Thin Films, Vol. 27: Non-crystalline films for device structures

Thin Films Non-Crystalline Films for Device Structures Volume 29 Serial Editors Inorganic Thin Films STEPHEN M. ROSSN...

Author: Francombe M.H.

23 downloads 1145 Views 14MB Size Report

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

DOWNLOAD PDF

Thin Films Non-Crystalline Films for Device Structures Volume 29

Serial Editors Inorganic Thin Films STEPHEN M. ROSSNAGEL

Organic Thin Films ABRAHAM ULMAN

IBM Corporation, T. J. Watson Research Center Yorktown Heights, New York

Alstadt-Lord-Mark Professor Department of Chemistry Polymer Research Institute Polytechnic University Brooklyn, New York

Honorary Editor MAURICE H. FRANCOMBE Department of Physics and Astronomy Georgia State University Atlanta, Georgia

Editorial Board DAVID L. ALLARA Pennsylvania State University ALLEN J. BARD University of Texas, Austin MASAMICHI FUJIHIRA Tokyo Institute of Technology GEORGE GAINS Rensselaer Polytechnic Institute PHILLIP HODGE University of Manchester JACOB N. ISRAELACHIVILI University of California Santa Barbara

JEROME B. LANDO Case Western Reserve University HELMUT MOHWALD University of Mainz NICOLAI PLATE Russian Academy of Sciences HELMUT RINGSDORF University of Mainz GIACINTO SCOLES Princeton University JEROME D. SWALEN International Business Machines Corporation

MICHAEL L. KLEIN University of Pennsylvania

MATTHEW V. TIRRELL University of Minnesota, Minneapolis

HANS KUHN MPI Gottingen

GEORGE M. WHITESIDES Harvard University

Recent volumes in this serial appear at the end of this volume

Thin Films Non-Crystalline Filmsfor Device Structures Edited by Maurice H. Francombe

Department of Physics and Astronomy Georgia State University Atlanta, Georgia

VOLUME 29

ACADEMIC PRESS A Division of Harcourt, Inc. San Diego

San Francisco New York London Sydney Tokyo

Boston

This book is printed on acid-free paper. | Compilation copyright 9 2002 by ACADEMIC PRESS All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. The appearance of the code at the bottom of the first page of a chapter in this book indicates the Publisher's consent that copies of the chapter may be made for personal or internal use of specific clients. This consent is given on the condition, however, that the copier pay the stated per copy fee through the Copyright Clearance Center, Inc. (222 Rosewood Drive, Danvers, Massachusetts 01923), for copying beyond that permitted by Sections 107 or 108 of the U.S. Copyright Law. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works, or for resale. Copy fees for pre-2002 chapters are as shown on the title pages. If no fee code appears on the title page, the copy fee is the same as for current chapters. 1079-4050/2002 $35.00. Explicit permission from Academic Press is not required to reproduce a maximum of two figures or tables from an Academic Press chapter in another scientific or research publication provided that the material has not been credited to another source and that full credit to the Academic Press chapter is given. The articles in this book are selected from the Academic Press multi-volume work titled

Handbook of Thin Film Materials, edited by Haft S. Nalwa, and are uniquely arranged to focus on current advances in surface science. Academic Press

A division of Harcourt, Inc. 525 B Street, Suite 1900, San Diego, California 92101-4495, USA http://www.academicpress.com Academic Press Harcourt Place, 32 Jamestown Road, London NW1 7BY, UK http://www.academicpress.com Intemational Standard Book Number: 0-12-533029-4 International Standard Serial Number: 1079-4050 PRINTED IN THE UNITED STATES OF AMERICA 01 02 03 04 05 CO 9 8 7 6 5 4 3 2 1

Contents

Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii ix

Ultrathin Gate Dielectric Films for Si-Based Microelectronic Devices C. Krug and I . J. R . Baumvol Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Requirements of Ultrathin Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . Ultrathin Gate Dielectric Film Processing . . . . . . . . . . . . . . . . . . . . . . . . . . Characterization of Ultrathin Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . Hydrogen and Ultrathin Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . Silicon Oxide Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Silicon Oxynitride Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . . . . 1.8. Alternative (High-k) Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . . . I .9. Final Rcmarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1. 1.2. 1.3. 1.4. 1.5. 1.6. I .7.

1 6 6 16 45 52 82 104 122 122 123

Electrochemical Passivation of Si and SiGe Surfaces J. Rappich and Th. Dittrich 2.1. 2.2. 2.3. 2.4. 2.5. 2.6. 2.7.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . In Situ Characterization of Surface Bond Configurations and Electronic Surface States . Electrochemically Hydrogenated Si Surfaces . . . . . . . . . . . . . . . . . . . . . . . . Hydrogenated Porous Silicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thin Anodic Oxides on Si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thick Anodic Oxides on Si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Enhanced Passivation of SiGe by Anodic Oxidation . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

V

135 137 159 182 200 224 233 249 249

261

This Page Intentionally Left Blank

Contributors

Ultrathin Gate Dielectric Films for Si-Based Microelectronic Devices: C. Krug,

I. J. R. Baumvol, Instituto de Fisica, Universidade Federal do Rio Grande do Sul,Porto Alegre, RS 91509-900, Brazil Electrochemical Passivation of Si and SiGe Surfaces: J. Rappich, Hahn-Meitner

Institut, Abteilung Silizium-Photovoltaik, Berlin D- 12489, Germany Electrochemical Passivation of Si and SiGe Surfaces: Th. Dittrich, Technische

Universit~it, Mtinchen, Physikdepartment E16, Garching 85748, Germany

vii

This Page Intentionally Left Blank

Preface

Volume 29 of Thin Films, titled Non-Crystalline Films for Device Structures, presents two chapters covering in detail the preparation and properties of noncrystalline layers and their critical role in the successful development of current and emerging semiconductor microelectronic components. Previous volumes in the series have generally stressed crystalline and epitaxial films or multi-layers performing the key, active functions in electronic devices. This is the first topical treatment of non-crystalline (or amorphous) films in relation to such structures. The review articles included in this volume have been selected from a new multi-volume work, Handbook of Thin Film Materials, edited by Dr. Hari Singh Nalwa for publication by Academic Press. The material used was chosen primarily for its relevance to high-performance semiconductor device technologies, which are strongly dependent on improvements in thin film processing and optimization of dielectric and electronic properties. The common theme in these articles focuses on the critical use of non-crystalline insulator layers, for example, for gate dielectrics and surface passivation in semiconductor devices. Chapter 1, authored by C. Krug and I. J. R. Baumvol, offers a comprehensive treatment of ultrathin gate dielectric films for Si-based microelectronic devices. Continued progress in the area of MOSFET arrays for ultralarge-scale integration (ULSI) requires further reduction of insulator thickness below the present limit of ~3 nm, coupled with improvements in dielectric quality to diminish current leakage. This is a key area of thin film technology and production, with enormous future growth potential. Sales of semiconductor integrated circuits for the year 2000 alone were estimated to be 150 billion dollars. After the general criteria for thin gate dielectrics in MOSFET devices are surveyed, thermal, physical, and chemical growth approaches and electrical and physicochemical characterization techniques are outlined. Next, the beneficial and detrimental effects of hydrogen during oxide growth are discussed. Following sections consider the properties and limitations of silicon oxide and silicon oxynitride gate dielectric layers and the potential and feasibility of high-k dielectric films. Finally, it is concluded that silicon-based dielectrics will continue to be used until 2005 or so, possibly to be supplanted by high-k dielectrics such as ZrSiO4, Ta2Os, or TiO2. Chapter 2, by J. Rappich and Th. Dittrich, presents an exhaustive status review of the electrochemical passivation of Si and SiGe surfaces. The subject matter is closely related to that discussed in the previous chapter, in that low-temperature approaches for growth of low-leakage oxide layers suitable for high-performance semiconductor device structures form a significant aspect of the outline. The advantages of electrochemical passivation are that specific chemical reactions are ix

X

PREFACE

locally activated at the Si surface by an applied electric potential, and they can be controlled by using the measured current for monitoring progress of the reaction. Furthermore, the low reaction temperatures used (relative to those needed for thermal oxidation) are ideally suited to passivation of chemically less stable semiconductors such as SiGe alloys, which are becoming very important for higher performance hetero-bipolar transistors or MOS structures. The authors discuss, among other aspects of this approach, the powerful advantages of in situ optical techniques, e.g., Fourier transform infrared spectroscopy (FTIR), surface photovoltage (SPV), and photoluminescence (PL) to monitor information on surface chemical bonds, electronic trap states, and non-radiative recombination centers at the Si surface, respectively. The ability to achieve high-quality passivation and low-leakage anodic dielectric layers by such low thermal budget approaches offers exciting advantages in the future manufacture of more powerful semiconductor memory systems. MAURICE H. FRANCOMBE

THIN FILMS, Vol. 29

Ultrathin Gate Dielectric Films for Si-Based Microelectronic Devices C.

KRUG

AND

I. J. R.

BAUMVOL

Instituto de Ffsica, Universidade Federal do Rio Grande do Sul, Porto Alegre, RS 91509-900, Brazil

1.1. 1.2. 1.3. 1.4. 1.5. 1.6. 1.7. 1.8. 1.9.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Requirements of Ultrathin Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . Ultrathin Gate Dielectric Film Processing . . . . . . . . . . . . . . . . . . . . . . . . . . Characterization of Ultrathin Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . Hydrogen and Ultrathin Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . . Silicon Oxide Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Silicon Oxynitride Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . . . . . Alternative (High-k) Gate Dielectric Films . . . . . . . . . . . . . . . . . . . . . . . . . . Final Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 6 6

16 45 52 82 104

122 122 123

1.1. Introduction A microelectronic device may be defined as one that relies on electrons (specifically, on their quantum behavior) for functioning. "Micro," although related to physical dimension, stands for a character of building block; an electronic device is usually an assembly of microelectronic components. Generally speaking, a device must be controllable to be useful. The possibility of controlling a significant physical quantity, electron "flow," is offered by the band structure of a crystal of semiconducting material suitably doped with an electron donor or acceptor. From a myriad of semiconductors, silicon has been chosen to form the vast majority of microelectronic devices, totally dominating the commercial market. However, if we are currently witnessing a silicon age, it is not solely because silicon is a semiconductor. (Gallium arsenide (GaAs), for example, shows superior electron transport properties and special optical properties [ 1]). It is perhaps mainly due to the fact that silicon has a stable electrical passivating oxide, which can be placed on top of it with a structurally sharp interface of incomparable electronic quality; i.e., there is an extremely low density of interface electronic quantum states. Electrical passivation of the semiconductor crystal surface is the key factor to THIN FILMS Copyright 9 2002 by Academic Press Vol. 29 ISBN 0-12-533029-4

All rights of reproduction in any form reserved ISSN 1079-4050/02

$35.00

2

KRUG AND BAUMVOL

FIG. 1. Three-dimensional sketch and cross section of an n-channel MOSFET on p-type Si, showing the channel width W and length L. The active region of the device under bias is also shown.

control, as surface properties can be the dominant influence on the electronic current in the device in opposition to the contribution of charge transport in the bulk. This is undesirable because devices with surface-dominated characteristics usually exhibit poorer performance and stability than devices with bulk-dominated characteristics. Even ideal surfaces can possess a density of electronic states that would make any device uncontrollable [2]. Finally, silicon is very abundant and can conveniently be converted into high-purity single crystals suitable for device manufacturing. A key component of present-day microelectronics built on the concept of the electrical passivation of silicon by its oxide is the metal-oxide-semiconductor field-effect transistor (MOSFET) [3, 4], shown schematically in Figure 1. It can be structurally divided into gate, oxide, source, drain, and bulk semiconductor. The gate is a conducting electrode used to control the device. The passivating oxide is also called the gate dielectric. The source and drain are islands doped opposite from the substrate. The semiconductor in contact with the gate oxide and between the source and drain forms the active region of the MOSFET. In such a device, an electric field produced by a voltage applied between the gate electrode and the silicon substrate (the gate voltage VG) can control charge flow (electrical current)

ULTRATHIN GATE DIELECTRIC FILMS

from the source to the drain (the drain current ID, produced by an applied drain voltage VD)mhence the "field effect" denomination. Under favorable gate bias, a channel appears, connecting the source and drain, and charge flows between them (Fig. 1). The metal/oxide (SiO2)/semiconductor (Si) or MOS structure is, without a doubt, the core structure of modern-day electronics. (Actually, heavily doped polycrystalline silicon (polysilicon, poly-Si) has been favored over metal (aluminum) electrodes since about 1970, eliminating the technological problem of aligning the gate to the insulating film.) This cannot be ascribed to a single characteristic of the MOSFET, but is due to a favorable combination of properties [3]. The two-terminal MOS capacitor (MOS-C) is the structural heart of all MOS devices. There lies the gate dielectric. The MOS-C consists of a parallel plate capacitor with a metallic plate and the semiconductor substrate as electrodes. The two electrodes are separated by a thin insulating layer of SiO2. The MOS designation is reserved for the metal-SiO2-Si system; a more general designation, metal-insulator-semiconductor (MIS), is used to identify similar device structures composed of an insulator other than SiO2 or a semiconductor other than Si. One key feature of the MOSFET is its planar geometry, which allowed the development of integrated circuits (ICs), i.e., the arrangement of many individual devices on the same semiconductor chip. The vast majority of discrete devices and ICs are based on MOSFETs. In complementary metal-oxide-semiconductor (CMOS) technology, both n-channel (p-bulk, electron current carrying) MOSFETs and p-channel (n-bulk, hole current carrying) MOSFETs are fabricated on the same chip. Today, CMOS technology is the dominant semiconductor technology for microprocessors, memories, and application-specific integrated circuits (ASICs). Sales of semiconductor ICs were estimated to be $150 billion for the year 2000. The main advantage of CMOS is its low power dissipation. A CMOS circuit has almost no static power dissipationmpower is only dissipated when the circuit actually switches. This makes it possible to integrate many more CMOS gates on an IC, resulting in much better performance. Significant expressions of MOSFET performance are drain current, charge carrier mobility, and switching speed. In part, these depend on properties more or less under the designer's control [3]. Both drain current and switching speed are proportional to the gate capacitance and to the device aspect ratio, given by the channel length divided by the channel width. The gate capacitance, in turn, is inversely proportional to the gate oxide thickness. The drive for increasing device performance has led to continuous MOSFET scaling in the direction of decreasing gate oxide thickness and increasing aspect ratio. Such scaling, involving such parameters as gate length, gate width, gate oxide thickness, source/drain junction depth, substrate dopant concentration, and supply voltage (Table I), must follow some rules [5] to cope with the "short-channel" effect. Briefly stated, the shortchannel effect refers to the tendency toward a lower gate voltage at which drain current is observed (threshold voltage) as the source-to-drain distance (channel

KRUG AND BAUMVOL

TABLE I MOS TRANSISTORSCALINGATA GLANCE Design parameter

Scaling factor Upon

Transistor dimensions Voltage Doping

Decreased by k Decreased by k Increased by k Result

Circuit area Speed Currents Power per circuit Power per unit area

Reduced by 1/ k2 Increased by k Reduced by 1/ k Reduced by 1/ k2 Held constant

length) is reduced, because of the two-dimensional electrostatic effect involving the source, the drain, the gate, and the channel regions. As a rule of thumb, the gate oxide thickness should be about 1/50 to 1/25 of the channel length to keep the short-channel effect under control [6]. So if one takes gate capacitance as the scaling parameter to increase device performance, channel length must be scaled concomitantly. As a result, the whole device shrinks, and more devices can be placed on the same chip area, increasing integration and decreasing the cost per transistor. Gate and drain voltages must be scaled at the same time, and one finds that power consumption is drastically reduced. With regard to the MOSFET, one is tempted to say that "the smaller, the better." Because of scaling, semiconductor device manufacturing has gone through medium- (MSI), large- (LSI), very large- (VLSI), and now ultralarge-scale integration (ULSI). Computing power has been doubling every 18 months since the 1970s [7], following what became known as Moore's law [8] (Fig. 2). Goals and strategies for developments in semiconductor technology appear in the biannual editions of the Semiconductor Industry Association's roadmap [9]. Reduction of the gate dielectric film thickness is one of the areas of device fabrication that now limits CMOS scaling. In active CMOS devices, the thickness of the gate insulator has been scaled down from 100 nm in the early 1970s to less than 3 nm for current devices with an effective channel length of less than 0.2 #m. Already at the present stage one cannot make a strict distinction between "surface" "bulk," and "interface" of the gate dielectric films, from a structural or compositional point of view. The quality and reliability of these films determine the performance of the ICs that contain them. The drive of ULSI establishes new and very strict material requirements. Outstanding scientific and technological

ULTRATHINGATEDIELECTRICFILMS

{3 .~_ 'E o ..Q E .-I Z

~ o

IOM

~

.,~'~48'~6

1M

.~

2000

P~n~ieusm;r

0.1M O.01M ~176

19'75

19;o

19'8s

19;o

Production year

19;~

FIG. 2. Moore's law illustrated by the time evolution of the number of transistors in Intel chips.

challenges have been created by the downscaling of CMOS devices. Most of them have been surpassed at the cost of massive research, but some remain at the very forefront, limiting further evolution of silicon science and technology [10-12]. Those concerning gate oxides, and more generally gate dielectrics, are close to fundamental limits, where the need for atomic-scale understanding and control becomes ever more critical. A limit to silicon oxide as a gate dielectric has already been foreseen, and alternative materials are now under intense investigation [ 13]. This chapter is intended to present and discuss the broad view of ultrathin gate dielectric films on silicon. The main purpose is to review the current status of ultrathin gate dielectric science and provide the means for the reader to keep pace with the new scientific literature in this very lively research area. Discussion in depth is restricted for the sake of brevity, with references offered to the interested reader. The material is intended for either continuous or sectional reading. Emphasis is placed on the correlation of dielectric quality, physicochemical issues, and processing parameters. Basic requirements of ultrathin dielectric films to be used in microelectronic devices are listed in Section 1.2. Film preparation methods are outlined in Section 1.3. Both electrical and physicochemical methods of characterization of dielectric films are presented in Section 1.4. The significance and effects of the hydrogen presence in gate dielectrics are presented in Section 1.5. Section 1.6 describes silicon oxide films thermally grown on single-crystalline silicon in dry oxygen. They have allowed and promoted semiconductor device development for 40 years. Silicon oxide is the gate dielectric p a r excellence, and all discussions take it as a reference. Understanding and simulating modem processing routes to the formation of gate dielectric films require an understanding of the atomic transport processes responsible for their growth. Atomic transport is the natural way to approach the growth of ultrathin films and so is explored in this

KRUG AND BAUMVOL

text. Because of ever-increasing constraints brought about by integration issues, silicon oxide has been at least partially replaced. Dielectrics generally referred to as silicon oxynitrides are addressed in Section 1.7, considering a variety of preparation methods. (Silicon nitride finds a number of applications in the semiconductor industry, but not in the form of ultrathin gate dielectric films, and so will not be discussed here.) Recently proposed dielectric materials that are alternatives to these are covered in Section 1.8. Section 1.9 summarizes the chapter and gives concluding remarks.

1.2. Requirements of Ultrathin Gate Dielectric Films Ultrathin dielectric films must fulfill a number of electronic, physical, and chemical requirements to be useful in microelectronic devices [ 14]. A major concern for the gate dielectric is minimization of leakage current, as this ensures the field effect. A large bandgap is desirable for the prevention of leakage and for ensuring a lower sensitivity to radiation. A high dielectric constant (permittivity) is becoming more and more important as integration increases. Uniform and high dielectric strength (high breakdown voltage) is also necessary. To obtain high-performance MOSFETs with fast turn-on characteristics and high carrier mobility, electrically active defects (and the associated charge) encountered in the gate insulator and at its interfaces must be minimized. The dominant criterion is still a low density of electronic states at the insulator-semiconductor interface, which makes SiO2 the first choice unless one of the other already mentioned aspects cannot be fulfilled. To make sure that the fabrication yield of ICs is kept acceptable despite the ever-increasing chip size, gate dielectric films must be homogeneous, uniform, and both physically and chemically stable. (Although individual devices keep shrinking, chips are growing because of increasing device integration.) For the sake of device stability, these dielectric films should prevent all foreign species from diffusing into active parts. Foreign species can be doping atoms, ions diffusing from a metal gate electrode, or any type of contaminant introduced during the fabrication process. One very illustrative and disturbing example is diffusion of boron from heavily doped gate electrodes made of polycrystalline silicon. Such characteristics are usually perfected through cleanliness and control of the film growth rate. The domain of ultrathin dielectric films is particularly tricky because as a general rule the first stages of formation are not fully understood.

1.3. Ultrathin Gate Dielectric Film Processing In current IC fabrication technology, passivation of the semiconductor surface with a suitable dielectric film is one of the first processing steps. It is therefore

ULTRATHINGATE DIELECTRICFILMS

considered as a "front end of line" process. With regard to manufacturing, dielectric films can be divided into two groups: "grown" and "deposited." The term "grown" designates films that are formed by means of a chemical reaction (oxidation, nitridation, etc.) involving the semiconductor substrate and whose growth depends on atomic transport, and "deposited" means that the dielectric has been formed without any chemical reaction with the semiconductor [ 14]. A film unintentionally grown by simple exposure of a clean surface to the ambient is usually called "native." Examples are SiO2 and A1203 films formed on clean Si and A1 upon exposition to air. This section addresses different approaches to grown and deposited dielectric film fabrication. The presentation is preceded by considerations on substrate cleaning, which is a key factor regardless of the particular method chosen for the preparation of the passivating film. 1.3.1.

SILICON WAFER CLEANING

Cleaning is the most frequently repeated step in IC production. One of the most critical cleaning steps is that which precedes ultrathin dielectric film growth or deposition. The RCA clean, developed in 1965, still forms the basis of most front-end wet cleans [15-19]. A typical RCA-type cleaning sequence starts with a sulfuric acid-hydrogen peroxyde mixture (SPM) step (HzSO4]H202) followed by a dip in diluted HE The SC1 step (NH4OH/H202/H20) then removes particles, and the SC2 step (HC1/HzOz/H20) removes traces of metals. Environmental concerns and the search for greater cost effectiveness have recently led to much research effort directed toward understanding cleaning chemistries [20]. A variety of organic contaminants can exist in IC processing (human skin oils, pump oil, silicon vacuum grease, cleaning solvents). A potential problem arising from this is the presence of an organic film at the silicon surface, preventing the action of cleaning solutions. Therefore, removal of organic contamination is often the first step in cleaning. SPM (H2SO4/I-I202) or sulfuric acid-ozone mixture (SOM) (H2SO4/O3) have successfully been applied to this end. Traditional cleaning sequences use mixtures based on H2SO4 to grow a sufficiently thick chemical oxide to obtain high particle removal efficiencies in the second step of the cleaning sequence. The effect of organic contamination on SiO2 film quality was tested by wafer exposure either between cleaning and oxidation or oxidation and poly-Si gate electrode deposition [20]. Charge-to-breakdown (see Section 1.4.1.2) measurements showed no difference in the intrinsic oxide breakdown, but the extrinsic tail was influenced by the different treatments. The presence of organic contamination is likely responsible for the extrinsic defects in the thin oxide layer. In the SC2 mixture the H202 can be left out completely, inasmuch as it has been shown that diluted HC1 is as effective in the removal of metals as the standard SC2 solution [20]. Moreover, at low HC1 concentrations, partictes are kept away from

8

KRUG AND BAUMVOL

the silicon surface. This is due to the fact that the isoelectric point for silicon and silicon oxide is between pH 2 and 2.5. For most particles in liquid solutions at pH values greater than 2-2.5, an electrostatic repulsion barrier is formed between the particles in the solution and the surface. An additional step can be added to the cleaning sequence to make the silicon surface hydrophilic. This allows easier drying without the generation of drying spots or watermarks. In the Marangoni dryer [21], the drying is performed by a strong natural force (the Marangoni effect) in cold deionized water, and the wafer is rendered completely dry without the evaporation of water. Trace amounts of noble-metal ions, such as from Ag, Au, and, especially, Cu, are present in HF solutions and can deposit on the Si surface. An optimized HF/HC1 mixture provides protection against metal outplating from the solution. Ca shows a pronounced tendency to deposit on the wafer surface, with a strong effect on the gate oxide integrity [22]. Low pH and high temperature were found to result in lower Ca surface concentration [20]. A transport-limited surface deposition was also observed, so low pH, high temperature, and fast cleaning minimized Ca contamination. The chemical state of the silicon surface after a cleaning process can be either an oxidized form or oxide-free, bare silicon, terminated by S i - H groups [23]. It has been shown that the presence of a thin chemical oxide (0.6 to 0.8 nm) on the Si surface before deliberate oxidation does not significantly influence final oxide thickness or thickness variation on a wafer. This chemical oxide forms a passivation layer with a dangling-bond defect density in the range of 1012 cm -2 at the SiO2-Si interface, two orders of magnitude higher than that of device-grade thermal oxides. Wet chemical oxides protect the Si surface from some recontamination or at least render the surface less sensitive. On the other hand, such thin oxides left on the surface are detrimental, for example, to epitaxial silicon deposition and might become increasingly critical for subsequent thermal oxidation as the gate oxide thickness decreases. For such purposes, cleaning sequences that end with an HF step might be preferred, as the surface becomes hydrophobic and the silicon dangling bonds are passivated with hydrogen atoms. However, no significant differences were found between wafers that received HF and SC2 final cleaning, with respect to electrical performance and defect density, provided metal and particle contamination was sufficiently low. So far experimental proof for the superiority of HF-last cleaning has not been reported [24]. Surface roughness of Si wafers received much attention over the last few years as a possible cause of gate oxide defects. It was recently reported that as far as yield loss is concerned, silicon surface roughness for very thin oxides (95%) is found near the SiO2-Si interface, as proposed in the model of Deal and Grove. In the film of intermediate thickness, 180 is also found near the gas-SiO2 interface. In this region it is distributed according to a complementary error function, a profile typical of diffusing species. In the thinnest film, enrichment of SiO2 in the 180 isotope at the interface with Si is no longer 100% and 180 becomes incorporated in the "bulk" oxide. Figures 55 [ 160] and 56 show extensive isotope intermixing after 1602, 1802 sequential oxidations. Figure 54 provides the basis for a "physical" definition of thick, thin, and ultrathin oxide film. Although the mechanisms responsible for the distinct features in Figure 54c as compared with its counterparts are certainly active during the growth of thicker films, their contribution to oxide growth is increasingly negligible. Continuing with reference to where silicon oxidation takes place, these mechanisms are now explored. The first systematic experimental investigation of isotopic substitution in ultrathin films was made by Rochet et al. [ 161 ]. Growth was promoted in ultra-dry 02 (less than 1 ppm of H20), in the temperature range from 810~ to 1090~ to total film thicknesses from 2 to 7 nm. Oxidations performed in the 1602, 1802 gas sequence showed most (between 98% and 80%) of the 180 incorporated in the near-surface region of the samples. This is illustrated in Figure 57, which shows results from etch-back NRA (Section 1.4). The steep dashed lines in the plots indicate the enhanced incorporation of 180 in the near-surface of the oxide film. Such 180 is found to be incorporated in two different modes, namely either in exchange for 160 originally in the samples or contributing to oxide film growth. This brings two new ideas not contemplated within the framework of the model of Deal and Grove: reaction of oxidizing species not at the SiO2-Si interface, but

74

KRUG AND BAUMVOL

FIG. 54. 180 profiles as determined by NNRP for silicon oxide films of different thicknesses sequentially grown in natural (quoted 160) and 180-enriched 02. The profiles are representative of original (Si1602) oxide films (a) thicker than 30 nm; (b) between 10 and 30 nm; and (c) thinner than 10 nm. Reprinted with permission from [31], 9 1996, Elsevier Science.

near the gas-SiO2 interface and resulting in (i) O exchange and (ii) film growth. Isotopic exchange between the oxygen atoms from the gas phase and those already existent in the oxide film was shown to be present at all temperatures above 800~ At all oxidation temperatures, the contribution to oxide film growth of 180 incorporated near the surface decreases with the increase in the total film thickness. Experiments also showed that the less dry the 02, the less favored is the fixation of 180 in the near-surface of the samples. This could be due to a fastdiffusing species combining H and 180, like H2180. Experiments involving oxide

75


I i ~ i 5 nm . . . .

