close

Вход

Забыли?

вход по аккаунту

?

Investigation of active channel formation processes in gallium arsenide by photoinduced microwave reflectometry

код для вставкиСкачать
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI
films the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter face, while others may
be from any type of computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality
illustrations and photographs, print bleedthrough, substandard margins,
and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete
manuscript and there are missing pages, these will be noted. Also, if
unauthorized copyright material had to be removed, a note will indicate
the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand corner and
continuing from left torightin equal sections with small overlaps. Each
original is also photographed in one exposure and is included in
reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6" x 9" black and white
photographic prints are available for any photographs or illustrations
appearing in this copy for an additional charge. Contact UMI directly
to order.
UMI
University Microfilms International
A Bell & Howell Information Company
300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA
313/761-4700 800/521-0600
•V
Order Number 9302686
Investigation of active channel formation processes in gallium
arsenide by photoinduced microwave reflectometry
Heimlich, Michael Craig, Ph.D.
Rensselaer Polytechnic Institute, 1992
Copyright ©1992 by Heimlich, Michael Craig. All rights reserved.
UMI
300N.ZeebR<L
Ann Arbor, MI 48106
INVESTIGATION OF ACTIVE CHANNEL FORMATION PROCESSES IN
GALLIUM ARSENIDE BY PHOTOINDUCED MICROWAVE REFLECTOMETRY
by
Michael Craig Heimlich
A Thesis Submitted to the Graduate
Faculty of Rensselaer Polytechnic Institute
in Partial Fulfillment of the
Requirements for the Degree of
DOCTOR OF PHILOSOPHY
Major Subject: Electrical Engineering
Approved by the
Examining Committee:.
Ronald J. Gumlann, Thesis Advisor
d
<y~
ose M. Borrego, Member
Edward W. Maby, Member
Peter D. Persans, Meihber
Mephber
^
Joho^vaughan, Mem
Rensselaer Polytechnic Institute
Troy, New York
August 1992
© Copyright 1992
by
Michael C. Heimlich
All Rights Reserved
ii
CONTENTS
USTOFTABLES
viii
LIST OF FIGURES
ix
ACKNOWLEDGEMENT
xiii
ABSTRACT
xv
1. INTRODUCTION
1
1.1 Motivation
1
1.2 Ion Implantation and Activation Annealing of GaAs
2
1.3 Characterization of In-Process GaAs
4
1.4 Thesis Scope
7
2. TECHNICAL BACKGROUND
9
2.1 Introduction
9
2.2 Microwave Photoconductivity Techniques
10
2.2.1 Overview of Others'Work
10
2.2.2 Microwave Reflectometry
11
2.2.3 GaAs Steady State Photoconductivity
17
2.3 UndopedSILECGaAs
28
2.4 Active Channel Formation Processing
35
2.4.1 Ion Implantation in GaAs
35
2.4.2 Furnace Annealing
37
2.4.3 Rapid Thermal Annealing
39
iii
2.5 Stoichiometrically-Constrained Defect Modeling
41
2.6 Non-Contacting Characterization Techniques
SO
2.6.1 Photoluminescence
SO
2.6.2 Infra-Red Transmittance
55
2.6.3 Raman-Scattering
57
2.7 Summary
58
3. PIMR SYSTEM AND MODELING ENHANCEMENTS
61
3.1 Introduction
61
3.2 Experimental Set-up
62
3.2.1 The Basic System and Its Operation
62
3.2.2 Two-Dimensional Mapping
71
3.3 System Performance
76
3.3.1 Dark Microwave Patterning Effect
77
3.3.2 Measurement Attenuation by Thin Conducting Layers . . .
87
3.3.3 Repeatability
94
3.4 Phenomenological Model
96
3.4.1 Undoped SI GaAs Characterization by Peak PIMR Transient
Response
97
3.4.2 Extension for Undoped SI GaAs with an Ion Implanted
Channel
109
3.4.3 Extraction of Diffused Net Shallow Acceptors
114
3.4.4 Extended Phenomenological Model Summary
117
iv
4. EXPERIMENTAL RESULTS AND DISCUSSION FOR PROCESSED GAAS
120
4.1 Typical Processing Sequence
120
4.2 Experimental Set-up
129
4.2.1 Furnace Annealing
129
4.2.2 Rapid Thermal Annealing
150
4.2.3 Ion Implantation Channeling
136
4.2.4 Coimplantation
158
4.3 Summary of Experimental Results
162
4.3.1 PIMR and Annealing
162
4.3.2 PIMR and Ion Implantation
163
5. POINT DEFECT DYNAMICAL INTERPRETATION
166
5.1 Introduction
166
5.2 Anneaing Phenomena
167
5.2.1 Annealing Data Review
167
5.2.2 Analysis of EL2 Loss
171
5.2.3 Analysis of Net Challow Acceptor Concentration
. . .
174
5.2.3a Nitride Capped RTA
174
5.2.3b Uncapped RTA
176
5.2.3c Silox Capped RTA
183
5.3 Ion Implantation Phenomena
187
5.3.1 Ion Implantation Activation with RTA
v
187
5.3.2 Substrate Effects
195
5.3.3 B Coimplantation Effects
198
5.4 Summary
203
6. SUMMARY AND CONCLUSION
206
6.1 PIMR: System Improvements and Functional Characterization
. .
206
6.2 Experimental Results and Process Monitoring and Development . .
209
6.3 Model Development and Extensions
212
LITERATURE CITED
215
Chapter 1
215
Chapter 2
217
Chapter 3
220
Chapter 4
221
Chapter 5
222
APPENDICES
225
A. PIMR Background Interference Removal Scheme
225
B. Simulation Software
226
B1. BBG Peak Transient Response using Three Defect Levels:
C Source Code
226
B2. ABG Peak Transient Response: C Source Code
C. Processing Procedures
. . . .
228
234
CI. GaAs Wafer Cleaning and Etching
234
C2. Silox and Nitride Deposition
234
vi
C3. RTA
235
C4. Rffi Silox and Nitride Strip
235
vii
LIST OF TABLES
Table 2.3.1
Dopant States of EL2 Elemental and Bidefect Constituents
. . .
33
Table 3.4.1
Typical Undoped SI GaAs Parameters for Simulations
. . . .
92
Table 4.1.1
PIMR Results for Basic Wafer SeriesfromBoule 9127 . . . . .
102
Table 4.2.1
Susceptor RTA Measured Results
147
Table 4.2.2
Summarized C-V Results for Channeling Experiment
151
viii
LIST OF FIGURES
Figure 2.2.1 Basic Microwave Reflectometry System
12
Figure 2.2.2 Two-Level Model of Undoped SI GaAs
21
Figure 2.2.3 Three-Level Model of Undoped SI GaAs
26
Figure 2.2.4 Franz-Keldysh Effect in GaAs
27
Figure 2.5.1 Morrow Activation Model Computed Results
46
Figure 2.6.1 PL Measurement of SI GaAs at 10 K
53
Figure 2.6.2 Example of SPL Measurement
54
Figure 3.2.1 Conceptual PIMR System
63
Figure 3.2.2 PIMR System
66
Figure 3.2.3 Tapered-tip Antenna in PIMR Test Facility
68
Figure 3.2.4 Detector Response Curve
69
Figure 3.2.5 Scan Pattern for Automated PIMR Linear Scan Procedure
Figure 3.2.6 Scan Pattern for Automated PIMR 2D Mapping Procedure
. . . .
73
. . . . 74
Figure 3.3.1 Typical Maps for Undoped SI GaAs
78
Figure 3.3.2 DC Maps for Materials Systems with Various Conductive Properties
and Geometries
79
Figure 3.3.3 Effect of Wafer Thickness on DC Patterning Effect for 3 inch,
Undoped SI GaAs
82
Figure 3.3.4 Microwave Patterning Effect Simulation for One Quarter of a 3 inch,
SIGaAsWafer
84
ix
Figure 3.3.S FFT-Corrected and Measured PIMR Scans for Undoped GaAs
Figure 3.3.6 Microwave Antenna Coupling to Conducting Channel
Figure 3.4.1
.
. . . .
BBG Peak Transient Photoconductance Simulation
Simulated ABG Peak Transient Photoconductance vs N^
90
99
Figure 3.4.2 ABG Numerical Solution Validation Tests
Figure 3.4.3
88
101
. . .
104
Figure 3.4.4 Simulated ABG Peak Transient Photoconductance vs N^
and Mobility
Figure 3.4.5
105
Normalized Simulated ABG Peak Transient Photoconductance
vsSRV
106
Figure 3.4.6 Measured ABG PIMR and Simulated ABG Peak Transient
Photoconductance vs Carbon Concentration
107
Figure 3.4.7 ABG and BBG Laser Absorption Depth Relative to Channel
Thickness
Ill
Figure 3.4.8
Surface Depletion Region with a Channel
Figure 4.1.1
PIMR Maps for Unannealed Wafer with Two Half Implants
112
. .
124
Figure 4.1.2 ABG and BBG PIMR versus Etch Depth for Wafer 9127-062 . .
127
Figure 4.2.1
131
PIMR and Sheet Resistance Measurements for FA Experiment . .
Figure 4.2.2 Measured Temperature Profile of Furnace Annealing Oven . . .
133
Figure 4.2.3 Uncapped RTA PIMR Results
137
Figure 4.2.4
Silox-Capped RTA PIMR Results
139
Figure 4.2.5
Diced GaAs Samples Photographed with Standard 35mm Flash
Lamp
140
x
Figure 4.2.6 Photoluminescence Scans of Uncapped and Silox-Capped RTA
GaAs Samples
142
Figure 4.2.7 Nitride-Capped RTA PIMR Results
144
Figure 4.2.8
Implanted and Nitride-Capped RTA Results
145
Figure 4.2.9
BBG PIMR Response versus Sheet Resistance for the Channeling
Experiment
153
Figure 4.2.10 Corrected ABG PIMR Response versus Sheet Resistance for the
Channeling Experiment
154
Figure 4.2.11 Channeling Experiment Etching Results
155
Figure 4.2.12 Modeled and Measurement-based PIMR vs Calculated Sheet
Resistance
158
Figure 4.2.13 B Coimplantation Measurements as a Function of Coimplant Dose . 160
Figure 5.2.1
Normalized Defect Concentrations vs Temperature for
Uncapped, Nitride- and Silox- Capped RTA
169
Figure 5.2.2 EL2 Reaction Rate vs. 1/kT for Measured and Modeled Data . . .
173
Figure 5.2.3 Corrected and Normalized Shallow Donor Concentration vs.
Temperature for Uncapped RTA
177
Figure 5.2.4 \ ^ s Diffusion Length vs. Temperature
180
Figure 5.2.5
181
Surface V^ Concentration vs. Temperature
Figure 5.2.6 Corrected and Normalized Shallow Acceptor Concentration
vs. Temperature for Silox-Capped RTA
Figure 5.2.7
Surface VQ & Concentration vs. Temperature
xi
184
186
Figure 5.3.1
Normalized Nitride-Capped RTA BBG PIMR vs. Temperature
Figure 5.3.2
Normalized Activation vs Annealing Temperature for
PIMR and Burial's Measured Data
. . 189
190
Figure 5.3.3 Normalized Nitride-Capped ABG PIMR versus Temperature. . . . 192
Figure 5.3.4 Modeled Sheet Resistance vs. B and EL2 Defect Concentrations . . 197
xii
ACKNOWLEDGEMENT
I wish to express my gratitude to Dr. R. J. Gutmann for his guidance,
encouragement, perseverance, and friendship. Dr. J. M. Borrego and Dr. P. D. Persans
have been very helpful in terms of suggested areas of inquiry and additional testing.
Gratitude is also expressed to Dr. E. W. Maby for his participation in the committee as well
as his insight into ion implantation. Special thanks and appreciation to Dr. J. Vaughan for
his never ending supply of GaAs wafers and interest in the author, PIMR, and RPI.
During the course of my studies at Rensselaer, financial support has been provided
by many institutions. M/A-COM and its employees at the Lowell ASD facility never failed
to provide access to GaAs material, fabrication facilities, and technological experience. I
am forever indebted to LeNore Kerber and Dan Seielstad for their personal and
professional commitment to this work. I would also like to thank the ECSE Department
and the New York Center for Advanced Technology for their financial support of myself
and the PIMR research. The interest and encouragement of IBM and its Yorktown Heights
staff--Dr. J. M. Halbout and Dr. G. Arjavalingam-was of great importance in providing
breadth to my studies.
The support of Pacific Monolithics during the preparation of this manuscript is
acknowledged. Discussions with Dr. B. Kendall proved fruitful in developing some of the
finer points of the PIMR modeling.
I thank my labmates, Sanjay Khan, Nitin Jain, and Ted Letavic, for their hours of
questions and comments which have honed my skills as a researcher. They have set a
xiii
standard of excellence for which I have strived, and will continue to do so. I also wish to
recognize Eugene Atwood for his efforts in providing me with the additional bits of
information so crucial in finishing this thesis.
And finally, to my wife, Luan, without whom my dream would have been
unrecognized and unrealized, and this thesis much longer in coming.
xiv
ABSTRACT
Photoinduced Microwave Reflectometry (PIMR) is used for process monitoring and
material characterization with undoped, semi-insulating GaAs processed through ion
implantation activation annealing.
The PIMR technique measures the peak
photoconductivity transient induced by either of two pulsed lasers, one with an energy
above the GaAs bandgap (ABG) and the other below (BBG), through the corresponding
loss in a coincident, CW microwave signal reflected from the GaAs wafer. The PIMR
system is automated and generates two-dimensional maps of up to 14,000 points for a
three-inch diameter wafer with a repeatability of approximately 1%.
Extraction of point defect information proceeds in two steps using a
phenomenological model of PIMR. The measured PIMR data for a wafer with a silicon ion
implanted channel is corrected for microwave attenuation, optical absorption, and altered
surface recombination velocity so that a direct comparison can be made with semiinsulating starting material. Defect concentrations are then extracted from the
photoconductivity based on separate solutions to the continuity equation for the ABG and
BBG excitation. Three defect levels, the singly and doubly ionized EL2 deep donor levels
and the carbon shallow acceptor, are required to adequately model the transient. Computer
simulations based on this three-level model agree with both steady-state calculations and
carbon acceptor concentration as measured by FTIR. This extended phenomenological
model (EPM) suffices for all of the material tested here.
A variety of processes and processing phenomena are investigated with PIMR. A
xv
series of wafers representing each aspect of MMIC processing through channel activation
validates the correction procedures used with the EPM. Data for an implant-only sample
from this series also suggests that partial implant activation occurs prior to annealing, while
etching experiments for an implanted and annealed wafer showed evidence of an
augmented defect structure below the channel. A temperature gradient in the furnace anneal
during tests with variable arsine overpressure correlates well with trends in the bulk EL2
concentration and masked any effect of the arsine at the wafer surface. Annealing
temperature dominates the variations seen in rapid thermal annealing (RTA) experiments,
both with and without implants. Although capping technique is a factor, RTA generally
yields samples with PIMR results similar to those from conventional furnace annealing.
Additional experiments with channeled implants suggest that PIMR can be used to screen
wafers with discrepant trapping phenomena, while boron coimplantation induces changes
in sheet resistance and PIMR which are a function of implant dose and energy.
A point defect dynamical interpretation of these experiments begins by extracting
the point defect concentrations with the EPM and proceeds with a consideration of the
process variables for both annealing and implant activation. The RTA temperature
dependence of EL2, independent of capping, agrees with experimental values for point
defect processes related to the dissociation and reformation of AsQ a \^ s V(j a . The net
shallow acceptor population for nitride-capped RTA evolves in a manner compatible with
this EL2 model while silox-capped and uncapped RTA produce excessive Ga- and As-loss,
respectively, at the surface. Experimentally, EL2 differences from boule to boule are found
to have a relatively minor effect on sheet resistance for MESFETs, as predicted by a
stoichiometrically-constrained, point defect activation model. PIMR data for implants
xvi
activated with RTA at various temperatures fails to agree with this activation model unless
the model is augmented by the addition of a subchannel defect arising from a reaction
between the dissociated EL2 and the implant damage. Based on this and other
experiments, VQSP^AS *S a likely candidate for the subchannel defect. Further additions to
this model explain the sheet resistance and PIMR results for coimplants of B, by requiring
the implanted B to inhibit the subchannel defect generation and mitigate the silicon
activation.
xvii
CHAPTER 1
INTRODUCTION
1.1
Motivation
The metal-semiconductor field-effect transistor (MESFET) has become the dominant
device in analog and digital GaAs integrated circuit (IC) technology. For economic
reasons, ion implantation is the method of choice in today's manufacturing environment for
introducing the doped channel layer into the first micron of the substrate surface.
However, several effects associated with implantation into semi-insulating (SI) GaAs
substrates limit MESFET performance. For example, digital IC yield relies heavily on
threshold voltage control which is hampered by inhomogeneities in the GaAs substrate and
non-uniformity introduced by implantation and subsequent processing [Miyazawa, Lee,
Saito]. In analog applications, MESFETs have been found to be plagued by low frequency
oscillations, transconductance dispersion, excessive noise, premature power saturation,
and switching gate lag [Wager, Fujisaka, Ladbrooke, Immorlica 1980a, Yeats, Soares].
While analog ICs are not as sensitive to threshold voltage, controlling parameters related to
threshold voltage contributes to greater reproducibility of higher performance device and
circuit designs.
Many factors contribute to the severity of these effects, but significant IC yield
improvements can be obtained by an assessment of the SI GaAs wafer before and after the
ion implantation and activation anneal processes [Holmes, Williams], particularly as more
1
emphasis is placed on yield enhancement through stringent IC process control.
Furthermore, a study of the properties of a MESFET should incorporate not only channel
depth and doping but also bulk SI wafer thermal history and native defect content
[Lagowski and Gatos], unannealed ion implantation damage [Mircea], and the interaction
of the bulk and channel during the anneal [Sato]. This bulk/channel interface, or
subchannel, region is poorly understood yet influences many of the problems in ion
implanted MESFETs.
Of great importance, then, to the economic fabrication of GaAs ICs is a diagnostic to
monitor the quality of the channel, subchannel, and bulk from the start of IC processing
through completion of the activation anneal. Non-destructive screening of the active layer
at these early stages of fabrication can remove questionable material before significant
resources are invested in lithographic and metallization steps. Photo-Induced Microwave
Reflectometry (PIMR) is a characterization technique which has been shown to have the
capability of non-destructively discerning subtle differences among wafers before and after
ion implantation and activation annealing. Improvement of the PIMR system and its
application to the investigation of a variety of IC-related fabrication processes are the focus
of this thesis.
1.2 Ion Implantation and Activation Annealing of GaAs
Ion implantation and subsequent activation annealing by furnace or rapid thermal
techniques have proven effective in GaAs IC technology; however, several drawbacks are
associated with this technology. The implants typically used for channel doping introduce
extensive damage in the first few tenths of a micron from the wafer surface which can
2
range from simple point defects to total amorphization of the crystal structure. In fact,
unannealed implants of a neutral species can be used for device isolation. The electrical
properties of an implanted channel are not a function of the implant parameters alone and
will be influenced by the SI substrate properties [Anholt 1988]. Coimplants of various
types have been used to limit this wafer dependence and alter the channel activation, but
have been met with varied results [Mickanin, Sadler, Canfield, and McNally].
An implant is typically identified by the implant species, dose, and energy. Depth
profiles of implants into GaAs do not typically conform to the LSS theory [Anholt 1989,
Tabatabaie-Alavi] but are somewhat deeper than LSS predictions due to implanted ions
colliding with lattice atoms and scattering statistically into some of the larger gaps between
the GaAs crystal planes [Christel].
An annealing process follows the implant to produce device quality channel mobilities
and move the implanted dopant to active, lattice positions. Furnace annealing (FA) of an
entire cassette of wafers for about 30 minutes at 800 "C to 850 °C in a gaseous arsenic
environment (which tends to control arsenic loss from sublimation) is mostly used in
foundries today. The heating and cooling of FA wafers is done gradually over several
minutes. In contrast to FA, rapid thermal annealing (RTA), one wafer at a time, employs
gradients as high as 100 "C/sec with the high temperature portion of RTA lasting only a few
seconds at 900 *C to 950 °C. Arsenic loss is limited by "capping" with a silicon nitride or
silicon dioxide layer, "proximity" placement of a GaAs or silicon wafer on top of the
implanted side of the sample wafer, or enclosing the wafer in a carrier, or susceptor, whose
interior has been charged with arsenic. Uncapped techniques are used where the loss of
arsenic can be tolerated.
3
Not surprisingly, the subchannel defect structure and population, and the activation
efficiency is very different between FA and RTA approaches to activation annealing [Bindal
1989a, and Fang]. Furthermore, neither process produces ideal results and both have
deleterious inherent limitations. Long anneal times and insufficient arsenic overpressure
for FA can significantly alter the properties of the bulk SI material as deep as tens of
microns from the surface [Ohkubo]. Variations in the thermodynamic equilibrium
conditions at opposite ends of a furnace can cause wafer-to-wafer variations within a FA
run and limits FA's throughput advantage over RTA. RTA typically produces lower quality
channels mainly because of the higher stresses in the latter [Bindal 1989a, and de Souza
1989]. Activation and defect structure in processes using RTA are also dependent on the
capping technique [de Souza 1990].
In this study, both RTA and FA are used for annealing implanted channels. FA
provides a stable process base for evaluating the PIMR technique and process variations,
like channeling and coimplantation. Samples with other parameter variations, such as
temperature variation, arsenic loss, or gallium out-diffusion, are created by different RTA
techniques to validate existing PIMR models and form the foundations for newer ones.
1.3 Characterization of In-Process GaAs
In Monolithic Microwave Integrated Circuit (MMIC) processing, the GaAs wafer can
be measured for several important parameters in a variety of ways at many steps in the
process even prior to the formation of any device structures. Many of these
characterization techniques are destructive or, at the very least, contacting. Non-destructive
wafer testing is preferred for several reasons: expensive GaAs wafers are not wasted or
4
destroyed in acquiring the data, extra monitor wafers do not have to be included with a
fabrication run, and additional process steps, solely for the purpose of test sample
preparation, can be eliminated. Even in cases where test structures must be fabricated,
non-destructive methods lend themselves more readily to monitoring wafer uniformity by
two-dimensional (2D) wafer mapping.
Starting material, typically undoped SI liquid encapsulated Czochralski (LEC) GaAs,
is characterized for electrical data and material defects usually with samples from the seed
and tail regions of a GaAs ingot. Electrical parameters, such as resistivity and mobility can
be determined by destructive Hall and magnetoresistance measurements [Look]. Point
defects which determine the electrical parameters are routinely measured by infrared
transmittance (IRT) for bulk phenomena or photoluminescence (PL) for defects near the
surface. IRT detects defects by their optical absorption spectra in the near infrared while
PL employs almost the opposite concept and identifies levels by the radiated spectrum from
electronic recombination after optical excitation. Dislocation density is especially important
in digital processes [Miyazawa] because of its indirect effect, via gettering, on electrical
uniformity. Dislocations are monitored by a destructive etch method, etch pit density
(EPD).
After implantation, the dopant profile can be determined destructively by Secondary
Ion Mass Spectroscopy (SIMS). SIMS combines a neutral ion sputter etch with mass
spectroscopy to identify impurities and their concentration as a function of depth into the
wafer. Raman scattering is a method of measuring the active phonon modes in a sample.
The degree of amorphization in the active layer region from implant damage can be
determined by Raman scattering through the presence of phonon modes forbidden by the
5
symmetry of a pristine sample [Holtz]. Raman scattering can be used after annealing to
indicate the level to which the crystal has been repaired relative to the two extremes of
before and after implantation [Wagner].
The resistivity of the implanted layer after activation annealing can be measured in any
of a number of contacting techniques. A capacitance vs. voltage measurement (C-V)
provides information on the activated carrier concentration and profile. Point defects which
are active as hole or electron trapping centers can be detected and quantified by Deep Level
Transient Spectroscopy (DLTS) after fabricating a diode structure. PL can be used here to
differentiate between the amphoteric states of the implanted silicon as well as tracking other
defects within the channel [Bindal 1989b]. FET structures are also used since they can
provide the full range of parameters needed for modeling the MESFET itself [Immorlica
1980b].
Noncontacting methods are also used after implant activation. The sheet resistivity can
be measured by an eddy current technique [Miller]. Changes in the bulk of the wafer,
away from the surface, brought about by the activation anneal can be determined by IRT
[Dobrilla]. Surface related measurements, like PL, can be used with the unimplanted, or
backside, of the wafer assuming that no contamination or arsenic out-diffusion occurs here.
While the above is by no means an exhaustive list of the measurements which could be
performed through these first few steps of GaAs IC processing, the majority of these
measurements are rarely done due to the need for expediency in a production facility and
the expense of acquiring the data and needed equipment. More importantly here, none of
these techniques allows for characterization of both the bulk and surface, once the channel
is formed, while still being useful for SI wafer characterization prior to implantation.
6
1.4
Thesis Scope
The main points to be emphasized in this work are: 1) the development of the PIMR
system into a more useful tool in monitoring GaAs wafers; 2) the application of the PIMR
technique to monitoring GaAs through various aspects of processing prior to MESFET gate
definition; and 3) an understanding of the point defect dynamics responsible for the
observed PIMR variations over process. Toward these ends, we begin in Chapter 2 by
reviewing microwave reflectometry and its use with steady state photoconductivity
measurements, the electrical and structural properties of SI GaAs after crystal growth and
boule annealing, the effects of ion implantation and annealing processing on these
properties, and the modeling and characterization of these material systems. The PIMR
system is presented in greater detail in Chapter 3 with particular attention to twodimensional (2D) wafer mapping and transient photoconductivity of undoped SI GaAs with
and without a conducting channel.
The experimental results are given in Chapter 4 along with some generalized
discussion as to how these results relate to one another and to the specific processing
involved. These results are derived from controlled wafer test sets composed of adjacent or
nearly adjacent wafers from a single boule which were processed in different manners.
The experiments carried out on the controlled wafer test sets are divided along various
process steps and phenomena (such as FA, RTA, channeling, and coimplantation) to
exemplify specific aspects of PIMR, as well as to provide more detailed data versus
process parameters for the point defect models discussed in Chapter 5.
Model development is presented in two places in this manuscript.
A
phenomenological model of PIMR for processed SI GaAs wafers through active channel
7
formation is discussed in Chapter 3. This model facilitates the extraction of point defect
concentrations from the PIMR data. The modeling in Chapter 5, on the other hand, builds
on the phenomenological model by relating the variations in the extracted point defect
concentrations to process variables via point defect dynamics.
Chapter 6 summarizes the thesis with conclusions on this work as well as suggestions
for future effort. The appendices provide details on simulation software and GaAs
processing procedures.
8
CHAPTER 2
TECHNICAL BACKGROUND
2J
Introduction
A variety of characterization techniques are used in evaluating the electrical properties
and defect structures of GaAs, but few of these have made their way onto the IC
production line. Three reasons for this are:
1. The contacting or destructive aspects of most methods.
2. Special preparation or measurement environment for the sample under test
(SUT).
3. Lack of depth sensitivity due to spectroscopic restrictions.
PIMR is not hampered by these constraints and is also well suited for use as a
characterization tool. The purpose of this chapter is to present a foundation for
understanding PIMR and its use in GaAs process monitoring and material characterization.
In this chapter, microwave photoconductivity methods and the underlying principles
of microwave reflectometry are reviewed. GaAs photoconductivity as it relates to
microwave detection schemes is introduced and forms the basis for discussing the PIMR
system in chapter 3. Subsequent sections in the present chapter look at the defect structure
of starting material, the effect of specific processes on this structure, and a model which
incorporates both starting material properties and active channel processing into a consistent
understanding of point defect evolution. The chapter closes with an overview of some
9
other nondestructive characterization techniques by way of a comparison with microwave
photoconductivity.
2.2 Microwave Photoconductivity Techniques
Microwave photoconductivity methods which characterize semiconductor materials
consist of two basic subsystems. A pulsed light source, typically a laser, induces a
transient or constant conductivity state in the SUT by optically generating carriers. This is
then detected in a nondestructive and contactless fashion by changes in a reflected or
transmitted microwave signal. Because wafer cooling and spatial mapping are facilitated by
microwave reflection schemes, and therefore provide more versatile measurement
capability, we will focus on microwave reflectometry.
Steady-state GaAs
photoconductivity will also be covered in this section.
2.2.1 Overview of Others' Work
While an extensive list of microwave reflection systems, with and without
photoconductive pumping, has been provided by Wang [Wang], a few of the more recently
published results from groups other than those at Rensselaer will be mentioned. A system
at 2 GHz has been used at AT&T for sometimeto measure surface recombination velocity
(SRV) [Yablonovitch 1986]. In the AT&T system, the microwave signal is coupled to the
wafer by a pick-up coil. A Japanese concern has developed a similar apparatus for spatial
mapping of GaAs photoconductivity [Hasegawa]. Both of these systems suffer from
limited spatial resolution in mapping because of the low operating frequency of the
microwave subsystem. In this respect, a microwave photoconductance system at X-band
[Kunst and Beck] offers greater potential, but is not realized as the silicon wafers under
10
examination are fully inserted into the waveguide. Mapping could still be achieved by
moving the optical source relative to the wafer/waveguide, but spatial variations in the
microwave field strength must be taken into account.
Work in this area has been continuing for many years at Rensselaer with several
variants of PIMR preceding the current system. Lo used the technique to measure lifetime
in oxygen-precipitated silicon as a method of determining the depth of the defect-free, or
denuded, zone [Lo]. PIMR was shown to have sufficient sensitivity to discern among
undoped and chromium-doped SI LEC GaAs samples [Heimlich] and undoped SI LEC
GaAs processed to various point in a MMIC process [Campbell]. The technique has been
used for determining resistivity, lifetime, and mobility in GaAs as well as InP and HgCdTe
[Bothra]. Wang studied SI LEC GaAs characterization by PIMR with a choke-flange
antenna [Wang].
2.2.2 Microwave Reflectometry
The central microwave features of an MR system are shown in Figure 2.2.1. The
source signal is split equally between the test and reference sections. These signals are
altered in their respective arms and then sent back into the coupler. The coupler splits the
reflected signals sending a portion of each toward the isolator, where they are prevented
from reaching the microwave generator, and a portion to the detector, where they are
summed vectorially. The voltage output by the detector is a function of this sum.
The signal coupling back into the antenna after interacting with the SUT is actually the
sum of two separate effects at the SUT which are characterized by their respective reflection
coefficients. The first, r o , is due to reflections under equilibrium conditions from the
11
rr+LexpG<|))r*Ar (t)]
&
-
h
—psolator
rr
Detector
3dB
Coupler
<—->
^
/
/
Coupling Element/
Antenna
^
r.+Ar(t)
SUT
Figure 2.2.1
Basic Microwave Reflectometry System
12
GaAs wafer and the supporting shorting plane. Even with a conductive channel of typical
design, very little power will be dissipated in the wafer and
irol«l
(2.2.1)
Under illuminated conditions, a second and transient reflection, Ar(t), arises due to losses
in the photoconductive regions of the SUT. Taking P^(t) as the power dissipated in the
sample during the light pulse and P 0 as the power incident on the sample
P
2
d( l >
IT - A T ( t ) r = l - ^ L o
P
o
(2.2.2)
Expanding (2.2.2) and applying (2.2.1) yields
2
PH(t)
1 - 2|Ar(t)|cos6(t) + |Ar(t)| = 1 — 5 -
(2-2.3)
o
*
where 6(t) is the angle of r o Ar(t). If the power dissipated by the transient phenomena is
small compared to the equilibrium reflection, that is
| A r ( t ) | « cos9(t)
(2.2.4)
then (2.2.3) can be simplified to
p
d(0
IAIT0I
(2.2.5)
2Pocos9(t)
By Ohm's law, the power dissipated during the photoconductive event is the conductive
13
loss in the SUT
Pd(t) = jo(x,y,z,t)E2(x,y,z) dV
v
(2.2.6)
If the conductivity distribution is relatively independent of position or the conductivity
transient is spatially confined to a region much smaller than that over which the electric field
varies, then
P d (t)« a(t)
(2.2.7)
and, therefore,
fAr(t)| = Ko(t)
(2.2.8)
This expression shows that the change in the microwave signal received by the antenna is
proportional to the optically induced loss in the SUT.
At low incident powers, the detector is in a square law regime such that the detector
output voltage, V ^ ^ ^ is proportional to the magnitude of the microwave power (i.e.
voltage squared) incident on the detector. Therefore, at the detector output,
V
de,ec«or = V D
+ V t
()
and
Vdetector = K D l r r
+
^ " P C W o + Ar(t)}| 2
(2.2.9)
where Lexp(j<|>) is the loss and phase shift due to the antenna and passive microwave
14
components between the SUT and the detector and Tr is the signal at the detector from the
reference arm. Generally, the square law regime is attained by adjusting the phase and
amplitude of the reference signal to bring Vj) below the level at which the detector departs
from a square law.
If we repeat the simplifications which led to (2.2.S) and use similar assumptions then
the transient portion of the detector voltage is found to be
v(t) = 2KclIAr(t)l2
(2.2.10)
where 2 K c j is the conversion loss of the transient associated with a square law detector as
well as the losses between the sample and the detector. It should be noted that Kcj is a
function of the magnitude and relative phases of the Tr and Lroexp(j<t>) signals as
determined by the Vrj versus incident power curve for the detector diode. The sum of
these two signals determines the quescient, or Q,point for the detector which, in general,
will bias the detector in a regime other than square law. (2.2.10) then becomes
v(t) = qKcl|AT(t)|q
(2.2.11)
where ^K^ is the conversion loss and q=2 for a square law detector, as in (2.2.10), or q<2
for a sub-square law detection regime.
In microwave photoconductivity applications where the transient decay time is a
measured parameter, the value of q determines the relationship between the actual and
measured lifetime. Assuming a conductivity transient
15
c(t) <* exp(—)
x
then the detector transient will be
v(t)~expA
by (2.2.11).
Without any attenuation in the reference arm, the microwave signal seen at the detector
will be large and VcjetecttM> will be dominated by the nontransient T r Thus, about the Q
point defined by Vp>
v
(t) _ ,1
LPd(t)exp(j<|>)
1 + - p . = (1 + —2—
V
D
)-
(2.2.12)
*D
where P p is the incident microwave power producing the V Q detector output. Expanding
this in a Taylor series and keeping the first two terms (since the transient is small) gives
v(t) = <hj2pd(t)I*xp(j<|>)
or
v(t) ~ Pd(t) - a(t)
(2.2.13)
Therefore, by biasing the detector with Prj so that P^(t) is considered as a small signal, the
detector transient output voltage is proportional to transient power loss in the SUT which,
in turn, is proportional to the photoinduced conductivity. This is the situation under which
16
PIMR measurements are made.
The two key assumptions in the analysis leading to (2.2.11) or (2.2.13) that have not
been justified are the spatial distribution of the photoinduced conductivity used to arrive at
(2.2.7) and the relative magnitude of P^t) represented by (2.2.4). The former will be
discussed in the next section, while we consider the latter in chapter 3.
2.2.3 GaAs Steady State Photoconductivity
The conductivity change in a GaAs sample brought about by optical excitation can be
calculated at any time during the light pulse by
d
G(t) = q L ) n ( x , t ) dx
o
(2.3.1)
where d is the wafer thickness, \i is the electron mobility, and n is the photoinduced
electron concentration. The integration, in general, is performed over a region defined and
weighted (e.g. in the case of the system of Kunst and Beck) by the microwave field
strength, but the expression can be reduced to an integration in one spatial dimension after
assuming that the optical spot size is smaller than the microwave spot size. Holes are not
considered here since their much lower mobility makes a negligible contribution to G(t).
The time-dependent excess electron concentration, n(x,t), can be determined by
solving the partial differential equation obtained by combining the continuity and current
equations, assuming no internal or applied electric field:
17
dn
32n n
— = D — j - - + <xNphexp(-ax)
(2.3.2)
where D,x,a, and N p n are the electron diffusion constant, the electron lifetime, the optical
absorption coefficient, and the photon flux density, respectively. The boundary conditions
at the front and back surface of the wafer are due to carrier recombination by the high
density of surface defects and are represented by a front and back SRV, Sf and S^,
respectively:
D ~ = S£n at x=0
D ^ - = S b n at x=d
(2.3.3)
For GaAs IC starting material, we take
Sf=Sb=S
(2.3.4)
Since there is no analytic solution to (2.3.2, a numerical technique is utilized. In the
remainder of this section, we will consider steady state photoconductive, thereby yielding a
more tractable problem and a basis for understanding the effect of defect concentrations on
GaAs photoconductance. Transient photoconductivity will be discussed in detail in section
3.4.1 where a numerical solution is presented.
