future for PV. The entire process of mining and refining raw materials, manufacturing PV modules and handling of obsolete product must be designed not only for low cost, but also for the environment. If module processing requires the use of toxic materials, then systems and procedures must be established to minimise the risk to employees. In the manufacture of a-Si PV devices, BP Solarex uses toxic doping gases such as diborane and phosphine only in a diluted form (-1-20 vol.% in silane). Trimethylboron (-1-5 % in

236

C. R. Wronski and D. E. Carlson

silane) is also used as a p-type dopant source and is less toxic than diborane. Silane is pyrophoric, so if a leak develops the dopant gas will be oxidised in the flame and a silicate glass powder will be formed, thus reducing the toxicity hazard. The silane and germane feedstock gases at the BP Solarex TF1 plant in Virginia are stored in an outdoor holding area and fed into the facility through stainless steel pipes. All exhaust gases are passed through a burn box and the powder is collected in a bag house for disposal. The powder consists mainly of silicon dioxide fused with small amounts of oxides of germanium, boron and phosphorus. Since all module interconnections are made using lasers, there are no wet chemicals such as acids or solvents used in the BP Solarex manufacturing process for tandem modules. Thus, there are no harmful waste products or effluents produced in the manufacturing process. In addition, since a-Si PV modules do not contain any toxic materials, there are no environmental risks associated with fires or with longterm disposal in landfills.

5.11

Challenges for the future

While a-Si photovoltaics has the potential to become a major source of low-cost electricity worldwide in the next few decades, the fulfillment of this potential depends on continued progress in improving the stabilised performance, reducing the total manufacturing cost and establishing the infrastructure necessary to create large-scale markets. There is tremendous potential to reduce the total system cost of thin-film PV arrays significantly by integrating them into new buildings. This can be accomplished by designing the modules to function as PV roofs or windows. The PV system cost is reduced since there is no need for land and support structures, and most of the material and labour can be credited against the cost of the building, as discussed in Chapter 15. Improvements in stabilised performance will require a better understanding of the growth kinetics, the nature of the intrinsic defects and the role of hydrogen in a-Si alloys. While a-Si alloys can be grown by a number of different techniques such as DC PECVD, RF PECVD, electron cyclotron resonance remote plasmas, hot-wire CVD, photo-CVD and sputtering in argon-hydrogen atmospheres, it is not clear what precursors or conditions are necessary to assure the best-quality films. Moreover, there is still no consensus on the microscopic origin of the Staebler-Wronski effect, or on the role of hydrogen in determining the metastability or doping efficiency of aSi alloys. Thus, there is a clear need for further experimental and theoretical work on understanding the growth and the microscopic nature of the defects in a-Si alloys.

Amorphous Silicon Solar Cells

237

While today's multijunction a-Si PV cells degrade by only -10-17% under prolonged light exposure, the efficiency could be greatly improved if the lightinduced degradation could be significantly reduced or eliminated. This is because the Mayers of current a-Si cells have to be quite thin to minimise the degradation. Conversion efficiencies of 17-20% might be possible with multijunction a-Si-based structures if the a-Si alloy j'-layers could be made thicker while remaining stable. While improving the stabilised performance would lead to lower manufacturing costs on a $/Wp basis, increasing the throughput will lower the costs associated with labour and overhead. The throughput of a-Si PV manufacturing plants is limited mainly by the deposition rate of the a-Si alloys. Currently, these are deposited at rates of about 0.1 nm s"' in most commercial manufacturing processes. Since about half of the cost of an a-Si PV manufacturing plant is associated with the cost of the a-Si deposition machine, a doubling of the throughput of that machine would significantly lower the capital cost of the plant on a $ per Wp of capacity basis. A number of organisations are investigating the rapid deposition of a-Si alloys by techniques such as hot-wire CVD, high frequency PECVD etc. The major challenge is to increase the deposition rate without increasing the concentration of the intrinsic and the metastable defects that lead to a reduction in stabilised performance. Another challenge is to understand the additional defects created when alloying aSi with germanium or carbon. The defect density increases when either of these is added to a-Si, and changes are observed in the kinetics of the light-induced degradation. It appears that additional defects associated with clustering or hydrogen complexes are introduced, but there is no detailed microscopic model or theory to account for these effects. Improvements in the microstructure of a-Si alloys should lead to further improvements in stabilised module performance. Significant progress has been made in recent years in developing high-quality films of more ordered a-Si:H, protocrystalline and microcrystalline silicon, and even thin polycrystalline silicon films. These materials have some significant advantages over other thin-film materials for PV applications. Silicon is very abundant and does not pose any environmental hazards. Moreover, especially in the case of a-Si alloys, the films are very easy to deposit with good uniformity over large areas. In summary, while amorphous and microcrystalline silicon alloy materials are complex and poorly understood, there are a large number of talented scientists and engineers in laboratories around the world who are working to make the promise of low-cost photovoltaic electricity a reality. It is this large pool of experienced,and talented people that makes it highly probable that this goal will be achieved in the next few decades.

238

C. R. Wronski and D. E. Carlson

References Bennett A. and Olsen L. C. (1978), 'Analysis of multiple-cell concentrator/ photovoltaic systems', Conf. Record 13th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 868-873. Branz H. M. and Silver M. (1990), 'Potential fluctuation due to inhomogeneity in hydrogenated amorphous silicon and the resulting charged dangling bond defects', Phys. Rev. B. 42, 7420-7428. Braymen S., Grimmer D., Jeffrey F., Martens S., Noack M., Scandrett B. and Thomas M. (1999), 'Monolithic amorphous silicon modules on continuous polymer substrates', AIP Conf. Proc. 462, 73-740. Brodsky M. H., Frisch M. A., Ziegler J. F. and Lanfard W. A. (1977), 'Quantitative analysis of hydrogen in glow discharge amorphous silicon', Appl. Phys. Lett. 30, 561-563. Carlson D. E. and Wronski C. R. (1976), 'Amorphous silicon solar cells', Appl. Phys. Lett. 28, 61'1-673. Carlson D. E., Wronski C. R., Triano A. and Daniel R. E. (1976), 'Solar cells using Schottky barriers on amorphous silicon', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 893-895. Carlson D. E. and Wronski C. R. (1979), 'Amorphous Silicon Solar Cells', in Topics in Applied Physics, Vol. 36, Brodsky M. H., ed., Springer-Verlag, Berlin, 289329. Carlson D. E. (1984), 'Solar energy conversion', in Topics in Applied Physics, Vol. 55, Joannopoulos J. D. and Lucovsky G., eds., Springer-Verlag, Berlin, 203-244. Carlson D. E. (1989), 'Low-cost power from thin film photovoltaics', in Electricity, Johansson T. B., Bodlund B. and Williams R. H., Lund University Press, Lund, Sweden, 595-626. Carlson D. E. and Catalano A. (1989), 'Improving the performance and reducing the cost of amorphous silicon solar cells', Optoelectronics 4, 185-193. Carlson D. E. and Rajan K. (1995), 'The degradation of amorphous silicon solar cells at elevated temperatures', Proc. 13th. European Photovoltaic Solar Energy Conf, Nice, H. S Stephens & Associates, Bedford, 617-619. Carlson D. E., Arya R. R., Bennett M., Chen L. F., Jansen K., Li Y. M., Newton J., Rajan K., Romero R., Talenti D., Twesme E., Willing F. and Yang L. (1996), 'Commercialization of multijunction amorphous silicon modules', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1023-1028.

Amorphous Silicon Solar Cells

239

Carlson D. E. and Rajan K. (1998), 'Evidence for proton motion in the field-induced recovery of light-induced degradation in amorphous silicon solar cells', J. Appl.Phys. 83, 1726-1729. Carlson D. E. and Ganguly G. (2000), to be published. Chittick R. C , Alexander J. H. and Sterling, H. E. (1969), 'The preparation and properties of amorphous silicon', J. Electrochem. Soc. 116,77-81. Collins R. W. (1994), 'Real time spectroscopic ellipsometry studies of the nucleation, growth and optical functions of thin films. Part 1: tetrahedrally bonded materials', in Physics of Thin Films, Vol. 19, Academic Press, New York, 49-125. Collins R. W. and Vedam K. (1995), 'Optical properties of solids', in Encyclopedia of Applied Physics, Vol. 12, Trigg G. L., ed., VCH Publishers, New York, 285-336. Collins R. W. and Fujiwara H. (1997), 'Growth of hydrogenated amorphous silicon and its alloys', Current Opinion in Solid State & Material Science 2, 417-424. Dawson R. M., Li Y., Gunes M., Nag S., Collins R. W., Bennett M. and Wronski C. R. (1992), 'Optical properties of the component materials in multijunction hydrogenated amorphous silicon-based solar cells', Proc. 11th. European Photovoltaic Solar Energy Conf, Montreux, Harwood Academic Publishers GmbH, Chur, Switzerland, 680-683. Deckman H., Wronski C. R. and Yablonovitch E. (1984), 'Optical enhancement of solar cells', Conf. Record 17th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 955-960. Forrest H. (1997), 'Efforts on commercialization of amorphous thin-film technology', Proc. 14th. European Photovoltaic Solar Energy Conf, Barcelona, H. S Stephens & Associates, Bedford, 2018-2020. Fritzsche H. (1997), 'Search for explaining the Staebler-Wronski effect', Mat. Res. Soc. Symp. Proc. 467, 19-31. Ganguly G. and Matsuda A. (1996), 'Role of hydrogen dilution in improvement of aSiGe:H alloys', J. Non-Cryst. Solids 198-200, 559-562. Guha S., Yang J., Pawliklewicz T., Glatfelter T., Ross R. and Ovshinsky S. R. (1989), 'Band-gap profiling for improving the efficiency of amorphous silicon alloy solar cells', Appl. Phys. Lett. 54, 2330-2332. Guha S. (1996), 'Amorphous silicon alloy solar cells and modules—opportunities and challenges', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1017-1022. Guha S., Yang J., Banerjee A., Hoffman K. and Call J. (1999), 'Manufacturing issues for large volume production of amorphous silicon alloy photovoltaic modules', AIP Conf. Proc. 462, 88-93.

240

C. R. Wronski and D. E. Carlson

Hanak J. J. and Korsun V. (1982), 'Optical stability studies of a-Si:H solar cells', Conf. Record 16th. IEEE Photovoltaic Specialists Conf., San Diego, IEEE Press, Piscataway, 1381-1383. Hirose M. (1984), 'Glow discharge', in Semiconductors and Semimetals Vol. 21A, Pankove J. I., ed., Academic Press, Orlando, 9-39. Ichikawa Y., Fujikake S., Takayama T., Saito S., Ota H., Yoshida T., Ihara T. and Sakai H. (1993), 'Large-area amorphous silicon solar cells with high stabilized efficiency and their fabrication technology', Conf. Record 23rd. IEEE Photovoltaic Specialists Conf., Louisville, IEEE Press, Piscataway, 27-33. Jackson W. B., Biegelsen D. K., Nemanich R. J. and Knights J. C. (1983), 'Optical absorption spectra of surface or interface states in hydrogenated amorphous silicon', Appl. Phys. Lett. 42,105-107. Jiao L., Liu H., Semoushikina S., Lee Y. and Wronski C. R. (1996a), 'Importance of charge defects in a-Si:H materials and solar cell structures', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1073-1076. Jiao L., Liu H., Semoushikina S., Lee Y. and Wronski C. R. (1996b), 'Initial, rapid light-induced changes in hydrogenated amorphous silicon materials and solar cell structures: the effect of charged defects', Appl. Phys. Lett. 69, 3713-3715. Kamei T., Hata N., Matsuda A., Uchiyama T., Amano S., Tsukamoto K., Yoshioa Y. and Hirao T. (1996), 'Deposition and extensive light soaking of highly pure hydrogenated amorphous silicon', Appl. Phys. Lett. 68, 2380-2382. Koh J., Lee Y., Fujiwara H., Wronski C. R. and Collins R. W. (1998), 'Optimization of hydrogenated amorphous silicon p-i-n solar cells with two-step i layers guided by real-time spectroscopic ellipsometry', Appl. Phys. Lett. 73, 1526-1528. Koval R. J., Koh J., Lu Z., Lee Y., Jiao L., Collins R. W. and Wronski C. R. (1999), 'Performance and stability of Si:H p-i-n solar cells with i layers prepared at the thickness-dependent amorphous-to-microcrystalline phase boundary', Appl. Phys. Lett. 75, 1553-1555. Lee Y., Jiao L., Liu H., Lu Z., Collins R. W. and Wronski C. R. (1996), 'Stability of a-Si:H solar cells and corresponding intrinsic materials fabricated using hydrogen diluted silane', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 1165-1168. Lee Y., Ferlauto A. and Wronski C. R. (1997), 'Contributions of bulk, interface and build-in potential to the open-circuit voltage of a-Si:H solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 683-686.

Amorphous Silicon Solar Cells

241

Lee Y., Ferlauto A.S., Lu Z., Koh J., Fujiwara H., Collins R. W. and Wronski C. R. (1998), 'Enhancement of stable open circuit voltage in a-Si:H p-i-n solar cell by hydrogen dilution of pli interface regions', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, European Commission, 940-943. Lu Y., Kim S., Gunes M., Lee Y., Wronski C. R. and Collins R. W. (1994), 'Process property relationships for a-Sii_^C^:H deposition: excursions in parameters space guided by real time spectroellipsometry', Mat. Res. Soc. Symp. Proc. 336, 595-600. Lu Z., Jiao H., Koval R., Collins R. W. and Wronski C. R. (1999), 'Characteristics of different thickness a-Si:H/metal Schottky barrier cell structures—results and analysis', Mat. Res. Soc. Symp. Proc. 557, 780-785. Nostrand G. E. and Hanak J. (1979), 'Method of removing the effects of electrical shorts and shunts created during the fabrication process of a solar cell', U.S. Patent No. 4,166,918. Powell M. J. and Dean S. C. (1993), 'Improved defect pool model for charged defects in amorphous silicon', Phys. Rev. B 48, 10815-10827. Roxlo C , Abeles B., Wronski C. R., Cody G. D. and Tiedje T. (1983), 'Comment on optical absorption edge in a-SiiH/, Solid State Commun. 47, 985-987. Shah A., Kroll U., Keppner H., Meier J., Torres P. and Fischer D. (1996), 'Potential of VHF-plasma for low-cost production of a-Si:H solar cells', Proc. 9th. Int. Photovoltaic Sci. Eng. Conf., 267-270. Spear W. E. and LeComberP. G. (1975), 'Substitutional doping of amorphous silicon', Solid State Commun. 17, 1193-1196. Staebler D. L. and Wronski C. R. (1977), 'Reversible conductivity change in discharge produced amorphous silicon', Appl. Phys. Lett. 31, 292-294. Staebler D. L. and Wronski C. R. (1980), 'Optically induced conductivity changes in discharge-produced hydrogenated amorphous silicon,' J. Appl. Phys. 51, 32623268. Stutzmann M.(1997), 'Microscopic aspects of the Staebler-Wronski effect', Mat. Res. Soc. Symp. Proc. 467, 37-48. Tanaka K. and Matsuda A. (1987), 'Glow-discharge amorphous silicon: growth process and structure',Mat Sci. Report!, 139-184. Tanaka M., Toguchi M., Matsuyama T., Sawada T., Tsuda S., Nakano S., Hanafusa H. and Kuwano Y. (1992), 'Development of new a-Si/c-Si heterojunction solar cells: ACJ-HIT (artificially constructed junction-heterojunction with instrinsic thin-layer)', Jpn. J. Appl. Phys. 31, 3518-3522. Tauc J., Grigorovici R. and Vancu A. (1966), 'Optical properties and electronic structure of amorphous germanium', Phys. Stat. Solidi 15, 627-631.

242

C. R. Wronski and D. E. Carlson

Tiedje T. (1984), 'Information about band-tail states from time-of-flight experiments', in Semiconductors and Semimetals, Vol. 21C, Pankove J. I., ed., Academic Press, Orlando, 207-238. Tsuda S., Takahama T., Hishikawa Y., Tarui H., Nishiwaki H., Wakisaka K. and Nakano S. (1993), 'A-Si:H technologies for high efficiency solar cells', J. NonCryst. Solids 164-166, 679-684. Uchida Y. (1984), 'DC glow discharge', in Semiconductors and Semimetals, Vol. 21A, Pankove J. I., ed., Academic Press, Orlando, 41-54. Watanabe T., Azuma K., Nakatani M., Suzuki K., Sonobe T. and Shimada T. (1986), 'Chemical vapor deposition of a-Si:H films utilizing a microwave excited Ar plasma stream', Jpn. J. Appl. Phys. 25-12, 1805-1810. Willing F., Bennett M. and Newton J. (1987), 'Thermal stability of interconnected aSi:H solar modules', Conf. Record 19th. IEEE Photovoltaic Specialists Conf., New Orleans, IEEE Press, Piscataway, 1086-1089. Woodcock J. M., Schade H., Maurus H., Dimmler B., Springer J. and Ricaud A. (1997), 'A study of the upscaling of thin film solar cell manufacture towards 500 MWp per annum', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 857-860. Wronski C. R., Carlson D. E. and Daniel R. E. (1976), 'Schottky barrier characteristics of amorphous silicon diodes', Appl. Phys. Lett. 29, 602-605. Wronski C. R., Abeles B., Tiedje T. and Cody G. D. (1982), 'Recombination centers in phosphorus-doped hydrogenated amorphous silicon', Solid State Commun. 44, 1423-1426. Wronski C. R. (1984), 'The Staebler-Wronski effect', in Semiconductors and Semimetals, Vol. 21C, Pankove J. I., ed., Academic Press, Orlando, 347-373. Wronski C. R. (1996), 'Amorphous silicon technology: coming of age', Solar Energy Mat. Solar Cells 41-42, 427-439. Wronski C. R. (1997), 'The light-induced changes in a-Si:H materials and solar cells—where we are now', Mat. Res. Soc. Symp. Proc. 467, 7-17. Wronski C. R., Lu Z., Jiao L. and Lee Y. (1997), 'An approach to self-consistent analysis of a-Si:H material and p-i-n solar cell properties', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 587-590. Yablonovitch E. and Cody G. D. (1982), 'Intensity enhancement in textured optical sheets for solar cells', IEEE Trans. Electron Devices 29, 300-305. Yang J. and Guha S. (1992), 'Double-junction amorphous silicon-based solar cells with 11% stable efficiency', Appl. Phys. Lett. 61, 2917-2919

Amorphous Silicon Solar Cells

243

Yang J., Banerjee A., Glatfelter T., Hoffman T., Xu, X. and Guha S. (1994), 'Progress in triple-junction amorphous silicon-based alloy solar cells and modules using hydrogen dilution', Conf. Record 24th. IEEE Photovoltaic Specialists Conf., Waikoloa, IEEE Press, Piscataway, 380-384. Yang J., Banerjee A., Lord K. and Guha S. (1998), 'Correlation of component cells with high efficiency amorphous silicon alloy triple-junction solar cells and modules', Proc. 2nd. World Conf. Photovoltaic Solar Energy Conversion, Vienna, Joint Research Centre of the European Commission, EUR18656EN, 387-390. Yang L. and Chen L. F. (1994), 'The effect of H2 dilution on the stability of a-Si:Hbased solar cells', Mat. Res. Soc. Symp. Proc. 336, 669-674. Zanzucchi P., Wronski C. R. and Carlson D. E. (1977), 'Optical and photoconductivity properties of discharge produced a-Si', J. Appl. Phys. 48, 5227-5236.

CHAPTER 6

CADMIUM TELLURIDE SOLAR CELLS DIETER BONNET ANTEC GmbH, D-65779 Kelkheim, Germany

Will you walk into my wavetrap? said the spiter to the shy. James Joyce, Finnegans Wake, 1939.

6.1 Introduction Until today, silicon has been used as base material for photovoltaic solar energy conversion with increasing success, albeit at relatively high cost. Only three additional semiconductors have shown real promise for replacing silicon as the primary material for PV power generation: amorphous silicon, CdTe and Cu(In,Ga)Se2. Other materials, including Se, Cu2S, Cu 2 0, InP, CdSe and Zn3P2, have been studied, but, because of disappointing results or high cost, are no longer intensively investigated. Only GaAs is being developed and used for special applications, where very high efficiency is required in spite of high cost. The discussion which follows describes the technical status and industrial prospects of CdTe thin-film solar cells. From its basic physico-chemical properties CdTe is an optimum material for use in such cells. Work to master the technology for large-scale production has been highly successful. The market introduction of commercial products is imminent. Thin-film solar cells are large area diodes tailored to enable and maximise the absorption of light within a short distance from its space-charge region. The absorbed photons create electron-hole pairs. The potential energy of an excited (minority) carrier is converted into electrical energy as it is swept through the built-in electric field of the diode. The separation from its opposite (majority) charge carrier leads to an electric voltage, which can drive a current through an external circuit, such as an electric motor. As a minority carrier device, a solar cell requires a material of good electronic base properties—mainly high minority carrier lifetime and mobility. This can be achieved only by good crystalline properties, chemical purity, suitable doping and low-resistance contacting. Although there are several varieties of solar cells, the following general description applies most directly to thin-film solar cells in which the diode is created by two materials designated as the window layer and the absorber—'heterojunction'

245

246

D. Bonnet

devices. The field region is generated at the interface between window and absorber and mainly resides in the absorber. In our case CdTe is the absorber.

6.2 Early work There have been many efforts in the past 35 years to design and realise good junctions to CdTe films for extraction and collection of light-generated charge carriers (Bube, 1988). In the case of thin-film CdTe p-n homojunctions, there has been very limited success, because of strong light absorption in CdTe, a direct-gap semiconductor, coupled with a high surface recombination rate that severely limits the minority carrier lifetime and results in low quantum efficiencies. Furthermore, it is difficult to manufacture CdTe p-n junctions in thin-film form as the interdiffusion of doping species along grain boundaries degrades and distorts the junction. CdTe tends to beptype and is difficult to manufacture in n-conducting form and therefore an n-type heterojunction partner is required in order to induce a strong space-charge region as a prerequisite for good efficiency. Heteroj unctions are therefore the most promising configuration. The first heteroj unction was the n-CdTe//?-Cu2Te junction (analogous to the CdS/Cu2S solar cell under study at that time: Cusano, 1963). Although efficiencies around 7% were achieved, stability problems arising from the diffusion of Cu stopped further development of this cell structure. A heteroj unction partner with wider bandgap than CdTe allows light to enter the CdTe material more readily, by the so-called 'window effect'. Around 1970, a new heteroj unction was identified for CdTe with CdS as the n-partner (Bonnet and Rabenhorst, 1972), and this has had much success. Over the course of time, a concentration process has taken place so that most research and commercial interest is now focussed on CdTe/CdS p-n heterojunctions. The rationale for this selection is discussed below, but it should be admitted that it is at least partly empirical {i.e. 'the CdTe/CdS structure works').

6.3 The potential of the base material 6.3.1 Energy gap CdTe has an energy gap of 1.45 eV, and is therefore very well adapted to efficient conversion of solar light into electricity. Furthermore, the energy gap is 'direct', resulting in an absorption coefficient of >105 cm"1 for visible light, so that the absorber layer needs to be only a few /xm thick to absorb >90% of photons at energies >1.45 eV.

247

Cadmium Telluride Solar Cells

Current densities of 27 mA cm 2 and open-circuit voltages of 880 mV, leading to AM 1.5 efficiencies of 18.5%, can be expected for cells made from CdTe (Sites and Liu, 1995).

6.3.2 Thermodynamic properties The phase diagram of CdTe is reproduced in Fig. 6.1 (Zanio, 1978). Above 400 C, the stoichiometric compound is the stable solid phase, because the constituting elements have a significantly higher vapour pressure than the compound. In the high-temperature phase a slight nonstoichiometry is present in the form of a slight Cd deficiency, which leads to a native p-doping of the material. This property makes it relatively easy to produce CdTe films suited for thin-film solar cells. No excessive care has to be taken in 1200 1000

" 1

1

1 — ' 1092 ± 1°'

—1

1

1

Liquidus

_ - .

^^wLiquidus Solidus

o E £

. _ ^ -

600 449 + 2°

400

200

-

324 + 2°

• 0

i

I

i

10

20

30

I 40

50

'

60

1 70

1 80

1, 90

100

Atom fraction Te

Figure 6.1

Phase diagram of CdTe (from Zanio, 1978).

preparing the CdTe films as long as the substrate temperature is sufficiently high. CdTe or Cd + Te can be used as starting materials. The only requirement is the absence of disturbing impurities, which might impair the doping. In practice, the compound can easily be prepared in sufficiently high purity, as the constituting elements—Cd and Te—can easily be purified by standard chemical procedures. Due to the material's high ionicity (72%) (Hartmann et al., 1981), fewer dangling bonds occur at grain boundaries and crystallites tend to be well passivated. The energy of all photons in the solar spectrum is lower than the bond energy (5.75 eV) of CdTe, and this strong bonding leads to extremely high chemical and thermal stability. The

248

D. Bonnet

energy of solar photons is used only for the photovoltaic effect or the generation of harmless phonons, and it cannot break chemical bonds and destabilise the material. 6.3.3 Crystal lattice The natural crystal lattice of CdTe (Fig. 6.2)—being formally cubic—is de facto hexagonal: if viewed perpendicular to the direction of the cubic 111 axis, stacked planes of hexagonally packed alternating Cd and Te layers can be identified. In most deposited CdTe films, these planes tend to lie in the plane of the substrate (the 111 axis being perpendicular to the substrate), leading to columnar growth of crystallites.

(a)

(b)

Figure 6.2 Crystal lattice of CdTe in (a) the cubic representation and (b) as seen perpendicular to the cubic (111) axis, illustrating its quasi-layer structure.