'"

~8 0

z

z -

-

-

~

Q

, t

5

i I i

i

i

i i e e

'! T

-1

i i i

, I

d

! I

C

g

.~

~r

'*

,

a

I 0

75.5

76.5 77.5 78.5 79.5 Backscattered proton energy (keY)

' I

80.5

FIG. 55. Oxygen section of MEIS spectra for (a) an initial Si1602 film sequentially exposed to 1802

at (b) 800~ 1 Torr, 21.5 h; (c) 800~ 1 Torr, 40 h; (d) 900~ 0.1 Torr, 5 h; (e) 900~ Reprinted with permission from [ 160], 9 1995, American Institute of Physics.

1 Torr, 5 h.

films in the thickness range between 20 and 300 nm are in agreement with such results [ 162, 163]. Figure 58a shows the 180 profile in a silicon oxide film thermally grown on Si in a 1602, 1802 gas sequence, as measured by NNRP (Section 1.4). The nearsurface 18O distribution follows a complementary error function (erfc). Figure 58b shows the 180 profile after a third oxidation step in 160 2. As expected, a region enriched in 160 is found in the near-surface. Also shown is a significant

76

KRUG AND BAUMVOL

1170K, 10 .2 T0rr

1.0 0.8

min (1802)160 min (1602)

0.6 0.4 0.2

~.. ,....

160

-~,80

~.,,.

b U~ r

s

"E3 N

0.8

4 rain (1B02)/300 rain (1802)

0.6 0.4 /

P"

....... "'o."

0,2 ~..~' O

z

~

%.....

0,0

4 min (~802)/

O.8

1860 min (~602)

O.6 0.4 0.2

0.0 ~""""""""~ "........ ""~ ..... -;

....

~'~ 3;0

Depth (A)

"

FIG. 56. Isotope depth distributions from MEIS measurements on silicon oxide films sequentially

grown in 1802 and 1602. Reprinted with permission from [82], 9 1995, American Physical Society.

loss of 180 previously existing close to the gas-oxide interface, accompanied by modification of the original erfc-like profile. Figure 59 shows the experimentally determined erfc-like and final 180 profiles, together with a calculated (labeled "predicted") final profile. The good agreement with experimental data was taken as a confirmation of the hypotheses used in the calculation, namely (i) the defects responsible for near-surface O incorporation can be thought of as diffusing in a semi-infinite medium; (ii) the gas-oxide interface constitutes a permeable interface; and (iii) the transport of the oxidizing species in the near-surface region follows a diffusion-exchange mechanism, related to step-by-step motion of


77

~rilllillililiilliill -

-25

i

I I f 1

2'0

!

"-" 2-0 '7 E u

,,,,

l . , , ~, l , ~ m..~l , . . , , , 20

v

20 -~"

E ~.s

-

O

0

m

D

-

o

-

o

x

~

I0

0 Z -

o

o

"

00,

_"2 •

0 0 5

0-5 -

Z v

_

~ o Z

' '2~0015 I ' ' '

( N~ + N~) x 10Isatoms cm -z

la)

E m

-

E

-~o

_o

-

O

0 _u,

0

0-5

0

00

0

I

!

o o

.

t

l I

0

!

0

, , , ,I t 5 , , ~ 10I ,

,._.. ~-~

-

f

I#1

"0"

-

!

ti

....

5

o

T .... S

-

! ....

10

15

,x

z

0 20

( N ~ + N~) x 101Satoms cm -~

Ib~

FIG. 57. Areal densities of 180 (circles) and of 160 (squares) as a function of the total amount of O from etch-back NRA in silicon oxide films grown at 930~ in ultradry gas. (a) 30 min in 1602 followed by 30 min in 1802; (b) 330 min in 1602 followed by 30 min in 1802. The arrows indicate the gas-oxide interface. The steep dashed lines indicate enhanced 180 incorporation at the sample surface. Reprinted with permission from [161 ], 9 1986, Taylor and Francis.

network oxygen atoms [ 164]. The latter has also been inferred from previous twostep sequential oxidation experiments [ 165]. There is more than one model capable of incorporating the experimental evidence found so far conceming silicon oxidation [ 158]. One picture is the reactive layer model [166]. The idea is that the oxide nearest to the silicon is not fully oxidized, so that species (perhaps 02) that diffuse interstitially through the outer oxide react within this layer, before reaching the silicon itself. Diffusion continues by some other mechanism. The reactive layer is therefore opaque to interstitial 02. The reaction does not have to be at the outside (stoichiometric oxide side) of the reactive layer, but this side will receive the largest flux of interstitial oxidizing species. Figure 60 shows a pictorial sketch of the basic idea announced in the reactive layer model [ 167]. Based on the reactive layer model, Stoneham et al. [ 166] were able to explain the results given in Figure 57 and equivalent data, concluding that the thickness of the reactive layer is between 1.5 and 2.0 nm. From the point of view of experimental evidence and confirmation, the reactive layer model is solidly based on results conceming growth kinetics, structure, composition, and isotope-exchange data. One possibility according to the reactive layer model that has not been addressed in this chapter so far is the mobility of species involving Si. The

78

KRUG AND BAUMVOL 2500

--

i'"

~

'

,'

,

[

'",

~

-r

I'-

'

'' '

'"

9

.

oL_

/~?o

t5oo

u

,

100 ~

2000

o

'

o

s

to

pt.h

iszoz~

[nm]

I000

500

0

2500

5

0

'1'

F., [k~v]

zi

,

'

i

'

(b)

2000

'

~

'

'i'

I0

t

l'~

i

7

9

-- 9

2~ .... o. i

1500

epth o

u

[n

]

1000 500

....

0

--

l

5

z-

i

L

j

~

z. [keY]

.I

10

15

Excitation curves of the 180(p,ot)15N nuclear reaction around the resonance at 151 keV for a silicon oxide film grown in (a) 1602, 1802 sequence and (b) 1602, 1802, 1602 sequence. Corresponding profiles are shown in the insets. Reprinted with permission from [164], 9 1997, American Institute of Physics. FIG. 58.

long-range mobility of Si was rendered very improbable by the initial results on oxygen isotopic substitution. Indeed, direct evidence of Si immobility was given by Si isotopic substitution experiments. Results of a 31Si isotopic substitution experiment with a rather poor depth resolution (approximately 50 nm) [ 168] were

79


FIG. 59. Near-surface 180 profiles from Figure 58 and the predicted (calculated) profile to be com-

pared with Figure 58b. Reprinted with permission from [164], 9 1997, American Institute of Physics.

Oz / 9> /

:-C'//,4 Reactive Layer

Indtvtdual Atoms ~n the oxide move out from the c - S t

/

Stotch,ometr~c con|in uous r a n d o m network

,..SlO z " , ./ / /, , ' / / / / / .

New

Reactive

Volume ._____.._ c o r r e s p o n d i n g to an Si atom to be o x i d i s e d

S,O

Si / oxide i n t e r f a c e moves

_~nt_o__s, [_..c_o i n___- ~ .

I t

~

~ S iO

Oz

This region may c o n t a i n Si c o o r d i n a t e d to S6 atoms as well

Reactive One or Two monolayers of SiO

St

]

as oxygens

corresponding ~ to one S, atom lll=-s~!i

Area

ifriil!i t

FIG. 60. Pictorial sketch of the reactive layer model. A column of silicon atoms is shown both before and after oxidation by one oxygen molecule. Reprinted with permission from [167], 9 1986, Taylor and Francis.

80

KRUG AND BAUMVOL

confirmed by a 29Si isotopic substitution experiment of much better resolution (approximately 0.7 nm near the sample surface) [169]. The profiles of 29Si before and after thermal oxidation in 1802 and of 180 revealed that no Si is lost and that it is immobile during oxide growth, in the sense that it does not diffuse across the growing oxide to react with 02 at the gas-oxide interface. These results do not exclude short-range Si transport from the substrate into the near oxide-silicon interface, as hypothesized in the reactive layer model. It is important to note, however, that Si mobility is not at the basis of the model: "whilst it is convenient to talk of out-diffusion of silicon, it suffices if interstitial oxygen stops at the outside of the reactive layer. Even diffusion through the layer would give an equivalent effect provided it is a vacancy mechanism or one involving exchange" [ 167]. The overall atomic transport picture in the growth of ultrathin oxide films on silicon comes from the relative thicknesses of the different regions relevant to the process: (i) the thickness of the reactive oxide layer, approximately 2 nm from the oxide-silicon interface [ 166]; (ii) the thickness of the oxygen-excess region, approximately 3-4 nm from the gas-oxide interface [170, 171]; and (iii) the oxide film thickness. Experimental findings and models discussed above revealed the atomic transport mechanisms acting in the ultrathin regime. Oxygen either (i) diffuses through the oxide network without interaction (interstitially) to react within the reactive layer or (ii) follows a diffusion-exchange mechanism, related to step-by-step motion of network atoms, and is fixed at the oxygenexcess region. On the basis of atomic transport, the ultrathin oxide film regime is conveniently defined as that at which film thickness becomes comparable to the sum of the reactive layer and the oxygen-excess zone, or equivalently, that at which these two regions overlap. Oxygen diffusing step-by-step in the oxygen-excess zone can then reach the reactive layer to react therein. As a direct consequence, the contribution to film growth of oxygen incorporated near the surface increases. The identity of O-containing species diffusing during silicon oxide growth is now addressed. Atomic transport during thermal oxidation of Si in dry 02 or water vapor was first described by Deal and Grove [33] as steady-state interstitial diffusion of molecular oxygen (02) or water (H20) across the growing oxide and subsequent reaction with Si at a sharp SiO2-Si interface. This has been supported by both experiment and quantum chemical calculations [159]. However, recent experimental results have shown that atomic oxygen is also a possible candidate for the transported species, and it is now claimed that while traditional arguments for molecular oxygen being the transported species are valid for atomic oxygen as well, more recent experimental results support atomic oxygen as the transported species [ 172]. The new results concern (i) the dissociation rate of oxygen molecules at the SiO2 surface compared with the oxidation rate [ 173]; (ii) coupling between the dissociation rate and oxidation kinetics [ 174];


81

and (iii) oxygen exchange at the oxide-silicon interface [175]. As for the stepby-step motion of oxygen atoms in the near surface, peroxyl bridges ( O - O bonds, equivalent to an excess-oxygen interstitial) have been identified as probably being responsible [165]. This is in good agreement with experimental EPR data concerning defects associated with Si depletion or O excess centers (the so-called EX centers) [170, 171]. The depth distribution of EX centers is remarkably similar to that of oxygen found in the near-surface region of an oxide film after the second step of a sequential 1602, 1802 oxidation. Oxygen transport through peroxy bridge defects in silicon oxide has been explored with the use of molecular dynamics and Monte Carlo combined to first-principles calculations [154, 176]. 1.6.3. REMARKS AND LIMITATIONS CONCERNING ULTRATHIN SILICON OXIDE FILMS AS GATE DIELECTRICS Silicon dioxide has remained the gate dielectric of choice because it has close to ideal properties: its dielectric strength is large and the SiO2-Si interface contains very few defects. The scaling method with some modifications has succeeded in downsizing MOSFETs for 30 years, to the 0.18-#m technology node with gate lengths of 0.12 #m. By thinning of the gate oxide to less than 2 nm, various advantages in the MOSFET performance were confirmed. Further downscaling, however, is being simultaneously threatened by different parameters. Among them, gate SiO2 thinning is thought to be the most severe. First, the presence of large quantum mechanical tunneling current is a serious scaling limitation in terms of standby power consumption. Second, breakdown characteristics for ultrathin oxides become even more critical because of the dramatic increase in electric field across the oxide during normal device operation. Third, poly-Si gate depletion effects are known to get worse with oxide scaling, as operating gate voltage normally does not scale proportionally to oxide thickness. Furthermore, as gate oxide thickness decreases, process integration issues emerge as new challenges. Boron penetration from p+-polysilicon gates into the thin gate oxide and the channel region in p-MOSFETs is one of the major concerns for CMOS technologies. It is now well established that ULSI reliability and electrical properties are strongly dependent on the quality of the SiO2-Si interface region. Channel mobility, leakage current, time-dependent breakdown, and hot electron-induced effects have all been correlated with the oxide structure and defects at the SiO2-Si interface. Although the electrical defects are controlled by fabrication conditions, oxidation ambient, etc., relatively little is known about the atomic configuration of these defects, especially for ultrathin oxide films. The mechanism of breakdown of ultrathin dielectrics is also not fully understood. An atomic-scale description of silicon oxidation in the ultrathin oxide film regime is also still to be developed.

82

K R U G AND B A U M V O L

1.7. Silicon Oxynitride Gate Dielectric Films At this time, silicon oxynitrides (SiOxNy or, more accurately, nitrogen-doped SiO2) are the leading candidates for replacing pure SiO2 in ultrathin gate dielectric films [177]. Oxynitride films are of great interest because they retain favorable features of both silicon oxide and silicon nitride while minimizing their drawbacks [ 14]--one takes advantage of the passivating and masking properties of Si3N4 while retaining the excellent electrical properties of the SiO2-Si interface. The three main reasons for the attractiveness of silicon oxynitrides as a replacement for pure SiO2 are (i) very good diffusion barrier properties (particularly against boron penetration from p+-polysilicon gates); (ii) a slightly higher dielectric constant that is reflected in some reduction of leakage current; and (iii) enhanced reliability. Even small amounts of N (1 • 1014 cm -2 or more) in the SiO2 network significantly improve its diffusion barrier properties. The dielectric constant of oxynitrides linearly increases with N content from Esi02 = 3.9 to E S i 3 N 4 - - 7.8 [178]. At first, nitrogen was introduced in thicker gate oxide films [ 179-181 ], increasing their reliability. It was shown [56] that N incorporation results in a reduced defect generation rate (Section 1.4.1.2). Many ultrathin (800~ the actual nitriding species is NO, but in an environment containing N2, 02, and atomic oxygen, O. The first is essentially inert; 02 can promote some film growth, but at reduced rates if compared with the continued oxidation of SiO2/Si structures because of the inhibiting effect of nitrogen incorporated into the film; atomic oxygen, in turn, is responsible for the partial removal of nitrogen from the film, as stated above. It has been found [231 ] that in a heated conventional furnace N20 rapidly decomposes into about 60 mol% N2, 30 mol% 02, and 10 mol% NO. Rapid thermal oxynitridation with N20 leads to slightly different results, as the gas decomposes only upon reaching the heated sample surface [223]. The growth kinetics of thermal silicon oxynitride films on Si in NO are selflimited to 2.5 nm at any temperature below 1100~ At the initial stages, NO adsorbs dissociatively on Si(001) as well as in Si(111), forming one monolayer of Si3N4 at the dielectric-Si interface, followed by several monolayers of subnitrides (nitrogen-defective silicon nitrides) and, most probably, suboxides as well. Isotopic substitution studies clearly indicate that NO diffuses toward the oxynitride-Si interface. The presence of one Si3N4 monolayer or even a fraction of a monolayer at the interface largely prevents the migrating NO molecules from reacting with Si atoms of the substrate. So direct thermal growth of silicon oxynitride films on Si in NO proceeds within a very limited atomic transport scenario, because of the diffusion barrier properties of a layer with an appreciable concentration of N at and near the oxynitride-Si interface. Most of the activity consists of replacement of N originally incorporated by O, as well as completion of the suboxide/subnitride network. The degree of replacement depends on processing time, NO pressure, and film thickness as a whole. Silicon oxynitride films can also be produced by thermal nitridation of silicon oxide films. When the nitriding gas is NO, nitrogen is introduced only in the nearinterface region. Contrary to what is observed in the direct thermal growth of oxynitrides in NO, the nitrogen concentration at and near the interface increases with increasing annealing time. No appreciable film growth (thickness increase)

96

KRUG AND BAUMVOL

occurs. XPS analysis indicates that nitrogen is predominantly bonded to Si, as in stoichiometric Si3N4. Only a minor portion of the N atoms presents a S i - N - O bond structure. Nitridation of SiO2 films in N20 leads to similar results, except that significant film growth can occur. This should be due to the presence of O2 as a product of N20 decomposition, as discussed above. Isotopic substitution results [222] indicate that the nitridation of silicon oxide films in NO takes place by two atomic transport mechanisms occurring in parallel, as in the case of dry oxidation of silicon oxide films in O2: (i) NO diffuses through the silica network and reacts at the SiOxNy interface to fix both N and O (this mechanism involves, in fact, a minor fraction of the NO molecules entering the oxynitride network) and (ii) step-by-step motion of network oxygen atoms induced by the presence of network defects leads to incorporation of O in the near-surface. As (ii) involves only O, one can presume that N is released in the form of a nonreacting molecule, like N2. Figure 71 [232] shows SIMS depth profiles for silicon oxynitride samples grown in a rapid furnace at 1050~ for 1 min with the use of (a) N20 or (b) a mixture of NO and O2. Profiles are not reliable within the top 0.5 nm because of the presence of an adventitious carbon surface layer. The film grown in N20 shows a characteristic nitrogen pile-up at the interface with the Si substrate, whereas the other sample has the nitrogen uniformly distributed throughout the oxynitride layer. The latter is typical of oxynitridation with NO. As the samples were prepared by rapid thermal treatment, the result is explained by the decomposition of NeO being restricted to the sample surface, in small amounts, so that a continuous supply of O radicals was available during film growth. The nitrogen and oxygen profiles in a sample nitrided in NO and subsequently annealed in O2 at high pressure and low temperature are shown in Figure 72. This sample corresponds to the I - V characteristics shown in Figure 62. Processing parameters are detailed in the figure. One observes relatively high nitrogen incorporation in the film (>5 at.%) at low processing temperatures. Moreover, the N profile is shifted to the sample surface as compared with the usual feature for an SiOxNy film grown in NO. These are all desirable characteristics for a nitrided gate dielectric. Figure 73 presents in schematic form a sequence devised to produce the ideal nitrogen profile in a gate oxynitride film. It consists of an oxynitridation step of Si in NO followed by oxidation in 02 and repeated oxynitridation in NO. The first step results in an oxynitride film with a given concentration of N at the interface with Si. The second step promotes growth of an underlying oxide film, displacing the N distribution in the direction of the sample surface. The third step introduces N at the new dielectric-silicon interface. One expects to be able to tailor N concentration at the sample surface and interface through the annealing temperature in NO in the first and third steps. The successful result of such an approach is shown in Figure 74.

97


N20 grown oxide

.~

o

4"0 [lO00eV 'i3s"75'deg'rees/

I 100

3.5

4

~ 3.o

~"

o

'~ i

0---

2.0 0

o

1.5

40 o

o

~ 1.0 ~

2O

0.5

0.0

0

10 20 30 40 Depth

50 60

70

(Angstroms)

0 80 90 100

Oxide grown with NO-O~ mixture

,1000eV Cs 75 degree~///~"

4.0

....

100

;g 3.5 r

E o

II

Si 3.0

"

98 0 0

~

0 ---'-'-

.< E 0

C

_o 2.S

60

I ":" "~'" C

.

o 2.0

C 0

,

;

.,

f

~'_o ~IIZ 2

"...

I~ 0

i ~ 40

~ 1.5 C 0 0

e

u o

-9 1 , 0 2o

0.5

0.0

0

10

20

30

40

50 60

70

Depth (Angstroms)

80

90 100

FIG. 71. SIMS depth profiles for silicon oxynitride samples grown in a rapid furnace at 1050~ for 1 min with (a) N 2 0 or (b) a mixture of NO and 0 2. Reprinted with permission from [232], 9 1999, Materials Research Society.

98

KRUG AND BAUMVOL

.~ o 5

[ . ~__.+ 02' 800~

15min

g 4

l

8E3 ~ ~' 6E3 ~ ~

3

4E3

!

2E3

0

20

40 60 Depth (A)

80

0

~

&

FIG. 72. SIMS profile of high-pressure NO oxynitride with subsequent high-pressure 0 2 annealing.

Reprinted with permission from [205], 9 1999, Materials Research Society.

Oxynitridation

N

:looxidation

N I

Reoxynitridation

NO

.0 - 2.0 nm

FIG. 73. Depiction of the process flow for creating the ideal nitrogen profile with NO gas. Reprinted

with permission from [202], 9 1998, Kluwer Academic Publishers.

1.7.4.

HYPERTHERMAL AND D E P O S I T I O N M E T H O D S

Thermal nitridation of SiO2 in NO or N20 generally results in a relatively low concentration of nitrogen in the films, on the order of 1015 cm -2 N atoms [189, 202, 230, 233]. Because nitrogen content increases with temperature, thermal oxynitridation is typically performed at high temperatures (i.e., >800~ Other nitridation methods, such as with the use of energetic nitrogen particles (plasma, nitrogen atoms, or ions) [46, 206, 234-247], can be used. These nitridation methods can be performed at lower temperatures, ,~300-400~ However, low-temperature deposition methods may result in nonequilibrium films, and subsequent thermal processing steps are often required to improve film quality and minimize defects and induced damage [199, 248]. Because the thermodynamics [249, 250] of the SiOxNy system and the kinetics [184, 188, 194, 202, 203,

99

ULTRATHINGATE DIELECTRICFILMS

E (J

4 t NO/O2/NO

/'a--~Qt"~--~'"

interface positions

c:) --9 3 v

x r

o

~176o

Z

10 MV cm -1 . C - V characteristics of Al-gate capacitors and 6.5 nm-thick AleO3 films deposited on n-type Si without an intentional intermediate layer are shown in Figure 87. It can be seen that the Dit is rather low. The quasi-static C - V measurements indicate a small d.c. leakage for the larger voltages. By changing the ramp rate, it is verified that this leakage does not distort the quasi-static C - V in the relevant interval - 1 to 0 V. The high-frequency C - V was ramped from - 2 to + 2 V and back. Very little hysteresis was observed in this voltage range. Scanning to larger voltages indicated the occurrence of some electron trapping, evidenced by a flatband shift toward more positive voltages. Finally, I - V characteristics are shown in Figure 88 for 12.5 nm GdeO3 films directly deposited on Si(001) by reactive e-beam evaporation [48]. The as-deposited film is leaky, but a 10-min annealing in Oe at 500~ or 700~ resulted in a dramatic improvement. Estimates of the thickness of the intermediate SiOe layers formed after annealing in Oe obtained from C - V characteristics are shown in Figure 89 [48] along with the average dielectric constants of the films. Electrical characteristics of several other alternative dielectric films are becoming available in the literature, some of them constituting excellent candidates to replace SiOe. Many different illustrations of the effects of the above-mentioned -

-

111


2O

.