(2.3.2) under steady state conditions yields an analytical solution for n(x). The sheet
photogenerated carrier concentration, N tota j, is found by integrating n(x) over the wafer
thickness [Wang]
18
N„J»(x)dx=-^ (,.,-" ^ y < l ^ )
„
l-(oL)
1+s
(23.5)
/xcoth(i-)
where L, the electron diffusion length, is (Dt) 1 ^. Under the condition d » L , which is
true in general for device quality GaAs channels and starting material, (2.3.5) reduces to
N ^ - J S i L (1 - . - - ^
^
HaL)
)
1+s
(2.3.6)
Ji
Three special cases of interest will be examined which lead to simplifications to
(2.3.6). The first two involve the extremes of S, zero and infinite SRV In the case of
large S, (2.3.6) reduces to
N
totai
^(l-e^-oLd+e'0"1))
l-(aL)z
for
S » . / V
(2.3.7)
x
while the latter gives
i-r«Lr
v x
Negligible surface recombination can be achieved by chemical or photo-chemical treatments
[Yablonovitch 1987], or by a thin, doped n-layer as would be found in the material system
19
composed of an ion implanted channel on an SI GaAs wafer. Infinite, or very large, SRV
is typically seen with GaAs material before any any processing.
The third case which simplifies (2.3.6) corresponds to small absorption coefficient,
ad « 1
(2.3.9)
Recognizing that for device quality GaAs wafers L « d and combining with (2.3.9),
<xL « 1
(2.3.10)
Simplifying (2.3.6) based on these two expressions, we arrive at
N
toui = N p h ^ d
C 2 - 3 - 11 )
As might be expected in this instance, carriers are generated independent of depth and,
therefore, the effect of the illumination is uniform throughout the thickness of the wafer.
Small absorption coefficients for IR excitation are typically seen for wavelengths above 900
nm where the photon energy is less than that of the 1.42 eV GaAs bandgap.
In order to use steady state photoconductivity as a measure of defect concentration in
GaAs, a relationship between the dominant defects and at least one of the parameters
accessible through a photoconductivity measurement must be established. In SI GaAs, the
dominant bulk defect is the deep donor EL2, which is found approximately near the middle
of the band. The next most prevalent defect is the shallow carbon acceptor and, in effect,
EL2 compensates C to form SI material. (Compensation involves the negation of a dopant
so that net free carrier concentration is very nearly intrinsic.) This is referred to as the two
level model (TLM) for GaAs (shown in Figure 2.2.2). Details on point defects,
20
ABG
BBG
TT TT
EL2
OOP
Carbon
E„
Figure 2.2.2
TWo-Level Model of Undoped SI GaAs
21
compensation, and carrier lifetime in GaAs will be discussed in the following section; here
we mention EL2 and C as a matter of reference for developing two basic photoconductance
models for undoped SI GaAs.
The first model is for the interaction of above bandgap (ABG) laser light with starting
material. In the current PIMR system, the ABG optical illumination is at 850 nm (and the
term "ABG" henceforth will be in reference to this specific source). ABG light generates
hole-electron pairs band-to-band which recombine through EL2 [Borrego]. Since the
absorption coefficient for the ABG laser is about 10^ cm"*, the majority of the carriers are
generated within a few microns of the surface. Thus the ABG photoconductivity can be
treated as a sheet conductance on the wafer surface and condition (2.2.7) is satisfied.
If we assume a low level injection condition of
=JNNAA(X
N.„,.,
(x) dx
total «
(2.3.12)
where N^(x) is the C concentration as function of depth, than the lifetime is unchanged
from its equilibrium value during the laser excitation and is given by
1
ABG
(2.3.13)
c NA
n A
where
(2-3-14>
C„ = V t h
Cn is the electron capture rate for the EL2 level, o n is its electron capture cross-section and
v^ is the electron thermal velocity. Substituting (2.3.7) into a time-independent (2.3.1)
22
and recognizing that ad » 1 (most of the generated carriers are found near the illuminated
surface) gives
for the ABG photoconductance. More succinctly, for our purposes:
Therefore, for ABG excitation in a low level injection regime, the TLM implies that GJ^Q
is inversely proportional to the root of N^ and can be used to measure the C concentration
in SI GaAs. Experimental results have shown that there is good agreement between carbon
concentration as measured by the local vibrational mode (LVM) technique and the results
obtained by the microwave ABG photoconductivity measurement using the TLM [Wang
and Borrego].
The second model of GaAs photoconductance is for below bandgap (BBG). We
similarly define "BBG" as that for the 904 nm laser source used with PIMR. In the TLM,
BBG promotes electrons from EL2 to the conduction band with the recombination being
carried out via the reverse process. Since the absorption process is through EL2, a is a
function of the concentration and occupancy of the deep donor:
a = o;(Nd"NA>
<2-3-16>
by the TLM, where N<j is the concentration of the deep donor and a n ° is the electron
optical ionization cross-section for EL2. In the absence of an electric field, the absorption
23
coefficient is approximately 1 cm"* and the condition for using (2.3.11) holds (carriers are
generated fairly independent of depth in a SOO \im thick SI GaAs wafer). In contrast to the
case for the ABG excitation, the BBG photoconductivity does not vary much over the
wafer thickness and can be assumed constant in compliance with condition (2.2.7).
Again using the time-independent form of (2.3.1) and substituting with (2.3.13) and
(2.3.16) gives the BBG photoconductance as
qjiN.CT°(N.- NJd
A
G = qnNphaTd = q ^ N d
n A
Reducing this to a relationship between steady state photoconductance and defect
concentration gives
°BB(i—V^
a3 17)
'
^A
Based on the TLM, GggQ is proportional to the ratio of neutral-to-ionized EL2 sites but
typically the EL2 concentration exceeds that of the C by an order of magnitude and we
therefore expect the BBG measurement to be proportional to the total EL2 concentration.
By measuring the photoconductive transient decay using, Wang found that the TLM
extracted EL2 concentration in disagreement with that measured by IRT. However, the
BBG photoconductivity measurement did yield acceptable transient results as well as
predicting the magnitude of the photoconductive response when an additional level was
taken into consideration: the second ionization, or double donor, level of EL2 positioned at
24
about 0.54 eV above the valence band (Figure 2.2.3). This three level model does not alter
the result given in (2.3.15) for Gy^gQ since in SI GaAs the only sites available for
recombination (under low level injection conditions) will be the singly ionized EL2 level
whose concentration will change one-for-one if the C concentration varies.
Two aspects of the photoconductivity model must be emphasized in light of the focus
in the following chapters on processed material. First, it is assumed that the dominant
recombination process is through the unoccupied EL2 sites. Second, it is not necessary
for the shallow acceptor to be carbon; just that the sum of all defects shallower than EL2,
the net shallow acceptors, are compensated by the EL2.
(2.3.17) is derived under the assumption of uniform absorption throughout the
thickness of the illuminated wafer, but this may not be the case if significant electric fields
are present within or impressed upon the GaAs sample. The absorption coefficient can
increase by several orders of magnitude for wavelengths below, but near, the bandgap
energy. This is referred to as the Franz-Keldysh effect and is essentially a tunneling
process: the photonic energy places the valence electron close enough to the band edge that
there is a substantial quantum mechanical probability of completing the band-to-band
transition. This augmented absorption coefficient is therefore a function of photon energy,
as well as electric field, as depicted in Figure 2.2.4.
In 10 *' cm~3 n-doped GaAs, an electric field is established at the surface due to the
lack of passivation. With the surface "pinned" near the midgap, a depletion region is
formed, roughly 0.1 \im deep, with a peak electric field of about 10-* V/cm. Based on
Figure 2.2.4, the absorption coefficient increases by almost four orders of magnitude and
25
ABG
B8G
TT T t T T
EL2
t
EL2
Carbon
Figure 2.2.3
Three-Level Model of Undoped SI GaAs
[Wang]
26
Wavelength (jim)
s
10
0.84
-T—
0.88
0.92
104
_
1.00
0.96
1.04
S„ -1.42 eV
103
E
u
E - 2 X 10s V/cm
102
1 X 10s
4
. 4 X 10
4
, 1 X 10
101
10°
-0.04
0.04
0.08
0.12
0.16
0.20
S„ - hw (eV)
Figure 7.8 Electric field and photon energy dependence of the band-to-band absorption for GaAs. [After G. E. Stillman and C. M. Wolfe, "Avalanche Photodiodes" in Semiconductors and Semimetals, Vol. 12, Infrared Detectors II, ed.
R. K. Willardson and A. C. Beer (New York: Academic Press, 1977), p. 291.]
Figure 2.2.4
Franz-Keldysh Effect in GaAs
27
the BBG photoconductance would not strictly conform to the requirements for using
(2.3.11). In this case, G Q S Q would be a superposition of (2.3.11), with a reduced
optical flux, and one or several expressions not too unlike (2.3.8).
2.3 IJndoned SI LEC GaAs
LEC ingots, or boules, start with a seed crystal dipped into a nearly stoichiometric,
As-rich melt of GaAs held in a pyrolitic boron nitride crucible. An inert liquid layer of
boric oxide (B2O3) floats on the melt surface and reduces As loss. Rotation of the seed
and/or the crystal reduces thermal gradients within the growth system while pull rate
determines the boule diameter.
Once crystallized, the GaAs ingot continues to cool, first to the ambient temperature
within the puller, or growth chamber, and then finally to room temperature. Control of the
thermal gradients during each of these phases is important in determining the final
dislocation density, particularly with the low thermal conductivity and critical resolved
sheer stress of GaAs. The temperature, as a function of position within the crystal during
any point of the growth process, is affected by factors such as latent heat of crystallization,
convection, and heat transfer from the melt to the crystal to the inert gas ambient within the
chamber. The generated dislocations themselves are electrically inactive, but they may act
as gettering sites for diffusing extrinsic dopants or excess As incorporated from the nonstoichiometric melt. Regardless of the degree to which the puller is thermally controlled,
contaminants and intrinsic defects will induce some minimum concentration of dislocations,
typically on the order of 10-* cm"^ [Jordan]. Impurity hardening with indium can yield
dislocation densities below this figure even without extraordinary attention paid to
28
temperature control, but indium doping sometimes reduces the mobility by about a factor of
two [Jordan].
Once growth is completed, the boule is typically annealed before wafer slicing. Boule
annealing is done below 950 °C for about 10 hours and radially homogenizes the ingot.
This heat treatment tends to create a dislocation cell structure with well defined dislocationfree regions surrounded by somewhat dense walls of line defects [Govorkov]. Point
defects are gettered within these cell walls. Most importantly for annealing or high
temperature processing done during processing, the boule annealing temperature
establishes a thermodynamic equilibrium temperature for the point defect constituency of
wafers cut from the boule.
Wafer slices are taken from the central core of the boule after minor and major flats are
machined to indicate the crystallographic orientation. Chemical and/or mechanical methods
are used to polish a single face or both sides of the slice. Subsurface damage can be
introduced by the polishing, but this is generally removed by etching the first few hundred
angstroms.
Even after removing the polishing damage, the wafer surfaces might still be
considered the most significant defect associated with GaAs devices. There is a very high
density of states at the surface made up of dangling bonds, Ga and As oxides, and
contaminants. These states effectively immobilize, or "pin", the position of the Fermi
energy near the middle of the bandgap. Passivation of GaAs is practically impossible,
thereby making SRV an almost unmitigated problem. (An additional consequence, the
surface depletion layer, has already been mentioned with regard to the presence of a
channel.) While this midgap pinning of the Fermi level is an important factor in the surface
29
and device properties, it does not directly influence events in the bulk.
The largest bulk contaminant by concentration is B; either from the B2O3 (borax)
encapsulant or the PBN crucible. B is usually found on a Ga lattice site and, in this
position, is isovalent in the SI LEC GaAs used for ion implanted IC technology.
Furthermore, B is relatively benign in the GaAs lattice itself and little performance
degradation is seen for concentrations in the 10*" through 10*' cm"* range. Borax water
content has been shown to influence the extent to which C is incorporated into the crystal
from the melt [Hunter].
A shallow acceptor at 26 meV, C has become the dominant electrically active
contaminant in LEC growth technology today. Over the last decade, the C concentration in
undoped SI LEC GaAs ingots has decreased from 3x10^ cm"-* to 4x10^ cm"3 and
below.
Previous to this, the prevalent extrinsic dopant in as-grown SI material was Si, a
shallow donor. Some Si is still present in GaAs melts, on the order of 10* •* cm"-* and
some degree of compensation results since Si is most often found on a Ga lattice site,
where it is a donor. The amphoteric nature of Si (Si^s, Si on an As site, is an acceptor) in
GaAs is typically not a factor for as-grown material, but does have an impact at
concentrations found in ion-implanted channels.
Other contaminants can be found in as-grown material in concentrations much below
that of C. Cu and Be are rarely a problem with LEC GaAs, but are mentioned here as they
are often cited in the literature for specific and separate reasons. Cu can corrupt a GaAs
wafer after contact with metal surfaces in handling. Cu introduces several acceptor levels
30
and can, therefore, affect compensation near the surface. Be is used to intentionally create
p-type regions during processing. The Be dopant sits on a Ga site and produces an
acceptor level at 28 meV above the valence band edge.
The SI state in undoped GaAs is a fortuitous consequence of the intrinsic, or native,
defect referred to as EL2. Historically, EL2 was thought to be associated with oxygen
incorporated into the boule from the melt during LEC growth. Later it was found that EL2
is indeed an intrinsic defect, initially identified as the ASQ & antisite. More detailed
investigations have shown that the annealing properties of EL2 are not easily explained by a
simple point defect [Look]. Noting that As-rich melts favored EL2 production, models
combining the antisite with single or paired Ga vacancies (VQ & ) [Lagowski and Gatos,
and van Vechten] or Asj [von Bardeleben] were proposed. The behavior of EL2 upon
quenching from high temperatures suggest an even more complex constituency, probably
involving V^s as well as VQ & [Wager].
As this study is concerned with the characterization of processed GaAs, we will rely
more heavily on the EL2 annealing experiments and thermodynamic considerations for
choosing a model of the deep donor rather than structural measurements, like positron
annihilation [Bourgoin], or calculations [Baraffj. As the ingot cools, there is a high
concentration of As and Ga vacancies above 950 to 1000 °C [Lagowski] most of which, at
slightly lower temperatures, will coalesce and react with the excess As to form AsQ a and
bidefect complexes [Look]. Quenching from anneal temperatures in the 500 to 750 °C
range have shown that EL2 is formed between 650 and 700 °C [Chichibu, and Suezawa].
Furthermore, long duration anneals in a vacuum, thereby generating V^ s near the surface,
have shown a change in the EL2 concentration corresponding to V^ s diffusion. This
31
combination of results have led many to suggest AsQ a \^ s VQ a as a model for EL2 [Zou,
and Chichibu]. Other data and process models are compatible with this model [Ohkubo,
and Morrow].
Native point defects can be present in forms other than EL2. The concentrations of the
elemental point defects (ASJ, Gaj, \ ^ s , V Q E , AsQa, and Ga/^) are determined mainly by
crystal stoichiometry and the thermodynamics of the growth and boule anneal processes.
The important defects for an As-rich melt are Asj, VQa, and ASQ E , and V^s, with the last
added due to the propensity of As to out-diffuse. Annealing history is an important factor
for processed GaAs and may significantly alter bulk compensation, implant activation, and,
ultimately, MESFET performance. Annealing will be considered in some detail in chapter
5. For future reference, Table 2.3.1 lists the elemental and bidefects which make up EL2,
and their dopant states, donor or acceptor [Look]. Ga^ s is included since it can be created
by the annihilation of V^s by As^ s .
An extensive list of native defects present in undoped SI GaAs could be compiled by
taking the elemental point defects and enumerating all of the possible combinations.
Naturally, some of these are not thermodynamically favorable in device quality material.
The only defects usually discussed are those which have been measured in the course of
material or device characterization—be they native, extrinsic, or a combination of the two.
An exhaustive list of defects associated with electrically active defects, or traps, was first
introduced by Martin [Martin], of which EL2 is one. Some of these defects have been
associated with changes in EL2 concentration in the course of processing [Fang, and Min]
or intentional crystalline damage [Bourgoin]. There remains some question as to the exact
relationship between the elemental defects and their electronic counterparts as reported by
32
Defect
Dopant State
VAs
donor
vGa
acceptor
As
Ga
donor
Ga
As
acceptor
donor
VGaVAs
v
As As Ga
donor
v
Ga G a As
acceptor
Table 2.3.1
Dopant States of EL2 Elemental and Bidefect Constituents
33
Martin, but Look has made tentative identifications based on the work of many others.
However, all of the defects detected in typical device quality starting material with the
exception of EL2 and C are found in sufficiently low concentrations that their presence can
be ignored in discussing compensation.
As stated previously, compensation occurs where one or more defects act in concert to
bring the carrier concentration nearly equal to the intrinsic carrier concentration, nj.
Because the intrinsic carrier concentration in GaAs is on the order of 10" cm"^, the intrinsic
resistivity is very high («1(P ohm-cm at room temperature) and the material is referred to
as SI rather than semiconducting. This condition corresponds to the Fermi level being at or
near the midgap (the intrinsic Fermi level, Ej); the same position where the EL2 level is
found (Figure 2.2.2). The simplest case for compensation would be if the total number of
acceptors equals the total number of donors. Thus in GaAs, we could add Si to negate the
effects of the dominant contaminant, C. As these quantities are typically on the order of
Kr^ cm"-*, even a small fraction of a percentage difference in the two could produce lightly
doped material.
A second and more practical approach is to add a deep donor with an energy close to
Ej and a concentration somewhat larger than that of the C acceptor (the larger concentration
is needed since the percentage of the deep level in the ionized state will drop as Ef
approaches Ej). This has the effect of balancing the charge of the acceptors without
allowing Ef to get very far from Ej. If an order of magnitude or more of the deep level is
added, Ef is pinned near the deep level's energy. In the past when Si was the dominant
shallow contaminant, the deep acceptor Cr was used for compensation. For a generalized
34
GaAs system of shallow donors, shallow acceptors, and EL2, this model has been
quantified [Johnson].
2.4 Active Channel Formation Processing
MMIC device technology today requires the introduction of many different dopant
species and profiles. Si doping on the order of 10^ cm"* and 0.25 |im deep is necessary
for a general purpose GaAs MESFET. Low current or enhancement mode devices have
shallower and more highly doped channels, while power devices need deeper and more
lightly doped layers. Improved device performance has been found with deep Be implants
below, or B or H adjacent to, the active device. Ion implant technology permits the use of
any of these, as well as all of these, on a single wafer, although some desired profiles
require several combinations of doses and energy.
The drawback of implantation is that significant lattice damage is incurred in bringing
the ions to rest. A post-implant anneal rectifies the situation by providing the
thermodynamic conditions for correction of the damage. However, annealing implanted
GaAs does not simply involve crystalline repair and donor activation, but also requires a
consideration of the GaAs wafer and any synergistic effects between the wafer, the
implant, and the anneal. Two techniques are discussed here and are in general use: furnace
annealing (FA) or rapid thermal annealing (RTA).
2.4.1 Ion Implantation in GaAs
Besides implant species, an implantation schedule is specified by the energy and
implanted concentration per square centimeter, or dose. While the energy for a given
35
implant species allows for an estimate of the average implant depth and dose gives an
average peak concentration, GaAs ion implantation is not as easily modeled as silicon.
Silicon ion implantation is predicted well by the LSS theory [Anholt and Sigmon
1989]. Depending upon the energy and the mass of the implanted species, the lattice will
absorb the energy of the ion by physical collisions (nuclear stopping) or by coulombic
interaction (electronic stopping). The resulting implant profile and silicon lattice damage
are adequately modeled by applying these two stopping mechanisms. For most implants
into silicon, the profile appears Gaussian.
Unfortunately, LSS theory does not work well with GaAs for several reasons, the
most significant of which we elaborate on here. Along several preferential directions in the
GaAs crystal lattice, there exists very large gaps between the intersecting atomic planes.
These conduits, or channels, are large enough for an implanted ion to travel relatively
unimpeded, and therefore, much further, than would otherwise be expected. The smaller
and lighter ions typically used with GaAs plus the high doses required for GaAs
amorphization compound this "channeling" effect. During implantation, the GaAs wafer is
oriented with respect to the ion beam so that the ions are less susceptible to channeling, but
because of the stochastic nature of ion/lattice collisions, wafer tilt and rotation adjustments
can only reduce the channeling. Thus, GaAs profiles are less Gaussian than their silicon
counterparts and can be characterized as having "tails".
Despite this problem, two systems are used to predict GaAs implantation results. The
first and most widely used is the GATES software package based on the work of Anholt
and Sigmon [Anholt 1989]. GATES is essentially a collection of look up tables based upon
empirical data and the Pearson-IV distribution. These data were derived by implanting the
36
common GaAs dopants over a wide range of doses and energies and measuring the profiles
by secondary ion mass spectroscopy (SIMS), as well as other methods. GATES will
predict the implant and activated carrier profile based upon the implant and wafer data
supplied by the user. The second is a more analytic approach based on curve fitting the
measured profiles. To a fairly accurate degree, most Si implants in GaAs can be fit with a
hyperbolic Gaussian profile [Tabatabaie-Alavi].
Damage from channel implants, while not extensive enough to form amorphous
GaAs, is nonetheless significant in its lasting effect on the material and any subsequently
fabricated devices. The first hundred angstroms or so of the wafer surface is sputtered as
secondary ions, leaving behind large concentrations of vacancies. Towards the bulk side of
the channel, Ga j and Asj can accumulate under the force of the incoming dopant. Other
defects, such as Frenkel pairs, are produced as well. Total induced defect density maybe
on the order of dopant concentration in some locations [Christel]. On their own or after
reacting with other defects, the defects originating with the implant damage can alter the
dopant activation by blocking a point defect reaction or reduce RF performance by
increasing the density of traps.
2.4.2 Furnace Annealing
Furnace Annealing (FA) has been used the most over the years as a GaAs annealing
method. A typical FA cycle lasts 20 to 40 minutes between 800 and 900 °C in a specified
gas environment. The arsenic can be from a variety of sources and over a range of
pressures. In heating to and cooling from the anneal state itself, the temperature within the
oven is varied slowly so as to minimize thermodynamic gradients.
37
The duration of the high temperature step of the FA must be a balance between several
factors. If this step does not last long enough, residual implant damage will degrade the
mobility and leave too many traps to yield good working devices. Long anneals have a
similar effect, but this is mainly due to As out-diffusion near the surface and changes in the
equilibrium point defect constituency in the bulk. The latter is a result of the wafer
attempting to establish thermodynamic equilibrium at the FA temperature rather than remain
close to the equilibrium state associated with the post-growth boule anneal. Under these
circumstances, the EL2 concentration typically decreases at the expense of shallower traps,
such as EL6 [Min, and Fang]. Long anneals may also alter the implant profile, especially
with p-type implants at higher doses.
In some respects, time and temperature can be traded off one another, as the damage
repair can be viewed as a diffusive process. There is, however, an activation energy
associated with this process which appears to be related to the Si moving from interstitial
sites to Ga sites in a model developed by Bindal [Bindal 1989a]. The activation energy sets
a lower limit of about 750 °C for FA temperature; an upper bound is established at about
850 °C due to excessive As out-diffusion. Above 850 °C, the equilibrium partial pressure
of arsenic increases rapidly above the values typically used for an arsine ambient [Arthur].
The presence of gaseous arsenic in the oven tends to reduce \ ^ s generation by As
loss. Partial pressures may be anywhere from a few milliTorr to tens of atmospheres
[Chichibu]. Solid sources maintained within the oven at a temperature different from that
for the wafers is sometimes used. The more popular method is an arsine (ASH3) ambient
diluted with hydrogen or forming gas. Some processes accept the inevitable As-loss and
remove As sources altogether from FA in deference to safety.
38
Ramping to and from the anneal temperature is done over 10 to 20 minutes. Just as
with the high temperature portion of the step, care is taken in uniformly heating and cooling
the wafers to minimize thermodynamic gradients which would cause variations within a
wafer and between wafers.
The latter concern is common in FA. Wafers at one end of the furnace may not be at
'the same temperature as those at the opposite end. As out-diffusion from wafers upstream
will create a higher As overpressure at the outlet end of the oven. In this respect, the larger
capacity of FA equipment can become a drawback.
2.4.3 Rapid Thermal Annealing
The concept behind rapid thermal annealing (RTA) is in direct contrast to that of FA.
The high temperature portion of the RTA step is roughly 100 °C higher (than that for FA)
but lasts for only 10 sec in an inert ambient. Heating and quenching are also done in a
matter of seconds. The result is crystalline repair affected by higher temperatures and
temperature gradients rather than extended annealing duration.
The heating within an RTA oven is accomplished by radiative heat transfer from a bank
of lamps. Problems with wafer uniformity can be traced to the lamp intensity spatial
distribution. For a bank of lamps of fixed length and width, there will be an intensity
nonuniformity across the wafer. This can be minimized by setting the individual lamp
intensities to form a more constant energy distribution at the wafer, but there will still be
end effects which can be severe when the oven is not sufficiently large.
This disadvantage is more than outweighed by the advantages of RTA. The short
anneal times reduce dopant diffusion problems, especially severe with p-type dopants. For
39
some anneal processes, the relative speed of RTA negates much of FA's load advantage.
As-loss is mitigated due to the limited time at high temperatures. Out-diffusion of As is still
a problem, though, and many options are available for reducing its impact
Deposition of dielectric layers, or capping, is a popular method of preventing As-loss.
Si02 (silox) or Si3N4 (nitride) can be deposited on the GaAs substrate at sufficiently low
temperatures, thereby minimizing As-loss during the deposition while maintaining film
integrity at the anneal temperature. Particular care must be taken in designing film
deposition processes to alleviate film stress. Slip from film stress can be reduced by
several methods incorporated in the RTA system itself, such as guard rings, but these are
generally not compatible with the demands of a production line.
Silox films present an additional problem. In silicon technology, Ga preferentially
diffuses into silox rather than remaining in the wafer, making Ga unacceptable as a dopant
in most applications. This behavior is duplicated in GaAs where significant and detrimental
Ga out-diffusion can be incurred during the RTA process.
Proximity capping methods rely on stacking a silicon or GaAs wafer on top of the
implanted side of the annealing wafer. This presents a physical barrier to As loss in the
case of the silicon cap. If a GaAs wafer is used as a cap, a portion of the As partial
pressure is generated by the cap which lessens the effect on the wafer of interest.
Proximity capping is typically used when lower activation efficiency is tolerable
Noncapping techniques can be divided into two categories. The As-loss can be
accepted in some applications and no extraneous effort is made to keep As in the sample.
Some post annealing treatment, like wet chemical etching or plasma treatments, may be
necessary in these cases to remove the As depleted region. The second category of
40
noncapping techniques maintains an As overpressure about the annealing wafer. This can
be accomplished in the same manner as with FA by using a solid or gaseous As source, but
another method is more desirable as As sources are not directly involved.
First, a silicon carbide-coated graphite susceptor is used to encapsulate the annealing
wafer thereby forming a much smaller annealing chamber. Before any implant annealing is
done, the internal walls of the susceptor are charged with As by annealing a sacrificial
GaAs wafer for an extended period of time to promote As out-diffusion. Implanted wafers
can than be annealed without any overt means of limiting As loss. Depending on the rate at
which gaseous As leaks from the susceptor itself, more than 50 wafers can be annealed
before recharging without much variation in the peak carrier concentration [Kazior],
implying that equilibrium As overpressure is achieved from the As-charge on the susceptor
walls.
Quench rate is of some concern in designing an RTA process because the results of
high temperature defect dynamics, such as EL2 dissociation above 700 °C, will not have
the opportunity to reverse themselves due to the nonequilibrium thermodynamic nature of
RTA. Defect concentrations have been found to be different between FA and RTA
treatments which yield approximately the same sheet resistance for implanted layers [Fang],
but the bulk region of RTA wafers is still good SI material and the channels are of device
quality.
2.5
Stnichiometricallv.Cnnstrained Defect Modeling
Numerous studies have shown the importance of substrate point defect population
and stoichiometry in determining activation efficiency [Anholt 1988, Sato, Kazior,
41
Morrow, Von Neida, Saito]. Coimplanted species must be located on lattice positions and
in so doing can alter the activation of the primary implant by direct intervention or through
intermediate reactions with as-grown defects [McNally, Morrow 1988b]. The anneal
further complicates matters by providing thermal energy for point defect reactions and
diffusion processes not otherwise possible. A robust model of active channel formation
must be capable of considering all of these effects individually as well as in concert.
A model based on the concept of stoichiometric constraint has been developed
which shows considerable success in modeling the point defect structure for a variety of
GaAs processes, such as post-growth cooling and Si implant activation [Morrow].
Stoichiometric constraint is the assumption that the degree to which a wafer deviates from
perfect, crystalline GaAs is determined by the composition of the melt from which the
boule was pulled. Any subsequent high temperature thermal processing typically used in
MMIC fabrication does not change the deviation from stoichiometry, s, unless mass
transport involving the GaAs itself, like As out-diffusion, is encountered.
Arbitrarily choosing an arsenic-rich condition as positive stoichiometric deviation,
we define
s
(NAS - NGa>
^
^
(2.5.1)
where N^ s , N(j a , and N are the concentration of arsenic, gallium, and GaAs molecules,
respectively. The difference in the numerator can be rewritten to reflect the fact that each
defect contributes to s and the sum of all of the defects present must account for s. That is,
2sN = X 3 D N D
D
42
<2-5'2>
where d p is the stoichiometric signature of defect D and Np is its concentration. The
stoichiometric signature is the deviation from local stoichiometry for an integral number of
unit cells in the immediate neighborhood of D; thus, ASQ E has a stoichiometric signature of
2, V^s is -1, and VAsAs@aVQa, the model for EL2 assumed in this manuscript, is also 2.
B and EL2 are the most abundant defects in undoped, SI GaAs so
2sN = adNd + dBGNBG
(2.5.3)
where defect d is EL2 and defect BG is BQa. dgQ is 1 while dg will, in general, depend
on the EL2 model. C should also be included in (2.5.3) on the basis of maintaining charge
neutrality and achieving compensation [Johnson], however its contribution to (2.5.3) is
negligible. The development of the model continues along the lines of most other point
defect models where a series of reaction equations are chosen which appear to account for
observed phenomena followed by a solution to these equations given some constraints,
such as charge balance, mass conservation, or stoichiometric conservation.
Implants are considered as the stoichiometrically neutral introduction of a defect
species into the crystal structure which must then be incorporated onto lattice sites without
directly altering s in the process. Thus, all implants are modeled as interstitials prior to the
activation reactions. Implant damage could also be accounted for without affecting s,
unless sputtering at the surface is a concern. For the simplest case,
2sN = dE N E +d BG N BG +3 SI N SI +a SA N SA +3 SG N SG
<2-5-4)
where SI is the implanted Si interstitial, SA is the implanted Si on an As site, and SG is the
implanted Si on a Ga site. (2.5.4) becomes more complicated when considering defect
43
complexes of B and Si.
Based on the results of many independent investigations with a wide range of
implants and anneals, Morrow proposed the following reaction equations to be in
equilibrium at the annealing temperature and constrained by charge balance and
stoichiometry (with 3JJQ = 1):
Si
Si
B
As~ S i Ga +
Ga + 2 V L
+ B
2V
As +
4e
"
Ga + 4 e"~ B GaSiAs
GaSiAs + «0d> E L 2 ~ BGa + SiGa +
2e
"
with the corresponding mass action constraints
n N
f SA = K VA N SG N VA n l
K
2
4
SBVNSGNVANBGnH =
N
7
SB n i
3
^d
2
i SB( N c/ n i>
= K BE N BG N SG n H
n N
where nj is the intrinsic carrier concentration at the annealing temperature, njj is the
electron concentration at the annealing temperature, and N ^ \ and N<jg are the
concentrations of V^s and BQa^^As' res P ect i v ely. The n j have been inserted to keep the
reaction constants, K^\, KsBV m^ ^BE> dimensionless.
The annealing temperature will determine the values for the reaction constants and
annealing conditions, such as variable arsine overpressure, may require the addition of
other reaction equations. The reaction constants are not known due to the lack of available
processing data and the model at this level of complexity can only be used at this time to fit
44
the reaction constants for a given set of wafers and implant data. Morrow did find,
however, that in many instances, two simplifications to the model still provided good fits
for low to moderate dose implants annealed in a furnace. The first is to use a simplified set
of reaction equations:
Si, + (l/3d)EL2 --> Si^ + e"
(2.5.5)
2SiI + B G a - > S i ^ a + BGaSiAs
Si
l" > S i Ga +
V
As
(2.5.6)
<2-5-7>
Associated with each of these is a mass action constraint. However, we invoke a second
simplification requiring that these reactions proceed to completion in the listed order, and,
therefore, the reaction constants for each mass action equation is zero.
Under these assumptions, (2.5.5) through (2.5.7) can be rewritten to give the
activated electron concentration
or
n = NSi,
NSi<aENE
n = aENE,
3 E N E <; N S i <£ 2N B + ^ N g
n = NSi,
2N B + 3 E N E < N S i
(2>5#8)
For high doses, an additional equation can be added to account for the activation efficiency
reduction seen in such implants by a reaction which produces Si^ s one-for-one with SiQa
so as to maintain s.
Computed results for this model are shown in Figures 2.5.1. Varying the EL2
45
•
*
•
i
i
i i
I
E
w
I 10
es
2xl0 1 6 cm" 3 EL2
k
s
e
o
U
lxl0iOcm
u 1
8
w
i
i
i
i i
•
i
1
10
2
Si Concentration, cm"
a)
"s
•2 10
2
lxl016cm"JB
e
c
o
U
2xl0 1 6 cm" 3 B
2 l
8
5xl0 1 6 cm" 3 B
i
i
i i
i
i
i
i
i i i
1
10
2
Si Concentration, cm"
b)
Figure 2.5.1
Morrow Activation Model Computed Results - concentrations in 1 0 ^ cm"3
a) Variable As-Grown EL2 Concentration (fixed B at 1E16 cm'3)
b) Variable As-Grown B Concentration (fixed EL2 at 1E16 cm'3)
46
concentration between the typically observed 1 and 2 x lO1^ cm~3 produces shifts in the
activation which are localized to silicon concentrations near that of the EL2. Larger EL2
concentration-achieve greater activation efficiency only in the portion of the implant profile
where the B is not totally consumed by (2.5.6). In contrast, varying the B concentration
produces shifts in the activation curve which are not totally reversed by larger Si
concentrations. This is mainly due to the stoichiometric constraint which tends to greatly
limit Si donor production (i.e. SiQa) with so much B already on Ga sites.
Sheet resistance trends among wafers from different boules can now be viewed in
light of EL2 and B concentration variations. Wafers with more EL2 will have more heavily
doped channel tails as reaction (2.S.5) staves off the compensating B reaction (2.5.6).
Large sheet resistance variations from differences in EL2 concentration are not typically
seen with MESFET channels as the deep donor concentration is controlled, generally, in the
1 to 2 x 10 *" cm"^ range. EL2 concentration is only a factor for shallow, or low dose
implants, and may be involved in substrate current [Ladbrooke]. Conversely, higher B
concentrations affect the tail and, perhaps, the channel itself (if the B concentration
approaches the Si concentration in the heart of the channel) by initiating and maintaining
compensation up to fairly high electron concentrations. Using a weighting function,
w(N(x)), where N(x) is the implanted Si profile, based on (2.5.8), the sheet resistance
predicted by this model can be calculated as
1
?s=qL(x)N(x)w[N(x)) dx
a
47
where b is the depth of the channel and a is the depth of the surface depletion layer.
w[N(x)] represents activation efficiency as predicted by Morrow's model and, as such, can
be used to fit B and EL2 concentrations.
This simplified model suffices for the effects of the implant and substrate without
an anneal. If the activation anneal produces conditions in which the second assumption of
the model does not hold (e.g. the activation reactions do not go to completion) than the
reaction constants must be determined and solutions to the original equations must be
obtained rather than using the simplified activation relationship (2.S.8). This, then, is the
drawback to the simplified stoichiometrically-constrained model. What we require to study
annealing effects is a second model which explicitly includes temperature yet retains at least
some of the point defect features of Morrow's model.