6.3.4 Growth and doping of films On heating in vacuum to about 700 C, CdTe sublimes congruently, liberating Cd and Te in equal amounts, the residue remaining stoichiometric CdTe. On arrival of Cd and Te on the substrate, even in a non 1:1 ratio, CdTe condenses stoichiometrically as long as the substrate is heated above 449 C, at which temperature excess Cd and Te are not stable (see Fig. 6.1). In many cases, films deposited at lower temperatures, and therefore not necessarily at stoichiometric ratio, can be heated to create the stoichiometric compound. This allows numerous film deposition technologies to be applied. Moreover, as the material grows natively p-doped in thin-film form, no additional doping has to be introduced. Oxygen, being isovalent with Cd, is not a critical impurity, and may even enhance p-doping (Tyan and Perez-Albuerne, 1982). In many cases, quite large crystallites (up to 10 fim in diameter) will grow. The best films have been grown at

249

Cadmium Telluride Solar Cells

substrate temperatures around 600 C and deposition rates of ~1 nm per minute (Ferekides etal., 1993).

6.4 Diodes and cells Like CdTe, CdS has a strong tendency to form stoichiometric films, but, unlike CdTe, CdS films are natively n-doped by a slight non-stoichiometry. CdS can be deposited by essentially the same techniques as CdTe, permitting compatibility of manufacturing. A potential disadvantage is that CdS has a significant lattice mismatch to CdTe. Fortunately, after the post-deposition treatments described below, the negative consequences of this are only mild. back contact

p-CdTe (3-5 prn)

l&

n-CdS(100nm)

i^sssssssssssssssssssssssssss'^a^TC0 < 200 nm >

incident light

Figure 6.3

Film sequence of the CdTe thin-film solar cell as used today.

The n-CdS/p-CdTe heterojunction solar cell must be illuminated through the CdS window, so that the light is absorbed in the CdTe close to the junction. In the preferred fabrication procedure, the n-CdS film is deposited onto a transparent conductive oxide (TCO) film, typically ln 2 0 3 or Sn0 2 . Next the CdTe is deposited onto the CdS, and finally a low-resistance contact is made to the CdTe followed by a back electrode, which can be opaque. Figure 6.3 shows the superstrate cell structure. Figure 6.4 shows the energy diagram of the heterojunction. CdS is heavily n-doped and its conductivity under cell operating conditions increases on illumination (an effect known as 'light doping'), whereas CdTe is lightly p-doped (typically to a level of p < 10IS cm"3). Therefore essentially all the electric field drops within the CdTe layer. This field extends to a depth of about 1 pm, a value comparable with the optical absorption length. Light-generated electrons in the CdTe experience a drift field and move toward the junction into the CdS. Due to the strong absorption of CdTe (a > 104

250

D. Bonnet

conduction band

metal contact

Figure 6.4

Energy band diagram of the typical CdS/CdTe thin-film solar cell.

move toward the junction into the CdS. Due to the strong absorption of CdTe (a > 104 cm"1), the majority of electrons is generated in the field region of the CdTe layer and under influence of the field drift rapidly towards the junction. Hence charge separation does not have to rely on diffusion, which would be much less effective owing to the small lifetime (< 1 ns) of minority carriers in /?-CdTe. Electronic defects are primarily located at the metallurgical junction between CdS and CdTe, and these can act as recombination centres for minority carriers. Fortunately they can be significantly reduced in number by a special 'activation/passivation' step, discussed below in Section 6.5.3. As light-generated holes in CdS—being minority carriers in this layer—have a short lifetime and experience no drift field, they do not contribute to the photocurrent. Therefore the thickness of the CdS layer should be reduced as far as possible, to allow as much light as possible to penetrate through the CdS 'window' and enter the CdTe film. Interface states also act as recombination centres for the majority carriers (electrons in CdS and holes in CdTe) that cross the junction under forward bias. This leads to increased dark currents and thereby decreased photovoltage. It is thus evident that the action of defects must be reduced for optimum performance of the solar cell. It is also possible to make a CdTe substrate cell in which CdTe is laid down on a substrate and then n-CdS and TCO are added successively by deposition. However, the TCO-superstrate configuration is more successful, probably because of the material properties of the films involved. TCO is often deposited at temperatures above 600 C and is relatively stable with respect to typical CdTe device processing. Device-quality CdS is readily deposited onto the TCO, and CdTe deposition and post-processing

Cadmium Telluride Solar Cells

251

(which often requires heat treatments above 400 C) can be performed with minimal damage to the CdS. In fact, although there is some interdiffusion between CdS and CdTe, there is a miscibility gap between the two compounds that limits the composition of the alloy to a few percent substitution of either chalcogenide. The final step in the fabrication of the superstrate cell is the deposition of a low-loss electrical contact to CdTe. Although there are many ways of doing this, contact fabrication is typically the most delicate step and no contact processing temperatures exceed 270 C. Thus use of the TCO-superstrate configuration enables use of process steps with decreasing temperatures as the device is fabricated, whereas the alternative deposition sequence would require the relatively delicate CdTe contact to be made early in the fabrication process. These basic materials aspects have led to remarkable success in research organisations and universities. Indeed, then-record efficiencies of 15.8% were obtained on 1 cm2 cells (Ferekides et al., 1993). Recently a new record efficiency of 16.0% has been announced by Ohyama et al. (1997).

6.5 Cell production If CdTe thin-film solar cells are to become a commercially successful product, their promising basic properties have to be retained while meeting the following criteria: • • • • • •

high cell efficiency (10-15%) high module production speed (100,000 m2 p.a.) robust, forgiving manufacturing processes cheap substrate (commercial glass) low materials consumption (8% for 30 x 30 cm2 modules (Woodcock et al., 1995). If many substrates are coated in parallel, the long deposition time of about 1 hour can be compensated by high throughput.

Cadmium Telluride Solar Cells

255

Screen printing CdS and CdTe films can be screen printed from slurries containing CdS and CdTe (or Cd and Te powders), respectively, plus CdCl2 as flux (Yoshida, 1995; Cleminck et al., 1992). Films are then given a heat treatment for about an hour in a controlled atmosphere at about 700 C (for CdS) or 600 C (for CdTe), to produce large-grained films with thickness from 15 to 30 fim. Efficiencies above 12% have been reported. This process has appeal for manufacturing due to the simplicity of the process and equipment. On the other hand, semiconductor film thickness is 3 to 6 times that of films made by other techniques, the process involves several hours of heat treatment to produce high-quality films, and high-quality substrates (made of borosilicate glass) are required. Chemical vapour deposition Chemical vapour deposition (CVD) has some resemblance to the spraying process, insofar as CdTe is formed by chemical reaction from thermally decomposable compounds. In the case of CVD, the compounds are gaseous and are injected into the reactor by a carrier gas, e.g. H2. Typically metal-organic compounds such as dimethyl cadmium and diethyl tellurium are used as precursors for the reaction (Ghandi et al., 1987; Rohatgi, 1992). CVD has the advantage that doping species such as P or As can also be introduced (e.g. in the form of thermally decomposable AsH3 or PH3) by a suitable gas-mixing system. This process, although slower than the fast physical vapour deposition processes (/*m If1 vs. ^m min"1 ) has wide process latitude in gas composition, allowing basic studies to be made. For example, CVD has been used at Georgia Institute of Technology, where one interesting result has shown that, even under very strong deviations of the Cd:Te ratio from 1:1, device-quality stoichiometric films can be made, again giving proof of the latitude available for CdTe processing. Efficiencies achieved on experimental cells have been well above 10%. However, due to the toxicity, high cost and low materials efficiency of the metal-organic gases, this process is generally considered less suited for large-scale production of CdTe thin-film solar cells. Atomic layer epitaxy (ALE) In this process alternate monolayers of Cd and Te are deposited on the substrate by alternately directing gas streams containing Cd or Te onto it. This allows very stoichiometric and pure films to be grown. Cd and Te are evaporated into the inert gas streams in a closed system at elevated temperatures. The gas streams are of high

256

D. Bonnet

temperature and are guided inside high-temperature tubing to avoid condensation. The substrate is also heated and the deposition is driven by the chemical bonding energy between Cd and Te. Cells of 14% efficiency have been reported, and modules of 5 x 5 cm2 area at efficiencies above 10% have been made by Microchemistry Inc. (Skarp et al., 1991 and 1992). This process has some similarity to that used for the very first CdS/CdTe cells around 1970. Here the compound CdTe had been evaporated into an inert gas-stream which had been guided onto a substrate at lower temperatures—but still around 500 C (Bonnet and Rabenhorst, 1972). ALE requires very low deposition rates, but enables multiple glasses to be coated in parallel, as does electrodeposition. The technology has not been pursued further at the time of writing this chapter. Sputtering Bombardment with argon ions of a solid target of CdTe leads to emission of Cd and Te from the surface of the target. The atoms move in the ambient vacuum and condense on the substrate, forming CdTe films at suitable temperatures of up to 300 C. This technology has led to good results in first experiments at NREL (Abou-Elfotouh and Coutts, 1992) and the University of Toledo (Compaan etai, 1993). Deposition rates are typically < 100 nm min"', lower by a factor of 10 than for CSS. This process may gain industrial application for deposition of semiconductor back contacts, e.g. ZnTe, as such contacts are typically very thin (Gessert et al., 1995).

6.5.3 The CdS/CdTe interface and activation of the cell CdS and CdTe in thermal equilibrium can form mixed compounds CdS^Te,^ only for limited ranges of x (0<x10 % efficient devices. In general this topic is treated as confidential by industrial companies, and is considered too technology-oriented by universities to merit intense study. However, the back contact is fundamentally important because it bears on the essential question of long-term stability. Unfortunately, the highest efficiency cells have been made using a Cu-doped graphite back-contact layer that is not well suited for monolithic integration of cells into a large-area (60 x 120 cm2) modules. There are two general principles for making ohmic contacts to p-type semiconductors: 1. To use a metal of work function higher than the electron affinity of the semiconductor in order to align the top of the valence band with the Fermi level of the metal. The electron affinity of CdTe is 4.3 eV. 2. To create a highly doped back-surface layer in the semiconductor. The barrier created by the back-contact metal in the semiconductor will then be thin enough for holes to tunnel through efficiently. The problems for CdTe are evident for both cases: low-cost metals of work function greater than 4.5 eV are not available, and p-doping in CdTe suffers from a strong tendency for acceptors to self-compensate. Furthermore, acceptors cannot be introduced by

260

D. Bonnet

diffusion doping from the surface, as dopants generally diffuse preferentially along grain boundaries, leading to shunting of the cell before sufficient doping levels can be achieved. The latter problem can be alleviated to some extent by using CdTe films of higher-than-necessary thickness, e.g. >5 /im. In practice, some methods may combine approaches 1 and 2 by first contacting CdTe with a more easily doped semiconductor (generating a p-p* junction), and then with metals of high work function. Efforts have primarily been directed towards four semiconductors: HgTe, ZnTe:Cu, Te and Cu2Te (Chu, 1988; Meyers etal., 1990; Tyan, 1980; Tang et al., 1991; Ferekides et al., 1997b). All three are p-type semiconductors with suitable work functions. Nevertheless, modification of the surface of a polycrystalline thin film has proved to be a delicate and often material- and morphology-dependent process (Pompon, 1985; Fahrenbruch, 1987). A very thin film of Te, created by chemical etching of the CdTe surface after activation or deposition of Te, seems to be especially beneficial to the back contact (Sasala, 1997). In many cases, such as the above-mentioned graphite contact, copper (an acceptor in CdTe) is added, which on annealing can diffuse into the CdTe film. An adverse consequence of this can be reduced stability, as Cu has high diffusivity in most semiconductors (McCandless, 1995; Chou, 1995) and may propagate into the semiconductor even at cell operating temperatures (-60 C). If Cu reaches the junction, it first reduces the junction width, and then it compensates donors in the CdS layer. The photocurrent is reduced in both cases. Another alternative has been to dope the graphite with HgTe, but this also has some unsatisfactory practical aspects (Niles et al., 1996). Fortunately, recent analysis (Sites and Lui, 1995) has shown that a completely nonrectifying back contact is not required. At room temperature, a contact barrier of 200 meV will not lead to a noticeable reduction of output. The thermally activated reverse currents in such junctions generate only a low series resistance. Only if the barrier height increases with time will degradation occur. This type of contact and its time-dependent behaviour can be easily identified by observing the l-V curves in the forward direction. Observations of an increasing 'rolloff (or current saturation) can be used as early warning for degradation, even before the power output suffers. Unfortunately, publications rarely show I-V curves at / > 0 in the forward direction, and it is recommended that they should (Bonnet et al., 1992). Medium-term measurements on technical pilot modules exposed outdoors, showing minimal degradation for over 20,000 hours of continuous illumination at 0.8 Sun intensity, have been performed for BP Solarex and Solar Cells Inc. (which has since changed its name to First Solar LLC) by Sasala et al. (1996). The success of these tests is a strong indication that long-term stability can be achieved by careful design of the back contact.

Cadmium Telluride Solar Cells

261

6.5.5 Substrates The substrate onto which thin-film solar cells are deposited will to a large extent determine the cost of the final module. Many experiments have been made on borosilicate glass, which is stable at temperatures up to 600 C. The cost of this type of glass is about twice that of standard soda-lime window glass, which costs about $5 per m2. Whereas the world-record cell was deposited at low speed and elevated temperature onto borosilicate glass, the use of soda-lime glass under suitably reduced substrate temperatures has also resulted in efficiencies above 13% (Ferekides, 1994). There exists strong evidence that a certain amount of diffusion of Na from the soda-lime glass into the growing film promotes improved structure and properties. Comparison of cells made on glass + TCO and glass + Na-diffusion barrier (Si02) + TCO shows improved results for the first option (Bonnet et al., 1994). All industrial efforts today use soda-lime standard window glass as substrate.

6.5.6 The TCO film The transparent conducting oxide (TCO) base contact of the diode is an essential part of the cell and its optimisation can lead to a significant improvement in cell and module efficiency. Generally, metal oxides based on indium and tin are used, which can be made highly conductive by either native or additive doping while still keeping the high optical transmission that results from their large energy gap. Free carrier absorption and impurities lead to reduced transmission. Generally a TCO film is considered good if its conductivity is

two-phase region

A

DTA heating

Figure 7.4 Quasi-binary phase diagram of CuInSe2 established by Differential Thermal Analysis (DTA) and microscopic phase analysis. Note that at 25% Cu no single phase exists. After Haalboom el at. (1997).

The existence range of the a-phase in pure CuInSe2 on the quasi-binary tie line Cu2Se-In2Se3 extends from a Cu content of 24% to 24.5%. Thus, the existence range of single-phase CuInSe2 is astonishingly small and does even not include the stoichiometric composition of 25% Cu. The Cu content of absorbers for thin-film solar cells typically varies between 22 and 24 at. % Cu. At the growth temperature this region lies within the single-phase region of the a-phase. However, at room temperature it lies in the two-phase a + /? region of the equilibrium phase diagram in

Cu(ln,Ga)Se2 Solar Cells

283

Haalboom et al. (1997). Hence one would expect a tendency for phase separation in photovoltaic-grade CuInSe2 after deposition. Fortunately, it turns out that partial replacement of In with Ga, as well as the use of Na-containing substrates, considerably widens the single-phase region in terms of (In + Ga)/(In + Ga + Cu) ratios (Herberholz et al, 1999). Thus, the phase diagram hints at the substantial improvements actually achieved in recent years by the use of Na-containing substrates, as well as by the use of Cu(In,Ga)Se2 alloys.

7.2.3

Defect physics of Cu(In, Ga)Se2

Basics The role of defects in the ternary compound CuInSe2, and even more in Cu(In,Ga)Se2, is of special importance because of the large number of possible intrinsic defects and the role of deep recombination centres in the performance of the solar cells. For insight into the defect physics of Cu(In,Ga)Se2, see Cahen (1987), and for a recent discussion see Burgelman et al. (1997). The challenge of defect physics in Cu(In,Ga)Se2, according to Zhang et al. (1998), is to explain three unusual effects in this material: (i) the ability to dope Cu(In,Ga)Se2 with native defects; (ii) the structural tolerance to large off-stoichiometries; and (iii) the electrically neutral nature of the structural defects. It is obvious that the explanation of these effects significantly contributes to the explanation of the photovoltaic performance of this material. It is known that the doping of CuInSe2 is controlled by intrinsic defects. Samples with ptype conductivity are grown if the material is Cu-poor and annealed under high Se vapour pressure, whereas Cu-rich material with Se deficiency tends to be n-type (Migliorato et al., 1975; Noufi et al, 1984). Thus, the Se vacancy Vse is considered to be the dominant donor in n-type material (and also the compensating donor in p-type material), and the Cu vacancy VCu the dominant acceptor in Cu-poor p-type material. Theoretical considerations By calculating the metal-related defects in CuInSe2 and CuGaSe2, Zhang et al. (1998) found that the defect formation energies for some intrinsic defects are so low that they can be heavily influenced by the chemical potential of the components (i.e., by the composition of the material) as well as by the electrochemical potential of the electrons. For VCu in Cu-poor and stoichiometric material, a negative formation energy is even calculated. This would imply the spontaneous formation of large

284

U.RauandH. W. Schock

numbers of these defects under equilibrium conditions. Low (but positive) formation energies are also found for the Cu-on-In antisite Cuto in Cu-rich material (this defect is a shallow acceptor which could be responsible for the />type conductivity of Cu-rich, non-Se-deficient CuInSe2). The dependence of the defect formation energies on the electron Fermi level could explain the strong tendency of CuInSe2 to selfcompensation and the difficulties of achieving extrinsic doping. The work of Zhang et al. (1998) provides a good theoretical basis for the calculation of defect formation energies and defect transition energies, which exhibit good agreement with experimentally obtained data. Further important results in Zhang et al. (1997) are the formation energies of defect complexes such as (2Vcu,InCu), (Cu^Inc) and (ZCu^Cu^), where CUJ is an interstititial Cu atom. These formation energies are even lower than those of the corresponding isolated defects. Interestingly, (2Vcu,InCu) does not exhibit an electronic transition within the forbidden gap, in contrast to the isolated InCu-anti-site, which is a deep recombination centre. As the (2Vcu,InCu) complex is most likely to occur in Inrich material, it can accommodate a large amount of excess In (or likewise deficient Cu) and, at same time, maintain the electrical performance of the material. Furthermore, ordered arrays of this complex can be thought as the building blocks of a series of Cu-In-Se compounds such as CuIn3Se5 and CuIn5Se8 (Zhang et al., 1997). Table 7.1 Electronic transition energies and formation energies of the twelve intrinsic defects in CuInSe2 Defect transition energies" and formation energies''

Trans -ition

Vcu

Vi„

HO)

0.03

0.17

0.03

0.04

H2-)

0.41

(2-/3-)

0.67

V«

Cm In,

Se,

Cum

Sec

Cus«

Se.„

0.75

0.08

Inse

0.29 0.04c

0.07

0.05 0.58

0.2

(0/+)

0.25

0.11'

0.08

0.07

2.6

2.88

9.1

0.06

0.04

0.09

0.44

(+/2+) Atf/eV

Incu

0.60

3.04

2.9

2.8

4.4

22.4

3.34

1.54

1.4

7.5

7.5

7.5

5.5

5.0

"Difference between the valence/conduction band energy for acceptor/donor states; formation energy At/ of the neutral defect in the stoichiometric material; ccovalent; rfionic. All energies in eV. Source: The ionisation energies in italics are derived from Abou-Elfotouh et al. (1991), and the formation energies in brackets from Neumann (1983). All the bold numbers are from Zhang et al. (1998).

Cu(ln,Ga)Se2

285

Solar Cells

Table 7.1 summarises the ionisation energies and the defect formation energies of the twelve intrinsic defects in CulnSe2. The energies given in bold type for VCu, V,„, Cm, Cm„ and Inc„ were obtained from a first-principles calculation, whereas the formation energies in italics were calculated from the macroscopic cavity model. The numbers in the first column represent the ionisation states of the defects. Device-relevant

defects

Let us now concentrate on the defects experimentally detected in photovoltaic grade (and thus In-rich) polycrystalline films. In-rich material is in general highly compensated, with a net acceptor concentration of the order of 1016 cm" . The shallow acceptor level Vc„ (which lies about 30 meV above the valence band) is assumed to be the main dopant in this material. As compensating donors, the Se-vacancy Vsc as well as the double donor InCu are considered. The most prominent defect is an acceptor level about 270-300 meV above the valence band, which is reported by several groups from deep-level transient spectroscopy (Igalson and Schock, 1996) and admittance spectroscopy (Schmitt et al., 1995; Walter et al., 1996b). This defect is also present in single crystals (Igalson et al., 1995).

high-energy tail D, ~exp(-l//l/*)

£ KT1

0.1

0.2 0.3 0.4 0.5 Activation energy C/eV

0.6

Figure 7.5 Defect density spectrum obtained from admittance spectroscopy of a ZnO-CdS-CuInGaSe2 heterojunction. The peaks Ni and N2 can be related to interface and bulk defects (see inset).

As an example. Fig. 7.5 displays a defect density spectrum obtained from admittance spectroscopy by the method of Walter et al. (1996a). The transition at -300 meV exhibits a broadened energy distribution with a tail in the defect density

286

U. Rau and H. W. Schock

towards larger energies. This tail-like distribution is best described by a characteristic energy U*, as shown in Fig. 7.5. This defect is detected, not only in In-rich, but in equal amounts also in Cu-rich polycrystalline materials (Herberholz, 1998). An assignment of this defect to the Cu^ anti-site is in agreement with the theoretical calculations of Zhang et al. (1998) as well as with the proposition of several experimentalists. The importance of this transition derives from the fact that its concentration is related to the open-circuit voltage of the device (Herberholz et al, 1997a) and that the defect seems to be involved in the defect metastability (Igalson and Schock, 1996) (cf. Section 7.4.6). The lower-energy transition in Fig. 7.5 is attributed to interfacial defects, rather than to a bulk defect (Herberholz et al., 1998) because its activation energy can vary between 50 meV and 250 meV depending on air-annealing prior to the measurement (Rau et al, 1999a). Thus, the activation energy of this transition measures the depth At/ Fn from the vacuum level of the (electron) Fermi level and the conduction-band energy at the Cu(In,Ga)Se2 surface (Herberholz et al, 1998), as shown in the inset of Fig. 7.5.

7.3 7.3.1

Cell and module technology Structure of the heterojunction solar cell

The complete layer sequence of a ZnO/CdS/Cu(In,Ga)Se2 heterojunction device is shown in Fig. 7.6. It consists of a typically 1 fjm thick Mo layer deposited on a sodalime glass substrate and serving as the back contact for the solar cell. The Cu(In,Ga)Se2 is deposited on top of the Mo back electrode as the photovoltaic absorber material. This layer has a thickness of 1-2 /an. The heterojunction is then completed by chemical bath deposition (CBD) of CdS (typically 50 nm) and by the sputter deposition of a nominally undoped (intrinsic) /-ZnO layer (usually of thickness 50-70 nm) and then a heavily doped ZnO layer. As ZnO has a band-gap energy of 3.2 eV it is transparent for the main part of the solar spectrum and therefore is denoted as the window layer of the solar cell. We will first mention four important technological innovations which, during the last decade, have led to a considerable improvement of the efficiencies and finally to the record efficiency of 18.8% (Contreras et al, 1999). These steps are the key elements of the present Cu(In,Ga)Se2 technology.

Cu(ln,Ga)Se2

287

Solar Cells

,ZnO:AI - /-ZnO .CdS - Cu(ln,Ga)Se2 .Mo

. glass

^ Figure 7.6

Schematic layer sequence of a standard ZnO-CdS-Cu(In,Ga)Se2 thin-film solar cell.

1. The film quality has been substantially improved by the crystallisation mechanism induced by the presence of CuvSe (y < 2). This process is further supported by a substrate temperature close to the softening point of the glass substrate (Stolt et al., 1993). 2. The glass substrate has been changed from Na-free glass to Na-containing sodalime glass.(Hedstrom et al., 1993; Stolt et al., 1993). This has led to an enormous improvement of the efficiency and reliability of the solar cells, as well as to a larger process tolerance. It was first assumed that this improvement was due to better match of thermal expansion coefficients, but the beneficial impact of Na—diffusing from the substrate through the Mo back contact—on the growth of the absorber layer and its structural and electrical properties was soon recognised. 3. Initially, the absorbers consisted of pure CuInSe2- The partial replacement of In with Ga (Devaney et al., 1990) is a further noticeable improvement, which has increased the band gap of the absorber from 1.04 eV to 1.1-1.2 eV for the highefficiency devices. The benefit of 20-30% Ga incorporation stems not only from the better band-gap match to the solar spectrum but also from the improved electronic quality of Cu(In,Ga)Se2 with respect to pure CuInSe2 (Hanna et al., 2000; Herberholz etal., 1999). 4. The counter electrode for the CuInSe2 absorber of the earlier cells was a 2 ,um thick CdS layer laid down by Physical Vapour Deposition (PVD). This has been replaced by a combination of a 50 nm thin CdS buffer layer laid down by chemical bath deposition (Potter et al., 1985; Birkmire et al., 1989; Mauch et al, 1991) and a highly conductive ZnO window layer.

288

U. Rau andH. W. Schock

The effect of items 1-4 on the electronic properties and performance of Cu(In,Ga)Se2 solar cells will be considered in detail below, as we discuss the preparation of a Cu(In,Ga)Se2 solar cell step by step.