~.-1

~

-,

t

,

-r

1

-

'

" " ' ~

'"

}-

--T

~:--

~r

"

.

I

9 "1

"

1-

,

I

"

'

(a)

E

.-::_ oo 2z 10 kHz

15

. ~

"

,,z,4 "

, , # , -

r,,O

c,-

Eb o c-

_~ 10 0

5

-2

-1

0

1

2

Gale Voltage (V)

"E ~6

(b)

::L

~4

- . - loo k . z

.,,f--

t

s

m 10

o

-g

F

8

~,'

i

9

!

=

02 / 500~ -2

-1

0 1 Gate Voltage (V)

9 .-

! 10 rain 2

FIG. 85. C-V curves for thin HfSixOy and ZrSixOy films, with Au electrodes (A --- 1.76 x 10 -4 cm2). (a) 50-,~ Hf6Si29065 film on n+-Si, deposited at 500~ and subsequently annealed in forming gas at 450~ for 30 min. The Cmax/A value in accumulation yields EOT = 17.8 ,&. (b) 50 ,~ Zr4Si31065 film on n+-Si, deposited at 25~ and subsequently annealed in 02 at 600~ for 10 min. The capacitance density in accumulation yields EOT = 20.8 A. These films have some dispersion near zero bias, which indicates the presence of interface traps. Reprinted with permission from [265], 9 2000, American Institute of Physics.

112

KRUG AND BAUMVOL 1 0 .2

E

gr,..3l ~ ,-

!',r

,

I'

'

9

"

+--I

.

.

.

.

l

'

+

"

"

+ Au / HfsSi 90+5 / n + Si 2 10 .3 9 N2: H 2 / 450~ / 30 min 0 -4

I

.

.

.

.

---r-

9 '"~I

Accumulation

.=.-

1 0-5

"~ 10 G t_

o

:i

10.7 10"•

i 0 -2

/.

~

i

.

.

.

.

!

-2

F

'

I

.

-1

.

.

.

.

.

.

.

1

.

~

.

.

.

.

.

,

,


i

'

"

'-

9

I

.

.

.

.

I

1

'

9

9

'

I

1

,

,1

'

,r

2

L Au / Zr 4Si3~Oss / n + Si 1 0-3 E o

;~ 10 -4

.# = 10" 5 O

10-s L_

~ ,o, 10 8

iol ~

t

-2

\ .

.

.

.

I

-1

. . . .

'%.

I

.


2

FIG. 86. I - V curves for films shown in Figure 85. (a) Au/50 ,~ Hf6Si29065/n+-Si. (b) Au/50 ik Zr4Si31065/n+-Si. These films show extremely low leakage currents, which are below 2 x 10 - 6 A cm -2 at 1.0-V gate bias in accumulation. The I - V curves are well behaved and appear nearly symmetrical about zero bias. This suggests nearly equal barrier heights in the two polarities. Reprinted with permission from [265], 9 2000, American Institute of Physics.

aspects like deposition process, intermediate layer, postdeposition annealing, and gate electrode on the electrical characteristics are also being explored. In this section some illustrative examples were given, mainly indicating that physicochemical stability against thermal annealing must be fully understood before

113


0.8

9

'

"

1

I-

8.0nm AI203

i

,'"

on n-type Si (HF last) E

0.6

"-"

0.4

0 U_ ',e, 0

i

,

]

". . . .

~

0 f0

~o ~

0.2 -

1:3..

(.9

.............. 0.0

.......

;--7 . . . . . . . .

9

~

~ I--.

~

-1

~

~

Af gate capacitors: o QS-CV (dV/dt = 0.1 V/sec) . HF-CV (100 kHz)

I

,

..[

0

.

1

Gate bias (V) FIG. 87. Quasi-static (open symbols) and high-frequency (solid symbols) C - V characteristics of an AI/A1203/n-Si structure. Reprinted with permission from [266], 9 2000, American Institute of Physics. '

r

'

i

,

!

i

,

i

i

,

,~e-

'~ .....

eeee ol~176

o .,

10 -2

L~ "~

i

10~

..-..

9

. - - 9

eol"~

-

el,

J

eoe ~

1 0 -4

1 0 "6

C.~ 104 10

"10

Gate Potential (V) FIG. 88. I - V data for a 12.5-nm-thick Gd203 film annealed for 10 min in oxygen: solid line, 700~ dashed line, 500~ dotted line, as-deposited; circles, calculated Fowler-Nordheim current for ideal SiO2 layer 2.05 nm thick. Reprinted with permission from [48], 9 2001, The Electrochemical Society, Inc.

any of these alternative dielectrics can be incorporated into Si-based device technology. One word of caution: the strict requirements of breakdown lifetime will also apply to novel high-k gate dielectrics. Because the breakdown properties are very material-dependent, systematic studies of reliability of the novel high-k materials will be needed to evaluate their applicability. Despite the reduced leakage, some of the new materials may not pass the reliability (lifetime) criterion.

114

KRUG AND BAUMVOL 3.0

,• 2.5 c~ 9

2.0

o

1.5

.O

e,i

o'3

'

|

.

,

.

,

i

| , i (6.8/ (6.6)' ._._.._----. 9 .__----,

(7.0) 1"I ~

(7.3)

(7.5)

(8.6)

1.0-

-

o.si 0.0

. , . , .

.

.

.

.

450 5oo sso 6;o 6;o 760 7;o 8;o 8so Anneal Temp. (~

FIG. 89. C - V analysis estimate of the thickness of the SiO2 interface layer as a function of annealing temperature for 8-nm-thick Gd203 films with (circles) and without (squares) a previous 10-min vacuum anneal at 700~ The anneals were done in oxygen for 10 min. The average dielectric constants are shown in parentheses at the thickness data points. Reprinted with permission from [48], 9 2001, The Electrochemical Society, Inc.

1.8.2.

PHYSICOCHEMICAL

CHARACTERISTICS

The structure and composition of alternative, high-k dielectric films of potential interest as replacements for SiOe that have been investigated so far may have rather different characteristics: (i) they can be amorphous or epitaxial with the c-Si substrate. Amorphous films will most probably be initially used, whereas epitaxial films will constitute a further improvement in latter stages; (ii) stoichiometry and chemical bonds can be different, depending on deposition methods and parameters; (iii) the sharpness of the dielectric-Si interface is a critical aspect, similar to what happens in the case of the SiOe-Si interface; (iv) the as-deposited structures will be submitted to thermal annealing in vacuum, forming gas, or Oe, and the resulting structures and compositions will be a direct consequence of the physicochemical stability of the different thin film materials on Si; (v) atomic transport and chemical reaction may be inhibited by ultrathin, intermediate oxide, nitride, or oxynitride layers. They can be intentionally grown or deposited on the c-Si substrate before high-k film deposition or be present unintentionally, as a result of the formation of silicon oxide on the c-Si surface after cleaning or during high-k film deposition. A few examples are given here with the only aim of illustrating the variety of cases and the need for detailed investigation of the above-mentioned characteristics. Figure 90 shows high-resolution transmission electron microscopy (HRTEM) micrographs of TazOs films deposited by CVD at temperatures below 400~ [258]. Si substrates were cleaned in HF solution before deposition to remove any native oxide, and even so the as-deposited films showed a 2 nm interfacial region with stoichiometry close to SiO2 and 7 nm of amorphous Ta205 on top. Thermal


115

FIG. 90. High-resolution TEM micrographs of tantalum oxide and interfacial region after (a) a lowtemperature plasma anneal and (b) a rapid thermal anneal in oxygen at 800 ~C. Crystalline regions and growth of the interfacial region can be observed after the thermal anneal. Reprinted with permission from [258], 9 1998, American Institute of Physics.

annealings at and above 800~ induced thickening of the SiO2 layer and partial crystallization of the Ta205 layer. A1203 films deposited by atomic layer chemical vapor deposition (ALCVD) on HF-cleaned Si substrates display atomically sharp interfaces and no detectable intermediate SiO2 layer, as shown in Figure 91 [266]. When a 0.5 nm SiO2 layer is intentionally grown on the surface of Si before A1203 deposition, the A1 and Si profiles are as shown in Figure 92 [267]. Thermal annealing in isotopically enriched oxygen (1802) leads to incorporation of 180 and transport of A1 and Si as shown in Figure 92. Angle-resolved X-ray photoelectron spectroscopy (ARXPS) of Si 2p electrons indicates the presence of SiO2 in the near-interface region in the as-deposited samples, and the formation of Si-A1-O compounds in the nearsurface region after thermal annealing in O2, as shown in Figure 93.

116

KRUG AND BAUMVOL

FIG. 91. Cross-sectional HRTEM of an A1203 film deposited on HF-treated Si. Reprinted with per-

mission from [266], 9 2000, American Institute of Physics.

Figure 94 illustrates the dependence of the morphology of T i O 2 films deposited by CVD on the c-Si substrate temperature during deposition [268]. Although they are much thicker than the approximately 5 nm films that will be necessary for FET insulators, these thick films allow relatively easy electron microscopy of the grain structure. At the lowest substrate temperature (~ 170~ completely amorphous films were obtained (Fig. 94a). At 200~ anatase crystals in an amorphous background were observed (Fig. 94b). Increasing the temperature further led to rough anatase films (Fig. 94c). At 500~ the films became considerably smoother (Fig. 94d), with a columnar structure and a grain width of about 10 to 20 nm. Films grown at a substrate temperature above 600~ were primarily rutile. Finally, Figures 95 and 96 show HRTEM images of Hf and Zr silicate films, respectively, deposited on HF-cleaned Si by reactive sputtering or e-beam evaporation [265]. There are no visible intermediate layers, and the films are amorphous with sharp interfaces with the Si substrates. Hf 4f and Zr 3d XPS analyses [265] of these films are shown in Figure 97, indicating the presence of H f - O or Z r - O bonds and no H f - S i or Z r - S i bonds.

117


1 0 0 ~

200 >-

150

f / / %% ~t_\

~ / E ~oo I.'~ E ~- ,,~,? (.9

5O /

20

~

'~_

~,..~ 1.0

2.0

~.- 800~ 30 sJ -~ I , , 4.0 5.0 6.0

3.0

surface ~80 i ;~:> li

(D >" 800 r-.m

._.20

'

1

AI2OJSi02

~

! ; ,~ i' ~~e.~e,.~._

400 -

4 6 8 10 Depth (nm) I"-'E- before '8021 /"~'" 700~ 30 st

I"~ 7oo~ 60 sl

~'~#)-

0.0

v

>

o_ -0.2

p-Si (10 ,o.cm) HF treated

z, = 337 nm AtFWHM

=

5

ns

lexe= 0.5 mJ/cm 2

[

~ buffer

I~

scope

-0.3 -0.4 ,

I

0

,

,

,

,

I

1

,

,

,

,

I.

2

,

,

,

,

I

3

,

,

i

,

I

4

,

,

,

,

I

,

5

Time (gs) FIG. 8. Typical photovoltage transient and experimental set-up for transient PV measurements (laser pulse/semitransparent electrode/mica spacer/sample).

the experiment. The application of well-defined laser pulses (with an intensity in the range of some W/cm 2 and a duration time on the order of nanoseconds) allows large signal SPV measurements with excellent reproducibility. A further advantage of application of nanosecond laser pulses is that trapping processes are less important. Time-dependent preferential trapping, for example, has been shown at the porous Si/Si interface [ 151 ]. Figure 8 shows a typical PV transient. The pulsed PV is measured in a parallelplate capacitor arrangement (inset of Fig. 8). For ex situ measurements, the parallel-plate capacitor consists of a semitransparent front electrode, a thin mica spacer (thickness of the mica, some tens of micrometers), and the sample. The PV is measured with an oscilloscope via a resistance in the Gg2 range and a high impedance buffer. The maximum of the photovoltage is reached at the end of the laser pulse for n-type Si or a little bit later for p-type Si. The reason for this behavior is that the surface photovoltage and Dember voltage have opposite signs for p-type Si and therefore the PV amplitude can increase with decreasing concentration of excess charge carriers (6p) for high values of 6p. The PV amplitude (Upv) is measured just after the laser pulse has finished, and the band bending (Y0) can be obtained from Upv if 3p is known. Figure 9 shows the intensity dependence of the measured PV amplitude for n-Si with high resistivity (band bending is negligible because bulk and surface Fermi levels are both near midgap). The measured data are well fitted for the Dember voltage by Eq. (2.3), with b -- 3.5 and 3p = 2 x 1016 cm -3. The obtained value of b is in good agreement with values published in the literature [152]. The experimentally obtained value of 3p is characteristic for a given surface recombination velocity (So). The influence of So on the maximum value of 3p

148

RAPPICH AND DITTRICH

o

5 kocm

fit (parameters 5p and b)

150

/ o

5p = 2 1018cm -3

> v

n-Si(100)

measurement

E

100

/

b=3.5 ~

m

~

> 13..

50

.,#

j

= 902 am AtFWHM = 100 ns Io = 100 W / c m 2

. .~,...I

.

. ..

1 0 "6

.... I

.

. ......I

.

..

.....

I

.

.......

I

.

. .....

.I

.

..

....

.I

.

10 .5 10 .4 10 -a 10 -2 10 -1 10 0

I/I 0 FIG. 9. Intensity dependence of the measured photovoltage amplitude for n-Si with high resistivity

(open circles) and of the fitted Dember voltage (solid line). 1016

n-Si

10 W / c m 2 i

1015

1 W/cm 2

c') |

E

014 0.1 W / c m 2

~'•

1013 P C 1 D simulation" -~-~ (Ps = 0.3 V u,se: Zbu~k= 10 laS laser p

1012

1011

S B = 300 cm/s n = 1014 cm -3 -'..l.i|l

|

10 o

i

i

.....I

i

101

~

X = 902 nm

AtFWHM = 200 as .

. .....I

.

10 2

.

. .....I

.

10 3

.

.....I

.

10 4

.

. .....I

.

.

. 1...

10 5

S o (cm/s) FIG. 10. Dependence of the simulated excess carrier concentration at the surface of n-Si on the sur-

face recombination velocity under pulsed laser excitation for different intensities [ 153].

at the surface of n-Si (@(x=0)) is illustrated for different laser intensities (I0) in Figure 10 (duration time of the laser pulse, 200 ns). These simulations are performed with a one-dimensional model for pulsed laser excitation (PC1D [ 153]). The values of @(x=0) scale with I0. The value of &p decreases significantly for

149

ELECTROCHEMICAL PASSIVATION OF SI AND SIGE

n-Si q)s = 0.4 v

0.4

laser pulse: I. . . . 3 W/cm 2 X= 902 nm

0.3 >

PC1D simulation: Zbu,k= 10 ItS SB= 300 cm/s n = 10 TM cm3

AtFWHM= 2 0 0 ns

q)s = 0.2 v

v

> Q.

0.2

0.1

.0

CPs= 0.05 V

,,,,,,i

100

,

,

,,,,,,i

101

,

,

,,,,,,i

102

,

,

,,,,.,I

,

,

,,,,,,i

103

104

,

,

,,.,,,i

,

,

,,.,,

10 ~

So (cm/s) FIG. 11. Simulated photovoltage (PV) amplitude at the surface of n-Si as a function of the surface recombination velocityunder pulsed laser excitation for different surface potentials [153].

So > 104 cm/s. S0 can be expressed by cr 9v 9Ns, where ~r, v, and Ns are the surface recombination cross section (on the order of 10 -15 cm-3), the thermal velocity of excess carriers, and the concentration of surface recombination defects, respectively. The surface recombination becomes important for the determination of the maximum excess carrier concentration if Ns >__ 1012 cm -2. Figure 11 shows the simulated surface PV amplitude of n-Si as a function of the surface recombination velocity under pulsed laser excitation for different surface potentials (PC1D simulations [153]). 6p is larger by one order of magnitude than the equilibrium carrier concentration, which is usually the case for our PV experiments. Upv is larger than ~s for low values in So due to the Dember voltage, and Upv starts to decrease remarkably for So > 104 cm -3. The decrease in Upv at larger So does not seem to be significant. However, it is a serious source of error in determining the distribution of the surface state density (Dit). This source of errors can be slightly reduced by increasing the equilibrium carrier concentration, which leads to a reduction of the Dember voltage. But, in this case, the accuracy of the determination of 6p decreases, and the sensitivity of the SPV technique for measuring Dit decreases. The ability to measure Y0 with high accuracy by PV was used to obtain Dit ( E Ei), where Ei is the bulk Fermi level of the intrinsic semiconductor [37, 154]. For such a measurement, a field voltage (UF) is applied to the back contact of the Si sample for a certain period of time. The buffer is opened only during the PV measurement to avoid destruction of the buffer during switching of UF. After the

150


PV transient is recorded, a UF pulse of identical time and amplitude but opposite sign is applied to discharge slow states [38]. The field voltage influences a charge at the semitransparent counter-electrode (QG). The value of Qc is given by UF and the thickness of the mica spacer (Ci, insulator capacitance). The condition of charge neutrality of the system is Qsc + Qit + QG + Qfix = 0

(2.5)

where Qsc(Y0), Qit, and Qfix are the space charge, the charge in surface states, and the fixed charge, respectively. A variation in QG causes variations in Qsc and Qit. A change in Qsc means that the surface Fermi level changes and, therefore, so does the charge in the surface states. The variation of Qit with Y0 determines the surface state distribution (Dit), which is given by Dit --

1 dQit q dYo

(2.6)

Using Eq. (2.5) as d Qsc + d Qit -k- d QG = 0, dUF = dUi + dYo, and d QG = Ci 9d Ui, where Ui is the voltage drop across the mica spacer, the following expression for Dit can be derived [154]:

( )dUE d Q s_f ( 1Y o )+

q 9Dit -- Ci 9 dY0

dY0

(2.7)

Equation (2.7) contains values that can be determined only experimentally. The SPV method works well when the surface is in depletion or weak inversion. The accuracy for determining 6p and Y0 can be increased by the so-called doublepulse method when the intensity of the exciting laser pulse is changed and 11/12 = 6pl/6p2 is considered [ 155]. It should be noted that the component of PV induced by preferential trapping cannot be separated in this kind of experiment, but its influence on the measurement can be minimized with the use of short laser pulses in the nanosecond range. The minimal density of surface states ( D mi ti n) is reached near midgap for Si/SiO2 interfaces. These so-called midgap states are fast traps with large capture cross section for both electrons and holes. The density of midgap states has to be minimized for electronic applications. The sensitivity of the SPV method to/-).min "'It (A/Dmin] is given by analysis of the turning point in the Upv(UF) dependence. " " it J ADit in is limited by second term in Eq. (2.7). A serious source of experimenA/)min tal errors is the imperfect homogeneity of the mica spacer. Values for "--"-'it of about 101~ and 109 eV -1 cm -2 can be reached for dopant concentrations of 1015 and 1013 cm -3, respectively. The absolute values of fixed charges and of charged surface states cannot be obtained easily. The calculation of Qit by integration over the acceptor and donor states demands a detailed knowledge of the distribution and character of involved defects. This renders an accurate determination of Qfix. For most applications a

ELECTROCHEMICALPASSIVATIONOF SI AND SIGE

151

FIG. 12. Experimental setup for in situ SPV measurements during electrochemical treatment of semiconductor surface processing. Reprinted with the permission of the American Institute of Physics [36],

copyright 2001.

knowledge of Oit is sufficient. Changes in Qfix can be measured on the basis of shifts of the Upv (UF) characteristics along the UF axis. The experimental situation changes for in situ SPV measurements during electrochemical treatments. A typical experimental setup is shown in Figure 12 for PV measurements during electrochemical treatments. The band bending in the semiconductor can usually be obtained from the PV amplitude. The capacitance of the Helmholtz layer at the semiconductor/electrolyte interface is much larger than the space charge capacitance of the semiconductor. This makes a SPV analysis according to Eq. (2.7) impossible. Figure 13 shows typical PV transients of p-Si in (NH4)2SO4 at 0 V (thick solid line) and - 8 V (thin solid line) during potentiostatic treatment plotted on a longer time scale. The PV transient at - 8 V is very different from that measured at 0 V or the PV transients measured ex situ. First, the PV amplitude (not shown) is much higher than the band gap, and, second, the PV transient exhibit damped oscillations. The value of q - Upv may be much larger than the band gap because the applied potential drops across the semiconductor. The oscillations in the PV transient at - 8 V are caused by the time constant of the potentiostatic control.

152


p-Si(100) 0.0

~

-0.5 [-

t

I

-~.01-

[[

II

II II

IIII IIII II II II# v

0

applied potential (Ag/AgCI):

u=ov

laser pulse:

~= 902 am_

AIFWHM=I00 ns

W = 150 W/cm 2 10

Time (las)

20

FIG. 13. Typical in situ measured PV transients during potentiostatic control of p-Si in (NH4)2SO4

at 0 V (thick solid line) and - 8 V (thin solid line). Reprinted with the permission of the American Institute of Physics [36], copyright 2001.

Therefore, the internal amplifier of the potentiostat should be slow enough if high Upv is measured with high accuracy. 2.2.3. PULSED PHOTOLUMINESCENCE Crystalline silicon (c-Si) is an indirect semiconductor. The forbidden band gap of c-Si is 1.1 eV. The rate of radiative interband recombination is very low for indirect semiconductors because phonons are involved during the transition process. Therefore, the radiative recombination lifetime is very large (more than 10 ms). Nonradiative recombination processes such as Shockley-Read-Hall (SRH) recombination are usually much faster. For this reason, the efficiency of the interband luminescence at room temperature is very low because of the high efficiency of nonradiative recombination processes. In other words, the recombination is dominated by nonradiative bulk and/or surface recombination processes. The radiative interband recombination can be measured by PL techniques. The quenching of the PL signal contains information about nonradiative recombination. This circumstance has been used to monitor the change in nonradiative surface recombination during surface treatments and processing of c-Si by considering no change in the bulk lifetime. The interband PL of c-Si can be excited with pulsed or cw lasers. The measured PL intensity is rather low for c-Si at room temperature. A high excitation intensity or extreme cooling of the Si sample is required to increase the PL intensity. The main disadvantages of PL excitation by cw lasers are (i) sample heating

153

ELECTROCHEMICAL PASSIVATION OF SI AND S I G E

(a) (b) (c) (d) (e)

absorption radiative band-to-band recombination (PL, ~ 6n 2) non-radiative surface recombination (~ 6n) non-radiative bulk recombination (~ 6n) non-radiative Auger-recombination (~ 6n3) l

l

**ooO. ,, l

l

l

l

".

I

r

i

I I

Ec EF(n)

9

I

(b),

, I I

\Oo - . . . . .

%'" +

+

,,I-

i "" I

,.~

&

. !(d)

(e)

9

+

+

EF(p) Ev

FIG. 14. Overview of elementary processes at a semiconductor surface under strong illumination.

for high excitation levels, (ii) distortion of the electrochemical process by the high amount of excess carriers, and (iii) the unsuitability of cooling below about - 10~ for electrochemical processing. These disadvantages are eliminated by excitation with laser pulses in the nanosecond range because of high excess carrier concentration for a very short period of time. The PL intensity can be measured by excitation with a single laser pulse, and it can be used very nicely for in situ stroboscopic probing of a c-Si surface during electrochemical processing. Figure 14 gives an overview of elementary processes at a semiconductor surface under strong illumination (~n >> n, p). The relevant processes are carrier diffusion, Auger recombination, nonradiative surface and bulk SRH recombination, and bimolecular radiative recombination. The efficiency of the radiative interband recombination is proportional to the product of the excess electron (~n) and hole concentrations (@), whereas the efficiency of the SRH nonradiative recombination is proportional to the excess electron or hole concentration. Therefore, the PL intensity increases much more strongly with increasing excitation intensity (W) than the nonradiative SRH recombination. The efficiency of the Auger recombination is proportional to ~ne,~p or ,~peSn, and nonradiative Auger recombination limits the PL intensity at high values of W. A typical setup for in situ PL measurements during electrochemical processing of Si surfaces is shown in Figure 15. The sample is placed in a quartz tube, and the electrolyte is pumped continuously through the tube. The Si sample serves as a working-electrode and a Pt wire as a counter-electrode. The reference-electrode is

154


FIG. 15. Experimentalset-up for in situ photoluminescence (PL) measurementsduring electrochemical treatment, after [35]. Reprinted with the permission of the Electrochemical Society, copyright 1997.

a Calomel electrode. The PL is excited with nitrogen lasers (wavelength, 337 nm; duration time of the laser pulse, AtFWHM -- 0.5 or 5 ns; W up to 10 mJ/cm2). The intensity of the N2 laser is changed over several orders of magnitude with glass plates as filters. The laser beam is slightly focused on the sample with a quartz lens (spot diameter about 3-4 mm). The light of the radiative interband recombination is collected using a lens with a short focal length and large diameter. A quartz prism monochromator is used to select the light at a wavelength of about 1.1/zm. Integrating InGaAs photodetectors with a high impedance preamplifier (EMM, integration time on the order of 10 ms) and Si avalanche photodiodes with fast amplifiers (EMM, time resolution 3 ns) are used for the detection of the integrated and transient PL signals, respectively. The oscilloscope is triggered with a photodiode by scattered light from the N2 laser. Sometimes a filter made of silicon is used as an optical band-pass filter instead of the monochromator for the light of the radiative interband recombination of c-Si. The duty cycle of the stroboscopic measurements is, with respect to the lifetimes of excess carriers, on the order of 10 -5. Therefore, electrochemical processes at the anodic oxide/p-Si interface are not remarkably influenced by the PL and PV measurements. The inset of Figure 16 presents the PL spectrum of c-Si at room temperature. PL transients are measured for photon energies at which the PL intensity has a maximum (about 1.1 eV). Figure 16 shows typical PL transients for c-Si at different excitation levels (N2 laser pulses, wavelength 337 nm, AtFWHM = 0.5 ns). The absorption of ultraviolet light is very strong in c-Si, and excess carriers diffuse from the near-surface region into the bulk. The fast decay of the PL intensity

155

ELECTROCHEMICAL PASSIVATION OF SI AND SIGE -4

c-Si

101

"~B=I ms

=>" r e-

100 v

d

.c_ d

10 -1

o_ 0.9

1.1

(eV)

.m

c c J

13_

1 0 .2

1.3

10 .3 1 0 .4 10 -s

[ i

0

,

I

20

,

I

,

40

I

,

I

60

,

80

I

,

100

I

120

Time (ps) FIG. 16. Measured PL transients for c-Si at different excitation levels (N2 laser pulses, wavelength 337 nm, AtFWHM -- 0.5 ns). The inset presents the PL spectrum for c-Si at room temperature. The dashed line shows the decay of a PL transient for a PL lifetime of 25 #s. Reprinted with the permission of the American Institute of Physics [ 156], copyright 1999.

at the shorter times is given by the fast reduction of the excess carrier concentration an due to diffusion and Auger recombination. The decay of the PL intensity at the longer time is given by the so-called PL lifetime, which is a combined lifetime for surface and bulk recombination. The PL lifetime is about 25/zs for the PL transients (shown in Fig. 16), regardless of the excitation level. This fact is crucial to the calibration of the pulsed PL technique. PL transients are simulated with a simple diffusion model in which band bending is not taken into account [ 156]. For high excitation levels, the values of excess holes and electrons can be considered as equal (an = @), and the bands are flat at the semiconductor surface. The one-dimensional kinetic equation describing the excess carrier concentration can be written as [ 157] Oan at

026n

= D "-2

Ox

q- G ( x ,

an t) -

~

rB

-

flan 2 -

van 3

(2.8)

where D is the ambipolar diffusion coefficient (15 cm2/s for c-Si), G ( x , t ) is the generation rate of nonequilibrium carriers, rB is the carrier lifetime in the bulk,/3 is the coefficient of interband radiative recombination (3 x 10 -15 cm3/s for c-Si), and 9/is the Auger recombination coefficient (2 x 10 -3~ cm6/s) [158].

156


The generation rate G ( x , t) can be expressed for pulsed laser excitation as Wot(1-R)

[

(t-t0)2

]

(2.9)

G ( x , t) -- hvp(AtFWHM/2)~/_ ~ exp --otx -- (AtFWHM/2) 2

where W , hvp, R (0.6 for hvp - 3.7 eV), ot (106 cm -1 for hvp - 3.7 eV), AtFWr~M, and to are the total energy density, the photon energy of the laser pulse, the reflection coefficient, the absorption coefficient, the laser pulse duration time (full width at half-maximum), and the time needed to reach the maximum light intensity, respectively. The boundary conditions are 06n _ Sf[3n(0, t) - 6no] -0x (x=0) D

(2.10a)

06n = Sb [3n(d, t) - 6no] Ox (x=d) D

(2.lOb)

where d is the thickness of the sample (0.4 mm for the calculations), 3no is the equilibrium carrier concentration (1014 cm -3 for the calculations), and Sf and Sb are the surface recombination velocities at the front and back surfaces of the c-Si sample. The surface recombination velocity depends on the concentration of surface nonradiative defects (Ns), their recombination cross section (o'), and the thermal velocity of the excess carriers (v, about 107 cm/s at room temperature): (2.11)

S f - - Sb = cr v N s

The recombination cross section (or) may change over several orders of magnitude, but cr is on the order of 10 -15 cm 2 for highly efficient recombination active centers [48]. Therefore, S is on the order of 100 cm/s for Ns = 10 l~ cm -2. For comparison, unusually low surface recombination velocities below 1 cm/s could be obtained for advanced c-Si (bulk lifetime larger than 100 ms) in HF [48]; i.e., Ns is below 108 cm -2 in this case. The transient and integrated PL intensities are given by

f

d

IPL (t) -- fl

6n 2 (x,

t)dx

(2.12a)

0 tm

pint L _

f 0

where tm >> rB.

IpL(t)dt

(2.12b)

157


p-Si(100),10

nm thermal

oxide

N s = 2 10 l~ c m 2

I ~

100

n-Si(100),

native oxide N s = 8 10

TM

c m -2

I

r

10 1

I

simulations

I

>,,

I I

03

t-

1

9

0_ 2

I ! I

t._1 13.

I I

1 0 .3

m

laser pulse

i 104

;L = 3 3 7 n m

I' 9

1

W = 1 mJ/cm 2

I

,,,,,I

I

1 0 .8

,

,

, ,,,,,I

,

,

1 0 -7

, ,,,,,I

1 0 .8

,

,

, ,

'

I

,

,

......

1 0 .5

I

1 0 .4

II

, ....

I

1 0 .3

Time (s) FIG. 17. Measured and simulated PL transients for Si surfaces covered with a thermal or native oxide.

The quantum yield of PL can be calculated as r]pL - -

pint L

(W/hvp)(1 -

R)

(2.13)

Measured and simulated PL transients on a log-log scale are shown in Figure 17 for Si surfaces covered with a thermal or native oxide. The bandwidth of the amplifier was 15 MHz for the measurements shown. The bulk lifetime was 0.9 ms for these samples. The first decrease in the PL intensity is mostly related to nonradiative Auger recombination. Diffusion of excess carriers dominates the decay of the PL intensity in the range between 100 ns and 20/zs, and nonradiative surface recombination limits the PL intensity at longer times for the wellpassivated sample (thermal oxide, 10 nm thick). The PL lifetime is about 40 #s for this sample. An excellent fit can be obtained for the sample covered with the thermal oxide (Ns = 2 x 10 l~ cm-2). The regions where diffusion or nonradiative recombination of excess carriers dominates are not well distinguished for the sample covered with the native oxide. Nevertheless, the PL transient is well fitted for Ns = 8 x 1011 cm -2. Thus, the absolute values of Ns can be obtained from PL transients if cr = 10-15 cm 2. The measurement of the integrated PL intensity is needed for stroboscopic in situ PL investigations of Si surfaces during electrochemical treatments. A theoretical analysis of the dependence of "pL lint and r/pL on W was made with the use of Eqs. (2.12b) and (2.13) for different rB and Ns and for AtFWHM in the nanosecond range. The PL intensity is proportional to W 2 for W up to 1 mJ/cm 2 and saturates for higher W because of increasing influence of Auger recombination. The PL el-

158


~(~., ,, "......... "'"-.. Islope -1/,

101 10 0 ~-~ 10 -1

10

~

..

W = 1 mJ/cm 2

.

...1 o-1 0_ 2

10

-3

10 .4 ,,

1|,|,I

10 8

........

i

10 9

........

i

........

101~

i

1011

........

i

........

1012

N~(cma)

i

........

1013

i

,

1014

FIG. 18. Dependence of the PL efficiency on Ns for W = 1 mJ/cm 2 and different values of rB. Reprinted with the permission of the American Institute of Physics [ 156], copyright 1999.

ficiency increases up to W of about 1-2 mJ/cm e and decreases for higher W. The excess carrier concentration is about 1017 to 1018 cm -3 for W of about 1 mJ/cm e. The value of the excess carrier concentration at 1 mJ/cm: was also confirmed experimentally by PV measurements. The PL efficiency can reach values in the range of 1% for low Ns, high rB, and optimized conditions of excitation (W in the mJ/cm e range for AtFwi-irvl in the nanosecond range). The dependence of the PL efficiency on Ns is shown in Figure 18 for W = 1 mJ/cm 2 for different values of rB. The sensitivity of PL efficiency to changes in Ns (Ns sensitivity) is limited by the bulk carrier lifetime to lower values of Ns. The Ns sensitivity is in the range of 1011 cm -2 for rB - 10/zs and can be improved to less than 108 cm -2 for rB -- 10 ms. From an experimental point of view, Si wafers with bulk lifetimes larger than 100/zs are needed to detect changes in Ns with a resolution better than 101~ cm -e. If rB > 100/zs, the PL efficiency is practically proportional to N s 1 for Ns > 1011 cm -2. The dependence of the integrated PL intensity of Si/SiOe samples on Dit is shown in Figure 19 for different excitation levels. The values of Dit are obtained by conventional capacitance/voltage (CV) measurements. As remarked, for thermally oxidized c-Si the value of Dit in the minimum corresponds quite well to Ns, because intrinsically back bonded Si dangling bonds act as rechargeable and recombination centers, which dominate the interface state distribution in the range near midgap. The solid circle in Figure 19 denotes an oxidized Si sample for which no CV data but the PL intensity could be obtained. The integrated PL intensity is proportional to Dff 1 regardless of W. Therefore, the measurement of Ns

ELECTROCHEMICAL

159

P A S S I V A T I O N OF S I A N D S I G E

10 5

>,.,

or) C

10 4

03

-..

c-J &. "O

10 2

~

101

W (mJ/cm 2)

1:o=1 ms (3_...

-~+"~.... .

1011

.

.

"'-

~-~,.

,...., c

10 0 .......

.

.

.

.

.

2

)J(

0.2

-I- 0.o5

"'x~-O "--.~. "'r - .++.+

0

I

-

~-. .

.

1012

.

)16 .

.

.

.

.

I

,

101~

Dit (eV-lcm -2) Dependenceof the integratedPL intensity of Si/SiO2 samples on Dit for different excitation levels. The values of Dit were obtained by conventional capacitance/voltage measurements.Reprinted with the permission of the American Institute of Physics [156], copyright 1999.

FIG. 19.

with the use of stroboscopic PL excitation can be calibrated by only one set of PL and CV measurements of a Si sample (Figs. 18 and 19). This makes in situ PL measurements very manageable.

2.3. Electrochemically Hydrogenated Si Surfaces Hydrogenation of Si surfaces takes place whenever an oxide layer on Si is etched back by an HF-containing solution. The formation of hydrogenated Si surfaces is one of the most important steps in device manufacturing. HF dip or buffered NH4F treatments produce different kinds of surface morphology (i.e., rough or smooth), which is of interest for further processing (deposition, oxide growth, etc.). Four types of hydrogenated Si surfaces can be distinguished: (i) HF dip (a partial step in the RCA clean [99]), (ii) treatment in buffered fluoride solutions [59], (iii) electrochemical hydrogenation in diluted fluoride solutions [73, 85], and (iv) formation of porous Si (por-Si) in fluoride solutions [ 159]. The hydrogenation of these surfaces has been investigated by FTIR [48, 52, 53, 55, 59, 85, 147, 160] and high-resolution electron loss spectroscopy [51, 58, 67, 73, 161-164]. In this section, we show that electrochemically prepared, microscopically rough Si surfaces have very low defect concentrations, which will be interpreted by a special kind of reconstruction of step facets. Furthermore, we show some results concerning the stability of such surfaces in the presence of oxidizing agents, acidic HF or alkaline solutions.

160

RAPPICHANDDITTRICH

2.3.1. ELECTROCHEMICAL HYDROGENATION IN DILUTED HF SOLUTIONS The ideally hydrogenated Si(111) surface consists of atomically flat and unreconstructed facets [42, 62] and is covered by Si-H bonds [53]. Such surfaces can be prepared in concentrated buffered NH4F solution (40%; pH 7.8) [42, 59]. The formation of fiat surfaces on Si is important, for example, for heteroepitaxial deposition of other semiconductor material on Si or for the formation of thin gate oxides. Steps and kink sites open the pathway for leakage currents through an enhanced electric field at tips, which, at least, decreases the breakdown voltage. Furthermore, hydrogenated Si surfaces are free of surface states that act as recombination centers. A well-controlled way to monitor the hydrogenation process can be performed by means of electrochemical processing. A current transient occurs at the end of the oxide dissolution in diluted HF solutions for n- and p-type material in the dark [85, 86, 164-168]. Figure 20 shows (photo)current-potential scans of n- and p-type silicon in 0.1 M and 0.2 M NHnF (pH 4) under illumination and in the dark. At the first strong increase in the current, Si is oxidized to a divalent state and dissolves into the electrolyte, leading to a rough and at least porous structure. At higher anodic potentials, four positive charges (denoted by h + (hole)), are consumed for the overall oxidation reaction [7] according to Si 4- 2H20 4- 4 h + --+ SiO2 4- 4H +

(3.1)

//

a) n-Si(111 ), in the dark b) n-Si(111 ), white light c) p-Si(lO0)

NH4F pH 4

o v

1

0

0

1

2

3

4

5

10

15

Potential (Vsc E) FIG. 20. Current-voltagecurves of n-type Si(111) in the dark (a) and under white light illumination in 0.1M or0.2 M NH4F (b) and of p-type Si(100) in 0.1M NH4F in the dark (c).

161


Therefore, n-type silicon needs illumination to ensure hole generation (electrons are majority carriers) by the incident light (curve b), whereas the hole concentration is high enough for p-Si (curve c), and no illumination is required (holes are majority carriers), as can be seen in Figure 20. There is no reaction (no current) of n-Si in the dark at anodic potentials (curve a). The Si surface is covered with oxide at potentials above the first current peak, and the thickness of the oxide layer depends on the applied potential [ 166, 168, 169]. Current oscillations occur when silicon is polarized around +6 to + 12 V. These oscillations are damped at higher anodic potentials [ 168, 170]. While the oxide is being formed, it is simultaneously dissolved by small amounts of HF present in the electrolyte, leading to the well-known electropolishing behavior in such solutions [81, 171,172]. Chemical etching of silicon oxide occurs according to the following reactions: SiO2 4- 6HF --+ SiF 2- 4- H20 4- 2H +

(3.2)

SiO2 4- 3HF 2 --+ SiF 2- 4- H20 4- O H -

(3.3)

The etch rate for SiO2 in fluoride solution is given by ke

-

-

(3.4)

a[HF] + b[HF 2] + c

where a and b contain activation energy-type terms, a = 2.5, b = 9.7, and c -- 0 . 1 4 [173]. It should be noted that etching by HF 2 dominates at pH values above 2.5, because b is 4 times higher than a. The HF and HF 2 concentrations can be calculated as follows. The solution contains F - and H + ions, which are coupled via dissociation reactions [ 174, 175], kl[HF] -- H+][F-], k2[HF 2] -- [HF][F-],

kl = 1.3 • 10 -3

and

(3.5)

k2 -- 0.104

(3.6)

[F] = 2[HF 2] + [HF] + [F-]

(3.7)

The total concentration of fluor is given by

The formation of an HF dimer, (HF)2, is discussed for fluoride concentrations higher than 1 M [176], so that Eqs. (3.5), (3.6), and (3.7) are no longer valid at higher fluoride concentrations. Figure 21 shows the pH dependence of the concentrations of F - , HE and HF 2 for 0.1 M total concentration of fluoride as calculated from Eqs. (3.5) to (3.7). HF and F - are the main components at lower and higher pH values, respectively. The concentration of HF 2 has a maximum at a pH of about 3 and diminishes at pH values below 1 and above 5, where only HF and F exist, respectively. A solution consisting of 0.1 M NH4F at pH 4 contains about 0.014% HE Figure 22 shows the behavior of the dark current transient during the etchback process of an anodic oxide formed at +3 V. The dark current transients are

162


0.10 O

E

0.08

E

._o 0.06 .i-, ..i-, c-

O tO

o

0.04

__•0.1 M NH4F iiii

I / ~ " -. .- --

.......

F-

~,__r:'f" __. . . . . .

HF HF 2"

0.02 0.00

0

i

4

7

pH FIG. 21. Dependence of F - , HE and HF 2 concentration on pH for 0.1 M total concentration of fluorine as calculated from Eqs. (3.5) to (3.7).

900

n-Si(111 ) 0.2 M NaF (pH 4.5)

= I .9 g l

/~ /,-~,.

o I

~

.Q

E ~j

light on +3 V 300

-o.6v

------ -0.9V

...2"

600

--I

"-

+0.5v

m ~

light off ~

2000 2100 2200 2300

Wavenumber (cm 1)

1

FTIR (reference) I

5O s

-0.9 V FTIR

I

Time FIG. 22. Current during the etch-back process of an anodic oxide in 0.2 M NaF (pH 4.5) monitored at different potentials (a, b, and c: +0.5 V, - 0 . 6 V, and - 0 . 9 V) after switching off the light is switched off. The anodic oxide is formed at +3 V under illumination. The inset shows the respective IR spectra after the dark current transient has leveled out.

monitored at different potentials after the light is switched off. At +0.5 V the wellknown current transient with the typical current peak is obtained. At a potential near the flatband potential of about - 0 . 6 V the dark current becomes slightly cathodic after a short period, and only a very small current peak can be observed. The dark current finally decreases to a constant negative value. The current peak disappears when a stronger cathodic potential ( - 0 . 9 V) is applied, and the dark current rapidly decreases to a constant value of about - 5 0 / x A / c m e. The inset of Figure 22 shows the IR spectra obtained after the dark current transients have

163

ELECTROCHEMICAL PASSIVATIONOF SI AND SIGE

rel. change of IR-absorption ~-E 2300o.9.. ..~ 2200E 2100- S i g H 20001900-"" E

40-

;~ =I.

20-

4

o

5"

-_

O

.m

0

n-Si(111) / 0.1M NH4F (pH 4.4 )

0

'

'

'

I

500

'

'

'

'

o'oo

Time (s)

'

'

'

'

o

og. 5'oo =

FIG. 23. Time dependence of the current (bottom) during the etch-back process of an anodic oxide in 0.1 M NH4F (pH 4.4), and the relative change in the IR absorption in the Si-H stretching mode region measured at different times during the decay of the current (spectra are normalized to the oxidized surface). The open circles denote the integrated IR absorption as calculated from the FTIR spectra.

reached constant values. The spectra are normalized with reference to the Sillfree and oxidized surface measured at +3 V under illumination. The IR spectra are quite similar to each other and show no distinguishing features. This means that the shape (slightly anodic or slightly cathodic) of the dark current transient has no influence on the hydrogenation process. Figure 23 shows a series of IR spectra (top) recorded during the current transient (bottom) as a function of time. The spectra are normalized to the oxidized surface. There are no Si-H species detectable up to the maximum of the current transient (the detection limit is about 5-10% of 1 ML Si-H). The hydrogenation process starts after the maximum of the dark current transient. When the current begins to decay, the hydrogenation sets in, and the IR absorption in the range of the Si-H stretching modes increases and saturates when the dark current transient levels out. This is demonstrated by the integrated IR absorption (2000-2200 cm -1), which is plotted at the bottom of Figure 23 (open circles). The H-termination is preserved over long etching periods. In addition, X-ray photoelectron measurements recorded when the current transient has decayed reveal an oxygen, fluorine, and carbon content below a tenth of a monolayer [86]. The current transient can be monitored in a high range of variation of anodic potentials at n-Si electrodes [ 166], and it was shown that the charge flow during the current transient increases slightly with increasing anodic potential at n-Si. The situation is different for p-type silicon, where holes are the majority carriers, and anodic oxidation occurs without illumination. Nevertheless, a narrow potential regime

164


Si-H l 1Si=H2 n-Si(111) 0.2 M NaF

80

••'60 IIA

.=o g '~ a::_~

pH 3

._ 40

II II

II/I

pH 4

~ , p H

..., ~....'

/ ' pH4.7~

ooo

~~,.d.~,,.~. H_5.~ ,oo

0

ado

2200

4oo

Wavenumber(cm")

H I I ] \ afteranodicoxidationat +10 V IIH / \ underillumination /" /L~---- ~ 0.1 M K-hydrogenphthay

20

pH3~ 419

0'00

pH 5.3 " "

"

5'00

Time (s) FIG. 24. Current during the etching process of an oxide-covered n-Si(111) surface in 0.2 M NaF at different pH values as a function of time (U = +0.5 V). The inset shows the respective IR spectra of the n-Si(111) surfaces (reference at pH 5.3nnonhydrogenated surface). The baselines of the spectra are shifted for better visualization.

exists, located near the flatband potential between - 0 . 4 and - 0 . 6 V (see Fig. 20), to control the hydrogenation process and to protect the p-Si against oxidation reactions [35, 167, 168]. Therefore, the process of H-termination on p-Si surfaces can also be well monitored by measuring the time behavior of the current. The anodic (or dark) current transient does not only depend on the potential, but also on the oxide thickness and the etch rate of the oxide, which is given by the pH of the solution used. Figure 24 shows the current transient in 0.2 M NH4F for different pH values measured at +0.5 V. The anodic oxide is prepared in a 0.1 M solution of potassium hydrogenphthalate (pH 4) up to 4-10 V. The resulting thickness of the oxide layer is about 80/k [ 164]. The transient occurs later in time with increasing pH (from 3 to 5.3), and the charge that passes the electrode increases. Surprisingly, the current did not decay at pH 5.3; moreover, it remains at a high level, and no hydrogenation takes place, as measured by FTIR spectroscopy [85]. The relative change in the IR absorption with respect to the Si-H-free surface is plotted in the inset of Figure 24 for different pH values of the electrolyte. The FTIR spectra are measured at a time that is two times larger than the transient width, to create comparable conditions for the experiments. The IR spectrum of the oxidized and 0.2 M NaF (pH 5.3) etched silicon surface serves as our Si-H-free reference spectrum to eliminate the influence of the electrolyte on the IR absorption. A shoulder can be seen in the high-energy part of the spectrum at pH 3,

ELECTROCHEMICAL

PASSIVATION

165

OF SI AND SIGE

which is attributed to the stretching mode of Si=H2 species. With increasing pH this shoulder diminishes. This result is in good agreement with the fact that steps on a Si(111) surface have one or two dangling bonds, depending on the step orientation. The smallest amount of Si=H2 was found after the two-step procedure. The treatment at pH 4 with a subsequent etching step at pH 4.9 leads to about 90-100% of a monolayer of hydrogen on the Si(111) surface [64, 76]. But there is still a slight asymmetry that points to a very small IR absorption due to Si=H2. At pH 5, only a very small concentration of hydrogen silicon bonds (about 25% of spectrum d) exists on the surface. The amount of hydrogen-silicon bonds at electrochemically hydrogenated Si surfaces depends very little on the way the oxide is formed before the etching back process. The anodic oxidation can be carried out with and without current oscillations. In the following, two different treatments are used: (i) anodic oxidation without current oscillation and subsequent etching back of the oxide layer (noOsc-surface) and (ii) anodic oxidation with current oscillation and subsequent etching back of the oxide layer (Osc-surface). The anodic oxidation and etching back of the oxide is performed in the same solution. The etch-back process is monitored by the current transient, and the hydrogenation is completed when the transient levels out. Figure 25 compares in situ FTIR spectra of the different treatments, process (i) and (ii), applied with 0.1 M NH4F (pH 4). The anodic oxidation is performed at

Si-H

o

.i

Si=H 2

0

..Q rr i o o

c~ etO

o -4

. i i

9

o

2o'oo

'

2 'oo

Wavenumber

'

='oo

(cm ~ )

In situ FTIR spectra of an electrochemically hydrogenated Si surface in 0.1 M NaF (pH 4.0) obtained just after the anodic current transient has leveled out. The anodic oxidation is performed at + 1.5 V (a) (no oscillations) and at +6 V (b) (with anodic current oscillation). The baselines are shifted for better visualization. Reprinted with permission of the Electrochem. Soc. Inc. [81], copyright 1994. FIG. 25.

166


FIG. 26. STM micrographs of electrochemically hydrogenated Si(lll) surfaces in 0.1 M NH4F

(pH 4.0) after anodic oxidation in the oscillating regime for 2 (a) and 30 (b) rain. Reprinted with the permission of Elsevier Science B. V. [ 178], copyright 2000.