Such a model of Si implant activation with variable annealing temperature has been
developed by Bindal [Bindal 1989]. Bindal models activation as the establishment of V^ s
and VQ a concentrations in equilibrium at the annealing temperature. In the presence of
implanted Si, again assumed to be found in an interstitial position immediately after
implantation, the GaAs vacancies are consumed by the activation reactions
Sil + V G a ~ > SiGa
and
Si
I + v As">
Si
As
The amphoteric nature of Si in GaAs and the migration of Si donors to Si acceptors are also
considered in the model as second order effects which ultimately reduce activation
48
efficiency for high implant doses. Stoichiometry of the wafer itself is not considered
outright; however, we have seen that for the doses typically encountered in MESFET
channels, the EL2 concentration need not be considered and B concentrations in the low
10*" cm"' range are of little consequence.
Bindal's model defines three temperature regimes for FA. Fairly stable activation in
the 80% range for typical MESFET channels is found from the 800 *C to 850 °C. At lower
annealing temperatures, the generation of insufficient V(j a limits the activation efficiency.
Temperatures above the stable processing region produce more \ ^ s and the activation
process becomes more self-compensating.
Bindal's basic ideas hold for RTA, but two modifications must be made. First, the
limited annealing time shifts the stable processing region to higher temperatures in the 900
to 950 °C range. Second, choice of RTA capping method-uncapped, silox- or nitridecapped, or susceptor—will alter the balance between V(-ja and V^ s at the annealing
temperature. Neither is implicitly accounted for in Bindal's work.
Bindal's model is, however, quite useful for practical processing problems as it
relates process variables, annealing temperature and annealing arsine overpressure, to point
defect reactions which ultimately yield activated Si. Morrow's model, on the other hand,
has to date been developed to consider processes occurring prior to the anneal. But
knowledge of the reaction constants and the addition of some surface conditions would
extend the application of Morrow's model to the entire range of implant and annealing
issues. In this sense, Bindal's model can be viewed as a subset of the stoichiometricallyconstrained model.
49
2.6 Non-Contacting Characterization Techniques
A plethora of characterization methods are available to the laboratory scientist for
GaAs analysis. A small fraction of these are of interest to the process engineer by virtue of
their nondestructive nature. We have already mentioned the fact that some nondestructive
techniques are undesirable for in-process GaAs evaluation, but in this section, we will look
at three techniques-photoluminescence (PL), infra-red transmittance (IRT), and raman
scattering (RS)-which have been selected for review here based on their acceptance in the
literature for use with GaAs process characterization.
2.6.1 Photoluminescence
PL spectroscopically identifies and quantities GaAs defects by measuring the intensity
of the light emitted by the recombination of photoexcited carriers. The basic PL apparatus
is fairly straightforward but variants such as scanning or spatially resolved PL (SPL or
SRPL) can lead to much more elaborate systems. Deep level measurements require cooling
which is a major drawback to using PL in a production environment.
A standard PL system is similar to PIMR in that it is a "pump and probe"
measurement. The exciting source is a laser, usually generating a few watts of red (above
bandgap) light, whose signal is mechanically chopped and then directed toward the SUT.
If used, focusing lenses or optical fibers will determine the optical spot size and, therefore,
the spatial resolution. With single mode fibers this can be on the order of microns. The reradiated signal, or luminescence, is filtered using a spectrometer and detected by a photomultiplier (PMT). The PMT and a reference signal from the chopper feed a lock-in
amplifier. The PL measurement itself corresponds to stepping the spectrophotometer over a
50
range of wavelengths while recording the output voltage of the lock-in.
The dominant physical process in PL is band-to-band, hole-electron pair generation
followed by carrier recombination and a corresponding emission of band energy photons.
There is, however, sufficient photon flux to also excite many carriers from deep and
shallow levels. Naturally these excitation processes are much less likely than the band-toband process, and the recombination events which could giveriseto luminous output occur
at rates many orders of magnitude below that at the band energy.
The majority of the carrier generation takes place within a few absorption depths
(about 0.5 (im or so for red light in GaAs) followed by diffusion on the order of a
diffusion length. Depth sensitivity is further limited by re-absorbtion of the luminescent
photons; not all recombination events which emit a photon get detected as the photon must
make its way from the spatial point in the wafer where the recombination took place to the
surface before it can be re-absorbed. Because of these absorption and diffusion limitations,
only the surface is characterized in a PL measurment.
A typical PL measurement for SI material taken at Rensselaer is shown in Figure
2.6.1. This scan was taken at about 10 K. The strong feature at 1.494 eV corresponds to
the recombination of a free electron with C^s [Skromme]. If the Si concentration where of
the order of that for C, a feature corresponding to the SiQ a -C^ s exciton could be seen at
1.492 eV[Bindal 1989b].
Interpreting PL results such as this are complicated by several factors. A key problem
in quantifying PL-detected defects comes about because some recombination processes do
not involve photon emission. In effect, these nonradiative centers make it difficult to
normalize one wafer's PL intensity to another's as luminescent intensity may decrease from
51
sample to sample due to a process which is not related to the particular defect of interest.
Nonradiadve centers can also reduce the overall sensitivity of the measurement. Sometimes
this is countered by increasing the laser intensity, but there is an upper limit to incident
power since the laser can physically alter the defect concentrations at intensities well below
that where ablation occurs. Another problem is that the most interesting PL features are so
close together in energy that cooling to at least liquid nitrogen, and typically liquid helium,
temperatures is necessary to discern one recombination process from another.
Identification of EL2, for example, requires that the sample is cooled to 10 K, or colder,
which is difficult and time consuming for an entire 3 inch wafer [Skromme].
To facilitate the use of PL as a process control tool, the cooling system is not used and
measurements are done at room temperature. This results in significant loss of capability as
individual levels can no longer be identified. In this mode, PL can be used to identify
problems with surface polish (subsurface damage), dislocations, and general implant
damage. The utility of room temperature PL is augmented when spatially resolved or beam
scanning systems are added.
SRPL or SPL systems allow for the 2D mapping of a wafer by rastering the laser
across the wafer surface and measuring the luminescences with a video camera. An
example of SPL is shown in Figure 2.6.2 for undoped SI GaAs [Krawczyk]. While no
longer capable of yielding information on individual defects, these changes to the basic PL
system provide a quick and relatively easy to use method of determining overall surface
related damage. SRPL/SPL still do not provide access to the bulk properties of the SUT
once processing has changed the surface.
52
PHOTOLUMINESCENCE
:
INTENSITY
VS
VRVELENGTH
1.4939
I
4/3/91
Material : Gafls «5
Temp <K> : 10
PM Uolts : 500
S l i t Uidth : 250
i.e -i
1.4916
e.8 .4989
0.6 -
0.4 -
e.2 -
e.e 8240
1.505
8273
1.499
CURSOR <Angstroms) : 8248.00
CURSOR (eU) : 1.5046
RtlPLITUOE I 0.048
8365
1.493
WAVELENGTH <Rngstoms> ->
<- ENERGY <eU>
8337
1.487
Maximum Signal <volts) i 3.997E-85
8376
1.481
Fig.l Typical SPL images (gray level and pseudo 3-D plots) of LEC Fe-doped S.I. InP crystals;
dark pixels correspond to high PL intensities; a) image of a 2 inch wafer; b) image of a 500um
x 500(i m area. These images reveal the presence of long-range siriations, short-range
striations, dislocations and sub-surface extended defects.
Figure 2.6.2
Example of SPL Measurement
[Krawczyk]
54
2.6.2 Infra-Red Transmittance
IRT is a complementary technique to PL in the sense that it is a characterization tool for
bulk properties only. This is a consequence of the fact that the transmitted signal must
interact with the entire wafer thickness before detection. Thus, the defect population will
have an integrated effect. The technique is mainly used to measure the neutral and ionized
EL2 concentration, but some IRT work has been done with stress and dislocation density.
A key advantage of IRT for deep level investigation is that the work can be done at room
temperature.
An IRT system consists of a broad band optical source (a tungsten filament bulb),
spectrophotometer, and an optical detector. The IRT system is calibrated by recording the
detector output without the SUT as the spectrophotometer is scanned across the
wavelengths of interest. With the SUT in place, this procedure is repeated and the percent
of the light transmitted is calculated. IRT spatial resolution will be a function of both the
radiating and detecting aperatures.
Below the bandedge at about 1.1 \im, the optical absorption within GaAs is dominated
by EL2. Reflection from the surface, however, contributes to most of the difference
between the incident and detected signals and this must be corrected for by a calibration
measurement at a somewhat longer wavelength of about 1.8 |im. The difference between
the detected signal at about 1.1 |im and the theoretical transmission signal (based on the 1.8
lim data) is due to absorption by EL2. Absorption by the deep level is the dominant
mechanism in altering the signal rather than scattering of the optical beam [Skolnik].
By doing two measurements near 1.1 |im and accounting for the difference in the EL2
optical ionization cross-sections at these wavelengths, the neutral and ionized EL2
55
concentration can be determined [Kazior]. This technique has the advantage that once the
the neutral and ionized EL2 concentrations are determined, the net shallow acceptor
concentration can be determined by using the TLM thereby providing a fairly accurate
picture of the bulk SI defect structure.
The difficulty with IRT arises from the high absorption coefficient for below bandgap
radiation in GaAs. Off-the-shelf spectrophotometers, such as the Perkin-Elmer 330, can
measure transmittance to about 0.1% while more advanced single wavelength IRT systems,
like that described by Dobrilla [Dobrilla], are accurate to about 0.01%. Multiple frequency
techniques, such as Brierly's, improve the measurement sensitivity for 20 mil wafers but
error can be still be on the order of 10% to 20% [Wang]. Based on thesefigures,accurate
mapping of the EL2 concentrations requires measurement samples thicker than the usual 20
mil wafer or combined bowing and flatness specifications of less than 50 p.m across the
wafer [Dobrilla]. The former solution is typically used as thick wafer slabs of about 120
mils are normally available from local-vibrational mode (LVM) measurements to determine
C concentration.
LVM is essentially an IRT method, although the physics involved for C measurements
are different from those involved with EL2. At 580 cm" , in the mid-IR range, C^ s will
absorb energy and vibrate in its lattice position. The C acceptor concentration can be
quantified by determining the absorption at this wavelength. The difficulty here is that the
equivalent to the EL2 optical ionization cross-section is not easily measured, and published
calibration factors can yield unacceptable variations in the C concentration.
Because transmission measurements are inherently a function of all the material
56
through which the signal passes, IRT is unsuitable for characterizing any parameters
associated with an implanted channel. IRT can be used to characterize bulk properties of
annealed sample as long as the annealing effects are fairly uniform through the wafer
thickness (extreme As loss from the surface or slip will change the reflected signal by
added scattering) and the bowing/flatness requirements can still be met.
2.6.3 Raman Scattering
The phonon modes of a crystalline structure have associated with them characteristic
frequencies which alter the electric polarizability of the SUT. Two different modes will
usually be identified by the differing degrees to which the polarizability is effected. RS
takes advantage of this uniqueness by allowing incident radiation to interact in this manner
with the phonons. Analyzing the scattered signal using the laws of energy and crystal
momentum conservation identifies the active phonon modes. RS is a form of modulation
spectroscopy, albeit with the modulation arising from sources internal, rather than applied,
to the SUT.
An RS system is similar in many ways to a PL system. An above bandgap laser is
incident upon the SUT at some non-normal angle. Since scattering events are involved, the
signals of interest will be at diffuse angles so the spectrophotometer and detector must be
positioned by some goniometrc arrangement. The spectrophotometer must be in the midto far-IR as GaAs optical phonons (with wavelengths in this range) are generated or
annihilated in creating the frequency shift. Cooling is also used to reduce noise associated
with lattice vibrations. Any apparatus, such as lenses or fiber optic, which tends to focus
the incident optical signal will ultimately determine the spatial resolution of the RS system.
57
In a perfect crystal, only certain phonon modes exist based on the group, or
symmetry, of the crystalline structure. If defects are present, certain normally forbidden
phonon modes will be activated. The result in a RS measurement is that a detected signal
will appear at a previously disallowed energy corresponding to the activated phonon mode
wavelength. The activated mode, transverse optical or longitudinal optical, is determined
from its energy and scattering angle, or reciprocal space value.
RS has been used by Yano [Yano] to measure the diffusion constant of excess As in
GaAs by monitoring a TO mode at 200 cm"* which has been attributed to crystalline As.
The self-compensation of implanted silicon was investigated by Wagner [Wagner] at about
378 cm-1 for SiQ a donor and 393 for the Siy^ acceptor. Nearly the same wavelength for
SiQa was found by Holtz [Holtz] after Si implantation but before any annealing. RS has
also been used to study the dose-dependent damage of implanted Si by following the
behavior of forbidden modes as the implant dose is varied.
The drawbacks of RS are similar to those for PL. The resolution necessary for
qualitative comparison of samples grown or processed under different conditions demands
a significant investment in time and equipment for the purpose of cooling the SUT. The
technique is also limited to surface characterization due to the use of an above bandgap
probe beam.
2.7
Summary
In this chapter, microwave photoconductance has been introduced and reviewed. The
relationships between ABG and BBG steady-state photoconductance and specific traps has
been identified and it has been demonstrated that, under the conditions in which PIMR
58
operates, these photoconductances produce linear changes in the detected microwave
signal. The models of ABG and BBG photoconductivity will be expanded upon and
improved in the next chapter.
The properties of SI GaAs have also been explored. The compensation mechanism
between EL2 and C in undoped, material, as well as the characteristics of other intrinsic
and extrinsic defects, was discussed. EL2 and C are important because they are electrically
active and are present at rather high concentration levels relative to other, active, defects.
Native defects in addition to EL2 also play a key role as they are most often associated with
trapping and recombination phenomena. Implanted MESFET technology relies heavily on
Si as an n-type dopant.
A model which predicts the electron concentration, shape, and residual defect
concentration for a GaAs ion implanted channel must contend with several significant
issues not previously encountered in silicon technology. Starting material stoichiometry is
a factor in determining the properties of the channel. Channeling adds an additional
variable which, while statistically predictable, increases the complexity of the problem.
The contribution of dislocations to channeling and the production of native point defects in
the process of stopping the implanted species must also be considered. Even after these
effects have been included, a complete model of active channel formation should must
consider the thermodynamic implications of a high temperature anneal.
A model which provides an adequate basis for considering all of these factors has been
introduced and developed by Morrow. Built around the concept of stoichiometric
constraint, this model has demonstrated its capability to model a wide range of GaAs active
layer issues, such as bulk defect constituency, coimplantation, and capping material.
59
While annealing conditions could be investigated with Morrow's formalism, Bindal's
model is developed in a more useful form for studying annealing effects mainly at the
expense of considering some of the more complicated defect interactions.
For the lack of a comprehensive, qualitative understanding of the relationship between
material and process variables and the properties of the active channel, process monitoring
is a useful tool in successfully fabricating MMICs. However, even here there is a problem
as some techniques are destructive or require extra time and equipment for cooling. Most
techniques are also limited to characterizing either the surface or the bulk, but not both.
PIMR, in contrast, does not suffer in any of these respects: it is fast, nondestructive and
can characterize both the surface and bulk of starting or processed GaAs.
60
CHAPTER 3
PIMR System and Modeling Enhancements
3.1
Introduction
Using the groundwork laid in the preceding chapter regarding microwave reflecting
(MR) systems, photoconductivity, and GaAs defects, the aim of this chapter is to provide
information on the specific PIMR system used along with a phenomenological
interpretation of the measurements of processed GaAs through active channel formation.
The discussion of the system focuses on its basic design and operation, and the
enhancements which have improved PIMR's measurement sensitivity and augmented its
functional capability.
Two-dimensional mapping results show a heretofore unknown
measurement effect in which the detected microwave signal varies with wafer position. A
second, wafer-to-wafer variation due to a conducting, channel layer which attenuates the
PIMR measurement is discussed. The repeatability of this upgraded PIMR system is
determined. The system is also calibrated against the same standards used in an earlier,
steady-state confirmation of the TLM and PIMR, except the focus here is on the peak
transient response.
The model quantifying EL2 and the shallow acceptors from PIMR data is modified
in several respects. First, a relationship between ggBG an<* ^d *s derived by a numerical
solution to the peak transient response. This requires adding a third defect level, the
doubly ionized EL2 level, to the GaAs photoconductance model discussed previously.
61
Similarly, a simulation for g ABG a s a Unction of shallow acceptors is presented. Finally,
modifications are introduced which extend the model to PIMR with ion implanted material
or where mass transport has occurred during annealing.
3.2
Experimental Set-up
In this section, the equipment used in the current generation of PIMR is described in
terms of the basic subsystems. Two simplifications made in the previous chapter while
analyzing GaAs photoconductivity and MR systems are justified based on PIMR
component parameters. This system represents a significant improvement in measurement
sensitivity from earlier versions [Campbell]. The majority of the effort on the system itself
focused on automating the equipment to speed data acquisition and facilitating repeatable
two-dimensional (2D) peak PIMR transient response maps.
3.2.1 The Basic System and Its Operation
PIMR is a pump and probe photoconductivity measurement shown conceptually in
Figure 3.2.1. The microwave probe subsystem is modulated by photoconductivity effects
induced in the SUT by the optical pump subsystem. This modulation can be either steadystate or transient; the latter being used here. In the transient mode, the peak magnitude, as
well as growth or decay time constant, can be acquired by the controlling microcomputer.
A motorized stage allows data to be taken along crystallographic directions of interest as
well as full-wafer maps. Measured PIMR data is transferred to a workstation for analysis.
The microwave portion of the set-up was chosen with several factors in mind. Since
sensitivity to an ion implanted channel is desired, the microwave energy should be spatially
distributed so that the photoconductive modulation having the greatest effect on the
62
Detection/
Measurement
Electronics
T
I
Controller
•
Data Processing/
Analysis
Figure 3.2.1
Conceptual PIMR System
63
I
microwave signal is at or near the surface of the wafer. Second, the microwave energy
should be focused in some way to help measurement sensitivity and mapping resolution.
Finally, we would like PIMR to be as sensitive as possible, which requires minimizing
passive losses and noise sources.
Several of the above requirements can be related to the microwave subsystem
frequency. In order for there to be maximum sensitivity to the channel, an electric field
maximum should be present at the wafer surface. The stage supporting the wafer will act
as a short so electric field maximums will be at odd multiples of quarter wavelengths in the
GaAs as measured from the stagetop (assuming that the microwave signal propagates as a
traveling wave). That is
^
W
*
(3-U)
where n=0,l,2,.... Taking the wafer thickness as 20 mils (500 (im) and the dielectric
constant of GaAs as 13.1, an operational frequency of about 41 GHz results for n=0 while
the frequency corresponding to n=l is 123 GHz, etc While larger values of n offer
potentially better spatial resolution by virtue of their smaller wavelengths, component costs
are prohibitive at these frequencies, and we therefore limit consideration to n=0.
One of the conditions for sensitivity to the transient is requiring the substrate to appear
as a good dielectric to the microwave signal. This requirement is just the condition for
negligible loss tangent at the frequency of interest:
-2-«l
(3.1.2)
64
where e = EjE0. For an SI GaAs wafer measured at about 41 GHz, a/ax, is approximately
1(T7.
Based on these considerations, 41 GHz appears to be a good frequency for GaAs
PIMR. We chose 36 GHz, however, for practical reasons since staying within Ka-band
(26.S to 40 GHz) further reduces the cost of components (which are available from a larger
supplier base than at 41 GHz) with only a slight sacrifice in depth and transient sensitivity,
or spatial resolution.
The microwave portion of PIMR, based on these design
considerations, is included in Figure 3.2.2. The differences in the microwave subsystem
between this and earlier systems is the absence of an attenuator in the reference arm, a more
stable and powerful microwave source, and a modified antenna; the remaining components
are the same and are discussed in detail elsewhere [Gutmann]. Furthermore, the dark
microwave patterning phenomena necessitated operation of the PIMR system in a nontransient mode without the lasers and modifications to the system reflect this need as well.
The microwave oscillator is a free running, mechanically-tuned, Gunn diode
operating at 35.74 GHz. The oscillator provides 150 mW of available power which is
significantly better than the 6 mW used previously [Campbell]. Thermal drift is minimized
with a cooling fan while the DC bias voltage is repeatably set by an external regulator. An
isolator limits feedback to the oscillator from the rest of the microwave system. Before
entering the measurement portion of the microwave set-up, a wavemeter (for frequency
measurement) and a directional coupler/power meter arrangement are included for
monitoring the oscillator output parameters.
One of the more critical components in any MR technique, and PIMR as well, is the
65
36 GHz
Qunn
Diode
I
Power
Meter
Thermistor
Mount
Sliding
short
Isolator
Wave
Meter
3dB
Coupler
£
Crystal
Detector
Laser
Diode
Optical transient
Digital
Osc.
Trigger
Source
Dark non-transient
IBM PS/2
COMPUTER
Image Processing
Macintosh
n
Figure 3.2.2
PIMR System
The equipment offers to modes of operation, as indicated by the "Optical transient" and
"Dark non-transient" arrows prior to the digital oscilliscope. The former is the actual PIMR
measurement system, while the latter is a non-photoconductive MR system with no optical
excitation (pulsed or CW).
66
antenna or coupling element. This part of the set-up has been the focus of several
investigations [Jensen, Heimlich 1987] with the intent of understanding and improving a
variety of configurations for MR systems. An open tapered-tip antenna has typically been
used with the PIMR systems. The tapered-tip portion of this structure is shown in Figure
3.2.3 and is composed of two conducting fins which guide the signal from a waveguide
transition structure to the wafer. Recently, the open tapered-tip configuration studied by
Jensen was modified by feeding the fin section with a rectangular-to-double-ridge
waveguide transition. The new transition provides a more gradual, and therefore lower
VSWR, transition from the waveguide mode to that propagating along the antenna.
Although this antenna has not been characterized to the extent typical of the earlier studies,
measurements have shown that PIMR sensitivity is increased by about a factor of 2 and the
microwave spot size is 1 mm~
Using this antenna under typical measurement conditions, the DC output of the
detector measured with a 50 ohm load is about 100 mV with the sliding short positioned for
peak transient PIMR response. Based on the measured characteristic curve for the detector
given in Figure 3.2.4, this is sufficient to bias the detector out of the square law detection
regime and into a linear regime. Thus, the signal detection is more accurately modeled as a
homodyne mixer. The LO signal is the sum of the reference signal, r r and the timeindependent portion of the test signal, Lexp(j<j>)ro. The position of the sliding short is
adjusted to put these two signals in phase, and thereby provide the highest CW LO power
possible (and the lowest loss in converting the RF signal to the IF frequency). Since P<j(t)
is responsible for modulating the time-dependent portion of the test signal, we expect the
67
Figure 3.2.3
Tipered-tip Antenna
68
1000d
I
I
I
I
I
100.
4)
10:
•25
15
-10
-5
10
Power, dBm
Figure 3.2.4
Detector Response Curve
Output DC voltage dependence upon input RF power.
69
RF frequency to be approximately
fRF = 35.74 GHz ± fo(t)
(3.1.3)
After mixing we get f ^ as the "downconverted" output. This output signal corresponds
to v(t), or the time-dependent detector output voltage. The difference between the
amplitudes of v(t) and the RF voltage signal is defined as the voltage conversion loss.
If the PIMR system is being used in the dark, non-transient mode, the lasers are not
pulsed and the conductivity transient, which would ordinarily modulate the CW microwave
signal, is absent. Instead, the signal at the detector remains constant in frequency but may
have an amplitude dependent on conditions at the SUT. The detector output of interest
under these "dark" measurement conditions is a DC level corresponding to the transfer
characteristics of Figure 3.2.4.
After downconversion, the transient signal enters the detection and measurement
subsystem (Figure 3.2.1). This area has also been improved over previous PIMR
implementations with a lower noise amplifier and a digital oscilloscope (Figure 3.2.2). The
oscilloscope is equipped with extensive automated measurement capability, such as signal
peak detection and falltime, as well as an IEEE-488 communication bus.
As described in the technical background, v(t) is determined by the photoconductive
dynamics within the GaAs. These dynamics are introduced by the optical portion of the
PIMR system. The ABG and BBG lasers are stacked arrays of individual semiconductor
lasers, with peak optical output power on the order of tens of watts during the 20 nsec
pulse used in PIMR. Optical fibers (-200 \im in diameter) guide the pulsed signal of the
70
wafer surface and to a great extent, determine the mapping resolution of the system. The
tips of the fibers are positioned relative to the antenna by movable, precision chucks
mounted on the same supports used for the antenna.
The lasers are controlled by separate integrated pulsing units which send trigger pulses
and, in the case of the BBG laser, regulated DC power to the laser diode units (the ABG
laser voltage is set externally using a voltmeter). These pulsers require an external trigger
source, adjusted for a repetition rate of a few KHz, and DC supply. The power supplies
and pulsers can not supply enough current to the laser diodes to create rectangular optical
pulses. Instead the 20 nsec "on" period is more ramp-like, but we will refer to the
photoexcitation period as the "pulse" nonetheless.
Operating under these conditions, the BBG laser was found to introduce about 10*"
photons/cm^ /sec at the wafer surface [Wang]. As the ABG laser operating point is similar
in energy and power to that of the BBG, the ABG is expected to have about the same
value. Wang also found that operation of these lasers in the specified regime allowed the
relative magnitude of P<j(t) to remain smaller than P 0 (condition (2.2.4)).
3.2.2 Two-Dimensional Mapping
The basic PIMR set-up, comprised of the microwave, optical and
detection/measurement subsystems, can be a useful research tool. For PIMR to be
successful in a manufacturing environment requires automating the equipment to quickly
produce 2D wafer maps. This required three additions to the previously described system:
1. An x-y-z translation stage (positioning subsystem) to repeatably position the
wafer with respect to the microwave antenna and optical fibers.
71
2. A control subsystem to synchronize wafer translation with data acquisition
and storage
3. An image processing subsystem to produce wafer maps from the raw
mapping data.
The motorized translation stage is basically a table top mounted to stepper motors in
the x,y, and z dimensions. All three axes can be repeatably positioned to within one motor
increment. This corresponds to roughly 0.5 \im based on the pitch of the drive screws
running from the motors to the table top. Registration guides (for placement of wafer flats)
occupy two adjacent corners of the table-top and can accommodate wafers up to 4 inches.
A separate controller unit decodes commands sent to the stage from the PIMR system
controller over an RS-232 bus.
The communication buses on the digital oscilloscope and the stage controller are tied to
an IBM PS/2 model 30 personal computer. Software coordinates automated measurements
within and data acquisition from the oscilloscope as well as wafer position via the stage.
Currently, software is available which automates three measurement options. One
program records complete transient waveforms at several points on the wafer along a scan
line. This program creates prohibitively large data sets because of the number of points
needed to represent the transient waveform. The second program, used most in this study,
acquires the peak photoresponse along 3 linear scan lines as shown in Figure 3.2.5. The
spacing between mapped points was chosen so that roughly 90 points are taken per 2 or 3
inch wafer after allowing for a 5 mm handling edge. The third can yield full 2D maps of
peak photoresponse for 2 and 3 inch wafers with a spatial resolution of up to 14,000 points
72
t
y
handling edge
N = total number of measured points
_ wafer diameter - 2 * handling edge
step size
Figure 3.2.5
Scan Pattern for Automated PIMR Linear Scan Procedure
73
t
y
handling edge
N-
7i (wafer radius - handling edge)
(x step) (y step)
Figure 3.2.6
Scan Pattern for Automated PIMR 2D Mapping Procedure
74
per wafer. The mapping is performed in the meander format depicted in Figure 3.2.6.
A second computer processes the data from the automated PIMR system. Another
system (other than the PS/2) was chosen for image processing compatibility. Due to the
availability of some public-domain software, an Apple Macintosh II ("Mac") was chosen.
The raw, measured data is ported from the IBM to the Mac and processed into a more
easily used form. The Mac is equipped with a disk drive compatible with both its own
format, as well as that used by the PS/2, and software provided with the Mac enables it to
convert and read the PS/2 PIMR data. Scan data is typically analyzed using a statistical
software package, while processing 2D maps is more involved.
Producing map images from the raw 2D data is done in two steps. The first step
converts the point-by-point data into a matrix format. The matrix represents a square with
sides scaled to the diameter of the wafer less twice the 5 mm handling edge. Peak data is
placed at the appropriate matrix element based on a position data file generated by the
mapping program. Points exterior to the wafer are set to zero. The matrix is saved as a file
in a format readable by the Image program, the second step in producing a map image.
The source code for the "Image" series of programs, first developed at the National
Institutes of Health, is in the public domain, and several modifications and enhancements to
the original have proliferated. The basic Image package combines simple data importing
and exporting with many image processing features, such as smoothing, equalization, and
noise reduction. The Image 1.30u program has extensive importing facilities and is used
here for this purpose as well as image manipulation and measurement.
This enhanced PIMR system has several benefits over previous systems. The
automated aspect of the PIMR measurement now makes available very large data sets with
75
negligible system drift for a wafer or series of wafers, thereby improving the quality of the
data. The automated features of the oscilloscope provide measurements which were
difficult, if not impossible, with previous systems. The oscilloscope also helps to reduce
error in the measurement through waveform averaging, as well as a subtraction scheme
which removes any stray signals which leak into the detection electronics from the laser
pulsers, due to poor shielding (Appendix A).
The current PIMR system is not without its shortcomings. The two lasers are not part
of the automated set-up, so any scan or map must be done twice-once for each laser. The
software has therefore been written so that the stage is not automatically re-initialized and
the ABG and BBG data for a given wafer still correspond to identical locations on a wafer
to within the repeatability of positioning the stage. The overall cycle of stage stepping and
data acquisition for the full 14,000 points capacity of the software takes about 10 hours per
laser. Useful maps only require about 1,000 points, or 45 minutes. This extended time is
not a fundamental limitation, but due to capture, storage, and peak detection of the full
transient response by the oscilloscope.
3.3 System Performance
Initially obtained 2D maps of peak transient PIMR response across a GaAs wafer
depicted a concentric ring structure, reminiscent of Czochralski growth rings. Further
investigation showed that this mapping feature is not related to growth rings nor, for that
matter, does it originate with a photoconductive mechanism. Instead, it is due to a timeindependent parasitic coupling between the SUT and the MR system-the PIMR system
without an optical subsystem. As shown in Figure 3.2.2, "dark" maps are taken with no
76
laser pulse and the detector directly connected to the oscilloscope. The dark data can be
used to remove the parasitic patterning from the PIMR map through image processing
techniques. This scheme is also useful for correcting the effect of table tilt which alters the
coupling between the antenna and the SUT as the wafer is mapped.
The presence of a conducting layer, or activated channel, on the face of the wafer
nearest the antenna can also change the PIMR measurement. This fact can be put to good
use as an indicator of sheet resistance. However, if other information is desired, such as
EL2 or shallow acceptor concentration, the effect of the conducting layer must be removed
from the data. A theoretical basis for this correction is discussed in section 3.3.2.
As this work focuses on interpreting peak PIMR transient data among wafers rather
than simply identifying intra-wafer defect distributions, it is informative to determine the
repeatability of the peak PIMR response and the sources of measurement error.
Measurement repeatability is determined in section 3.3.3.
3.3.1 Dark Microwave Patterning Effect
An example of typical maps with and without a laser pulse are shown in Figure 3.3.1
for a 3 inch as-grown, undoped SI wafer and will serve as a basis for comparison with
dark maps for other samples. The peak-to-valley difference in PIMR maps can be as high
as 80% of the wafer average while the dark maps have no more than 10% variation. The
rings, spaced 3 to 4 mm apart, are clearly evident in both PIMR and dark images,
indicating a one-to-one correspondence between the two maps.
Three other material systems are shown in Figure 3.3.2. An implanted and annealed
wafer with a 500 ohm-cm channel channel exhibits a much fainter patterning effect (a peak
77
b)
Figure 3.3.1
Typical Maps for Undoped SI GaAs
a) Peak PIMR Transient Response Map
b) Dark Response Map
78
a)
b)
c)
Figure 3.3.2
DC Maps for Materials Systems with Various Conductive Properties and Geometries:
a) Ion Implanted and Annealed GaAs
b) 1 ohm-cm Silicon
c) SI InP
79
to-valley difference of a few percent of the dark map average) while patterning is
nonexistent in a 1 ohm-cm silicon wafer. A D-shaped SI InP sample grown by the vertical
gradient freeze technique also displays the patterning effect, but the patterning follows the
boundary of the wafer. All three of these samples were 20 ± 1 mils thick.
Further investigation with non-circular samples showed that not only was the
patterning conformal to the wafer edge but it was dependent on its relative orientation to the
antenna and the stage top registration guides. If the major flat of a sample is positioned
parallel to the gap between the antenna tips, the patterning effect is fainter than if the sample
is rotated 90 degrees. Moving the sample away from the registration guides removes the
flattening found on the top and the left portions of the ring halfway between the center and
edge of the wafer (see Figure 3.3.1 for an example of the flattening).
The antenna and wafer-shape dependence suggest a parasitic coupling between the
antenna and the sample which launches a wave parallel to the stage top. This wave
propagates out from the antenna position, reflects off the edge of the wafer, and travels
back to the antenna where it is received and sent to the detector. As the relative position of
the wafer edge and antenna are altered during the wafer mapping routine, the phasing
between this radial patterning wave and the vertically-reflected, main signal changes and a
variable power is returned to the detector. Time-dependent, or PIMR, measurements
would also experience a position-dependent conversion loss as the total RF power received
at the detector varies with position.
To complete the qualitative understanding of the phenomena, an experiment was
performed to ascertain whether the parasitic mode was propagating laterally within the
wafer with a Poynting vector parallel to the stage top. The fundamental concept behind the
80
experiment is the relationship between waveguide height and effective wavelength for a
fixed frequency; that is, if the energy is propagating as assumed than increasing the wafer
thickness will increase the distance between the interference rings in the dark map since the
propagation constant will also change. The results are shown in Figure 3.3.3 where a 20
mil thick, 3" SI GaAs wafer is compared to a 120 mil slab of 3" SI GaAs. As the wafer
thickness increases, the ring structure becomes coarser.
Qualitatively, then, the antenna couples a microwave signal into the GaAs wafer which
propagates in a lateral direction, parallel to the stage top. Any phenomena which tends to
alter the antenna-wafer separation, such as table tilt, will contribute to the patterning
intensity. The wafer and conducting stage top act as a guiding structure with the air/wafer
dielectric discontinuity as the second boundary. The energy in this parasitic mode
propagates within the waveguide radially outward from the antenna. The signal then
reflects from the boundary at the edge of the wafer, setting up a 2D standing wave pattern
which is detected at the source point (i.e. the antenna).
The intensity of the detected parasitic signal is, therefore, a function of the antenna
position relative to the boundaries, mechanisms which reduce the intensity of the signal as
it propagates within the wafer, such as attenuation and transmitted energy at the
boundaries, and the antenna radiation patterning. In this planar, cylindrical waveguiding
system (neglecting the wafer flats, for the moment), energy from a point source will return
to the point source along pathways defined by a diameter, an equilateral triangle, a square,
etc. [Heimlich 1992b]. The distance and number of reflections associated with a pathway
is indicative of the intensity lost by radiation and attenuation, since any reflection of the
parasitic signal from the air/dielectric interface will be accompanied by some amount of
81
a)
b)
Figure 3.3.3
Effect of Wafer Thickness on Dark Patterning Effect for 3 inch, Undoped SI GaAs
a) 20 mil wafer
b) 120 mil slab
82
transmission into free space. Furthermore, the antenna is not a true point source in the
sense that the radiation pattern weights the energy propagated in certain directions. Finally,
based on the results with the implanted GaAs and the doped silicon, the signal is attenuated
when the guiding system is lossy.
Based on this understanding of the dark patterning effect, a first-order simulation was
developed which considered the radiation pattern of the antenna, modeled as a dipole, and
reflections from a perfectly circular wafer along the diameter only. At any given point, the
relative phase of the two signals (launched in opposite directions along a diameter, reflected
by an ideal, open boundary, and returned to the launch point) is determined. The results
for a quarter of a wafer in Figure 3.3.4 using two different values of wafer thickness: one
a nominal thickness, and the other, twice this nominal value. A radially varying pattern is
generated by the diametric reflections while the rotational dependence comes from the
radiation pattern.