7.3.2

Absorber preparation techniques

Basics The preparation of Cu(In,Ga)Se2-based solar cells starts with the deposition of the absorber material on a Mo-coated glass substrate (preferably soda-lime glass). The properties of the Mo film and the choice of the glass substrate are of primary importance for the final device quality, because of the importance of Na, which diffuses from the glass through the Mo film into the growing absorber material. Some processes use blocking layers such as SiN^, Si0 2 or Cr between the glass substrate and the Mo film to prevent the out-diffusion of Na. Instead, Na-containing precursors like NaF, Na2Se or NaS are now deposited prior to absorber growth to provide a controlled, more homogeneous, incorporation of Na into the film. The control of Na incorporation in the film from precursor layers allows the use of other substrates like metal or polymer foils. There seems to be no fundamental efficiency limitation due to the substrate provided a proper supply of sodium is provided. During absorber deposition, a MoSe2 film forms at the Mo surface (Wada et ah, 1996; Takei etal., 1996). MoSe2 is a layered semiconductor withp-type conduction, a band gap of 1.3 eV and weak van der Waals bonding along the c-axis. If the layer were oriented parallel to the plane of contact, the MoSe2 would inhibit adhesion of the absorber as well as leading to unfavourable electronic transport. Fortunately, the caxis is found to be in parallel with, and the van der Waals planes thus perpendicular to the interface (Wada et al., 1996). Because of the larger band gap of the MoSe2 compared with that of standard Cu(In,Ga)Se2 films, the MoSe2 layer provides an electronic mirror for the photogenerated electrons and at the same time provides a low-resistance contact for the holes (see Section 7.4.1). Photovoltaic-grade Cu(In,Ga)Se2 films have a slightly In-rich overall composition. The allowed stoichiometry deviations are astonishingly large, yielding a wide process window with respect to composition. Devices with efficiencies above 14% are obtained from absorbers with (In + Ga)/(In + Ga + Cu) ratios between 52 and 64% if the sample contains Na (Ruckh et al., 1994a). Cu-rich Cu(In,Ga)Se2 shows the segregation of a secondary Cu2.ySe phase preferentially at the surface of the absorber film. The metallic nature of this phase does not allow the formation of efficient

Cu(In,Ga)Se2

Solar Cells

289

Vapour

Fte-es/aporation Conderealion ,

Liquid

QoAth /

Solid

/ QjInSea

Substrate

Figure 7.7 Schematic illustration of the growth of a Cu(In,Ga)Se2 film under Cu-rich conditions. A quasi-liquid Cu-Se phase acts as a flux in a vapour-liquid solid growth mechanism.

heterojunctions. Even after removal of the secondary phase from the surface by etching the absorber in KCN, the utility of Cu-rich material for photovoltaic applications is limited, probably due to the high doping density of 1018 cm -3 in the bulk and the surface defects. 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 /an whereas In-rich films have much smaller grains. A model for the film growth under Cu-rich compositions comprises the role of Cu2o,Se as a flux agent during the growth process of co-evaporated films (Klenk et al., 1993). This model for the growth of Cu(In,Ga)Se2 in the presence of a quasi-liquid surface film of CuySe is highlighted in Fig. 7.7. For Cu(In,Ga)Se2 prepared by selenisation, the role of C^-^Se is similar (Probst et al., 1996), therefore growth processes for high quality have to go through a copper-rich stage and end with an indium-rich composition. Co-evaporation processes The absorber material yielding the highest efficiencies is Cu(In,Ga)Se2 with a Ga/(Ga + In) ratio of -20%, prepared by co-evaporation from elemental sources. Figure 7.8 sketches a co-evaporation set-up as used for the preparation of laboratoryscale solar cells and mini-modules. The process requires a maximum substrate temperature of -550 C for a certain time during film growth, preferably towards the end of growth. One advantage of the evaporation route is that material deposition and film formation are performed during the same processing step. A feed-back loop

290

U.RauandH. W. Schock substrate heater thermo element

chalcogen sources Figure 7.8 Arrangement for the deposition of Cu(In,Ga)Se2 films by co-evaporation on a heated substrate. The rates of the sources are controlled by mass spectrometry.

based on a quadrupole mass spectrometer or an atomic absorption spectrometer controls the rate of each source. The composition of the deposited material with regard to the metals corresponds to their evaporation rates, whereas Se is always evaporated in excess. This precise control over the deposition rates allows for a wide range of variations and optimisations with different sub-steps or stages for film deposition and growth. These sequences are defined by the evaporation rates of the different sources and the substrate temperature during the course of deposition. Figure 7.9 illustrates some of the possibilities, starting with a simple single-step process where all rates as well as the substrate temperature are kept constant during the whole process (Fig. 7.9a). Advanced preparation sequences always include a Cu-rich stage during the growth process and end up with an In-rich overall composition in order to combine the large grains of the Cu-rich stage with the otherwise more favourable electronic properties of the In-rich composition. The first example of this kind of procedure is the so-called Boeing or bilayer process (Mickelsen and Chen, 1980), which starts with the deposition of Cu-rich Cu(In,Ga)Se2 and ends with an excess In rate, as illustrated in Fig. 7.9b. Another possibility is the inverted process where first (In,Ga)2Se3 (likewise In, Ga, and Se from elemental sources to form that compound) is deposited at a lower temperatures (typically around 300 C). Then Cu and Se are evaporated at an

Cu(ln,Ga)Se2 Solar Cells

291

elevated temperature until an overall composition close to stoichiometry is reached (Kessler et al, 1992). This process leads to a smoother film morphology than the bilayer process. The most successful version of the inverted process is the so-called three-stage process (Gabor et al., 1994) shown in Fig. 7.9c. This process puts the deposition of In, Ga, and Se at the end of an inverted process to ensure the overall Inrich composition of the film even if the material is Cu-rich during the second stage. The three-stage process currently leads to the best solar cells. Variations of the Ga/Inratio during deposition, as shown in Fig. 7.9d, allow the design of graded band-gap structures. (Gabor et al., 1996).

a) single layer

h.Go

F.

Cu

6C0 3C0 Time b) bi-layer

Time a) g r a d e d b a n d g a p

Time

Time

Figure 7.9 Schematic rate and substrate temperature profiles for co-evaporation processes. All processes lead to single-phase films, (a) Single-layer process without Cu-rich growth step; (b) bilayer process ('Boeing recipe') with Cu-rich growth at the start; (c) three-stage inverted process with intermediate Cu-rich growth; (d) growth of graded-gap films under Cu-poor conditions

Selenisation

processes

The second class of absorber preparation routes is based on the separation of deposition and compound formation into two different processing steps. High efficiencies are obtained from absorber prepared by selenisation of metal precursors in H2Se (Binsma and Van der Linden, 1982; Chu et al, 1984; Kapur et al, 1987) and by rapid thermal processing of stacked elemental layers in a Se atmosphere (Probst et al.,

U. Rem and H. W. Schock

292

1996). These sequential processes have the advantage that approved large-area deposition techniques such as sputtering can be used for the deposition of the materials. The Cu(In,Ga)Se2 film formation then requires a second step, the selenisation. The very first large-area modules were prepared by the selenisation of metal precursors in the presence of H2Se more than ten years ago (Mitchell et al., 1988). Today, a modification of this process is providing the first commercially available Cu(In,Ga)Se2 solar cells, manufactured by Siemens Solar Industries. This process is

DEPOSITION

heating

SELENISATION

H2S inlet

Figure 7.10 Illustration of the sequential process. Stacked metal layers are selenised and converted into CulnSe2 in FhSe atmosphere.

schematically drawn in Fig. 7.10. First, a stacked layer of Cu, In and Ga is sputterdeposited on the Mo-coated glass substrate. Then selenisation takes place under H2Se. To improve device performance, a second thermal process under H2S is added, resulting in an absorber that is Cu(In,Ga)(S,Se)2 rather than Cu(In,Ga)Se2. A variation of this method that avoids the use of the toxic H2Se during selenisation is the rapid thermal processing of stacked elemental layers (Probst et al., 1996). Here the precursor includes a layer of evaporated elemental Se. The stack is then selenised

Cu(In,Ga)Se2 Solar Cells

293

by a rapid thermal process (RTP) in either an inert or a Se atmosphere. The highest efficiencies are obtained if the RTP is performed in an S-containing atmosphere (either pure S or H2S). 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. However, on the module level, co-evaporated and sequentially prepared absorbers have about the same efficiency. Sequential processes need two or even three stages for absorber completion. These additional processing steps may counterbalance the advantage of easier element deposition by sputtering. Also the detailed and sophisticated control over composition and growth achieved during co-evaporation is not possible for the selenisation process. Fortunately, the distribution of the elements within the film grown during the selenisation process turns out to be close to what one could think to be an optimum, especially if the process includes the sulphurisation stage. Since the formation of CuInSe2 is much faster than that of CuGaSe2, and because film growth starts from the top, Ga is concentrated towards the back surface of the film. An increasing Ga content implies an increase in band-gap energy. This introduces a socalled back-surface field, improving carrier collection at the same time as minimising back-surface recombination. In turn, S from the sulphurisation step is found preferentially towards the front surface of the film, where it reduces recombination losses and also increases the absorber band gap in the space-charge region of the heteroj unction. Other absorber deposition processes Besides selenisation 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. Methods that are used to form epitaxial III-V compound films, such as molecular beam epitaxy (MBE) (Niki et al., 1994) or metal organic chemical vapour deposition (MOCVD) (Gallon et al., 1998) have revealed interesting features for fundamental studies, such as phase segregation and defect formation, but cannot be used to form the base material for high-efficiency solar cells. Attempts to develop so-called low cost processes include electrodeposition, (Abken et al., 1998; Lincot et al., 1998) screen printing and particle deposition (Eberspacher et al., 1998). Electrodeposition can be carried out in either one or two steps. The crucial step is final film formation in a high-temperature annealing process. The recrystallisation process competes with the decomposition of the material, so process optimisation is quite difficult. Cells with good efficiencies were obtained by electrodeposition of a Cu-rich CuInSe2 film and subsequent conditioning by a vacuum

294

U.RauandH. W. Schock

evaporation step of In(Se) (Ramanathan et al, 1998a). Films prepared by spray pyrolysis did not lead to high-performance devices. Influence of sodium The outstanding role played by Na in the growth of Cu(In,Ga)Se2 films was realised some years ago (Hedstrom et al, 1993; Stolt et al, 1993; Ruckh et al, 1994a). In most cases, the Na comes from the glass substrate and diffuses into the absorber. But there are also approaches where Na is incorporated by the use of Na-containing precursors such as NaSe (Holz et al., 1994; Nakada et al, 1997), Na 2 0 2 (Ruckh et al, 1994a), NaF (Contreras et al, 1997a) or Na2S (Nakada et al, 1998). Other alkali precursors have been investigated by Contreras et al. (1997), who found that Nacontaining precursors yielded the best cell efficiencies. The most obvious effects of Na incorporation are better film morphology and higher conductivity of the films (Ruckh et al, 1994a). Furthermore, the incorporation of Na induces beneficial changes in the defect distribution of the absorber films (Keyes et al, 1997; Rau et al, 1998b). The explanations for the beneficial impact of Na are manifold, and it is most likely that the incorporation of Na in fact results in a variety of consequences. During film growth, the incorporation of Na leads to the formation of NaSe* compounds. This slows down the growth of CuInSe2 and could at same time facilitate the incorporation of Se into the film (Braunger et al, 1998b). Also the widening of the existence range of the oc-(CuInSe2) phase in the phase diagram, discussed above, as well as the reported larger tolerance to the Cu/(In + Ga) ratio of Na-containing thin films, could be explained in this picture. Furthermore, the higher conductivity of Na-containing films could result from the diminished number of compensating Vse donors. Wolf et al. (1998) investigated the influence of Na incorporation on the formation of CuInSe2 films from stacked elemental layers by means of thin-film calorimetry. The addition of Na inhibits the growth of CuInSe2 at temperatures below 380 C. The retarded phase formation is responsible for the better morphology in the case of Na-containing samples. Another explanation put forward by Kronik et al. (1998) is that Na promotes oxygenation and passivation of grain boundaries. This could account for the observed enhancement of the net film doping by Na incorporation, through the diminished positive charge at the grain boundaries. It has in fact been observed that the surfaces of Na-containing films are more prone to oxygenation than are Na-free films (Braunger et al, 1998a).

Cu(In,Ga)Se2 Solar Cells

295

The above explanations deal with the role of Na during growth. However, the amount of Na in device-quality Cu(In,Ga)Se2 films is of the order 0.1 at%, which is a concentration of 1020 cm"3 (Niles et al., 1997), and one may ask the question: where are these tremendous quantities of Na in the finished absorber? The electronic effect, i.e. the change of effective doping resulting from Na incorporation, is achieved at concentrations of ~1016 cm"3, four orders of magnitude below the absolute Na content. It has long been believed that the main part of the Na is situated at the film surface and the grain boundaries. Final evidence for this hypothesis was recently found by Niles et al. (1999) with the help of high spatial resolution Auger electron spectroscopy. Heske et al. (1996, 1997) investigated the behaviour of Na on the surface of polycrystalline Cu(In,Ga)Se2 films by X-ray photoelectron spectroscopy (XPS). They found two different species of Na: (i) The first, denoted 'reacted', was observed on the air-exposed sample or after storing the sputter-cleaned sample for three days in an ultra-high vacuum (UHV). The second, denoted 'metallic', was found on clean samples either after annealing at 410 K in UHV or after deliberate Na deposition from a metallic source. The latter species is considered to be the active one during crystal growth. In addition, Heske et al. found an increase of band bending of -150 meV induced by the deposition of Na. This finding, as well as the occurrence of two different Na species, is consistent with results obtained from vacuum-cleaved single crystals (Klein and Jaegermann, 1996). Another interpretation of the beneficial effect of Na is based on the incorporation of Na into the Cu(In,Ga)Se2 lattice (Niles et al., 1997). Niles and co-workers identified Na-Se bonds by means of XPS and concluded that the Na is built into the lattice, replacing In or Ga. The extrinsic defect Nain/Ga should then act as an acceptor and improve the p-type conductivity. The incorporation of Na into the Cu(In,Ga)Se2 lattice is supported by X-ray diffraction measurements that indicate an increased volume of the unit cell (Contreras et al, 1997a). Here, the authors assume that Na in a Cu site prevents the formation of the deep double donor InCu- Schroeder and Rockett (1997) found that Na driven into epitaxial Cu(In,Ga)Se2 films at a temperature of 550 C decreases the degree of compensation by up to a factor 104. Schroeder and Rockett attributed their findings to an Na-enhanced reorganisation of the defects, which allows them to build electrically passive clusters. We see from these numerous approaches that, despite the significance of Na incorporation, the benefit is far from being explained in terms of simple models. However, we feel that in view of the amounts of Na (~0.1 at%) necessary for optimum film preparation, arguments based on its effect on film growth are slightly favoured over those based on the incorporation of Na into the completed film.

296

U. Rau andH. W. Schock

Influence of oxygen Air annealing has been an important process step, crucial for the efficiency especially of the early solar cells based on CuInSe2. Also, though often not mentioned explicitly, an oxygenation step is still used for most of the present-day high-efficiency devices. The beneficial effect of oxygen was explained within the defect chemical model of Cahen and Noufi (1989). In this model, the surface defects at grain boundaries are positively charged Se vacancies V&. (Fig. 7.11a). During air annealing, these sites are passivated by O atoms (Fig. 7.11b). Because of the decreased charge at the grain boundary, the band bending and the recombination probability for photogenerated electrons are reduced. The surface donors and their neutralisation by oxygen are important for the free Cu(In,Ga)Se2 surface as well as for the formation of the surface states

Uc

\

dangling bonds

/ grain boundaries

/

c

\

•

ln;

''

)qVb

^V

Figure 7.11 Band diagram of the conduction and valence band energies across a single grain of Cu(In,Ga)Se2. (a) The electronic states at the grain boundaries are positively charged. This surface charge is compensated by the negative charges in the depleted grain. This induces the band-bending electronic states at the grain boundaries shown at the left-hand grain boundary, and the defect chemical equivalent, dangling bonds, shown at the right-hand boundary; (b) oxygen passivates these dangling bonds and reduces the band bending.

CdS/Cu(In,Ga)Se2 interface (Kronik et al, 2000). Electrical analysis of oxidised and unoxidised samples revealed the validity of the Cahen-Noufi model for the earlier CdS/CuInSe2 devices (Sasala and Sites, 1993), as well as for the more recent ZnO/CdS/Cu(In,Ga)Se2 heterostructures (Rau et al, 1999b; Kronik et al, 1998).

Cu(In,Ga)Se2

297

Solar Cells

(a)

(b)

(c)

Film surface I Grain boundary SCR

Substrata

Substrate

Substrate

Figure 7.12 Illustration of grain boundary and interface charges during the fabrication of a Cu(In,Ga)Se2/CdS heterojunction. (a) On as-prepared films, both grain boundaries and the film surface are positively charged because of the presence of the dangling bonds, (b) Air exposure or air annealing neutralises these charges, (c) The oxide passivation of the surface is removed by the chemical bath deposition process of CdS. The re-established positive charges give rise to a type inversion of the CuInSe2 surface.

The intriguing interplay between surface oxygenation and the deposition of the CdS buffer layer is visualised in Fig. 7.12. In the initial state (Fig. 7.12a), the film surface as well as the grain boundaries are electrically active owing to the positively charged dangling bonds. These charges create a large space-charge region within the grain. Air annealing passivates the dangling bonds at both interfaces. The bands become essentially flat, and space charge essentially vanishes (Fig. 7.12b). Eventually, the chemical bath removes the passivating oxygen and thus re-establishes the beneficial type inversion of the film surface (Fig. 7.12c).

7.3.3 The free Cu(In,Ga)Se2 surface The surface properties of CIGS thin films are especially important, as this surface becomes the active interface of the completed solar cell. However, the band diagram of the ZnO/CdS/Cu(In,Ga)Se2 heterojunction, especially the detailed structure close to the CdS/Cu(In,Ga)Se2 interface, is still under debate. Figure 7.13 depicts three different possibilities corresponding to three different approaches: (a) the ordered defect compound model of Schmid et al. (1993); (b) the surface-state model of Rau et al. (1999a); and (c) the defect layer model of Niemegeers et al. (1998) and Herberholz et al. (1999). The free surfaces of as-grown Cu(In,Ga)Se2 films exhibit two prominent features:

298

U.RauandH. W. Schock

1. The valence band-edge energy Uv lies above the surface-Fermi level £/F by about 1.1 eV for CuInSe2 films (Schmid et al, 1993). This energy is larger than the bandgap energy U"g of the bulk of the absorber material. This was taken as an indication for a widening of band gap at the surface of the film. For example, for the surfaces of Cu(In1_JGax)Se2 thin films it was found that UT-UV = 0.8 eV (almost independent of the Ga content if x > 0) (Schmid et al, 1996b). 2. The surface composition of Cu-poor CuInSe2, as well Cu(In,Ga)Se2 films, corresponds to a surface composition of (Ga + In)/(Ga + In + Cu) of about 0.75 for a range of bulk compositions of 0.5 < (Ga + In)/(Ga + In + Cu) < 0.75. Both observations have led to the assumption that a phase segregation of Cu(In,Ga)3Se5, the socalled Ordered Defect Compound (ODC), occurs at the surface of the films. The segregation of this /3-phase would be compatible with the phase diagram (see Fig. 7.3) and, as this material displays n-type conductivity, could yield the explanation for the surface type inversion. Unfortunately, the existence of a separate phase on top of standard Cu(In,Ga)Se2 thin films has, to our knowledge, not yet been confirmed by structural methods such as X-ray diffraction, high resolution transmission electron microscopy or electron diffraction. Furthermore, if the surface phase exhibited the weak n-type conductivity of bulk Cu(In,Ga)3Se5, simple charge neutrality estimates (Herberholz et al, 1999) show this would not be sufficient to achieve type inversion. The space-charge width in a CIGS absorber of doping density 3 x 1016cm"3 is approximately 300 run. This would require a charge density of 2 x 1018 cm"3 in a 15 nm thick ODC layer to warrant charge neutrality. This required n-type doping density is considerably more than that usually found in Cu(In,Ga)3Se5 compounds. Based on these arguments, another picture of the surface of Cu(In,Ga)Se2 thin films and of junction formation has emerged. As sketched in Fig. 7.13b, the type inversion can be viewed as resulting from the presence of shallow surface donors. This is the classical Bardeen picture (Bardeen, 1947) of Fermi level pinning by electronic states at semiconductor surfaces. Here, a surface-charge density of a few times 1012 cm"2 eV"1 is sufficient to pin the Fermi level at the neutrality level of free semiconductor surfaces. The positively charged surface donors in Fig. 7.13b are expected to be present in the metal-terminated (112) surface of CuInSe2 because of the dangling bond to the missing Se (Cahen and Noufi, 1989). The type inversion vanishes on air exposure because the surface donors are passivated by the reaction of oxygen with the metal-terminated surface, as discussed in Section 7.3.2 in the context of the Cahen-Noufi model.

Cu(In,Ga)Se2

299

Solar Cells

defect layer

a s

Figure 7.13 Models for the CuInSe2 surface: (a) Segregation of a Cu-poor CuInjSes ODC on the surface; (b) band bending due to surface charges; (c) band bending induces Cu-depletion at the surface, creating a surface defect layer. The valence-band energy is lowered due to Cu depletion. The defect layer provides an internal barrier v of the Cu(In,Ga)Se2 absorber, the CdS buffer layer and the ZnO window. The latter consists of the intrinsic and the highly Al-doped layer. Here, we completely neglect the polycrystalline nature of the semiconductor materials, which in principle requires a two- or threedimensional band diagram. We will restrict ourselves in the following to the implication of the one-dimensional diagram of Fig. 7.18. Even in the one-dimensional model, some details of the band diagram are still not perfectly clear. The diagram in Fig. 7.18 concentrates on the heteroj unction and does not show the contact between the Mo and Cu(In,Ga)Se2 at the back side of the absorber. Another feature under debate but neglected here is a 10-30 nm thick defect layer on top of the Cu(In,Ga)Se2 absorber, already discussed in Section 7.3.3. The energetic quantities describing the band diagram in Fig. 7.18 are the band gap energies £/*, where x = a,b,w for the absorber, buffer and window, respectively. The conduction/valence band offsets between the semiconductors are denoted MJf/v. The built-in or diffusion voltage of the p-type absorber is Vg whereas that of the n-type window/buffer is the sum of the contributions V£b from the buffer V£b and V^, from the window layer. Note that the quantities Vg/n as drawn in Fig. 7.18 are zero-bias quantities and change when an external voltage is applied. The important barriers £ and ®n„ can be calculated from 4>£ = qVb" + qp and Ug hv>Ug , i.e., the photons that can contribute to the short-circuit current of a semiconductor of band-gap energy Ug. For pure polycrystalline CuInSe2 with Ug = 1.04 eV, this value is 46.8 mA cm"2. For Ug= 1.11 eV, the band-gap energy of the best Cu(In,Ga)Se2 solar cell, J'SC = 43.6 mA cm"2. Now we estimate how much absorber material is needed to achieve this photocurrent. The light absorption in a semiconductor is described by the Lambert-Beer law. The irradiance E decays exponentially with depth x into the semiconductor according to E(x) = E0 exp(-ax)

(7.1)

where E0 is the incident irradiance and a the absorption coefficient. For direct semiconductors, a depends on the photon energy hv according to

a{hv) = a±

£_ hv

(7.2)

Cu(ln,Ga)Se2 Solar Cells

309

50 40 30

1 ^8

20 10 0t

J

•

1

1

1

1

20

•

1

1

L

L—l

I

I

l__J

•

•

2.5

Band-gap energy q/eV

Figure 7.19 Short-circuit current density from the AM 1.5 solar spectrum and corresponding to the band -gap energies of various chalcopyrite compounds, of a typical Cu(In,Ga)Se2 alloy (Ug = 1.12 eV), and of the heterojunction partners. The inset shows the losses that occur if less than 1.25/0.5 [im material is available for light absorption.

The absorption coefficient of Cu(In,Ga)Se2 with a low Ga content is reasonably described by eq. 7.2 and a ~ 8 x 104 eV1/2cm_1. By reorganising eq. 7.2 we can calculate the excess energy Mv = /iv-{/„s =1 — a ' a

(7.3)

of photons that have an absorption coefficient larger than a given a. For instance, with absorption lengths La = a~l = 1.255 /jm and 0.5 /zm, we have A/iv = 10 and 62.5 meV. The inset in Fig. 7.19 shows that the losses A/sc corresponding to the photons that are not absorbed within 1.25 or 0.5 jjm of CuInSe2 are 0.9 mAcm - 2 and 3.0 mAcm , respectively. The lower-energy photons are either absorbed at the backmetal/absorber interface or reflected out of the cell. Thus a typical absorber of thickness 1.5 /an absorbs all the light from the solar spectrum except for a negligible remnant corresponding to less than 1 mA cm - .

310

U. Rau and H. W. Schock

Next, we have to recognise that light absorbed in the ZnO window layer does not contribute to the photocurrent. This loss affects absorption and photogeneration for photons of energy > 3.2 eV (the band-gap energy of ZnO). As shown in Fig. 7.19, this loss of high-energy photons costs about 1.3 mA cm"2. In addition, photons in the energy range hv 1.4 eV this loss can be neglected. Another portion of the solar light is absorbed in the buffer layer. If, for instance, a CdS buffer layer caused a sharp cut-off of the spectral response at the band gap of 2.4 eV, only a total of 38 mA cm-2 or 35.5 mA cm""2 would be available for the shortcircuit current of a CuInSe2 or Cu(In,Ga)Se2 (Ug = 1.11 eV) solar cell, respectively. However, measurements of the External Quantum Efficiency (EQE) of a typical ZnO/ CdS/Cu(In,Ga)Se2 heterostructure reveal that the EQE typically drops by a factor of only -0.8 in the wavelength range between the band gap of CdS and that of the ZnO window layer. About 70-80% of the photons in the wavelength range 440-510 nm contribute to i^ because the thin buffer layer does not absorb all photons and about 50% of the electron-hole pairs created in the buffer layer still contribute to the photocurrent (the hole recombination probability at the buffer/absorber interface is relatively low (Engelhardt et al, 1999). Collection loss analysis The most common characterisation method of solar cells other than current-voltage analysis is the measurement of the quantum efficiency. The EQE at a given wavelength X is defined as the number of electron-hole pairs contributing to the photocurrent divided by the number of photons incident on the cell. A quantitative evaluation of the EQE can be used to determine the diffusion length Le if the data are corrected for reflection losses and absorption losses in the window material and if the absorption data of the absorber material are known for the wavelength regime where the absorption length is in the order of Le (Arora et al., 1980). This analysis has been performed in the past by several authors for different types of devices (Klenk and Schock, 1994; Parisi et al, 1998). An alternative way to determine the diffusion length in solar cells is provided by Electron Beam Induced Current (EBIC) measurements. Two approaches are possible: planar EBIC, where the electron beam is scanned over the device surface, and junction EBIC, where the device is cleaved and the beam is scanned along the cross section

Cu(In,Ga)Se2 Solar Cells

311

(Jager-Waldau et al., 1991). For CIGS, the values for Le extracted from EQE and EBIC measurements are -0.5-1.5 /J.m.