+1.5 V for process (i) (Fig. 25a) (noOsc-surface) and at +6 V for process (ii) (Fig. 25b) (Osc-surface). The IR absorption due to hydrogen silicon bonds is a little bit stronger for the hydrogenated Si surface after process (i), the noOscsurface. This stronger absorption is particularly prevalent in the region of the Si=H2 bonds and, therefore, points to a higher microscopic roughness of the hydrogenated Si surface with prior electropolishing without oscillations. In addition, an increasing IR absorption due to Si=H2 and S i - H species has been obtained with increasing time of etching after the dark current transient has leveled out. At first, the amount of Si-H2 increases and, finally, the total amount of S i - H and Si=H2 increases with increasing time. These processes are a result of roughening of the hydrogenated Si surface [ 177]. Figure 26 shows scanning-tunneling-microscopy (STM) micrographs of Si(111) surfaces hydrogenated electrochemically in 0.1 M NH4F (pH 4) after anodic oxidation in the oscillating regime, process (ii), for 2 (a) and 30 (b) min [ 178]. For the STM studies p-type samples have been cut from a B-doped Si(111) wafer (resistivity 1 S2 cm) with a misalignment of 0.25 ~ off the (111) orientation toward the (112) direction. The ultra-high-vacuum STM images are acquired at

167


c6

2 0

v

m

c 10

t p-poI

C~ o0

9

. m

n" o

5

tO

0

(1) c"

(1)

"

2ooo"

Si-H A !~~]

Si-H on A) (111) terrace B) (111) step atoms

i

C) (100) 1x2 reconstr.

,i# /..i., ISi-H2 |

?dad"

'2 'oo "

'2 'so "

72, )0

Wavenumber (crn -~) FIG. 27. Typical ex situ FTIR spectrum of an electrochemically hydrogenated Si(111) surface after anodic oxidation in the oscillating regime. The spectrum is normalized to the anodically oxidized surface. Reprinted with the permission of Elsevier Science B. V. [302], copyright 1999.

a constant current of 0.2 nA and a bias voltage of +3 V (for details see [178]). The overall surface morphology appears rough and is characterized by the formation of very fiat hole-like structures with lateral dimensions of some 10 nm up to about 100 nm. The holes become more pronounced and tiny protrusions are dissolved with increasing time of anodic oxidation in the oscillating regime, resulting in a smoother surface structure at the microscopic scale. There is no indication of facet formation. Surface structures very similar to those observed in our experiments can also be produced during electrochemical oscillations in fluoride-free electrolytes [ 179]. It should be recalled that hydrogenated Si(111) surfaces are macroscopically flat but very rough on the microscopic scale after immersion in HF (40%). The NH4F-treated (40%, pH 7.8) Si(111) surface is also macroscopically smooth; however, it exhibits a different microscopic structure. It consists of atomically fiat terraces that are spaced from one another by 0.31-nmhigh bilayer steps. These terraces are only disturbed by point defects and triangular holes [62, 79, 178, 180]. Typical ex situ FTIR spectra of an electrochemically hydrogenated Si(111) surface, recorded after processing (ii) at +6 V, are presented in Figure 27 for s- and p-polarization of the IR light. These spectra are very different from the wellknown FTIR spectra of flat hydrogenated Si(111) surfaces prepared in NH4F (40%, pH 7.8) or HF-treated S i ( l l l ) surfaces [42, 52, 59]. The most striking feature of the FTIR spectra of electrochemically hydrogenated Si surfaces is their

168


line broadening. The broad IR absorption peak gives evidence of a high degree of disorder at the hydrogenated Si surface region. In addition, there is a small signal for s-polarized IR light. Furthermore, the p-polarized spectrum contains two narrow peaks in the range of the Si-H stretching modes (2081.4 and 2087 cm-1), which are shifted with respect to the stretching mode of Si-H on terraces (2083 to 2084 cm -1 [68, 132]). The IR absorption peaks at 2078 and 2081.4 cm -1 may be due to Si-H on steps and Si-H groups on positions similar to Si-H on terraces, respectively. The peak at 2087 cm -1 seems to be due to Si-H groups that are at positions similar to Si-H species on 1 • 2 reconstructed Si(100) surfaces [45]. The broad spectrum smears out in the range of the Si=He stretching modes at higher wavenumbers. There is hardly any Si -- H3 detectable on the surface. The existence of the two narrow Si-H peaks, the positions of which are different from that of the ideally hydrogenated Si(111) surface, indicates that the electrochemically hydrogenated Si(111) surface is free of well-oriented and ideally hydrogenated Si(111) facets. 2 . 3 . 2 . HYDROGENATED SI SURFACES IN ALKALINE SOLUTIONS

Alkaline solutions lead to a high etching rate of silicon [ 181-185], and the silicon remains hydrogenated [ 148]. Etching in alkaline solutions is of great interest for microstructuring of silicon devices (see, for example, [186]). (100) oriented Si surfaces become a pyramid-like structure in alkaline solutions that is used for light-trapping systems in solar cell devices. The reason for this behavior is that the ratios of the etch rate of dihydride on steps (SD) to monohydrides on steps (SM) and monohydride on (111) terraces (TM) in NaOH solution are about 40 and 5000, respectively [ 187, 188]. Therefore, (100) facets are etched much faster than (111) oriented step facets. The etching process can be stopped by applying an anodic current to the sample, which leads to a decrease in the amount of hydrogen silicon bonds [ 148, 184, 185]. Obviously the Si surface becomes oxidized and SiOe is etched much more slowly than silicon. This section shows some selected results concerning the stability of hydrogenated silicon surfaces in alkaline solutions inspected by FTIR spectroscopy. The hydrogenation of a Si(111) surface is stable even at low anodic currents up to +3 #A/cm e, as can be seen in Figure 28, where IR spectra, with the use of a MIR-ATR sample, are recorded at different current densities in 0.5 M NaOH. The spectra are scaled to the IR spectrum of the oxide-covered surface in the same solution. There is, however, only a very narrow anodic potential range in which the hydrogen-terminated surface (a) remains stable. At slightly increased anodic current above +4/xA/cm e (b and c) passivation occurs [148, 185], and the Sill absorption gradually disappears until it vanishes (d). The electrode potential increases dramatically from - 0 . 9 2 V (a) to - 0 . 6 V (b), finally reaching +2 V (c), when passivation sets in at constant current. Obviously, a very sudden change

169


I

tO .4..-, {3.. x._ 0 0) t'3

10 .4 per reflection

rr 0 (7) tcO > .4..-,

a) -200 up to +a g~cm 2, c) +6.7 g~cm 2 b) +4 gA/cm 2, d) +8.5 gA/cm 2

0.5 M NaOH 1900

'

2dO0

'

21~00

'

22~00

Wavenumber

'

23~00

'

2400

(cm -~ )

FIG. 28. Relative change in IR absorption of the hydrogenated n-Si(111) surface at different current densities in 0.5 M NaOH. (a) -200 #A/cm 2 up to +3 #A/cm 2. (b) +6.7/zA/cm 2. (c) +8.5 #A/cm 2. (d) As in (c), but 5 min later. The spectrum of the oxidized surface in the alkaline solution serves as the reference.

in the mechanism of the corrosion occurs in this anodic potential range, which can be attributed to the formation of Si-OH surface bonds and their condensation to Si-O-Si bonds [148]. It should be noted that Si-OH could not be detected by the ATR-FTIR techniques, as described in Section 2.2. Figure 29 shows the integrated intensity of the IR absorption in the region of the Si-H stretching mode (2020 c m - 1 to 2200 c m - l) as a function of pH. The values are plotted in relation to the integral obtained at pH 4.9 with a pretreatment at pH 4.5 (open down triangle in Fig. 29, A0), which corresponds to about 95% of a monolayer, as deduced from FTIR, UPS, and HREELS measurements [64, 73, 76, 85, 164]. The integral of the Si-H/Si=He stretching mode region decreases with increasing pH. This decrease is attributed to the decrease in the Si=H2 surface species due to a reduction in microscopic roughness of the silicon surface. At a pH above 4.7, the competition between the very slow etching process of the oxide layer and the etching or oxidation of the Si surface is reflected by a strong suppression of the formation of hydrogen silicon bonds, which is completely suppressed at pH 5.3 in a 0.2 M NaF solution. A final etching step in a solution with pH 4.9 with a pretreatment at pH 4.5 leads to the smallest amount of Si=He oscillators on the Si(111) surface. This result is in agreement with ex situ HREEL spectra, where Si=H2 was present at pH 4.5, which disappeared after a subsequent dip in a solution with pH 4.9 [73]. In addition, Figure 29 reveals the stability of the hydrogenated Si surface with respect to alkaline etching processes when the acidic

170


// 1.5

,~

O

a{ C

1.0

< 0.5

2200 cm -1

A

-

-

.

J'IR absorption, d V

V=2000 cm -1 0.0

I// w

I

'

pH FIG. 29. Integrated IR absorption due to Si-H/Si=H2 in relation to the integrated intensity at pH 4.9 with a pretreatment at pH 4.0 (A 0, open down triangle) as a function of pH (calculated from the spectra of Figs. 24 and 28). Solid circles: direct etching of the oxide in 0.2 M NH4F; open circles: etching at pH 4.0 (a, c) or 4.5 (b) before etching at pH 5 (a), 5.3 (b), or 12.5 (c).

0.1 M NaF (pH 4) solution is replaced with a solution of pH 5.3 or 12.5. Nevertheless, no H-terminated surfaces can be formed in solutions with such high pH values (the oxide is etched back, but no hydrogen-silicon bonds are detected). 2.3.3.

E L E C T R O N I C STATES AT H Y D R O G E N A T E D SI SURFACES

Another interesting point is the evaluation of defects on such H-terminated surfaces. The defect concentration, Dit and Ns, is related to the surface structure and surface morphology after the hydrogenated Si surface is formed. Recall that Ns is measured in situ by a pulsed PL technique and Dit is obtained from ex situ SPV experiments. Figure 30 shows Ns and PL intensity (top), Dit at midgap (middle), and the current (bottom) as a function of time during the dark current transient after the anodic oxide on n-Si(111) is etched back in 0.1 M NH4F (pH 4) solution. The Dit value of wet anodic oxides measured with SPV is on the order of 1013 eV -1 cm -2, which is one order of magnitude higher than the value obtained from PL measurements. This behavior is due to the different kind of defects measured by SPV (rechargeable defects) and PL (nonradiative recombination active defects). Both Ns and Dit start to decrease when the current tends to decay; i.e., hydrogenation of the surface sets in and saturates when the current transient levels out. Note that the PL intensity is limited by the lifetime of the excess charge carriers in the wafer material, so that the Ns of high-quality float zone Si could be 1010 cm -2 or less.


171

101~ 3~--, 011

z~ 1

E3u

1012

EEg::p:D[~

~'E 1013

9

o

._

99

1011~

~" 1~

ex-situ SP 9

9

9 01

Oxide" //~ hydrogenation cover~ ~_ Current

~E 40 0

~

2

9

> 1012

o

in-situ PL

20

0

0

.

.

.

.

I

500

.

.

.

Time (s)

.

I

1000

FIG. 30. Ns and PL intensity (top), ex situ obtained Dit at midgap (middle), and the current (bottom) as a function of time during the dark current transient after an anodic oxide was etched on n-Si(111) in 0.1 M NH4F (pH 4) solution.

Anodic oxidation in the oscillating regime is important as the prior step for the formation of a hydrogenated Si surface with low density of surface states. As shown above, prolonged anodic oxidation in the oscillating regime leads to locally smoother surfaces. We note that the anodic current is lower and the maximum of the PL intensity is higher after the longer anodic oxidation in the oscillating regime. Therefore, the value of the anodic current transient, as well as the PL intensity, is a quantity describing the microscopic roughness of the Si surface under identical electrochemical conditions. It is important to note that the in situ PL intensity usually decreases in time after the hydrogenated Si surface is formed. This decrease in the PL intensity after the maximum is reached is caused by the onset of chemical etching at the Si surface in the electrolyte. The chemical etching is a dynamic process at the Si surface during which the well-passivated Si surface is disturbed and nonradiative recombination active surface defects are generated. The maximum of the PL intensity can be strongly increased after repetition of the electrochemical treatment of anodic oxidation with current oscillations and hydrogenation (process (ii), Osc-surface). Repeating this process (ii) leads only to a slight increase in the macroscopic surface roughness while tiny protrusions

172


0.6

f n-Si(111)

el.chem.hydrogenated (osc.) ~... ~ ( n o o s c . )

0.5

,I~

~NH4F treated

0.4 /

~.0.3

/

/ / /

//

HFtreated

0.2 0.1 0.0

-600

-300

0

300

UF (V)

600

900

FIG. 31. Examples of SPV measurements of differently hydrogenated n-Si surfaces (solid lines) and

of an n-Si surface covered with a native oxide (4 months of oxidation in air after treatment in NH4F). Reprinted with the permission of Elsevier Science B. V. [302], copyright 1999.

are dissolved or rounded. But Ns can be much more strongly reduced by this procedure than by prolonged anodic oxidation with only one oxide etch step [ 189]. The maximum of the PL intensity corresponds to a density of nonradiative surface defects on the order of 1 • 101~cm -e. Hydrogenation was also performed on Si surfaces after anodic oxidation when anodic current oscillations do not appear (process (i)). In this case, the microscopic roughness remains unchanged and the PL intensity does not depend on the repetition of process (i), and only a value of Ns -- 4 x 101~ cm -2 could be reached [189]. The surface morphology of hydrogenated Si surfaces can be correlated with the density of surface states measured ex situ by SPV. Figure 31 shows examples of ex situ SPV measurements in an N2 atmosphere of differently hydrogenated n-Si surfaces (solid lines) and of an n-Si surface covered with a native oxide (4 months of oxidation in air after treatment in NH4F). For the oxidized surface, the neutral point (NP) (UF = 0 V) is close to midgap, and the Uph (UF) characteristic is symmetric around NE This is caused by the amphoteric character of the electronic states at the Si/SiO2 interface, which are predominantly determined by Si dangling bonds [190, 191]. There is no hysteresis in the slope of the Uph (UF) characteristic of the oxidized Si surface. The Uph (UF)characteristic of hydrogenated Si surfaces generally shows a hysteresis, and NP is shifted to positive values of UF. The latter is caused by an accumulation of fixed positive charges (Qf) at the sur-

173


0.5

n-Si(111):H electrochemical preparation

= =

UFfrom 40 to 90 V the same sample UFfrom 0 to 200 V

0.4

>

o.3 c-

0.2

/ ~

0.1 0.0

(D O~c "-~1 i

J , , ,

0.0 . . . .

0

, . . . . , . . . . ,

50

1100

.

. . . .

50

U~ (v)

.

.

.

i

E,

0.5

.

.

.

.

Ev (eV)

. . . . . . . . .

200

,

250

1.0

,

, . . .

300

FIG. 32. Uph (UF) characteristics for different extensions of the UF range for the same Si surface hydrogenated after anodic oxidation in the oscillating regime. The inset shows the distribution of surface states. Reprinted with the permission of Elsevier Science B. V. [302], copyright 1999.

face, so that the surface is in accumulation for p-type and in inversion for n-type Si (see also [192]). An n-type behavior of HF-treated Si surfaces has been found by Buck and McKim, who made surface conductivity measurements [193]. The value of Qf ranges between 101~cm -2 in the case of electrochemical hydrogenation after anodic oxidation in the oscillating regime and 1012 cm -2 for HF-treated surfaces. The hysteresis tends to decrease with decreasing Qf and with increasing slope of Uph vs./-IF. It can be concluded that the microscopic rough hydrogenated Si surface (HF treated) has the largest concentration of surface defects, and the best passivation can be reached on the macroscopically relatively rough but microscopic smoothest surface after process (ii). Slope, shift, and hysteresis of the Uph (UF) characteristics of hydrogenated Si surfaces sensitively depend on the regime of the SPV measurement. Figure 32 shows Uph (/-IF) characteristics obtained at different extensions of the UF range for the same Si surface treated with process (ii), hydrogenation after anodic oxidation in the oscillating regime. The inset gives the distribution of surface states obtained from the SPV measurement with the lower extension of UF. For the more extended UF range, the value of UF in the turning point is shifted to higher values, the hysteresis is increased, and the slope is reduced. Therefore, Qf, hysteresis, and Dit are larger for larger extensions of UF. Consequently, the distribution of Dit strongly depends on the condition of the measurement. To conclude, the surface states measured ex situ by SPV on hydrogenated Si surfaces are of the donor type, which have a broad distribution of trapping and detrapping times.

174


Nevertheless, the lowest values of/-)min --it can be obtained with high accuracy by turning point analysis from Uph (UF) characteristics, except when neither accumulation nor strong inversion is reached at the Si surface. The value of "'It /-)min is about 101~ eV -1 cm -e for the n - S i ( l l l ) surface after process (ii) (see inset of Fig. 32). For chemically hydrogenated Si(111) surfaces (preparation in buffered NH4F), the lowest reported value of/-)min is 2-5 X 1010 eV -1 cm -2 (ex situ measure"'It ments [194, 195]). Slightly lower values of/-).min "-'it are reached on electrochemically hydrogenated Si surfaces after process (ii). This is surprising for this surface morphology and shows that roughness on the microscopic scale is crucial. Efforts have been devoted to correlating "'It /-)min of chemically hydrogenated Si surfaces with surface roughness [195] determined by ellipsometry [78]. The obtained resuits have been interpreted in terms of a dangling bond model [ 196]. This model may be suitable for explaining the development of the distribution of Dit during the initial stages of oxidation [ 197], because the formation of dangling bonds at Si back bonds plays a major role [ 198]. However, this model is not applicable to hydrogenated Si surfaces, because the concentration of dangling bonds at a hydrogenated Si surface (in terms of dangling bond centers measured in porous silicon by electron paramagnetic resonance [ 199]) is much lower than Dit obtained by ex situ SPV. The dangling bond concept can be applied to the in situ investigation of Si surfaces by PL during electrochemical processing. Dangling bonds are formed and passivated in the electrolyte during oxidation of Si surface atoms and etching depending on the chemical equilibrium. The surface recombination velocity is extremely low for hydrogenated Si surfaces in acidic solutions [48]. Therefore, the chemical equilibrium in acidic fluoride solution is shifted toward low oxidation rates and highly efficient passivation of surface states, probably by the reaction of protons with defects at the Si surface. Indeed, Trucks et al. [200] proposed a mechanism for the hydrogenation process on Si surfaces, which results in an extremely low etch rate for hydrogenated Si surfaces in fluoride-containing solutions. The concentration of dangling bonds on HF-treated Si surfaces is below the detection limit (less than 5 x 101~ cm -e) of an electron spin resonance spectrometer and starts to increase during the initial oxidation process [201]. The development of dangling bonds, which act as nonradiative surface defects, can also be probed in situ and ex situ by measurement of the PL quenching. 2.3.4. R O L E OF THE E T C H RATE FOR SURFACE STATE F O R M A T I O N

The decrease in the PL intensity the maximum value is reached is correlated with a roughening of the surface on a microscopic scale during chemical etching in acidic fluoride solution. As outlined before, the IR absorption after electrochemical hydrogenation increases with increasing etch time in 0.1 M NH4F (pH 4), and

175


0.1 M NH4F (pH 4.2) v

>" 1

!

,

oO C

n

0

E

8

< E

4

0

"--

",~. ,. p-S i(100)

I !

C _1

%

(b) +3v dicoxidation . . . .

-0.4v hydrogenation ~~ x l O 0

(a )

0

0

200

400 Time

600

800

(s)

FIG. 33. Time dependence of the current (a) and PL intensity (b) for a cycle of anodic oxidation in 0.1 M NH4F (pH 4.2) at +3 V followed by hydrogenation at - 0 . 4 V for p - S i ( l l l ) and p-Si(100). The wafers are cut from the same Si ingot.

the increase is stronger for the Si=H2 than for the Si-H bonds. The influence of surface orientation, oxidation rates, and temperature on the PL intensity measured in situ will be discussed in the following section. The stability of the hydrogenated Si surface depends strongly on the surface orientation, as illustrated in Figure 33, which shows the current (a) and PL intensity (b) as a function of time for a cycle of anodic oxidation in 0.1 M NH4F (pH 4.2) at +3 V followed by hydrogenation at - 0 . 4 V for p-Si(111) and p-Si(100). The samples are cut from the same Si ingot to ensure the same bulk properties. Usually the maximum of the PL intensity is about 10-20% higher for Si(111) than for Si(100). The subsequent decrease in the PL intensity is much faster for the Si(100) than for the Si(111) surface. This behavior can be attributed to the higher etch rates in the (100) than in the (111) direction, as obtained in alkaline and buffered NH4F solutions [ 188]. Figure 34 shows the influence of dissolved oxygen in the HF-containing electrolyte on the current transient and the PL intensity during and after hydrogenation of p-Si(111). The Si surface is oxidized at +8 V followed by the current transient recorded at - 0 . 5 V in 0.1 M NHaF (pH 3.5). The PL intensity increases drastically when hydrogenation occurs (decrease in Ns) and decreases with time in the HF-containing electrolyte. There is only a very narrow time window where the PL intensity is at a constant, high value directly after the hydrogenation is completed

176


0.1M NH4F (pH 3.5)

1.5

~~lUmuOoononi'o,O~00

9

010m0

electrolyte exchanged ~

9 _(00A

9

,dR . . . .

>,, .m

o~

t--

1.0 I

c-. "--

0.5

0.1 M K2SO4 (pH 3) at-0.7 V

O_

'•_.

0.0 E

10~

o

10-1

"-su ~l~,,,,av~ _14,u

exchange of electrolyte by

I

!

-0.5 V

non-purged

~: 10-2 "-

10 -a

N2 purged :

~

(a)

,, '

'

o I

'

'

'

'

1;o

'

'

'

'

2;o -

'"

'

'

'

'

I

'

'

300

Time (s)

FIG. 34. Time dependence of the current (a) and PL intensity (b) for a cycle of anodic oxidation in 0.1 M NH4F (pH 4.2) at +8 V followed by hydrogenation at -0.5 V. The electrolyte is either purged by N 2 (thin solid line) or is not (dashed line). The electrolyte is replaced with 0.1 M K2SO4 (pH 3) at - 1/zA/cm 2 after the maximum of the PL intensity is reached (dotted line).

in the nonpurged solution (dashed line). This time region is somewhat prolonged when the solution is purged with N2 before the experiment (solid line). Purging with N2 leads to a strong reduction of oxygen in the solution. In both cases, the PL intensity also decreases with time. The time dependence of the current does not differ significantly for the purged and nonpurged solutions. The decay of the PL intensity in the diluted HF solution is due to etching of the surface (formation of nonradiative defects), which is enhanced by dissolved oxygen in the electrolyte. Oxygen in solution promotes the formation of Si-O bonds [88, 202], which are then dissolved by HF, leading to a roughening of the surface. It is not only of interest from a practical point of view to preserve the low level of nonradiative surface defects for longer times. The decrease in the PL intensity after the maximum is reached is related to a partial destruction of the hydrogenated Si surface due to chemical etching of the Si surface. This can be avoided by replacing the acidic fluoride electrolyte with an acidic solution of, for example, 0.1 M K2SO4 (pH 3) at a small cathodic current density of - 1 / x A / c m 2 to protect the Si surface from oxide formation (dotted line in Fig. 34). Therefore, highly efficient etch stops can be integrated into electrochemical passivation procedures [203]. However, the cathodic current should not be too large, to avoid hydrogen evolution. The incorporation of hydrogen at cathodic potentials from

177


~ , ,

1.5 v

1.0 fao e-

._1

n 04

-0.3V

+8v

/

,_

0.5 0.0

no peroxide

!..,'\

(a)

E o

,, .m oo c

p-Si(lO0) / 0.2 M NH4F (pH 3.2)

O t--

0.2. 3

0.08-

-', e,-i-

o

2 3

0.06-

t.'.m J 13.. 0.04-

1

0.020.00

,

-0.5

,

0.0

0

,

0.5

1.0

Potential (VscE)

Current-voltagescan and PL intensity of p-Si(100) in 0.2 M NH4F (pH 3.2) during the voltage scan (scan velocity 20 mV/s). Reprinted with permission from the American Institute of Physics [229], copyright 1998.

FIG. 49.

The potential dependence of the photovoltage, Upv, gives information about the charge transfer at the Si surface. Figure 50 shows the current-voltage scan (a) and the corresponding Upv scan (b) of p-Si(111) in 0.1 M NH4F (pH 3.9). The anodic current changes the slope at a value that is about a third to a half of the anodic current at the first maximum. This is similar to the PL and current-voltage scan shown in Figure 49. The Upv amplitude decreases linearly with increasing potential up to a potential of about 0 V. At this potential the anodic current changes the slope and the slow decrease in the PL intensity, as shown in Figure 49, begins. There is no hint about any signature in the Upv or the sharp onset of the process of porous silicon formation with increasing potential as observed by PL measurements (Fig. 49). Therefore, the charge transfer at the Si surface remains unchanged during H-termination or porous silicon formation. The Upv amplitude increases slightly with increasing potential between 0 and 0.2 V and decreases again toward accumulation with a further increase in the potential. The increase in the Upv amplitude of p-Si(111) gives evidence for an increasing positive charge at the Si surface. The increase in positive charge at the Si surface is finished after the first maximum of the anodic current only. The onset of the increase in the Upv amplitude with increasing potential coincides with the change in the slope of the anodic current and the onset of the slow decrease in the PL intensity. Therefore, the electrochemical reactions leading to electropolishing cause a dynamical storage of positive charge at the Si surface [228]. Smith and Collins [218] distinguished the current potential scan in diluted HF solutions in a region of hydrogenation, a transition region, and a region

194


CM

E

< E

p-Si(111)

1.0

0.5

0.1 M NH4F pH 3.9 0.3 M (NH4)2s04

0.0 0.0 v>

> a_

(a)

A

3 Dcm

I

,I

(b)

I I

-o.1 -0.2

-0.3 i

i

i

-0.2

'

'

'

I

0.0

'

'

'

i

0.2

'

'

'

i

0.4

|

|

'

I

0.6

'

'

Upo, (V) FIG. 50.

Current-voltage scan (a) and corresponding PV scan (b) of p-Si(lll) in 0.1 M NH4F

(pH 3.9).

of electropolishing. The region of hydrogenation contains the two regimes of H-termination and porous silicon formation, which are characterized by two plateaus with the high and medium PL intensities (Fig. 49), respectively. The transition region is characterized by the slow decrease in the PL intensity and by a storage of positive charge at the Si surface. The PL intensity is at a low level during anodic oxidation, the region of electropolishing. As known, holes are consumed at the Si surface during anodic oxidation. This positive charge is transferred to the electrolyte by a retarded process. Such a process should be mediated by a complex supporting an electron that is injected into the Si and recombines with the hole. The positive charge that remains on the complex polarizes the Si surface within the inner Helmholtz layer. In contrast, there is no influence of charged surface complexes during porous silicon formation. Lehmann and G/3sele [216] proposed a dissolution mechanism where a Si-H surface bond is weakened by the capture of a hole. Subsequently, fluoride ions react with the destabilized Si surface atom. One electron is transferred from a fluoride ion into the Si bulk during this reaction. The former surface hydrogen atoms lead to hydrogen gas evolution. The Si-Si back bonds are now strongly polarized and can be very quickly attacked by other polar molecules like HF, leading to a dissolution of a SiF4 species, which further reacts with SiFt- ions, with F - ions

195


~-,

"T" r

10 (a)

, D

~, ~

nSi(111) A,

v ,,i-, t-

,-, "1-

0.1 M NaF (pH 4.0), 0.1 mA/cm 2 0.2MNaF(pH3.2),0.4mA/om 2

A, F"I 0.2 M NaF (pH 3.2), 0.8 mA/cm 2

r-i

1

r c.-

.

.

.

.

.

.

.

.

v

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

nSi(100)

0.1

pSi(100) / 0.2 M NaF (pH 3.2), 1 mA/cm 2

"-I

o.~

"~ I"

.

nanopor-Si~

0oo%o

0

l

c-si

.=_ 0.01 _.1

a.

(b) 0.01

'

lID '

'

' '"'I

0.1

'

'

'

''"'1

1

'

'

'

''"'1

10

'

'

'

''"'1

100

'

'

'

''"'1

1000

,

,

i

,

Charge (mC/cm 2 )

FIG. 51. Dependence of the Sill/Sill2 ratio (a) and the PL intensities of c-Si (measured at 1.1/zm) and nanopor-Si (measured at 0.6/zm) (b) on the passed charge during anodization of Si(111) or Si(100) at different current densities and in different fluoride solutions. Reprinted with permission from the American Institute of Physics [229], copyright 1998.

present in solution (see reaction scheme (4.1)). Hence, the Si surface is again hydrogenated by a process similar to the chemical hydrogenation of Si surfaces as proposed by Trucks et al. [200]. Si surfaces can be roughened in a well-controlled manner. The surface area increases with increasing charge flow during anodization. Figure 51 shows the dependence of the ratio of SiH/SiHe on the passed charge during anodization of Si(111) or Si(100) at different current densities and in different fluoride solutions. A constant ratio of SiH/SiHe = 0.4 is reached after a charge flow of 10 mC/cm e (a). This ratio is independent of surface orientation and remains unchanged for the formation of porous silicon. In addition, Figure 51 shows the dependence of the PL intensity of c-Si (measured at 1.1 /zm) and nanoporous Si (measured at 0.6/zm) (Fig. 5 lb) on the charge that passes the electrode [229]. The PL intensity of c-Si remains constant up to a passed charge of about 100 mC/cm e, despite the strong increase in the surface area of the Si sample. This means that the parts of the surface that are not related to the anodization process do not contribute to the nonradiative surface recombination at all. In other words, the concentration of reactive surface sites remains constant during the surface roughening and porous silicon formation. This is not surprising because the thickness of a porous layer is proportional to the passed charge.

196


The PL signal of por-Si arises after a passed charge of about 300 mC/cm 2 and increases strongly with further anodization. The PL intensity of c-Si starts to decrease at about 200 mC/cm 2, i.e., when the PL signal of por-Si occurs. The reason for this is that a certain amount of the exciting light is absorbed in the porous surface layer, where the excess carriers may recombine radiatively.

2.4.3.

ELECTRONIC STATES AT INTERNAL SURFACES OF POROUS S I AND LOCAL RECONSTRUCTION

Transport of excess carriers of charge is important for the PL of por-Si. Usually, the PL intensity is negative correlated with the electric conductivity [230]. Figure 52 presents PL spectra of c-Si and por-Si after anodization in 0.2 M NH4F (pH 3.2) solution and after replacing the electrolyte with nitrogen and ethanol atmospheres. The thickness of the por-Si layer is on the order of 70 nm, and the exciting light of the N2 laser (wavelength 337 nm) is almost completely absorbed in the porous surface layer. The PL signal of por-Si increases strongly after the electrolyte is replaced with an ethanol atmosphere, whereas the PL signal of c-Si decreases. This negative correlation shows the influence of the ambient on the diffusion of charge carriers. The diffusion length of the excess carriers of charge is larger than about 50 nm in the electrolyte, whereas it is much shorter than 50 nm for por-Si in the ethanol atmosphere. For comparison, diffusion coefficients

0.3

o

~ ' ~9 ..,A

~

0.2

"~ r-

.B

in electrolyte 9 in nitrogen atm. in ethanol atm.

~ 0

9

0.1

)1(

o 1.0

1.5

2.0

2.5

3.0

Photon e n e r g y (eV) FIG. 52. PL spectra of c-Si and por-Si after anodization in 0.2 M NH4F (pH 3.2) in solution and after

replacement of the electrolyte with nitrogen and ethanol atmospheres. Reprinted with permission from the American Institute of Physics [304], copyright 1997.

197


1014 c7 E 0 "7 >

a) n-Si(100):UPSL b) n-Si(100):H c) n-Si(1 t 1):H

1013

a .. 1012

s

. _

1011 lO l~ ,,.

-0.5

.

.

.

.

.

.

.

.

010

0.5

E - E i (eV) FIG. 53. Interface state distribution obtained from ex situ SPV measurements for electrochemically hydrogenated n-Si(111) and n-Si(100) surfaces and for a Si(111) surface covered with a 20-nm-thick por-Si layer.

of excess carriers in mesoporous silicon measured by optical grating techniques amount to about 30-90 nm [231 ]. The interface state distribution is an important parameter for the characterization of the surface passivation from an electronic point of view. Ex situ SPV measurements can be used for ultrathin porous silicon (thickness 10-20 nm), for which the exchange of charge carriers with the c-Si bulk is not limited by transport [232]. Figure 53 compares the interface state distributions for electrochemically hydrogenated n-Si(111) and n-Si(100) surfaces and for a Si(111) surface covered with a 20-nm-thick por-Si layer. The lowest value of Dit is reached on a Si(111) surface after electrochemical hydrogenation with prior anodic oxidation in the oscillating regime, as shown and discussed in Section 2.3. The density of surface states is higher for the n-Si(100) surface prepared under identical conditions than it is for the n-Si(111) surface. The overall surface state density of ultrathin por-Si is comparable to that of HF-treated Si surfaces, whereas the normalization to the internal surface area, which is about 600 me/cm -3 [233], leads to a value of Dit on an order similar to that of the n-Si(100) surface. The excellent passivation of hydrogenated Si surfaces in por-Si can also be demonstrated for mesoporous Si prepared on highly doped p+-Si substrates. An important question concerns the existence of free carriers of charge in the mesoporous Si. Usually, mesoporous Si has a large resistivity, and there is no evidence of free carriers of charge [234]. However, under certain preparation conditions absorption of infrared light by free carriers of charge is observed [235], as shown in Figure 54, where ex situ IR transmittance spectra are plotted for as-prepared

198


100 c"

=. 10 -1 d3 10 .2 0 c-"

"1

,.I...I

E 1

0-3

~

0.4

t

cO t-"

,

,

I

1000

,

m

p§ (120 lam) p+-Si (32 lam) meso-PS (as prepared) meso-PS (NO2 adsorption) I

I

i

I

2000

i

i

i

i

I

3000

,

,

,

,

I

,

,

4000

Wavenumber (cm-~) FIG. 54. Ex situ IR transmittance spectra for as-prepared free-standing mesopor-Si film (thickness 75 /zm), the same film of mesopor-Si with adsorbed NO2 molecules, and the p+-Si substrate for thicknesses of 120 and 32 #m. Reprinted with permission from Physica Status Solidi [235], copyright 2000.

free-standing mesopor-Si films (thickness 75/xm, without and with adsorbed NO2 molecules) and for the p+-Si substrate (thickness of 120 and 32 #m). The spectra show the typical absorption peaks of Si-H modes, and there is no evidence of oxide species at the internal surface of the mesoporous Si. The continuous underground of the IR spectra is characteristic of absorption by free carriers of charge. Our measurements give evidence of a high amount of free holes in the mesoporous silicon. One can conclude that the concentration of amphoteric or donor-type surface states is below 1011 eV -1 cm -2 with respect to the large internal surface area (about 600 m2/cm -3 [233]). This shows that the concentration of reactive surface sites that may adsorb water molecules (donor type molecules) is very low for hydrogenated Si surfaces in mesoporous Si. The low concentration of compensating defects in mesoporous silicon was also shown by intensity-dependent surface photovoltage measurements [236]. A space charge region is formed in the mesoporous silicon surface region, which depends on the concentration of free carriers. The porosity of the mesoporous Si modifies the dielectric constant, which allows us to handle this region in a manner similar to the method used for the space charge region of c-Si. The number of electronic surface states (or reactive surface sites) at the inner hydrogenated Si surface of porous silicon is extremely low. This, together with

199

ELECTROCHEMICAL PASSIVATION OF SI AND S I G E

co "o to ..Q

e,.,,,, ; ' ' 2,,.,,,, /e;' '

0 co

9. * " . . . . .

,,

t~

rr SiHa/SiH

0.1

i

I

i

i

i

I

1

9

,,,

I

,

i

9

~

i

i

i

i

i

I

,~411j~llilill/li

i

i

I

i

i

10

i

e i

I

i

100

R a d i u s of s p h e r e (nm) Dependenceof the calculated relation of SiH2/SiH and SiH3/SiH bonds on the radius of a sphere of c-Si when reconstruction is absent [238].

FIG. 55.

the hindrance of carrier transport, is the reason for the highly efficient photoluminescence in nanoporous Si. The PL efficiency of nanoporous silicon can be strongly increased by initial oxidation in water [237]. Reactive surface sites are created during the initial oxidation. However, water molecules passivate effectively reactive surface sites. It is known from electron spin resonance experiments on hydrogenated por-Si surfaces that adsorption of water molecules can decrease the concentration of dangling bonds by at least one order of magnitude [199]. A similar effect is known from experiments in ultra-high vacuum on Si(111) 7 • 7 surfaces [212] or Si(100) surfaces covered with a thin oxide [213] for which the concentration of nonradiative recombination surface defects decreases strongly during the adsorption of water. In Section 2.3, we relate the formation of disordered hydrogenated Si surfaces to electrochemically induced local reconstruction of the Si surface. The nanoparticles in porous silicon have round shapes; no faceting is observed. The argument of local reconstruction is supported by the fact that the ratio of Sill, Sill2, and Sill3 bonds saturates. Nevertheless, the ratio of dangling bonds at unreconstructed Si spheres is very different from the experimentally observed one. Figure 55 shows the dependence of the calculated relation of SiH2/SiH and SiH3/SiH bonds on the radius of a sphere of c-Si if reconstruction is absent [238]. A similar approach of passivation of dangling bonds by hydrogen is used for theoretical calculations of the electronic structure of Si nanoparticles. The calculated ratio Sill/Sill2 saturates at about 1, whereas the measured ratio Sill/Sill2 saturates at about 0.4

200


for the as-prepared porous silicon. The difference means that the shapes of the Si nanoparticles are not really sphere-like, but have some preferential orientation in the (100) direction and/or that atomic steps at the surfaces of the spheres are smeared out by local reconstruction. In fact, porous Si is preferentially etched in the (100) direction. Obviously, more theoretical work is needed to finally solve the question of local reconstruction in porous silicon. Local reconstruction can be understood as a kind of amorphization of the hydrogenated Si surface, which keeps all surface atoms in more or less identical positions from the point of view of surface chemical reactions. A very thin amorphous Si surface layer with an extremely high amount of hydrogen would act as a passivation layer.

2.5. Thin Anodic Oxides on Si The most widely used method in Si device passivation is thermal oxidation, in a dry or wet oxygen atmosphere in a range of 700-1100~ for some minutes, depending on the thickness of the oxide [4, 9, 239]. Recently, low thermal budget processing like PECVD in any kind of variation [17, 117, 124, 240, 241], electrochemical oxidation procedures [2, 3, 5, 12-27], and other more or less exotic treatments like, for example, oxidation in ozone at 200~ [242] have been developed for device quality passivation of Si and even for SiGe epitaxial layers [32-34, 243-245]. The surface roughness is very important for the formation of thin gate oxides [120-123]. The influence of initial stages of oxide deposition differs from that of "bulk oxide" formation. In the former, the initial Si surface mainly defines the electronic properties; i.e., damage to the surface can be observed during the beginning of the deposition process by the needed plasma source. No change in the interface occurs if the oxide layer is thick enough to head off the energy that is incorporated by the plasma [ 124, 125]. The situation is sharply different when the oxide layer is grown into the silicon bulk, where the interface is permanently changed by the formation process, i.e., diffusion of atoms, ions, or molecules through the oxide layer [2, 12]. The initial stages of formation of oxides are important for the electronic properties of the interface of thin oxide layers and have been widely investigated [ 133, 213,246-255]. These properties include the space-resolved variation of tunneling current investigated by STM or atomic force microscopy (AFM) techniques [256], the homogeneity and morphology of the layer and the interface [257], the dependence of the current density on the number of steps at the SUSiO2 interface [258], and the strain at the interface during oxide growth [ 118, 125], which is correlated with the defect concentration [259]. Even effects of the initial electrode potential [260], amount of water and temperature [27, 261], and organic impurities in water [249] on the oxide growth conditions have been observed.

201


2.5.1. I N I T I A L STATES OF A N O D I C OXIDATION

X-ray photoelectron spectroscopy and high-resolution electron energy loss spectroscopy [ 198] revealed the formation of Si-OH groups on Si surfaces in contact with water. Infrared absorption experiments in the UHV chamber have recently been performed and show the formation of Si-H species with different types of oxide back bonds in the beginning of oxide formation [ 133]. The oxidation process of a hydrogenated Si surface starts with the breaking of Si-Si back bonds, leaving the Si-H bonds untouched at first [133]. Similar results have been obtained for electrochemical oxidation of H-terminated Si surfaces. Figure 56 shows FTIR spectra during anodic oxidation of Si(111) (A) and Si(100) (B) surfaces in 0.2 M NazSO4 (pH 3) normalized to the hydrogenated state. The samples are cleaned and smoothed before oxidation by an electropolishing step at +3 V in 0.1 M NH4F (pH 4) followed by the hydrogenation procedure described in Section 2.3. In the beginning, the Si-Si back bonds of the Si-H surface species are converted into Si-O-Si bonds, leading to a decrease in the IR absorption in the Si-H region (peak around 2090 cm -1) and to an increase in Si-H species with an increasing amount of oxygen back bonds (OSi-H 2118, OzSi=H2 2200, and O3Si-H 2255 cm -1 [133-136]). Whereas the (111) oriented Si surface shows the subsequent formation of Si-H species with one, two, and three oxygen back bonds with increasing potential, the Si(100) surface shows the formation of Si-H species with

(a)

c-

.o

o_ o x_

or}

I (b)

I nSi(100) I

.., ..../~i1~~._~..,..,.,,~. ~ reference ~ e f e r e n c e lJl~k-i,~. . . . . . . . . . Z H-term" ~ ~ / A ~ _ . _ H-term ,4.

cr O

O O~

v v

0s v .

~-,,j. k,-,k t~--'--0.~

~

20'00

'

'

I

'

2200

'

I

2400

0

V *' [ !-~-:~=r

"~'"~"~"~'~-~-~-0.5 '

-

V '~ i ~ # : . . , , . = =

~ - 0 . 4

x...

, ~ ~

v-

co

O

!22j.

"

'

I

'

2600

.

+o. v

+~.~ v

3 V ~.-.~"-..-.-.~.~',:;:-0.36V -0.5 V

20'00'22'00'24'00'26'00

Wavenumber (cm ~) FIG. 56. IR absorption spectra in the range of Si-H stretching modes of n-Si(111) (a) and n-Si(100) (b) during anodic oxidation at different potentials. The spectra are normalized to the hydrogenated

surface (reference).

202


n-Si(111 )

tO Q_ 0 a3 I

n"

//

anodic oxidation 0.2 M Na2SO4 (pH 3.0)

anodization potential (VsoE)

_. "~,~.,,~........,z~ \

ta3 tO

+5.0 +2.0 .o

(D L!/~/~

I

1000

-. . . .

~

11 O0

-0.25

1200

Wavenumber

1300

;0.a

1400

(cm-')

FIG. 57. Relative change in IR absorption in the range of Si-O-Si stretching modes during anodic oxidation of n-Si(111) at different potentials. The hydrogenated and oxide-free surface serves as the reference spectrum.

two oxygen back bonds (O2Si--H2) only. This behavior points to a faster oxidation of the (100) surface, which is less stable than the Si(111) surface. The IR absorption due to Si-H completely disappears when the potential increases above +2 V. This relatively high potential is indicative of a preferred island formation of the oxide (3D growth) with a later onset of 2D growth, as also proposed by other authors [262]. In addition, Figure 57 shows FTIR spectra during oxide formation on Si(111) in the regime of the Si-O-Si asymmetric stretching mode, obtained with single internal reflection techniques. A broadened IR absorption with a maximum at about 1120 cm -1 occurs at small anodic currents at -0.25 V because of the formation of Si-O-Si groups [137-139]. This absorption peak decreases slightly in intensity with increasing anodic potential and splits into two peaks (easily seen at + 1 V and above) that are centered at 1050 cm -1 and 1240 cm -1, respectively. This indicates the formation of a thicker oxide layer (about 23 ML), where a disorder-induced vibrational effect is present [ 139], which leads to the LO-TO split [117, 139, 142, 144]. The former peak at about 1120 cm -1 (-0.25 V) seems to be due to a preferential order of the oxide formed during the beginning of the oxidation process (the LO4 mode is located at about 1140 cm-1 [117]); i.e., the bond angle of a Si-O-Si group at a Si surface bond is well defined by the crystalline lattice of the bulk material. The amount of this kind of ordered Si-O species, which exists only at the oxide/Si interface, decreases in relation to

203

ELECTROCHEMICALPASSIVATIONOF SI AND SIGE c 0

n-Si(111) / 0 1 M N a F (pH 4)

Q. 0

.-.

'0

. m

Cas SiO2 TC '

.13

rk

0.4

"E 0.3 '~. tE 0.2 "-" 0.1

-

,,,.,.=

I

b

'

d

o.o ,-1~,, 'a . . . . . . . .

I

0.0 0.5 1.0 1.5 2.0 2.5

~H20 Potential (V) aft. .,,a~-~,~k,~.,,, ....~ ....., . - , - , ~ i ~ ~

0

T 3.1 o-a

c co

a ~r",r,--'v,r','+-r,,,-,~ 1

d O

0

'

'

'

'

I

1500

'

'

'

'

I

2000

Wavenumber

b,-x,~

~, '

'

'

'

I

2500

(cm-')

FIG. 58. IR absorption spectra during anodic oxidation of n-Si(111) in 0.1 M NaF (pH 4) at different potentials. The spectra are normalized to the hydogenated Si surface obtained after etch-back of a thin anodic oxide layer (about one monolayer Si-H).

the growing thickness of oxide layer, and the IR absorption due to "normal" Si-O groups, which are highly disordered, predominates. From Figures 56 and 57 one can conclude that a mixture of fully (SiO2) and partially (xO-Si-H) oxidized parts coexists on the Si surface in this potential regime. Figure 58 shows the relative change in IR absorption in reference to the hydrogenated and oxide free n-Si(111) surface at different anodic potentials under illumination in 0.1 M NaF (pH 4) with the application of a single intemal reflection. The potential was stepped from - 0 . 2 V to +0.6 V, +0.9 V, +1.7 V, or 4-2.6 V, respectively, as indicated by the thick arrows in the current-potential curve of the inset in Figure 58. Hydrogenated and porous Si is formed at the first strong increase in the current (+0.6 V, 130 mC/cm2). The FTIR spectrum (a) exhibits typical IR peaks around 2100 cm -1, which are broadened because of different kinds of Si-Hx species of the porous layer (x - 1, 2, 3; mono-, di-, and trihydride, respectively). Nevertheless, no oxide species could be detected at this potential. The well-known split of the IR absorption at the asymmetric Si-O-Si stretching mode into parallel (TO, 1050 cm -1) and perpendicular (LO, 1230 cm -1) components can be seen at anodic potentials above the first current maximum. This split is not well resolved at +0.9 V (spectrum (b)). Moreover, the IR absorption of the asymmetric stretching mode of Si-O-Si is somewhat broadened, which is in contrast to the IR spectrum recorded at - 0 . 2 5 V in fluoridefree solution (see Fig. 57), where a maximum is observed around 1120 cm -1,

204

R A P P I C H AND D I T T R I C H n-Si(100) 3 sqcm 10 ML

>

IJJ O 03

s starting with hydrogenated surface

v m co 13_

1 laA/cm 2 1 laA/cm 2 2 laA/cm 2

- 9-- -

9 9 9 9

9 9 9 S'* ~

|

.-.." ~

'

'

SS

9

5 ML

2 ML

' 20100 '

'

' 4000

'

'

' 60100

Time (s) FIG. 59. T i m e d e p e n d e n c e o f the p o t e n t i a l d u r i n g g a l v a n o s t a t i c o x i d a t i o n o f n - S i ( 1 0 0 ) in p u r e water. T h e a n o d i z a t i o n starts w i t h h y d r o g e n a t e d surfaces.

which is attributed to a ordered Si-O-Si layer at the interface. The oxide/Si interface in fluoride-containing solution is no longer well defined. Moreover, the interface is permanently renewed because of the electropolishing behavior in such solutions [81, 171,172], and, obviously, no well-ordered interface can be formed. Furthermore, Figure 58 shows a reduction of the IR absorption in the H-O-H bending mode region around 1650 cm -1 . This finding is correlated with the increasing amount of silicon oxide from spectra 58b to 58d (film growth), which replaces the surrounding water from the electrolyte. The amount of oxide at +2.6 V (Fig. 58d) is nearly three times greater than at + 1.7 V (Fig. 58c). The time dependence of the potential during galvanostatic oxidation of an ntype Si(100) wafer in pure water is plotted in Figure 59. The anodization starts with the hydrogenated surface. The potential decreases by about 0.7 V during the first 100 s of anodic oxidation. The increase in the potential after about 200 s gives evidence of the formation of a homogeneous anodic oxide layer, and the applied potential drops across the layer (the charge flow is equivalent for 1 ML). The potential increases by about 0.25 V after the formation of 2 ML. The anodic oxides, with thicknesses of 5 and 10 ML, are completed after the potential reaches +0.8 and +2.4 V, respectively. The linear increase in the potential with oxidation time is due to an increase in the potential drop across the oxide layer. The rate of oxide formation (dU/dt) for very thin anodic oxides cannot be increased significantly by increasing the anodization current (see thick solid and dashed lines). This behavior reveals that the formation of Si-O-Si bonds is the rate-limiting step. The PL intensity of the c-Si is constant during the growth of very thin anodic oxides regardless of the current density (i.e., for low current densities). As remarked,

205

ELECTROCHEMICALPASSIVATIONOF SI AND SIGE anodic oxidation of 1 ML ii m.,m at 1 laA/cm2 ---o--- at 2 laA/cm2 ......,i ..... at 4 laA/cm2

I

0.0 -0.1 ~"

p-Si

f

(4 acm)

Lted ];t7 I hydrogenated

-0.2

""~.'~:~" ~ l l

7

>

23Q- -0.3

-0.4 -0.5 i

I

-900

,

,

I

-600

,

i

I

-300

,

,

I

~

0

,

I

300

,

,

I

600

,

,

I

i

900

U F (V) FIG. 60. Dependence of the photovoltage amplitude on the field voltage for Si surfaces covered with

thin anodic oxides (thickness about one monolayer) prepared at 1, 2, and 4/zA/cm 2. The dependence of the hydrogenated Si surface is shown for comparison.

a similar behavior has been observed for initial porous silicon formation when the PL intensity, which is a measure of the rate of dangling bond formation, is constant with increasing current (see Section 2.4). Figure 60 shows the ex situ measured dependence of the photovoltage amplitude, Upv, on the field voltage, UF, for Si surfaces covered with a very thin anodic oxide. The dependence of the hydrogenated Si surface is shown for comparison. The hydrogenated Si surface is in slight inversion because of positive charging of the surface. The slope of the Upv (UF) characteristics decreases strongly, and the band bending tends to stronger inversion after formation of the very thin anodic oxide layer. Therefore, interface states are generated, and the density of slow hole traps increases because of initial anodic oxidation. Interestingly, the corresponding density of interface states has a minimum for a current of about +2 #AJcm 2 (highest slope of Upv (UF) characteristics among the very thin anodic oxides), whereas the inversion is the strongest. Positive charge may be partially compensated for by negatively charged acceptor states (for example, complexes associated with OH- ions). The probability of formation of negatively charged complexes increases with increasing surface state density, provided that the surface is covered by an oxide. Figure 61 compares the distributions of surface states, Dit, for Si surfaces covered with anodic oxides of different thicknesses. Dit of the hydrogenated Si surface is shown for comparison. The value of D ~ i tmin of the H-terminated surface is on the order of 1011 eV -1 cm -2 (analysis of the Upv/UF data from

206

RAPPICH AND DITTRICH anodic oxidation at 1 pA/cm 2 --Z&-- 1 ML oxide - - O - - 2 ML oxide -r-I3 ML oxide

1013 E o > (D "~= 1012

p-Si (4 ~cm)

B

1013

anodic oxidation with 400 p.C/cm2

I^ ,ro0enated I, I -0.3

,

,

I 0.0

,

E-E i (eV)

1012

1 IJA/cm2, 400s --4"-- 2 laA/cm2, 200s -X-

v~>,O_O_O_.O_OI~""

1011

-~"~_.x~:~e~T

z--~Z~,..A_A..~-,,~ I anodic oxiaes I

,

I 0.3

4 IJNcm 2, 100s

-~1~- 10 I~A/cm2, 40s -0.3

J

i

ol.o

I

,

01.3 1011

E-E i (eV)

FIG. 61. Interface state density distribution (SPV analysis) for Si surfaces covered with anodic oxides prepared at an anodic current density of 4-1/zA/cm 2 (left, nominal thickness, 1, 2, and 3 ML) or as a function of the current density (right, nominal thickness, 1 ML). The dependence of the hydrogenated S i surface is shown for comparison.

Fig. 60). Dit increases after initial anodic oxidation to values in the range of 2-5 x 1012 eV-1 cm-2. The lowest Dit has been observed after anodic oxidation with 4-2/zA/cm 2. The density of states in the minimum is practically independent of the thickness of the anodic oxide, whereas the density of states in the range toward the conduction band increases remarkably with increasing thickness of the very thin anodic oxide layer. This behavior is very similar to the formation of native oxides on hydrogenated Si surfaces in air or chemical solutions [197, 247]. The development of surface states in the range toward the conduction band can be interpreted as an increase of the number of acceptor-type surface states (exchange of charge with the conduction band) with increasing thickness of the very thin anodic oxide layer. 2.5.2. PASSIVATION BY E L E C T R O N I N J E C T I O N AT C A T H O D I C P O T E N T I A L S The density of electronic states at the anodic oxide/Si interface is usually quite high (> 1012 cm-2), and thermal post-treatments are needed to improve the electronic passivation of the anodic oxide/silicon interface [27, 263]. The electronic states at the anodic oxide/Si interface depend strongly on the chemical equilibrium in which reactive surface sites such as dangling bonds are involved. It is shown in this section that the density of states at the anodic oxide/Si interface can be strongly reduced by optimizing the electrochemical equilibrium at the cathodic potential when electrons are injected into the thin anodic oxide layer. For this pur-


0

(D O > O

o r-n

decreasingpotential

--

....

increasingpotential

(NH4)zSO 4 anodic oxide formed at +8 V I

,

,

,

i

,

,

,

|

,

,

,

i

,

,

,

(a) ,

(b)

decreasing potential

0

|

i

-2 -4

J

laser diode:

tj~' I ! - w

-8

,

902nm, lOOns, 150W |

I

-6

,

,

,

I

-4

,

,

,

I

-2

,

Potential (V)

,

,

I

,

0

FIG. 62. Current-potential (a) and photovoltage-potential (b) characteristics of the anodic oxide/p-Si

structure in (NH4)2SO 4 electrolyte. The photovoltage was excited with single pulses of a laser diode. Reprinted with the permission of the American Institute of Physics [36], copyright 2001.

pose, a thin anodic oxide (thickness about 6 nm) was formed on Si(100) samples of p-type doping (resistance 1 f2 cm) in 2 M (NH4)2SO4 at 4-10 V, and the photovoltage and photoluminescence were probed stroboscopically during electron injection. Figure 62 shows the current-potential (a) and photovoltage-potential (b) characteristics of a 6-nm-thick anodic oxide in 2 M (NH4)2SO4 electrolyte for decreasing and increasing potential scan, Upot. The PV transients were excited by single light pulses from a laser diode (wavelength 902 nm, duration time 100 ns, 10 #J/cm2). Electron injection starts at Upot -- -2.1 and - 1.6 V for the branches of decreasing and increasing potential, respectively. The values of Upv amount to -1.1 and - 0 . 8 V at the potentials o f - 2 . 1 and - 1 . 6 V. Breakdown fields of 1.8 and 1.3 MV/cm can be obtained for decreasing and increasing potentials if it is taken into account that Upv corresponds to the potential drop across the p-Si sample. The flatband potential is shifted to lower Upot for increasing potential by about 0.2 V in comparison with decreasing potential. This shows that the positive charge is decreased by about 3 x 10 ll q/cm 2 after electron injection. The specific role of the potential and current in the density of nonradiative recombination defects at the interface can be investigated by switching between

208

RAPPICH AND DITTRICH >

.~

5

E~0

0

13_

_=

p-Si(100) r..I ~ H 4 ) z s ~ ,

~

i,_ L,_

i ~

-1 q

1

,O~

-

0..

,

,

I :

,

/:

o t, u

. . . .

0

,

, i

I

,

,

',

: ',

" ' ' ' '

. . . .

1000

,

I

. . . .

exc~tatwon.

337

,

I

i

(a)-

! i

,!

;

J

,

nm,

10

, ....

2000

',

t

',

I

,

,

n

~

ns

i

,,

.

3000 T i m e (s)

.

.

J

, -,"L" .

,

,

,

(b

I"

.,r

.

4000

(c)

5000

FIG. 63. Time dependence of the potential (a), current (b), and photoluminescence (c) of the anodic

oxide/p-Si structure during switching experiments between +8 V and cathodic potentials o f - 1, - 2 , - 3 , and - 4 V. The photoluminescence was excited with single pulses of a N 2 laser. Reprinted with the permission of the American Institute of Physics [36], copyright 2001.

anodic and cathodic directions of the current. Figure 63 shows an example of experiments with switching between anodic (+8 V), zero, and increasing cathodic potential. The anodic potential switch is used to create the same initial experimental conditions before the switch to the cathodic potential. The time dependence of the potential (a), current (b), and PL intensity (c) are plotted in Figure 63. The PL was excited with pulses from the N2 laser (wavelength 337 nm, duration time 10 ns, 100/zJ/cm2). The PL intensity is low during the application of the anodic potential and decreases further after switching to zero potential. The corresponding value of Ns is on the order of 1012 cm -2. The PL intensity remains almost unchanged during the application of low cathodic potentials (Upot > - 2 V). The Si surface is in strong inversion under these conditions as can be seen from Figure 62. Therefore, the influence of band bending on the recombination processes can be neglected and Ns is constant. In other words, the electrochemical equilibrium at the thin anodic oxide/silicon interface cannot be changed by a simple shift of the surface Fermi-level position from accumulation (anodic potential) to strong inversion (cathodic potential). The situation changes when electron injection becomes significant (Upot < - 3 V). The PL intensity starts to increase after switching from zero potential to a cathodic potential

209


of Upot = - - 3 V. Furthermore, the PL intensity increases with increasing time, i.e., with increasing injected charge. The PL intensity decreases again at a high anodic potential of +8 V. The increase in the PL intensity is much more significant at a potential o f - 4 V. Therefore, we can conclude that the electrochemical equilibrium at the thin anodic oxide/silicon interface is strongly affected by injected charge. The PL intensity can be increased by up to more than one order of magnitude by electron injection. Nevertheless, the PL intensity saturates at high amounts of injected charge, as plotted in Figure 64. The PL intensity saturates faster at higher cathodic potentials. This is evident for the adjustment of a common chemical equilibrium that is characterized by a given value of Ns during the injection of electrons. We have to remark that the amount of injected charge that is needed for a certain increase in the PL intensity increases with increasing density of the cathodic current. The PL intensity at zero potential also strongly increases after electron injection but reaches only about half of the PL intensity at high cathodic

> v

8

FI (a)

t t ~(NH4)2804

r--]

I-I

, m

c 0 O_

< ~

0 -8

I-/i,,I

....

I , , , , I

.

.

.

.

I ....

I,,R,,

p-si(lOO)

I

.

.

.

.

I ....

I,,,

0

v

r-"

-2

~ 23

-4

0

-6

FE

o:i =

2 ',(c) ~

N

1

t---

0

%-J kll lU excitation: 337 nm, 10 ns

tim[~ ~ 9

0

2000

4000 6000 Time (s)

8000

FIG. 64. Time dependence of the potential (a), current (b), and photoluminescence (c) of the anodic oxide/p-Si structure during switching experiments between -+-8 V and cathodic potentials of - 4 , - 6 , and - 8 V. The photoluminescence was excited with single pulses of a N 2 laser. Reprinted with the permission of the American Institute of Physics [36], copyright 2001.

210


> v .

tl:i m

(D o 13_

>

8 0

p-Si(lO0)

(NH4)2SO 4

6 nm anodic oxide

(a)

-8

o.o

zero potential branch

g -0.2 > o o

_c: -0.4 13_

(b)

4

excitation" cathodicpotential branch 337 am, 10 as imp lIP ~ I ~ 3 ql~

0

300

~

-r

600 Time (s)

900

Time dependence of the potential (a), photovoltage (b), and photoluminescence (c) of the anodic oxide/p-Si structure during switching between cathodic (-8 V) and zero potential. The experiment starts with anodic potential at +8 V. The photovoltage is shown only for the zero potential branch. The photovoltage and photoluminescence are excited with single pulses of a N2 laser. Reprinted with the permission of the American Institute of Physics [36], copyright 2001. FIG. 65.

potentials. This behavior reflects the influence of strong inversion on the surface recombination velocity. Surprisingly, the PL intensity can be further increased by switching between high cathodic and zero potential at shorter time intervals without the application of anodic potential (Fig. 64 time around 8000 s). This points out the importance of transport phenomena like drift or diffusion of ions through the oxide to establish the chemical equilibrium at the thin anodic oxide/Si interface. The electron injection can be interrupted for short times during the ongoing passivation process. This allows a correlation between Ns and the projected charge at the thin anodic oxide/Si interface (Qox). Figure 65 presents an example of such a switching experiment where the applied potential (a), photovoltage (b), and PL intensity are plotted as functions of process time. The photovoltage is shown in the case where Upot = 0 V. The p-Si surface is in inversion after switching from anodic to zero potential; i.e., there is a positive Qox of about 2 x 1011 cm -2 at the anodic oxide/Si interface. Despite the high band bending, the PL intensity is the lowest after switching from anodic to zero potential as mentioned above. Upv


p-Si(100) (NH,)2SQ

1012

O A E o

z

O3

211

I~

Ns Ns-81010cmz

1011

Oj," 0 0 Ai -

O0 /"

10 ~~

.A

.A"

slope 1 ,

0

,

,

|1

=

|

|

|

|

,

,

,I

4011

10l~

Qox (q/cm2) FIG. 66. Correlation between the concentration of nonradiative defect centers at the anodic oxide/pSi interface, Ns, and the positive oxide charge, Qox (open circles). The triangles show Ns for Nit: 8 • 1010 cm -2. The dependencies are obtained by analyzing the data of the zero potential branches in Figure 65. The slope 1 is indicated. Reprinted with the permission of the AmericanInstitute of Physics [36], copyright 2001.

at zero potential decreases from inversion to depletion during repetitive switching from cathodic to zero potential. The PL intensity increases at the same time. Hence, the electron injection neutralizes and/or anneals positive charges at the anodic oxide/p-Si interface and decreases Ns by more than one order of magnitude. This correlation between Qox and Ns is given in Figure 66 for the zero potential branch (open circles). The data are obtained from Figure 65. Ns and Qox are reduced from 1012 to 1011 cm -2 and from 2 x 101] to 2 • 101~ q/cm 2, respectively. Ns decreases monotoneously with decreasing Qox and tends to saturate at the lower values of Qox ( N s nin is about 8 x 101~ cm-2). The triangles show the dependence of Ns - N~nin on Qox. The dashed line indicates the slope 1. It can be seen that Ns - N~nin is nearly proportional to Qox for values of Qox lower than 1011 q/cm 2. The observed passivation of nonradiative recombination centers at the anodic oxide/p-Si interface is caused by injected electrons passing through the thin anodic oxide layer. The injected electrons change the chemical equilibrium at the anodic oxide/p-Si interface. There are two possible routes of chemical reaction that are induced by the injection of electrons. The first mechanism is connected with the drift of protons from the electrolyte/anodic oxide to the anodic oxide/Si interface. Injected electrons can react with protons near the anodic oxide/Si interface, and hydrogen atoms can diffuse back to the electrolyte/anodic oxide in-

212


terface. Hydrogen could passivate Si dangling bonds, which act as nonradiative recombination defects. However, this mechanism does not explain the correlation between Ns and Qox. We favor a second mechanism that is related to the high amount of water in anodic oxides and its role of destabilization of H-Si - Si3 bonds. The correlation between Ns and Qox can be well explained by this second mechanism as follows. The correlation between Ns and Qox is very similar to the negative-biastemperature instability (NBTI) of SiO2/Si interfaces [264]. On the bases of many experimental findings, Poindexter proposed the following NBTI reaction [210]: H-Si-- Si3 + H20 + h + ~ o S i - Si3 + H30 +

(5.1)

In accordance with reaction (5.1) we propose the following reaction for passivation of Qox and Ns at anodic oxide/p-Si interfaces by electron injection: ~

Si3 + [H502] + + e- --~ H - S i - Si3 + 2H20

(5.2)

Reaction (5.2) is, in principle, the inverse reaction of (5.1). Reaction (5.2) is controlled by the number of injected electrons and the amount of water present at the interface. The activation energy could be supported by the recombination energy of the electron. Reaction (5.1) dominates at anodic potentials, whereas reaction (5.2) is initiated by the injection of electrons at cathodic potentials. Ns cannot be reduced below a certain level, whereas Qox decreases further. Therefore, a competitive reaction to (5.2) that neutralizes charged [H502] + complexes and creates new oSi -- Si3 bonds should take part in the chemical equilibrium. Such a reaction could be H - S i - Si3 + [H5Oe] + + e- ~ o S i - Si3 + 2H20 + H21"

(5.3)

The passivation of the anodic oxide/Si interface is limited by reaction (5.3). Reaction (5.3) can be partially suppressed by switching between cathodic and zero potentials as shown in Figure 64. We suggest that capture times of charge and/or polarization of Si-Si bonds are important for an inhibition of local reconstruction, which should be important for the activation of reaction (5.3). 2.5.3. PASSIVATION BY PROCESS OPTIMIZATION AT ANODIC POTENTIALS

The density of nonradiative recombination centers at the anodic oxide/silicon interface is usually on the order of 1012 cm -2. However, in this section we show that Ns can be decreased by more than one order of magnitude, even during anodic oxidation in diluted fluoride solution, with the use of a certain regime of anodic oxidation in the oscillating regime. Figure 67 shows the maximum and minimum PL intensities during the anodic oxidation process of p-Si(100) in 0.1 M NH4F (pH 4.2) as a function of the applied potential. The PL intensity has the lowest value at the lowest anodic


213

FIG. 67. Maximum and minimum PL intensity of p-Si(100) during electropolishing as a function of the applied potential (0.1 M NH4F (pH 4.2)). Reprinted with the permission of Wiley VCH [168], copyright 1997.

potential (+3.5 V) during electropolishing and increases by more than one order of magnitude with increasing anodic potential in the oscillating regime. The current oscillations vanish for potentials higher than + 10 V, and the PL intensity decreases for higher potentials. The strong increase in the PL intensity cannot be explained only by the increasing potential (p-type Si is in accumulation during anodic oxidation), because the PL intensity decreases for higher anodic potentials. The main reason for the strong increase in the PL intensity should be a structural change at the anodic oxide/p-Si(100) interface [178, 189]. As outlined in Section 2.3, an anodic current transient can be observed after complete removal of an oxide layer in a diluted fluoride solution at about - 0 . 4 V. The dependence of the anodic oxide thicknesses that have been prepared on pSi(100) at different potentials is monitored by anodic current transients in Figure 68. The potential is switched from the anodic oxidation state (a, b, c: +3.5 V, +9 V, +12 V) to -0.45 V in the same solution, and the time base is scaled to the interruption of the current after the potential is switched. The onset of the anodic current transient appears at a later time with increasing oxidation potential because of the increased thickness of the oxide layer. The time needed to reach the peak maximum of the anodic current transient correlates well with the period of the current oscillations for the potential range between +4 and + 10 V, but it is slightly longer than the respective period [35, 168]. Therefore, the anodic current oscillations depend sensitively on the thickness of the anodic oxide layer. The oscillation period is given by the amount of

214


10

4

- - -:...7....7...

-

-0.45 v

O4

E

~

0_2

~

10_3

Potentials

~

ti i!,. I ,:.~.."

23

0

10.4 10 s

a

'

'

'

0

'

'

' Time

........ ..~." -..... ~ '.,

'

I

100

'

C

\ '

'

'

\ I

200

(s)

FIG. 68. Anodic current transients of p-Si(100) at -0.45 V after different anodic potentials (a, b, c:

+3.5 V, +9 V, +12 V) in 0.1 M NH4F (pH 4.2). Reprinted with the permission of Wiley VCH [168], copyright 1997.

oxide generated during one oscillation period (i.e., the oxidation potential) and by the etch rate of the electrolyte [35, 170, 179, 265, 266]. The frequency of the oscillation is proportional to the etch rate at fixed potential [ 170]. The anodic current peak that appears during the etch back of the oxide layer is caused, in our opinion, by the oxidation of partially oxidized Si atoms at the Si surface during the hydrogenation process, as proposed by Gerischer and Ltibke [ 166]. Figure 69 shows the dependence of the integrated electric charge of the anodic current transient at -0.45 V on the applied oxidation potential [ 168]. The integrated electric charge decreases with increasing oxidation potential-showing that the number of partially oxidized Si atoms at the interface decreases. It is important to note that the integrated electric charge of the anodic current transient is not correlated with the concentration of nonradiative recombination defects at the Si interface. It can be concluded that the quenching of the PL intensity is not related to the formation of partially oxidized Si surface atoms. Unfortunately, there is no direct experimental proof of this conclusion at the moment. The PL intensity correlates with the anodic current oscillations during the anodic oxidation. An example is shown in Figure 70 for p-Si(100) at +10 V in 0.1 M NH4F (pH 4.2). As usual, the maximum current corresponds to a minimum in the PL intensity and vice versa. The modulation of the PL intensity is related to a modulation of the concentration of dangling bonds at intrinsically back-bonded Si surface atoms. Breaking of Si-Si back bonds is one elementary step during the anodic oxidation of Si, and the concentration of dangling bonds at intrinsically

215


50 t'Xl

E r

O v

O O

40

O

:I.

(1)

30

to r

20

O

O

.m

o

10

O

UJ

Time I

I

1

2

I

3

I

4

O I

5

I

I

6

7

I

I

8

I

9

I

10 11

I

12 13

Potential (V) FIG. 69. Dependence of the integrated electric charge of the anodic current transient, Qt, for p-

Si(100) on the applied potential during the preceding electropolishing in 0.1 M NH4F (pH 4.2). The inset illustrates the determination of Qt. Reprinted with the permission of Wiley VCH [168], copyright 1997. 10 1.2

,~'~_

,.-i. tl)

o

1 o~ m.

0.4. 0.2 It~

u#

.

0.0 0

. 200

. 400

'~11~ I 600

~,

~t~

'

JD

I

'~11/1~.

800

0.1 1000

Time (s)

FIG. 70. Anodic current oscillations and oscillations of the PL intensity of p-Si(100) during anodic oxidation at +10 V in 0.1 M NH4F (pH 4.2). Reprinted with the permission of Wiley VCH [168], copyright 1997.

back-bonded Si surface atoms is limited by the oxidation rate. As remarked, the anodic current is not necessarily correlated with the PL intensity and therefore with the oxidation rate [ 168].

216

RAPPICH AND DITTRICH c: o

1.2

-9 o r

1.0

m n-

0.8

c

06

04

n-Si(111)

O

__.

!

A

r

'-

1000

1100

1200

1300

E o

0.4

Wavenumber (cm-~)

E

o.2

]~'~~-----~"~' gilh~ off IRreference,,,~

0.0

0

0

~60

2;0

'

I

300

'

Time (s)

4;0

'

S60

'

600

FIG. 71. Anodic current oscillations (solid line) and integrated IR absorption of the Si-O-Si bonds for

n-Si(111) in 0.1 M NaF (pH 4). The inset shows the FTIR spectra normalized to the hydrogenated Si surface.

The amount of oxide correlates with the oscillation period during anodic oxidation as shown by in situ ellipsometric measurements [267] or in situ FTIR spectroscopy [86, 92]. Figure 71 shows the time dependence of the current for n-Si(111) during two oscillation periods and the corresponding relative amounts of oxide. The inset shows the FTIR spectra normalized to the hydrogenated Si surface. The amount of oxide has a maximum (minimum) in the decreasing (increasing) part of the oscillating current. The relative change in the amount of oxide is about 50%. This finding reveals that the rate of formation of anodic oxides changes strongly during the anodic current oscillations and that the maximum of the current is not simply related to a pure injection current. As remarked, the FTIR spectra did not show any partially hydrogenated Si surface. The homogeneity of the thin anodic oxide layer during the current oscillations is important for a better basic understanding of the processes and for possible applications. The anodic oxidation in the oscillating regime can be interrupted at the two characteristic points when the oxidation rate is minimal (at the minimum of oscillation) or maximal (at the maximum of oscillation). Figure 72 shows, as an example, the anodic current transient at - 0 . 4 V (a) and PL intensity (b) for p-Si (111) in 0.1 M NH4F (pH 4) after anodic oxidation at +8 V and interruption of the oscillations in the maximum (solid lines) or minimum (dashed lines). The anodic current transient appears somewhat earlier in time after interruption in the minimum than after interruption in the maximum of a current peak. This means that the overall thickness of the anodic oxide is lower at the minimum of the cur-

217


-q

100-

p-Si(111)/0.1 M NH4F

=0"; 1 0-1 (D t._1 Q_

100 <E E

(1) ~--

10 -~

-0.4 v

10-2

(a)

-3

~- 10 0 10 .4 0

50 Time (s)

100

FIG. 72. Anodic current transient at - 0 . 4 V (a) and PL intensity (b) for p-Si (111) in 0.1 M NH4F (pH 4) after anodic oxidation at +8 V and interruption of the oscillations in the maximum (solid lines)

or minimum(dashed lines).

rent peak than at the maximum. The current transient exhibits a shoulder after an etching time of about 60-70 s, which gives evidence of a slight second maximum. This behavior points to an inhomogeneous thickness of the anodic oxide layer at the current maximum. The PL intensity decreases just after interruption of the anodic oxidation, probably because of the change in the electric field (Fig. 72b). The PL intensity increases within the first 20 s as a result of the ongoing chemical reactions, which further passivate nonradiative recombination centers at the anodic oxide/Si interface; i.e., these processes are not interrupted after the potential is switched. The PL intensity reaches a constant level with time and increases monotonously after the maximum of the anodic current transient is reached, after interruption of the anodic oxidation at the current minimum (dashed lines in Fig. 72). The PL intensity saturates when the anodic current transient levels out. The situation is slightly different for interruption of the anodic oxidation at the current maximum (solid lines in Fig. 72). In this case, the PL intensity increases in two pronounced steps after the current transient maximum is passed, which are related to the main and second slight maximum of the anodic current transient (indicated by the dotted lines). This behavior demonstrates that the thickness of the anodic oxide at the current maximum of an oscillation has two different mean values. The maximum of the PL intensity is practically independent of the time of interruption of the oxidation process.