More detailed simulations would include the triangular, square, etc. reflections,
attenuation by a conducting layer, radiative losses at the wafer surface, reflection from
registration guides and other non-ideal reflection coefficients, such as the major flat of the
wafer. In the qualitative model of this phenomena, the major flat will tend to cause a
flattening of the nearby rings (just as in Figure 3.3.1) while the registration guides will
introduce an added phase shift due to a reflection coefficient of -1.
Since the parasitic patterning effect is a nearly periodic perturbation to the actual data,
we expect that by averaging the measurements for a large number of points across the
wafer we can arrive at a consistent PIMR measure which can be used to compare one wafer
to another. SO or more points taken using the linear scan software produces an average
83
a)
b)
Figure 3.3.4
Microwave Patterning Effect Simulation for One Quarter of a 3 inch, SI GaAs Wafer
a) Nominal wafer thickness
b) Twice nominal wafer thickness
84
value for a dataset which is independent of wafer/antenna orientation. This is the basis for
the data presented in the next chapter.
But the dark patterning effect must be removed for the 2D PIMR maps to properly
reflect defect structure of the wafer. Removal of the patterning can be affected by one of
three possible methods:
1. eliminating the excitation of the lateral wave by reconfiguring the coupling
between the antenna and the wafer.
2. preventing the parasitic signal from reflecting from the edge of the wafer.
3. correcting the PIMR map with a map of the dark patterning effect.
The first two fall into the category of hardware solutions since changes to the PIMR
equipment would be required. In the case of the former, an improved antenna design may
suffice, with emphasis on definition and control of a narrow, main beam for reflection from
the stage shorting plain. A structure which has met with success in other MR applications
is a thin film antenna with a collimating lens [Heimlich 1992a]. The concept in the second
hardware approach recognizes that the coupled signal is a nuisance only when a standing
wave pattern forms in the absence of a matched boundary conditions at the wafer's edge.
'Terminating" the wafer edge could be accomplished by milling the shape of a 3" wafer into
the center of a large diameter, doped GaAs wafer (a silicon wafer may also prove useful as
the dielectric constant of GaAs and silicon are similar). This absorbing annulus would be
impedance matched to the SUT, completely couple the outwardly traveling parasitic signal,
and sufficiently attenuate any significant boundary reflections before re-entering the SUT.
Finally, the third solution--a software method-relies on the similarity between the transient
and dark maps to remove the patterning effect and leave the defect data. This approach is
85
undesirable since measurement time is increased by 50% over a typical two laser, PIMR
mapping run.
As the mapping data must already be processed to some degree to generate 2D images,
the software correction technique was investigated. Two methods for this are possible.
The first is a direct correction of the transient data at each point based on the measured dark
value. The second is to take both data sets into the frequency domain via a fast Fourier
transform (FFT), deconvolve the dark map from the PIMR map, and inverse transform the
result back into real space.
The direct correction technique is attractive because it is simple to implement. A
correction function is first determined based on the AC output characteristics of the diode
versus input power. At any point, the true transient response is determined by multiplying
the dark-dependent correction and the PIMR measured value. Generating this correction
function, however, proved to be quite difficult since determining the precise effect of the
dark measurement on the photo-induced transient required performing the time-independent
and -dependent measurements simultaneously, which is not possible in the current system.
The FFT approach is also simple in concept. Since the same pattern is seen in both the
PIMR and dark maps, a software filter built from the dark data removes the unwanted
frequency components from that for the PIMR. As a proof-of-concept test, dark and PIMR
peak transient linear scans for an undoped SI GaAs wafers were processed in this manner.
The filter was constructed from the dark transform by adjusting the amplitude of the dark
data to values between 1 and an arbitrarily chosen attenuation factor of 3 to 100, and then
taking the inverse of the scaled dark data (e.g. all of the filter varies between 0.12S and 1
for an attenuation factor of 8). Multiplying the filter and PIMR data, point by point, in the
86
frequency domain is equivalent to deconvolution. The corrected data is then brought back
into real space by an inverse transformation.
Results for this procedure are shown in Figure 3.3.5. The measured scan is seen to
have a rather large peak-to-valley ratio which is reduced by more than an order of
magnitude after FFT correction. Furthermore, subtle features in the original, which were
masked by the patterning phenomena, can now be seen (e.g. dip at scan position 10). The
drawback to this rudimentary image processing is that the wafer edge introduces a
tremendous discontinuity in the data and degrades both the filter and the PIMR transform
before filtering.
Currendy, the FFT correction method suffers in several respects. 2D FFT routines
with sufficient matrix capacity for the more detailed maps are not available. This requires
software which is capable of manipulating very large double precision, floating point
matrices-not a trivial matter with either the Mac or the PS/2-but work is continuing in this
area and has shown considerable progress [Atwood]. Secondly, the attenuation factor used
in building the filter must be determined to enable turn-key data correction in the course of
generating maps from measurements. Finally, more advanced image processing algorithms
must be employed to eliminate the effects of the numerical discontinuity at the wafer edge.
3.3.2 Measurement Attenuation bv Thin Conducting Layers
It has been regularly observed that GaAs samples which have a conducting layer have
significantly lower peak transient BBG PIMR responses than their anneal-only
counterparts. For a 600 ohm/sq channel, the PIMR measurement decreases by 35% to
40%. The relative immunity of the BBG measurement to the channel itself suggests that
87
1.9-
-
1.85
1.8
>
06
1.75
1.7-
s
1.65
0-
1.6-
c
a
CQ
1.55
1.5
1.45
1.4-
0
—T"
5
i
10
•
15
i
20
•
•—•!——»•
25
30
Scan Position
Figure 3.3.5
FFT-Corrected and Measured PIMR Scans for Undoped GaAs
Data corresponds to a scan from the wafer edge (Position 1) to the center (Position 28).
this is not dominated by an altered defect structure, but rather by the conducting layer
affecting a reduction in the microwave and/or optical signals in the bulk of the SUT. In this
section, we explore several mechanisms in the PIMR system itself, as well as related to the
SUT, which contribute to this effect
We begin by considering the ramifications of a conducting layer in close proximity to
the radiating antenna structure. If the material is in the near-field region of the antenna, it is
possible that the efficiency of the antenna could decrease. A simple model of the
antenna/SUT arrangement incorporates a transmission line for the fin section of the antenna
as a transmission line and a shunt resistance for the channel (Figure 3.3.6a). More exactly,
the air gap between the the antenna and the wafer adds a series capacitance between the
transmission line and the shunt element, thereby reducing the degree to which the signal is
attenuated, but we will ignore this to simplify the model and arrive at an upper bound for
the signal loss. Finally, the portion of the signal propagating through the SI GaAs and
reflecting from the shorting plane is modeled as a terminating impedance. Assuming TEM
plane wave propagation, the reflected signal presents an impedance of j474 ohms at the
surface of the wafer.
Without the channel, the resistor is very large compared to the characteristic impedance
of the line, and the signal is not attenuated, but as the conductance of the channel increases
and has a resistance on the order of the line's characteristic impedance, the propagating
signal will be reduced. Although the modified antenna used here has not been thoroughly
analyzed, it is similar to the that previously studied in detail [Jensen]. The fin section of
this earlier antenna was shown to have a characteristic impedance of 170 ohms as well as a
possible near-field component for the antenna/wafer separation used here.
89
Air Gap
Hf
j470 ohms
500 \un GaAs wafer
+ shorting plane
170 ohms
Antenna Fin Section
a)
1
.95
9
.9
•
€>B .85
.8
fa
.75
55
"O .7
4)
£ .65
9>
33 .6
M
4> .55
tf
.5 500
^
<W
•
750
I
•
I
•
I
1000 1250 1500 1750 2000
Sheet Resistance, ohms/sq
I
— 1 — ^ ^
2250
2500
b)
Figure 3.3.6
Microwave Antenna Coupling to Conducting Channel
a) Transmission Line Model
b) Microwave Field Strength at Wafer Surface vs. Channel Sheet Resistance
(field strength normalized to field strength without a channel)
Air gap impedance chosen to be large compared to the j470 ohms of die wafer.
90
The reduction in microwave power in the fin section for SUTs with a channel can be
determined by first finding the equivalent impedance for the channel and wafer,
_ (Sps)(j470)
Zload=
8p s + j470
where 8 is the number of squares of material interacting with the antenna and p s is the
active layer sheet resistance, and then solving for the average power at the load (wafer)
relative to the average power in the absence of a channel. The parameter 5 is a function of
the length, 1, (in the direction of current flow, or parallel to the spacing between the antenna
tips) and width, w, of the channel region interacting with near-field
A plot of this relative field electric field strength at the wafer is shown in Figure 3.3.6b
calculated as a function of channel sheet resistance. The number of squares used to
calculate the load impedance for Figure 3.3.6b was chosen to be two-thirds which gives
fairly good agreement with the attenuation for the 600 ohm/sq channel. A 8 less than 1 is
reasonable based on the 2D mapping results which show a a stronger field component
perpendicular to the spacing between the antennatipsthan parallel; that is 1 > w.
Confirmation of this model should follow directly from a comparison of dark, nontransient data from adjacent, or nearly adjacent, samples with and without conducting
channels. In some instances, the DC detector voltage did decrease in the presence of a
channel; however, this was not the case for all such comparisons. Wafer thickness
91
variations (±10%) and the adjustment of the sliding short may combine to produce this lack
of consistent agreement. A transmission line analysis of a similar test system using doped
silicon wafers demonstrated excellent agreement with measured data [Bothra].
This is not the only potential source of PIMR attenuation. The optical signal may also
be subject to attenuation by a thin conducting layer [Blakemore] in which free electrons
absorb the photon energy. In this case (free carrier absorption), the attenuation of the
optical intensity goes as
~ = exp(-<nia)
o
where I is the optical intensity after passing through the channel, I 0 is the incident optical
intensity, a is the channel conductivity at the optical frequency, r| is the characteristic
impedance of lossless GaAs, and a is the channel thickness. The channel conductivity at
such high frequencies must be adjusted accordingly when the ac frequency is of the order
of the momentum relaxation time [Seeger], which in this case reduces the low frequency
conductivity by roughly an order of magnitude. The resulting attenuation of the optical
signal is about 1% for a 600 ohm/sq channel.
The Franz-Keldysh effect (FK) is also a factor (for BBG PIMR) since the high
concentration of surface states will deplete a part, or all, of the channel, resulting in an
electric field. The surface depletion layer (SDL) is typically about .1 |xm deep and FK
increases the absorption depth to no less than 1 |im (about the same value as for ABG
excitation), thereby resulting in additional attenuation of the optical intensity by no more
than 5%. The impact on PIMR depends on whether or not these carriers immediately
92
recombine in the SDL or at the unpassivated surface. For this reason, as well as the
similarity between the FK-augmented BBG absorption coefficient and the intrinsic ABG
absorption coefficient, we expect both the ABG and BBG PIMR to be attenuated by no
more than 5%.
Other mechanisms for PIMR attenuation are possible depending on the material and
the processing. For implanted but unannealed wafers, the implant damage contributes to an
increase in the BBG absorption coefficient of about four orders of magnitude [Brierly].
Localized electric fields within the damaged, unannealed channel will also increase a by
FK. But for MMIC-quality material, implant damage of this extent is not present and we
can expect a negligible effect on PIMR.
As a practical matter, the sheet resistance must be determined so that PIMR can be
corrected for these attenuation effects. This can be accomplished in one of two ways. A
sheet resistance measurement could be be performed using a standard contactless
technique, such as the eddy current method [Miller]. The other alternative is the dark MR
measurement which could be used to determine sheet resistance.
In chapters 4 and 5, a third option is employed which is to determine an attenuation
factor by comparing the BBG PIMR data for a sample with a channel to an anneal-only
sample from the same or an adjacent wafer and then use this to correct the ABG data. The
obvious disadvantage is that an anneal-only sample is required which presumably is
identical in all respects to the implanted and annealed sample, except for the active layer.
To summarize, the presence of a conducting layer in the near-field of the microwave
antenna is the major contributor to the observed attenuation of PIMR with implanted and
annealed wafers. This microwave attenuation effect is a function of the active layer sheet
93
resistance and is on the order of 40% for a 600 ohm/sq channel. Free carrier absorption,
an attenuation of the PIMR optical signal, contributes an additional 1%. Photogenerated
carriers in the SDL will decrease the PIMR measurement to the extent that these carriers
recombine instantaneously (relative to the PIMR optical pulse duration) within the SDL or
at the surface. The effects within the SDL attenuate PIMR by no more than 5%. These
results will be compared to experiments involving both annealed and unannealed implants
in section 4.1.
3.3.3 Repeatability
With the improvements made to the PIMR system, we would expect that the peak
transient response measurement is more repeatable. Several experiments were performed
to quantify the uncertainty in a PIMR measurement and identify potential sources of error.
As part of the experiment, the entire PIMR system was shut-down for at least 20 hours
before beginning a new experiment. This allowed the entire start-up procedure to be
included in measuring the repeatability. Most importantly, the position of the stage relative
to the antenna was reset so that stage initialization is also a variable in determining day to
day error.
In the first experiment, 10 consecutive linear scans ("scans") with the ABG laser were
performed on a typical undoped SI GaAs wafer. Nothing in the system was altered
between runs. The standard deviation for this data was 0.8%. This deviation is attributed
to measurement error in the oscilloscope's 6 bit A-to-D converter which is expected to
produce a minimum error equivalent to half of the least significant bit (1/2 LSB) or 0.78%.
The manufacturer's specification for the scope is slightly greater than this since some
94
internal reference voltages must be set, depending on the voltage offset and resolution of
the oscilloscope.
A point-by-point correlation coefficient was also determined, comparing the first run
to the subsequent nine runs. The correlation coefficient was 0.95 or greater, indicating
excellent repeatability of the variations within the dataset, and, therefore, with the
positioning of the stage.
In the next experiment, 11 data sets were generated. The first run in this data set was
used as a calibration run to normalize this second experiment set to the first Between each
of the remaining ten runs, the system error signal (used to subtract stray signalsfromthe
transient response-see Appendix A) was remeasured with insufficient accuracy (by not
allowing enough time for the oscilloscope to extensively average out the random noise in
this signal). The standard deviation within this experiment increased to 0.9% while the
correlation coefficient between runs remained 0.95 or better, implying that the procedure
used to acquire the error signal is sufficiently stable to yield high quality measurements.
When combined, these first two experiments had a standard deviation of 1.0% and a
95% confidence interval of 0.4% for the two-tail, student t-test. From this we conclude
that the calibration procedure provides adequate reproducibility. That is, by calibrating the
system to a "standard" wafer before an experiment and then normalizing the experimental
data to the calibration, the peak transient PIMR response is repeatable to less than 1%.
Without the calibration procedure, the standard deviation for the combined data sets was
2% and a 95% t-test confidence interval within 0.8% of the mean.
In addition to remeasuring the system error signal between runs, the third and final
experiment reset the ABG laser output voltage by momentarily turning the laser off before
95
starting a run. A calibration scan was again performed before the 10 measurement runs
making up this experiment. The standard deviation of this data set increased to 1.5%. The
correlation coefficient between runs remained unchanged from the previous two
experiments.
Datasets for the first and last experiment were combined and analyzed. With the
calibration procedure based on normalization to a standard wafer, the standard deviation
was 1.2% and, without calibration, 2.5%. The 95% confidence intervals were 0.5% and
1.1% with and without the calibration, respectively.
From these repeatability tests we conclude that we can expect the one "sigma", or
standard deviation, accuracy of PIMR to be 1.2%. The biggest contributions to inaccuracy
are digital resolution in oscilloscope and error in setting the ABG laser voltage. BBG
performance is expected to be repeatable to within 1% since an internal regulator sets the
laser drive voltage. In both ABG and BBG measurements we expect further degradation
on a point-by-point basis as measured by the correlation coefficient across experiments
mainly due to the re-initialization of the stage between experiments.
3.4 PhenpmenplQgical Model
The Two-Level Model, presented in section 2.2.3, is applicable to unprocessed,
undoped SI GaAs photoconductance in the steady state: G ^ B Q is a function of N^ and
GgBG is determined by N ^ and Nj. Wang found that this model could not be used to
explain the steady state BBG data, and made suggestions for its improvement [Wang]. In
this section, a model for the peak transient BBG photoconductance, gjjBG' *s developed
from computer simulations for (undoped SI) starting material based on these suggestions.
96
A similar approach is used for g^BG which *s m e n refined by cross-calibrating the
simulated results with actual PIMR measurements of samples with known carbon
concentration.
The models for g£BG
anc
* SABG when combined form the
phenomenological model (PM) of PIMR. Diffusion from the surface and its effect on the
extraction of N a is also presented. Finally, we combine several of the discussions in this
chapter with the PM to arrive at an extended PM (EPM) for use with ion implanted GaAs.
3.4.1 Undoped SI GaAs Characterization bv Peak PIMR Transient Response
The BBG portion of the two-level model (TLM) was uncorrelated to EL2
concentration [Wang]. Based on transient decay results, Wang deduced some of the
properties of a third level, which he identified as the doubly ionized EL2 level. Here, we
begin by developing a relationship between the EL2 concentration and gBBG' ^ l a t e r d°
the same for carbon and gABGDuring the BBG pulse, electrons are promoted from both the singly, E + , and doubly,
E + + , ionized EL2 levels, independent of depth. Assuming low level injection, the carrier
continuity equations for free electrons generated from each level can be written as a
nonlinear initial-value problem,
TT = <fWfl*A-*J
^
- Cnn(NA+n+)
= °> P h<VV>-C„2 n V
n(0) = n+(0) = n++ (0) = 0
(3.4.1)
(3.4.2)
(3.4.3)
[Wang] where n = (n + + n + + ) is the total photoexcited electron concentration, and the o°
97
and C terms are the optical ionization and capture cross-sections, respectively, for electrons
associated with the two EL2 levels.
The nonlinearity of this problem together with the generation, or forcing, term makes
an analytical solution difficult. (Note: since "pulse" is actually a ramp, the N p h above is
implemented as tN p }/r, where t is time and T is the "pulse" duration.) Therefore, a
numerical solution (Appendix B) was implemented using a simple finite difference
implementation [Press, and Sewell] of (3.4.1), (3.4.2) and (3.4.3) to find the peak BBG
photoconductance for the 20 nsec pulse as a function of EL2 and shallow acceptor
concentrations. Values for the various constants are taken from the curve fittings by Wang.
Results for the numerical solution are shown in Figure 3.4.1 as a photoconductance
surface plot in arbitrary units as a function of carbon concentration in units of lO1^ cm"-*
and EL2 concentration in units of 10 *" cm"3. In the region of typical SI material, a half
order of magnitude change in N ^ has about the same effect as a 10% change in N^. Curve
fitting these results gives
gBBG = 3 - °- 7N A + 5 - 2 N d
<3-4-4)
as a more tractable form, where gBBG is in units of sheet conductivity (sq/ohm), N ^ is
units of 10*5 cm"3 and N<j is in units 10 ^ cm"3. (3.4.4) is applicable so long as the
material is SI by the three-level model and the EL2 concentration remains high enough to
insure low-level injection. Thus, the changes in the peak transient BBG photoconductance,
ggBG' a r e
t0
f*rst or(ter, dominated by N<j and the basic interpretation of the TLM for
G Q B G *S applicable to that for gBBG-
98
M
•
•
r
,-
.
.
.
.
.
.
.
.
t
J
I ;.t 1 4
11
1.5-
i
i
/
/ -
I
U
I
/
/
•• l . - f i
I
/
/
i
*
il
i
*
/
i 'i
i
1
I
i
•
i
'
'
I
10
4
i
I
\ *
i
\
I
:- L -i
<
;,i :
'*
\ *
I
}r—{
i\
i
I
1
il
i
j
1
il
*
'*
I i
\o
i!
'i
,'i
*
*j
i
i
!•:
I
S
(
S
{
I
1.5
i.l
I.L
1.:
1.4
J.
1.^ l.o
1./
•. /
1.3
1.9 5.0 2.1
U
EL2
Figure 3.4.1
BBG Peak Transient Photoconductance Simulation
Photoconductance in arbitrary units as a function of carbon concentration, 1 0 ^ cm"3, and
EL2 concentration, lO 1 ^ cm"3.
99
Moving on to a solution for gABG» we r e t u r n t o m e continuity equation for the free
electrons generated by laser excitation, (2.3.2), and modify this for the ramp-like optical
excitation, giving
3n _ 3 n n
. .^oh*
•ST = D-^r - - + <xexp(-ax)-£L
T
*
3x 2 x
After solving for n(x,t), g^BG *s f° un d by
SABG
~ ^J n ( x ' t ) <**
(3.4.5)
<3-4-6)
(3.4.5) is a nonhomogeneous Sturm-Liouville problem [Morse and Feshbach] and can be
solved by finding the eigenfunctions for the impulse response for the homogeneous version
of (3.4.5) then convolving this with the ramped generation function. A solution for the
impulse response is available [Lo], but in the case of the ABG response, it involves a
slowly converging series. Therefore, a numerical solution to (3.4.5) was also developed
for finding gABQ.
The gABG numerical solution was validated in two ways, both involving the
determination of a steady state value by this transient method. The validation testing
required using a true pulse for the simulated optical excitation, rather than a ramp. The first
test simulated the ABG steady state photoelectron distribution for an SI wafer and
compared this to an analytical solution to the one-dimensional continuity equation
[Campbell]. Figure 3.4.2a shows that the carrier distribution for the simulation is to within
about 10% of the calculated steady state distribution for typical SI GaAs (parameters in
Table 3.4.1) over the first 5 Jim of material-the principle region of interest for the ABG
PIMR measurement The simulation takes about a microsecond of simulated time to reach
100
5
Depth, um
a)
Z.6-
2.4<
1 2.2g.
^^o
slope = -0.5
1 *
3 1-8-
1.61.4.25
.5
.75
1
1.25
1.5
1.75
2
2.25
Iog(Normalized Carbon Concentration)
b)
Figure 3.4.2
ABG Numerical Solution Validation Tests
a) Simulated and Calculated Steady State Carrier Distribution
b) Simulated Steady State ABG Response vs. N^
101
2.5
lifetime
200 nsec
electron mobility
6500 cm^/Vsec
surface recombination velocity
5 x 10" cm/sec
optical absorption coefficient (850 nm)
10^* cm"*
optical absorption coefficient (904 nm)
1 cm"*
photon flux rate
10*" cm^/sec
Table 3.4.1
Typical Undoped SI GaAs Parameters for Simulations
102
99% of the steady-state value. The second validation test involved steady state simulations
over a range of lifetimes to see if the simulated G ^ B Q obeyed an inverse square root
dependence on N^. The simulation program also passes this test as (G/^QT
is linear
with N A over a one and a half order of magnitude range (Figure 3.4.2b).
It should be noted that the numerical results for the BBG and ABG peak transient
response are valid for determining electron concentration and photoconductance. The
simulations will only give peak transient PIMR response to within an arbitrary constant
since there are several unquantified linear factors associated with the microwave system,
such as antenna/wafer coupling efficiency and waveguide mismatch.
Transient simulation results of gABG (w**
m e ram
P "Pu^se") f° r a uniform SI wafer
are shown in Figure 3.4.3 as a function of N^. In contrast to the steady state results,
SABG var * es inversely with shallow acceptor concentration, but only exhibits the inverse
square root relationship to N ^ for those lifetimes which are on the order of the 20 nsec
optical excitation. Insufficient time is available for recombination and diffusion where the
lifetime is much larger than the pulse duration, so the response in this regime is dominated
by SRV and generation, and is independent of lifetime and N^ (Figure 3.4.4). Simulations
with respect to variable SRV (Figure 3.4.5) show that a passivated surface will increase
SABG by almost 35% over a surface where SRV is limited by the electron thermal velocity.
A few of the samples used by Wang to confirm the predicted steady state functional
dependence of G ^ B Q on carbon concentration were obtained for cross-calibration of the
peak transient response. Carbon concentration was determined by FTIR [Wang]. The
measured results are shown in Figure 3.4.6a and do not match the general features of the
simulated results, mainly in the region of very low carbon concentration. This may be due
103
2 -1.5 -1 -.5
0
.5
1
1.5
2 2.5
3
3.5
log(Normalized Carbon Concentration)
Figure 3.4.3
Simulated ABG Peak Transient Photoconductance vs N ^
104
70
60.
|i-7000
H»6500
H«6000
50
-
40
2
30
5
H
8 20'
In
's
la-
i
K
-T"
0'
-
2
-
1
0
1
2
3
log(Normalized Carbon Concentration)
Figure 3.4.4
Simulated ABG Peak Transient Photoconductance vs N^ and Electron Mobility
(mobility in units of cnr/Vsec)
105
1.4-
1.35-
1
1-3'
$
1.25-
3
i-2;
<
1.15-
^ ^ ^
V
1 1.1cs
S 1.05.
B
a
\
1-
.95-
3
4
5
log (SRV, cm/sec)
Figure 3.4.5
Normalized Simulated ABG Peak Transient Photoconductance vs SRV
106
8
1
.9
^ H J L M A M M I H H A M H J ^ H ^
Simulated peak transient ABG
>M
7
«-
|.6
« 5J
.3
.2
14.2
14.4
14.6 14.8 15 15.2 15.4
log(Carbon Concentration)
15.6
15.8
16
a)
.9
1
I
1
I
•
1
I
1
I
.8
> .7
as
S .6
Measured ABG PIMR
Simulated peak
transient ABG
E
o .5
CO
<
>
.3
14.2
m^mtmm^^mmmm^^
14.4
14.6 14.8 15 15.2 15.4
log(Carbon Concentration)
15.6
15.8
16
b)
Figure 3.4.6
Measured ABG PIMR and Simulated ABG Peak Transient Photoconductance
vs Carbon Concentration.cm"^
a) Without mobility correction.
b) Withfirstorder mobility correction.
107
to an increase in mobility as ionized impurity scattering will vary monotonically with
carbon concentration over this range [Duncan]. Using (3.2.12), we can get a first order
correction to the simulated results by using a mobility with a carbon concentration
dependence which varies |X from 6000 to its intrinsic value of 8500 cm^/Vsec [Sze] over
this range of carbon concentration. As can be seen in Figure 3.4.6b, the correction to the
simulation is adequate to reproduce the measured data to first order.
A similar set of samples was not available for gBBG' however, the numerical solution
used to arrive at (3.4.4), the relationship between the defect concentrations and gBBG' *s
based on the measured data of Wang.
To relate gBBG'
anc
*> therefore the EL2
concentration, to peak transient BBG PIMR photoresponse, we note that 2.0 x 10* *> cm"^
is at the high end of reported EL2 concentrations and a value of about 3.00 V is at the high
end of peak transient BBG PIMR seen over about 100 unprocessed, undoped SI GaAs
samples. Taking a nominal N ^ of 1 x 1 0 ^ cm-3
an( j
solving (3.2.7) for EL2
concentration gives
Nd
4.2v R R r -3.0 + 0.7NA
22£__
A
(3 4 ?)
V
BBG *s *he P ^ transient BBG PIMR response in volts, N<j is in units of 10*" cm"-* and
N ^ is in units of 10 ^ cm"-*. While this does not provide an exact method of determining
EL2 and a true calibration should be performed, (3.4.7) will serve to quantify relative
changes in EL2 among the samples reported in Chapter 4 and analyzed in Chapter 5.
108
3.4.2 Extension for Undoped SI GaAs with an Ion Implanted Channel
In the photoconductivity model developed in the previous section, the dominant levels
within the GaAs bandgap are the EL2 deep donor and the C shallow acceptor. During
processing other defects are introduced: first, by ion implantation, then by annealing.
Since the processed GaAs wafers have bulk regions which remain SI during the entire
fabrication process, a single model which accommodates both the SI, or starting, and
processed material is desired.
To this end, we begin by considering the charge sum of all of the levels compensated
by EL2, or the net shallow acceptors Na, of which C is the dominant level,
E*E d
where E is the energy of level, Ng is the concentration of that level, and (J> is the charge
state. Since the material remains SI, N a is always positive in the bulk, however the
calculation of N a does include donors shallower than EL2, such as Si incorporated during
growth or \ ^ s from post-implant annealing.
Thus, the phenomenological interpretation of photoconductance for starting material
can also be used with an annealed sample if Na is substituted for N^. As N a increases, for
example, the ionized EL2 concentration increases and gBBG decreases since fewer
electrons are available for photoexcitation at the deep donor. Substituting N a for N ^ also
permits considering sources of disagreement between LVM carbon measurements and
PIMR and could explain some of the discrepancy seen between the two (Figure 3.4.6b).
Once the material is implanted, the ABG photoconductivity depends on the properties
109
of the channel region because of its surface sensitivity. Figure 3.4.7 shows that, to first
order, the majority of the carriers generated by the ABG laser are below the channel in
relatively SI material. Thus, the assumption that EL2 is the dominant defect is still
applicable. Associated with the active layer is an electric field which tends to keep electrons
away from the surface recombination centers (Figure 3.4.8). The primary effect of the
channel on the photoconductance can therefore be approximated as a change in the effective
surface recombination velocity, Seff, seen by carriers in the subchannel brought about by
the surface depletion layer. The photoconductive model developed in 2.2.3 and 3.4.1 still
applies; only the boundary condition and, potentially, the quantities of various defects have
changed. Of course, PIMR involves more than just GaAs photoconductance and the
phenomena discussed in section 3.3.2 will also affect measurements with a channel.
Campbell suggested an analysis similar to that used with solar cell high-low junctions.
Seff is a function of n^, D, a (channel thickness), Np (the doping concentration), and the
original SRV
s
-"s^7Tr
(3A9)
The intrinsic carrier concentration term, nj, in (3.4.9) represents the free carrier
concentration on the low, or bulk, side of the junction, but since the subchannel carrier
concentration at the ABG peak transient is dominated by the photogenerated electrons, we
should use a value of n j on the order of 10 " «m"^. For Np « 10*' cm"*, and a channel
thickness of about 0.2 |im, Seff is very low and surface recombination is not a factor. All
110
ii
i
channel depth
6
1
i
^
.absorption depth @ 904 nm
« 1 cm
§
absorption depth @ 850 nm
2
3
4
depth, microns
Figure 3.4.7
ABG and BBG Laser Absorption Depth Relative to Channel Thickness
111
'surface
«E, .
Vsurfaas E f f e c t s
£
surfaceoe N D X
S
effoe"N"
=^
=
^
Franz
Keldysh Effect
Effective Passivation
Figure 3.4.8
Surface Depletion Region with a Channel
112
other effects remaining the same, a channel should increase g^BG by almost 35% (from
Figure 3.4.5) as far fewer carriers will be lost to recombination at the surface.
But additions may be brought on by processing. If residual implant damage manifests
itself as donor levels (shallower than EL2), for example, than g^BG w°ukl increase since
N a is reduced. In the case of anneal-only processing, the point defect constituency can be
altered by several mechanism, such as EL2 dissociation, point defect
generation/annihilation, or As out-diffusion. The net effect of these processes on gABG
depends on the charge state and concentration associated with each process.
Separate effects in the bulk and surface contribute to the BBG photoconductance in the
presence of an activated channel. Bulk effects are identical to those with anneal-only
material. At the surface, the activated channel attenuates the optical signal reaching the bulk
via FK, as discussed in section 3.3.2. The fraction of the total optical signal which is
eventually incident on the bulk is
J S t = exp(-ocSDLb)
(3.4.10)
inc
where 1 ^ ^ is the optical intensity immediately below the SDL, 1 ^ is the optical intensity
incident on the wafer, otsDL *s m e °P^ca^ absorption coefficient in the SDL, and b is the
SDL width. A typical MESFET channel doped at 10 ^ cm"^ will produce a surface electric
field of about 7 x Mr V/cm, resulting in an ocsDL °^ aD0Ut 5000 cm"* (Figure 2.3.3) or a
5% decrease in bulk optical intensity.
A comparison of BBG PIMR data for
implanted/annealed and anneal-only adjacent GaAs wafers from a single boule will show
113
somewhat less than this 5% attenuation as some of the carriers generated in the SDL by FK
contribute to the total number of photogenerated electrons.
Wafers which have only been implanted, but not annealed, will have lower peak ABG
PIMR than that for SI material since the subchannel electrons see a Seff of about 10^
cm/sec (limited by thermal velocity) from the implant damage. Furthermore, any carriers
generated within the implanted layer recombine immediately (on the PIMR time scale) and
introduce a small deficit in N tota j. BBG photoconductance, already reduced slightly by the
higher Seff, decreases further if a significant SDL exists in much the same way as that for
an annealed channel. Implant damage augments the absorption coefficient [Brierly].
3.4.3 Extraction of Diffused Net Shallow Acceptor Concentration
The interpretation of the PIMR measurement represented by (3.4.7) and Figure
3.4.6, yields values for N a and N^ averaged over a portion of the wafer thickness. The
way in which this average is taken depends on the microwave wavelength, the absorption
depth for the optical source, and the electron diffusion length in the GaAs sample. The
PM, as it stands, is incapable of distinguishing in- or out-diffusing N a and N^ from the N a
and N4 which are stationary.
This is especially true for N a and ABG PIMR because of the propensity of the
GaAs surface to be altered by temperature processing and the measurement's sensitivity to
the surface. For defect diffusion with ABG PIMR, the microwave wavelength is not a
factor since the penetration depth limitations of the ABG PIMR measurement is determined
by the optical absorption depth. The electron diffusion length is already considered to first
order if we utilize the PM (Figure 3.4.6) as an intermediary between the PIMR-measured
114
N a and diffused Na. Therefore, we must only consider the diffusion length relative to the
optical absorption depth.
More specifically, PIMR measurements of a sample with defect diffusion length
greater than the optical absorption depth yields an N a associated with the defect diffusion
process while the N a determined by PIMR for a wafer where the defect diffusion length is
less than the optical absorption depth is dominated by other processes. In this section, we
derive expressions for extracting Arhennius data for the diffused N a from the N a
determined by the PM at these two diffusion length extremes.
The net shallow acceptor concentration determined by the ABG PIMR measurement
can be approximated by averaging the actual net shallow acceptor concentration over the
wafer thickness with a weighting function due to the absorption of the PIMR optical signal:
d
Na(T) = a f Na(x,T) exp(-ax) dx
o
(3.4.11)
where Na(x,T) is the actual net shallow acceptor concentration in the SUT at a depth x after
processing at temperature T, while N a (T) is the net shallow acceptor concentration
>
determined by the PM. N a (T) is actually what has been previously referred to as N a ,
except here the temperature dependence is noted explicitly and the prime,', has been added
to emphasize that the PM gives an amalgamated N a when diffusion is involved.
Na(x,T) itself can be reduced to
Na(x,T) = Nbg + N^x.T)
(3.4.12)
or the sum of a constant, background concentration, Njjg, and in-diffusing, N^x/T), net
115
shallow acceptor populations. N^g is the acceptor population PIMR would measure in the
absence of any N ^ diffusion. N^g is typically a function of annealing temperature, but we
will assume that we know N^g for all T of interest. Assuming that the source of vacancies
at the surface is of infinite extent, we therefore take N^x.T) as
NJx/T) = No(T) e r f c ( - £ - )
(3.4.13)
where N0(T) is the surface vacancy acceptor concentration and L(T) is the diffusion length
of the vacancy during the anneal. The two temperature dependent terms in (3.4.13) can be
related to two activation energies: Ej^j for the surface concentration and EJL for the
diffusion length via its relationship to the diffusivity,
L T
( )= N / D o teX P(w )
Assuming that negligible diffusion occurs relative to an optical absorption depth,
(3.4.11) is approximated by
N
ttt> - Nbg + 2txL(T)N0(T),
2L(T) « a"1
(3.4.14)
This reduces to N^g in the limit of small diffusion length or large optical absorption depth,
as would be expected. At the other extreme, if significant diffusion with respect to an
optical absorption occurs, then we find
N
«TO * Nbg +
N
0CT).
2L
CT> » a"1
(3A15)
or that the diffused acceptors are not spatially confined on the scale of the ABG PIMR.
Arhennius analysis can be used to identify point defect processes responsible for
116
the observed variations of the ABG PIMR data with temperature, as was done with EL2
earlier. Here, however, the activation energy must be interpreted with regard to (3.4.14)
and (3.4.1S). Arhennius analysis of (3.4.14) and (3.4.1S) leads to expressions which are
a function of either E ^ or E L , depending on whether surface concentration or diffusion is
the limiting mechanism in the given temperature regime. If diffusion is small compared to
the depth sensitivity of the ABG PIMR measurement, than the Arhennius slope will go as
El/2 or E j^-the factor of 1/2 coming from the square root relationship between diffusivity
and diffusion length. The regime with limited diffusion will also be accompanied by a
time dependence through the diffusion length. For the case where substantial diffusion
occurs, the Arhennius slope will be determined be EL or E^.