7.4.3

Open-circuit voltage

Diode characteristics At open circuit, no current flows across the device and all photogenerated charge carriers have to recombine within the solar cell. The possible recombination paths for the photogenerated charge carriers in the Cu(In,Ga)Se2 absorber are indicated in the band diagram of Fig. 7.20. Here we have considered recombination in the neutral bulk (A) and at the back surface of the absorber (A'), recombination in the space-charge region (B), and recombination at the buffer/absorber interface (C). The dotted lines indicate that the latter two mechanisms may be enhanced by tunnelling in the presence of a high built-in electrical field. Cu(ln,Ga)Sft.

Figure 7.20 Recombination paths in a CdS-Cu(In,Ga)Se2 junction. The paths A and A' represent bulk and back-contact recombination. B and C result from space-charge and interface recombination. The dotted arrows indicate tunnelling.

At the back contact we have drawn the thin MoSe2 layer which forms during the first minutes of absorber deposition. As drawn here, the MoSej has a small conduction-band offset with respect to the Cu(In,Ga)Se2 bulk material and a small Schottky barrier at the Mo back contact. Both features are beneficial for device performance, because the conduction band offset between the Cu(In,Ga)Se2 absorber and the MoSe2 acts as an electronic mirror (the so-called back surface field) for the

U.RauandH. W. Schock

312

photogenerated electrons and diminishes back-surface recombination, and the narrow Schottky barrier provides no substantial resistance for holes between the absorber and the metallic back contact. We emphasise, however, that the details of this band diagram are still under debate. The basic equations for the recombination processes (A-C) can be found in Bube (1992) or Chapters 1 and 2 of this book. All recombination current densities irec for processes A-C can be written in the form of a diode law

W =UexP

PkT t

(7.4)

-1

where V is the applied voltage, /3 the diode quality factor, and kTlq the thermal voltage. The saturation current density i0 is in general a thermally activated quantity and may be written in the form '«. ^ h = 'oo exp kT

(7.5)

where ua is the activation energy and the prefactor j,*, is only weakly temperaturedependent. The quantities «0and /3 depend on the details of each recombination mechanism. Since mechanisms A-C are connected in parallel, the strongest one will dominate the recombination loss. At open circuit, the total recombination current density irec exactly compensates the short-circuit current density iK. Hence we can write the open-circuit voltage in the form

v^SL-BE.^

'L^

(7.6)

where we have assumed that V^ >3/3kT/q, which allows us to consider only the exponential term in eq. 7.4. We have also replaced the activation energy ua by Ua =fiua, which will prove in the following to be the 'true' activation energy of the carrier recombination processes. We shall now discuss the recombination processes A-C in more detail.

313

Cu(In,Ga)Se2 Solar Cells Recombination in the absorber

In the following we shall assume a n -p junction, i.e. that the doping density on the nside is much higher than on the p-side. Shockley's diode equation for such a singlesided junction yields the saturation current density for recombination in the neutral region of the (p-type) absorber. Knowing the square of the intrinsic carrier density n] = NCNV exp(-Ug IkT) we calculate the open-circuit voltage as

U* q

kTjqDeNcN^ q

LN*Le

(7.7)

where De is the diffusion constant for electrons, and Nclv the effective density of states in the conductance/valence band. (For our calculations we have used the values Nc = 6.7 x 1017cm~3 and Nv=l.5x 1019cm"3 resulting from the density-of-states effective masses me =0.09m 0 and mh = 0.71 m0 for electrons and holes, respectively, where m0 is the free electron mass (Neumann, 1986). The quantity NA is the acceptor density, and Le is the diffusion length of the electrons. If this becomes comparable with the thickness d of the quasi-neutral region (QNR) of the absorber, the recombination velocity Sb at the back contact has to be taken into account (recombination path A' in Fig. 7.20), and Lt in eq. 7.7 has to be replaced by cosh sh cosh (z-1) + s i n h ( r l ) where Sb = Sb Le IDe and / = Le Id. Since the width of the space-charge region in thin-film solar cells is comparable with the film thickness, recombination in the space-charge region is important. The Vx -limitation due to recombination in the space-charge region (SCR) of the absorber may be written in a form comparable to eq. 7.7, namely

V.-2L-=II. q

q

kTDenl2^NcNv i E I2

(7.9)

where £m = (2qNAVbm /es) is the electrical field at the position of maximum recombination. The quantity £m depends on the doping density NA, the band bending Vbm, and the dielectric constant es of the absorber. The dependence of eqs. 7.7 and 7.9 on

314

U. Ran and H. W. Schock

the doping density NA is equal in that an increase of NA by one order of magnitude yields an increase of Vx of &VX = (kT In 10) / q ~ 60 raV . However, improving the open-circuit voltage by increasing the doping density is limited by the increased Auger recombination in the QNR and the enhancement of tunnelling in the SCR (Green, 1996a). In eq. 7.9, the activation energy U„ is given by U„ = Aua = 2ua = Ug, whereas in the diode equation for space-charge recombination, the saturation current density is io °= exp (Ug/2kT) and the activation energy is only Ug/2. This demonstrates that we have to correct the activation energies obtained from, for example, Arrhenius plots of the temperature dependence of /„ for the effect of non-ideal diode behaviour in order to obtain the activation energy relevant to VK. Diffusion length L„/ fim

1

05

1

5

' '

' >^1

Sb=10'cmj^£^

.,.'. ...

0.40

0.70 s8

1 0.65

s£^°^S*=\tfcm

y

i 1

Ope

-Cf I

/ /

i

0.60

- 0.45

/

1 i

!

sH

0.50

8

• K.

i:

//

• 0.55

0.55

.... 0.1

.

: i

10

100

0.60

Lifetime t„/ns

Figure 7.21 Correlation of the open-circuit voltage with lifetimes and diffusion lengths for a device with a band gap of 1.12 eV. Solid lines are the results of eq. 7.9 for £,.Q.l fim, eq. 7.7 holds if effective diffusion lengths that take back-surface recombination into account are introduced. The lines with symbols arc the results of a complete device simulation.

In Fig. 7.21 we display the open-circuit voltage limitations given by eqs. 7.7 and 7.9 for a Cu(In,Ga)Se2 solar cell with an absorber layer thickness of 1.5 /jm, a bandgap energy Ug of 1.11 eV and a short-circuit current density /„. of 35.4 mA cm - . The top and the bottom axes, showing the electron diffusion length Le and the lifetime re, are connected by Lr =(z>,T,)"2and a diffusion constant which is here assumed to be De = 2.59 cm2s"'. As the open-circuit voltages in eqs. 7.7 and 7.9 can be shifted by the band-gap energy, we have used the right-hand axis of Fig. 7.21 to display the difference (Ug/q) - V^.

Cu(In,Ga)Se2 Solar Cells

315

For the record Cu(In,Ga)Se2 solar cell (Contreras et al., 1999), this difference is only (1.12- 0.68) V = 0.44 V. The open-circuit voltage of this device requires a lifetime of 30 ns or more, corresponding to a diffusion length of over 2 /an, thus exceeding the absorber thickness. Hence recombination at the back contact also has some influence on V,*—a recombination velocity Sb >105 cm s"1 would hardly allow a VocOfeSOmV. Since the open-circuit voltage of reasonable Cu(In,Ga)Se2 devices (VQC ~ 0.5 V) is just at the threshold between SCR and QNR recombination, we have also conducted some numerical simulations using the software package SCAPS-1D (Niemeegers and Burgelman, 1996). The results for assumed back-surface recombination velocities 5„= 102 cms"1 and 105 cms"1 are also displayed in Fig. 7.21. Here we see that recombination can be well described only outside a transition regime of 1 ns < T„ < 30 ns (0.5 /an < Le < 3 fJm) by the analytical approaches for SCR or QNR recombination. Within this parameter range, the recombination paths A, A' and B contribute to recombination. Note that we have suppressed interface recombination (path C) by setting the recombination velocity for holes at the front contact to Sp = 102 cm s"1 and assuming a hole barrier <X>£ = 1 eV. Effective lifetimes for polycrystalline semiconductors Cu(In,Ga)Se2 solar cells are based on polycrystalline absorbers. Electronic transport in such devices is not completely covered by one-dimensional models. However, quasione-dimensional approaches are possible as long as the influence of grain boundaries on the recombination and charge distribution is not too strong. A first-order approximation is the replacement of the minority-carrier lifetime te by an effective lifetime T^ y , which includes the interface recombination velocity Sg at the grain boundaries. This is given by 1 Tpoly

- U ^ T*

e

(7.10)

where xbe is the minority-carrier lifetime within the grain volume and g denotes the grain size. With the help of eq. 7.10, we can still use eqs. 7.7 and 7.9 if we also use the effective diffusion length Z^Jy for polycrystalline materials, given by .poly

_

'(C°)- 2 + 2S g /(Ds)J" 2

(7.11)

316

U. Rau and H. W. Schock

instead of Leff = L™°"°. For more details, and the limitations of eqs. 7.10 and 7.11, see Green (1996b), Brendel and Rau (1999) and Jensen et al. (2000). Distribution of recombination centres An approach to describing the temperature dependence of current-voltage curves which is useful for Cu(In,Ga)Se2 devices was introduced in Walter et al. (1996b). This approach does not use recombination centres of a single energy within the forbidden gap, but rather a distribution of the form DT(U) = DTOexp(-U I kT*), where the centres are exponentially distributed in energy. The defect density DT(U) has units of cm"3 eV~' and kT* denotes the characteristic energy of the exponential distribution. The characteristic energy U* = kT* is also seen in the defect density spectra obtained from admittance spectroscopy (see Fig. 7.5). A rigorous mathematical treatment for the recombination current density under this assumption is given in Rau et al. (2000). The recombination current density can be written in the form

,exp

(fikTJ exp

fikT

(7.12)

where the pre-exponential term i^ is weakly temperature dependent, and the diode quality factor is given by f

T \

1+ T

(7.13)

The importance of this approach is on the one hand that a defect distribution with a characteristic energy kT* of the order of 100-150 mV is often observed in Cu(In,Ga)Se2 as well as in CuGaSe2. On the other hand, it has been shown by Walter et al. (1996b) and Engelhardt et al. (1998) that the temperature dependence of the current-voltage characteristics of high efficiency Cu(In,Ga)Se2 solar cells in the temperature range 2O0K
(7.14) kT

with t;=UF-Uv in the QNR of the Cu(In,Ga)Se2 absorber. In general, a voltage applied over the heteroj unction does not drop only across the p-type part of the junction. Rather, a change AK in the externally applied bias is shared between the ptype and n-type part according to &V = (l-a)AV£ +CCAVQ where AVg/AVp is the ratio of the diffusion potential in the /?-type and n-type components. Note that the calculation of the voltage share between the two heteroj unction partners in a complicated heterojunction like that shown in Fig. 7.18 is not straightforward. Here we simply use a linear approach with the coefficient a (0 < a < 1). The diode law can be written equivalently by the use of a voltage-dependent barrier pb (V) = £0 +aV , where "

- exp kT

(

qV{\-aj

(7.15)

kT

By comparison of the coefficients we find that the coefficient a is linked to the diode quality factor by f3=(l—a)~l. Finally, we write the open-circuit voltage for interface recombination as

Of

fikT1

•In

(qSpNv

(7.16)

•/sc

where - "b becomes important as shown by experiments (Schmidt, 2000, p. 682) and numerical simulations (Topic et al., 1997).

7.4.5

Electronic metastabilities

The long time relaxation (over hours and days) of the open-circuit voltage of Cu(In,Ga)Se2 based solar cells during illumination is a commonly observed phenomenon (Ruberto and Rothwarf, 1987; Sasala and Sites, 1993). Fortunately, it turns out that in most cases the open-circuit voltage increases with illumination time, a situation which is more favourable than that encountered in a-Si:H (Staebler and Wronski, 1977). A first model for the open-circuit voltage relaxation of Cu(In,Ga)Se2 solar cells was proposed in Ruberto and Rothwarf (1987). This model relies on the reduction of interface recombination at the CdS/Cu(In,Ga)Se2 interface by additional charges introduced into the CdS buffer layer either by illumination under open-circuit conditions or by application of forward bias in the dark. The model is based on the assumption that interface recombination is the dominant recombination mechanism in the solar cells. The increase of positive charges in the buffer layer is assumed to increase the barrier £ and thus reduce interface recombination. However, as we noted above, the open-

fa)

(b)

Figure 7.24 Illustration of persistent changes of (a) the density of free charge carriers in the bulk, and (b) the charge density in the space-charge region.

324

U.RauandH. W. Schock

circuit voltages of the recent high-efficiency devices are limited by recombination in the bulk {i.e., in the SCR) rather than at the interface. Since these devices also show light-soaking effects, another mechanism, possibly additional to that proposed in Ruberto and Rothwarf (1987), must be at work. An important observation is that of persistent photoconductivity in Cu(In,Ga)Se2 thin films (Rau et al., 1998c) and single crystals (Seifert et al., 1997). Meyer et al. (1999) relate the persistent trapping of electrons as the origin of persistent photoconductivity (Fig. 7.24a) to the persistent increase of the charge density in the SCR of the heterojunction, as shown in Fig. 7.24b. This leads to another model for the open circuit voltage relaxation in Cu(In,Ga)Se2 solar cells: the gradual decrease of the electrical field in the SCR leads to a decrease of space-charge recombination, and finally to the increase of the open-circuit voltage during illumination. The band diagrams in Fig. 7.25 schematically compare the model of Ruberto and Rothwarf

(a)

(b)

Figure 7.25 Metastability effects in CIGS-based heteroj unctions, (a) Light-generated excess positive charges are persistently captured in the buffer layer and lead to an increase of the barrier &pb . The full (dashed) lines correspond to the band diagram before (after) illumination; (b) light-generated excess negative charges persistently trapped in the Cu(In,Ga)Se2 absorber layer lead to a decrease of the width of the p-side part of the space-charge region.

Cu(In,Ga)Se2 Solar Cells

325

(1987) with the more recent suggestion of the consequence of persistent photoconductivity in bulk Cu(In,Ga)Se2 (Meyer et al., 1999). In Fig. 7.25, the solid and dashed lines represent the band diagram before and after illumination, respectively. As shown in Fig. 7.25a, an increase of positive charge in the buffer layer increases the barrier 1.3 eV have so far failed. Table 7.3 compares the output parameters of the best chalcopyrite-based solar cells. This compilation clearly shows the superiority of Cu(In,Ga)Se2 with a relatively low Ga content, which leads to the actual world champion device. The fact that the best CuInSe2 devices has an efficiency of 3% below that of the best Cu(In,Ga)Se2 device is due not only to the less favourable band-gap energy but also to the lack of the beneficial effect of small amounts of Ga on film growth, discussed above.

326

U.RauandH.

W. Schock

Table 7.3 Operating parameters of the best Cu(In,Ga)Se2, CuInSe2, CuGaSe2 and CuInS2 cells, and the best pentenary Cu(In,Ga)(S,Se)2 cell

VJmV /a/niA cm"2 FFf%

4/cm 2

Ref.

78.6

0.449

1

41.2

72.6

0.38

2

861

14.2

67.9

0.471

3

13.9*

775

24.3

74.0

0.5

4

11.1"

728

21.24

70.9

0.48

5

Material

tfg/eV f?/%

Cu(In,Ga)Se2

1.12

18.8"

678

35.2

CuInSe2

1.04

15.4*

515

CuGaSe2

1.68

8.3"

Cu(In,Ga)(S,Se)2

1.36

CuInS2

1.57

"Confirmed total area values; 'effective area values (not confirmed). References: 1. Contreras et al. (1999); 2. Stolt et al. (1993); 3. Nadenau et al. (1997); 4. Friedlmeier and Schock (1998); 5. Klaer et al. (1998).

The difficulty of obtaining wide-gap devices with high efficiencies is also illustrated by plotting the absorber band gap of a series of chalcopyrite alloys vs. the attained open-circuit voltages. Figure7.26 shows that below Ug = 1.3 eV, the data follow the straight line Voc = (Ug 14) - 0-5 eV, indicating a proportional gain in V^ with increasing Ug, whereas at Ug> 1.3 eV the gain is much more moderate. At the high band-gap end of the scale, the differences between the band-gap energies and the open-circuit voltages of CuInS2 and CuGaSe2 amount to 840 mV and 820 mV, respectively, whereas (Uglq) - Voc is only 434 eV in the record Cu(In,Ga)Se2 device. 1.2

5

!

'

11

(D D)

jS 0.9 §

4'

1 0.8

•

^

^

'o c 0.7 °

-

-?&• m ? •

0.6 W ,

0.5 1.0 CulnSe 2

i

1.1

1

i

I

1.2

1 ,,„!

1.3

1

1.4

, i ...

1.5

Band-gap energy L/g/eV CulnSz

i

1.6

1.7 CuGaSe 2

Figure 7.26 Open-circuit voltages of different Cu-chalcopyrite based solar cells of different band-gap energies. Full squares correspond to Cu(In,Ga)Se2 alloys, open squares to CuIn(S,Se)2, red circles and black crosses to Cu(In,Ga)(S,Se)2, downward triangles to CulnS2, and upward triangles to CuGaSe2.

327

Cu(In, Ga)Se2 Solar Cells

One reason for the large differences in Ug/q - Vc in wide-gap devices is the less favourable band offset constellation at the absorber/CdS-buffer interface. Figure 7.27 shows the band diagram of a CuGaSe2-based heteroj unction. As the increase of band gap in going from CuInSe2 to CuGaSe2 takes place almost exclusively by increase of the energy of the conduction band, the positive band offset AUf between the absorber and the buffer in Fig. 7.18 turns into a negative one in Fig. 7.27. This implies that the barrier $£ that hinders the holes from the absorber from recombining with the electrons from the buffer does not increase proportionally with increase in the band-gap energy. Thus the importance of interface recombination (dominated by the barrier ,)2 (Friedlmeier and Schock, 1998). Among the materials listed in Table 7.3, the pentenary system is the only one with an open-circuit voltage larger than 750 mV and an efficiency above 13%, outperforming CuInS2 in both these respects. The advantage of Cu(In,Ga)(S,Se)2 could arise from the mutual compensation of the drawback of CuGaSe2 (too high a charge density) and that of (Cu-poor) CuInS2 (too low a conductivity). Electrical analysis of Cu(In,Ga)(S,Se)2 demonstrates that even with a band-gap energy of 1.3 eV and more, this material still preserves the main features of T

•

J

i

I

'

I

•

r

, § •D

2 io 1B 0.0

0.1

L

0.2

0.3

0.4

0.5

Activation energy 14/ eV

Figure 7.29 Defect density spectrum obtained from admittance spectroscopy of a Cu(In,Ga)(S,Se)2based heterojunction solar cell.

Cu(In,Ga)Se2. Figure 7.29 shows that the defect density spectrum of a Cu(In,Ga)(S,Se)2 solar cell is close to what we observe in high-efficiency Cu(In,Ga)Se2 cells. Compared with the high-Ga-content Cu(In,Ga)Se2 device shown in Fig. 7.28, the Cu(In,Ga)(S,Se)2 device displays a relatively moderate value of D™" = 2 x 1016 cnT3eV_1 for the maximum bulk defect density at the activation energy Ua fv 300 meV relevant to recombination. As in standard Cu(In,Ga)Se2, an interfacerelated transition appears at energy Ua ~ 200 meV, indicating the preservation of type inversion at the buffer/absorber interface. Thus, it seems that the overall positive features present in Cu(In,Ga)Se2 with Ug > 1.3 eV can be maintained for larger bandgap energies if one makes use of the full alloy system Cu(In,Ga)(S,Se)2.

Cu(In,Ga)Se2 Solar Cells 7.5.4

331

Graded-gap devices

An interesting property of the CuIni^Ga^S^Se^ alloy system is the possibility of designing graded-gap structures that optimise the electronic properties of the final device (Gray and Lee, 1994; Dhingra and Rothwarf, 1996; Gabor et al., 1996; Dullweber et al., 2000). Such band-gap gradings are achieved during co-evaporation by the control of the elemental sources, but selenisation/sulphurisation processes also lead to beneficial compositional gradings. The art of designing optimum band-gap gradings is to push back charge carriers from critical regions, i.e. regions with high recombination probability within the device. Such critical regions are 1) the interfaces between the back contact and the aborber layer; 2) the heterojunction, including the absorber/buffer interface. Figure 7.30 shows a band diagram of a grading structure that fulfils the requirements for minimising recombination losses. 1. To keep the back contact region clear from electrons, one can use a Ga/In grading. The increase of the Ga/(Ga + In) ratio x causes a movement of the conduction-band minimum upward with respect to its position in pure CuInSe2. An increase of x towards the back surface leads to a gradual increase of the conduction-band energy, as illustrated in Fig. 7.30. The resulting back-surface field, as in the Cu(In,Ga)Se2/MoSe2 heterocontact, drives photogenerated electrons away from the metallic back contact towards the buffer/absorber junction. 2. The minimisation of junction recombination, both at the point of equal capture rates of holes and electrons and at the metallurgical interface between absorber and buffer, requires a larger band gap towards the front contact to the absorber. If one had the choice, one would clearly favour a decrease of the valence-band energy, as shown in Fig. 7.30, over an increase of the conduction-band energy. This favours a grading with the help of S/Se alloying, as at least a part of the increasing band-gap energy is supported by a decrease of valence-bandedge energy. The decreased valence-bandedge energy in Fig. 7.30 leads to an increase of the barrier 30% should be possible using lattice-mismatched materials if the dislocation density can be reduced to

Figure 8.15 l-V curves of a mechanically stacked three-junction cell, consisting of an InGaP/GaAs two-junction cell and an InGaAs bottom cell, under 1 Sun AMI.5G illumination (Takamoto et al., 1997b).

M. Yamaguchi

362

AlGaAs (InGaP)/GaAs/InGaAsP/InGaAs four-junction cell Beaumont et al. (1990) have proposed a mechanically stacked four-junction cell consisting of an AlGaAs/GaAs two-junction cell on a GaAs substrate and an InGaAsP/InGaAs two-junction cell on an InP substrate, and Sharps et al. (1997) a mechanically stacked four-junction cell consisting of an InGaP/GaAs two-junction cell on a GaAs substrate and an InGaAsP/InGaAs two-junction cell, also on an InP substrate. The maximum theoretical efficiencies for a 1.91/1.43/1.05/0.75 eV four-junction cell are 35.1% at 1 Sun AMO, 39.3% at 1 Sun AMI and 44.5% at 500 Suns AMI. Figure 8.16 shows a cross section of a proposed InGaP/GaAs/InGaAsP/InGaAs four-junction solar cell, with a projected efficiency of 34.8% at 1 Sun AMO (Sharps et al, 1997). AR coaling V

I n*-GaAs

n-AI!nP n-GalnP p-GalnP

• contact metal • cap layer window layer emitter layer base layer back surface field

High band-gap tandem

• tunnel junction n-GaAs p-GaAs p-GalnP

window layer emitter layer base layer back surface field ' substrate

p-GaAs

n-lnP n-GalnAsP p-GalnAsP

• metal-bond interconnect window layer emitter layer base layer back surface field tunnel junction

Low band-gap landem

n-lnP n-GalnAs p-GalnAs

- window layer - emitter layer - base layer back surface field

p-lnP p-lnP

Figure 8.16

-substrate -contact metal

Schematic cross section of a proposed InGaP/GaAs/InGaAsP/InGaAs four-junction solar cell

(Sharps era/., 1997).

Super-High Efficiency III-V Tandem and Multifunction Cells

363

8.4 Epitaxial technologies for growing III-V compound cells Figure 8.17 shows chronological improvements in the efficiencies of GaAs solar cells fabricated by the LPE, MOCVD and MBE methods. LPE was used to fabricate GaAs solar cells in 1972 because it produces high quality epitaxial film and has a simple growth system. However, it is not as useful for devices that involve multilayers because of the difficulty of control over layer thickness, doping, composition and speed of 30

& |

20

|

15

|

10

1970

1975

1980

1985

1990

1995

Year Figure 8.17 Chronological improvements in the efficiencies of GaAs solar cells fabricated by the LPE, MOCVD and MBE methods.

throughput. Since 1977, MOCVD has been used to fabricate large-area GaAs solar cells because it is capable of large-scale large-area production and has good reproducibility and controllability. Using large MOCVD systems (for example, AIXTRON AIX-3000 or EMCORE Enterprise 400) which can simultaneously process up to 25 wafers, each of 4 inch diameter, two-junction InGaP/GaAs cells and three-junction InGaP/GaAs/Ge cells are now commercially produced by TECSTAR (Yeh et al., 1996) and Spectrolab (Chiang et al., 1996). In the research stage, InGaP/GaAs two-junction solar cells with efficiencies of 30.3% at 1 Sun AMI.5 and 26.9% at 1 Sun AM0 have been fabricated using the MOCVD method (Takamoto et al., 1997), while an efficiency of 21.1% at 1 Sun AM0 has been reported for MBE-grown InGaP/GaAs two-junction cells (Lammasniemi etal., 1997) and efficiencies of 27.5% at 140 Suns AM1.5 and 24.6% at 100 Suns AM0 have been reported for LPE-grown AlGaAs/GaAs two-junction cells (Andreevefa/., 1997).