218


A new approach has been proposed by F611 et al. for the interpretation of the development of current oscillations during anodic oscillations [268]. Frill postulated local current bursts through a closed oxide. The current bursts occur for short times, and they may under certain conditions interact space-resolved to their synchronization in time (critical potential and etch rate). Current bursts take place on the ,~ to nm scale [268], and their synchronization should be responsible for the conditioning of the anodic oxide/silicon interface. The anodic oxidation rate decreases with decreasing current, and therefore the generation rate of nonradiative defects at the Si interface exhibits a minimum. The oscillation period and the oxidation rate increase linearly with the oxidation potential at constant etch rates of the solution [168]. This shows that the thickness of the oxide layer is important for damped or sustained current oscillations. To our understanding, the microscopic reason for triggering of the synchronized current bursts could also be related to the development of a space charge region inside the anodic oxide layer due to injected charge. The condition for the development of current bursts can be changed by the structure at the anodic oxide/Si interface. The anodic oxidation in the oscillating regime should be interrupted by switching the potential to the hydrogenation state and restarting the oxidation process again to change the structure at the interface. The second cycle of anodic oxidation in the oscillating regime then starts with a microscopically smoother surface, and the probability of the development of current bursts due to structural nonuniformity on the microscopic level should be decreased. An example of anodic cycling between oxidation in the oscillating regime and hydrogenation is given in Figure 73 for p-Si(111) in 0.1 M NH4F (pH 4). The current and the value of Ns are shown on the left and right sides, respectively. The potential is switched between +8 V and - 0 . 4 V. As expected, the current at the maximum of the oscillations (starting from the second oscillation) decreases for repetitive cycling. The integrated charge of the anodic current transient decreased from the first to the second cycle, which is reflected by the reduced current density during the transient. The values of Ns at the maximum and minimum of the oscillations as well as for the hydrogenated surface decrease with each cycle. Figure 74 summarizes the results obtained from Figure 73" (i) decrease in the integrated charge of the anodic current transient (A), (ii) decreasing value of Ns at the hydrogenated Si surface down to 101~ cm -2 and lower (B, open square), and (iii) decrease in Ns at the oxidized Si surface to values below 1011 cm -2 (C) with increasing number of cycles. This demonstrates the great potential of anodic oxidation for low-temperature passivation of Si surfaces. Figure 74B also shows that repetitive cycling of electropolishing and hydrogenation does not lead to any improvement of Ns for anodic oxidation in the nonoscillating regime at +3 V (B, solid squares). We remark that the difference between the PL intensities at different potentials is not as strong for the p-Si(111) surface as it is for the p-Si(100) orientation.

219


+_,v o4vl

(cycle)

0.5

ixlO 0

o

1011

(3)

10 10

I

z

03

El.O --0.5

', i

0.0

1011

(2)

,

101~

,

0 3

1011 __0"5 0.0

~

,

.... , .... , .... , .... , .................

100 200 300 400

,.

100 200 300 400

10 ~~

Time (s) FIG. 73. Time dependence of the anodic current, I, and the density of nonradiative surface defect centers, N S, as obtained from the PL intensity of p-Si(111) in 0.1 M NH4F (pH 4) for three subsequent oxidation (+8 V)/hydrogenation ( - 0 . 4 V) cycles (1 to 3). The horizontal dotted line marks the value of N s during anodic oxidation at +3 V (no oscillations).

The values of Ns at the anodic oxide/Si interface and the hydrogenated Si surface could be, to our opinion, further decreased by optimization of the electrolyte and the potential. There is a strong indication that interruption even after the first heavy current peak, before regular oscillations start, allows a further decrease in Ns. Obviously, any current minimum that follows a strong current maximum, which is induced by well-synchronized current bursts, is characterized by a very low concentration of reactive surface sites. This state of the anodic oxide/silicon interface can be used to optimize the interface passivation. The current bursts occur preferentially at places where the electrical field is enhanced. A nonuniformity on the microscopic scale leads to the enhancement of the electrical field. The local input of energy during the current burst causes a fast local oxidation. This fast oxidation process is far from equilibrium and leads to a local reconstruction in such a way that the nonuniformity on the microscopic scale will be smoothed. The locally reconstructed interface is quite uniform and practically free of reactive surface sites, which would cause the development of new current bursts. 2.5.4.

F O R M A T I O N OF O X I D E S IN A L K A L I N E S O L U T I O N

Figure 75 shows a typical treatment of the p-Si(111) surface (bottom: current density, i; top" PL intensity). At first, the oxide-free surface (HF dipped) is anodically oxidized in 0.1 M NH4F (pH 4) at +3 V for about 15 min to obtain the

220


&

during hydrogenation at-0.4V

(A)

o E O v

~1 -0.4V

0

Time

I

A

I

1011 O4 n

v

(B)

E

nn

o

e-

E Z

o'}

10 ~~

H-term. after I-! +8V n +3V I

O4

|

v

E

to

1011

x_ d~

z

I

10 ~~

nn

El

17

I

O O Oxidation at +8V O max. Ns ~ I min. N s

O 9

I

(c) +3V

O 9

I

2 3 Number of oxidation cycle

FIG. 74. The charge that passes the electrode during etch-back of the oxide formed at +8 V (A), the minimum values of N S after hydrogenation at -0.4 V (B), and the average minimum and maximum values of NS during anodic oxidation at +8 V and +3 V (C) (values of I and PL are taken from Fig. 73). Reprinted with the permission of Elsevier Science B. V. [ 189], copyright 2000.

same starting condition for each experiment. The PL intensity is low during oxidation, where the breaking of Si-Si bonds leads to a high amount of nonradiative recombination centers (i.e., to a high defect concentration, Ns, which is about 1013 c m - 2 ) . The oxide is then etched back in the same solution by switching the potential to - 0 . 4 V, and the well-known current peak occurs, which indicates the

221

ELECTROCHEMICAL PASSIVATION OF SI AND SIGE ~.

0.4

v

0.3

o~ 0.2 C

.~

_.1 13_

p-Si(111)

Pllh -0.4V A) [ '~. ' ]9 "~ il ' ~L,....

+3V, A) - - - - - -

i 2 E-lpA/om , B) i i pH: 9.1

0.1 0.0 400

E 300

A) 0.1 M NH4F (pH 4) ,B) alkaline solution

o

10 nm) on Si or SiGe. A small amount of water (typically 0.35%) is necessary to reduce the rate of oxidation with respect to pure water, where the rate is too fast at higher anodic potentials. This leads to a worse passivation of the SiO2/Si interface [261]. The total reactions in such a solution are as follows [7]. Oxidation at the Si anode: Si 4- 2H20 + 4h + --+ SiO2 + 4H +

(6.1)

CH2 - CH2

~ 2H2CO + 2H + 4- 2e-

(6.2)

--+ O z t + 4 H + + 4 e -

(6.3)

I

I

OH

OH

2H20 Reduction at the Pt cathode: 4H + + 4 e -

--+ 2Hzt

NO 3 + H20 + 2e- -+ NO 2 + 2 O H -

(6.4) (6.5)

The efficiencies of reactions (6.1) to (6.3) are about 1%, 98%, and 1%, respectively [7]. This means that only 1% of the current leads to the oxidation reaction of Si with water to form silicon oxide. The anodic oxidation of SiGe leads to an additional reaction, Ge + 2H20 4- 4h + --+ GeO2 4- 4H +

(6.6)

where it is known that GeO2 is soluble in water [274-277]. Figure 80 shows a photograph of the front view of the quartz chamber for anodic oxidation for 3-inch (maximum) wafers (left) and a schematic drawing of the cell from a side view (right). The Si sample is sucked up onto an O-ring seal by a vacuum in the quartz tube and electrically contacted with a small A1 plate. A P t ring is used as a counter-electrode. The electrolyte in the chamber is stirred, and the temperature is controlled by a thermostat. It should be noted that the layout of an oxidation chamber is defined by process parameters so that, in principle, several Si wafers can be oxidized at the same time with the use of another type of electrochemical cell. After the anodic oxidation is completed, the oxidized sample is cleaned in HCI:H202 :H20 (1:1:6), dried under nitrogen steam, and annealed in nitrogen or forming gas up to 800~ (maximum). The oxidized samples are characterized by a great number of analytical and imaging techniques, and by capacitance-voltage (CV), surface photovoltage (SPV), and pulsed photoluminescence (PL) measurements.

226


Front and sid~ views of the electrochemical cell used for preparation of thick anodic oxides. Reprinted with the permission of Elsevier Science B. V. [27], copyright 2000.

FIG. 80.

Thermal oxidation in dry 02 steam at 1000~ for Si and rapid thermal oxidation (RTO) at 800~ for SiGe epilayers are used for comparison. 2.6.2. PREPARATION OF THE OXIDE LAYER Figure 81 shows the time dependencies of the potential and current on the temperature of the solution (A) and the initial current density at 40~ (B) during the oxidation process in ethylene glycol with 0.35% water. The oxidation starts with a galvanostatic process. In the beginning, the potential increases slowly. The increase in potential becomes faster with time because of the increasing thickness of the oxide layer. The thicker the oxide layer, the higher is the potential drop across the oxide. When the desired final potential is reached (in this case + 120 V, oxide thickness about 54 nm), the process is continued potentiostatically until the current has decayed to an almost constant value, as can be seen in the bottom part of Figure 81. Increasing both the temperature of the solution and the current density decreases the time of the oxidation process. The amount of electric charge that passes the electrode is plotted in Figure 82 as a function of the temperature of the electrolyte (A) and the initial current density (B) (values of current and time are taken from Fig. 81). The charge flow is reduced either by an increase in the

227

ELECTROCHEMICALPASSIVATIONOF SI AND SIGE final potential +120 V

>S. -E

1oo 50

A)

t:1i (

~"'-'" ,, ~ " " -"'""'~" " ~ ' ~ Y ----41

:~"~ - ~

-

~c

-

&. o

...... 0

1

'

I

1.0

'

'

i

'

-

I

'

-""

i

|-

~" 0.8 E ~o 0.6

'

i

'

I

'

i

I . . ('B!" ? " ' " ~ '......... iii~'~1 ...

t

I

..... ........

~ I

........

I

40~ i

........

I

-

........

i

initial current I-I" /

. . . . . . . . . . . . . . . . . . . . . . .

', ',

-

'.~

.........

E 0.4

,

~

1

....

47 f 64

: ~

0.2

0

'

'

i

'

4000

i

.

.

8000

.

.

.

12000

10 ~

Time (s)

,

101

,

102

8 2

:":.... J0

103

10

4

Time (s)

Time dependence of the current density and potential during galvanostatic preparation of anodic oxides (A) at different temperatures and a constant current density of 1 mA/cm2 and (B) at 40~ and different initial current densities. The final potential is 120 V (oxide thickness ---54 nm). Reprinted with the permission of Elsevier Science B. V. [27], copyright 2000. FIG. 81.

12 O4

E o

O

rO

10

8

O.Oo

(A)

40~

I mA/cm 2

4'o'do'6'o

Temperature

(~

12

10 A

Xo

o ' ~'o'2'o'3'o

(B)

o

\zx~zx____~ '~'4'~'&'1'0 initial current (mA/cm 2)

The electric charge that passes the Si anode as a function of the temperature of the electrolyte (A) and the initial current density (B) (values of current and time are taken from Fig. 81).

FIG. 82.

temperature or by an increase in the current density. The charge flow saturates at a current density of about 7 m A / c m 2 at 40~ and does not change at higher current densities (Fig. 82B), whereas the charge decreases almost linearly with the temperature of the electrolyte and shows no tendency to saturate up to 60~

228


500

E

anodicoxidation ethyleneglycol/ 0.04MKNO3

40O-

toto

E 300._~

~

~

S

9 n-Si(100) 0.35% HzO p-Si(lOO) o.6 %H~O ~

.

__

11.5 A/V

a~ 200 .-g_ x O 100 , , , - ~ " " ~ - . . ~ 4.4 Nv 0

100

200

3;0

400

5;0

Potential (V) FIG. 83. Dependence of the thickness of an anodic oxide on the final potential for anodization in

ethylene glycol with low contents of water.

The oxide thickness is independent of temperature and the current density; it depends only on the final potential and the amount of water in the electrolyte (see Fig. 83). Therefore, a constant charge at higher current densities, which is due to a reduction of the oxidation time, is a result of constant current efficiencies for reactions (6.1) to (6.3). The efficiency of oxide formation due to reaction (6.1) decreases with decreasing current densities, the competition of the other reactions overcomes this effect, and the net charge flow increases. The efficiency of oxide formation also shows a strong dependence on the temperature of the electrolyte; the efficiency at 60~ is twice that at 8~ A reduction of the charge flow reduces the overall power consumption during anodic oxidation. The oxide thickness depends linearly on the applied potential and on the amount of water in the supporting electrolyte, as can be seen from Figure 83 (4.5 and 11.5 ~ for 0.35% and 0.6 % water, respectively). 2.6.3. ELECTRONIC CHARACTERIZATION Figure 84 shows the interface state density distribution, Dit, as obtained from CV measurements. Dit is high for the as-prepared and cleaned oxide layers (about 5 x 1012 cm -2 eV -1 at midgap), regardless of the applied final potential during the growth process. Dit decreases by about one order of magnitude after drying in nitrogen at 400~ and annealing in forming gas at about 450~ This procedure leads to a Dit value at midgap of about 1-2 x 1011 cm -2 eV -1, whereas the Dit of thicker oxide layers (400 V, 180 nm, 3 x 101~ cm -2 eV -1 at midgap) is slightly below the value of the thinner layer prepared at 120 V (54 nm).

229


1013

p-Si(lO0) CV analysis

as anodized +120 V

"=='li~

retiNa

,,=,I i'= 9== o

99

O4

E 1012

T

0

after FG anneal at 450~ AA

~- 1011

(~&

A,J'A & ~i,j, A A A'AAA

0 O (if)~

0%0 CPOooo6>Cbo ooCP~176

anodized at 400 V

10 l~

FG anneal at 450~ ,

I

-0.3

,

,

I

0.0

,

,

I

0.3

E-E i (eV) FIG. 84. Energetic distribution of the interface state density, Dit, for anodically oxidized p-Si(100)

samples after different drying and annealing procedures.

The dependence of Dit and of surface nonradiative defects (Ns) of anodically prepared oxide layers is investigated by CV and PL measurements, respectively. Figure 85 shows PL transients of differently annealed anodic oxide layers prepared on the same substrate, in the same solution, at the same current density and final potential to ensure comparative conditions. The inset shows the annealing condition and the Dit value at midgap as observed from CV measurement. No CV data have been obtained for sample F6. The lower the interface state density at midgap, the higher is the decay time of the time dependence of the PL. The integrated PL intensity scales with the reciprocal of Dit (see Section 2.2). Therefore, an increase in Dit by a factor of 10 decreases the PL intensity by the same factor. The as-anodized samples have high defect concentrations of about 1013 cm -2 eV -1 and a very low PL intensity. Annealing in nitrogen (or in a more pronounced way in forming gas) decreases the defect concentration at midgap to a typical value of about 1011 cm -2 eV -] or slightly below. The oxide charges are in the range of 10 ]2 cm -2. Figure 86 shows Dit at midgap and PL intensity as a function of the annealing condition. There seems to be a plateau for the values of Dit and PL between 400~ and 600~ The forming gas anneal

230


100 -

E3

>~ 10 ~

p- Si (100)

F6- FG 800~ (no CV data)

anodically oxidized in ethylene glycol excit. 337 am, O.5 ns I =0.2 mJ/cm

~3 FFG~00:C((~~1100::I

~"~'~~

\~X ~

O3 t"

F5 - FG 600~ (1.5 1011)

RE - N2 4SO~ (7.0 101112 )

~

N3-N 2300~

as.an,

9-~ 10 .2

L N3~N6 ~F1 ~ F3 ~

.J 0..

~ F5. F

0

II

20

40

Time

1013)

6

as.an.

10 "3

)

as-oxidized (1.1

60

(ps)

I

80

100

"

120

FIG. 85. PL transients of p-Si(100) covered with an anodic oxide processed in forming gas and N2 at various temperatures. Reprinted with the permission of the American Institute of Physics [156], copyright 1999.

p-Si(100) prior anodicoxidation in ethyleneglycol

i E 1013 as anodized O 1012

> r c:i~ 1011 (a) ,

,

i

,

(b) --.-- 100 (/)

= c-

_J

n

O1

]II [ 10-2 as anodized 3C)0

~ N2gas anneal I 9 forminggas anneal 660

Annealing temperature (~

900

FIG. 86. Density of interface states, Dit, (a) and PL intensity (b) of anodically oxidized p-Si(100) as a function of the annealing condition.

decreases Dit by a factor of 10 with respect to the values obtained for the N2 anneal. Note that the samples are dried in nitrogen at 400~ before the forminggas anneal.


231

2.6.4. PASSIVATION OF S T E P S AND T R E N C H E S Figure 87 shows SEM images of trenches on Si at a tilt angle of 30 ~ (A: as prepared by CF4-plasma etching; B: thermally oxidized at 1000~ C: anodically oxidized). The as-prepared trench shows a sharp step that is also visible after thermal oxidation. The electric field-enhanced anodic oxidation leads to a constant thickness of the oxide layer at the trench and to a rounding of the step where the oxide layer is thicker. The thickness of the thermal oxide (B) is somewhat smaller at the step. A similar behavior can be observed by comparing thermal and

FIG. 87. SEM side view (tilt angle 30 ~ of concave edges of c-Si (A) before and after (B) thermal

and (C) anodic oxidation.

232


FIG. 88. SEM side view (tilt angle 30 ~ of convex edges of c-Si after (A) thermal and (B) anodic oxidation.

anodic oxidation of a inner comer of a trench, as visualized by the SEM images in Figure 88. The arrows point to the narrowing of the thermal oxide at a comer (A), which is not the case for anodic oxides (B). The kinetics during thermal oxidation that lead to smaller oxide thickness at concave and convex steps is discussed intensively by Wolters and Duynhoven [ 110, 278]. The open circles in Figure 89 represent the radius of a circle, which follows the curvature of the rounding of the step (see inset), as a function of the applied potential (oxide thickness) as obtained from SEM images (like Fig. 87). The ratio of x / r is always constant and amounts to about 0.4. This means that the anodic potential at steps is enhanced by about 40% with respect to the applied potential, which leads to a 40% thicker oxide layer at the step. Recall that the thickness of anodic oxides depends linearly on potential (Fig. 83). This result is not surprising, because it is known that the potential is increased at steps and tips. In addition, such a constant oxide thickness and rounding of steps after anodic oxidation leads to a high breakdown voltage of about 12 MV/cm, whereas the breakdown voltage of thermal oxides is only about 8-10 MV/cm [270, 273]. The surface smoothing effect during anodic oxidation can easily be seen from AFM images presented in Figure 90. This figure shows microcrystalline silicon surfaces before and after anodization (A: as-deposited layer; B, C: after oxidation at + 100 V and + 160 V). The oxide layer (thickness about 50 nm (B) and 90 nm (C)) has been etched back by 2% HF before image recording. The surface of the microcrystalline silicon becomes smoother with increasing oxidation potential.

233


n-Si(100) / ethylene glycol/0.04 M KNO3 / 0.35% H20

40

E t'-"

x/r=0.4

...._.~__-

30 X

v

P

20

I --

t

t

10

5'o ....

i

,

|

|

,

,

,

2;0

,

Potential (V) FIG. 89. Dependence of the radius of rounded concave shapes on the potential during anodic oxidation of Si edges (ratio x / r is constant, ~0.4).

Figure 91 shows a highly structured Si surface (poly-Si) covered with a uniform anodic oxide layer, even in the very narrow tails and holes, which can be reached very well by the liquid electrolyte contact. This procedure is also applicable to repassivation, especially of etched structures on silicon, as presented in Figure 92. First, grooves are opened in a thermal oxide on Si down to the Si surface. These grooves are then etched by KOH, which leads to an underetching of the thermal oxide layer, whereas the etch rate for Si is much faster than that for the oxide (Fig. 92A). The partly free-standing oxide layer can be seen at the top of the image. Such etched structures could then be completely repassivated by anodic oxidation due to the liquid contact (Fig. 92B). The anodic oxide layer connects very well to the underetched thermal oxide layer above.

2.7. Enhanced Passivation of SiGe by Anodic Oxidation SiGe is of great interest for modem semiconductor devices because of the enhanced capability of bandgap engineering [279] and the increased mobility of charge carriers created by the incorporation of Ge into the Si lattice as used in heterobipolar transistors [28, 29] or metal-oxide-semiconductor (MOS) structures [30]. Thermal oxidation of SiGe alloys above 600~ leads to Ge segregation at the oxide/SiGe interface [31 ] and relaxation of the strained SiGe lattice of an epitaxially grown thin film of SiGe on c-Si [33]. These processes induce defects in

234


FIG. 90. AFM images of microcrystalline silicon surfaces (A: as-deposited layer, B, C: after oxidation at + 100 V and + 160 V, respectively). The oxide layer (thickness about 50 nm (B) and 90 nm (C)) was etched back by 2% HF before image recording.

the crystal lattice and at the interfaces, which strongly affect the electronic behavior of the device [32, 33]. Therefore, passivation of modem devices on the basis of SiGe requires methods of low thermal budget processing below 500~ [29]. Among these are plasma-assisted oxidation techniques [280-282], ion beam deposition [283], oxidation at high pressures [245], and anodic oxidation [32, 33].


235

SEM side view (tilt angle 30~ of a highly structured surface region of polycrystalline Si after anodic oxidation.

FIG. 91.

Some of these techniques involve new problems. For example, plasma oxidation is usually performed in a UHV chamber where electrons and oxygen ions are excited and speeded up in the direction of the positively polarized SiGe samples. The deposition temperature is low (

Thin Films, Vol. 27: Non-crystalline films for device structures

Thin Films for Optical Systems

Thin films for optical systems

Ferroelectric Thin Films

Photoreactive Organic Thin Films

Photoreactive organic thin films

Preparation of Thin Films

Photoreactive Organic Thin Films

Waveguide Spectroscopy of Thin Films, Volume 33 (Thin Films and Nanostructures) (Thin Films and Nanostructures)

Nanostructured Thin Films and Surfaces

Order in thin organic films

Thin films of soft matter

Metal Based Thin Films for Electronics

Interface Controlled Organic Thin Films

Adhesion Aspects Of Thin Films

Rare Earth Oxide Thin Films

Materials Science of Thin Films

Adhesion Aspects of Thin Films

Order in Thin Organic Films

Thin Films and Heterostructures for Oxide Electronics

Organic Thin Films for Photonic Applications

Ionized Physical Vapor Deposition, Volume 27 (Thin Films)

Field Guide to Optical Thin Films (SPIE Vol. FG07)

Principles of Vapor Deposition of Thin Films

Complex Wave Dynamics on Thin Films

Handbook of Thin Films, Five-Volume Set

Microstructure evaluation of thin films during growth

Solid surfaces, interfaces and thin films

Domains in Ferroic Crystals and Thin Films

Handbook of Thin Films, Five-Volume Set

Solid Surfaces, Interfaces And Thin Films

Thin Films, Vol. 27: Non-crystalline films for device structures

Thin Films for Optical Systems

Thin films for optical systems

Ferroelectric Thin Films

Photoreactive Organic Thin Films

Photoreactive organic thin films

Preparation of Thin Films

Photoreactive Organic Thin Films

Waveguide Spectroscopy of Thin Films, Volume 33 (Thin Films and Nanostructures) (Thin Films and Nanostructures)

Nanostructured Thin Films and Surfaces

Order in thin organic films

Thin films of soft matter

Metal Based Thin Films for Electronics

Interface Controlled Organic Thin Films

Adhesion Aspects Of Thin Films

Rare Earth Oxide Thin Films

Materials Science of Thin Films

Adhesion Aspects of Thin Films

Order in Thin Organic Films

Thin Films and Heterostructures for Oxide Electronics

Organic Thin Films for Photonic Applications

Ionized Physical Vapor Deposition, Volume 27 (Thin Films)

Field Guide to Optical Thin Films (SPIE Vol. FG07)

Principles of Vapor Deposition of Thin Films

Complex Wave Dynamics on Thin Films

Handbook of Thin Films, Five-Volume Set

Microstructure evaluation of thin films during growth

Solid surfaces, interfaces and thin films

Domains in Ferroic Crystals and Thin Films

Handbook of Thin Films, Five-Volume Set

Solid Surfaces, Interfaces And Thin Films

Recommend Documents