3.4.4 Extended Phenomenological Model Summary
The TLM developed previously has been modified in several respects for use here. A
third energy level has been added to correct for shortcomings observed by Wang in the
BBG interpretation for SI GaAs. Numerical solutions to the transient problems have been
developed to investigate the peak transient response, rather than that for the steady state or
transient decay. Processed, as well as as-grown, undoped SI GaAs can now be considered
by expanding the definition of the acceptor level and correcting the data for various effects
associated with the channel. Those precepts of the TLM retained plus these modifications
comprise the Extended Phenomenological Model (EPM) of PIMR. Although technically
we could include the effects of diffusing N a in the EPM, we defer this and consider it as a
separate issue in Chapter 5.
In order for the EPM to be useful with the PIMR system, we must consider the effects
117
of the various interactions between the equipment and the SUT discussed earlier in this
chapter. The dark patterning phenomena is not a factor so long as a sufficient number of
points on the SUT are measured. The attenuation of PIMR by a conducting channel is
another matter. While the model outlined in section 3.3.2 appears to account for the
attenuation, it must be more thoroughly tested before PIMR data for material with
implanted channels can be quantitatively analyzed.
The channel can also alter the photoconductive aspects of the measurement. However,
an examination of the channel thickness relative to the optical absorption depths shows that
the majority of the photoconductive processes occurs in SI material and, therefore, the
model of SI photoconductance (with EL2 as the dominant recombination site) developed in
3.4.1 applies. The challenge in applying this model lies in quantifying Seff and r s so that
corrections can be made for surface recombination, microwave attenuation, and FK effect.
To summarize the EPM for GaAs material processed through active channel formation:
1. As-grown
Substitute N a for N^, although C is still the dominant shallow level. N a
and N^ are extracted from the peak transient PIMR response by
Figure 3.4.6 and (3.4.7), respectively.
2. Anneal-only
Consider N a rather than just N^. N a andN^ are extracted as in 1). If
in- or out-diffusion of a defect is suspected, the defect may be
identified by activation energy information via ABG PIMR, the EPM,
and (3.4.14) or (3.4.15).
118
3, Implant without annealing;
ABG PIMR is a measure of implant damage and Seff. BBG provides
information on the charge state of this damage via FK.
N a and N<j determined as in 1), except with the PIMR data corrected for
FK and Seff.
4. Implant with annealing;
Both ABG and BBG PIMR must be corrected for attenuation by the
conducting channel using the methods discussed in section 3.3.2.
Additional corrections as specified in 3) are also necessary.
N a and N^ determined as in 1), except with the PIMR data corrected for
sheet resistance, FK and Seff.
119
CHAPTER 4
Experimental Results and Discussion for Processed GaAs
The experimental focus of this study is in-process GaAs from starting material through
active channel formation. The purpose of this chapter is to outline the experimental results
for various processes and process phenomena, and interpret these results using the
extended phenomenological model (EPM). Samples have been divided into five major
groups which lead to a progression of understanding about PIMR and the effects of
processing.
We start by examining the results for a series of adjacent wafers from the same boule
which have been processed to various points in a GaAs MMIC fabrication line. This
provides a validation of the basic premises in the EPM. Some of the specific processes and
phenomena of GaAs IC manufacturing, such as furnace annealing (FA), rapid thermal
annealing (RTA), channeling, and coimplantation, are then investigated in more detail and
discussed with regard to the use of the EPM. At the end of this chapter, these samples are
grouped together and the results summarized to provide an overall perspective of the
experimental effort and form the basis for a discussion in the following chapter on point
defect dynamics.
4.1
Typical Processing Sequence
Prior to definition of the MESFET gate, there are two points in the manufacturing line
at which the properties of the GaAs wafer are desired: prior to any processing (staring
120
material) and after formation of the doped layer(s). Here, we form a defined set of wafers,
the "basic wafer series" or BWS, from these two points as well as after several intermediate
processes. The BWS is comprised of an as-polished sample, a wafer which has been
implanted only, a wafer which has been annealed only, and one which has been both
implanted and furnace annealed.
Five adjacent wafers from boule 9127, a 3 inch ingot, were processed to form a BWS,
with the additional wafer being a second implant/anneal sample. Si was implanted in the
appropriate wafers with identical dose and energy of 4 x 1 0 ^ cm"^ at 135 keV. Assuming
80% activation, a mobility of 5000 cnr/Vs, and an implant profile based on interpolated
SIMS data [Tabatabaie-Alavi], this implant should produce a 580 ohm/square channel with
apeak electron concentration of 2.3 x 10*' cm'^ and a depth of about .25 urn. Annealing
was done in a furnace at about 825 °C for 20 minutes in an arsine environment. Measured
PIMR data for these samples is given in Table 4.1.1.
We begin by comparing the as-grown sample to the anneal-only wafer and see that the
thermal treatment decreases N a and N^. Quantifying these data with the EPM shows that
the N a and N<j are reduced by about 45%. These defect losses are compatible with a
model of EL2 dissociation into its point defect constituents which will be expanded upon in
the next chapter.
The implant-only BBG PIMR is about 92% of the as-grown BBG PIMR, or an 8%
loss in amplitude. As discussed in the development of the EPM, the Franz-Keldysh effect
(FK) can be a cause of this attenuation, with the electric field arising from two possible
sources. If the implanted Si is on Ga lattice sites prior to the anneal, than a surface
121
fiojllfi SK&ffeE
Descrintinn
ABG
mV
mV
9127
9127
059
060
as-polished
570
1670
anneal
650
1230
9127
061
implant
120
1430
9127
062
imp&ann
840
900
9127
063
imp & ann
840
940
Table 4.1.1
PIMR Results for BWSfromBoule 9127
122
depletion layer (SDL) with an associated electric field will induce FK and attenuate PIMR
by less than 5%. Additionally, the implant damage itself will induce internal fields of
unknown magnitude locally about individual defects [Blakemore] and FK will exist over
most of the channel region.
To investigate these hypotheses, one half of a wafer from a different qualified boule
was implanted with Si (4 x 1 0 ^ cap- at 135 keV) while the other half was B implanted
with a profile similar to that for the Si. The Si self-annealing hypothesis should manifest
itself as different BBG peak PIMR transient values between the two halves, while the
implant damage hypothesis requires that both halves be identical. The ABG response
should be relatively independent of implant species and dominated by the damageaugmented Seff.
In the case of the ABG measurement, the B side is 3% greater than the Si (Figure
4.1.1) which could be related to different implant damage, table tilt, or experimental error.
The wedge down the center of the map, parallel to the major flat, corresponds to a region
which was masked from both implants and has a PIMR response similar to as-grown
material. The wedge region, measuring about 600 mV with ABG PIMR, is much less
pronounced in the BBG map.
The B side of the wafer, adjacent to the major flat, exhibits a BBG photoresponse 4%
lower then the undamaged wedge which is due to the combined effects of higher Seff,
infinitesimal lifetime in the damage-region, and measurement error. The Si half of the
sample decreases by 11% from the BBG PIMR response in the wedge region. Attributing
3% of decrease in the Si half to the same factors creating the ABG decrease between the
two implants, leaves a 4% difference in BBG PIMR between the two implants or
123
ABGMap
BBG Map
Figure 4.1.1
PIMR Maps for Unannealed Wafer with Two Half Implants
Left Half is B, right half is Si
Darker regions represent higher PIMR response
124
approximately the same decrease associated with FK for a similar Si implant discussed in
section 3.4.2.
Apparently, the two implants affect a difference in BBG PIMR. A possible
explanation is that the implanted Si resides, to some extent, on Ga lattice sites prior to any
high temperature annealing process. Confirmation of the Ga-site occupancy by the Si was
attempted using Raman-scattering, as was done by Holtz. However, the system available
did not have the equipment necessary to resolve the phonon modes allowed by the presence
ofS^.
Prior to applying the point defect portion of the EPM to the implanted and annealed
samples, the PIMR data must be corrected for the attenuating effect of the activated
channel. The microwave antenna transmission line model (Figure 3.3.6) predicts about a
40% decrease in the PIMR response for a 580 ohm/sq channel while FK from the SDL
attenuates the measurement by another 3%, assuming total recombination of electrons
generated in the SDL. The measured difference between the BBG response for the annealonly and (an average of the values for the two) implanted and annealed samples is 25%.
Reasons for this discrepancy will be discussed shortly.
Based on the ABG PIMR simulation, the ABG response for the implanted and
annealed wafers should increase by 35% (Figure 3.4.5) from the reduction of Seff which
compensates the 40% decrease from the antenna transmission line model and yields a
theoretical decrease of 19%. The measured data, however, shows a 30% increase over the
ABG PIMR for the anneal-only sample suggesting that the two wafers have been annealed
differently or that some alteration in the subchannel defect structure has occurred during the
implant, the furnace anneal, or both.
125
To investigate this further, the 9127-062 implant/FA wafer was etched using the
method outlined in Appendix C and then measured with PIMR. This procedure, performed
repeatedly, provides a depth profile of the effect of the channel~or, more exactly, removing
the channel.
As the channel is etched back to a position just short of the peak doping
concentration, the ABG PIMR response increases rapidly (Figure 4.1.2). The attenuation
of the PIMR response between the ABG maximum of 1350 mV and that found with no
etch is about 40%, the value predicted by the antenna transmission line model.
Furthermore this attenuation factor is relatively independent of optical source. Etching
beyond the peak doping position is accompanied by a sharp decrease of about 37% in the
ABG PIMR measurement which then remains roughly constant for etches beyond 2000 A.
This 37% is higher than the 26% predicted by the EPM for Seff goingfromnearly zero to a
value limited by electron thermal velocity (Figure 3.4.5). The variations in ABG and BBG
PIMR response for etches deeper than 2000 A is about 3.6% of the mean (for these etch
depths) corresponding to 3a of the PIMR repeatability.
The constant subchannel ABG PIMR value is significantly different from the ABG
PIMR for the anneal-only sample and corresponds to a decrease in N a assuming that
carrier recombination occurs via EL2. As 9127-060 and -062 are nearly adjacent and,
therefore, are expected to have identical as-grown properties, we attribute the ABG PIMR
difference to defects related to the implant or differences associated with positioning in the
annealing oven. The former is treated in the next chapter and the latter is discussed in detail
in the next section.
Similarly and as noted earlier, a discrepancy exists between the BBG PIMR of 9127060 and that for 9127-062 after correcting with Figure 3.3.6. The BBG PIMR for 9127126
1IQQi
700
-
4
0
^
i
•
-
1000
-
^
^
•
^
2000
-
F
.
1
-
.
^
^
3000
^
4000
"
"
^
!
5000
Etch Depth, Ang
Figure 4.1.2
ABG and BBG PIMR versus Etch Depth for Wafer 9127-062
127
1
6000
060 would also be expected to be similar to that for 9127-062 after etching the channel,
however, this is not the case. The EL2 concentrations for each prior to annealing are
identical due to their proximity in the boule. Since they differ after annealing, we must
conclude that they did not receive the same anneal. Furthermore, since all three wafers in
this BWS were annealed simultaneously, we must also conclude that the FA oven is
nonuniform in some respect. Perhaps, then, the difference between the BBG response for
the two implanted and annealed wafers is attributed to the anneal and not the implant. This
has ramifications for the ABG PIMR differences among the annealed wafers (after
correcting for the channel) since any EL2 lost during the anneal may manifest itself as a
change in Na. This will be discussed further in the next chapter.
The etching results validate several essential features of the EPM. The first is the
attenuation of PIMR by the conducting channel. Shallow etching of the channel increases
both ABG and BBG nearly equally, implying an attenuation mechanism dominated by the
microwave, rather than the photoconductive, aspects of PIMR. Second, deeper etching
eventually leaves insufficient carriers in the channel to support an SDL which isolates the
photogenerated electrons in the subchannel from the surface recombination sites. The
result is a decrease in ABG PIMR slightly greater than that predicted in Figure 3.4.5,
which may stem from the error in the simulation developed in Chapter 3 (see Figure
3.2.4a). The corresponding BBG PIMR measurement is relatively unchanged after
reaching its peak value (as a function of etch depth) since BBG excitation generates carriers
equally through the cross section of the wafer. Finally, the consistent interpretation of these
two facets of the EPM supports the hypothesis that the dominant PIMR recombination
process occurs through the EL2 deep donor. Furthermore, we do not need to explicitly
128
consider the value of Seff so long as it is sufficiently low with a channel (Figure 3.4.S).
Based on the results for this BWS, the EPM has been validated and appears adequate
to describe the basic trends in the PIMR data versus active channel processing. Annealing
reduces the EL2 and N a concentrations in all the samples processed over temperature
following corrections for Seff, FK, and the interaction between the SUT and PIMR. The
B/Si implant-only wafer reveals a photoconductive difference between an isovalent and an
electrically active impurity which may be attributed to Si on Ga sites before the high
temperature anneal. Wafer-to-wafer variations appear to be induced by the processing
based on comparing the anneal-only wafer to the implanted, annealed, and etched sample.
Causes for these process variations will be discussed in the next section.
4.2
Specific Processes and Processing Phenomena
The experiments in the previous section provide a coarse, or macroscopic, view of
processing effects of GaAs as interpreted by PIMR and the EPM. Over time, individual
steps in any IC manufacturing line can drift for a variety of reasons. Furthermore, desired
improvements in IC performance often require altering a process or tightening process
specifications. This section, therefore, takes a more focused look at a few processes,
processing phenomena, and technologies from the point of view of PIMR's use as a
process development and monitoring tool.
4.2,1 Fwrnace Annealing
An experiment using 10 adjacent wafers from a single MMIC qualified boule was
performed to ascertain whether arsine overpressure variations could improve wafer-towafer uniformity within a single furnace annealing (FA) run. A wafer at the back of the
129
oven (near the gas outlet) is exposed to a different As environment than a wafer near the
front due to distance from the arsine inlet as well as temperature variations through the
oven. Thus, each run used one test wafer in the "upstream" position and a second
"downstream". The wafers were implanted with a double implant schedule of 3.41 x 10 ^
cm2 at 30 keV and 6.05 x 10 12 cm2 at 200 keV to from an N+/N structure with an
expected activated sheet resistance of 180 ohms/sq. Arsine overpressure, in percent
deviation from the standard process, was varied from -10% to +100%. The arsine
comprises a small portion of the gas at the inlet, with hydrogen as the carrier gas. Sheet
resistance was measured by the contactless eddy current technique [Miller] and is repeatable
to less than 1%. The backside and the implanted frontside of each sample, the former
being equivalent to an FA-only sample, were measured with PIMR. Results for the
experiment are shown in Figure 4.2.1.
The backside data provides an opportunity to look at the effect of FA at two points in
the furnace. From the plot of backside BBG versus arsine pressure for up- and
downstream wafers (Figure 4.2.1a), it is evident that a significant and consistent difference
occurs corresponding to an upstream EL2 loss. Similarly, the ABG data (Figure 4.2.1b)
shows that N a is always lower at the downstream position. While the difference in EL2
between upstream and downstream is nearly constant with pressure, the same is roughly
true for Na, except when the pressure is twice its nominal value. From the backside data, it
is evident that the upstream and downstream positions do notreceivethe same anneal.
This is substantiated to some degree by a temperature profile of the oven (Figure
4.2.2). The boat carrying the wafers is six inches long and resides in the "hot zone" of
130
1250
40
60
120
80
% Change Arsine
a. BBG PIMR ys Arsine: Backside - Upstream and Downstream
1125
-
1100
1075
1050
1025
l> 1000.
f
975.
CO 950
^ 925
900.
upstream
875
850
•20
i
0
20
40
•
i
•
60
80
100
% Change Arsine
b. ABG PIMR vs Arsine: Backside - Upstream and Downstream
Figure 4.2.1
PIMR and Sheet Resistance Measurements for FA Experiment
131
120
190
-*_—_^^__^___i__^_i___i.
185
I180
«
170
165-
upstream
160
•20
i
20
40
60
•
i
•
100
80
120
% Change Arsine
c. Sheet Resistance, ohms/sq, vs Arsine: Implant side - Upstream and Downstream
280
260
-
S 220
0* 200
18
I °
S 160-1
1 14 °
ftS 120-|
100
80
•20
20
40
60
80
100
120
% Change Arsine
d. Upstream, Implant Side Sheet Resistance, ohms/sq, and PIMR, mV, vs Arsine
Figure 4.2.1 (con't)
PIMR and Sheet Resistance Measurements for FA Experiment
132
900-1
800
y
'S #C/in
700-
e
jy 600
la
9
g 500.
Hot
Zone
j" 400-1
H
3002001 DO-
—T"
5
10
15
20
25
Position, inches
Figure 4.2.2
Measured Temperature Profile of Furnace Annealing Oven
133
30
during the high temperature portion of the anneal. A 15 to 20 *C gradient can be expected
across the boat even with it perfectly positioned within the hot zone. Temperature
differences of this scale have been associated with variations in point defect structure
[Bindal] of the order necessary to induce the observed differences in PIMR.
Considering the implanted side of the wafers, all of the wafers have sheet resistances
within the process window (Figure 4.2.1c). The upstream implants consistently yield
channels below the specified resistance while those downstream vary around the expected
resistivities. The difference between upstream and downstream sheet resistance narrows
for the higher gas pressures which would seem to suggest more consistent annealing
conditions under these regimes. However, this is not substantiated by the backside data.
In the case of the upstream, implant-side PIMR data, both the ABG and BBG data
track that for the sheet resistance (Figure 4.2. Id). This is expected based on the attenuation
of PIMR discussed in developing the EPM, however, the downstream, implant-side PIMR
data does not track the sheet resistance suggesting point defect structural changes in the
bulk region of the downstream wafers. While the dominant factor in differentiating PIMR
with and without a conductive channel is the attenuation of the microwave signal by the
channel, other factors are at work in differentiating the PIMR response of samples with
small variations in the conductivity of a low sheet resistivity channel. Based on the lack of
agreement between the upstream and downstream results and the measured temperature
gradient in the furnace, the variations in channel conductivity and PIMR appear to be
related to the temperature within the oven and, perhaps as well, the positioning of the boat.
To summarize, the PIMR data shows that arsine overpressure is not the dominant
factor in determining the observed effects. Backside PIMR indicates a difference in the
134
anneal at the two ends of the furnace which is supported by a temperature profile of the
oven itself. The upstream end of the oven yields annealed wafers with lower N^ and
higher N a than the downstream end. While the EPM can reveal the extent to which
annealing reduces N4 or sheet resistance attenuates PIMR, it can not address the
relationship between N<j loss and annealing process parameters, such as temperature. This
requires considering the point defect dynamics, which will be discussed in Chapter 5.
The difference between the PIMR values for the etched implant/anneal and the annealonly BWS samples can be reviewed in light of the furnace temperature gradient. Under the
worst case where these two BWS samples were annealed at opposite ends of the oven with
the nominal arsine overpressure, we would expect a difference between the two of less than
10% (based on Figure 4.2.1b). This would more than account for the variations seen in
the BBG data. However, the difference for the ABG is more like 20% and we therefore
conclude that the majority of these ABG differences are related to unannealed implant
damage or effects related to activating the channel.
4.2.2 Rapid Thermal Annealing
Experiments with RTA can be used to study both anneal-only and channel formation
processes without the need to vary arsine overpressure. However, RTA does require
consideration of a capping technique. Here, three RTA experiments for annealing alone and
a fourth with an implanted channel were performed using designed experiments. A
screening experiment was carried out on a fifth sample set using susceptor RTA.
For each experiment, anneal temperature, anneal time, and quench rate were varied
identically. A D-optimal experiment with cubic curve fitting [Mason] was chosen for the
135
design as this provides maximum sensitivity to parametric coupling and variability with a
minimum experimental set. The temperature was varied from 550 to 1050 *C so that
various annealing phenomena discussed in Chapter 2 might be observed. The annealing
time and quench rate were varied from 5 to 30 sec and 10 to 75 *C/sec so that information
could be gathered over a relatively wide range of operating conditions. The ramp rate for
heating the RTA to the anneal temperature was fixed at 50 "C/sec. Each experiment called
for two adjacent wafers diced into 16 pieces, for a total of 32 runs per experiment. All of
the wafers used here came from boules qualified for MMIC processing. Backside testing
was not possible since the samples were scribed on this side for tracking and identification.
Four adjacent wafers from a 3 inch boule chosen for the silox-capped and uncapped
experiments were measured with PIMR prior to any processing. The average as-grown
BBG and ABG values of for these samples were 2000 and 440 mV, respectively. CVD
silox was deposited and stripped per the procedures outlined in Appendix C. After
completion of the RTA run, described in Appendix D, the pieces from each wafer were fit
back together on the PIMR stage top and mapped The average PIMR response across each
diced wafer piece was determined and the data for the entire experiment was fit to the
process variables using the experimental design software.
Results for the uncapped experiment are shown in Figure 4.2.3 as a surface plot of
ABG (solid line) and BBG (dashed line) PIMR versus anneal time and temperature. These
results are relatively independent of quench rate. Temperature is the dominant effect for
both PIMR measurements, with the region of maximum process stability around 870 *C for
the ABG and 650 °C for the BBG. In the case of the BBG PIMR, there is a slight
dependence on anneal time which favors longer anneals for greater stability.
136
30'
H-
*
s.
^
i
-./
28--
/
/
/
/
/
/
'
/
/
/
/
26'
A
N
N
2+
22"
i
i
£ 20
A
L
13--
-16-
T
114-
M
E
J%
/
12-•
\
\
\
\
\
400 450 500 550 600 650 700 75^ \B00
850
850 W75i)70O6C
\
\
•
/
/
\
n
\
\
\
\
\
\
\
\
\
\
\
\ \ \ \
\ \ \
i
\
\ \
\ ' \
\
\
\ } n iLL_1_\
\J
l\
I
l\
560 580 600 620 640 660 680 700 720 740 760 760 BOO 820 840 860 880 900 920 940 960 980 100010201040
TEMPERATURE
ABG
circled
Figure 4.2.3
Uncapped RTA PIMR Results
Surface plot of ABG and BBG PIMR versus anneal time in seconds and temperature in °C
137
Interpreting these results with the EPM, we find that the EL2 concentration decreases
for higher temperature anneals by as much as half of its original, as grown value. N a also
decreases with increasing temperature until reaching a stable plateau at 70% of the asgrown value around 870 *C. These results are compatible with EL2 dissociation in the
bulk and As-loss at the surface below 950 *C, both of which will be discussed in greater
detail in Chapter 5. Both N4 and N a decrease rapidly for temperatures above 950 *C.
Figure 4.2.4 shows the results for the silox-capped samples in the same format used
with the uncapped experiment. The BBG temperature features found with the uncapped
test are also found with the silox, except that the magnitude of the silox-capped BBG
response is cut in half and the sensitivity to anneal time is roughly doubled. The ABG
PIMR is again only a function of temperature, but the magnitude of the response is reduced
by a factor of 2 and the plateau region has shifted to a lower temperature which roughly
coincides with that for the BBG response.
The factor of 2 loss in silox-capped PIMR response found with both lasers was
investigated further. Using differential IRT scans over wavelengths from 1.0 to 1.8 Jim,
the signal transmitted through pieces from the silox annealed wafer was also consistently
lower than that for an as-grown sample by roughly a factor of 2. Visual inspection of the
samples revealed a "halo" about the edge of the samples (Figure 4.2.5) which is compatible
with the formation of a polymer layer from the byproducts of the silox REE strip [Wolf].
Doubling the silox-capped BBG PIMR to account for this polymer yields values
which are almost identical to that for the uncapped anneal, except for the doubling in the
anneal time dependence. This is consistent with stress and plastic deformation generating
EL2 [Vignaud]. Stress relief in these silox films occurs at temperatures above 700 *C
138
'>1"1'HI"1 I l"l F 1
lENPERATURE
A66
circled
Figure 4.2.4
Silox-Capped RTA PIMR Results
Surface plot of ABG and BBG PIMR versus anneal time in seconds and temperature in °C
139
a)
b)
c)
Figure 4.2.5
Diced GaAs Samples Photographed with Standard 35mm Flash Lamp
a) As-grown; b) 1050 °C uncapped RTA; c) 550 °C silox-capped RTA
140
[Blauuw], The film stress in the lower temperature anneals generates EL2 and counteracts
the deep donor loss associated with annealing. Once the film stress is relieved, at annealing
temperatures above 700 'C, the time-dependent effect is less of a factor.
Similarly correcting the silox-capped ABG PIMR for the reflective coating yields N a
values which increase for temperatures above 700 *C. This is compatible with Ga loss, or
V Q S acceptor generation which increases N a and decreases the ABG PIMR response. The
decrease in ABG PIMR below 600 *C is most likely due to some loss of As during the
silox deposition process as the wafer spends some time stabilizing at the 425 °C deposition
temperature without a protective cap or arsine overpressure.
PL scans of the silox and uncapped samples where also taken (Figure 4.2.6). Only
transitions germane to undoped SI GaAs are seen in the scan for the the uncapped RTA
piece. An excitonic donor-acceptor transition can be found at 1.492 eV with the siloxcapped sample. This level becomes stronger as the anneal temperature increases which
implies that Si from the cap is diffusing into the GaAs and taking up donor positions.
Since the donor state for Si in GaAs is Si@a, VQ a creation would be a highly likely
precursor. Furthermore, incursion of Si into the GaAs at a level sufficient to produce a
channel of any significance is highly unlikely because of limited Si diffusion, V Q & surface
concentration, and VQ & diffusion. The latter two will be investigated in greater detail in the
next chapter. The total absence of a time dependence in the ABG PIMR data shows that the
SiQa concentration is limited to a very thin layer at the surface.
The nitride-capped experiment was done in the same manner as the silox and uncapped
RTA experiments, but with a different RTA oven. Silicon nitride was deposited on and
stripped from (Appendix C) four adjacent wafers from a different 3 inch, MMIC qualified
141
1.0 -I
0.8 -
Uncapped
0.6 -
0.4 -
0.2 -
0.0 8273
1.499
824Q
1.503
8305
1.493
URUELENCTH ( f l n g r t o m s ) ->
<- ENERCY < t U )
8337
1.487
8378
1.481
1.4939
1
1.0 -
Silox Capped
/
\
1.4916
0.8 -
-
r
\
.4989
0.6 -
0.4 -
0.2 -
0.0 -
1
8240
1.505
1
I
8273
1.499
I
1
1
8305
1.493
URUELENCTH (Rngttoas) ->
<- ENERGY («U>
i
8337
1.487
i
i
8370
1.481
Figure 4.2.6
Photoluminescence Scans of Uncapped and Silox-Capped RTA GaAs Samples
(Vertical scale is Normalized PL Intensity in arbitrary units)
142
boule than that used for the previous two experiments. Prior to nitride deposition, two
wafers were given Si implants with a target sheet resistance of 620 ohms/sq.
As was the case with the silox-capped experiment, stress was also a factor with the
nitride-capped samples. All of the samples annealed above 750 *C with the nitride cap
clearly exhibited slip lines along <110> and<100>. This is due to the stress of the nitride
film on the GaAs and is exasperated by the speed of the thermal cycle and the mismatch in
thermal expansion coefficient. The longer anneal times and higher temperatures produced
more slip, while the implanted samples tended to have fewer slip lines.
Figure 4.2.7 shows the BBG and ABG PIMR results for the nitride-capped RTA
without an implant. The BBG PIMR response is still dominated by temperature and
steadily decreases from an as-grown value of 2200 mV to 400 mV for a 1050 °C anneal.
Using the EPM, these data correspond to a loss of about 80% of the EL2 over the entire
temperature range. The plateau regime for BBG PIMR is again found at about 660 °C.
In contrast to the ABG data for the uncapped and silox-capped wafers, the ABG data
for nitride-capped annealing depends on both temperature and anneal time. ABG PIMR
appears to be a more complicated function of the process variables; perhaps related to the
excessive slip inducing some variation into the surface-sensitive ABG PIMR. The stress in
the nitride film may be related to any one of a number of factors such as film thickness or
hydrogen content or excessive thermal gradients during deposition or annealing.
The implanted, nitride-capped RTA samples (Figure 4.2.8) appear to exhibit roughly
the same BBG PIMR trend as the unimplanted samples, except here the response begins to
decrease near 650 *C rather than 750 to 800 *C. This is attributed to improvements in the
channel resistivity which attenuate the microwave signal. The ABG PIMR signal steadily
143
560 580 600 620 640 660 680 700 720 740 760 780 800 820 840 860 880 900 920 940 960 980 100010201040
TEMPERATURE
ABG
— — BBG circled
Figure 4.2.7
Nitride-Capped RTA PIMR Results
Surface plot of ABG and BBG PIMR versus anneal time in seconds and temperature in *C
144
•iHll I
t+
\
28"
•H-
\
\
26"
\
\
\
A*'
\
N 22t
!
N
\
l
1 ,
[ 20'
I
A
L
18"
-16"
T
DO
7
114-
H
E
500 700
12"
10"
20C
pm
500
aoo HOC
JiCC
BC0 1CG0
9Cp
.000 800
J50C
W
400 300
@
£oo(40(j)"
B"
1
••
6"
4
560 580 600 620 640 660 680 700 720 740 760 780 800 820
TEMPERATURE
880 800 920 940 960
000 1020 1040
ABG
BBG circled
Figure 4.2.8
Implanted and Nitride-Capped RTA Results
.
Surface plot of ABG and BBG PIMR versus anneal time in seconds and temperature in C
145
increases with temperature from 100 mV before the anneal to 1100 mV after an 800 *C
anneal as the implant damage is reduced. Above this temperature, a combination of effects
related to both the annealing of the bulk and the activation of the channel cause the ABG
PIMR to decrease. This will be revisited in Chapter 5.
Temperature is the dominant factor with the implanted samples but slip also contributes
to the results. In general, slip is related to nitride film thickness and quality, as well as the
size and shape of the particular wafer piece. Yet, stable, repeatable RTA processes have
been developed with nitride capping, such as a two step anneal [Bindal], This process was
used here to anneal two additional 3 inch wafers with the 620 ohm/sq implant, but the
warpage was too great to permit the use of PIMR. These wafers did, however, warp to a
lesser extent than wafers similarly annealed but with a single step RTA process.
Susceptor annealing [Kazior] is much more attractive than capped anneals for just this
reason. In the final RTA experiment, 8 adjacent wafers from a single boule were implanted
with Si for a target resistance of 620 ohms/sq. Annealing took place in an As-charged
susceptor at 850 or 1050 *C, for 2 or 20 seconds, and a 1 or 6 °C/sec quench. Sheet
resistance was measured by the eddy current method.
The dominant effect was again temperature (Table 4.2.1). The 850 °C anneals
produced sheet resistivities around 800 ohm/sq while the 1050 °C anneal varied from 560
to nearly 1400 ohm/sq. Eliminating the two extreme sheet resistances, the backside BBG
correlated fairly well (R=.8) to implanted side BBG, as was the case with the FA
experiment. Lower backside ABG and BBG are both characteristic of the 1050 °C anneal,
similar to what was seen with the other RTA experiments.
Based on the EPM, the frontside ABG PIMR is a function of the channel sheet
146
A BG
(rr»V)
front back
BBG
Resistance
(mV)
front back
(ohm/sq)
Processing
Temp Time Ouench
779
372
966
912
562
H
S
S
941
407
1090 947
714
H
L
F
505
259
1086 544
715
H
L
S
686
626
935 1861
758
L
L
F
330
711
604 1846
775
L
S
F
406
495
626 1585
776
L
L
S
725
799
1078 874
847
L
S
S
608
269
1056 571
1376
H
S
F
....
900
....
._
..
._
2000
Legend
Temperature:
H=1050°C L=850eC
Time:
L=20 sec
Quench
F=6'C/sec S=l'C/sec
Table 4.2.1
Susceptor RTA Measured Results
147
S=2 sec
resistance and the subchannel Na. We see that 850 *C anneal sheet resistance extremes do
not correspond to the 850 *C anneal frontside PIMR extremes, as would be the case if the
attenuation of the microwave signal by the conducting channel was the only effect
underlying the variations in ABG PIMR with annealing conditions. This general lack of
correlation between frontside ABG PIMR and sheet resistance implies that changes in the
subchannel N a contribute to the ABG PIMR variations with annealing conditions, as was
seen in the BWS and FA samples. The 850 *C anneals tend to have lower frontside ABG
responses than their 1050 °C counterparts, which corresponds to more subchannel N a
generated at 850 °C.
The sheet resistance data suggests that acceptable activation may be achieved by
annealing at a temperature and for a time intermediate to those used here, such as 950 °C
for 10 sec. This is compatible with RTA processes used elsewhere [Davies]. Furthermore,
based on an approximate interpolation of this data, Davies' RTA schedule would yield
PIMR responses which are similar to those for wafers annealed by FA: wafers which meet
their sheet resistance specifications.
IRT scans of the susceptor anneal wafers were also performed and the measurement
used to determine EL2 concentration [Wang]. All of the annealed samples had EL2
concentrations about 1/3 to 1/2 below that for the unannealed wafers, as would be
expected. No apparent trend was evident in these data, which may be as much from the
lack of a detailed sample pool as from the error introduced by the IRT equipment.
Furthermore, limitations in the IRT system's ability to handle small and odd-sized samples
precluded obtaining a similar set of data for the capped RTA experiments.
To summarize, RTA temperature and capping technique are the dominant effects on
148
PIMR, with anneal time as a secondary parameter. The designed experiments show that
independent of capping technique, BBG PIMR decreases as the temperature increases and
is most stable, or least dependent on temperature, for anneals below 700 *C. ABG PIMR
results for the silox, nitride, and uncapped RTA experiments all initially increase, reach
some maximum, and than decrease as a function of temperature. The stable process region
for the ABG PIMR measurements are always in the neighborhood of the maximum ABG
PIMR response and occur at a lower temperature for the capped methods than for the
uncapped RTA. Stress effects were also noted and in some cases were so severe as to warp
a wafer beyond the point acceptable for continued processing.
Using the EPM, the RTA data can be interpreted with respect to N a and N<j. The
silox-capped and uncapped RTA produce VQ & and V^s, respectively based on the ABG
response.
An EL2 loss with increasing temperature is indicated by the BBG
measurements. The processes producing these point defect changes will be studied in
detail in Chapter 5.
The greatest variability in the PIMR measurements was found with the implanted
samples. The implanted/nitride-capped samples reveal the temperature dependence of the
activation process via the progressive attenuation of the BBG PIMR response.
Although not performed in as much depth as the other RTA experiments, the susceptor
experiment showed that reasonably stable sheet resistances can be obtained without
warping, or other stress effects, while maintaining PIMR responses similar to those found
with successfully processed FA wafers. Based on these results, susceptor annealing is a
viable alternative to its RTA counterparts as well as FA.
149
4.2.3 Ion Implantation Channeling
Channeling of implanted ions occurs with all implants and tends to reduce the peak
doping concentration and shift the channel to greater depths. The effect can be exacerbated
or exploited by poorly chosen wafer orientation with respect to the ion beam. The latter is
the case in this experiment.
A dozen, adjacent, 2 inch diameter wafers from each of two boules were selected and
divided into four adjacent groups of three wafers per boule. Each group was implanted
with Si for a target sheet resistance of 580 ohm/sq, at a different rotation angle with respect
to the implanted beam. The tilt angle was held constant at 7°. Prior to PIMR
measurements, sheet resistance was determined by the contactless eddy current method. CV curves were used to profile the activated channel.
The two ingots used here were grown in the mid-1980's and are therefore expected to
have somewhat higher carbon concentration than the material reported earlier in this
chapter. The polish also differs somewhat from that used more recently, but it is of
sufficient quality so as not to be a factor with PIMR. Overall, the material was good
enough in its day to be qualified for MMIC processing.
The sheet resistance and C- V data show that significant channeling only occurred for
the 0* degree rotation while the 15°, 30°, and 45° rotations yielded about the same results
(Table 4.2.2). For the channeled samples, sheet resistance was reduced to about 530
ohm/sq. Channeled peak electron concentration decreased by 15% while its location was
15% deeper. The channeled profile is also broader since the electron concentration one
decade below the peak is 15% deeper into the wafer than the unchanneled profiles.