364

M. Yamaguchi

Table 8.2 compares the advantages and disadvantages of the various epitaxial technologies. While the LPE method can produce high-quality epitaxial films, the MOCVD method is effective for large-scale large-area production of solar cells. MBE and CBE are advantageous for realising novel multilayer structures such as multijunction solar cells because they provide excellent controllability of monolayer abruptness and thickness due to the nature of beam (Yamaguchi et al., 1994). However, there have been few reports of CBE-grown solar cells. Table 8.2

Advantages and disadvantages of epitaxial technologies

Characteristics

LPE

MOCVD

MBE

CBE

Quality MQW Abrupt interface Heavy doping Large-area Throughput Efficient use of Source materials Equipment cost

**** * * ** * *** ***

*** *** *** #* **** **** **

** **** **** *** ** ** **

*** **** **** **** *** *** ####

#*#

**

**

*

**** Excellent, *** very good, ** fairly good, * bad.

8.5 Monolithic vs. multi-terminal connection modes Figure 8.18 shows various configurations of two-junction cells. For example, in the case of two-junction cells, two cells can be connected to form either two-terminal, three-terminal or four-terminal devices. In a monolithic, two-terminal device, the cells are connected in series with an optically transparent tunnel junction intercell electrical connection. In a two-terminal structure, only one external circuit load is needed, but the photocurrents in the two cells must be equal for optimal operation. Key issues for maximum efficiency monolithic cascade cells (two-terminal multijunction cells series connected with tunnel junction) are the formation of tunnel junctions of high performance and stability for cell interconnection, and growth of optimum band-gap top-and bottom-cell structures on lattice-mismatched substrates, without permitting propagation of deleterious misfit and thermal stress-induced dislocations.

Super-High Efficiency Ill-V Tandem and Multifunction Cells

metal interconnect

Z=E P

365

c ^///////////////W/. o — "I

n

n

-o -

P' n

r 2-terminal cell

3-terminal cell

4-terminal cell

Figure 8.18 Schematic diagrams of various configurations of two-junction cells. In contrast, in three-and four-terminal cells the photocurrents do not have to be equal. However, because it is very difficult to connect three-terminal devices in series, three-terminal tandem cells do not appear to be viable. In the four-terminal case, two separate external circuit loads are used. Since the two individual cells are not coupled, the photocurrents do not have to be the same. Consequently, a much larger selection of band-gap energy combinations is possible, and the changes in photocurrents with changing solar spectral distributions do not pose serious limits. This approach avoids the problem of lattice-mismatched epitaxial growth, current matching and the internal electrical connection of the two-terminal device. Important issues for obtaining high efficiency mechanically stacked cells are the development of multijunction cell fabrication techniques such as thinning the top cell, bonding the bottom cell to the top cell, and cell connections

8.6 Cell interconnection One of the most important factors in making high-efficiency monolithic-cascade type multijunction cells is to achieve optically and electrically low-loss interconnection of two or more cells. There are two main approaches to providing low-resistance intercell ohmic contacts: degenerate doping (tunnel-junction interconnection) and localised metallisation (metal interconnection). The use of a degenerately doped p*/n* tunnel junction is attractive because it only involves one extra step in the growth process. To minimise optical absorption, formation of thin, wide-band-gap tunnel junctions is

366

M.

Yamaguchi

10000

f

0

1

1000

< >. —

100

-

10

-

ID

ti

-o r

A

CBE

A MOCVD

^ v calculated

0

MBE

1

P

*o

0.1

-

0.01

-

&

..Y:

S P c Q.

-1

H

< > \

0.001 0.0001 0.5

i

.

.

.

.

i

. . . . < >

1.5

l/„(eV) Figure 8.19 Changes in tunnel peak current densities of III—V compound tunnel diodes fabricated by various epitaxial growth methods as a function of band-gap energy of lll-V materials (Yamaguchi el al., 1994).

E o

top cell

P"-GaAs /i"-GaAs "•AI,Ga,.,As n-GaAs sub bottom cell

13

500

600 700 800 Annealing temperature (C)

Figure 8.20 Dependence of tunnel peak current density on annealing temperature for double hetero (DH) structure GaAs tunnel diodes; x indicates the Al mole fraction of the sandwichingAl.tGai_rAs layers.

Super-High Efficiency III-V Tandem and Multifunction Cells

367

necessary. However, the formation of a wide-band-gap tunnel junction is very difficult, because the tunnelling current decreases exponentially with increase in band-gap energy, as shown in Fig. 8.19. In addition, impurity diffusion into a highly doped tunnel junction during overgrowth of the top cell increases the resistivity of the tunnel junction and degrades the top cell performance. This was a severe problem in the past, but it has been reduced by the use of lower growth temperatures, as shown in Fig. 8.20, the advent of new dopants including carbon and the introduction of the double-hetero (DH) structure (Sugiura et al., 1988).

Grid Cap

Top metal 0.50fim

n'-GaAs

0.14nm

n'-AlxGa.xAs

0.15um

n • AI0.37G ao.63As

3.80nm

p

O.Uym

-AlosvGaosiAs

p'-AUGai.xAs

jfo/$0$0%&0fr

0.12nm

n*-AlxGai.xAs

&##:••••*"*»*»*> n*-Alo»Gao.iAs

Window _Ernjtter_ Upper ce.i bottom bottom metal metal Lower cell \ \ top metal V /T\ I f~\ S/^^J

Base Dd&e / J

StopetcrT

Slop etch Lower cetl-top-cohlct Window

0.15(im

n-GaAs

Emitter

3.60|un

p

-QaAs

Base

Figure 8.21 Cross section of an AlGaAs/GaAs tandem cell, showing details of the metal interconnecter (MacMillan et al., 1989).

368

M. Yamaguchi

The second approach uses a metallisation scheme to connect the top and bottom cells in the metal-interconnected cascade cell (MacMillan et al., 1989), as shown in Fig. 8.21. The grooves are formed using a sequence of wet chemical etches to remove the epitaxial layers selectively. The metal-interconnect approach has the problem of a complex fabrication process and difficulty in obtaining a low-resistance, reliable contact to the top cell materials. However, this approach may be effective when formation of a tunnel-junction interconnection is difficult. For mechanically stacked structures, adhesive bonding is used to connect the cells. Adhesive materials must be optically transparent over the wide wavelength range 350-1700 nm, have a high thermal conductivity and be mechanically resistant.

8.7 Possible applications of multijunction cells Concentrator operation of two-junction and three-junction cells fabricated on inexpensive substrates such as Ge, Si and polycrystalline materials are being considered as a way to achieve high-efficiency and low-cost cells. In Japan, the super-high efficiency solar cell R&D project including multijunction cells started in fiscal year 1990 (Yamaguchi and Wakamatsu, 1996). The objective of the project is to reach conversion efficiencies of about twice the 1990 values at the laboratory level by the year 2000 and production of such cells for terrestrial applications by 2010. As markets for direct-to-home broadcast, mobile telephone and data communications are growing, commercial satellite power requirements have increased by 200-400% during the early 1990s, and this increasing demand requires continuous efforts to improve solar cell performance and reduce solar cell array cost. In September 1995, the US Air Force Joint Wright Lab./Phillips Lab./NASA Lewis Multijunction Solar Cell Manufacturing Technology (the so-called Man Tech) Program for the development and fabrication of large-area InGaP/GaAs/Ge two-junction and threejunction cells started (Keener et al., 1997). This aims to improve InGaP/GaAs/Ge cell performance (average efficiency > 24-26%) and scale up to production size, quantity and yield while limiting the production cost per watt to not more than 15% over GaAs cells. The average efficiencies of InGaP/GaAs-on-Ge two-junction cells and InGaP/GaAs/Ge three-junction cells made to date were 22.4% and 24.2% at AM0, respectively. OANAMSAT5, the first satellite powered by InGaP/GaAs two-junction cells on Ge substrates, was launched in August 1997 into geosynchronous orbit and is operating nominally with 10 kW of multijunction power (Brown et al, 1997). The average two-junction cell efficiency of this array is 21.6%.

Super-High Efficiency II1-V Tandem and Multifunction Cells

369

8.8 Predictions For super-high-efficiency cells to come into wider use, it will be necessary to improve their conversion efficiency and reduce their cost. In this section, the possibility of obtaining efficiencies of over 40% by using multijunction cell structures and thin-film technologies on inexpensive substrates such as Si and polycrystalline materials is discussed. Figure 8.22 shows the theoretical and realistically expected conversion efficiencies of single-junction and multijunction solar cells reported in the past by some researchers (Fan et a/., 1982; Wanlass et al., 1989; Kurtz et al., 1997) compared with experimentally realised efficiencies. Clearly, concentrator three-junction solar cells have great potential for realising efficiencies of over 40%. 55 50

5 •

>

theoretical values

45

•

0

\ 40 a % 35

r

O

realistically expected values

lf>

$

30

experimental values

lb 1

1.5

2 2.5 3 3.5 Number of junctions

4

4.5

Figure 8.22 Theoretical and realistically expected conversion efficiencies of single-junction and multijunction solar cells reported by Fan et al. (1982), Wanlass et al. (1989) and Kurtz et al. (1997). compared with experimental values.

So far, we have focused mainly on improvement of conversion efficiency. To reduce costs, the use of cheaper substrates is necessary. Figure 8.13 shows the calculated grain size d and dislocation density iVd dependencies of the AM 1.5 conversion efficiency of GaAs single-junction cells, two-junction cells and concentrator two-junction cells compared with experimental values. The calculations were carried out using the following expressions for minority-carrier diffusion length L as a function of d and NA (Yamaguchi and Itoh, 1986): 1/L2 = 1/Lo2 + ASIDd UL2 = l/Lo2 + x3Nd/4

370

M. Yamaguchi

where L is the minority-carrier diffusion length in the solar cell active layers, Lo is the radiative-recombination-limited value of L, S is the surface recombination velocity at the edge of the grain boundary depletion region (assumed to be 5 x 106 cm s_l in the case of GaAs), and D is the minority-carrier diffusion coefficient. It follows that concentrator thin-film multijunction solar cells fabricated on inexpensive substrates such as Si and polycrystalline materials have great potential for realising efficiencies of more than 35% at low cost if one can reduce the dislocation density to less than 5 x 105 cm"2 and increase the grain size to more than 0.1 cm. Cost reduction of III-V compound solar cells is also necessary for their widespread application. To this end, cell fabrication using inexpensive substrates such as Si and Ge, large-scale epitaxial growth equipment and concentrator systems are needed. In addition, an increase in conversion efficiency reduces the cell cost per Wp directly. Figure 8.23 shows an analysis of the energy cost of a 50 MW concentrator system (Whisnant et al., 1994). This suggests that tandem solar cells on Ge substrates under concentrator operation with efficiencies as high as 35% are promising for cost reduction.

T

H

-

-

II GaAs 1-junclion 25%/27.5%/30% 'A

Si point contact 27.4%

!l

i

sc-GaAs sc-Ge pc-Ge

I

GaAs 2-junctio i 30% / 32.5% / 3 5%

sc-GaAs sc-Ge pc-Ge

Si

s c : single-crystalline p c : poly-crystalline

Figure 8.23

An analysis of the energy cos! of a 50 MW concentrator system (Whisnant et al., 1994).

Super-High Efficiency III-V Tandem and Multifunction Cells

371

Acknowledgment This work was partly supported by the New Energy and Industrial Technology Development Organization as part of the New Sunshine Program under the Ministry of International Trade and Industry, Japan.

References Amano C , Sugiura H., Yamamoto A. and Yamaguchi M. (1987), '20.2% efficiency Alo.4Gao.6As/GaAs tandem solar cells grown by molecular beam epitaxy', Appl. Phys. Lett. 51, 1998-2000. Amano C , Sugiura H., Yamaguchi M. and Hane K. (1989), 'Fabrication and numerical analysis of AlGaAs/GaAs tandem solar cells with tunnel interconnections', IEEE Trans. Electron Devices ED-36, 1026-1035. Ando K., Amano C., Sugiura H., Yamaguchi M. and Saletes A. (1987), 'Nonradiative e-h recombination characteristics of mid-gap electron trap in AljGai., As (x = 0.4) grown by molecular beam epitaxy', Jpn. J. Appl. Phys. 26, L266-L269. Andreev V. M., Khvostikov V. P., Rumyantsev V. D., Paleeva V. E. and Shvarts M. Z. (1997), 'Monolithic two-junction AlGaAs/GaAs solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 927-930 Beaumont B., Garabedian P., Nataf G., Guillaume J.-C, Gibart P. and Verie C. (1990), 'Mechanically stacked two-tandem consolar cell concept', Conf. Record 21st. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 47-52. Bedair S. M., Hutchby J. A., Chiang J. P. C , Simons M. and Hauser J. R. (1981), 'AlGaAs/GaAs high efficiency cascade solar cells', Conf. Record 15th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 21-26. Bertness K. A., Kurtz S. R., Friedman D. J., Kibbler A. E., Kramer C. and Olson J. M. (1994), '29.5%-Efficiency GalnP/GaAs tandem solar cells', Appl. Phys. Lett. 65, 989-991. Bowler D. L. and Wolf M. (1980), 'Interactions of efficiency and material requirements for terrestrial silicon solar cells', IEEE Trans. Components, Hybrids Manufacturing Technol., CHMT-3, 464-472. Brown M. R., Goldhammer L. J., Goodelle G. S., Lortz C. U., Perron J. N., Powe J. S. and Schwartz J. A. (1997), 'Characterization testing of dual junction GaInP2/ GaAs/Ge solar cell assemblies', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 805-810.

372

M. Yamaguchi

Chiang P. K., Kurt D. D., Cavicchi B. T., Bertness K. A., Kurtz S. R. and Olson J. M. (1994), 'Large-area GaInP2/GaAs/Ge multijunction solar cells for space applications', Proc. 1st. World Conf. Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 2120-2123. Chiang P. K., Ermer J. H., Nishikawa W. T., Krut D. D., Joslin D. E., Eldredge J. W., Cavicchi B. T. and Olson J. M. (1996), 'Experimental results of GaInP2/GaAs/Ge triple junction cell development for space power systems', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 183— 186. Chung B.-C, Virshup G. F., Hikido S. and Kaminar N. R. (1989), '27.6% efficiency (1 Sun, air mass 1.5) monolithic Alo.37Gao.63 As/GaAs two-junction cascade solar cell with prismatic cover glass', Appl. Phys. Lett. 55, 1741-1743. Chung B.-C, Virshup G. F., Klausmeier-Brown M., Ristow M. L. and Wanlass M. W. (1991), '25.2%-efficiency (1 Sun, air mass 0) AlGaAs/GaAs/InGaAsP threejunction, two-terminal solar cell', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 54-57. Fan J. C C , Tsaur B. Y. and Palm B. J. (1982), 'Optical design of high-efficiency tandem cells', Conf. Record 16th. Photovoltaic Specialists Conf, San Diego, IEEE Press, Piscataway, 692-701. Flores C. (1983), 'A three-terminal double junction GaAs/GaAlAs cascade solar cells', IEEE Electron Device Lett. EDL-4, 96-99. Fraas L. M , McLeod P. S., Cape J. A. and Partain L. D. (1984), 'Monolithic two-color, three-terminal GaAsP/GaAsSb solar cells', Conf. Record 17th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 734-739. Fraas L. M., Avery J. E., Martin J., Sundaram V. S., Girard G., Dinh V. T., Davenport T. M., Yerkes J. W. and O'Neill M. J. (1990), 'Over 35-percent efficient GaAs/GaSb tandem solar cells', IEEE Trans. Electron Devices 37, 443-449. Gale R. P., McClelland R. W., Dingle B. D., Gormley J. V., Burgess R. M., Kim N. P., Mickelsen R. A. and Stanbery B. J. (1990), 'High-efficiency GaAs/CuInSe2 thin-film tandem solar cells', Conf. Record 21st. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 53-57. Gee J. M. and Virshup G. F. (1988), 'A 31%-efficient GaAs/silicon mechanically stacked, multijunction concentrator solar cell', Conf. Record 20th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 754-758. Hutchby J. A., Markunas R. J., Timmons M. L., Chiang P. K. and Bedair S. M. (1985), 'A review of multijunction concentrator solar cells', Conf. Record 18th. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 20-27.

Super-High Efficiency III-V Tandem and Multijunction Cells

373

Jackson E. D. (1955), 'Areas for improving of the semiconductor solar energy converter', Trans. Conf. on the Use of Solar Energy 5, University of Arizona Press, Tucson (1958), 122-126. Keener D. N., Marvin D. C , Brinker D. J., Curtis B. H. and Price P. M. (1997), 'Progress toward technology transition of GaInP2/GaAs/Ge multijunction solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 787-792. Kurtz S. R., Myers D. and Olson J. M. (1997), 'Projected performance of three-and four-junction device using GaAs and GalnP', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 875-878. Lammasniemi J., Kazantsev A. B., Jaakkola R., Toivonen M., Jalonen M., Aho R. and Pessa M. (1997), 'GalnP/GaAs cascade solar cells grown by molecular beam epitaxy', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 823-826. Lamorte M. F. and Abbott D. H. (1980), 'Computer modeling of a two-junction monolithic cascade solar cell', IEEE Trans. Electron Devices ED-25, 831-840. Loferski J. J. (1976), 'Tandem photovoltaic solar cells and increased energy conversion efficiency', Conf. Record 12th. IEEE Photovoltaic Specialists Conf, Baton Rouge, IEEE Press, Piscataway, 957-961. Ludowise M. J., LaRue R. A., Borden P. G., Gregory P. E. and Dietz W. T. (1982), 'High-efficiency organometallic vapor phase epitaxy AlGaAs/GaAs monolithic cascade solar cell using metal interconnects', Appl. Phys. Lett. 41, 550-552. MacMillan H. F., Chung B.-C, Hamaker H. C , Kaminar N. R., Kuryla M. S, Ladle Ristow M., Liu D. D., Partain L. D., Schultz J. C , Virshup G. F. and Werthen J. G. (1989), 'Recent advances in multijunction III-V solar cell development', Solar Cells 27, 205-217. Mitchell K. W. (1981), 'High efficiency concentrator cells', Conf. Record 15th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 142-146. Matsubara H., Tanabe T., Moto A., Mine Y. and Takagishi S. (1998), 'Over 27% efficiency GaAs/InGaAs mechanically stacked solar cells', Solar Energy Mat. Solar Cells 50, 177-184. Nell M. E. and Barnett A. M. (1987), 'The spectral p-n junction model for tandem solar-cell design, IEEE Trans. Electron Devices ED-34, 257-266. Olson J. M., Kurtz S. R. and Kibbler A. E. (1990), 'A 27.3% efficient Gao.5Ino.5P/GaAs tandem solar cell', Appl. Phys. Lett. 56, 623-625.

374

M. Yamaguchi

Sharps P. R., Timmons M. L., Hills J. S. and Gray J. L. (1997), 'Wafer bonding for use in mechanically stacked multi-band-gap cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf., Anaheim, IEEE Press, Piscataway, 895-898. Sugiura H., Amano C , Yamamoto A. and Yamaguchi M. (1988), 'Double heterostructure GaAs tunnel junction for AlGaAs/GaAs tandem solar cells', Jpn. J. Appl. Phys. 27, 269-272. Takahashi K., Yamada S., Unno T. and Kuma S. (1998), 'Characteristics of GaAs solar cells on Ge substrate with a preliminary grown thin layer AlGaAs', Solar Energy Mat. Solar Cells 50, 169-176. Takamoto T., Ikeda E., Kurita H. and Ohmori M. (1997a), 'Over 30% efficient InGaP/GaAs tandem solar cells', Appl. Phys. Lett. 70, 381-383. Takamoto T., Ikeda E., Agui T., Kurita H., Tanabe T., Tanaka S., Matsubara H., Mine Y., Takagishi S. and Yamaguchi M. (1997b), 'InGaP/GaAs and InGaAs mechanically stacked triple-junction solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 1031-1034. Timmons M. L. and Bedair S. M. (1981), 'AlGaAsSb/GaAsSb cascade solar cells', Conf. Record 15th. IEEE Photovoltaic Specialists Conf, Kissimmee, IEEE Press, Piscataway, 1289-1293. Umeno M., Soga T., Baskar K. and Jimbo T. (1998), 'Heteroepitaxial technologies on Si for high-efficiency solar cells', Solar Energy Mat. Solar Cells 50, 203-212. Vernon S. M., Tobin S. P., Wojtczuk S. J., Keavney C. J., Bajgar C , Sanfacon M. M. Daly J. T. and Dixon T. M. (1989), 'III-V solar cell research at Spire Corporation', Solar Cells 27, 107-120. Wanlass M. W., Emery K. A., Gessert T. A., Horner G. S., Osterwald C. R. and Coutts T. J. (1989), 'Practical considerations in tandem cell modeling', Solar Cells 27, 191-204. Wanlass M. W., Coutts T. J., Ward J. S., Emery K. A., Gessert T. A. and Osterwald C. R. (1991), 'Advance high efficiency concentrator tandem solar cells', Conf. Record 22nd. IEEE Photovoltaic Specialists Conf, Las Vegas, IEEE Press, Piscataway, 38^15. Whisnant R. A., Hutchby J. A., Timmons M. I., Venkatasubramanian R. and Hills J. S. (1994), 'Silicon and GaAs/Ge concentrator power plants: a comparison of cost of energy produced', Proc. 1st. World Conf. Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1103-1106. Wolf M. (1960), 'Limitations and possibilities for improvement of photovoltaic solar energy converters', Proc. Inst. Radio Engineers 48, 1246-1263.

Super-High Efficiency III-V Tandem and Multifunction Cells

375

Yamaguchi M. (1991), 'Dislocation density reduction in heteroepitaxial III-V compound film on Si for optical devices', J. Mat. Res. 6, 376-384. Yamaguchi M., Uemura C. and Yamamoto A. (1984), 'Radiation damage in InP single crystals and solar cells', J. Appl. Phys. 55, 1429-1436. Yamaguchi M. and Amano C. (1985), 'Efficiency calculations of thin film GaAs solar cells on Si substrates', /. Appl. Phys. 58, 3601-3606. Yamaguchi M. and Itoh Y. (1986), 'Efficiency considerations for polycrystalline GaAs thin-film solar cells', J. Appl. Phys. 60, 413^117. Yamaguchi M., Warabisako T. and Sugiura H. (1994), 'CBE as a breakthrough technology for PV solar energy applications', J. Crystal Growth 136, 29-36. Yamaguchi M. and Wakamatsu S. (1996), 'Super-high efficiency solar cell R&D program in Japan', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 9-11. Yamaguchi M., Okuda T., Taylor S. J., Takamoto T., Ikeda E. and Kurita H. (1997), 'Superior radiation-resistant properties of InGaP/GaAs tandem solar cells', Appl. Phys. Lett. 70, 1566-1568. Yeh Y. C. M., Chu C. L., Krogen J., Ho F. F., Datum G. C , Billets S., Olson J. M. and Timmons M. L. (1996), 'Production experience with large-area, dual-junction space cells', Conf. Record 25th. IEEE Photovoltaic Specialists Conf., Washington D.C., IEEE Press, Piscataway, 187-190.

CHAPTER 9

ORGANIC PHOTOVOLTAIC DEVICES JONATHAN J. M. HALLS and RICHARD H. FRIEND Cavendish Laboratory, Cambridge, CB3 OHE, UK jjmh I @cam. ac. uk

"I just want to say one word to you —just one word. Plastics. There's a great future in plastics. Think about it. " Mr Maguire to Ben Braddock in The Graduate, 1967.

9.1 Introduction Despite much effort, semiconductor photovoltaic devices made with traditional inorganic semiconductors have remained sufficiently expensive that their uses are confined to a number of niches. Much effort is currently directed towards the use of thin-film semiconductors, in place of silicon wafers, since the direct fabrication of thin devices on substrates offers the prospect of lower manufacturing costs, particularly for larger area applications. The development of amorphous silicon solar cells in 1976 by Wronski and Carlson had the potential of making photovoltaic cells cheaper to produce, and other techniques have been developed to make larger devices possible, including polycrystalline silicon, cadmium telluride, and copper indium diselenide, as described elsewhere in this volume. Despite these advances, the cost of fabricating photovoltaic cells remains prohibitively high for many applications, particularly when large areas are required. One of the factors that keeps system costs relatively high for these technologies is the requirement for high-temperature processing of the semiconductor in a high vacuum environment. This largely restricts fabrication to batch processing onto glass substrates, with associated costs.

9.1.1 Molecular semiconductor devices An alternative approach is the use of organic, molecular semiconductors, which can be processed over large areas at relatively low temperatures, either by vacuum sublimation of molecular materials, or, preferably, by processing from solution of 377

378

J. J. M. Halls and R. II. Friend

film-forming materials such as polymers. If the many issues that we discuss in this chapter concerning the photovoltaic performance and stability can be satisfactorily resolved, then there is the prospect of considerably lower manufacturing costs. The reduction will result in part from the low cost of the small volume of the thin active semiconductor layers, but more importantly, from the lower cost of the other materials used, such as substrates, and the reduced costs that can be realised, for example, by roll-to-roll manufacturing. The challenges in developing organic semiconductors for use in photovoltaic applications arc considerable, requiring new materials, new methods of manufacture, new device architectures and new substrate and encapsulation materials. The most realistic approach is to make use of available know-how that has been developed in related technologies, and it is our view that this is necessary here. Molecular semiconductors are in fact widely used-they are the dominant technology for xerographic copying and laser printing, as we discuss in Section 9.2. More relevant to photovoltaic devices is the development more recently of molecular semiconductor light-emitting diodes (LEDs). These devices are manufactured on a transparent substrate {e.g. glass) as a layer of molecular semiconductor sandwiched between a transparent bottom electrode (e.g. indium tin oxide) and a top metallic electrode. Figure 9.1 shows such a structure.