A subset of the two dozen wafers, consisting of all the channeled wafers and seven
150
Rotation
Sheet
Angle Resistance
(deg)
(ohms/sq)
0
530
15
Peak
C-V Results
Depth Depth/10
(1017cm"3)
(|im)
(\xm)
1.5
0.15
0.35
570
1.8
0.12
0.23
30
580
2.0
0.13
0.24
45
575
1.9
0.12
0.23
Legend
Peak:
Peak electron concentration
Depth:
Depth of peak electron concentration
Depth/10: Depth where electron concentration is .1 of that at the peak
Table 4.2.2
Summarized C-V Results for Channeling Experiment
151
unchanneled wafers, was chosen for PIMR testing based on the sheet resistance and C-V
results. The measured PIMR data is corrected for the effect of sheet resistance by using
Figure 3.3.6 to yield "corrected" PIMR data. The BBG PIMR (Figure 4.2.9) seems to be
an indicator of the boule from which the samples originate, however another possibility is
that the boules were annealed at different times and the corrected BBG PIMR variation
across the boules represents a process variation. This is unlikely based on the FA
experiment where repeated use of the furnace apparently introduces variations of less than
10% rather than the nearly 20% seen here. The corrected BBG PIMR is independent of the
sheet resistance and therefore confirms the attenuation correction technique.
On the other hand, the corrected ABG PIMR (Figure 4.2.10) is quite different with a
strong sheet resistance relationship remaining for the channeled samples. These data
represent differing concentrations of N a in the subchannel after the implantation and
annealing processes and as such may be an indicator of aberrant, time-dependent MESFET
behavior. Except for the unchanneled wafer with 30° rotation and a corrected ABG PIMR
below 700 mV, the corrected ABG PIMR data could be used as an indicator of a poor
channel for this implant if values much different than 760 mV are obtained. In a fabrication
facility, wafers with corrected ABG PIMR values falling outside this PIMR acceptance
region would warrant further testing before continuing, or scrapping the wafer altogether.
These samples provide an other opportunity to investigate the phenomena found with
the BWS etching. In Figure 4.2.1 la, we see that the ABG responses for channeled and
unchanneled wafers from boule 7027 corroborate the tabular C-V data. The ABG PIMR
versus depth for the channeled wafer increases much more gradually as it approaches a
lower peak electron concentration at a greater depth. In contrast, the unchanneled ABG
152
Boule9021
0
S 34°
330
3201
310
520
D
530
Boule7027
540
550
560
570
Sheet Resistance, ohms/sq
580
590
600
540
550
560
570
580
Sheet Resistance, ohms/sq
590
600
a)
680
•
i
i
i
-
| 660]
g 640
S 620
Boule 9021
CD
§ 600J
Boule 7027
"2
| 5801
540
520
• •
•
| 5601
530
b)
Figure 4.2.9
BBG PIMR Response versus Sheet Resistance for the Channeling Experiment
a)Uncorrected b) Corrected for channel conductivity
Correction for sheet conductivity effect on PIMR microwave signal using antenna
transmission line model.
153
900-1
•
•
•
•
1
B
800-
PIMR Acceptance
Region
CO
Odeg (Channeled)
15 deg
30 deg
45 deg
•
B
i
+
<
•8
B
B
700-
^^Poor RF performance
possible
e
u
B
600
20
540
560
580
i
600
Sheet Resistance, ohms/sq
Figure 4.2.10
Corrected ABG PIMR Response versus Sheet Resistance for the Channeling Experiment
Correction for sheet conductivity effect on PIMR microwave signal using antenna
transmission line model. Experiment demonstrates that channeled samples can be
identified by low sheet resistance or corrected ABG PIMR outside the "acceptance
window". One unchanneled sample is found outside this window and may be indicative of
poor or discrepant RF MESFET performance due to trapping effects.
154
1100
1000
channeled
900'
E
CO
<
8 0
*
700
600
500'
400
•500
i
1000
1500
2000
2500
•
i
3000
3500
Etch Depth, A
a.
6U0-
^D^,^
550-
^y^
unchanneled
^^'^'O—^^
,-Tl
50C
450-
/
^r
channeled
g400M 350.
CO
300250.
200-
-500
500
1000
1500
2000
2500
3000
Etch Depth, A
b.
Figure 4.2.11
Channeling Experiment Etching Results firom Boule 7027
a) ABG PIMR and b) BBG PIMR
155
3500
PIMR peaks quickly and more sharply. The BBG PIMR (Figure 4.2.11b) increases
steadily for both samples until the peak electron concentration is reached.
The etch results for both wafers are similar to those for the etched BWS sample and
are compatible with the EPM. Since the channeled implant is more broadly distributed than
its unchanneled counterpart, the sheet resistance increases more gradually with etching,
thereby inducing a slower increase in the PIMR response versus depth via the attenuation
of the microwave signal.
In Figure 4.2.12, we plot the interpolated BBG PIMR responses for the two etched
wafers using data from etch depths less than that where the ABG PIMR response begins to
decrease. This criteria was chosen since it corresponds to the regime over which the
attenuation of the microwave signal by the channel is the dominant factor in the etching
data. Etch depth has been converted to sheet resistance based on calculations using the C-V
data and the interpolated BBG PIMR response was than normalized to the value predicted
by the transmission line model (Figure 3.3.6) at the maximum calculated sheet resistance,
2000 ohms/sq. In comparison, the transmission line model with the 5% adjustment for the
Franz-Keldysh effect for all but the 2000 ohms/sq data point is in fairly good agreement
with the measurement-based results.
These same sheet resistance calculations were used to quantify the channel
characteristics at the etch depth where the ABG PIMR response reaches a maximum.
Maximum ABG PIMR response was found to occur when the electron concentration at the
surface dropped below 9 x 10 ^ cm"-*. This was not only the case for both the channeled
and unchanneled samples etched in this experiment, but was also found to be true for the
the etched BWS sample. Although we would assume some dependence on the details of
156
.95
.9
9
.85H
.8
.754
.7
C6
.65
Modeled
.6
Measured
.55
.5
.45
400 600 800 1000 1200 1400 1600 1800 2000 2200
Calculated Sheet Resistance, ohms/sq
•
Modeled
O
Measured
Figure 4.2.12
Modeled and Measurement-based PIMR vs Calculated Sheet Resistance
The sheet resistance is calculated as a function of etch depth for the two implants in Figure
4.2.1 lb using the C-V data of Table 4.2.2. The modeled data is from Figure 3.3.6 with a
5% correction for the Franz-Keldysh effect for all but the largest sheet resistance. The
measurement-based data is taken from the BBG PIMR measurements in Figure 4.2.11b
and normalized to the BBG maximum for the corresponding wafer. This normalized BBG
data for the two wafers is then combined into a single data set and interpolated as a function
of calculated sheet resistance relative to the value predicted by the modeled data at 2000
ohms/sq.
157
the channel profile, carrier concentrations at the surface less than 9 x 10*6 cm'* are
indicative of a (remaining) channel layer too resistive to produce the SDL electric field
sufficient to isolate the subchannel photo-electrons from the surface recombination sites.
In the FA and RTA experiments, PIMR and the EPM revealed material variations
induced by changes in processing. The results for the channeling experiments demonstrate
that processing phenomena can also be investigated with PIMR. The correspondence here
between a ABG PIMR value of 760 mV (after correcting for the microwave attenuation by
the conducting layer) and an acceptable channel suggests that subchannel N a evolution is
subject to implant, as well as annealing, phenomena. The corrected BBG PIMR data
shows that the starting material properties can affect measurable differences in PIMR after
active channel formation. The channeling data has also proven useful in experimentally
quantifying two aspects of the EPM: microwave attenuation and reduced Seff.
4.2.4 Coimplantation
Processes using Be coimplantation to form a deep, buried p-type layer have been
gaining in popularity [Canfield, Sadler]. The addition of the p-type layer below the channel
tends to sharpen the channel profile as well as provide a buffer layer between the channel
and the substrate. The extra deep, energetic implant also increases the total damage during
processing and alters the channel and subchannel point defect structure by the activation
and annealing of a second extrinsic defect. The evaluation of such MESFET designs
discussed in the literature make no attempt to differentiate the effects of an activated p-type,
subchannel layer from those due to an alternate point defect constituency brought about by
the additional implant.
158
In an attempt to differentiate the effects of the p-layer from the subchannel defects,
samples with either Be or B were prepared. Wafers were first implanted with the same
silicon implant schedule used with the FA experiment and then implanted with several
different B or Be coimplants per wafer. The B coimplants introduce implant damage, as
seen with the BWS experiment, and activation reaction products similar to those for the Be
[Heimlich], but without the formation of a p-type layer. The coimplantation was done in
quadrants so that each wafer had a silicon-only region and low, moderate, and high dose
coimplant regions. In this way, all of the wafers could be cross-calibrated by normalization
to the silicon-only regions. Following implantation, annealing was done using FA. 2D
PIMR maps were taken with the results averaged over each quadrant. The contactless eddy
current technique was used to measure to sheet resistance in each quadrant.
As revealed in the FA experiment, the furnace itself can introduce a large degree of
variability between wafers at the back and front ends of the oven. The B and Be
coimplanted wafers were placed at opposite ends of the annealing boat with the unfortunate
consequence of introducing a larger than tolerable temperature gradient across the Be
coimplanted wafers. These wafers were therefore removed from the study.
The remaining wafers were B coimplanted prior to annealing with implant schedules
chosen to place doses ranging from 5 to 200 x lO1^ cm"^ within the channel itself (120
keV) or below the channel (300 keV). Each coimplant energy was used with three adjacent
wafers for a total of eight different boron doses and one repeated dose, each.
For the Si-only quadrants of the six wafers, the sheet resistance was 183 ± 2 ohms/sq.
Sheet resistance for the coimplanted regions is shown in Figure 4.2.13a. The peaked sheet
resistance for the low dose, low energy B coimplant has been seen previously
159
120 keV
i
•
75
i
•
100
i
•
i
125
150
12
2
Dose, 10 cnT
a.
•
•
i
i
i
•
i
i
•
i
— - » — t — « —
175
i
200
225
200
225
-
120 keV
300 keV
50
75
100 125 150
Dose, 10 1 2 cm" 2
175
b.
Figure 4.2.13
B Coimplantation Measurements as a Function of Coimplant Dose
a) Sheet Resistance for High and Low Energy B Coimplants
b) Normalized and Corrected ABG PIMR for High and Low Energy B Coimplants
160
[Heimlich] and attributed to the B inhibiting Si activation by mechanisms similar to those
outlined in section 2.5. The earlier study, however, involved a very limited selection of
doses. A similar reduction in activation occurs, but to a lesser extent, with the same doses
for the high energy. In general, the coimplant energy effects sheet resistance for the lower
B doses but has no influence on the higher B doses.
The data here show somewhat different results. The measured BBG PIMR data
showed a strong agreement with sheet resistance, but after correcting for the conducting
layer attenuation, no distinct trend with dose or energy was evident. The lack of a
correspondence to sheet resistance is expected based on the EPM since the BBG PIMR
measurement should be independent of the surface after correcting for the attenuation by the
conducting layer and assuming that the wafers received the same anneal. The normalized
ABG PIMR, after correcting for the sheet resistance, is highly dependent on both dose and
energy (Figure 4.2.13b). The large increase and subsequent decrease with low doses for
the shallow implant, seen with the sheet resistance, is repeated in the ABG PIMR
measurement. Suggestions of a similar feature in the high energy implant are muted, but a
local maximum appears at a dose of 2 x 10* * cm"~ A small energy dependence exists past
the low dose region until a dose of 2 x 10^ cm"~
While the differences in ABG PIMR among all of the B coimplant data here are small
compared to the much larger changes seen with the previous experiments, they are,
nonetheless, expected for several reasons. Such doses should, at the very least, introduce
additional trapping or time-dependent phenomena in the surface region due to the extra
implant damage. But the rather large deviation from the expected sheet resistance with low
161
dose B coimplantation is both interesting and unexpected because of the isovalent nature of
B. This sheet resistance and its similarity to the PIMR results suggest that both the B and
Si are involved
An argument could be made for insufficient correction of the effect of the sheet
resistance on the measured ABG PIMR. However, this is not supported by the corrected
BBG PIMR which shows no correlation to sheetresistanceafter the correction procedure.
PIMR results such as these were previously attributed to B passivating subchannel
defects [Heimlich]. Larger doses of B reduce the defect population, thereby increasing the
PIMR response, but only insofar as the introduction of the B itself did not greatly enhance
the total implant damage. This model predicts a continually decreasing PIMR response
with higher B doses, which is not seen in the data here. Mechanisms for the PIMR and
sheet resistance coimplantation results will be revisited in section 5.3.3.
4.3 Summary of Experimental Results
A large amount of data has been presented in the previous two sections. These
experiments show that PIMR and the EPM can be used to nondestructively characterize and
meaningfully interpret a variety of samples, processes, and process phenomena. Here, the
data is split into annealing and ion implantation so as to provide a concise summary with
respect to the EPM as well as a review in preparation for a discussion of point defect
dynamics beyond the EPM.
4.3.1 PIMR and Annealing
The oven used for FA apparently provides a different anneal to each wafer within the
annealing boat. The difference is independent of arsine pressure and appears to be similar
162
to changes expected from temperature gradients. This was confirmed by a temperature
profile of the oven. Since the completion of this experiment, steps have been taken to
minimize FA lot size and reduce the temperature gradient. In the BWS, FA alone is
associated with a decrease in BBG PEVIR which is attributed to decreased EL2 while ABG
PIMR increases due to lower Na.
RTA was used to investigate PIMR over a wide range of temperature. Above 700 *C,
BBG PIMR steadily decreases, which is again attributed to EL2 loss. Nitride capping can
introduce excessive stress, complicating the design and analysis of such processes.
Uncapped and silox-capped methods produced relative increases and decreases,
respectively, in N a which can be associated with As-loss and Ga-loss,respectively.Above
950 °C, a second mechanism for N a generation takes over as ABG PIMR decreases in this
regime regardless of capping technique.
Most importantly for the model presented in chapters 2 and 3, the similarities between
the RTA and FA results, as well as among the various capping techniques, confirms the
supposition that dominantly observed PIMR recombination mechanism is through EL2.
That is to say that even though there may be changes in the quality of the GaAs near the
surface due to processing, most of the photocarriers are affected by the EL2 level.
4.3.2 PIMR and Ion Implantation
Introduction of a channel layer requires corrections to the PIMR data as outlined in
section 3.3.2. Etching experiments revealed the degree to which these phenomena affect
PIMR. The experimental data showed that once these corrections are performed, the BBG
PIMR measurement for an implanted/annealed wafer are close to that for an anneal-only
163
sample, but some discrepancy remains when FA is used for reasons mentioned above. An
even greater discrepancy was found for the ABG PIMR between an implanted/annealed
wafer with just the channel etched and a nearly adjacent wafer which was only annealed
and seems to be related to effects in the subchannel after channel activation. The implantonly BWS data led to another experiment which demonstrated, via FK and BBG PIMR,
that partial Si activation is achieved prior to any high temperature annealing.
The channel formation process involves damage repair as well as Si activation, and the
FA experiments showed a strong sensitivity to anneal conditions. Sheet resistance varied
directly with the implanted-side PIMR and inversely with backside PIMR. The same,
however, was not true at the back end of the oven. And while the N a and N^ differences,
as measured by PIMR, received at the two ends of the annealing oven are only on the order
of a few percent, they are nonetheless significant since the temperature gradient producing
these differences appears to yield different sheet resistances.
RTA processes one wafer at a time, and so temperature gradients across the oven do
not tend to produce wafer-to-wafer variations. Using a susceptor, it was found that the
most uniform results came from annealing below 1050 °C. Front and backside BBG data
were well correlated, as was the case with the FA experiment, while temperature had the
strongest effect on all PIMR data, similar to what was found in the RTA annealing tests.
Nitride-capped RTA was also used to anneal implanted layers. Warping is a problem with
nitride-capped RTA of entire wafers, but a trend of improved sheet resistance with
increasing anneal temperature is evident in the wafer pieces based on the PIMR attenuation.
The channeling experiment reconfirmed some of the conclusions made with the BWS
and FA experiments. The lack of a correlation between sheet resistance and corrected BBG
164
PIMR for a single boule demonstrated that the model of the antenna/channel interaction is
adequate to correct the PIMR data. The clearly different corrected BBG PIMR responses
for wafers from separate boules demonstrates that the properties of the starting material are
important in determining the material properties after processing. The corrected ABG
PIMR data from the channeling experiment shows the promise of PIMR in a production
MMIC environment as a screen immediately after annealing by revealing an apparent
"acceptable" corrected ABG PIMR value when the channel is within its target sheet
resistance window.
In the etching experiments we see confirmation of several of the tenets of the EPM.
The interpretation of the results with etch deep shows that the carriers resided mainly in the
subchannel and that the carrier dynamics in this region are consistent with processes
dominated by EL2. Furthermore, we do not need to explicitly correct for Seff so long as it
is sufficiently low with a channel present (Figure 3.4.5).
Finally, the B coimplants also demonstrated that we should not expect implant/anneal
and anneal-only wafers to have precisely identical ABG PIMR measurements even after
correcting for the channel. Subtle differences exist in the subchannel due to unannealed
implant damage and point defect reactions which may or may not appear as changes in
sheet resistance. Therefore, p-type coimplants which are merely used to adjust sheet
resistance by controlling the doping at the backside of the channel can also introduce timedependent phenomena which may ultimately affect MESFET performance.
165
CHAPTER 5
Point Defect Dynamical Interpretation
5.1
Introduction
The experiments in the previous chapter demonstrate that PIMR and the EPM can
be useful for investigating GaAs processing phenomena. Differences among PIMR
responses for the measured samples indicate that processing temperature is the most
dominant factor.
Interpreting these measurements with the EPM yields N a and N^
associated with each change in the PIMR data. Such information would allow a process
engineer to identify shortcomings in a piece of equipment, like the furnace, or establish
process windows which lead to acceptable parametric performance, as perhaps in the case
of the susceptor annealing. In this capacity, PIMR and the EPM could contribute to highyield GaAs IC manufacturing technology.
The EPM does not, however, relate process variables, such as temperature,
annealing method or coimplant dose, to the observed point defect concentrations. It does
not bridge the gap between changes in point defect concentrations and the physical
mechanisms inducing these changes.
In contrast to IC manufacturing process
development, semiconductor material characterization focuses on the point defect dynamics
and their relationships to the processing variables.
In this section, the PIMR experimental results are studied from a material
characterization viewpoint. The results from the previous chapter are divided between
166
annealing and ion implantation characterization and are analyzed as such. The analysis and
discussion of the data are first attempted with respect to published models and
corroborating data, with improvements introduced to existing interpretations where
necessary.
S.2 Annealing Phenomena
This section focuses mainly on the results taken from the anneal-only RTA
experiments. The data are first reviewed as a function of the dominant process variables,
temperature and capping technique. Mechanisms for the observed changes in the EL2
concentration are then explored. The silox-capped and uncapped RTA N a appear to involve
the diffusion of defects and are considered using the results of section 3.4.3, while the
nitride-capped data can be regarded as changes in bulk Na.
5.2.1 Annealing Data Review
The BWS was the simplest experiment performed regarding any of the processes.
FA alone exhibits very definite behavior with respect to an as-grown wafer, the BBG PIMR
peak transient response decreases while ABG PIMR increases. Based on the EPM, these
changes correspond to a 45% loss of both N a and N<j but may vary somewhat with
position in the furnace. These losses are similar to that seen for the uncapped RTA samples
at a similar temperature.
The RTA experiments can also be viewed in terms of loss relative to starting
material by normalizing the extracted N^ and N a (± 5% from the experimental fitting) to
those for the as-grown, or preprocessed, samples. This normalization procedure will be
repeated throughout this chapter so that we may compare the results across different
167
boules. Plotting these normalized data reveals that the EL2 concentration decreases by
approximately 45% for an 850 *C (Figure 5.2.1a), as was the case with the BWS. This
loss of EL2~independent of capping technique-is even more interesting since over the 700
to 900 °C, the changes in N<j versus temperature for each of the three capping methods are
nearly identical. Therefore, the same processes are assumed responsible for the decrease in
EL2 over this temperature range for the three capping techniques.
We also conclude that the differences in the EL2 loss below 700 8C for the three
RTA experiments are due to previous annealing history introduced by the capping
deposition processing. The uncapped samples are not exposed to any extra high
temperature processing and have the least EL2 lost, while the silox, processed at 425 °C,
has the highest EL2 lost. The nitride capped samples receive a 250 *C temperature
treatment during the PECVD nitride deposition and lose a quantity of EL2 intermediate to
both the uncapped and silox capped samples. FA samples are expected to exhibit EL2 loss
behavior similar to that for the uncapped RTA since the FA wafers have no extra thermal
steps between the boule and activation anneals.
At the upper end of the annealing temperature range, greater than 900 °C, the
processes affecting changes in N^ are identical for the silox and uncapped samples, but
appear different for the nitride capped samples. Thus, we could infer that the nitride cap is
superior to the other two methods; however, this is not supported by the excessive slip
previously noted in a visual inspection of the nitride-capped samples. Instead, a more
likely cause for the difference is boule-related phenomena since the uncapped and silox
capped samples are cut from adjacent positions of a single boule while the nitride samples
come from a different boule. This is supported by the fact that, for temperatures above 950
168
1 •
R
1«rf •»-
Uncapped
2 .8.
**
|.7-
•
.
§ -6U
Silox
5
^"^§5 ^^^^
J^j^Nt
J
Cd .4-
^-\v&s» Nitride
^^rV*""*^ .
^
^f\
'•§ -2-
"T^^_
^^^jf\
E
/
u .1-
o 0-
*
500
600
700
800
900
1000
1100
1000
1100
Temperature , deg C
a)
10
es
u
B
V
uc
o
U
1
'
eo
Z
•o .1
J
E
u
o
z .01500
Uncapped
•
i
600
•
i
700
•
i
800
•
i
900
Temperature, deg C
b)
Figure 5.2.1
Normalized Defect Concentrations vs. Temperature for Uncapped, Nitride and Silox
Capped RTA
a) EL2 concentration b) N a concentration (without diffusion effects)
PIMR data from Chapter 4 using a 12 sec anneal time.
Normalization is done with respect to an unprocessed, as grown sample.
169
*C, all point defects and point defect complexes are resolved into VQ a and VA S [Lagowski
and Gatos] concentrations determined by the melt composition.
N a (within the first few microns of the wafer surface) changes over the 700 to 900
*C range in the manner expected for each particular capping technique. Uncapped samples
generate V^ s donors near the surface due to arsenic out-diffusion which decreases the N a
measured by PIMR. The flattening of the uncapped N a curve versus temperature may or
may not be real since these values of N a are extrapolations of the cross-calibration
performed with FTIR (section 3.4.1 and Figure 3.4.6). In contrast, samples which are
silox capped exhibit N a greater than the as-grown value as Ga preferentially diffuses into
the silox and leaves VQ a acceptors in the GaAs. The nitride capped samples tend to
maintain. N a at a constant level over most of the moderate annealing temperature region,,
since no loss of either GaAs component is typically experienced with this process.
The N a annealing behavior below 700 *C has two interesting features. First, N a
for all three capping methods is greater than that for the starting material. Second, and in
light of the constant N a for the nitride-capped RTA between 700 and 900 *C, it is evident
that the reactions responsible for the 45% loss in N a seen with both FA and (nitridecapped) RTA are activated well below the 850 "C annealing temperature. The reduction in
N a appears to be initiated at a temperature above that for the nitride deposition (250 *C) and
goes to completion at 600 °C where N a becomes constant with increasing temperature. A
lack of data below 600 °C makes it difficult to identify a specific point defect process or
point defect processes which are responsible, but it is not related to the EL2 lost over this
temperature regime based on the varied EL2 results for the. three, capping techniques..
Furthermore, it is not related to (the lack of sufficient) arsine overpressure with FA since
170
the capped techniques and FA give the same N a loss.
The temperature regime beyond 900 *C introduces more N a in all of the RTA
processes. This trend is initiated in the nitride capped samples at a somewhat lower
temperature than the other two methods, again suggesting that this phenomena originates
with properties of the individual boule rather than the process.
5.2.2 Analysis of EL2 Loss
Taking the behavior of EL2 above 700 *C in Figure 5.2.1a to be associated with its
dissociation [Suezawa] and that above 950 °C to be due to point defects going back into
solution [Lagowski and Gatos], we analyze the region in between these two phenomena to
study the deep donor.
Assuming that EL2 is composed of three, not necessarily unique, elemental point
defects, its dissolution would go as
EL2<-»A+B+C
(5.2.1)
EL2 is the dominant deep level in GaAs and the concentrations of A, B, and C are therefore
mainly determined by the degree to which EL2 disassociates; if n(T) is the EL2
concentration after annealing at temperature T normalized to the as-grown EL2
concentration, than (l-n(T)) is the normalized concentration of A, B, and C. Since the RTA
data was nearly independent of annealing time in this temperature regime, the EL2
dissociation reaction is in equilibrium and the equilibrium constant, ^£L2*
determined by:
_ [A][B][C] _ (l-n(T))3 rET „ ,i
EL2
[EL2]
nW~ [
o1
K
K
171
(
}
can
De
where [EL2Q] is the as-grown deep donor population. Determining the activation energy
for K EL 2 from the annealing data only requires the temperature dependent information
„ (l-n(T))3
K
EL2~
(5
n(T)
*23)
which yields an Arhennius activation energy for the data in Figure 5.2.1a of 1.5 eV (Figure
5.2.2). This is lower than the 2.5 eV obtained for a similar model of EL2 decomposition in
this temperature regime [Suezawa] and is due to the fact that the 2.5 eV value for the
process activation energy was obtained after quenching the wafers from the anneal, thereby
freezing in the thermodynamic state. In comparison, the RTA experiments performed here
did not involve a true quench and a large portion of the dissociated EL2 reforms by the
reverse reaction of (5.2.1). Moreover, since FA produced a normalized EL2 loss similar to
that for the RTA, we can conclude that the EL2 dissociation process is dominated by
processing temperature rather than the anneal times, quench rates, or capping techniques
used here.
The activation energy for the reverse, EL2 dissociative reaction can be derived from
k
E E
K
EL2 = k p " P t - T F ^
{52A)
where k j and E j are the reaction rate constant and activation energy, respectively, for the
forward or EL2 dissociation process, and ^ and E2 are the corresponding values for the
reverse or EL2 formation process. Taking Suezawa's result for the E2 as 2.5 eV, we find
that E J is 4.0 eV. And, whereas EL2 formation is limited by V<-ja diffusion, based on the
2.5 eV activation energy found by Suezawa, the 4.0 eV associated with EL2 dissociation
corresponds to the activation energy for V/^ bulk diffusion [Swaminathan].
172
2i
0
eg
OS
•
measured
4>
-2
•
.4.
06
- • » . •
-6-
•
i^
-8-
modeled
•
10-
•
1?
S
9.5
10
10.5
11
11.5
1/kT, e V 1
12
12.5
Figure 5.2.2
EL2 Reaction Rate vs. 1 AT for Measured and Modeled Data
Measured slope = -1.5 eV. Modeled slope=-4.0 eV.
Measured data, extracted with the EPM, assumes a tri-defect reaction.
Modeled results assume instantaneous quench.
Reaction rate in arbitrary units.
173
13
To summarize, the EL2 lost during RTA was found to be relatively independent of
capping technique but strongly dependent on temperature. Analysis of these results,
assuming a tri-defect EL2, suggests that EL2 dissociation during heating is controlled by
\ ^ s diffusion while the degree to which EL2 reforms during the post-anneal cooling is
determined by the diffusion of VQ & . Since EL2 has been unquestionably related to As@a,
these results are consistent with a tri-defect model for EL2 of VQ a AsQ a \^ s .
5.2.3 Analysis of Net Shallow Acceptor Concentration
Each of the RTA capping techniques produced unique ABG PIMR and N a results.
These data will be analyzed in this section using the previously derived relationships
between diffused net shallow acceptors and that determined by the EPM. The activation
energies extracted from the measured data will be used to identify the mechanisms which
cause the observed changes in Na.
Figure 5.2.1b shows the N a extracted with Figure 3.4.6, after normalization to the
as-grown values, for the three RTA anneal-only experiments with no attempt made to
extract the effects of defect diffusions. The apparent loss in N a with increasing temperature
below 600 °C is attributed to a mechanism common among all of the RTA, and the FA,
methods used here, while the increase in N a above 950 °C is due to a wholesale change in
point defect constituency [Lagowski and Gatos]. We therefore focus on the region between
these two temperature extremes and begin with the nitride capped experiment.
5.2.3a Nitride Capped RTA
An understanding of the changes in N a for the nitride capped RTA experiments
must be compatible with the observed steady decrease in EL2 with annealing temperature as
174
we assume that the nitride cap acts as a barrier to both Ga and As loss and therefore do not
utilize the results of section 3.4.3. N a remains fairly constant from 600 to 850 *C implying
that the decomposition of EL2 does not produce a net change N a . The steady increase in
N a for anneal temperatures above 850 "C (Figure 5.2.1b) has an activation energy of 1.4
eV. This value is in agreement with the 1.5 eV found for EL2 dissociation and reformation
during RTA to suggest that the byproducts of the EL2 decomposition eventually go on to
form shallow acceptors above 850 °C. The similarity in activation energy between the EL2
and shallow acceptor processes further suggests that the availability of the EL2 constituent
point defects is the limiting factor in the production of the N a . This is even more plausible
after reflecting upon the fact that EL2 is the dominant defect in SI LEC GaAs.
A possible scenario for the production of N a from EL2 above 850 °C requires a
two- step process to be compatible with the PIMR/EPM data. The EL2 lost by annealing
below 850 °C becomes V^ S ASQ & donors and V Q S acceptors after cooling, yielding no net
change in N a . Above 850 *C, the YA S ASQ 3 donor must either react to from a charge
neutral configuration or one which is electron consuming. A possibility falling into the
former category requires two steps:
v
AsAsGa"> vAs + As Ga + e "
which then further decomposes by
v
As + As Ga + *'">
As
As + v Ga
It should be noted that V^sAsQa has a binding energy of 1.8 eV [Look] which is close to
the observed 1.4 eV for the nitride capped N a process above 850 "C.
On the basis of the 1.4 eV activation energy, we must consider, as a matter of
175
completeness, a reaction with YAsV@a since the binding energy for this defect is 1.4 eV
and the defect can be formed from the assumed constituency for EL2.
Y\SVQ&
and the
remaining EL2 constituent, As@a, are donors so that the EL2 decomposing reaction below
850 °C must include an additional reaction which maintains both the integrity of V^sVQa
and charge neutrality. Unfortunately, the antisite is a more stable defect than the vacancy
pair and is only destroyed by rapidly quenching from annealing temperatures in excess of
1000 °C [Look]. Therefore, we eliminate Y\s^Ga ^
a react n
i° Patn-
5.2.3b Uncapped RTA
The measured PIMR results for the uncapped RTA, after interpretation with Figure
3.4.6, show that the net shallow acceptor concentration is reduced by the processing over
temperature. As we suspect that the in-diffusion of V^s donors is involved, we will look at
this data as the production of donors shallower than EL2 rather than a loss of Na.
Considering the nitride capped RTA ABG PIMR results as indicative of bulk
phenomena generally due to the annealing process itself (and not the capping technique)
since no apparent mass transfer with annealing oven occurs, we correct the uncapped RTA
shallow donor concentration determined from PIMR/EPM for these bulk effects by
subtracting the normalized nitride capped shallow donor concentration derived from Figure
5.2.1b from that for the uncapped RTA samples. This is equivalent to using (3.4.14) or
(3.4.15) with the nitride capped results as N^g and solving for the temperature-dependent
terms involving NQ(T) and L(T). The resulting corrected uncapped shallow donor
concentration as a function of temperature is shown in Figure 5.2.3.
Two temperature-dependent regimes can be seen in Figure 5.2.3. Below about 750
176
550 600 650 700 750 800 850 900 950 1000 1050 1100
Temperature, deg C
Figure 5.2.3
Corrected and Normalized Shallow Donor Concentration vs. Temperature
for Uncapped RTA
Correction for bulk effects with nitride capped results.
Normalization with respect to as-grown data at each temperature.
177
*C the activation energy is 0.5 eV (Process 1) while above 800 *C, the activation energy is
1.0 eV (Process 2). The factor of two difference between the two processes offers a
tempting basis for attributing this change in donor concentration to diffusion only. A
doubling in the measured activation energy is expected, based on (3.4.14) and (3.4.15), if
the diffusion length of the donor defect goes from being much less than an ABG optical
absorption depth to one which is much greater. This, however, would require Figure
5.2.3 to continuously transition from the slope for Process 1 to that for Process 2 over the
intermediate temperature regime where the diffusion length is of the order of the optical
absorption depth. We reject this hypothesis based on the data not exhibiting a gradual
transition between the two diffusion length extremes.
Instead, we recognize that significant defect diffusion relative to the optical
absorption depth is not likely given both the limited time involved in the RTA anneal and the
lack of a time dependency in the ABG PIMR data. Therefore, Process 1 is determined by
the generation of a surface V^s concentration in the regime of negligible diffusion. The 0.5
eV activation energy measured is comparable to the 0.7 eV found by others in experiments
involving annealing for tens of hours [Chiang and Pearson].
Process 2 must involve a diffusion process if the analysis done in the previous
section is valid since Process 1 has already exhausted E^. Still keeping the diffusion
length small requires that Ej^ for Process 2 be twice the measured activation energy, or 2.0
eV. This is similar to the 2.5 eV found by Chin for V^s diffusion [Chin] but differs
significantly from Chiang and Pearson's 4.0 eV.
But as pointed out by others
[Swaminathan], Chin's use of LEC (vs. Chiang and Pearson's horizontal Bridgeman)
includes a large number dislocations which would increase the diffusivity by "pipeline"
178
diffusion phenomena. Pipeline-assisted diffusion could be further enhanced by the nonequilibrium thermodynamic nature of the RTA process and could explain the lower
activation energy observed here.
Using (3.4.14), the L(T)N0(T) product can be extracted from the data in Figure
5.2.3 by recognizing that this figure is a plot of 2aL(T)N0(T) (normalized to the as-grown
values) versus temperature. Figure 5.2.4 shows L(T) versus annealing temperature for the
diffusion dependent regime greater than 800 *C, where L(T) was determined by using the
measured L(T)N(T)0 product and dividing by Chiang and Pearson's fit of NQ(T). The
diffusion length remains small compared to the optical absorption depth over the whole
temperature range and is therefore in agreement with the suppositions made in applying
(3.4.14). Furthermore, the diffusion length appears to reach a minimum of about 50 A at
lower temperatures.
The annealing temperature for Process 1, controlled by surface vacancy
concentration, can be similarly analyzed by using this minimum diffusion length as an
effective depth over which the surface vacancy concentration manifests itself. This is
compatible with the 50 A minimum diffusion length found above, as one would expect a
"surface" concentration to be located within several molecular distances from the surface.
Figure 5.2.5 shows the surface vacancy concentration determined from the PIMR data as
well as a comparison to Chiang and Pearson's data. The difference in slopes is due to the
previously mentioned discrepancy in activation energy, but the similarity is nonetheless
significant.
The analysis of Process 2, however, suffers from the lack of a time dependency in
179
750
850
900
950
1000
Temperature, deg C
1050
1100
Figure 5.2.4
\ ^ s Diffusion Length vs. Temperature
\ ^ s determined from PIMR-measured L(T)N(T)0 product and Chiang and Pearson'sfitto
their N(T)0 data.
180
1.2E17
I
E
1.0E17O
PIMR
e 8.0E16
o
<**
S 6.0E16
s
g 4.0E16-I
o
U
"2.0E16
0
600
Chiang & Pearson
620
640
660
680
700
Temperature, deg C
720
740
760
Figure 5.2.5
Surface V As Concentration vs. Temperature
Derived from Uncapped RTA PIMR data and Chiang and Pearson's fit to their data IChiang
and Pearson].
181
the measured data which should be evident if diffusion is involved. A possible remedy is
to consider L(T) not as a diffusion length but rather as the length over which the surface
concentration manifests itself. That is to say that a surface (i.e. no depth dimension)
vacancy population per se is an absurdity and we should really consider a vacancy
concentration within some small, but finite, region within the wafer and adjacent to the
surface.
A better solution is to consider Process 2 as being similar to the thin oxide regime in
the Deal and Grove Si(>2 growth model [Ghandhi]. By this analogy, greater penetration of
the VA S into the wafer is required before the As mass transport is the limiting factor. For
deeper V^s penetration (longer annealing time) a diffusion-limited region would be seem
[Chichibu] Three process regimes would than exist, with the first two found here:
1. Process 1: Increasing vacancy concentration with temperature as the dominant
process over some fixed depth perhaps determined by the free surface
energy.
2. Process 2: Fixed or slightly increasing vacancy concentration over a depth
which increases with temperature but is limited by the reaction out-gasing
As (generating V^ s ).
3. Process 3: (not observed) Fixed or slightly increasing vacancy concentration
over a depth which increases with temperature but is limited by the diffusion
of As to surface ( V ^ into the wafer).
In summary, the uncapped RTA results are compatible with a model of V^ s surface
vacancy generation. Using Arhennius plots of the EPM derived shallow donor
concentration corrected for bulk annealing effects, two separate mechanisms which
182
determined the donor concentration and profile (as a function of temperature) were
identified. At lower annealing temperatures, the introduction of V^s into the sample is
dominated by the equilibrium surface vacancy concentration. At higher temperatures the
\ ^ s population moves deeper into the wafer, but is limited by the reaction forming V^s.
We expect to see time-dependent, diffusion-limited behavior as the V/^ moves deeper into
the wafer. Whereas the analysis here of Process 1 gives an activation energy similar to that
in the studies with longer anneal time, Process 2 would not be seen because the V^s
penetrates to depths deeper than that where the proposed reaction rate limitation occurs.
5.2.3c Silox Capped RTA
The GaAs samples annealed using silox capped RTA exhibit a net shallow acceptor
concentration as determined by PIMR/EPM which eventually increases monotonically with
temperature after an initial, small decrease in N a for temperatures below 600 °C. The loss
of N a has already been attributed to processes other than Ga-loss to a silox cap and we,
therefore, concentrate here on the temperatures in excess of 600 *C.
Proceeding in a manner similar to that used for shallow donors with the uncapped
RTA, we remove the bulk effects from the silox-capped shallow acceptors with the data
from the nitride-capped experiment. The corrected shallow acceptor concentration,
attributed to V Q E generated at the GaAs-silox interface, is shown in Figure 5.2.6 as a
function of temperature. Over the regimefromabout 750 to 900 °C, the shallow acceptor
concentration increase steadily. Arhennius analysis of this data yields an activation energy
of0.4eV.
Again assuming that significant diffusion relative to an optical absorption length
183
1.1E16
E
1.0E16
B
9.0E15
«
u 8.0E15
B
4>
B
O
7.0E^
u
6.0E15
u
o 5.0E151
a
v
w
4.0E15
550
650
700
750
800
Temperature, deg C
950
Figure 5.2.6
Corrected and Normalized Shallow Acceptor Concentration vs. Temperature
for Silox Capped RTA
Correction for bulk effects with nitride capped results.
Normalization with respect to as-grown data
184
does not occur, we rely on (3.4.14) to aid in interpreting the activation energy fit to the
PIMR data. The 0.4 eV activation corresponds to that found for V Q E surface generation
[Chiang and Pearson]. This result is also compatible to that for the uncapped RTA
experiment in that a surface vacancy population is established prior to diffusion becoming
the dominant force.
The surface vacancy concentration itself can be extracted from the data using
(3.4.14), as was done with the uncapped RTA data. Assuming that the surface vacancy
population is concentrated in the same region as was the case for the V^g in the uncapped
RTA experiment-first 50 A of the sample-N0(T) is shown in Figure 5.2.7. There is a
large discrepancy, in this instance, between the surface V@a concentration determined by
PIMR and that from Chiang and Pearson's fit to their data. While the diffusivity for VQ &
is somewhat greater than that for V^ s (and, therefore, likewise for the diffusion length) the
more likely source for the difference in the two surface concentration results is the silox
capping for the PIMR data versus the vacuum for Chiang and Pearson's. The order of
magnitude increase with the silox attests to the oxide's well-known affinity for Ga.
Reviewing the silox capped RTA results, we have found an activation energy of 0.4
eV for the increase in shallow acceptor concentration over temperature. This is in
agreement with a model of generation of VQ & at the wafer surface. Annealing in a vacuum,
rather than with a silox cap, produces a lower concentration of V(j a , which is in agreement
with the known properties of silox.
185
10 18 -,
PIMR
f>
c
i io'.17.
e
o»
e
o
U
Chiang & Pearson
CO
>
10
1H 6
l — i
1 1 — | — i 1 — i |
760 780 800 820 840
i
| — i
1
860 880
1
1
900
Temperature, deg C
Figure 5.2.7
Surface V Q 3 Concentration vs. Temperature
Derived from Silox Capped PIMR data compared with Chiang and Pearson's fit to their
data [Chiang and Pearson].
186
5.3 Ion Implantation Phenomena
The point defect processes which eventually lead to the successful formation of an
ion implanted MESFET channel find their origins in the properties of the starting material,
the parameters of the IC processing, and synergies between the two. We have already seen
that at least a fraction of the implanted Si finds its way to Ga lattice positions prior to a high
temperature processing step, but the annealing method and temperature are more significant
factors in determining the final sheet resistance for the channel [Kanber]. Other studies
have demonstrated the importance of starting material parameters, such as EL2
concentration and As mole fraction [Von Neida], in influencing the eventual quality of the
active region.
PIMR's sensitivity to IC processing and starting material variability was
demonstrated in Chapter 4. In this section, the ion implanted results are analyzed using
available models--with modifications where necessary. The ion implanted RTA data is
viewed in light of Bindal's model as described in Chapter 2. Morrow's model, also
discussed in Chapter 2, is used to elucidate the the lack of a relationship between EL2 and
sheet resistance found with the channeling experiment. Analysis of the B coimplanted
phenomenarequiresadditions to Morrow's model for consistency with PIMR and sheet
resistance data.
5.3.1 Ion Implantation Activation with RTA
The experiment utilizing nitride capped RTA to activate a single implant schedule
aims to study sheet resistance variations induced by changes in a process variable, anneal
temperature. When the implanted and nitride capped RTA PIMR data is viewed with regard
187
to the unimplanted nitride capped RTA samples, the effects of the implant and associated
activation can be elucidated.
The normalized BBG data for the two nitride capped experiments shows that the
implant introduces an attenuation factor as a function of temperature (Figure 5.3.1a). The
attenuation steadily increases with temperature up to about 850 *C (Figure 5.3.1b), after
which it remains roughly constant until the point defect structure goes through its
rearrangement above 950 *C. In the low temperature region, the properties of the channel
are dominated by the implant damage. The 10% attenuation factor is therefore controlled
by effects such as Seff and FK, discussed in the previous chapter. In contrast, the 40%
attenuation factor once the activation process is complete (higher annealing temperatures)
represents microwave attenuation by the conducting channel and is consistent with the
results obtained when comparing the BWS anneal-only and implant/FA samples. Yet, it
can not be discerned from this data whether the improved sheet resistance up through 850
*C is due to either improved mobility, as the implant damage is annealed, or an increase in
the activation of the implanted Si.
Using the EPM relationship between sheet resistance and PIMR attenuation
developed in 3.3.2, we can deduce the sheet resistance from the BBG attenuation data in
Figure 5.3.1b. Figure 5.3.2 shows the activation as a function of temperature derived
from the EPM after normalizing to the activation maximum. Included with the derived
PIMR activation data is an activation curve from Bindal [Bindal 1989] which demonstrates
some similar features to the PIMR data, but differs markedly in the fact that the PIMR data
shows peak activation at a lower temperature, as well as a region of activation stability over
a fairly wide range of annealing temperatures. Nitride capping techniques have been
188
1.1
1
•
i
i
i
i
i
Unimplanted
s.7i
g.6
Implanted
73 .5
!•<
z
.3^
500
6(X)
700
800
900
Temperature, deg C
1000
1100
1000
1100
a)
500
600
700
800
900
Temperature, deg C
b)
Figure 5.3.1
Normalized Nitride Capped RTA BBG PIMR vs. Temperature
a) Implanted and Unimplanted Normalized BBG PIMR - Normalization with respect to
maximum measurement for the particular set of samples
b) Channel Attenuation Factor - Calculated as the ratio of the two plots in a).
189
e
©
'S
.£
*
1-95'
.9.85-
PIMR
o
< .8|.75.
E .65©
y^indal
fa
ft
z .55-6"
.5500
600
700
800
900
1000
1100
Temperature, deg C
Figure 5.3.2
Normalized Activation vs. Annealing Temperature for
PIMR and Bindal's Measured Data
PIMR dataset is determined from Figure 5.3.1b using the microwave attenuation model
developed in 3.3.2. Bindal's and the PIMR datasets are normalized to the maximum value
in the respective dataset.
190
known to reduce the activation efficiency as hydrogen incorporated into the nitride during
PECVD diffuses into the channel and neutralizes some of the activated Si [Pearton]. The
observed wafer warping supports this hypothesis as severe slip is also indicative of an
overly hydrogenated nitride; however this single data point (at 820 *C) could be due to the
error incurred by fitting the experimental data.
Examining the PIMR data, we find that initially activation increases with increasing
annealing temperature. Above 800 'C, the activation efficiency decreases slightly and then
levels off. We attribute this to the hydrogen diffusing out of the nitride and into the
channel, and reducing the ability of the Si to activate [de Souza]. It is interesting to note
that de Souza found very similar behavior to that depicted in Figure 5.3.1b (between 800
and 875 °C) except as a function of annealing time, thereby supporting the involvement of
hydrogen diffusion. Longer annealing time or higher annealing temperature drives the
hydrogen out of the channel andreturnsthe activation to its expected level, but temperature
cannot be traded with time over the entire annealing temperatureregimehere since (Bindal's
model predicts and his data illustrates that) the activation will eventually be reduced at high
annealing temperature by a preferential shift toward higher V^ s production. More V^ s
leads to a higher percentage of Si acceptors and the activation efficiency is reduced, as
observed in the high temperature portion of Figure 5.3.1b.
The ABG data versus temperature for the implanted samples (Figure 5.3.3a)
increases steadily up to 800 °C. Since the 6BG data indicated an increasing sheet
resistance over this same domain, the extremely low ABG PIMR values for the implanted
samples below 800 *C underscore the dominance of the reduced Seff over the attenuating
effect of the conducting channel. Above 800 *C, and continuing beyond 850 *C, the
191
1.1
1
I&-*.7^8
Unimplanted
J.6
g .5
Z .3
.2
.1
500
700
800
900
Temperature, deg C
1100
a)
700
800
900
Temperature, deg C
b)
1000
1100
Figure 5.3.3
Normalized Nitride Capped ABG PIMR versus Temperature
a) Implanted and Unimplanted ABG PIMR
b) Implanted ABG PIMR Corrected for Sheet Resistance and Bulk Annealing - calculated
from Figure 5.3.3a
192
implanted ABG PIMR data continues to decrease. Figure 5.3.2 suggests that the sheet
resistance is relatively constant over the 850 *C to 950 *C regime, so the large decrease in
implanted ABG PIMR compared to the unimplanted data leads us to conclude that these
changes in the implanted ABG PIMR are related to the subchannel defect constituency.
This conclusion is supported by the BWS etching experiment which showed that N a
immediately below the channel differs from that for the unetched, anneal-only wafer, as
well as the channeling and B coimplantation data. Hydrogen is probably not involved as it
should be uniformly distributed in the channel (by diffusion at 900 °C) and a discontinuity,
of the type observed in Figure 5.3.1b, is not found in Figure 5.3.3b
The implanted ABG PIMR, after correcting for the attenuating effect of the channel
and bulk annealing phenomena (Figure 5.3.3b) emphasizes the decrease in the corrected
ABG PIMR and implies a significant increase in net shallow acceptor concentration based
on the EPM. This augmented N a may be associated with the generation of electron traps in
the subchannel. We consider only electron traps since PIMR is a photoconductivity
measurement and hole phenomena are mitigated by a reduced mobility. Formation of
electron traps, such as EL3, EL6, and EL 12, in the subchannel region has been seen in
many RTA studies [Bindal 1991, Min, and Fang].
Identification of individual traps is not possible with the limited spectroscopy and
absence of a temperature stage in this implementation of PIMR. However, Bindal's model
would suggest that the increase in PIMR comes from either V^s, VQ S , or a defect complex
involving these elemental defects. V g a is an acceptor, and is therefore compatible with the
PIMR data. Returning to Bindal's model, we recognize that the increased VQ&
concentration in the channel would be associated with a sheet resistance decrease which is
193
not observed. The conflict can be resolved by requiring the VQ & to be located below the
0.3 |im channel but within the 5 \im range of the ABG PIMR measurement. Locating the
phenomena and the responsible point defects in the subchannel is supported by the BWS
etch and channeling data. Furthermore, subchannel phenomena are not unprecedented as
there are several examples in the literature of this, such as dislocation loops [Kanber] or
Gaj or Ga As diffusion [Bindal 1991].
Since this subchannel N a increases with annealing temperature, it is not due to the
implant alone. The effect was seen to a lesser degree with the unimplanted, nitride-capped
samples (Figure 5.3.3a) in section 5.2.3a, where it was attributed to the production of V Q S
by EL2 dissociation. A possibility for the subchannel N a is the formation of VQ a Ga As ,
acceptor, which is somewhat more stable at high annealing temperatures than several other
bidefects [Look]. VQ a Ga As is a compelling choice for a subchannel acceptor because of
the expected availability of antisites in the subchannel from collisions with the implanted Si
which presumably react with the "V*Qa from the dissociated EL2. However, Arhennius
analysis of the corrected implanted ABG data gives an activation energy of 0.9 eV for the
process which does not appear to be related to the nearly 4.0 eV binding energy of
VQaGaAs [Look] or the roughly 2 eV activation energy for V Q E diffusion [Suezawa].
A comparison of Figures 5.3.2 and 5.3.3 shows that the sheet resistance eventually
decreases in the same temperature range (800 to 850 °C) that the subchannel N a is
increasing. This suggests that the VQ a Ga As is forming at the expense of activated, Si
donors. This will be discussed in greater detail in section 5.3.3.
To summarize, the implanted nitride-capped RTA data can be corrected for the
microwave attenuation effect by comparison to their unimplanted counterparts. Bulk
194
annealing or anneal-only phenomena, discussed in section 5.2.3a, are also removed from
the implanted data in the same manner. This corrected data for the implanted nitride-capped
RTA shows an improvement in sheet resistance for annealing temperatures up to 800 *C as
indicated by the attenuation of the BBG measurement The maximum attenuation is similar
to that seen in comparing the anneal-only and implant/anneal wafers from the BWS. The
activation exhibits several departures from a published, temperature-dependent, point defect
model of Si activation which are consistent with nitride capped annealing side effects. The
corrected ABG data demonstrates that N a generation, perhaps VQaGa^s, in the subchannel
occurs with increasing anneal temperature and that a reaction related with origins in both the
implant and annealing processes produce this subchannel N a . The existence of subchannel
N a is supported by other data reported in Chapter 4, as well as that in the general literature.
5,3,2 Substrate Effects
Numerous studies have shown the importance of substrate point defect population
and stoichiometry on activation efficiency [Morrow, Von Neida, Saito]. The channeling
experiment reported in Chapter 4 (Figure 4.2.9a) demonstrated a boule-dependent
correlation between BBG PIMR and sheet resistance: linear regressions for the datasets
from the individual boules have nearly identical slopes but vastly different intercept points.
The slope for the two regressions is predicted very well by the EPM model of sheet
resistance and PIMR microwave attenuation. In this section, we analyze the difference in
intercept point with regard to an activation model incorporating both point defect and
stoichiometric constraints.
A significant difference in BBG PIMR versus sheet resistance suggests that EL2 is
195
responsible for the discrepancy.
There is a wealth of evidence that deep donor
concentration can affect activation efficiency [Anholt, Sato, Morrow, Brierly]. Since the
sheet resistance data was nearly identical for the two boules and the two boules possess
distinctly different EL2 concentrations, we wish to interpret this data with a model which is
capable of a muted, or moderated, EL2-dependent activation for the particular implant used
in the channeling experiment. Bindal's model does not incorporate such a feature, but
Morrow's model was specifically intended for just this situation.
Applying the model outlined by reactions (2.5.5) through (2.5.7) and summarized
in (2.5.8), sheet resistance for the channeled and unchanneled implant profiles used in the
experiment can be calculated as a function of B concentration for different EL2 and B
concentrations. Figure 5.3.4 demonstrates that B generally controls the sheet resistance.
In contrast, the EL2 concentration contributes to a large sheet resistance variation only
where activation is stifled by excessive boron contamination. Otherwise, typical variations
in EL2 concentration only amount to about a 1% change in sheet resistance, although the
channeled implant is slightly more sensitive to EL2 concentration than its unchanneled
counterpart. This is not to say that EL2 variations are not important to device uniformity,
just that gross changes in DC sheet resistance are not dominated by EL2 variations alone.
The model predicts a boron concentration of about 3.3 x 1 0 ^ cm"* for the boule (from
which the two wafers used in Figure 5.3.4 were sliced). The B concentration was not
confirmed by independent measurement, but is certainly well within the bounds of the
expected boron contamination from LEC growth [Morrow 1987a & b].
The channeling experiment demonstrated a 10% to 15% difference in BBG PIMR
between the two boules involved. Correcting for the microwave attenuation effect by the
196
MOODS'1300:
I 1200;
£ 1100tflOOO| 900;
« 800'8 700Z 600| 500c« 400--
^ M b M l M > * i
1E16EL2
B Concentration, cm" 3
a)
1500
1400
•s 1300
E 1200
©
1100
8 1000
2E16EL2
900
800<
700
8
V3
6(XH
500
4000
•
8
B Concentration, cm" 3
b)
•
i
9
•
•
10
11
Figure 5.3.4
Modeled Sheet Resistance vs. B and EL2 Defect Concentrations
B concentration in units of 10'" cm"3
a) Unchanneled Implant
b) Channeled Implant
197
conducting channel, this BBG difference translates into about half as much EL2 in Boule 2
as was found in Boule 1, yet the sheet resistances are nearly identical. As shown in Figure
S.3.S, the model predicts a negligible effect on sheet resistance for this implant. Thus,
even though PIMR shows different EL2 concentrations for the two boules, Morrow's
model confirms that no significant difference in activation should be observed based on the
starting material.
5.3.3 B Coimplantation Effects
As mentioned earlier, there are many reasons for attempting to gain greater control
of the doping profile than that afforded by Si implantation alone. Coimplants of Be are
routinely used when a sharper implant tail is needed. B was studied here to attempt to
separate the effects of having a p-type layer in the subchannel from those due to simply
performing an additional implant. As Morrow's model involves B directly, it is natural to
continue with its use. Moreover, this model has also proven successful with coimplants of
As, N, and P [Morrow 1988b].
The B coimplant data presented in Chapter 4 raises a questions regarding sheet
resistance which has been seen previously in a separate, but related, study [Heimlich].
Low dose implants into the channel itself degrade the activation efficiency, but this trend
corrects itself for sufficiently high B doses. Lower activation with such implants has been
seen elsewhere [Itoh].
Furthermore, the ABG PIMR presents two additional feature of interest. On each B
coimplant wafer, one quadrant was reserved for an Si-only implant and this region has
consistently lower ABG PIMR by at least 10%. Finally, the ABG PIMR data (Figure
198
4.2.13b) is dominated by a coimplant dose dependence for doses up to about 25 x 10 1U
cm~3, but this changes to an energy dependence for larger doses. During the dosedependent phase, the ABG PIMR increases with dose while the energy-dependent phase
sees the ABG PIMRresponsedecreasing with higher energies.
The EPM interpretation of the ABG measurement suggests that the increase in the
ABG PIMR for the B coimplanted regions over that with Si only is due to fewer net
shallow acceptors. Reducing the net shallow acceptor concentration can be accomplished
by generating donors or consuming acceptors. The latter is more likely since the difference
between the Si-only ABG PIMR and the B coimplanted ABG PIMR is monotonically
increasing with dose. The energy dependence also implies that fewer of these acceptors are
present as one goes deeper into the wafer. Combining the consumed acceptors with the
depth dependence suggests a model of VQ & or Ga^s implant damage, or an implantinduced defect complex, dominated by V(j a or Ga^s, reaction and diffusion. Implant
damage passivation at the root of this ABG PIMR discrepancy is not only supported by the
energy (or depth) dependence, but also by the increase in subchannel N a found in the
implanted RTA activation analysis (section 5.3.1). We therefore tentatively identify the
passivated defect as VQaGa^s.
For a model of B coimplantation to be consistent with the data, the sheet resistance
changes must also be accommodated. A sheetresistancemodelrequiresthat when the Si is
plentiful, the sheet resistance must be controlled by the B dose in such a way that the sheet
resistance increases and then returns to a value commensurate with the absence of
coimplanted B; when negligible Si is available, the B must have little effect.
199
A model which meets these requirements places B on Ga sites at the expense of
silicon and then effectively reverses the process when more B is available. The mechanism
driving the initial process involves the Si concentration so that when B coimplants into the
channel tail or subchannel are benign. More exactly, we can require that a portion of the Si
is activated as donors prior to any annealing, as suggested in section 4.1, and that this preanneal free electron concentration determines whether the B will hinder the activation of the
remaining Si. We therefore add a preemptive equation to the (2.5.5) through (2.5.7)
activation sequence of
Si + V
I
Ga~>SiGa
t5-3'1)
The vacancy presumably comes from the implant damage: predominantly in the front part
of the channel [Christel] and not considered by Morrow. (5.3.1) suggests that the implant
damage trap passivated by the B coimplant is related to a Ga^s-based trap [Bindal 1991].
Returning to the data, the change in sheet carrier concentration needed to go from
180 to 250 ohms/sq is about 40 x 1 0 ^ cm"3, based on calculations, which is nearly
identical to twice the B dose at 245 ohms/sq. This free electron concentration is also
approximately 1/4 of the sheet EL2 concentration contained in the channel for the Si implant
used with the coimplant experiment. Recalling from Morrow's activation model that the
first stage of activation (2.5.5) involves EL2 consumption by the Si in a ratio dependent on
the EL2 stoichiometric signature, the increase in sheet resistance due to the low energy B
coimplant can be associated with the coimplanted EL2 reacting with B rather than the Si:
B, + (l/9d)EL2 --> BGa
200
(5.3.2)
This has the effect of increasing the BQ & concentration by an amount equal to twice the
EL2 concentration. The alternative to (5.3.2) is (2.5.5) which is assumed to be
thermodynamically less attractive in this early stage of activation because it yields electrons
in an already doped environment.
The factor of two relationship between the B dose and the lost free electron sheet
concentration relates to Morrow's second stage of activation. Once all of the EL2 is
consumed, the Si begins to react according to (2.5.6) producing compensated, but
activated, Si and reducing the potential electron concentration by a factor of 2 for every
B Q E present. In this way, the EL2 concentration reduces the sheet carrier concentration by
an amount four times greater than its own.
If unactivated B is still present within the channel after (5.3.2) and (2.5.6) have
gone to completion, than the B reacts in a countermanding fashion, similar to P in
Morrow's coimplant model [Morrow 1988b], and releases complexed Si complexed
B
I
+ B
Ga S As " > B Ga B As +
Si
I
<5-3'3>
where thereactantcomplex is an acceptor and the product complex is assumed neutral. The
end result of this, as far as the sheetresistanceis concerned, is that, with sufficient B in
the channel, it is as if no coimplant were performed. Eventually, of course, other effects,
such as implant damage, will not appear so benign in the resulting sheet resistance.
The coimplant data, as well as that for the RTA-activated implants, show that the
subchannel N a is a factor even without a B coimplant. We have assumed that VQ a Ga^ s is
involved and that a reaction between dissociated EL2 remnants and the implant damage
forms this complex. RTA activation studies show that the trends in the subchannel N a are
201
reflected, to some extent, in the sheet resistance data. This implies that the production of
v
G a G a A s i s m competition with the activation of the Si implant We may further suppose
from this that VQ a Ga^ s production competes with any implant which prefers activation on
the Ga sublatdce. In either case, the competition between V Q ^ a ^ and activation leads us
to conclude that the source of the dissociated EL2 in a model of subchannel N a is that
which is represented by (2.5.5) and (5.3.2). A possible candidate for the reaction
producing the subchannel defect is
GaAs + (l/3 d )EL2->V Qa Ga As
(5.3.4)
This reaction and (5.3.2) compete for EL2, but go to completion before (2.5.5) begins.
Therefore, coimplanting B can be used to alter the sheet resistance and reduce some of the
trapping phenomena in the subchannel.
To summarize the activation with B coimplantation:
Si
I + V Ga" > S i Ga
+ e
t5"3'1)
'
B, + (l/3d)EL2 - > B Ga
(5.3.2)
Si, + (l/ad)EL2 --> Si*a + e"
(2.5.5)
2SiI + B G a ->Si^ a + B Ga Si As
(2.5.6)
B
<5'3-3>
I + B Ga Si As" >B Ga B As
Si
r->SiGa
+ V
As
+ Si
I
<2'5-7>
With the exception of (5.3.1), each reaction goes to completion before the next begins. If
subchannel N a is involved (5.3.4) will compete with reaction (5.3.2) and (2.5.5) in the tail
202
of the implant where we expect to find implant damage. B coimplants can be used to
reduce the effects of the subchannel defect.
S.S Summary
This chapter focused on material characterization using PIMR as the measurement
tool, the EPM as the link between the measurement and the point defect concentrations, and
additional models, such as stoichiometrically-constrained activation, to relate the observed
point defect dynamics to process variables. Both annealing and implantation effects were
studied. EL2 dissociation and bulk N a phenomena were discussed as well as the effects
which can occur with silox or uncapped samples. The ion implantation analysis included
temperature effects from the activation anneal, substrate effects from the starting material,
and implant-related phenomena with the B coimplant.
The importance of tracking the EL2 concentration during the MMIC process cannot
be understated since this defect gives undoped GaAs its SI property. The annealing
analysis showed that the same processes control the changes in the EL2 concentration
during RTA for all of the capping techniques. Annealing reduces the EL2 concentration,
and while the dissociation of EL2 at the annealing temperature is due to V^s diffusion, its
reconstitution during the quench is limited by V(j a diffusion.
The EL2 dissociation is not without its effects on Na. For properly capped wafers
or those in an appropriated gaseous arsenic environment, the dissociated EL2 has no net
effect on N a . Assuming a tri-defect EL2, VQ a AsQ a \^ s , the EL2 breaks up into VQ &
(acceptor) and AsQa"V^s (donor) below 850 *C. The donor bidefect is reduced to V Q 3 for
higher temperature anneals, and the net effect is an increase in Na.
203
The use of uncapped or silox capped RTA tends to remove As or Ga, respectively,
from the wafer surface. This required reconsidering the integrating nature of the ABG
PIMR measurement, discussed in Chapter 3, to elicit diffused defect information from the
measured data. In the case of the uncapped samples, a two-step process was found as a
function of temperature; the first, below 750 *C, is limited by the generation of a YA S
surface concentration and the second, above 800 *C, is limited by the reaction forming V^ s
at the interface between the V^s surface population and the unaffected GaAs. The latter
process is analogous to the initial, reaction rate limited regime in the Deal and Grove
thermal silicon oxidation model. The silox capped samples showed that the process of
creating the V Q & surface concentration was the factor which limited the change in the
measured N a as a function of annealing temperature.
The analysis of ion implanted samples revealed effects relating to the anneal, the
substrate, the implant, and combinations of all three. The RTA-activated implant data was
compared to Bindal's activation data and showed that the activation process here reaches a
plateau at lower temperatures. Hydrogen diffusion from the nitride cap may be partially
responsible as the hydrogen will inhibit activation until it becomes dispersed at higher
annealing temperatures. An eventual increase in activation with annealing temperature,
however, is not found. Examination of the ABG data for these samples revealed that N a
increases quite dramatically in the subchannel for annealing temperature above the
activation plateau. The source of the N a increase was tentatively assigned to the production
°f ^Ga^aAs fr°m bo™ t n e anneal and the post-implant damage in the subchannel. The
presence of increasing quantities of VQaGa^s may also explain the muted activation in
these samples.
204
Sheet resistance variations were not found to be a function of the substrate EL2
concentration in an analysis of the channeling experiment. Different, but apparently
MMIC-quality, EL2 concentrations were found by BBG PIMR in the two boules used in
the experiment. Wafers from these boules did not yield activated channels with vastly
different sheet resistances. Morrow's activation model showed that, from a theoretical
standpoint, this is also the case, suggesting that control of the as-grown B concentration is
much more important for reproducing sheet resistance.
Morrow's model was extended to account for the sheet resistance variations induced
by the B coimplantation and to include trapping phenomena. The presence of B in small
quantities apparently inhibits Si activation, but larger B doses leads to the formation of
benign complexes without a reduction in activation. The ABG data for the B coimplant
experiment also showed subchannel N a phenomena. Here, the subchannel N a , as a
function of dose and energy, followed the sheet resistance, thereby suggesting that the
production of the VQaGa^s competes with both the Si and B activation processes. Since
B coimplantation was found to affect sheet resistance and subchannel lifetime (defect
concentration), we conclude that at least some of the contribution made by Be coimplants to
the activated profile characteristics and the RF MESFET performance is independent of the
p-type nature of the Be.
205
CHAPTER 6
Summary and Conclusions
This work has encompassed several areas relevant to PIMR and processed GaAs
characterization. The PIMR technique itself has been expanded upon and an understanding
of its use with ion implanted and annealed GaAs has been developed. A wealth of
experimental data has been gathered using PIMR and other techniques on samples
processed in many different ways. These data were analyzed using well-developed
techniques, published interpretations, and modified models in an effort to gain insight into
the evolution of GaAs material properties by MMIC processing.
In this chapter, the work reported is summarized and future directions are proposed.
We begin with the PIMR technique itself, discuss the improvements made to the system
and the measurement methodology, and then move on to using PIMR to investigate
processing phenomena. The application of PIMR to processed GaAs, in general, is
followed by a closer look at the data on RTA activation and B coimplantation from the
perspective of material characterization discussed in Chapter 5. Finally, we review the two
models used most in this study, the EPM and Morrow's model with extensions.
6.1 PIMR: System Improvements and Functional Characterization
Enhancements to the PIMR equipment were aimed toward improving the sensitivity,
repeatability and ease of use. In the microwave subsystem this was mainly limited to
replacing a 5 mW sweeper with a 150 mW Gunn diode oscillator and installing a constant
206
bias circuit for the Gunn. Also added was a new antenna structure with a lower VSWR
match to the microwave waveguide. A more detailed experimental study, with some
theoretical understanding, of the new antenna would prove useful in investigating several
of the limiting effects in this implementation of PIMR.
The optical subsystem also received attention. The pulsing electronics were improved,
especially in the area of repeatability. The ABG laser power supply circuitry could be
enhanced further, either by obtaining a high quality pulser similar to that controlling the
BBG laser or by regulating the supply current. Both lasers would benefit from procuring a
high current power supply so that true pulses are produced, rather than the "ramped" output
now available.
The capability of the measurement/detection subsystem was augmented with the digital
oscilloscope. The new oscilloscope produces the averaged waveforms more quickly than
the equipment used previously. The multiple-channel processing features improve the
quality of the data by allowing the removal of stray periodic signals from the data using a
subtraction scheme. The built-in measurement capability readily lent itself to automating
the system.
The automation of PIMR has mainly been achieved with the motorized stage and the
PS/2 computer. Under computer-control, the PIMR system with the motorized stage is
almost a fully-automated, turn-key characterization tool which produces high quality, 2D
maps, with the Macintosh. Improvement in the speed of the mapping function is necessary
before PIMR can become a widely used technique. Increasing the speed could be realized
by replacing the oscilloscope with custom-made circuitry for peak detection and allowing
the stage to continuously scan. A PIMR system of this kind, with the PS/2 controlling the
207
lasers, could map a wafer in minutes rather than the 20 hours now needed (for full wafer
maps with 2 PIMR scans and 1 dark map).
The PIMR system used throughout this study was shown to have a one sigma
repeatability of 1.2% which could be reduced to less than 1% by fixed regulation of the
ABG laser power supply. However, the greatest future improvement to PIMR can be made
by removing the parasitic patterning effect, seen in the 2D mapping. The simplest solution
is to reduce the magnitude of the effect through the use of a conducting GaAs or Si
annulus, but an antenna structure which has very little near-field component at the surface
of the wafer could remove the patterning altogether, and may also reduce the degree to
which a conducting channel attenuates PIMR.
The attenuation of the PIMR measurement in the presence of a conducting channel is
both a nuisance and a blessing. We would ultimately like to probe the bulk and subchannel
properties of processed GaAs and not need to correct the data. However, this sensitivity
adds another dimension to the technique. The BBG laser was used here several times as an
indicator of the sheet resistivity via the attenuation effect, but we could just as easily have
used another BBG laser at 1.1 or 1.3 |im, which would also provide additional
spectroscopic information. The dark, non-transient use of the PIMR system could also be
applied to the measurement of sheet resistance, but this will require a DC amplifier to
improve the voltage sensitivity during mapping operation. Both options should be
investigated with a detailed series of adjacent wafers (from a single boule) with implanted
channels of different sheet resistivity.
We conclude that the PIMR system used in this study is of greater utility than that
previously available based on the installed automated testing and mapping capabilities (for
208
both transient and dark responses) and the improved repeatability.
6.2
Experimental Results and Process Monitoring and Development
It is clear that PIMR can differentiate between all of the basic processes through active
channel formation, as seen with the BWS. Furthermore, PIMR was used to investigate
several processes and process phenomena. The analysis based on these data allowed us to
gain new insight into GaAs processing, as well as GaAs behavior during processing. In
this section, the experimental results and analysis are summarized.
The BWS provided a solid justification for more fully researching PIMR's use with
GaAs processing. Compared to an as-grown wafer, the implant only sample shows a 10%
reduction in BBG PIMR while the ABG PIMR decreases by 80%. The anneal-only sample
has an increase in ABG PIMR, but a decrease in BBG PIMR while the implanted and
annealed wafers have relatively large increases in ABG PIMR and decreases in BBG
PIMR. It should be noted again that this is in comparison to an as-grown sample and that
PIMR cannot yet absolutely differentiate wafers of unknown origin. In a MMIC
fabrication facility, this is not a concern since the processing received by any given wafer is
known prior to testing.
Two related tests were also reported in the BWS data section. The results of BBG
PIMR for a wafer with one half B implanted and the other implanted with a similar Si
profile showed signs of the Franz-Keldysh effect prior to a high temperature annealing
step. This was interpreted as partial activation of the Si without the benefit of a traditional
activation annealing. The second test involved etching one of the BWS wafers. The ABG
PIMR data for this sample showed evidence of an altered point defect structure in the
209
region below the active channel layer. More detailed experimentation of both of these
phenomena would be beneficial in attempting to quantify these effects in terms of the
implant and, in the case of the etching experiment, annealing parameters.
PIMR has proven itself a useful tool in a process monitoring role. Problems with the
furnace anneal procedure, which seem to be related to the furnace itself were seen in the
data presented in section 4.2.1. The placement of wafers between the up- and downstream ends of the oven could provide more information on the temperature gradient and its
effects as a function of wafer position. The repeatability of the furnace annealing process
could also be established, with PIMR and sheet resistance measurements, by running this
experiment several times.
RTA process development can be even more complicated than that for furnace
annealing, and PIMR is also of use in monitoring this annealing method. The uncapped
data showed that V^s was incorporated in a two-step process involving establishing a
surface concentration followed by reaction-rate limited growth of the surface V^ s region.
Silox capping leached Ga from the wafer which was seen in the PIMR data as the
generation of VQ & at the surface. EL2 dissociation and reformation during annealing was
found to be limited by vacancy diffusion using an EL2 model of VGaAsGa^AsThe channeling experiment yielded two interesting results. First, PIMR was able to
select the wafers with acceptable sheet resistances (i.e those without channeling).
Channeled wafers had ABG PIMR values outside a regime where only unchanneled wafers
were found. An unchanneled wafer was also outside the "acceptable" region which implies
that there may be some time-dependent trapping phenomena in any MESFETs subsequently
fabricated on this wafer, even though the wafer passed a sheet resistance specification.
210
Second, PIMR revealed a difference in EL2 concentration across boules which did not
translate into a change in sheet resistance. It was later shown that this is compatible with an
accepted model of Si activation which includes the as-grown B concentration as a factor in
activation.
Implant activation by RTA was also investigated. Here, susceptor annealing was
shown to have promise and repeating the designed experiment used with the uncapped and
capped RTA would be of great interest. This would be helpful in ascertaining the degree to
which the nitride stress contributed to the bulk EL2 and N a annealing phenomena. The
temperature dependence of the EL2 dissociation does not require a contribution from the
nitride stress, but investigation of the anneal time variability could benefit from more
experiments with the susceptor.