\ > organic layer Photovoltaic mode

metal

LED mode

Figure 9.1 Schematic diagram of a molecular semiconductor diode which, according to the selection of electrodes and semiconductor layers, can iunction as a light-emitting diode or as a photovoltaic diode. Fabrication is by successive deposition of bottom transparent electrode (e.g. indium tin oxide) onto the transparent substrate (e.g. glass), the semiconductor layer or layers (by vacuum sublimation and/or solution processing) and top metal electode (by vacuum deposition).

Organic semiconductor LEDs, or OLEDs, have advanced very rapidly over the past five years, and now provide a full range of colour, high efficiency (of order 10% quantum efficiency), and, very importantly, have been engineered to give good shelf life and operational lifetime (10,000 hours operation is a minimum requirement for

Organic Photovoltaic Devices

379

most display applications). Work on sublimed molecular devices originating from the Kodak group of Tang in the mid 1980s (Tang and VanSlyke, 1987) has been developed by a number of companies including Pioneer, Japan, who have developed LED displays for automobile audio systems and for mobile electronics. The use of semiconductor polymers for LEDs was developed in Cambridge, UK (Burroughes et al, 1990). The state-of-the-art polymer LEDs are very efficient and are also being commercialised, for example by Philips, Eindhoven (Friend et al, 1999). Solution processing of polymers is particularly attractive for large-area coating, as will be required for photovoltaic devices. Organic transistors networked to form small integrated circuits are being developed by Philips, and are expected to be used in smart cards and electronic barcode labels within the next few years. Organic photodiodes are further from commercial exploitation than these other applications, although there is considerable interest in using these novel semiconductors to fabricate cheaper photovoltaic panels and photodetector arrays. However, using a combination of new materials and novel device structures the efficiency of organic photovoltaic cells continues to rise, and their comparatively low fabrication cost makes these cells increasingly attractive. The general structure of the diode as used for LEDs, shown in Figure 9.1, is directly transferable to operation in a photovoltaic mode (though the electrodes and semiconductor layers need to be correctly designed). There is therefore scope for the direct transfer of know-how from OLEDs to the manufacture of practical, durable and efficient photovoltaic devices. This know-how includes materials synthesis and purification, electrode manufacturing, semiconductor layer deposition and encapsulation.

9.1.2 Photovoltaic properties of molecular semiconductors Molecular materials show semiconducting properties when constructed so that the carbon atoms present in the molecule or polymer chain are bonded as sp2 + pz hybrid orbitals. The pz orbitals form delocalised n and 71* molecular orbitals, which are conventionally recognised as the alternation of carbon-carbon 'single' and 'double' bonds in the molecule. A range of such materials is shown in Figure 9.2. The semiconducting properties of these materials have been very extensively investigated over many decades; we review in more detail the properties relevant to photovoltaic properties in later sections, and refer the reader to a number of monographs (Borsenberger and Weiss, 1993; Greenham and Friend, 1995; Pope and Swenberg, 1999). However, we can summarise the salient characteristics briefly here.

380

J. J. M. Halls and R. H. Friend

molecules metalloporphyrin

C&H13

polyacetylene

MEH-PPV

Figure 9.2 Chemical structures of a range of organic semiconductors. Perylene derivatives are used extensively as electron acceptors and charge-transport layers for xerography. Porphyrins can be made with a range of metal ions (M) at their centres; magnesium, copper and zinc are common choices.

Organic Photovoltaic

381

Devices

Photoabsorption in these materials creates an excited state which is generally confined to a molecule or a region of a polymer chain. This localised excitation is generally termed an 'exciton'. This can be considered either as a neutral excited state of a molecule, or, using a semiconductor description, as an electron-hole pair, bound together by Coulomb and lattice interactions. The electron-hole binding is usually very strong, of order 0.5 eV or above, so that at room temperature {kT = 25 meV), there is little likelihood of electron-hole separation. These materials are therefore commonly strongly luminescent, with emission resulting from radiative decay of the exciton. Electron-hole separation is clearly required for photovoltaic operation, and can be achieved by a number of extrinsic processes. The most important of these is the use of a heterojunction formed between two molecular semiconductors (which can be deposited one on top of the other). The two semiconductors must be chosen so that one can act as electron acceptor and the other as hole acceptor, as is shown schematically in Figure 9.3. Charge separation (often termed photo-induced charge transfer) requires that the offsets in the energies for hole states (n valence band) and for electron states (n* conduction band) at the heterojunction exceeds the binding energy of the electron-hole pair when present on one or other molecular semiconductor (Halls et ai, 1999). This approach has been developed over several years, and is found to be effective both with molecular structure (Tang, 1986), using perylene/phthalocyanine heterojunctions, and with polymer devices (Halls et ai, 1999), using for example MEH-PPV and CN-PPV (see Fig. 9.2).

vacuum level glass ITO interpenetrating polymer network LUMO

acceptor Exciton

donor (a)

acceptor

HOMO

(b)

JHole

Electron

energy acceptor

donor

Figure 9.3 (a) Schematic diagram of photoinduced charge transfer at the interface between two semiconductors with different ionisation potentials and electron affinities, (b) Schematic showing how a mixture of electron- and hole-accepting polymers can be used to provide heterojunctions distributed throughout the polymer composite layer.

382

J. J. M. Halls andR. H. Friend

This double-layer structure separates charge, which can be collected at the two electrodes without difficulty in the strong internal electric field present for these thin devices (Note that optical absorption depths are less than 1 /an for most molecular semiconductors.) The structure functions well, provided that light is absorbed sufficiently close to the heterojunction that the excitons produced can diffuse to the heterojunction in order to ionise. Unfortunately, typical diffusion ranges are an order of magnitude lower than the thickness required to absorb the incident light, so that only about 10% of incident light can be captured in such device arrangements (Halls etal, 1996). Much of the current interest in organic photovoltaic devices is therefore directed to finding new architectures which allow all absorbed light to produce excitons which do reach heterojunctions. One approach is the use of phase-separated polymer blends, which provide a 'distributed heterojunction' throughout the layer thickness (Halls et ai, 1995a; Yu et ai, 1995). Quantum efficiencies up to nearly 30% are achieved in this way. The structure of such a device is shown schematically in Fig. 9.3b.

9.1.3 Overview of this chapter In Section 9.2 the development of organic photovoltaic cells is put into historical context. In Section 9.3 we shall consider why certain organic molecules and semiconductors behave like semiconductors, and look at some of their characteristic properties. Section 9.4 covers the development of simple molecular and polymeric photovoltaic cells based on metal-semiconductor-metal sandwich structures. In Section 9.5 the physics that underlies the charge separation and charge transport properties is discussed, and in Section 9.6 the photocurrent action spectra and currentvoltage characteristics are interpreted in the context of these phenomena. Section 9.7 introduces techniques to improve the performances of these simple calls, beginning with the fabrication of heterojunctions. In Section 9.8 the use of dispersed heterojunctions is introduced; in these the donor and acceptor materials are scrambled together, using, for example, phase separated polymer blends. In Section 9.9 the use of diffuse interface heterojunctions is considered, in which the surface are is increased by intermixing over a limited part of the semiconductor layer, as may be achieved by lamination. In Section 9.10 we look at the technological benefits and drawbacks of these new devices, and speculate on future uses of what promises to be a low-cost avenue to the production of large area photovoltaic cells. Section 9.11 brings the chapter to a close with some general conclusions.

Organic Photovoltaic Devices

383

9.2 Background—early work on photoresponsive organic semiconductors Molecular semiconductors were incorporated into light-sensitive electronic devices at an early stage in their development; indeed, the solid-state photovoltaic effect was first observed in a cell using selenium. Until the middle of this century, molecular, and primarily organic, materials provided the basis of research for the photovoltaic effect. It was not until 1954, when Pearson, Chapin and Fuller invented the silicon photocell at Bell Laboratories, that inorganic materials were destined to become the material of choice for commercial applications. The first report of solid-state photoconductivity is that by Smith (1873) who observed the phenomenon in selenium. The first detailed study of the subject was carried out in the 1920s by Gudden, Pohl and co-workers, on diamond, ZnS and alkali halide single crystals (Borsenberger and Weiss, 1993). The phenomenon was originally interpreted as a radiation-induced structural effect. It was not until the full understanding of the Hall effect that photoconductivity was attributed to the creation of free electrons by the absorption of light. Anthracene was the first molecular material in which photoconductivity was observed, in work by Pochettino (1906) and Volmer (1913). Covalently bonded solids formed the basis for investigations in the 1940s and 1950s, when research into organic materials was limited by the need for single-crystal samples. Interest in organic photoconductors was renewed by the discovery that common artificial pigments and dyes, such as malachite green and methylene blue, had semiconducting properties (Bube, 1960). The photovoltaic phenomenon was observed in cells containing thin films of these organic pigments and dyes in amorphous, crystalline and microcrystalline phases (Merritt, 1982). At the same time it was realised that many biological compounds, and their synthetic analogues, had photoconductive properties. These included carotenes, chlorophylls and other porphyrins, phthalocyanines, cyanines, merocyanines and porphyrins, many of which are important in biological systems. Most of the understanding of the photovoltaic effect in organic photocells comes from the study of devices fabricated from these molecular materials. In recent years semiconducting polymers have been applied to organic photovoltaic cells. Their electronic properties are, in the main, very similar to those of the smaller molecules described above, but their physical properties tend to make them easier to process. Much of the interest in molecular semiconductors was driven by the search for organic photoconductors for xerographic applications in laser printers and photocopiers. In these applications an image is projected onto a statically charged photoconductive drum, and the drum is discharged wherever the drum is exposed. Toner is picked up by the areas of the drum that remain charged, and is subsequently

384

J. J. M. Halls and R. H. Friend

transferred to the paper (Borsenberger and Weiss, 1993). A wide range of organic materials has been investigated for this application, many of which have been exploited in organic photovoltaic cells. Present day xerographic devices almost universally use organic molecular photoconductive materials, dispersed in a polymer binder, rather than more 'traditional' selenium-based photoconductors. In a photoconductive material, charge carriers are created by the optical absorption. An externally applied field is required to produce current by extracting these photogenerated charges from the photoconductor before they recombine. The photovoltaic effect is an extension of photoconductivity in which this field is 'builtin' to the system and exists in the dark. In a photovoltaic cell, this field typically arises from the interaction ofp- and w-type semiconductors (such as in silicon, GaAs, CdTe and CuInSe2), or (less commonly in the case of inorganic semiconductor devices) from the interaction of the photoconductor with the metal. Photovoltaic materials are necessarily photoconductors, but the converse is not always true.

9.3 Conjugated molecules: a new class of semiconductors 9.3.1 Introduction Until recently, carbon-based molecules and polymers have been considered to be insulating materials, and as such have been exploited as electrical insulators in numerous applications. Although it was known from the turn of the twentieth century that certain organic materials were photoconductive, it was arguably the extensive development of molecular electronic materials such as anthracene (Fig. 9.2) by Pope and Swenberg (1999) and the subsequent discovery in 1974 that doped polyacetylene, the simplest conjugated polymer, can exhibit metallic levels of conductivities (Chiang et al, 1977), that initiated an exciting and rapidly expanding field of research into these materials. The novel electronic properties of both molecular and polymeric semiconductors arises from their conjugated chemical structure, and on a molecular level the physical processes behind their properties can be dealt with in the same way. In their undoped state, molecular semiconductors are generally medium to wide bandgap semiconductors. Conjugated polymers have the additional processibility advantages of engineering plastics. Carbon has the electronic structure Is2 2s22p2, and forms hybrid orbitals with its four valence electrons (2s12p2). In conjugated materials, which have alternating double and single bonds in their canonical structures, three sp2 hybrid orbitals form covalent bonds: one with each of the carbon atoms either side of it, and the third with

Organic Photovoltaic

Devices

385

a hydrogen atom or other group. The remaining electron occupies a pz orbital. Collectively, thepz orbitals overlap to create delocalised % bonds which, in the case of a polymer, extend along the full length of the backbone. The most stable configuration of a conjugated polymer is planar, since this maximises the overlap of the pz orbitals (Fig. 9.4), and so these materials tend to be rigid, insoluble and intractable. This originally made it difficult to produce samples in a useful form; fabrication techniques were limited to gas-phase polymerisation and electrochemical growth.

Figure 9.4 Schematic diagram of (a) trans-polyacetylene and (b) poly(p-phenylenevinylene) showing the pz orbitals which overlap to provide the extended delocalised n-system.

The 7i-bonds are weaker than the strong covalent bonds formed by the sp2 electrons, and the electrons in the delocalised 71 system therefore have a smaller binding energy. These electrons dominate the electronic and optical properties, whereas the .^-derived bonds maintain the physical structure of the molecule when electrons are excited from the bonding % orbital to the anti-bonding 71* orbital. The development by Wessling of the sulphonium polyelectrolyte precursor route to poly(p-phenylenevinylene) (PPV) made available thin, high-quality conjugated polymer films (Wessling and Zimmerman, 1968; 1972). The precursor does not have a fully conjugated % system, and is therefore soluble in organic solvents. The final polymer is formed by thermal elimination of the alkyl sulphonium leaving group, which conjugates the links between the benzene rings, as illustrated in Fig. 9.2. Spincoating, a technique commonly used to deposit thin layers of photoresist onto silicon wafers, can therefore be used to deposit uniform thin films of PPV onto planar substrates. By substituting alkyl chains into the benzene rings of PPV, polymers can be synthesised which are soluble in common organic solvents, and can therefore be spin-cast directly. Poly(2-methoxy-5-(2'-ethyl-hexyloxy)-/>phenylenevinylene), MEH-PPV, was one of the first derivatives synthesised for this purpose (Wudl et al, 1991). By using electron withdrawing or donating substituents the electronic energy levels of these materials can be adjusted (Bredas and Heeger, 1994).

386

J. J. M. Halls and R. H. Friend

The physical properties of organic semiconductors depend to a large extent on the morphology of the bulk material, and the differences between small molecules and conjugated polymer-based devices are largely a result of this dependence. Much of the early work on small molecules was carried out on single-crystal samples, in which relatively high electronic mobilities can be obtained. Amorphous samples have a lower mobility, as charges must hop between adjacent molecules, a process that involves an activation energy. Despite this, amorphous films of molecular semiconductors have turned out to be advantageous for applications of organic lightemitting devices. Uniform amorphous films can be produced over large areas by vacuum sublimation. Small molecules have a tendency to crystallise, a process associated with premature device failure in organic molecular displays, and preventing this from occurring has been the focus of much recent research. Crystallisation can create pinholes in the organic films, and grain boundaries along which diffusion of impurities may occur. Bulk samples of conjugated polymers tend to be highly amorphous, although the morphology of a particular material depends critically on its chemical structure, and on the method of synthesis and film preparation. PPV provides a relevant example. Electron diffraction studies of PPV by Granier et al. (1986; 1989) revealed the presence of microcrystallites with a monoclinic unit cell containing two monomer units. Masse et al. (1990) demonstrated that the microcrystallites were on a typical length-scale of 50 A. A number of attempts have been made to increase the degree of molecular order, including stretch-alignment of heated polymer films (Briers et al., 1994) and the use of a precursor polymer consisting of rigid conjugated chain segments separated by flexible spacer groups (Halliday et al, 1993; Pichler et al., 1993). Optical measurements by McBranch et al. (1995) suggest that, in the case of thin spin-cast films made from soluble polymers, the molecular chains lie primarily in the plane of the film.

9.3.2 Electronic properties of conjugated molecules The novel electronic properties of conjugated molecules arise from the overlap of the pz orbitals. The interaction between these orbitals on two adjacent carbon atoms causes their degeneracy to split, and a pair of 7i-type molecular orbitals are formed, as illustrated in Fig. 9.5. In a polymer chain, several electrons contribute to the 7i system, and the bonding and anti-bonding orbitals become broad quasi-continuous energy bands, analogous with the conduction and valence bands of inorganic semiconductors. As the overlap between adjacent pz orbitals and the number of electrons participating

Organic Photovoltaic

387

Devices

in the Ji-system increases, the bands become wider, and the energy gap between them decreases. Thus larger molecules and longer polymer chains tend to have smaller band gaps. In the context of molecular semiconductors we shall define the band gap as the energy difference between the top of the valence band (the highest occupied molecular orbital, or HOMO) and the bottom of the conduction band (the lowest unoccupied molecular orbital, or LUMO). The band gap controls the optoelectronic properties of conjugated materials, and its value is typically in the range 1-4 eV. As a caveat, however, we note that the exact energy of the optical transition may differ from the band gap due to excitonic effects; we will discuss this later. anti-bonding

orbital




f

U

9

_S

f

1

t

valence band

well material

barrier material

nistanrp

Figure 10.2

Band profile of a Type I quantum well.

If the conduction band edge is lower in energy, and the valence band higher in energy, in the well material than the barrier, then electrons and holes are both confined in the well material. This is known as a Type I QW (Fig. 10.2). If only one carrier type is confined in the well, the QW is Type II. Many QWs together form a multi-quantum well (MQW), and if the barriers are thin enough for neighbouring wells to be electronically coupled the structure is known as a superlattice (SL). Only Type I QWs have so far been studied for solar cells, although SLs have been proposed as a means of improving carrier transport in high-resistivity InP solar cells (Varonides and Berger, 1997).

453

Quantum Well Solar Cells 10.3.2 Density of states

The QW forms a quasi-two-dimensional system. Confinement of electrons and holes in the growth (say, z) direction leads to quantisation of the z component of their momentum and kinetic energy. The quantised energy U„ of the nth level is related to the z component of the wavevector kn through

U.

2m

where m* is the effective mass of the carrier in the growth direction. The carriers are confined to a set of subbands of minimum energy Un, but are free to move in the xy plane of the well where the symmetry of the crystal is maintained. Hence a carrier in the nth. subband has total energy

U(k) = Un+^X

(10.1)

2m

\\

where k is the total wavevector, k\\ is the component in the xy plane (such that k2 = k\2 + k„2), and my* is the effective mass of the carrier in this plane. In the envelope function approximation, the shift V(z) in the conduction or valence band edge due to the QW is considered as a perturbation to the periodic crystal potential, and the wavefunctions as crystal eigenfunctions modulated by an 'envelope function'. The confined state energies U„ and envelope functions Fn(z) are solutions to an 'effective mass' equation, which resembles Schrodinger's equation for a onedimensional potential well. They are analogous to the energy levels and wavefunctions of a one-dimensional quantum system. For a QW of width L and depth V, h2 d2'Fn(z) + U(z)Fn(z)-= [/„, 2m lm dz dz"2 where

U(z) = 0, U(z)

= v,

-LI2\LI2\

(10.2)

LI2

This equation holds for both electrons in the conduction band and holes in the valence band, but with different values of m and V. Energies Un are measured up from the bottom of the QW in the conduction band for electrons, and down from the top of

454

J. Nelson

the valence band for holes. The well depth V depends on the composition of the barrier and well materials and on how the difference in band gap is divided between the valence and conduction bands. The effective mass m for each carrier type is in general different for well and barrier. In III-V semiconductors two different types of hole, heavy and light, need to be considered. In the bulk crystal, heavy and light holes are carriers with different effective mass associated with two degenerate crystal bands. For a QW in unstrained material, heavy and light holes occupy the same potential well in the valence band, but with different sets of confined-state energies on account of their different effective masses. In a strained QW, the well depths for heavy and light holes can be different. The number N of confined states contained in the QW for each carrier type is given by (

I—r~

L-J2mV N = int nh

s

i

+1

(10.3)

where int(x) means the integer part of x. N increases with increasing well width and depth, and carrier effective mass. The well is normally narrow enough to admit only a few confined states. At energies U > V the carriers are no longer confined and a continuum of states becomes available, as in the bulk material. These continuum states will not be considered here. In accordance with the Uncertainty Principle, the lowest energy level is always shifted away from the bottom of the well, by an amount that increases with increasing quantum confinement. This means that the ground-state energy, and hence the absorption edge, can be controlled simply by varying the well width. The corresponding envelope functions have well-defined parity and penetrate further into the barrier as energy is increased. Energy levels and envelope functions for a typical AlxGa1_jrAs/GaAs QW are shown in Fig. 10.3. In the SL configuration, neighbouring wells are coupled, extended state envelope functions span the entire SL, and the previously discrete energy levels of the QW broaden into bands. These effects improve carrier transport in the growth direction. The density-of-states function can be constructed from the energy spectrum in the usual way. For a QW of width L the density-of-states per unit volume V is given by D(U)=^d[U-U(k)] = - ^ i > 0 / - £ / „ ) V k nn L „=1

(10.4)

455

Quantum Well Solar Cells

0.2 e2 81

o -0.2 -0.4 |

I-

:;i Well widths

Figure 10.3 Calculated energy levels and envelope functions for a 100 A GaAs QW in AlojGao.vAs. The relative energies of confined states and band gaps are to scale, and the bottom of the conduction band is taken as the zero of energy. Quantum number is measured up from the bottom of the well for electrons, and down from the top of the well for holes.

where 5 is the Dirac delta function and 0 is the Heaviside function. As shown in Fig. 10.4, D{U) has the staircase structure characteristic of quasi-two-dimensional systems. D(U) finite well,

bulk

Figure 10.4 Schematic density-of-states function D(U) for a finite QW, compared with that for an infinitely deep QW and for the well material in the bulk. The first three confined state energies, Ui, U2 and Ui are shown.

This allows us to calculate the concentrations n (of electrons) and p (of holes) in the QW, assuming a local quasi-thermal equilibrium. For electrons with density-ofstates function DC\,(U) in the conduction band,

456

J. Nelson

n = J £>cb([/(k))/rott/(k),7\&)d£/ U

(10.5)

cb

where/FD(U) is the Fermi-Dirac distribution function, Ucb is the conduction band-edge energy, p,e the quasi-Fermi level for electrons in the QW and T the effective electron temperature. When £/cb > fie,nis given by n = niqexp[(jie-Ui

+ en)/lcT]

(10.6)

where niq is the intrinsic carrier density of the QW material in the bulk, Ut is the intrinsic potential energy—the level at which the Fermi level would lie in a perfectly intrinsic material—and 9„ is a measure of the shift in n due to quantum confinement. This is analogous to the expression for n, namely nt, exp[(/2 ? -£/ ( )/W], in a nondegenerate bulk semiconductor. For the remainder of the discussion, we will assume that the QW is described by a quasi-two-dimensional density of states and by a local quasi-Fermi level that is not necessarily continuous with that in the barrier material. electric field

Figure 10.5 Band profile for a QW subject to an electric field S in the growth direction. As the field is increased the right-hand barrier is reduced, increasing the probability of electron escape by thermionic emission or tunnelling.

In operating conditions, QWs placed in the space-charge region of a p-n junction will be subject to a (small) electric field. The field tilts the QW, as shown in Fig. 10.5, distorts the confined-state functions and shifts their energies. The energy of the lowest confined state is reduced. Strictly speaking, in the presence of the field these wavefunctions are no longer confined—carriers penetrate further into the barrier on the side of decreasing potential energy and can tunnel out. For solar cells, the electric field is small enough for the 'flat band' approximation to the band structure to be adequate. However, it is relevant that tunnelling through the barrier is possible.

457

Quantum Well Solar Cells

10.3.3 Photogeneration In a solar cell, photon absorption across the band gap is important. Fermi's golden rule gives the absorption coefficient a in terms of the confined-state energies and overlap integrals. For transitions between a valence-band state \i) of energy £/, and a conduction-band state | / ) of energy Uf , under the influence of an electromagnetic field of angular frequency co and polarisation e, we have (Bastard, 1988) a(U) = - X

\(f\e.p\ifs[uf

-U, -u]

(fm(U,)-

fm{Ufj)

(10.7)

where U is the photon energy ha, p is the momentum operator and A is a sampledependent optical constant. In the usual case where the light is incident normal to the plane of the QW, the matrix element is proportional to the overlap integral Mtm between the valence subband / and conduction band m envelope functions Mlm=jFel(z)Fhm(z)dz

(10.8)

This means that optical transitions are allowed only between subbands of the same parity (/ and m both even or both odd), and are strong only when / = m. In addition, Coulombic bound states (excitons) are formed at an energy just below the minimum for each optically allowed subband-to-subband transition. The excitons appear as strong peaks in the spectrum, even at room temperature, because of their higher binding energy in two-dimensional systems. Including only the principal (Is) exciton and summing eq. 10.7 over initial and final state energies for the Ith. electron - wth. hole subband pair, we have alm(U) = alhlhhMlm [flmS(U -Ulm -Blm) + Q(U -UJ]

(10.9)

where Utm is the electron-hole transition energy before Coulombic effects are included, fi/m and fim are the exciton binding energy and oscillator strength, and the constants QLihm represent the absorption coefficient on the first step edge. In III-V semiconductors, optical transitions occur between both electron-heavy hole (hh) and electron-light hole (Ih) states. The total absorption is the sum of contributions from all such transitions:

a(U) = J X ^ (U) +YJ%K (U) l,m

l,m

(10.10)

458

J. Nelson

where each electron-hole subband pair contributes a step function and a set of excitons to the total absorption spectrum. The absorption coefficient for a typical AltGa,_,As/GaAs QW is shown in Fig. 10.6. GaAs 5

-Alo.aaGao.67As/GaAsMQW -Alo.33Gao.67As

I 1.6

1.7

1.1

Photon energy/eV

Figure 10.6 Calculated absorption coefficient for a 100A Alo.33Gao.67As/GaAs QW compared with the absorption of bulk GaAs and bulk Alo.33Gao.67As. (For the QW, the absorption coefficient is per unit thickness of well material, not including barrier thickness.)

The QW absorption spectrum thus reflects the step-like form of the density-ofstates, modified by strong excitonic peaks. (Because of the strong exciton, the QW spectrum may have a steeper absorption edge than the equivalent bulk alloy, which could be useful for certain PV applications.) The absorption edge or effective band gap Ua is blue-shifted from the absorption edge U,. of the well material in the bulk by the joint confinement energies Uuh of the lowest electron and heavy hole subbands less the corresponding exciton binding energy Bnh.

U„ =U.+U„k-B,lit

(10.11)

The effective band gap U„ is most strongly influenced by QW width and varies from the band gap Ug of the well material for very wide wells, to the band gap Ub of the barrier—or host—material for very narrow wells. This tunability of the absorption edge is one of the most important features of the QWSC. At photon energies above Ub, photogenerated carriers are no longer confined in the QW and the simple quantum mechanical model of absorption becomes unhelpful. In this range the absorption spectrum of the QW begins to resemble that of the bulk material.

Quantum Well Solar Cells

459

10.3.4 Transport and recombination As in any semiconductor device, the electron and hole, once excited, may be transported away from the point of creation, or recombine with each other or with trap states in the band gap. In the steady state, these processes are described by the continuity equations for electrons and holes: 1 di q dz and q dz where r is the volume recombination rate, g the volume generation rate, ie the electron current density and ih the hole current density. The materials parameters that are normally used to quantify these processes in a bulk crystalline semiconductor device—the recombination lifetimes and diffusion constants—are properties of the bulk material and only have meaning in a material many times the thickness of a QW. Level quantisation affects not only the generation term through the QW absorption, discussed above: it also affects the rate of recombination and the mechanism of transport in the direction of the built-in field. We shall discuss these effects next.

10.3.5 Recombination The processes that govern recombination in bulk materials apply to QWs. For III-Vs the most important, in practice, is nonradiative recombination through traps. For a single trap state in the band gap, the Shockley-Read-Hall recombination rate is given by r„=

^ Te(p + p,)+T„(n + nt)

(10.12)

where p„ n, are the equilibrium populations of trap states occupied by holes and electrons, and TP, r„ are the respective carrier trapping times. This formulation should be appropriate to a QW provided that n and p are defined using the quasi-Fermi level of the carriers in the QW (eq. 10.6). The lifetime parameters are properties of the

460

J. Nelson

material and so as a first approximation to a QW we may take the same values as for the well material in the bulk. However, the accumulation of defects at the QW interface may affect the location and density of trap states, and quantum confinement may reduce the trapping times. In the limit of ideal material, radiative recombination is the process that determines solar cell efficiency. The excess radiative recombination in the biased device (i.e. in addition to the recombination that balances thermal generation in equilibrium) then constitutes the dark current. In any volume element 8V the radiative recombination rate rrad depends on the local absorption spectrum a(U) and the local quasi-Fermi level separation A/iF , according to rrad5V = \a(U)j(U,T,^f)dU8V

(10.13)

The emitted flux density j is given by the generalised Planck equation (Wurfel, 1982;Tiedjee/a/., 1984)

j(U,T,AfiF) = 1 ^ ^ - J J j L —

(10.14)

where nr is the local refractive index, h is Planck's constant and c the speed of light.

10.3.6 Transport in the growth direction In a homojunction solar cell, electron and hole currents are normally described by the drift-diffusion equations. The electron current is given by (Sze, 1981; Hovel, 1985) dn 0

(10.27)

The advantage increases with the number of QWs since, while A/sc increases approximately linearly with the number of QWs, the decrease in open-circuit voltage due to AiDk changes only logarithmically. In Figs. 10.13 and 10.14,1-V characteristics are presented for an Alo.3Gao.7As p-in cell with and without 30 GaAs QWs and a GaAs p-i-n cell with and without 10 Irio.i6Gao.84As QWs. In both cases, introducing the QWs has increased /sc and reduced VQC. The latter results from the increased dark current, which is evident from eq. 10.25. In the case of the Alo.3Gao.7As host cell, where the host band gap exceeds the optimum for solar energy conversion, the net effect of QWs is to increase the cell efficiency. This is as expected, since QWs added to a wide-gap host cell reduce its effective band gap towards the optimum. In the case of the GaAs host, the efficiency decreases, which is again the result expected simply from arguments about the optimum band gap for photoconversion: the addition of a lower band-gap material to GaAs will reduce the effective band gap for absorption, and from detailed balance arguments this is expected to reduce the efficiency of the solar cell.

471

Quantum Well Solar Cells

-AIGaAsp-;"-ncell

so 7 £


U^ and, if the quasi-Fermi level separation AfiF in the QW is equal to qV, then eq. 10.26 becomes identical to eq. 10.28 above, and the optimum QWSC will be identical to the optimum single band-gap homojunction cell. There has been some debate about whether the detailed-balance theory applies to the QWSC in practice (Corkish and Honsberg, 1997; Anderson, 1995; Araujo et al., 1994b). Measurements of radiative recombination currents from biased single QW test cells suggest that App is smaller in the QW than in the surrounding host material. Irreversible carrier escape from the QW under the small electric field which is present at the operating point has been suggested as a reason for this (Nelson et al., 1995; Corkish and Honsberg, 1997). It is now of great interest to establish whether the same effect can be observed in the light. If so, then the studies mentioned above showing that VQC in a QWSC is higher than expected for the effective band gap may be explained by a reduced AfiF in the QW, since a smaller Ap,F implies a smaller dark current, and a smaller dark current implies a higher V^. If the apparent reduction in A/JF were carried over to ideal solar cells, then a small improvement in Voc, of perhaps a few per cent, could be expected.

474

J. Nelson

Another interesting idea is the possibility of exploiting 'hot' carrier effects (Ross and Nozik, 1982) in QWSCs. At high carrier densities the relaxation of excited carriers to the band edge can be slowed down by quantum confinement in a QW. The carrier populations then appear to have a higher effective temperature than the lattice, and recombination is reduced. Retarded relaxation has already been observed in QW photoelectrodes (Rosenwaks et al, 1993), and attempts have been made to design hotcarrier superlattice solar cells (Hanna et al., 1997).

10.6 Applications Because QWSCs are as costly to produce as high efficiency III-V home-junction cells, we may expect them to be interesting only in those applications where III-Vs are preferred. At the present time that means space, concentrator and thermophotovoltaic systems. Finally, we mention certain applications where QWSCs are particularly promising.

10.6.1 Tandem cells The efficiency of a monolithic tandem cell is highly sensitive to the combination of band gaps, and to the requirement of current matching between the wide and narrow band-gap components. Compared with wide-gap bulk alloys such as Al^Gaj.^As and InGaP, QW structures in Al^Gai^As/GaAs and InGaP/GaAs offer the advantages of (i) tunability of the band gap through the QW width and (ii) control of the current through the number of QWs. Although the band gap of bulk Al^Ga^As can be adjusted simply by varying the aluminium fraction x, nonradiative recombination increases rapidly with increasing x and degrades collection efficiency. QWSCs offer the alternative possibility of controlling the band gap through the width of the GaAs QWs. Since recombination will occur primarily in the lower band-gap GaAs QWs, where recombination lifetimes are longer than in AljGa^As, it may be possible to design a QWSC of superior practical performance to the A^Ga^As homojunction cell of the same effective band gap (Connolly, 1998).

475

Quantum Well Solar Cells

10.6.2 Concentrator cells In a homojunction cell, efficiency decreases at high levels of light concentration when the increased temperature causes the band gap to shrink and the open-circuit voltage, which is directly related to the band gap, to fall. In a QWSC, although the band gaps of the well and host material still reduce with increasing temperature, the effect on V,*. is less marked. Figure 10.16 compares the temperature dependence of V,* and efficiency for a pair of QWSC and homojunction cells. Although the mechanism is not fully understood, clearly the efficiency of carrier escape from the QWs will increase, or remain at unity, as T is increased. Faster carrier escape is likely to reduce the relative probability of recombination in the QWs, and so offset the effect of the decreasing QW band gap. 8 -I 76-

"r;

1

~^~~—U-^--

5-

4321 0

1

1

1

1

1

1

1

10

20

30

40

50

60

70

1

1

90

1

100 110

Temperature/C

Figure 10.16 Temperature dependence of the efficiency of an InP/In.tGai-tAs QW cell (full line) in comparison with a homogenous InP p-i-n device (squares) and an lnP/In.,Gai-,As heterostructurc device with an InjGai-.tAs /-region (triangles). The measurements were made in a 3000 K blackbody spectrum and scaled by correcting the photocurrent to the standard terrestrial AM 1.5 spectrum using the measured spectral response.

10.6.3

Thermophotovoltaics

In f/iermophotovoltaics (TPV, fully discussed in Chapter 11), low band-gap photovoltaic cells are used to produce electricity from the long-wavelength radiation emitted by a hot (2000-3000 C) source. The source is usually provided through fossil fuel combustion in a combined heat and power system. Often a selective emitter is used to reabsorb the very low energy photons and re-emit them at higher energies to prevent heating. The reshaped spectrum is concentrated around certain bands characteristic of the emitter. For such a spectrum, control of the band gap of the PV

476

J. Nelson

cell is essential for good power conversion efficiency. The flexibility of band gap makes QWSCs of great interest for TPV. It is also possible that Auger recombination, a longstanding problem in low band-gap solar cells, is suppressed in the QW device. QWSCs in InGaAsP/In/jai-jAs have already been shown to produce a higher V^ than the comparable In/ja^As homojunction cell (Griffin et ah, 1997).

10.7 Conclusions We have reviewed the use of novel quantum-well semiconductor heterostructures in solar cells. QW structures are of interest as a means of enhancing the photocurrent and efficiency of crystalline solar cells. Photocurrent enhancement has been demonstrated in a range of materials and is well understood. Efficiency enhancement has been observed in materials whose band gap is larger than the optimum for solar energy conversion. In materials of band gap close to the optimum, experimental tests on QW cells of equivalent quality to homojunction cells have not yet been possible. Nevertheless there is some evidence that the effect of QWs in increasing recombination within the device is smaller than expected from arguments based on a quasi-thermal equilibrium distribution of carriers. If this is true under operating conditions, then higher efficiencies may also be available with optimum band-gap cells. QW structures have the advantages over homojunction cells that the effective band gap can be controlled by tuning the width of the QW, rather than by varying the composition of a bulk alloy. This means that QWs may produce better cells of better material quality than bulk alloys when particular band gaps are required. This is relevant for high-efficiency tandem cells and for thermophotovoltaic cells, and QW structures are being researched for both these applications. A further important advantage is that QW structures have a better response to temperature and consequently are expected to perform better under concentrated light. Some of the major challenges that remain are: to find and verify a theoretical explanation for the observed dark currents and V^ behaviour; to establish whether the suppressed recombination behaviour observed in the dark occurs under solar cell operating conditions; and to prepare an optimum band-gap QWSC of equivalent quality and superior efficiency to a GaAs homojunction solar cell. More generally, work on QW structures has stimulated a range of new ideas about the role of quantum nanostructures in photovoltaics and the limits to efficiency of solar cells.

Quantum Well Solar Cells

477

References Anderson N. G. (1995), 'Ideal theory of quantum-well solar cells', J. Appl. Phys. 78, 1850-1861. Araujo G. L. and Marti A. (1994), 'Absolute limiting efficiencies for photovoltaic energy conversion', Solar Energy Mater. Solar Cells 33, 213-240. Araujo G. L., Marti A., Ragay F. W. and Wolter J. H. (1994), 'Efficiency of multiple quantum well solar cells', Proc. 12th. European Photovoltaic Solar Energy Conf, Amsterdam, H. S. Stephens & Associates, Bedford, 1481-1484. Barnes J. M. (1994), 'An experimental and theoretical study of GaAs/InGaAs quantum well solar cells and carrier escape from quantum wells', Ph.D. Thesis, University of London. Barnes J. M., Nelson J., Barnham K. W. J., Roberts J. S., Pate M. A., Grey R., Dosanjh S. S., Mazzer M. and Ghiraldo F. (1996), 'Characterization of GaAs/ InGaAs quantum wells using photocurrent spectroscopy', J. Appl. Phys. 79, 77757777. Barnham K. W. J. and Duggan G. (1990), 'A new approach to high-efficiency multiband-gap solar-cells', J. Appl. Phys. 67, 3490-3493. Barnham K. W. J., Braun B., Nelson J., Paxman M., Button C , Roberts J. S. and Foxon C. T. (1991) 'Short-circuit current and energy efficiency enhancement in a low-dimensional structure photovoltaic device', Appl. Phys. Lett. 59, 135-137. Barnham K., Connolly J., Griffin P., Haarpaintner G., Nelson J., Tsui E., Zachariou A., Osborne J., Button C , Hill G., Hopkinson M„ Pate M., Roberts J. and Foxon T. (1996), 'Voltage enhancement in quantum well solar cells', J. Appl. Phys. 80, 1201-1206. Barnham K., Ballard I., Barnes J., Connolly J., Griffin P., Kluftinger B., Nelson J., Tsui E. and Zachariou A. (1997), 'Quantum well solar cells', Appl. Surf. Sci. 113/114, 722-733. Bastard, G. (1988), Wave Mechanics Applied to Semiconductor Heterostructures, Editions de Physique, Les Ulis. Connolly J. P., Barnham K. W. J., Nelson J., Griffin P., Haarpaintner G., Roberts C , Pate M. and Roberts J. S. (1995), 'Optimisation of high efficiency Al^Gai_xAs MQW solar cells', Proc. Int. Solar Energy Society 1995 Solar World Congress, Harare, Zimbabwe. Connolly J. P. (1998), private communication. Corkish R. and Green M. (1993), 'Recombination of carriers in quantum-well solarcells', Conf. Record 23rd. IEEE Photovoltaic Specialists Conf, Louisville, IEEE Press, Piscataway, 675-680.

478

J. Nelson

Corkish R. and Honsberg C. B. (1997), 'Dark currents in double-heterostructure and quantum-well solar cells', Conf. Record 26th. IEEE Photovoltaic Specialists Conf, Anaheim, IEEE Press, Piscataway, 923-926. Ekins-Daukes N. J. (1998), private communication. Freundlich A., Rossignol V., Vilela M. F. and Renaud P. (1994), 'InP-based quantum well solar cells grown by chemical beam epitaxy', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1886-1889. Greenham N. C , Peng X. G. and Alivisatos A. P. (1997), 'A CdSe nanocrystal MEHPPV polymer composite photovoltaic', in Future Generation Photovoltaic Technologies—First NREL Conf., McConnell R. D„ ed., American Institute of Physics, New York, pp. 295-301. Griffin P., Ballard I., Barnham K., Nelson J. and Zachariou A. (1997), 'Advantages of quantum well solar cells for TPV, Thermophotovoltaic Generation of Electricity, Coutts T. J., Allman C. S. and Benner J. P., eds., American Institute of Physics, New York, pp. 411^22. Griffin P., Barnes J., Barnham K. W. J., Haarpaintner G., Mazzer M., ZanottiFregonara C , Grunbaum E., Olson C., Rohr C , David J. P. R., Roberts J. S., Grey R. and Pate M. A. (1996), 'Effect of strain relaxation on forward bias dark currents in GaAs/InGaAs multiquantum well p-i-n diodes', J. Appl. Phys. 80, 5815-5820. Hanna M. C , Lu Z. H. and Nozik A. J. (1997), 'Hot carrier solar cells', in Future Generation Photovoltaic Technologies—First NREL Conf, McConnell R. D., ed., American Institute of Physics, New York, pp. 309-316. Hovel H. J. (1975), Semiconductor and Semimetals, Volume 11—Solar Cells, Willardson R. K. and Beer A. C , eds., Academic Press, London. Kitatani T., Yazawa Y., Minemura J. and Tamura K. (1995), 'Vertical transportproperties of photogenerated carrier in InGaAs/GaAs strained multiple-quantum wells', Jpn. J. Appl. Phys. 34, 1358-1361. Mazzer M. (1997), private communication. Meyer M. and Metzger R.A. (1996), Compound Semiconductor, November/ December 1996, p. 22. Nelson J., Paxman M., Barnham K. W. J., Roberts J. S. and Button C. (1993), 'Steady state carrier escape from single quantum wells', IEEE J. Quantum Electron. 29, 1460-1467. Nelson J., Barnham K., Connolly J. and Haarpaintner G. (1994), 'Quantum well solar cell dark currents—a theoretical approach', Proc. 12th. European Photovoltaic Solar Energy Conf, Amsterdam, H. S. Stephens & Associates, Bedford, 13701373.

Quantum Well Solar Cells

479

Nelson J. (1995), 'Multiple quantum well structures for photovoltaic energy conversion', Physics of Thin Films 21, Francombe M. H. and Vossen J. L., eds., pp. 311-368. Nelson J., Kluftinger B., Tsui E. and Barnham K. (1995), 'Quasi-Fermi level separation in quantum well solar cells', Proc. 13th. European Photovoltaic Solar Energy Conf., Nice, H. S. Stephens & Associates, Bedford, 150-153. Nelson J., Barnes J., Ekins-Daukes N., Kluftinger B., Tsui E., Barnham K., Foxon C. T., Cheng T. and Roberts J. S. (1997), 'Observation of suppressed radiative recombination in single quantum well p-i-n photodiodes', J. Appl. Phys. 82, 62406246. Nelson J., Barnes J., Ekins-Daukes N., Barnham K. W. J., Kluftinger B., Tsui E. SM., Foxon C. T., Cheng T. S. and Roberts J. S. (1998), 'Reduced radiative currents from GaAs/InGaAs and AlGaAs/GaAs p-i-n quantum well devices', Conf. Record 24th. IEEE Int. Symposium on Compound Semiconductors, IEEE Press, Piscataway, 413-416. Nelson J., Barnham K., Ballard I., Connolly J. P., Roberts J. S. and Pate M. (1999), 'Effect of QW location on quantum well photodiode dark currents', J. Appl. Phys. 86,5898-5905. Paxman M., Nelson J., Barnham K. W. J., Braun B., Connolly J. P., Button C , Roberts J. S. and Foxon C.T. (1993), 'Modelling the spectral response of the quantum well solar cell', J. Appl. Phys. 74, 614-621. Pearsall T. P. (1989), 'Optical properties of Ge-Si alloys and superlattices', J. Luminescence 44, 367-380. Ragay F. W., Wolter J. H., Marti A. and Araujo G. L. (1994), 'Experimental analysis of the efficiency of multiple quantum well solar cells', Proc. 12th. European Photovoltaic Solar Energy Conf, Amsterdam, H. S. Stephens & Associates, Bedford, 1429-1433. Renaud P., Vilela M. F., Freundlich A., Bensaoula A. and Medelci N. (1994), 'Modeling p-/(multi quantum well)-n solar cells: a contribution for a near optimum design', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, Waikoloa, IEEE Press, Piscataway, 1787-1790. Rosenwaks Y., Hanna M. C , Levi D. H., Szmyd D. M., Ahrenkiel R. K. and Nozik A. J. (1993), 'Hot-carrier cooling in GaAs—quantum-wells versus bulk', Phys. Rev. B. 48, 14675-14678. Ross R. T. and Nozik A. J. (1982), 'Efficiency of hot-carrier solar-energy converters', J. Appl. Phys. 53, 3813-3818.

480

J. Nelson

Sah C.-T., Noyce R. N. and Shockley W. (1957), 'Carrier generation and recombination in p-n junctions and p-n junction characteristics', Proc. Inst. Radio Engineers. 45, 1228-1243. Scott C. G., Sands D., Yousaf M., Abolhassani N., Ashenford D.E., Aperathitis E., Hatzopoulos Z. and Panayotatos P. (1997), 'P-i-n solar cell efficiency enhancement by use of MQW structures in the /-layer', Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 24992502. Sze S. M. (1981), Physics of Semiconductor Devices, Wiley, New York, 790-838. Tiedje T., Yablonovitch E., Cody G. D. and Brooks B. G. (1984), 'Limiting efficiency of silicon solar-cells', IEEE Trans. Electron Devices 31, 711-716. Varonides A. C. and Berger A. W. (1997), Proc. 14th. European Photovoltaic Solar Energy Conf., Barcelona, H. S. Stephens & Associates, Bedford, 1712-1715. Venkatasubramanian R. Timmons M. L., Sharps P. R., Hutchby J. A., Beck E. and Emery K. (1994), 'Material and device characterization toward high-efficiency GaAs solar-cells on optical-grade polycrystalline Ge substrates', Proc. 1st. World Conf. on Photovoltaic Energy Conversion, IEEE Press, Piscataway, 1692-1696. Weisbuch C. and Vinter B. (1991), Quantum Semiconductor Structures, Academic Press, San Diego. Vogel R., Hoyer P. and Weller H. (1994), 'Quantum-sized PbS, CdS, Ag2S, Sb2S3, and Bi2S3 particles as sensitisers for various nanoporous wide-band-gap semiconductors', J. Phys. Chem. 98, 3183-3188. Wurfel P. (1982), 'The chemical potential of radiation', J. Phys. C15, 3967-3985. Zachariou A., Barnham K. W. J., Griffin P., Nelson J., Button C , Hopkinson M., Pate M. and Epler J. (1996), 'A new approach to p-doping and the observation of efficiency enhancement in InP/InGaAs quantum well solar cells', Conf. Record 25th. IEEE Photovoltaic Specialists Conf, Washington D.C., IEEE Press, Piscataway, 113-117. Zachariou A., Barnes J., Barnham K. W. J., Nelson J., Tsui E. S.-M., Epler J. and Pate M. (1998), 'A carrier escape study from InP/InGaAs single quantum well solar cells', J. Appl. Phys. 83, 877-881. Zory P. S. (1993, ed.), Quantum Well Lasers, Academic Press, London.

CHAPTER 11

THERMOPHOTOVOLTAIC GENERATION OF ELECTRICITY T. J. COUTTS National Renewable Energy Laboratory Golden, Colorado 80401 [email protected]

Our energy is in proportion to the resistance met. We can attempt nothing great, but from a sense of the difficulties we have to encounter. William Hazlitt, Characteristics, 1823.

11.1 Introduction Thermophotovoltaic (TPV) generation of electricity has recently re-emerged after many years of stagnation. The subject was vigorously investigated in the 1960s, by Eisenman et al. (1963), by Guazzoni et al. (1968), and by Kittl (1966), and up to the early 1980s but languished for a period of about ten years because of lack of funding, at least partially due to the absence of high-performance semiconductor converters. In the view of the author, the re-emergence of the topic is largely due to advances made in technologies based on the III-V family of converters made by Wanlass et al. (1994), Bertness et al. (1994), and others. Strictly, TPV generation of electricity ought not to be included in this volume, because it is questionable whether it will ever be a solar-based technology. Indeed, in some respects, it could be regarded as merely another means of generating electricity using conventional fossil fuels. However, the Sun could be used as the source of power, as discussed by Stone et al. (1994), Guazzoni and Pizzo (1996) and Stone et al. (1995). In fact, the versatility of the fuel source is one of the main attractions of the technology. Equally, in the spirit of reducing the impact of human activities on global climate change, TPV could be driven by industrial waste heat. This is very much a speculative idea at present, but it is being actively investigated by several groups in the United States. The glass industry is a particularly good example.

481

482

T. J. Coutts

Eisenman et al. (1963) indicated that the manufacture of float-glass in the USA has a total power requirement of 1.5 GW (electrical plus thermal)1. Industry reports indicate that two-thirds of this is wasted in one form or another. Given that the temperatures involved in radiation from the melt process (to select just one aspect of float-glass manufacture) are extremely high, it is not difficult to appreciate that the waste of energy is probably considerable. It is estimated that about 16% of the total "wasted" energy is lost by radiation from the high-temperature surface of the glass melt region of the production line. Some of this heat may already be recovered by conventional methods, but there would appear to be further opportunities. With sufficient financial incentive, it seems probable that the engineering problems, sure to be encountered in a major application such as this, would be surmountable and that the potential for additional recovery of energy is considerable. We hope that the inclusion of TPV in this volume is justifiable on the basis of waste heat recovery! As mentioned above, the Sun could be used to heat radiant surfaces that could be radiation sources for TPV conversion, but it would certainly be more efficient to use photovoltaic cells designed to utilise concentrated sunlight. By going through an intermediate step, the efficiency is inevitably degraded. Despite this, solar-driven TPV has been considered very seriously, as discussed by Demichelis and Minetti-Mezzetti (1979/80) and by Stone et al. (1995). Other sources of fuel include propane, which has been investigated by Fraas et al. (1995), diesel, by DeBellis et al. (1997), natural gas, by Pelka et al. (1986), and nuclear radiation by Schock et al. (1997). Applications are already expected to be diverse, with many in the military arena and they have been reviewed by Rose (1996) and Rosenfeld (1994). Most of the R&D funding to date has derived from the military sources. Equally many applications could emerge in the non-military field. These could include sailing boats, recreational vehicles, stand-alone gas furnaces, see for example, Krist (1994), remote homes, community co-generation of heat and electricity, considered by Broman and Marks (1994) and many others not yet implemented, but mentioned by Coutts and Fitzgerald (1998), and Johnson (1996). Once the potential of TPV becomes more widely appreciated, it seems probable that many applications will emerge that are as yet undreamed. In many potential applications the attractions could include high power density, quietness, low pollution, low maintenance, fuel versatility, light weight, and reliability. Some of these attractions, however, may ultimately prove to be wishful thinking! ' Much of the thermal waste is already recovered, including some of the radiative losses. However, our calculations suggest that there is great potential for significant increases in the magnitude of recovered energy.

Thermophotovoltaic Generation of Electricity

483

TPV generation of electricity is based on precisely the same physical principles as photovoltaic (PV) generation. These principles have been fully discussed elsewhere in this volume, but, in brief, photons are absorbed by a semiconductor p-n junction and generate excess electrons and holes in both sides of the electrostatic junction. However, only the minority carriers are important to the operation of PV-based devices. After their creation, the minority holes in the n-type region drift to the p-type side of the junction under the influence of the built-in electric field. The minority electrons in the p-type region drift to the /i-type side. In the absence of external connections, the drift current is exactly offset by a reverse current caused by the carrier concentration gradient (and many other possible mechanisms). These two effects lead to the open-circuit voltage (V Combustion system

-o Radiator

Sub- banc gap pr otor s

DC electricity

Above-bandgap photons

Photons

Hea

i

Converter shield/Filter

—>

i7 TPV converter

"

—>

Power conditioning

V Electrical power output

Figure 11.1 Schematic of a TPV system with the key components shown.

In designing a TPV generator, attention must be paid to several individual component parts of the system. These include the combustion system, the radiator, the means of optical control (discussed in detail later), the converter, and the recuperator of waste heat. A schematic of generic TPV systems is shown in Fig. 11.1, showing the individual processes from combustion of fuel to power conditioning. An alternative diagram of the linear arrangement of the components is shown in Fig. 11.2. In this figure, the thermal management and power conditioning are also indicated. In the case of the broadband spectrum in Fig. 11.2a, the emissivity is taken as unity and the radiator temperature as 1500 K. The spectral emittance of the selective radiator is taken from a paper by Lowe et al. (1994) for mixed rare earths. Notice that the emittance in Fig. 11.2 is in absolute units whereas it is relative in Fig. 11.2b. To compare these, the spectral emittance data should be convoluted with that from an appropriate blackbody spectrum. Ideally, the components in a TPV system should be optimised collectively, although this has seldom been done. To maximise the efficiency2 of the system, a selective radiator is likely to be used, but to maximise the electrical power output a broad-band system would probably be used.

" Throughout this chapter, we shall use the term efficiency in several different ways. The efficiency of a TPV cell is equal to the electrical power out divided by the optical power absorbed. We do not express it in terms of the 'optical power incident' because it is generally considered that an optical control element will be used to return unusable photons to the radiator.

486

T. J. Coutts

Ultimately, development of high-performance systems will involve teams of engineers with collective experience in all required areas. Generally, individual researchers and groups focus on that component with which they are most familiar. To date, this has mainly been the radiator, the optical control element and the converter. This chapter will discuss relatively recent work on each of these components, with an emphasis on the sub-bandgap photon reflection and the semiconductor converter, as well as briefly reviewing the projected and actual performance of some systems.

Black-body radiator at 1500 K

40% Er: 1.5% Ho-YAG selective radiator

Re-circulated subbandgap photons

Re-circulated subbandgap photons

Low-bandgap converter

Spectrally-matched converter

Sub-bandgap photon reflector

Sub-bandgap photon reflector

0.5

1

1.5

2

Wavelenath/um

(a)

2.5

! 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2 Wavelenath/um

(b)

Figure 11.2 Schematic diagrams of generic TPV systems showing (a) a broadband and (b) a narrow-band radiator-based system. The choice of band gaps is strongly influenced by the type of radiator used. The ideal selective radiator would not radiate outside the characteristic emission band and would not require a component to reflect the sub-bandgap photons.

Thermophotovoltaic Generation of Electricity

487

11.2 Radiators The radiator in a TPV system is heated by a fuel such as those mentioned earlier. Its temperature governs the radiated power density, as defined by Planck's law. While the radiation from a flame may be somewhat erratic in intensity, because of fluctuations in temperature etc., the hot surface will radiate relatively uniformly. This is critical so that the intensity received by the semiconductor converter is also uniform and constant with time. Figure 11.2 shows that the radiator lies between the source of heat and the converter. All surfaces at a temperature above absolute zero radiate energy. The spectral dependence of the radiated power density (measured in W cm"2 /rnf') is given by Planck's law, which includes the absolute temperature and the emissivity of the surface. Blackbody and greybody radiators have emissivities that are independent of wavelength across the entire spectrum. The former has an emissivity of unity whereas the latter has a constant emissivity of less than unity. The welldefined intensity makes it relatively straightforward to calculate the optimum band gap of the semiconductor. The broadband radiator approach requires an optimum band gap in the range 0.5-0.7 eV, as will be shown later. Certain materials, such as the rare earth oxides, radiate in relatively narrow bands of wavelengths and, for these, the band gap of the semiconductor is chosen to match the emission 'band of the radiator.

11.2.1 Broadband radiators The selection of optimum band gap for a broadband irradiance, depends on optimising the product of short-circuit density and open-circuit voltage. At lower band gaps, the former increases, whereas the latter increases at higher band gaps. When the determination has been made, it is found that many of the incident photons have below-band-gap energies and are, therefore, not useful to the converter. Based on this reasoning, Fig. 11.3 shows the percentage of convertible flux as a function of converter band gap, with the radiator temperature being treated parametrically. Clearly, the percentage increases as the band gap decreases and the radiator temperature increases. However, Fig. 11.3 also shows that silicon, with a band gap of about 1.14 eV, combined with a radiator at 2000 K, can only convert about 15% of the incident flux. Even though silicon cells are supposedly low-cost, this makes the point that their performance could only ever be modest, when used with a broadband radiator.

488

T. J. Coutts

Band gap of converter (eV)

Figure 11.3 Percentage of convertible photon power from blackbody radiators at the temperatures shown as a function of the band gap of the converter.

Practical broadband radiators Silicon carbide is useful as a broadband radiator because it has an emissivity of about 0.9-0.95 that is essentially independent of wavelength, as shown by the data of Pernisz and Sana (1994). It does not melt until a temperature of above 2000 K, which is probably above the practical temperatures envisaged for radiators.

11.2.2 Selective radiators As mentioned above, only a small percentage of flux radiated from a blackbody is convertible by a silicon cell. There are several incentives for developing selective radiators with relatively narrow emission bands. In the broadband spectrum, there is a large proportion of flux at sub-bandgap wavelengths. The long-wavelength photons are not usefully absorbed by the semiconductor, and they do not contribute to the electrical output of the device. In fact, they may be absorbed by free carriers already in the conduction or valence band and cause heating of the device, thereby reducing device performance. Likewise, the short-wavelength photons exceed the band-gap energy, which also causes heating of the device because of thermalisation of hot carriers. In principle, selective radiators should eliminate both of these deficiencies. Principles of selective radiators The chemical and physical properties of elements depend on their outermost (valence) electrons. If all electron sub-shells are

Thermophotovoltaic Generation of Electricity

489

completely filled, then the element is one of the inert gases. If there is only one selectron in the outer orbital and no other partially filled sub-shells, then the element is, in general, metallic. For most elements with lower atomic numbers, the filling sequence of the electron sub-shells is relatively simple but, for the heavier elements, there are options that offer lower energy configurations than the apparently simplest sequence, which is important to the rare earth series of elements. The lanthanide series is usually defined as the elements from cerium (58) to lutetium (71). The characteristic valence of the elements is three throughout the series because the 6s2, the 5d\ and either one or two of the 4/electrons are the outermost shells not already completely filled. Thus, Ce3+, with one 4/electron, has the same valence as Lu3+, with fourteen 4/ electrons. With lanthanide ions, the valence electrons are no longer present, and the optical properties of the compounds are dictated by the 4/ electrons via/-/transitions. The 4 / electrons lie within the orbit of the filled outer sub-shells of 5s and 5p6 electrons of the [Xe] core. These outer electrons screen the inner 4/ electrons electrically. The screening prevents the 4/ electrons from interacting with other ions in the solid, which prevents the formation of energy bands. Hence, even lanthanide ions in solid matrices that have formed bonds to oxygen, for example, radiate individually more like the ions of a gas than a typical broadband solid. When such materials are heated, the emission spectra consist of relatively sharp lines in a limited portion of the spectrum, rather than being like a blackbody spectrum, because bands are unable to form. This phenomenon was first exploited by Auer von Welsbach (1896) in the Coleman-type lantern. In this case, the mantle consisted of thoria mixed with a precise amount of ceria. When heated, this material radiated in the central portion of the visible spectrum. Practical selective radiators The selective radiator (the lanthanide ion) is usually incorporated in a host material, an example of which is a rare-earth atom incorporated in a matrix of yttrium aluminium garnet. Unless the combination is well-designed, the radiative properties of the host can dominate the combination. Pioneering work on the emissive properties of the rare earth oxides was performed by Guazzoni (1972), who recognised the need to characterise the optical properties of these materials at high temperatures. Nelson (1992) realised that the background radiation, which is usually blackbody (or greybody) in nature, could be suppressed by making the thickness of the material less than its optical absorption depth, which is typically about 100500 fim thick. It is necessary to do this to achieve the high degree of selectivity of photons with energies approximately equal to the band gap of the converter, thereby optimising the efficiency. There is also very little thermal radiation from materials

490

T. J. Coutts

less than this thickness. For significantly lower thicknesses, the selective radiator properties are excessively suppressed. For greater thicknesses, the background radiation and internal scattering becomes excessive. The problem of producing such a thin radiator was solved by making bundles of fibres of a fabric impregnated with a solution of the lanthanide salt of interest. Thinfilm approaches have also been explored and will be discussed later. Most work was done on fibres of Yb 2 0 3 , but Er 2 0 3 and Ho 2 0 3 were also used. The fibre was dried and then heated to high temperature to remove the fabric and to oxidise the rare earth metal; the remaining structure consisting only of rare earth fibres. These were supported by a ceramic substrate, and inserted in bundles into holes in the latter. Fuel was fed through the holes from the rear of the substrate and burned just below the tip of the fibres, rather than near the substrate. This arrangement minimised the background blackbody radiation from the substrate. Approaches based on radiators made in similar ways have been developed in recent years and are still benefitting from significant federal funding in the USA. Reports on the topic are found in the NREL conference proceedings by, for example, Nelson (1994), Chen et al. (1996) and Goldstein et al. (1997). Another interesting approach developed by Lowe et al. (1994) from the NASA Lewis Research Center was based on films of thickness about equal to the absorption depth of the radiator materials, Yb 2 0 3 , Er 2 0 3 , and Ho 2 0 3 . However, instead of fibres, the oxide was incorporated in a thin film of yttrium aluminum garnet (YAG), a material widely used in laser technology, the film thickness of the YAG being substantially less than the absorption depth. The emissivities at the peak of the emission band reached about 80%, but the off-band emissivity was undesirably high at about 20%. Figure 11.4 shows the variation of the integrated emissivity with the thickness of an erbium-doped YAG film. Clearly, there is considerable scope for further improvement, which is expected in the near future. A successful radiator will lead to an increase in system efficiency, because of reduced fuel consumption, or to a higher power density output for the same fuel fuel consumption, because of a higher radiator temperature.

11.3 Optical control elements As mentioned earlier, there is a significant proportion of sub-bandgap photons in the broadband spectrum of radiators at temperatures in the 1000-2000 K range. In the next section, we shall discuss optimisation of the semiconductor converter band gap but, anticipating the results, these are in the range 0.5-0.7 eV. Figure 11.3 showed

Thermophotovoltaic

Generation of Electricity

491

0.6 0.5 o | 0.4 (d

I 0.3 >

I

02

LU

0.1 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 Radiator film thickness/cm

0.2

Figure 11.4 Variation of the integrated emissivity of erbium-doped YAG films as a function of film thickness, with the radiator temperature being treated parametrically.

that, even with a band gap of 0.5 eV, at least 50% of the photon flux is sub-band gap for a radiator of 1500 K. If this fraction is included in the calculation of efficiency, the result is discouragingly low, at about 8-10%. Consequently, it is vital to return the sub-bandgap photons to the radiator, to minimise fuel consumption. In this section, we shall briefly describe the methods that have been used to achieve the re-circulation of the sub-bandgap photons.

11.3.1 Dielectric stacks An ideal filter, used in conjunction with a broadband spectrum, would have a transmittance of unity up to a wavelength equivalent of the band gap of the semiconductor, and a reflectance of unity for all wavelengths above this. Alternatively, a very narrow band-pass filter centred at a wavelength roughly equivalent to the band gap could be used, although this would result in a lower power density output, because the incident optical energy would necessarily be limited to the bandwidth of the wavelengths passing through the filter. These filters depend on interference between rays reflected from the front and the back of the films. Filters with almost two hundred individual layers may be designed (using standard software design packages such as TFCalc™3), but fabricating them may be both practically 3

The email address of the company Software Spectra, Inc. that sells TFCalc™ is [email protected].

492

T. J. Coutts

difficult and expensive, because of the close control that must be maintained over film thickness, film density and interface coherence. Given that it will eventually be necessary to fabricate large areas of converters to generate significant amounts of energy, the problem may be excessively difficult. Clearly, a compromise must be achieved between performance and cost. In addition, the larger the area, the greater will be the distribution of angles of incidence of the radiation from the source on its way to the converter. Although these may seem to be insurmountable difficulties, it must be remembered that thin-film filters are used effectively in infrared detector technology. If the application is not highly cost-sensitive, then dielectric thin-film stacked filters may be feasible.

11.3.2 Plasma filters Investigations into plasma filters have been conducted by many researchers, such as Coutts et al. (1996), and the theory was originally proposed by Drude almost one hundred years ago. More modern and detailed explanations are given in many text books, an example of which is that by Born and Wolf (1985). The basis of these filters is that the electrons in a conductor are set into oscillatory motion by the electric field component of an electromagnetic wave. Their behaviour may be described by a linear differential equation of motion, the solution of which gives the time-dependent position and velocity of the electrons. The velocity then gives the AC conductivity of the material as a function of the effective mass, scattering time and density of the electrons, and the high-frequency permittivity of the material. The real and imaginary parts of the permittivity may then be calculated, and they give the optical constants of the material as a function of frequency or wavelength. Using these, it is then possible to calculate the optical constants of the material. From these, and an appropriately chosen film thickness, it is straightforward to calculate the reflectance, transmittance and absorptance as functions of wavelength. This elementary theory was compared with practical results for cadmium stannate (CTO) by Mulligan (1997) and the agreement was amazingly good for such a simple, single oscillator, approach. The key conclusion of the modelling studies is that a high electron mobility is essential to achieve high-performance filters. Without high mobility, both the selectivity of the filter and the free-carrier absorption are non-ideal. Figure 11.5 shows the modelled variation of free carrier absorptance with wavelength and mobility. The rate of turn-on of the filter with increasing wavelength in the transition from high transmittance to high reflectance, also improves with mobility. The transition occurs in the vicinity of the resonance and is due to the changing phase

Thermophotovoltaic Generation of Electricity

493

of the oscillating electrons relative to the electric field. This feature is also predicted by the simple Drude theory. At the peak of the absorptance, the free carriers are exactly in phase with the electric field vector i.e. the phenomenon is one of resonance as opposed to the increase to a more-or-less permanent high absorptance at the fundamental band gap. The wavelength of the resonance is designed to be slightly longer than the wavelength-equivalent band gap of the semiconductor to ensure that all higher energy photons are absorbed by the converter cell. 25 u = 100 cm* v " s" u = 500cmV's"'

20

u= 1000 cm V~'s"' 0 = 2000 cm \T' s"'

a 9-

8 10 < 5

0 0

•n'l 11111117 n~i 1 r 1 I~I 111 i i 1 r• irr -n-M-^-;2 4 6 8 10 12 14 16 18 20 Wavelength//ym

Figure 11.5 Modelled variation of free-carrier absorptance with wavelength, the mobility u being treated parametrically. The free-carrier concentration is taken as 3 x 10'° cm"'. M o r e recently, similar work has been performed on Ino.53Gao.47As, which is

lattice-matched to InP. This material can be doped heavily /i-type (up to about 1020cn-f3) and still retain a very high electron mobility, as discussed by Charache et al. (1999). This is possible because the effective mass of electrons in this material is low (~0.05»ic), which ensures high mobility, at least for material of reasonable quality. For carrier concentrations as high as 1020 cm"3, the mobility can be as high as 1000 cm2 V - ' s~' at 300 K, as discussed by Eastman (1993). A carrier concentration of 1020ctrf causes the plasma edge to appear at about 3 ftm, and the high mobility ensures extremely good selectivity and minimal free carrier absorption. At wavelengths longer than that of the plasma edge, the reflectance remains high (about 95-100%). The reduced effective mass is also low, thus ensuring a large BursteinMoss shift, because the available states at the bottom of the conduction band are filled very rapidly once the carrier concentration exceeds the degeneracy limit. Thus, for a

494

T. J. Coutts

carrier concentration of 1020cirf3, the optical band gap increases from the fundamental band gap of 0.73 eV, for non-degenerate material, to around 1.5 eV. This is equivalent to a wavelength of 0.8 \im, which is small enough to ensure essentially total transmittance of photons in the range 0.8-3 Jim. Meanwhile, photons with a wavelength greater than 3 pm are reflected by the front-surface Ino.53Gao.47As plasma filter. Modelling of these two systems (CTO and Ino.53Gao.47As) shows that the latter should perform far better than the former. However, these calculations assume that the states that are filled all appear in the same band, rather than spilling over to a nearby indirect band. Additionally, the back-surface filter, to be discussed in Section 11.3.4, appears to perform at least as well. Other materials are also being investigated.

11.3.3 Resonant array filters A band-pass filter can be made using a resonant antenna array, and was originally developed for use with sub-millimetre waves, these application having been discussed by Rhoads et al. (1982) and Tomaselli et al. (1981) and further developed by Home et al. (1980) specifically for solar and TPV applications. The filters are based on a dense array of thin-metal-film antennae deposited on a dielectric substrate, with the array consisting of either metal crosses or crosses etched in a metallic film. An array of crosses etched in a metallic film has the electrical characteristics of an inductive filter, whereas an array of metal crosses deposited on a dielectric substrate behaves like a capacitive filter. Inductive and capacitive filters give a band-pass or a band-reject function, respectively. Oscillating currents are induced in the filters by interaction with electromagnetic radiation of wavelength comparable with the dimensions of the filter. The magnitude of the electric field is different at different points on the surface of the filter, which leads to circulating currents. The magnitude of the transmittance is a function not only of the dimensions of the antennae, but also of their density on the substrate, of the conductance and thickness of the metal film, and the dielectric and optical properties of the substrate. The transmittance of the filter may be calculated by representing it by an AC equivalent circuit, an approach first discussed by Whitbourn and Compton (1985). These filters have the interesting property that they transmit only in a narrow band. Photons of both longer and shorter wavelengths are reflected back to the radiator. Originally, the arrays were made using direct-write electronbeam lithography, although this was slow and expensive. More recently, they have been made using a silicon stencil, fabricated using masked ion-beam lithography, and through which the metallisation was deposited. The stencil may be used many times as a mask over a gold film in which the array of micron-sized antennae is etched

Thermophotovoltaic

495

Generation of Electricity

I 40 E c 2 H

20

0

0

1

2

3

4

5

Figure 11.6 Variation of transmittance with wavelength of a resonant antenna array filter. This filter was designed for use in a radioisotope-fuelled TPV system. From Home et al. (1992).

using a proton beam. With this approach, it is claimed that the arrays can be made for under $1 cm"2. The performance of a typical filter is shown in Fig. 11.6. The transmittance within the required band is less than desirable, although improvements are being made through better edge resolution of the etched features and other means.

11.3.4 Back-surface optical reflectors Back-surface reflectors (BSR) were first used in connection with silicon devices by Borden (1980) and gallium arsenide devices by Boettcher et al. (1982). The principle is that the photons with sub-bandgap energies pass straight through the active layers of the device, and the substrate, which is semi-insulating (SI). The substrate does not absorb sub-bandgap photons because there are no free carriers to do so. The back of the substrate is metallised with a specular metal mirror that reflects the photons back through the substrate, the device layers, and out of the cell. The photons that have near-band-gap energy are thereby given a second opportunity for absorption, of possible value to devices with indirect band-gap converters. The photons with significantly less energy than that of the semiconductor band gap are still not absorbed on the second pass and they are returned to the radiator, from which they originated. This achieves the desired recirculation of the sub-bandgap photons. Ward et al. (1997), and others including Fatemi et al. (1997a) and Wilt et al. (1997), used the same concept in the development of low-band-gap TPV devices. These were based on semi-insulating InP substrates and consisted of a lateral conduction layer,

496

T. J. Coutts

followed by grading and device layers. These were In/ja^As, with x chosen to give a band gap of about 0.6 eV. The devices had lattice-matched Iny\si_yP window layers with band gaps of ~1 eV for electronic passivation, thus forming a fully passivated double heterojunction. All layers were grown using atmospheric pressure organometallic chemical vapour deposition (OMCVD), as outlined by Wanlass et al. (1998).

11.3.5 Summary of optical control approaches Each of the filters discussed has scope for improvement in performance, although each has already been incorporated, partially successfully, in prototype systems. It seems unlikely that cadmium stannate, or indeed any plasma filter based on a transparent conducting oxide, will be able to achieve the required free-carrier mobility. However, it may be possible to achieve this mobility using a degenerate single-crystal semiconductor such as In/ja^As. Although it is possible to design and fabricate dielectric stacks that exhibit the required optical functionality, such designs are probably costly to make in large area, as well as being sensitive to variations in angle of incidence, film thickness and specularity of the interfaces. In fact, on the basis of a Lambertian distribution of angles of incidence, it is necessary to design the filters for the average angle of incidence. Combinations of plasma and dielectric filters have been used successfully and may be less prone to these variations. Resonant array filters do not appear to be as costly as expected to fabricate. More complicated multilevel structures may be made that can result in improved transmittance and narrower bandwidth, but Chan (1995) pointed out that improvements in edge resolution are also required. In principle, the narrower bandwidth of the resonant array filter should be ideal because it would eliminate thermalisation of hot electrons, as well as free-carrier absorption of sub-bandgap photons. However, it is vital to ensure that sufficient power is contained within the transmitted band to yield adequate power density output from the device. The filter is essentially a means of converting a broadband into a narrow-band spectrum, equivalent to that emitted by a selective radiator. At present, the back-surface reflector appears to have the best near-term prospects for success, given the results already achieved. On the other hand, it is vital to maintain a high degree of specularity at the metallic back surface on the semi-insulating substrate. A high degree of parallelism between the front and back surface must also be maintained to avoid trapping sub-bandgap photons by reflection within the semiconductor substrate. The function of the back-surface reflector is not to increase the optical path length and optical absorption, but to reflect the sub-bandgap photons back to the radiator. Light

Thermophotovoltaic Generation of Electricity

497

entering the cell at a non-normal angle from between the grid lines has a high probability of being trapped by multiple reflections between the back-surface reflector and the grid lines themselves. In addition, total internal reflection from the surface of the semiconductor is also an important issue.

11.4 Device modelling The approaches reviewed in this section were developed by Cody (1998) (who considered devices limited only by radiative recombination i.e. the principle of detailed balance originally developed by van Roosbroeck and Shockley (1954), and used in predicting maximum efficiencies of solar cells by Shockley and Queisser (1961), by De Vos (1992) (who treated the converters as endoreversible heat engines), and by Gray and El-Husseini (1995). In addition, we also review the work of Wanlass et al. (1989) (who used an empirical approach based on measurements of the reverse saturation current of many devices as a function of band gap). The latter approach was based on an empirical equation relating the reverse-saturation current density of the device to its band gap. The conclusions obtained from each of these approaches could be useful in designing future-generation devices, and they are briefly reviewed in this section.

11.4.1 Radiative recombination Recombination refers to the process by which a photogenerated carrier returns to the ground state. This may occur via defects in the semiconductor crystal such as point defects, grain boundaries, vacancies, interfaces and surfaces, within the space-charge region of a p-n junction, by Auger processes involving three particles, or by direct recombination of excited charges of opposite sign with the accompanying emission of a photon. The latter may or may not generate further electron-hole pairs, a process that is known as 'photon recirculation' (not to be confused with the function of optical control elements described in Section 11.3). To maintain the steady state, the recombination rate of the excess charge must also be equal to the absorbed flux within the absorbed volume. The rate of recombination is inversely related to the concentration of dopants in the semiconductor, with a constant of proportionality known as the B-factor, which is governed by properties of the crystal and was described in detail by Ahrenkiel (1993). The rate of recombination is increased if any of several other possible recombination mechanisms is significant. Radiative

498

T. J. Coutts

recombination determines the fundamental upper limit on the lifetime of photogenerated carriers, unless they can be extracted before recombination occurs. Tiedje et al. (1984) and Cody (1998) used the theory of radiative recombination to model the limiting efficiency of silicon solar cells and TPV cells, respectively. They showed that this is significantly greater than that achieved with the highest-quality laboratory cells. On the other hand, they made the interesting point that the best laboratory cells now have efficiencies greater than those predicted by the semiempirical models used by earlier workers, to be discussed in Section 11.4.3. An important conclusion may be drawn from Cody and Tiedje (1992), whose argument applied to silicon solar cells. A well-funded, sustained and well-managed effort ought to be made to establish experimental upper limits to the efficiency of TPV devices, rather than to predicting that which may be achievable using present-day methods and approaches. To achieve the very high efficiencies of modern silicon cells, achieved by (for example) Zhao et al. (1995), a long-term program of research and development into materials, as well as device design and technology, has been necessary. The same philosophy could be used to the benefit of TPV. Using this argument, Tiedje et al. (1984) concluded that silicon cells have now reached about 60-80% of the limit predicted by radiative recombination theory. If the same could be achieved for TPV devices, then many more markets could become accessible to TPV generators. This important point was first made by Cody (1998), although the results of Gray and ElHusseini (1995), discussed below, were identical.

11.4.2 Endoreversible heat engines De Vos (1992) developed a generalised theory of endoreversible heat engines and applied it to several devices. The term 'endoreversible' means that the heat engine is fully reversible in its internal processes and connections, the losses occurring only because of external interactions. Two functions were derived initially and these were completely general to any reversible conversion process. These two general equations were then made specific to each device. An arbitrary spectrum may be used, although it is assumed to be a broadband radiator of temperature Temit in the present context. In the analysis of Gray and El-Husseini (1995) the output power density was based on the expression derived by De Vos (1992), an equivalent form of which is

p =

2* «V«, c'h3

r x A\\ 2

iu. exp U /kT „ )

^ )u, exp((£/-~i

s^^s"

^ ^ 7 ^ .

/

y^.-' > / >

•y

r

'

••'"'

' *

-

^ ^ \2000K

~ :

35 30

25

C O •

40

j

J&V,

'

~^"~

«