RTA of the implanted nitride-capped pieces demonstrated clear signs of activation, but
the most interesting result here was the apparent subchannel point defect evolution.
Although VQaGa^s was suggested based on the compatiblity with the assumed EL2 model
and the implant damage, more direct proof is required. The apparent increase in N a below
the channel warrants further investigation, perhaps using susceptor RTA or the furnace
anneal, under a variety of conditions. DLTS prior to etching followed by PIMR and PL
after etching would be most informative in attempting to isolate the nature of this defect, or
defects.
The subchannel N a was also seen in the B coimplant data. This data further supported
^ e ^Ga^ a As model* Dut raised interesting questions as to the source of the V Q E and the
relationship between the Si activation and the subchannel N a . More importantly however,
the analysis of the B coimplant data pointed out that the p-type nature of Be subchannel
211
coimplants is not the only factor contributing to the observed changes in MESFET
performance. Repeating this experiment with a similar set of Be coimplanted wafers would
be very helpful in isolating various aspects of the buried p-layer, especially if these samples
are eventually tested as completed MESFETs.
Based upon the data reviewed above, we conclude that PIMR is suitable for process
monitoring and development in a MMIC fabrication line. It is well-suited to this role
because of its demonstrated ability to identify differences among starting wafers as well as
those at any point in the active channel formation processing. There is no doubt, however,
that PIMR should be more fully investigated by testing wafers prior to gate definition and
then attempting to correlate this to device performance.
6.3 Model Development and Extensions
Finally, we discuss the modeling that was presented. The EPM represents the bulk of
the model development presented in this manuscript and supported the extraction of point
defect data from the PIMR measurement. Although not done in nearly as much detail,
several extensions were made to Morrow's Si activation model.
A central tenet of the photoconductive model for unprocessed, SI material is the
recombination via EL2 as the major effect driving the decay in the carrier concentration.
This is not necessarily the case for processed material as the channel is not compensated
and additional, perhaps more highly concentrated, defects can be introduced by the
implantation and annealing. Based on the etching results, RTA experiments with various
capping methods, and a similarity of the FA and RTA data, the photoconductive model for
SI material suffices for processed material so long as we make the proper corrections for
212
Seff and and PIMR (microwave) effects. In the case of the former, it is only important that
the channel is doped sufficiently high so that we are operating in the constant ABG PIMR
region of Figure 3.4.5.
The EPM departs in several respects from the TLM. An additional defect level has
been added to account for the findings of Wang. The model now accounts for a finite
optical pulsewidth rather than just the steady-state response. The effect of the conducting
layer has been added, in terms of the photoconductive implications as well as those
associated with the microwave portion of the PIMR techniques.
Overall the EPM yielded interpretations of the experimental results which were selfconsistent and contributed to a point defect analysis which was compatible with data in the
literature, but a few items could be improved upon. While the EPM was compared to
measured data to calibrate the ABG excitation, the same was not true for the BBG. This
should be done as a matter of completeness. The EPM performed well in predicting the
attenuation of the conducting channel, however improvements in its ability to model the
SRV effects are desirable and could be achieved with the currently available simulation
software and a more powerful computer. Although not formally included in the EPM, the
extraction of N a where mass transport from the surface was involved was fairly successful
but should be reviewed with a more extensive set of data involving longer anneal times.
The model of B coimplantation allowed us to account for the observed changes in
sheet resistance. In the model, the presence of low doses of B inhibits the activation of Si
associated with the loss of EL2. Larger doses of B activate as B complexes and leave the
EL2 free to activate the Si. This work represents an evolutionary extension to the body of
work developed by Morrow, as he has considered coimplants previously. Naturally, Be
213
coimplant data would further extend the model in this respect, and with MESFET testing
upon completion of processing, would lead to a greater understanding of the role that ptype dopants play in buried layer processes.
It was also shown that Morrow's model could be used to predict time-dependent
phenomena, as in the case of the subchannel Na. In the process of consuming EL2, the
low doses of B also block the reaction which forms the subchannel acceptor. The higher
doses, again, counteract the effect. The testing proposed in the previous section regarding
the etched wafers would also be helpful here.
To conclude, the EPM has proven adequate in extracting point defect concentration
data from the PIMR measurement for SI LEC GaAs processed through active channel
formation. Furthermore, the EPM is useful in determining sheet resistance via the
microwave attenuation. We also conclude that the extensions made to Morrow's model are
sufficient to explain the sheet resistance and time-dependent effects of B coimplantation.
214
Literature Cited
Chanter 1
Anholt, R. and T. W. Sigmon, ED 26, p. 250 (1989).
Anholt, R. J., and T. W. Sigmon, J Electr Mat 1, p. 5 (1988).
Bindal, A., K. L. Wang, S. J. Chang, M. A. Kallel, and P. K. Chu, J Electrochem Soc
J2& p. 2414 (1989a).
Bindal, A., K. L. Wang, S. J. Chang, M. A. Kallel, and O. M. Stafsudd, JAP £5, p. 1246
(1989b).
Canfield, P. and L. Forbes, ED 23_, p. 925 (1986).
Christel, L.A. and J. F. Gibbons, JAP 52, p. 5050 (1981).
de Souza, J. P., D. K. Sadana, and H. J. Hovel, MRS 144, p. 495 (1989).
de Souza, J. P., D. K. Sadana, H. Baratte, and F. Cardone, APL 52, p. 1129 (1990).
Dobrilla, P., and J. S. Blakemore, JAP 60, p. 169 (1986).
Fang, Z-Q., H. Yamamoto, and D. C. Look, unpublished (1991).
Fujisaki, Y, and N. Matsunaga, GaAs IC Symposium, p. 235 (1988).
Holmes, D. E., M. Kuwamoto, C. G. Kirikpatrick and R. T. Chen, Conference on SI IIIV Compounds (1984), p. 204.
Holtz, M., R. Zallen, A. E. Geissberger, and R. A. Sadler, JAP 52, p. 1946 (1986).
Immorlica, A.A., D. R. Ch'en, D. R. Decker, and R. D. Fairman, GaAs and Related
Compounds 1980a, p. 423.
Immorlica, A. A., D. R. Decker, and W. A. Hill, ED 22, p. 2285 (1980b).
215
Lagowski, J. and H. C. Gatos, 13th Int'l Conf on Defects in Semiconductors, p. 73
(1984).
Lee, R.E., T. A. Hunter, R. P. Bryan, H. M. Olsen, H. V. Winston, and R. S. Beaubien,
GaAs IC Symposium (1984), p. 45.
Look, D. C , Electrical Characterization of GaAs Material and Devices. Wiley, 1989.
McNally, P. J., EDL 5 p. 126 (1984).
Mickanin, W., P. Canfield, E. Finchem, and B. Odekirk, GaAs IC Symp 1989, p. 211.
Miller, G. L., D. A. H. Robinson, and J. D. Wiley, Rev Sci Instrum 4Z, p. 799 (1976).
Mircea, A., A. Mitonneau, L. Hollan, and A. Briere, App Phys 11, p. 153 (1976).
Miyazawa, S., in Defects in Semiconductors, ed. von Bardeleben, p. 1 (1986).
Ohkubo, N., S. Chichibu, and S. Matsumoyto, APL 5_3_, p. 1054 (1988).
Sadler, R. A., G. D. Studtmann, A. E. Geissberger, H. P. Singh, and R. A. Burrier,
GaAs IC Symposium 1989, p. 219.
Saito, Y, JAP 6J5, p. 846 (1989).
Sato, T., M. Tajima, and K. Ishida, APL 5J., p. 755 (1987).
Soares, R., GaAs MESFET Circuit Design. Artech House, Boston, 1988.
Tabatabaie-Alavi, K., ED 22, p. 96 (1990).
Wager, J.F., and A. J. McCamant, ED 24, p. 1001 (1987).
Wagner, J., and C. R. Fritzsche, JAP 64, p. 808 (1988).
Williams, J. S., R. J. Elliman, S. T. Johnson, D. K. Sengupta, and J. M. Zemanski, MRS
144, p. 355 (1989).
Yeats, R., D. C. D'Avanzo, K. Chan, N. Fernandez, T. W. Taylor, and C. Vogel, ffiDM
1988, p. 842.
216
Chapter 2
Anholt, R. and T. W. Sigmon, ED 26, p. 250 (1989).
Anholt, R. J., and T. W. Sigmon, J Electr Mat I, p. 5 (1988).
Arthur, J. R., J Phys Chem Solids 2&, p. 2257 (1967).
Baraff, G.A., and Schluter, M.L., Phys Rev Lett 55, p. 1327 (1985).
Baraff, G.A., and Schluter, MX., Phys Rev B 25, p. 6154 (1987).
Baraff, G.A., and Schluter, M.L., Phys Rev Lett £2, p. 2156 (1989).
Bindal, A., K. L. Wang, S. J. Chang, M. A. Kallel, and P. K. Chu, J Electrochem Soc
136. p. 2414 (1989a).
Bindal, A., K. L. Wang, S. J. Chang, M. A. Kallel, and O. M. Stafsudd, JAP 65, p. 1246
(1989b).
Borrego, J. M., R. J. Gutmann, C. S. Lo, M. C. Heimlich, and O. Paz, Conference on
Defect Recognition and Image Processing in III-V Compounds n, p. 177 (1987).
Bourgoin, J. C , H. J . von Bardeleben, and D. Stievenard, JAP £4. P- R65 (1988).
Bothra, S., Ph.D. Thesis, Rensselaer Polytechnic Institute (1988).
Brierly, S. K., and D. S. Lehr, APL 55, P- 2426 (1989).
Campbell, C. S., Master Thesis, Rensselaer Polytechnic Institute (1988).
Chichibu, S., N. Ohkubo, and S. Matsumoto, JAP 64, p. 3987 (1988).
Christel, L.A. and J. F. Gibbons, JAP 52, p. 5050 (1981).
Dobrilla, P., and J. S. Blakemore, JAP 5S, p. 208 (1985).
Fang, Z-Q., H. Yamamoto, and D. C. Look, to be published (1991).
Govorkov, A. V., A. V. Markov, E. M. Omeljanovsky, A. J. Polyakov, V. I. Raihstein, V.
A. Fridman, Conference on SI III-V Compounds, p. 145 (1988).
217
Hasegawa, H., H. Ohno, H. Shimizu, and S. Seki, J Electr Mat 13., p. 931 (1984).
Heimlich, M.C., Master Thesis, Rensselaer Polytechnic Institute (1987).
Holtz, M., R. Zallen, A. E. Geissberger, and R. A. Sadler, JAP 52, p. 1946 (1986).
Hunter, A. T, H. Kimura, J. P. Bukus, H. V. Winston, and O.J. Marsh, APL 41, p. 74
(1984).
Johnson, E. J., J. A. Kafalas, and R. W. Davis, JAP 54, p. 204 (1983).
Jordan, A. S., A. R. Von Neida, and R. Caruso, J Cryst Growth 2& p. 243 (1986).
Kazior, T. E., S. K. Brierly, and E J. Piekarski, JEEE Semicond Manuf 4, p. 21 (1991).
Kunst, M. and G. Beck, JAP£Q, p. 3558 (1986).
Kunst, M. and G. Beck, JAP £3_, p. 1093 (1987).
Krawczyk, S. K., J. Y Longere, K. Schohe, M. Ganigues, J. Tardy, R. Coquille, H.
L'Haridon, and P. N> Favennec, Conference on SI III-V Compounds, p. 287
(1990).
Ladbrooke, P., and S. R. Blight, ED 25, p. 257 (1988).
Lagowski, J., and H. C. Gatos, 13th Int'l Conf on Defects in Semiconductors, p. 73
(1984).
Lagowski, J., H. C. Gatos, C. H. Kang, M. Skowronski, K. Y Ko, and D. G. Lin, APL
42, p. 892 (1986)
Lo, C. S., Ph.D. Thesis, Rensselaer Polytechnic Institute (1988).
Look, D. C, Conference on SI III-V Compounds, p. 1 (1988).
Martin, G. M., A. Mitonneau, and A. Mircea, Electronic Lett 12, p. 191 (1977).
McNally, P. J., EDL 5p- 126 (1984).
Min, S-K., E, K Kim, and H. Y Cho, JAP £3_, p. 4422 (1988).
218
Moore, F. G., H. B. Dietrich, E. A. Dobisz, and O. W. Holland, APL 52, p. 911 (1990).
Morrow, R. A., J Mater Res 2, p. 681 (1987a).
Morrow, R. A., JAP £2, p. 3671 (1987b).
Morrow, R. A., Conference on SI III-V Compounds, p. 65 (1988a).
Morrow, R. A., JAP &, p. 1889 (1988b).
Morrow, R. A., APL 55, p. 2523 (1989).Ohkubo, N., S. Chichibu, and S. Matsumoyto,
APL 52, p. 1054(1988).
Sato, T„ K. Terashima, S. Ozawa, M. Nakajima, T. Fukuda, and K. Ishida, GaAs IC
Symp 1984, p. 53.
Saito, Y, JAP £5, p. 846 (1989).
Skolnik, M. S., and M. R. Brozel, APL 4£, p. 341 (1986).
Skromme, B. J., S. S. Bose, B. Lee, T. S. Low, T. R. Lepkowski, R. Y DeJule, G. E.
Stillman, and J. C. M Hwang, JAP 58, p. 4685 (1985).
Suezawa, M., and K. Sumino, Jap JAP, p. LI8 (1987).
Tabatabaie-Alavi, K..ED22, p. 96 (1990).
van Vechten, J. A., J Electrochem Soc 122, p. 423 (1975).
von Bardeleben, H. J., Phys Rev B 24, p. 7192 (1986).
\on Neida, A. R., S. J. Pearton, M. Stavola, and R. Caruso, APL 49_, p. 1708 (1986).
Wager, J. E, and J. A. van Vechten, Phys Rev B 25, p. 2330 (1987).
Wagner, J., and C. R. Fritzsche, JAP £4, p. 808 (1988).
Wagner, J., H. Seelewind, and W. Jantz, JAP SL p. 1779 (1990).
Wang, M. S., Ph.D. Thesis, Rensselaer Polytechnic Institute (1990).
Wang, M. S., and J. M. Borrego, and P. Pearah, J Electrochem Soc 122, p. 3648 (1990).
219
Yablonovitch , E., R. M. Swanson, W. D. Eades, and B. R. Weinberger, APL4&, p. 245
(1986).
Yablonovitch, E., C. J. Sandroff, R. Bhat, and T. Gmitter, APL 51, p. 439 (1987).
Yano, K., and T. Katoda, Conference on SI UI-V Compounds, p. 105 (1988).
Zou, Y, Z. Jicheng, L. Yian, W. Kuangyu, H. Binghua, L. Bingfang, L. Cuncai, L.
Liansheng, S. Jiuan, and s. Chi, 13th Int'l Conf on Defects in Semiconductors, p.
1021 (1984).
Chanter 3
Atwood, E. A. and R. J. Gutmann, J Electrochem Soc, to be published (1992).
Blakemore, J. S., JAP 52, p. R123 (1982).
Bothra, S., Ph.D. Thesis, Rensselaer Polytechnic Institute (1988).
Brierly, S. K, D. S. Lehr, and A. K. Grabinski, JAP 62, p. 5085 (1988).
Brierly, S. K, and D. S. Lehr, Conference on SI III-V Compounds, p. 407 (1990).
Campbell, C. S., Master Thesis, Rensselaer Polytechnic Institute (1988).
Duncan, W. M., G. H. Westphal, and A. J. Purdes, JAP 66, p. 2430 (1989).
Gutmann, R. J., C. S. Campbell, M. C. Heimlich, J. M. Borrego, and D. Bliss, J
Electrochem Soc 126, p. 3795 (1989).
Heimlich, M.C., Master Thesis, Rensselaer Polytechnic Institute (1987).
Heimlich, M., W. M. Robertson, G. Arjavalingam, and J. M. Halbout, Electr Lett 28, p.
249 (1992a).
Heimich, M. C , Atwood, E. A., N. Jain, and R. J. Gutmann, to be submitted to IEEE
MTT (1992b).
220
Jensen, N, Master Thesis, Rensselaer Polytechnic Institute, 1986.
Miller, G. L., D. A. H. Robinson, and J. D. Wiley, Rev Sci Instrum 47_, p. 799 (1976).
Lo, C. S., Ph.D. Thesis, Rensselaer Polytechnic Institute (1988).
Morse, P. M., and H. Feshbach, Methods of Theoretical Phvsics. McGraw-Hill, 1953.
Press, W. H., et al, Numerical Recipes in C: The Art of Scientific Computing. Cambridge
University, 1988.
Seeger, K., Semiconductor Phvsics: An Introduction. 3rd ed., Springer-Verlag, 198S.
Sewell, G., The Numerical Solution of Ordinary and Partial Differential Equations.
Academic Press, 1988.
Sze, S. M., Physics of Semiconductor Devices. 2nd ed., Wiley, 1981.
Wang, M. S., Ph.D. Thesis, Rensselaer Polytechnic Institute (1990).
Chapter 4
Bindal, A., K. L. Wang, S. J. Chang, M. A. Kallel, and P. K. Chu, J Electrochem Soc
I2& p. 2414 (1989).
Blaauw, C , JAP 54, p. 5064 (1983).
Blakemore, J. S.. Semiconductor Statistics. Dover. 1987.
Brierly, S. K, D. S. Lehr, and A. K. Grabinski, JAP £2, p. 5085 (1988).
Brierly, S. K, and D. S. Lehr, Conference on SI III-V Compounds, p. 407 (1990).
Canfield, P. and L. Forbes, ED 22, p. 925 (1986).
Davies, D. E., Nuclear Instruments and Methods B7/8. p. 387 (1985).
Heimlich, M. C , R. J. Gutmann, D. Seielstad, and D. Hou, Conference on SI III-V
Compounds, p. 367 (1990).
221
Kazior, T. E., S. K. Brierly, and F. J. Piekarskd, IEEE Semicond Manuf 4, p. 21 (1991).
Look, D. C, Conference on SI III-V Compounds, p. 1 (1988).
Mason, R. L., R. F. Gunst, and J. L. Hess, Statistical Design and Analysis of
Experiments with Applications to Engineering and Science. Wilev. 1989.
Miller, G. L., D. A. H. Robinson, and J. D. Wiley, Rev Sci Instrum 4Z, p. 799 (1976).
Sadler, R. A., G. D. Studtmann, A. E. Geissberger, H. P. Singh, and R. A. Burner,
GaAs IC Symposium 1989, p. 219.
Tabatabaie-Alavi, K., ED 22, p. 96 (1990).
Vignaud, D., and J. L. Farvacque, JAP £5_, p. 1516 (1989).
Wang, M. S., Ph.D. Thesis, Rensselaer Polytechnic Institute (1990).
Wolf, S. and R. N. Tauber, Silicon Processing for the VLSI Era: "Nfolume I - Process
Technoloby. Lattice Press, 1986.
Chapter 5
Anholt, R. J., and T. W. Sigmon, J Electr Mat 1, p. 5 (1988).
Bindal, A., K. L. Wang, S. J. Chang, M. A. Kallel, and P. K. Chu, J Electrochem Soc
X&, p. 2414 (1989).
Bindal, A., K. L. Wang, S. J. Chang, and O. M. Stafsudd, J Electrochem Soc 128. p. 222
(1991).
Brierly, S. K, T. E. Anderson, and A. K. Grabinski, Conference on SI III-V Compounds,
p. 21 (1988).
Chiang, S. Y, and G. L. Pearson, JAP 4£, p. 2986 (1975).
Chichibu, S., N. Ohkubo, and S. Matsumoto, JAP £4, p. 3987 (1988).
222
Chin, A. K., I. Camlibel, R. Caruso, M. S. S. Young, and A. R. von Neida, JAP 52, p.
2203 (1985).
Christel, L.A. and J. F. Gibbons, JAP 52, p. 5050 (1981).
de Souza, J. P., D. K. Sadana, and H. J. Hovel, MRS 1&, p. 495 (1989).
de Souza, J. P., D. K. Sadana, H. Baratte, and F. Cardone, APL5J, p. 1129 (1990).
Fang, Z-Q., H. Yamamoto, and D. C. Look, to be published (1991).
Ghandhi, S. K., The Theory and Practice of Microelectronics. Krieger, 1968.
Heimlich, M. C , R. J. Gutmann, D. Seielstad, and D. Hou, Conference on SI III-V
Compounds, p. 367 (1990).
Itoh, T., and M. Nomura, Jap JAP, p. L389 (1983).
Ladbrooke, P. H., MMIC Design: GaAs FETs and HEMTs. Artech House, 1989.
Lagowski, J., and H. C. Gatos, 13th Int'l Conf on Defects in Semiconductors, p. 73
(1984).
Kanber, H., Nuclear Instruments and Methods B24/25. p. 575 (1987).
Look, D. C, Conference on SI III-V Compounds, p. 1 (1988).
Min, S-K., E, K Kim, and H. Y. Cho, JAP £3_, p. 4422 (1988).
Morrow, R. A., J Mater Res 2, p. 681 (1987a).
Morrow, R. A., JAP 61, p. 3671 (1987b).
Morrow, R. A., Conference on SI III-V Compounds, p. 65 (1988a).
Morrow, R. A., JAP 64, p. 1889 (1988b).
Morrow, R. A., APL 55, p. 2523 (1989).
Pearton, S. J., J. M. Poate, F. Sette, J. M. Gibson, D. C. Jacobson, and J. S. Williams.
Nuclear Instruments and Methods B19/20. p. 369 (1987).
223
Sato, T., K. Terashima, S. Ozawa, M. Nakajima, T. Fukuda, and K. Ishida, GaAs IC
Symp 1984, p. 53.
Saito, Y, JAP £5, p. 846 (1989).
Suezawa, M., and K. Sumino, Jap JAP, p. LI 8 (1987).
Swaminathan, V, R. Caruso, and S. J. Pearton, JAP £2, p. 2164 (1988).
\bn Neida, A. R., S. J. Pearton, M. Stavola, and R. Caruso, APL 42, p. 1708 (1986).
224
Appendices
A. PIMR Background Interference Removal Scheme
The PIMR system operates in a relatively intense area of electromagnetic interference
(EMI). EMI arises from sources outside, as well as internal to PIMR. The PIMR
contributions to EMI include poorly grounded pulse generators and inadequately shielded
leads and electronics. Other experiments in the area also add to the EMI. All of these are
picked up by the microwave detection system and introduce an error in the measurement.
Fortunately, the disturbances are fairly periodic. Because of this, and some of the
features of the digital oscilloscope, these sources of measurement error are removed before
data acquisition by the computer. The procedure is relatively straightforward and involves
using the oscilloscope to sample the time-periodic EMI with the MR portion of the PIMR
system (no pulsed lasers). This "error" signal is than stored in the memory of the
oscilloscope. During an actual PIMR measurement, the error signal is subtracted from the
PIMR signal received at the detector to form a "noise-reduced" PIMR signal. After
sufficient averaging to remove random noise, the peak transient of the noise-reduced signal
is measured (with the automated measurement function of the oscilloscope) and reported to
the computer.
225
B. Simulation Software
Bl. BBG Peak Transient Response using Three Defect Levels: C Source Code
/*
This program calculates the BBG peak transient photoconductance
assuming a three-level model for undoped, SI GaAs proposed by M. S. Wang (Wang, M.
S., PhD Thesis, Rensselaer Polytechnic Institute, 1990. The numerical routine is an initialvalue solver takingfrom"Numerical Recipes in C: The Art of Scientific Computing", by
W. H. Press, et al, Cambridge University Press, 1988.
*/
#include
#inchidc
<stdio.h>
<math.h>
#definc MIN(a,b) ((a) < (b) ? (a): (b»
#dcfincABS(a) ((a) < 0.0 ? -(a): (a))
double Nph=9.5el9, CC=4.9e-9, CC2=2e-5, sigma=1.5e-16, sigma2=le-15;
double N_trap=3el5, N_carbon=2el5;
double Ntotal, N2, Ntotal_prev, N2_prev,
double gen, gen2, rec;
long kfact = 0, kprint = 10000;
float k;
double dt = le-12;
FILE *fopen0, *FichOut;
mam()
{
FichOut = fopen("PIMR904.dat", "w");
N2 = 0;
Ntotal = 1 ;
whfle(N trap <= 2.2el6)
{
fprintfJFichOut,"\n\n*
RUN PARAMETERS
*");
fprintf(FichOut,"\nPhotonflux:%c", Nph);
iprintf(FichOut, "\nEL2 +/0 capture cross section: %e", CC);
fprintf(FichOut,"\nEL2 ++/+ capture cross section: %e", CC2);
fprintf(FichOut,H\nEL2 +/0 optical ionization xsection: %e", sigma);
226
fprintfl[FichOui,"\nEL2 ++/0 optical ionization xsection: %e", sigma2);
fprintf^ichOut,"\nEL2 concentration: %c", N_trap);
fj>rintf(FichOut,"\nCarbon concentration: %e", N_carbon);
gen = (qgma*(N_trap - N_carbon) + «gma2*N_trap)*Nph/40e-9;
gen2 = (rigma2*Njrap)*Nph/40e-9;
rec = CC*N_carbon;
kfact = 0;
k=dt,
Ntotal = 0.0;
N2 = 0.0;
Ntotal_prev = 0.0;
N2jwev = 0.0;
while(k <= 40e-9)
{
Ntotal = Ntotal_prev+(gen*k - (rec + CC2*N2_prev)*Ntotal_prev)*dt;
N2 = N2_prev+(gen2*k - CC2*N2_prev*Ntotal_prev)*dt;
Ntotal_prev = Ntotal;
N2_prev = N2;
if (++kfact = kprint)
{
^nintf(FichOut,w\ntime, n, n2: %e, %e, %e", k, Ntotal, N2);
kfact = 0;
)
k = k + dt;
printf("\n%e, %e", N_trap, Ntotal);
N_trap = Njrap + lel5;
>
fclose(FichOiit);
227
B2, ABGPeaK Transient Response; C Source Code
/*
This program calculates the peak transient ABG photoconductance for an undoped,
SI GaAs sample with given carrier lifetime and mobility, surface recombination velocity,
absorption coefficient, depth into the wafer, and spatial gridsize. Many of the subroutines
used are take from "Numerical Recipes in C: The Art of Scientific Computing", by W. H.
Press, et al, Cambridge University Press, 1988.
*/
#include
^include
<stdio.h>
<math.h>
#define MTN(a,b) ((a) < (b) ? (a): (b))
#define ABS(a) ((a) < 0.0 ? -(a): (a))
int n, length;
double *e;
double *f;
double+r,
double al, bl, cl, a2, b2, c2, bn;
double dx, x, tkp, Esum = 0., S;
double ml, tl, m2, temp, xtemp;
double t2,11,12, slope, Nph, alphal, Nal, alpha2, Na2;
double pOl, pll, p21, p02, pl2, p22, p03, pl3, p23;
int i, xchan;
long kfact=0, kprint=100;
longk;
double d t = l e - l l ;
FILE *fopen(), *FichOuf,
double *dvector (nLnh)
intnl,nh;
{
double *v,
v=(double *)malloc((unsigned) (nh-nl+l)*sizeof(double));
if (Iv)
{
printfC^nmemory allocation error in dvectorO");
exit(l);
>
228
return v-nl;
void frcc_dvector(v,nl,nh)
double *v,
intnl,nh;
{
fiee((char*)(v+nl));
>
tridagO
{
nrtj;
double bet, *gam;
gam = dvector(0,n+2);
f[l]=r[l)/(bet=bl);
for(j-2;j<=n;j++)
{
if(j>xchan)
{
if(j<n)
{
gam[|]=c2/bet;
bet=b2-a2*gam[j];
flj] = (r[J]-a2*fD-l]ybct;
>
else
{
gam[j]=c2/bet;
bet=bn-a2*gam[j];
flj] = (r[jl-a2*f[j-l]ybet;
>
>
else
{
gam[j]=cl/bet;
bet=bl-al*gam[j];
flJ) = (rD]-al**lH]ybet;
>
>
for(j=(ii-l);j>=l;j-)
fUl-=g»mD+l]*ftJ+l];
free_dvector(gam,0,n+l);
changc_dt(delt)
double delt;
{
x = 0.0;
11 = sqrt(inl*0.025*tl);
12 = sqrt(in2*0.025*t2);
cl = -11*11;
bl = 2*tl*dx*dx/delt + 2*11*11 + 2*x*dx;
al = cl;
pl3 = -al;
pl2 = 2*tl*dx*dx/delt - 2*11*11 - 2*dx*dx;
pll=pl3;
c2 = -12*12;
b2 = 2*t2*dx*dx/delt + 2*12*12 + 2*dx*dx;
a2 = c2;
p23 = -a2;
p22 = 2*t2*dx*dx/delt - 2*12*12 - 2*dx*dx;
p21= p23;
p01=0;
p02 = 2*tl*dx*dx/delt - 2*11*11 - 2*dx*dx - S*12*12*dx/(ml*0.025);
p03 = 2*11*11;
bn = 2*t2*dx*dx/dett + 2*12*12 + 2*dx*dx - 12*12*exp(-dx/12);
>
main()
printf("Number of spatial subdivisions: ");
scanfC%d", &n);
printf("Length of simulated region (um):");
scanfC%d", &length);
printf(Thotonflux:");
scanf("%r, &Nph);
printfCOptical absorption coefficient region 1:");
scanfC%lf\ &alphal);
230
printfCOptical absorption coefScient region 2:");
scanfCWf, &alpha2);
printf("Surface Recombination Velocity (cm/sec):");
Bcaiif("%ir, &S);
printfCPosition of boundary 1 - 2 (urn):");
scanfl"%lf\ &xtemp);
printfC\nRegion 1 mobility:");
scanf("%ir, Ami);
printf("
lifetime:");
,
scaittT %ir, &tl);
printf("Region 2 mobility:");
s c a i u f W , &m2);
printf("
lifetime:");
w
scanf( %lf\ &t2);
xchan = (int) n * xtemp / length;
FichOut = fopen("PIMR.dat", "w");
tprintf(FichOut,"*
RUN PARAMETERS
•");
fprintf(FichOut,"\nlength (in microns): %d", length);
fprintf(FichOut, "\nnumber of subdivisions: %d", n);
fprintf^FichOu^-^photonflux:%6.1e", Nph);
fprintf(FichOut,''\noptical absorption coefScient 1: %6.1e", alphal);
fprintfCFichOut^optical absorption coefScient 2: %6.1e", alpha2);
fpiintf(FichOut,w\nstart of region 2: %6.1e", xtemp);
fprintf(FichOut,n\nSurface recombination velocity: %2.1e", S);
fprintf(FichOut,"\nRegion 1 lifetime and mobility: %3.2e\t%4.0e",\
tl,ml);
fprintf(FichOut,"\nRegion 2 lifetime and mobility: %3.2e\t%4.0e",\
t2,m2);
prfntfC*
RUN PARAMETERS
*");
printf("\nlength (in microns): %d", length);
printf("\nnumber of subdivisions: %d", n);
printf("\nphotonflux:%6.1e", Nph);
printf("\noptical absoiption coefficient 1: %6.1e", alphal);
printfCVnoptical absorption coefScient 2: %6.1e", alpha2);
printf("\nstart of region 2: %6.1e", xtemp);
printf("\nSurface recombination velocity: %2.1e", S);
printf(M\nRegion 1 lifetime and mobility: %3.2e\t%4.0e",\
tl,ml);
printf("\nRegion 2 lifetime and mobility: %3.2e\t%4.0e",\
t2,m2);
dx = le-4*length/n;
231
Nal = 2*Nph*afohal*tl*dx*dx*dt/40e-9;
Na2 = exp(-temp*alphal*le-4);
Na2 = Na2*2*Nph*alpha2*t2*dx*dx*dtf40e-9;
e=dvcctor(0,n+2);
f=dvcctor(0,n+2);
r=dvector(0,n+2);
for(i=0;i<=n+l;i++)
{
c[i] = 0.;
fM = e[i];
>
change_dt(dt);
k=l;
while (k> 0)
{
x = 2*dx;
ifl]« p01*e[0]+p02*e[l]+p03*e[2]+Nal*exp(-alphal*dx)*k;
for(i=2;i<=n;i++)
{
if (i > xchan)
{
r[i] = p21*e[i-l]+p22*e[i]+p23*e[i+l]+Na2*exp(-x*alpha2)*k;
>
else
{
temp = Nal*exp(-x*alphal)*k;
rfj] = pll*ep-l]+pl2*e[i]+pl3*e[i+l)+temp;
>
x = x + dx;
}
tridagO;
f[0] = f[2]+e[2]-(e[l]+fll])*S*dx/(0.025*ml)-e[0];
if (f[n-l] > 0.)
{
temp = f[n]/f[n-l];
if (temp <= 0.0)
<
printf("\n ABEND - log error at infinite boundary, %8.4e, temp");
exit(0);
}
slope = log(f[n]/f[n-l]);
232
ftn+1] = fln-l]*exp(2*dx/slopc);
bn = 2*t2*dx*dx/dt + 2 + 2*dx*dx - 12«12*exp(dx/glope);
}
for(i=0;i<=n+l;i++)
if (++kfact = kprint)
{
Esum = 0.0;
for(i=0;i<=n;i-H-)
{
if(i>n)
{
Esum = Esum + f[i]*ml*dx;
>
else
{
Esum = Esum + f[i]*m2*dx;
>
tkp = k*dt;
fprintf(FichOut,"\ntmie, mu*n: %6.2e, %4.3e", tkp, Esum);
printf("\ntime, mu*n: %2.1e, %4.3e", tkp, Esum);
i=0;
x=0.0;
kfact = 0;
if(tkp>39e-9)
{
k=-l;
fprintf(FichOut,"\n");
for(i=0;i<=50;i=i+2)
fprintf(FichOut,"\n %d\t %8.4e\t %d\t %8.4e",i,e[i],i+l,e[i+l]);
>
>
k++;
>
ftee_dvector(e,0,n+2);
free_dvector(f,0,n+2);
free_dvector(r,0,n+2);
fclose(FichOut);
>
233
C. Processing Procedures
CI. GaAs Wafer Cleaning and Etching
•As-grown wafer clean and general wafer etch (200 A/min).
-prepare a 1:1:500 mixture of I ^ O ^ N f y O H ^ O at +25 #C.
-immerse wafer in this mixture for 1 minute,
-rinse with DI water and blow dry with dry N2.
•Processed wafer clean.
-immerse wafer in for 1 minute in 100 *C TCE.
-rinse with IPA.
-blow dry with dry N2.
C2. Silox and Nitride Deposition
•Silox (RPICIE - Applied Materials)
-place samples in silox reactor,
-set N2 flow to 14.0 ccm.
-allowreactionchamber to reach 425 "C (8-10 minutes),
-turn off N2, turn on silane and O2.
-allow deposition to run for 7 minutes (=3000 A),
-turn off silane and flow O2 for 3 minutes.
-turn off N2 and flow O2 for 3 minutes.
-remove samples from reactor and let cool.
•Nitride (M/A-COM ASD)
-Nitride was deposited by personnel at M/A-COM ASD, Lowell MA
using a PEDCVD nitride reactor at 250 °C for 13.5 minutes
which produced a 2535 A film.
234
C3. RTA (RPICIE - AG Associates HeatPulse)
-clean all samples to be annealed using the appropriate procedure
above,
-turn on RTA and computer,
-make sure dry N2 for annealing chamber is flowing at maximum
flow rate.
-run RTA control program.
-select warmup procedurefromRTA program menu and run (to precondition oven before use).
-allow RTA to cool to 175 *C.
-open RTA door and place sample on quartz support pins.
-close door.
-select annealing time and annealing temperature using the RTA
program edit mode.
-run the RTA program.
-allow RTA to cool to 175 *C.
-remove sample and repeat
-prior to shutting down RTA, allow to cool to room temperature.
C2. RIESilox and Nitride Strip
•Silox (RPI CIE - Technics)
-place samples in reactor.
-pump reactor down.
-set flow for 300 mTorr composed of 90% CF4 and 10% 0 2 .
-Turn on RF power and set to 150 W.
-etch rate is 150 A/min for a total time of =20 minutes.
-turn off RF power, CF4, and O2.
-flow N2 for several minutes to clear reactor.
-vent reactor and removal samples.
•Nitride (RPI CIE - Technics)
-same procedure as above except 300 mTorr is composed of 41
SCCM of N 2 with 4% 0 2 at 250 mW RF for 3 1/2 minutes.
235
Документ
Категория
Без категории
Просмотров
0
Размер файла
9 075 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа