Graded injector, wide bandgap light emit

Graded injector, wide bandgap light emitters and XPS studies of the InAs/GaSb heterointerface - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Graded injector, wide bandgap light emitters and XPS studies of the InAs/GaSb heterointerface
Citation
Wang, Michael Wei-Ching
(1995)
Graded injector, wide bandgap light emitters and XPS studies of the InAs/GaSb heterointerface.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/653J-FX31.
Abstract
NOTE: Text or symbols not renderable in plain ASCII are indicated by [...]. Abstract is included in .pdf document.

This thesis consists of two parts: research performed towards the development of
a II-VI graded electron injector wide bandgap light emitter and characterization
of the InAs/GaSb heterointerface, for application to InAs/[...] infrared
superlattice detectors. Background and motivation for both projects are presented.

Investigations of the II-VI graded electron injector light emitter consist of simulations to test the feasibility of a number of LED designs, band alignment measurements, materials characterization and device performance studies. The simulations, based on the Drift-Diffusion model with modifications to account for heterojunctions, demonstrate that the graded electron injector LED design is feasible.
Improvements to the basic design can be implemented by incorporating confining
layers and active layers with wider bandgaps using ternaries and quaternaries. The
XPS measurements of the MgSe/[...] and MgTe/[...] valence
band offsets show a deviation from the common anion rule with the valence band
edge of the Mg-based semiconductor lower in both cases. The significance of these
results is discussed. XRD, TEM and SIMS characterization of the graded injector
devices reveal potential problems in material quality. Device performance characterization show good current-voltage and electroluminescence properties, but poor
external quantum efficiency and device lifetimes.

Characterization of the InAs/GaSb heterointerface consists of surface exchange
reaction studies, band alignment measurements and interface abruptness studies.
The XPS/RHEED investigation of surface exchange reactions shows monolayer
exchange and Sb island formation for the Sb soaks of InAs surfaces, and As exchange with Sb past the terminating monolayer of Sb into the underlying GaSb
for the As soaks of GaSb surfaces. The XPS band alignment studies show that
the InAs/GaSb valence band offset is independent of interface composition, but
changes with growth direction. Possible mechanisms for this behavior are discussed. Finally, the XPS, RHEED, cross-sectional STM and SIMS studies of the
abruptness of the InAs/GaSb interface show that the GaSb-on-InAs growth direction is more abrupt than the InAs-on-GaSb growth direction. Mechanisms for this
asymmetry in the interface abruptness are presented.
Item Type:
Thesis (Dissertation (Ph.D.))
Degree Grantor:
California Institute of Technology
Division:
Engineering and Applied Science
Major Option:
Applied Physics
Thesis Availability:
Public (worldwide access)
Research Advisor(s):
McGill, Thomas C.
Thesis Committee:
Unknown, Unknown
Defense Date:
23 August 1994
Record Number:
CaltechETD:etd-10262007-091414
Persistent URL:
DOI:
10.7907/653J-FX31
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
4266
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
07 Nov 2007
Last Modified:
21 Dec 2019 02:01
Thesis Files
Preview
PDF (Wang_mwc_1995.pdf)
- Final Version
See Usage Policy.
9MB
Repository Staff Only:
item control page
CaltechTHESIS is powered by
EPrints 3.3
which is developed by the
School of Electronics and Computer Science
at the University of Southampton.
More information and software credits
Graded Injector, Wide Bandgap Light Emitters
and
XPS Studies of the InAs/GaSb Heterointerface

Thesis by
Michael Wei-Ching Wang

In Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy

California Institute of Technology

Pasadena, California
1995

(Submitted August 23, 1994)

ili

Acknowledgements

This thesis would not have been possible without the support and friendship of a
number of people. Without Tom McGill, my advisor, I would not have survived
the first few years at Caltech. His support and faith in my abilities kept me going
at times when I was sure that I would not succeed. The education that he has
provided to me goes well beyond what I have learned in the lab or the classroom.
For that I am in his debt.

Besides being the most crucial person to the operation of the group, Marcia
Hudson has always been able to make me smile. Her charm and friendliness made
frequent visits to her office commonplace (the snacks were nice too). Thanks also
to the adminstrative assistance of Sandy Brooks and Phil Cheetham.

A number of former students have provided considerable guidance to me in my
earlier years in the group. Mike Jackson eased the transition into the group with
his helpful advice. David Chow got me started on my first project. Ed Croke was
just fun to be around. Ed Yu taught me more than can be listed here. Much of
the work presented in this thesis was based on what he has taught me. Besides
guiding me on the II-VI project, Mark Phillips, and his wife Becky, took me on
bike rides in the hills that were . . . truly breathtaking.

More recently, I have enjoyed working closely with Doug Collins and Johanes
Swenberg. During our “two week project” together, Doug has taught me all I know
about II]-V MBE growth and RHEED. His patience in explaining things two or

three times to me, and his expertise in UHV systems have been greatly appreciated.

Working together with Jo on the I-VI project has been a great experience. Despite
the long hours, the deadlines, and the infamous four hour Saturday meetings, I
have thoroughly enjoyed the time we’ve spent together on the II-VI project. His
good taste in restaurants also came in handy.

Another of Tom’s achievements is the quality of senior researchers that he
brings to the group. I have been lucky enough to interact with a few of them.
Dr. Jim McCaldin has provided invaluable guidance during the course of the II-VI
project. I have benifitted greatly from Dr. Ron Grant’s expertise on XPS. Ogden
Marsh has provided support and advice in a number of ways. Drs. Tom Keuch
and Randy Feenstra have shown me how much more there is to learn; my respect
for their abilities continues to grow.

I have greatly enjoyed my interactions with the other members of the group.
I have gone to David Ting for advice on a number of issues over the years, and I
respect his opinion greatly. His patience with my swimming excuses is also appre-
ciated. Yixin Liu has provided me with many interesting stories and reflections on
Chinese culture. He also taught me how painful badminton can be. Besides being
an ever helpful computer guru, Harold Levy’s unique brand of humor and outlook
on life have kept me entertained throughout the years. Ron Marquardt’s good
nature and friendship have been invaluable over the years. In contrast to his ap-
parent conservative reputation within the group, Ron’s partying and womanizing
tendencies outside of the group have amazed me over the years. Rob Miles, more
than anyone else, kept things lively in the group. His comments and queries in the
coffee room will be sorely missed. I have benefited greatly from my interactions
with Per-Olov Pettersson (Peo); his energy and enthusiasm are to be admired.
Chris Springfield has given much of his time helping me with numerous computer
and video related problems. I have also made much use of his ever-ready movie
review capabilities.

I have also enjoyed my interactions with some of the more recent members of

the group; Alicia Alonzo, Xiao-Chang Cheng, Erik Daniel, and Eric Piquette. I

look forward to future interactions with them.

My interactions outside of the group have helped to keep me sane these years
at Caltech. I have enjoyed all of the relaxing times I’ve had with John, Gemma,
and Ron. Thanks.

To Barb, my fiancee, I owe my present happiness. Her love, support and
understanding have changed my life.

Finally, without the unquestioning love and support of my parents, brother and

sister (in-law) I would not be here today.

List of Publications

Work related to this thesis has been, or will be, published under the following
titles:

Effect of Interface Composition and Growth Order on the Mixed
Anion InAs/GaSb Valence Band Offset,

M.W. Wang, D.A. Collins, R.W. Grant, R.M. Feenstra, and T.C. McGill, to
be submitted to Appl. Phys. Lett.

X-ray Photoelectron Spectroscopy Measurement of Valence Band

Offsets in Mg-based Semiconductor Compounds,
M.W. Wang, J.F. Swenberg, M.C. Phillips, E.T. Yu, J.O. McCaldin, R.W.
Grant, and T.C. McGill, Appl. Phys. Lett. 64, 3455-3457 (1994).

Measurement of the MgSe/CdZnSe Valence Band Offset by X-ray
Photoelectron Spectroscopy,

M.W. Wang, J.F. Swenberg, R. J. Miles, M.C. Phillips, E.T. Yu, J.O. Mc-
Caldin, R.W. Grant, and T.C. McGill, J. Cryst. Growth 138, 508-512 (1994).

XPS Investigation of the Mixed Anion GaSb/InAs Heterointerface,
M.W. Wang, D.A. Collins, T.C. McGill, and R.W. Grant, J. Vac. Sci. Tech-
nol. 11, 1418-1422 (1993).

vii

n-CdSe/p-ZnTe Based Wide Bandgap Light Emitters: Numerical
Simulation and Design,

M.W. Wang, M.C. Phillips, J.F. Swenberg, E.T. Yu, J.O. McCaldin and T.C.
McGill, J. Appl. Phys. 73, 4660-4668 (1993).

Study of the Abruptness of the InAs-GaSb Interface,
D.A. Collins, M.W. Wang, R.W. Grant, R.M. Feenstra, and T.C. McGill, to
be submitted to Appl. Phys. Lett.

Scanning Tunneling Microscopy of InAs/GaSb Superlattices: Sub-
bands, Interface Roughness, and Interface Asymmetry,

R.M. Feenstra, D.A. Collins, D.Z.-Y. Ting, M.W. Wang and T.C. McGill, J.
Vac. Sci. Technol. (1994).

Interface Roughness and Asymmetry in InAs/GaSb Superlattices
Studied by Scanning Tunneling Microscopy,

R.M. Feenstra, D.A. Collins, D.Z.-Y. Ting, M.W. Wang and T.C. McGill,
Phys. Rev. Lett. 72, 2749 (1994).

Status of Ga;_,In,Sb/InAs Superlattices for Infrared Detection,
R.H. Miles, D.H. Chow, J. N. Schulman, D.A. Collins, M.W. Wang, R.W.
Grant, and T.C. McGill, submitted to J. Elect. Materials.

RHEED Observation of Exchange Reaction Dynamics of InAs Sur-
faces,

D.A. Collins, M.W. Wang, R.W. Grant, and T.C. McGill, J. Appl. Phys. 75,
259-262 (1994).

Vili

Advances in the Development of Graded Injector Visible Light
Emitters,

J.F. Swenberg, M.C. Phillips, M.W. Wang, R. J. Miles, J.O. McCaldin, and
T.C. McGill, J. Cryst. Growth 138, 692-696 (1994).

Investigation of Crystal Quality and Surface Morphology of ZnTe:N
Epilayers Grown on ZnTe and GaSb Substrates,

R. J. Miles, J.F. Swenberg, M.W. Wang, and T.C. McGill, J. Cryst. Growth
138, 523-528 (1994).

A New Approach to Wide Bandgap Visible Light Emitters,
M.C. Phillips, J.F. Swenberg, M.W. Wang, J.O. McCaldin, and T.C. McGill,
Physica B 185, 485-489 (1993).

Interfacial Reactions and Band Offsets in the AlSb/GaSb/ZnTe
Material System,

E.T. Yu, M.C. Phillips, D.H. Chow, D.A. Collins, M.W. Wang, J.O. Mc-
Caldin, and T.C. McGill, Phys. Rev. B 46, 13379-13388 (1992).

Proposal and Verification of a New Visible Light Emitter Based on
Wide Bandgap II-VI Semiconductors,

M.C. Phillips, M.W. Wang, J.F. Swenberg, J.O. McCaldin, and T.C. McGill,
Appl. Phys. Lett. 61, 1962-1964 (1992).

Forming of Al-doped ZnTe Epilayers Grown by Molecular Beam
Epitaxy,

M.C. Phillips, J.F. Swenberg, Y.X. Liu, M.W. Wang, J.O. McCaldin, and
T.C. McGill, J. Cryst. Growth 117, 1050-1054 (1992).

Proposal for the Formation of a Minority Carrier Injecting Contact
on Wide Bandgap Semiconductors,

Y.X. Liu, M.W. Wang, J.O. McCaldin, and T.C. McGill, J. Cryst. Growth
117, 913-917 (1992).

Schottky Barrier Induced Injecting Contact on Wide Band Gap
Semiconductors,

Y.X. Liu, M.W. Wang, J.O. McCaldin, and T.C. McGill, J. Vac. Sci. Technol.
B 10, 2072-2076 (1992).

Abstract

This thesis consists of two parts: research performed towards the development of
a II-VI graded electron injector wide bandgap light emitter and characterization
of the InAs/GaSb heterointerface, for application to InAs/In,Gaj_,Sb infrared
superlattice detectors. Background and motivation for both projects are presented.

Investigations of the II-VI graded electron injector light emitter consist of sim-
ulations to test the feasibility of a number of LED designs, band alignment mea-
surements, materials characterization and device performance studies. The simu-
lations, based on the Drift-Diffusion model with modifications to account for het-
erojunctions, demonstrate that the graded electron injector LED design is feasible.
Improvements to the basic design can be implemented by incorporating confining
layers and active layers with wider bandgaps using ternaries and quaternaries. The
XPS measurements of the MgSe/Cdo.54Zng.465e and MgTe/Cdo sgZno.12Te valence
band offsets show a deviation from the common anion rule with the valence band
edge of the Mg-based semiconductor lower in both cases. The significance of these
results is discussed. XRD, TEM and SIMS characterization of the graded injector
devices reveal potential problems in material quality. Device performance charac-
terization show good current-voltage and electroluminescence properties, but poor
external quantum efficiency and device lifetimes.

Characterization of the InAs/GaSb heterointerface consists of surface exchange
reaction studies, band alignment measurements and interface abruptness studies.

The XPS/RHEED investigation of surface exchange reactions shows monolayer

exchange and Sb island formation for the Sb soaks of InAs surfaces, and As ex-
change with Sb past the terminating monolayer of Sb into the underlying GaSb
for the As soaks of GaSb surfaces. The XPS band alignment studies show that
the InAs/GaSb valence band offset is independent of interface composition, but
changes with growth direction. Possible mechanisms for this behavior are dis-
cussed. Finally, the XPS, RHEED, cross-sectional STM and SIMS studies of the
abruptness of the InAs/GaSb interface show that the GaSb-on-InAs growth direc-
tion is more abrupt than the InAs-on-GaSb growth direction. Mechanisms for this

asymmetry in the interface abruptness are presented.

xii

Contents

Acknowledgements

List of Publications

Abstract

List of Figures

List of Tables

Glossary of Acronyms

1 Introduction

1.1

1.2

1.3

Introduction to Thesis .. 2... 2... . eee ee ee
1.1.1 General Motivation... ...... 0.2... 00000008.
1.1.2 Summary of Results ............ 02... 00004
Wide Bandgap H-VI Light Emitters... 2... .......0.08.
1.2.1 Motivation... .. 2... ee
1.2.2 Commercially Available LED Products ............
1.2.3 Wide Bandgap LED Design Issues... ............
1.2.4 History and Current Status of Research Efforts .......
1.2.5 Introduction to Graded Electron Injector ...........
The Mixed Anion InAs/GaSb Heterointerface ............
1.3.1 Motivation... . 2... 2. ee ee

iii

XVi

xix

xiii

1.3.2 InAs/In,Gay_,Sb IR SL Detectors .............. 30
1.3.3 Challenges and Issues for Mixed Anion Interfaces ...... 32
1.3.4 Characterization Techniques .................. 35
1.4 Outline of Thesis ........... 0... 0000000 eee eee 39
References... 2... ee Al

I Graded Electron Injector, II-VI Wide Bandgap Light

Emitters 48
2 Light Emitter Simulation and Design 49
2.1 Introduction and Outline... 2.2... 2... ...2..0..0...2.00.4 49
2.2 Design Methodology ................ 00000080 0 50
2.3 Model ........ 56
2.4 Simulation Results .. 2... 2... 2... 2.2 ee ee ee ee 63
2.4.1 CdSe/ZnTe .. 2... ee 63

2.4.2 Graded Injector. .........0.........00000, 67

2.4.3 Graded Injector with Confining Layer............. 72

2.4.4 Tunable Bandgap LED...................0.,. 76

2.5 Recent Developments... ........0..0..0. 000000008 4 81
2.5.1 Current LED Structure... .............2..00. 81

2.0.2 Possible Quaternary LED Designs. ............2.., 82

2.6 Summary ....... 0... 00000 ee ee 85
References... 0. ee 87

3 XPS Measurements of Band Offsets for Mg-based Semiconductor

Compounds 90
3.1 Introduction and Outline... 2... 000.000.000.200. 004. 90
3.2 Experiment ...... 0... .0..0. 0000000000004 eae 92

3.2.1 Sample Growth ...........0.... 0.000000 8 3 92

Xiv

3.2.2 XPS Measurements..................0..004 93

3.3 XPS Data Analysis... ........0..0...000.0-0000,4 94
3.4 Results and Discussion ................0. 00000004 103
3.4.1 MgSe/Cdo54Zno.4gSe Valence Band Offset .......... 103
3.4.2 MgTe/Cdo.8gZmo.12Te Valence Band Offset .......... 104

3.0 Summary ....... 0... 00. ee ee 105
References... 2 107
4 Status of the Graded Electron Injector Project 110
4.1 Introduction and Outline... 2... 20... ...0.00.00004 110
4.2 Material Characterization ..............0...0.00040., 111
421 XRD... 2... ee en 111
42.2 TEM... ... en 113
4.2.3 SIMS...... 0.0.00. 00.00 ee ee ee 115

4.3 Device Performance................ 0000000 cea, 117
4.3.1 Current-Voltage Characteristics ............00.., 117
4.3.2 Electroluminescence ...............0000004 119
4.3.3 External Quantum Efficiency ................. 120
4.3.4 Device Degradation Studies .............2.0.0.., 123

4.4 Summary ....... 0... 0000 e eee eee en 123
References... 2... eee 126

II XPS Studies of the Mixed Anion InAs/GaSb

Heterointerface 127
5 Surface Exchange Reaction Studies 128
5.1 Introduction and Outline... ......00..0002.0....00., 128
5.2 Experiment .............0.0 020.0000 000 ee eee 130

5.2.1 Sample Growth ...........0.......0.0..0040. 130

5.2.2. XPS and RHEED Measurements ............... 130
5.38 Data Analysis... 2.0... ee ee 131
5.4 Results... 0... ee 135
5.4.1 Sb Soaks of InAs Surfaces ................04. 135
5.4.2 As Soaks of GaSb Surfaces... ............200. 139
5.5 Discussion... ... 141
5.5.1 Sb Soaks of InAs Surfaces ..............2.-.004.4 141
5.5.2 As Soaks of GaSb Surfaces... ............000. 144
5.6 Summary .. 2... 2... ee 146
References... 2. 149
InAs/GaSb Heterojunction Studies 151
6.1 Introduction and Outline... 2... eee en 151
6.2 InAs/GaSb Band Alignment Studies ................. 152
6.2.1 Motivation... 2... 0... 2... ee ee en 152
6.2.2 Experiment ............0. 000000. eeeeey 153
6.2.3 Results and Discussion... ............2.20004 156
6.3 Abruptness of the InAs/GaSb Heterointerface ............ 161
6.3.1 Motivation... 2... 0... ee ee en 161
6.3.2 Experiment ...........0 00.00.00 00000004 162
6.3.3 Observation of Interfacial Asymmetry. ............ 165
6.3.4 Underlying Causes of Interfacial Asymmetry ......... 172
6.4 Summary ...... 0... 000. ee ee 178

References... 0.0.0.0. ke 181

xvi

List of Figures

1.1

1.2
1.3
1.4
1.5

1.6
1.7
1.8

1.9
1.10
1.11

2.1
2.2
2.3
2.4
2.9
2.6

External quantum efficiency of existing commercial technologies and

spectral eye response .. 1... ee 9
Luminous efficiency of existing commercial LED technologies .... 10
p-n homojunction LED operation ...........0.... 0004 13
Use of McCaldin diagrams in LED design. .............. 15
External quantum efficiency of commercial and developing LED

technologies ©... we ee 24
Schematic of graded injector device .. 2... oe ee 27
Band edges for InAs/GaSb/AISb and InAs/GaInSb ......... 29
Band edges and carrier probability densities for InAs/GaSb and

InAs/GalnSb superlattices... es 33
Interface composition and growth order in InAs/GaSb heterojunctions 34
Schematic of X-ray photoelectron spectroscopy. ........... 37
Schematic of reflection high energy electron diffraction setup .... 38
McCaldin diagram for common semiconductors ........... 51
McCaldin diagram for common II-VI semiconductors ........ 53
Flatband diagram of graded injector device. ............. 55
Bandgap versus lattice constant . 2... 2. ee ee 57
CdSe/ZnTe: energy band diagrams ...............00. 64

CdSe/ZnTe: charge distributions ................4.0. 65

2.7

2.8

2.9

2.10
2.11
2.12
2.13
2.14
2.15
2.16
2.17
2.18

2.19

3.1
3.2
3.3
3.4
3.9

3.6
3.7

4.1
4.2
4.3
4.4
4.5

xvii

CdSe/ZnTe: current densities ©... 0.0... .... 0... ..0040. 66
Graded device: energy band diagrams................. 69
Graded device: charge distribution .................. 70
Graded device: current densities... ..........-.22000. 71
Graded device with confining layer: energy band diagrams .... . 73
Graded device with confining layer: charge distributions ...... 74
Graded device with confining layer: current densities ........ 75
Flatband diagram of tunable bandgap LED ............. 77
Tunable bandgap LED: energy band diagrams ............ 78
Tunable bandgap LED: charge distributions ............. 79
Tunable bandgap LED: current densities ............... 80

Bandgap versus lattice parameter for Zn,_,Mg,SeyTe;_, and
Cd,_,Mg,SeyTe;_y quaternaries ... 2... 0... 0... ..000. 83

Schematic diagram of fully lattice matched graded injector structure 85

XPS spectra for MgSe/Cdo.54Zno.4gSe band offset measurement... 95
XPS spectra for MgTe/Cdo,sgZno.12Te band offset measurement .. 96
Fitting of Zn 3p 3/2 and Mg 2s core level peaks ........... 97
Fitting of Cdo54Znp.4gse and MgSe VBDOS ............. 100

Cdo.gZNo.12Te and MgTe VBM separation as function of fitting in-
terval. 2. 101
Cdo.54Zng.4g9e and MgSe VBM separation as function of fitting interval 102
Measured MgSe/Cdo54Znp.4g5e and MgTe/Cdo.sgZng.12Te valence

band offsets 2... ee 103
XRD data for graded electron injector device... .......... 112
TEM image of graded electron injector device ............ 114
SIMS impurity profiles ..............0.000 000004 116
Current-voltage characteristics... 2 ee ee 118

Electroluminescence data... ......0...0 0000 eee ene 120

xviii

4.6 Light extraction and contacting issues... ............-..

4.7 Device lifetime studies .............0.....00. 0000048

5.1 XPS spectra: Sb2 soak of InAs surface, and InAs-on-GaSb hetero-
junction . 2... ee ee
5.2 Sb 4d core level with two chemically shifted peaks ..........
5.3 Effect of cracker power on exchange reaction rates ..........
5.4 Sb-As exchange as a function of soak time and cracker power... .
5.5 RHEED/XPS studies of Sb soaks of InAs surfaces. .........
5.6 Ase soak of GaSb surface as a function of soak time .........
5.7 Comparison of chemically shifted Sb 3d and Sb 4d core levels... .
5.8 Other XPS peak intensity ratios for As soaks of GaSb surfaces .. .

5.9 Schematic diagram of surface exchange reactions. ..........

6.1 XPS measurements for InAs/GaSb valence band offset .......
6.2 Fits for As 3d, Sb 4d, Ga 3d and In 4d core level peaks .......
6.3 Schematic diagram of cross-sectional STM setup...........
6.4 Schematic diagram of SIMS setup ................00.
6.5 RHEED reconstruction and intensity oscillation studies of interface
asymmetry. 2. eee
6.6 Cross-sectional STM images of the InAs/GaSb SLs .........
6.7 XPS/ion sputtering studies. ............00.0.0.00000.
6.8 XPS observation of Sb riding on InAs growth front. .........
6.9 SIMS analysis of InAs-on-GaSb interface abruptness.........
6.10 Schematic diagram of asymmetry in the InAs/GaSb interface... .

175

Xix

List of Tables

1.1

6.1
6.2
6.3

Recent progress of ZnSe-based and GaN-based LEDs ........ 23

Core level energy separations as a function of interface composition 158
Core level energy separations as a function of growth order ..... 159

XPS peak intensity ratios ............0.0..-.-.00040, 166

Glossary of Acronyms

CW
FWHM
HBT
HH

ITO

LED
LEEBI
LH
MBE
MOCVD
MOMBE
MOVPE
OMVPE
PL

Qw
QWIP
RHEED

compact disc

continuous wave

full width at half maximum

heterojunction bipolar transistor

heavy hole

infrared

indium tin oxide

laser diode

light-emitting diode

low energy electron beam irradiation

light hole

molecular-beam epitaxy

(=OMVPE=MOVPE) metalorganic chemical vapor deposition
metalorganic molecular-beam epitaxy
(=MOCVD=OMVPE) metalorganic vapor-phase epitaxy
(=MOCVD=MOVPE) organometallic vapor-phase epitaxy
photoluminescence

quantum well

quantum well infrared photodetector

reflection high energy electron diffraction

SCH
SIMS
SL
SRH
STM
TEM
UHV
XPS
XRD

separate confinement heterostructure
secondary ion mass spectrometry
superlattice

Shockley-Read-Hall

scanning tunneling microscopy
transmission electron microscopy
ultra-high vacuum

X-ray photoemission spectroscopy

X-ray diffractometry

Chapter 1

Introduction

1.1 Introduction to Thesis

This thesis consists of two parts. Part I describes research performed towards
the development of a new II-VI wide bandgap light emitter design. Part II
presents characterization of the InAs/GaSb heterointerface, for application to

InAs/In,Ga1_,Sb infrared superlattice (IR SL) detectors.

1.1.1 General Motivation

Recent advances in semiconductor growth capabilities have stimulated interest in
the development of a wide variety of innovative semiconductor heterostructure
devices. In some cases, these device applications rely on the recently developed
ability to grow semiconductor layers with extremely small length scales and near
monolayer control of growth thicknesses. At these length scales, structures can
be fabricated which make use of quantum mechanical phenomena such as quan-
tization of states, resonant tunneling through closely spaced, thin barriers, and
wavefunction penetration into barriers. Devices which exploit these phenomena
include laser diodes, a variety of inter- and intra-band tunneling structures, and

various structures which implement superlattices. Other devices currently under

development rely more heavily on the ability to grow semiconductor materials
which previously could not be grown at all, or could not be grown with the desired
characteristics. II-VI and nitride semiconductor compounds fall into this category.

In this thesis, various aspects of two semiconductor heterojunction device ap-
plications are studied: specifically, the design, fabrication and characterization
of novel II-VI wide bandgap light emitters and investigation of the mixed anion
InAs/GaSb interface used in IR SL detectors. The II-VI wide bandgap light emit-
ter makes use of the recently developed capability to grow certain IJ-VI compounds
with the necessary material quality and dopant levels, while development of the
InAs/In,Ga,_,Sb IR SL detectors relies on the ability to grow very thin (< 100 A)
alternating semiconductor layers with the desired interfacial properties.

The II-VI wide bandgap light emitter design studied in this thesis makes use
of an innovative graded electron injector design [1] to overcome fundamental dif-
ficulties inherent to II-VI light emitter fabrication. The originality of this light
emitting diode (LED) design made it necessary to investigate various design pa-
rameters in order to obtain and then optimize proper LED device performance.
Additionally, the device incorporates semiconductor materials that are relatively
uncharacterized, hence characterization of a number of material parameters was
required to ensure optimal performance of the LED.

The InAs/GaSb interface is of interest not only because of its application to
InAs/In,Ga ,_,Sb IR SL detectors and other device technologies, but also because
studies yield insight into the physics of mixed anion interface formation. Since
both the anions (As and Sb) and the cations (In and Ga) change across the
InAs/In,Ga,_,Sb interface, control of the interfaces in this system is difficult.
Add to this the differing bond lengths, bond strengths, and surface free energies in
this system, and control of the interfaces becomes even more complicated. Since
the interfaces comprise a significant fraction of the layers in the IR SL detector, it

is critical to have strict control over both their abruptness and their composition.

1.1.2 Summary of Results

In the first part of this thesis (Chapters 2-4) the simulation, characterization, and
fabrication of a graded electron injector H-VI wide bandgap light emitter are dis-
cussed. The simulations, based on the Drift-Diffusion model with modifications
to account for the effects of heterojunctions, are used to test for LED design fea-
sibility. Results for a variety of devices are presented to demonstrate the design
procedure and to explain in detail the operation of the graded electron injector de-
vice. The first device is an n-CdSe/p-ZnTe heterojunction LED, which simulations
show does not function correctly as an LED due to high levels of interfacial recom-
bination and carrier injection into the narrow bandgap CdSe layer. The second
simulated device contains an added graded electron injector layer (Mg,Cd4_,Se)
between the CdSe and the ZnTe layers. In this case, simulations demonstrate that
the device should function correctly as an LED, with electron injection into the
wide bandgap ZnTe layer and minimal interfacial recombination. The next two
device designs are modified from the standard graded injector device to increase
electron confinement, and to shorten the wavelength of emitted light. Finally, the
most recent device structure, and a proposed design which allows for a completely
lattice matched graded electron injector structure are presented.

Some of the more important parameters affecting LED device operation are the
band alignments between the various semiconductor layers used in the LED; how-
ever, these parameters are essentially unknown for the the Mg-based semiconductor
compounds used in the graded electron injector device. A measurement of the band
alignment for two Mg-based semiconductor heterointerfaces was performed using
in situ X-ray photoemission spectroscopy (XPS) analysis. Specifically, the two het-
erojunction interfaces were MgSe/Cdo54Zno.4gSe and MgTe/Cdo,sgZno,42Te, both
of which are lattice-matched and share a common anion. The measured valence
band offsets are 0.56 + 0.07 eV and 0.43 + 0.11 eV for the MgSe/Cdo.54Zino 46Se
and MgTe/Cdo sgZno.12Te heterojunctions respectively, with the valence band of

the Mg-based semiconductor lower in each case. These results are significant not
only because band alignments are important for device applications but also be-
cause the results deviate from the well known common anion rule. A possible
explanation for the cause of this deviation is presented.

In addition to band offset measurements, other materials characterization stud-
ies are discussed, as well as the latest device performance results. X-ray diffrac-
tometry (XRD), transmission electron microscopy (TEM) and secondary ion mass
spectrometry (SIMS) studies of graded electron injector devices are presented. The
XRD and TEM studies reveal materials quality problems in the devices, attributed
in part to the strain induced in the highly lattice mismatched graded region. The
SIMS analysis also brings to light potential problems in the form of increased im-
purity levels in the neighborhood of the graded region. The latest current-voltage
and electroluminescence studies of the graded electron device show excellent elec-
trical characteristics and green luminescence. The best external quantum efficiency
exhibited to date by our devices is 0.007%. Device degradation studies show evi-
dence for long lifetimes, but our recent devices show lifetimes on the order of only
100s of hours.

The second part of the thesis presents studies of the mixed anion InAs/GaSb
heterointerface. Surface exchange reaction studies are made during both Sb in-
terrupts (or soaks) of InAs surfaces, and As soaks of GaSb surfaces as a function
of soak time and soak species. XPS analysis of the Sb on InAs exchange reaction
shows that Sb exchanges for the terminating As layer with some Sb island forma-
tion. For the As on GaSb exchange reaction, it was observed that As exchanges not
with only the terminating Sb layer, but also exchanges into the underlying GaSb
layer. The mechanism responsible for the difference between the two different ex-
change reactions is attributed primarily to the interplay between the different bond
strengths and the steady state fluxes involved in each exchange reaction. The im-
pact of the difference in soak reaction behavior on subsequent interface abruptness

is discussed.

Finally, XPS measurements of the effect of interface composition and growth
order on the InAs/GaSb valence band offset and investigations of asymmetry in
the abruptness of the InAs/GaSb interface are presented. The InAs/GaSb va-
lence band offset was found to be independent of changes in interface composition,
but not independent of changes in growth order. Possible mechanisms for this
effect are discussed. XPS, reflection high energy electron diffraction (RHEED),
cross-sectional scanning tunneling microscopy (STM) and SIMS investigations of
the InAs/GaSb heterojunction show that interfaces for the InAs-on-GaSb growth
direction are more extended than for the GaSb-on-InAs growth direction. These
analysis techniques provide near-surface chemical information, real-time structural
information, high spatial resolution imaging and spectroscopy and high sensitivity,

low depth resolution chemical information respectively.

1.2 Wide Bandgap II-VI Light Emitters

1.2.1 Motivation

The development of short wavelength light emitting diodes (LEDs) and laser diodes
(LDs) is of significant commercial and technological interest. To better understand
why this is so, it is helpful to have some background on the advantages, disadvan-
tages, applications, and markets of currently available LEDs and LDs.

LEDs have a number of clear performance advantages over conventional in-
candescent lamps. The features that immediately come to mind are efficiency,
reliability, fast switching speeds, and narrow linewidths. Typical efficiencies are
roughly 2% for incandescent bulbs compared to 18% for the best high brightness
commercial red LEDs [2]. Reliability is another area where LEDs have the advan-
tage with lifetimes as high as ~ 10° hours compared to the ~ 104 hours typical for
incandescent lamps. Another advantage of LEDs is the fast switching speed attain-

able, especially when compared to fluorescent lamps. Finally, the narrow linewidth

can also be a very desirable quality in many applications where a specific color of
emitted light is desired.

Currently, the visible LED market is roughly 20 billion units per year for a total
market value of roughly $1 billion per year [3], with the primary application being
indicator lamps. Should high brightness, high efficiency green and blue LEDs be-
come available, they will increase the size of this market considerably, even without
accounting for new applications. There are many uses for bright green or blue indi-
cator lamps (e.g., outdoor settings or automobile displays) where ambient lighting
requires high brightness LEDs, and where colors other than red and amber are
desired. The main markets sought after by developers of high efficiency green and
blue LEDs, however, are those where the development of these LEDs is expected
to open up potentially very large markets that currently do not even exist. As an
example of this, the introduction of the ultra-bright, high efficiency red AlGaAs
LEDs found an immediate, very large market in automobile brake lights which
could not have been accessed with lower efficiency, lower brightness LEDs.

Two potentially large markets are in the areas of full color displays and traffic
signal lights. Full color displays based on LEDs would have the advantages of
being flat, bright, efficient and scalable to large sizes. The market for this is very
large, given the demand for full color, flat panel displays for portable computers,
and the potential demand for large area flat panel displays to replace conventional
CRI TVs and computer screens. Red and green LEDs also have a large potential
market in traffic signal lights. Because lamps used for traffic signal lights must
have reasonably fast switching speeds, standard low efficiency incandescent lamps
must be used instead of higher efficiency fluorescent lamps. A simple analysis
shows that by replacing existing lamps with high efficiency red and green LEDs,
an energy savings of over $100 per year per traffic signal head can be achieved [4].
This does not include reduced maintenance and replacement costs associated with
the longer LED lifetimes. The energy savings alone would pay for the cost of the
LED lamps years before they would need to be replaced. Multiply the savings per

lamp by the number of lamps in a typical city [5], and one can see why LEDs are
a very attractive light source for traffic signal lights.

It is estimated that the current market for flat panel displays is roughly $5
billion, projected to reach $40 billion by the year 2000 [6]. To replace all of the
traffic signal lights in the United States would require an estimated 4 billion each of
discrete red, green and amber LEDs [7]. Though these numbers are approximate,
it is clear that enormous markets for high brightness, high efficiency blue and green
LEDs exist if only they can be fabricated.

The market for LDs is much smaller than that for LEDs, with 1992 sales of
only roughly $287 million/year at an average cost of just under $6/unit [8]. The
motivating force for companies such as Sony, which are leading the development of
ZnSe-based LDs, is actually the market for systems utilizing the LDs. For instance,
although the market for LDs used in compact disc (CD) players and laser printers
is small, the market value for CDs, CD players, and laser printers that incorporate
these LDs is enormous. Thus, the motivation for producing LDs is not to produce
and sell LDs, but to increase sales of systems such as CD players, which have
a much larger market value, by providing unique features unattainable by other
vendors. The unique feature provided by shortening the wavelength of lasing to
the blue for CD players is that the storage density in optical recording can be
increased by factors of four or more due to the reduced bit size attainable with
shorter wavelengths. With current video compression techniques, this would allow
storage of full length movies on standard sized CDs. For laser printers, shorter
wavelength LDs would increase both resolution and throughput. Thus, despite the
smaller dollar value of the LD market compared to the LED market, there is still

very good reason to develop shorter wavelength LDs.

1.2.2 Commercially Available LED Products

In order to properly understand the motivation for the tremendous efforts invested
in wide bandgap light emitter research, it is important to evaluate the perfor-
mance of existing commercial LED technologies. Commercial LDs in the blue
and green wavelength regimes simply do not exist, so the discussion will center
on commercial LED technologies. Figures 1.1 and 1.2, published by researchers
from Hewlett-Packard [9] and modified to include SiC, show the performance of
existing commercial LEDs. The AlGalnP LEDs shown are not currently avail-
able, but are expected to be commercialized in the near future. Figure 1.1 shows
both the external quantum efficiency of the LEDs and the relative response of a
light-adapted human eye, as functions of wavelength. The photometric luminous
efficiency, expressed in lumens/watt, accounts for the eye response by combining
the relative eye sensitivity with the external quantum efficiency. Figure 1.2 shows
a plot of the luminous efficiency versus peak wavelength.

In the red wavelength regime, the highest performance LEDs are Al,Ga,_,As-
based double heterostructure LEDs. External quantum efficiencies of 7% at 645
nm and 18% at 650 nm for absorbing GaAs substrates and transparent AlGaAs
substrates, respectively, have been reported [2]. The luminous efficiency of the
transparent substrate Al,Ga;_,As LEDs is only slightly below that of a tungsten
incandescent lamp and is far superior to a red filtered incandescent lamp.

If we now study the shorter wavelength yellow LEDs, we start to see a drop
in the external quantum efficiency, although this is more than compensated for by
the higher eye sensitivity to the shorter wavelengths. Two technologies are shown
in this wavelength range: GaAs,P,_, on GaP substrates and Al,_,Ino5,Gag5.P
on GaAs substrates. The GaAs,P,_, LEDs perform relatively poorly as seen in
Figures 1.1 and 1.2. This is because GaAs,P _, becomes an indirect semiconductor
at wavelengths less than 627nm. The only reason these devices perform as well

as they do is because they are doped with nitrogen which acts as an isoelectronic

Yellow Red
ue Green a Orange <+——_»
>> >
Sa
AIGaAs/AIGaAs J 1.9
10.0 +
| ; |
AIGaGAs/GaAs
S ; J
& AlGainP/GaAs 0.8
eye Fe
— a od
in 10] 06
ro} ©
53 : in
C6 GaP-N/GaP * x
_& GaAsP:N/GaP 3
fe) . ~
2 0.1 + ‘.
ih e c>— ~=IJ1o2
GaP/GaP ,
_ oy SIC/SIC

460 480 500 520 540 560 580 600 620 640 660 680
Peak Wavelength (nm)

Figure 1.1: External quantum efficiencies of existing commercial LED technologies
as a function of peak wavelength at forward biases of 20 mA [9]. Notation for
labels is active layer/substrate. Where applicable, nitrogen isoelectronic centers
are indicated after the active layer. Dashed line shows the relative eye sensitivity

at the different wavelengths.

Blue
<—————» Green

100

Yellow
<————» Orange <——»
<>

Red

~ AllnGaP/GaAs
oa
ie)
310}
ae
7p)
Cc

‘3
6 fT
Cc
uu e
$ GaP/GaP
fe)
awed

0.1F

SIC/SIC

AIGaAs/AIGaAs

AIGaAs/GaAs

GaP:N/GaP
AA
GaAsP:N/GaP

460 480 500 520 540 560 580 600 620 640 660 680

Peak Wavelength (nm)

< Ruorescent
(40W)

+ — Halogen (30W)
<— Tungsten «OW

«— Red-Fiitered
Tungsten

Figure 1.2: Luminous efficiency of existing commercial LED technologies as a func-

tion of peak wavelength at forward biases of 20 mA [9]. The luminous efficiency

is obtained by applying the spectral response of the human eye to the external

quantum efficiency. The luminous efficiency of conventional fluorescent and incan-

descent lamps are also indicated for comparison purposes.

center, allowing radiative recombination without the involvement of a phonon.

The other technology for yellow LEDs is based on Aly_,Ing5,Gag5,P, grown
by organometallic vapor-phase epitaxy (OMVPE) and lattice matched to GaAs.
Progress in this material system has been substantial since the first orange LEDs
were demonstrated in 1984 [10]: first transparent Al,Ga;_,As window layers [11]
and lattice mismatched GaP window layers [12] were introduced, then the thickness
of the GaP window layer was increased to 45 microns [9], and most recently (not
shown in Figures 1.1 and 1.2) the removal of the GaAs substrate and subsequent
wafer bonding of a transparent GaP substrate was achieved [13]. Although the
most recent of these devices has not yet been commercialized, no degradation in
device reliability has been observed, hence they are expected to be commercialized
soon. The maximum luminous efficiency observed is 41.5 L/W, which is greater
than even unfiltered halogen or tungsten incandescent lamps and is only roughly
a factor of two lower than the luminous efficiency of fluorescent lamps.

If we now look at the green wavelength regime, we start to understand why
so much research has been directed toward developing wide bandgap LEDs. The
performance of both the GaP- and Al,_,Ino5xGao.5.P-based green LEDs is sig-
nificantly worse than that of the yellow LEDs. GaP, as mentioned before, is an
indirect semiconductor, so the green LEDs made from it expectedly have very low
quantum efficiencies. The addition of N to act as an isoelectronic center improves
the efficiency significantly, but also shifts the wavelength into the yellow-green,
leaving its performance below that of Al,_,Ino5,Gao5,P. The Al,_xIno.5,Gag5.P
technology, although somewhat improved over GaP, still has much lower lumi-
nous efficiency than LEDs in the yellow wavelength regime, despite the improved
eye response to green. The dropoff in efficiency for Al,_,Ino5,Gag5,P LEDs is
not entirely understood yet, but is expected to be related to the approach of the
direct-indirect transition at shorter wavelengths [14].

The decrease in performance continues as even shorter-wavelength, blue SiC

LEDs are examined (recently announced, high performance GaN/In,Ga,_,N/GaN

LEDs are discussed in the following section). The low quantum efficiency of
these devices is compounded by the sharp dropoff in the eye response for wave-
lengths shorter than that of green light. The best SiC LEDs today have lumi-
nous efficiencies of roughly 0.03 lumens/watt [15], 1000 times lower than the best
Al,_xIno.5xGaos5xP LEDs. Again, the low efficiency is due to the indirect nature
of SiC. Because of the fundamentally poor radiative recombination efficiency of
indirect bandgap semiconductors, it is unlikely that the SiC, GaP, GaP:N or
Ali_xIno.5xGao5xP technologies will be able to achieve significant improvements
in the efficiencies of green or blue LEDs in the near future. Thus, new materials
which have direct bandgaps at the shorter blue and green wavelengths must be
investigated. The history and current status of these research efforts are discussed

in Section 1.2.4.

1.2.3. Wide Bandgap LED Design Issues

For better understanding of Sections 1.2.4 and 1.2.5, a brief discussion of wide
bandgap LED design issues is given below. In its simplest from, an LED consists
of only a p-n homojunction as shown in Figure 1.3. Figure 1.3(a) shows the
conduction and valence band edges and Fermi levels for the n- and p-type regions
before charge redistribution occurs. Upon charge redistribution, at zero bias, the
band profiles are as shown in Figure 1.3(b). Applying a forward bias to the device
forces electrons into the p-type material, resulting in radiative recombination of
electron-hole pairs, as show in Figure 1.3(c). Similarly, holes forced to the left in
Figure 1.3(c) recombine radiatively with the electrons in the n-type material.
This simple p-n homojunction design is sufficient for narrow bandgap (long
wavelength) LEDs. However, it is much more difficult to implement this design
for a wide bandgap LED, since wide bandgap semiconductors are generally very
difficult to dope both n- and p-type. Until very recently it was not possible to

dope direct wide bandgap semiconductors both n- and p-type to sufficient levels

Figure 1.3: Schematic diagram of the operation of a simple p-n homojunction
LED. The band profiles and Fermi levels are shown prior to and after charge
redistribution in (a) and (b) respectively. At forward bias carrier injection occurs,
resulting in radiative recombination, as shown in (c). Dashed lines denote the hole

and electron Fermi levels.

to afford low resistance ohmic contacts and low series resistance devices. As a
result of the inherent difficulties in amphoteric (both n- and p-type) doping of
wide bandgap semiconductors, it has been necessary to go to more sophisticated
LED structures. Some of these approaches address contacting problems directly
using, for example, highly doped graded regions [16]. Other approaches address
the doping problem by using heterojunctions of easily doped semiconductors.

In either case, we must now examine material parameters that were not of con-
cern when dealing only with p-n homojunctions. The three primary parameters
that need to be assessed are the lattice parameter, band alignment and dopability,
which can be displayed together in a McCaldin diagram [17], shown in Figure 1.4.
In McCaldin diagrams each semiconductor is represented by a vertical line with
length proportional to its bandgap. The vertical position of each line is determined
either by experimental measurements or theoretical predictions of band alignments:
the horizontal position is determined by the lattice parameter. Dopability is in-
dicated by filled triangles (n-type dopable) and filled squares (p-type dopable).
Empty symbols indicate poor dopability. Using McCaldin diagrams then, it is a
straightforward procedure to determine if a combination of semiconductor materi-
als satisfies LED design constraints, at least to first order. These constraints require
that: 1) the LED structure be closely lattice matched, 2) the band alignments of
the semiconductor be amenable to efficient carrier injection into a semiconductor
with the desired bandgap, and 3) both n- and p-type doping be sufficient for ohmic
contacting and low series resistance. For example, an n-ZnSe/p-ZnTe heterojunc-
tion is unlikely to perform well because of the large lattice mismatch between ZnSe
and ZnTe which will result in a high defect density. Similarly, an n-AlSb /p-ZnTe
heterojunction has band alignments that favor carrier injection into the indirect,
narrower bandgap AlSb layer. More detailed usage of McCaldin diagrams will be
given in Chapter 2.

Another design issue that is important for LEDs and critical for LDs is that

of availability of quantum wells (QWs) and confining regions. In LEDs, these can

ee — a —— — —
i A
2 a & ; i:
> 1 Ff Sn
~ | Ge GoSb InSb

Lu 4 InP ZnTe CdTe
BeTe AlAs | 4

AIP
-1L GaP ZnSe ticdse MgTe
CiMgSe
D 2ns O cas
-2 7 Ci Mgs 1
Pa oe oe or er ee [a oe oe eT | ee a er oe | a oe ee a ee a a eo
5.4 5.6 5.8 6 6.2 6.4 6.6

Lattice constant (Angstroms)

Figure 1.4: Each material in a McCaldin diagram is represented by a vertical line
with a square at the bottom, corresponding to the VBE, and a triangle at the
top, corresponding to the CBE. The horizontal position of each line is determined
by the lattice parameter, and the length of each line scales with the bandgap of
the material that it represents. The vertical position of each line is determined by
predicted or experimental band alignments. Open triangles and squares represent
semiconductors that cannot be doped n or p type respectively using conventional

doping techniques.

be used to significantly increase the efficiency of the device, while in LDs they
are essential to achieve lasing. Separate confinement heterostructure (SCH) LDs
require a QW, confining layers and cladding layers, all with the same, or nearly
the same lattice parameter.

Extraction of the emitted light is an issue that will be dealt with in more
detail in Chapter 4. The only point to be made here is that the absorption of
light by semiconductors with bandgaps smaller than the energy of the emitted
light is detrimental to device performance and must be accounted for in the design
evaluation. This is important both for the epilayers and for substrates, since as
will be seen in the next section, the difference between absorbing and transparent
substrates can lead to a significant change in external quantum efficiency.

One other technical issue is the availability of lattice-matched substrates. Grow-
ing a device on a highly mismatched substrate will result in heavily dislocated layers
which, in general, leads to poor device performance. A notable exception to this
is the GaN/In,Ga,_,N/GaN developed by Nichia [18]. Details of the exceptional
performance of these devices despite relatively poor material quality will be given

in Section 1.2.4.

1.2.4 History and Current Status of Research Efforts

In Section 1.2.2, the status of existing commercial LED technologies was described,
and the need for research efforts into new materials systems in order to achieve
high efficiency blue and green LEDs was justified. Progress for two of these efforts
has been very rapid over the past few years, so the history and current status of
both efforts will be described. The first is the ZnSe-based system which had early
successes with the development of LDs and high efficiency blue and green LEDs.
The second effort is based on the GaN/InN/AIN material system, for which recent
results report high efficiency, long lifetime blue and green LEDs.

Znse

The majority of II-VI wide bandgap light emitter research is focussed on the
ZnSe material system. There are a number of features of ZnSe that initially made
it a very favorable material to investigate, including its close lattice match to GaAs
substrates, a bandgap in the blue and favorable band alignments to n-GaAs for
electron injection. The main challenge initially encountered was to achieve p-type
doping. Many different dopants were tried, but only two were even moderately
successful. The first successful p-type doing of molecular-beam epitaxy (MBE)
grown ZnSe was reported by researchers at 3M [19], who used lithium doping to
obtain a maximum hole concentration of 3.6 x 1013 cm~3. Eventually, net acceptor
levels of roughly 1 x 10” cm~? [20] were obtained; however, this was still insufficient
for good ohmic contacting, and the possibility of Li diffusion cast doubt on the
long term suitability of Li as a p-type dopant for ZnSe. In addition, efforts to dope
ZnSe with nitrogen were beginning to achieve some impressive results, so as a result
interest in Li doping waned. By 1990 N-doped p-ZnSe with hole concentrations

greater than 1 x 10’? cm~? was obtained using both metalorganic molecular-beam

epitaxy (MOMBE) [21] and MBE [22, 23]. Since then, despite the efforts of many
research groups around the world, the hole concentrations achieved in ZnSe using
N have saturated at roughly 1 x 10'® cm~? [24], with observation of high levels
of compensation. While these hole concentrations are sufficient for the epilayers
within LED and LD structures, they are insufficient for good ohmic contacting, as
will be discussed below.

Despite the contacting problems, the successful p-type doping of ZnSe led to
the first ever blue-green LD, operating under pulsed current injection at 77 K and
a threshold voltage of 20V [25]. Understandably, research into ZnSe intensified
considerably as a result of this report, and soon groups from around the world
were presenting ZnSe-based LD results. Many publications reported marginal im-
provements in device structure, and/or performance; however, it wasn’t until the

issues of cladding layers, ohmic contacts and obtaining pseudomorphic structures

were addressed that truly significant improvements in device performance were
obtained.

The proposal and characterization of Zn;_,Mg,5,Se;_, quaternary layers as a
suitable cladding layer for ZnSe LDs was first presented by researchers at Sony [26].
The use of the Zn,_,Mg,SySe,;_y cladding layers was advantageous because the
bandgap and lattice constant could be independently varied. Also, the wavelength
of light emission could be more freely adjusted to shorter wavelengths because of
the large bandgap attainable by Zn,;_,Mg,S,Sei;_,. These benefits soon resulted
in the first blue LD, operating under continuous wave (CW) operation at 77 K
[27], the pulsed operation of pseudomorphic blue-green LDs up to temperatures
as high as 394K [28], and the lasing of photopumped blue (464.5 nm) double
heterostructure lasers up to 500 K [29].

The issue of ohmic contacting continued to be a problem in these improved
devices, which still operated at voltages in excess of 10 V. In addition to annealing
schemes, a number of other more sophisticated approaches have been proposed
to improve the contact to the p-type ZnSe. The first of these was proposed by
Schetzina et al. at North Carolina State University [30]. Taking advantage of the
semimetallic nature and large electronegativity of HgSe, they were able to fabricate
LEDs with low turn-on voltages using a metal/HgSe/p-ZnSe contacting scheme.
The 4.4 V operating voltage of the LEDs produced was significantly better than
the 6-10 V typically required at the time. A further improvement, reported in the
same paper, makes use of a metal/HgSe/graded ZnTe;_,Se,/p-ZnSe contacting
scheme.

Shortly thereafter, researchers at Brown/Purdue [16] reported a pseudograded
bandgap contacting scheme using alternating layers of p-ZnTe/p-ZnSe in which
the thickness of the ZnTe layers became progressively narrower and the thickness
of the ZnSe layers, progressively wider as the bulk p-ZnSe was approached. This
technique was successful in producing contact resistances of 2 to 8 x 107? Qcm?,

which is considered acceptable for contacts to LEDs and LDs. The operating

voltage of the LDs using this contacting scheme was reduced considerably to 15-
17 V from approximately 30 V. However, this was still higher than the operating
voltages obtained at that time by other groups which were not using a graded
contacting scheme.

Researchers at Sony also presented an approach using ZnTe/ZnSe multiquan-
tum wells to achieve low contact resistances [31]. Their approach was similar to
that used at Brown/Purdue, except that the thicknesses of the ZnTe QWs were ad-
justed so that, accounting for bandbending, the lowest energy level of holes in each
ZunTe QW was aligned with the top of the ZnSe valence band. By doing this, reso-
nant tunneling through the multiquantum well structure is enhanced over a simple
pseudograded ZnTe/ZnSe structure. However, the contact resistances achieved
using this method were 5 x 107? 0. cm’, higher than those obtained using the pseu-
dograded structure. This effect may have been due to the overall lower doping
levels used in both the ZnTe and ZnSe layers for the multiquantum well structure.

Finally, a Be,Zn,_,TeySe;_y graded bandgap ohmic contacting scheme was
proposed by Philips [32] and Xerox [33]. Although no experimental data was
presented, this approach is theoretically feasible and has all the benefits of the
ZnTe/ZnSe contacting schemes with the additional advantage of maintaining full
lattice match to the device layers throughout the contacting layers.

The improvement in LED performance using the HgSe and ZnTe/ZnSe con-
tacting schemes is shown in Table 1.1. This table shows the progression of LED
results for both the ZnSe-based material system and the GaN-based material sys-
tem. In 1992, researchers at Brown/Purdue reported both blue and blue-green
LEDs [34, 35]. These LEDs did not implement the pseudograded ZnTe/ZnSe con-
tacting scheme as evidenced by their high operating voltages (7-20 V), and had
relatively poor efficiencies. Later, in 1993, green and yellow LEDs implementing
the ZnTe/ZnSe pseudograded contact were reported with operating voltages of
approximately 6 V [36]. At roughly the same time, blue and green LEDs incorpo-
rating the HgSe contact were reported with low operating voltages (roughly 3.7 to

3.8 V) but again having relatively low efficiencies [37].

More recent results by the same group show similar current-voltage character-
istics, but much improved external quantum efficiencies of 5.7 to 6.2 x 107°, at-
tributed to improved growth procedures which reduced the number of non-radiative
centers in the active layers [38]. These efficiencies were much better than current
commercial LEDs; however, the device lifetimes were still inadequate, with even
the best devices degrading by 60-65% after 250 hours. The latest devices from
the same group, also shown in Figure 1.5, use ZnSe instead of GaAs substrates.
Efficiencies as high as 3.2% have been obtained, with lifetimes increasing to over
1000 hours despite higher defect densities than in previous structures [39].

The incorporation of both the pseudograded and the multiquantum well
ZnSe/ZnTe contacting schemes has also resulted in improved performance of LD
structures. Recent results for the pseudograded contact show respectable threshold
voltages and current densities of 4.2 V and 600-1100 A/cm?, respectively for 509
nm LDs. Device lifetimes at room temperature were > 20 seconds [40]. Blue (489.9
nm) and green (523.5 nm) LDs incorporating the multiquantum well ZnSe/ZnTe
contacts show thresholds of < 4 V and 400 A/cm?, and room temperature CW
operation for greater than 9 minutes [41].

The main challenge remaining for both LEDs and LDs is the issue of device
degradation. One study by the researchers at 3M [42] showed that the LED and
LD degradation was caused by the propagation of crystal defects from the vicinity
of pre-existing defects, such as stacking faults. This was shown to lead to the
formation of dark line defects. Another study [43] showed that N-doping of ZnSe
can lead to a higher occurrence of stacking faults. Thus, it may be that the poor
device lifetimes in the ZnSe-based structures are related to the N-doping of the Se
layers, especially considering that highly p-doped ZnSe is very difficult to achieve
and is heavily compensated. Another possibility is that strain and interdiffusion
at the GaAs/ZnSe interface results in increased defect densities which adversely

affect device lifetimes. LED results presented recently [39] support this, and LD

structures grown on ZnSe substrates are currently being investigated to test this
hypothesis [44]. It is also possible that greatly improved lifetimes will be achieved

simply by improving growth and fabrication procedures to reduce defect densities.

GaN/AIN/InN

GaN/AIN/InN is both a promising and a challenging material system for LED
and LD design. Some attributes of this system include the large range in bandgaps
(1.9 eV for InN, 3.4 eV for GaN, and 6.2 eV for AIN, all for wurzite crystals) and the
robust nature of the materials. The challenges lie in the lack of a lattice matched
substrate, obtaining p-type dopability, as well as in processing difficulties resulting
from the resistance of the materials to wet chemical etches. Detailed reviews of the
nitride material system and early results can be found in References [45, 46, 47].
We now give a brief review of the recent progress in nitride development.

As late as 1989, there still remained many obstacles to the development of
nitride LEDs and LDs: p-type doping was still unattainable, material quality was
poor and heterostructure growth was still undeveloped. Also, most structures at
the time were grown on sapphire substrates which has a large lattice mismatch to
GaN. Metal insulating semiconductor (MIS) structures had been fabricated, but
no p-n homojunction LEDs had yet been produced. In 1989, however, Amano et
al. [48] made a significant advance with the announcement of p-type conduction
in Mg-doped, OMVPE grown GaN, using a low-energy electron beam irradiation
treatment (LEEBI). As grown, the Mg-doped GaN is highly resistive, but after the
LEEBI treatment the resistivity dropped by more than five orders of magnitude.
Using this LEEBI treatment Amano et al. were able to fabricate the first ever p-n
homojunction LED in the nitride system.

At the time, a thin AIN buffer layer was required to obtain reasonable GaN
growth [49], due to the large lattice mismatch between GaN and the sapphire
substrates being used. Using this technique, x-ray rocking curves with full width

half maximas (FWHMs) as low as 2 arc minutes were attainable. In 1991 Nakamura

showed that a low temperature GaN buffer layer could also be used to obtain
GaN layers of reasonable quality (FWHMs of approximately 5 arc minutes) [50].
Improvements on this technique led to the development of GaN p-n homojunction
blue LEDs with external quantum efficiencies of 0.18% and good current-voltage
characteristics [51] (see Table 1.1). The efficiencies of these devices were already
much better than the efficiencies of commercial SiC blue LEDs. At roughly the
same time, investigation of InGaN growth was yielding films of relatively high
quality [52]. These advances, combined with the growth of high quality Si-doped
n-InGaN films [53], eventually led to the demonstration of p-GaN /n-InGaN /n-GaN
blue [54] and violet [55] LEDs. The active region in these devices was Si-doped
InGaN, producing luminescence between 411 and 440 nm with FWHMs of 22-
26 nm. While the external quantum efficiencies were 0.15-0.22%, current-voltage
characteristics were poor. This was attributed to high resistivity of the p-type GaN
layer, even after LEEBI treatment. Work on LED structures using AIN buffer
layers had also progressed considerably during this time, with a report in 1993
by Akasaki et al. of ultra-violet LEDs with efficiencies of 0.33% under a forward
current of 90 mA and bias of 5.0 V [56]. These results were quite respectable but
would soon be bettered by the results of Nakamura and co-workers.

Their first development was the improvement of p-type conductivity in GaN
using N2-ambient thermal annealing [57, 58]. It was concluded that acceptor-H
neutral complexes was the cause of the high resistivity in Mg-doped GaN, and
that both the No-ambient thermal anneal and the LEEBI treatments lower the
resistivity by removing the hydrogen from the GaN film. The second development
was the successful Zn-doping of p-InGaN [18]. These two developments led to the
first demonstration of candela-class InGaN/AlGaN blue LEDs [18] with external
quantum efficiencies of 2.7%. This is roughly two orders of magnitude better than
commercial blue SiC LEDs. Recent versions of this device [59] (see also Figure 1.5)
have an external quantum efficiency of 4.4% and luminous intensity of roughly 2.8

candela, operating at an injection current of 30 mA at 3.8 V. Blue-green LEDs in

Ref. Active Comment d Ext. Qu. | Vop Jop Lifetime
Layer (nm) | Eff. (%) | (V) | (A/cm?)| (hours)
(a) ZnSe-based
[34] | CdZnSe* ~ 506 | 0.01 | 7 4 _
[35] | ZnSSe* ~ 494 0.1 20 1 =
[36] | CdznSe:Te*| — |nase| — |>10} — —
— ~ 960 0.03 >10 — —
[36] | CdZnSe:Te | “graded” | ~ 486] <0.03 6 4 —
~ 560} <0.03 6 4 —
[37] | ZnCdSSe HgSe 480 0.03 3.0 4 —
ZnTeSSe ~ 504 0.05 3.6 20 —
[38] | ZnCdSSe | HgSe | 490 | 0.57 | 3.7 4 10?
ZnTeSSe ~ 506 0.62 3.8 4 10?
[39] | ZnTeSSe HgSe | 510 32 | 4.0 16 108
ZnSe sub.
(b) Nitride-based
[51] GaN GaN buf. | ~ 430 0.18 4 6.7 —
[54] | InGaN:Si 440 0.22 19 6.7 —
|56} GaN AIN buf. 372 0.33 5 — —
[18] | InGaN:Zn | GaN buf. |~450] 2.7 | 3.6] 6.7t 104
Anneal

[59] | InGaN:Zn | GaN buf.} ~450| 44 | 3.8) 107 104
Anneal | ~500] 14 | 3.7] 10T 10¢

Table 1.1: Recent Progress of (a) ZnSe-based and (b) GaN-based LEDs. Tildes

before a specified wavelength indicate a broad emission peak (> 50 nm). Dashes

indicate that the quantity was not specific in the reference.

* Indicates p-type GaAs substrates used so no contact was made to p-ZnSe.

T Assumes device size unchanged from Reference [54].

External Quantum Efficiency (%)
(photons/electron)

10.0

1.0

0.1

lo R
Blue Green < elow Orange «et,
—_—_—__——> a
Sa
ee AlGaAs/AIGaAs + 1.0
ZnSSeTe/ZnSe ‘ \ .
AlGaAs/GaAs
\ AIGainP/GaAs
< i + 0.8
/ \
/ \
L InGaN/Saphire / \ + 0.6
| ; \
< / \
/ ad \ 4+ 0.4
/ GaAsP:N/GaP
/ \
L / \
/ e Ga | 0.2
/ GaP/GaP \
/ AN
/ NX
zy SIC/SIC \

Wt gg
450 475 500 525 550 575 600 625 650 675

Peak Wavelength (nm)

Relative Eye Sensitivity

Figure 1.5: External quantum efficiencies of both commercial and developing

LED technologies as a function of peak wavelength. Notation for labels is ac-

tive layer/substrate. Dashed line shows the relative eye sensitivity at the different

wavelengths.

this system have also been recently demonstrated with injection currents of 30 mA
at 3.7 V and external quantum efficiencies of 1.4%. Device lifetimes for both the
blue and the blue-green LEDs are several tens of thousands of hours.

Despite these spectacular results, there are still many issues that need to be
addressed in the nitride system. For example, the emission linewidths in the recent
nitride devices are very broad (70-80 nm) which is not conducive to achieving
lasing. Also, the material quality is still very poor, primarily because of the lack
of a lattice matched substrate. Some early work in this area on the growth of
thick (200-400 zm) GaN layers appears promising [60]; however, no commercial
lattice matched substrates for GaN exist as of yet. Another issue that needs to
be addressed is device processing; since GaN cannot be etched by standard wet
chemical processes, reactive ion etching procedures must be explored. Finally,
there is still much work to be done on the lower temperature growth of nitrides

using MBE, which is still in the early stages of development.

1.2.5 Introduction to Graded Electron Injector

A brief description of the graded electron injector is given here as an introduction;
a more detailed description of its operation is given in Chapter 2. As was pointed
out above, the initial challenge for groups pursuing wide bandgap LEDs was to
obtain both p- and n-type dopability. In the case of GaN efforts, an additional
difficulty exists due to lack of lattice-matched substrates. In both the ZnSe and
nitride systems, amphoteric (both p- and n-type) doping has been achieved at
least to first order. However, it is still possible that the poor device lifetimes
in the ZnSe-based devices will remain a problem for some time, and that broad
luminescence from deep acceptor levels as well as poor material quality in the
nitrides will preclude LD structures.

In anticipation of the difficulties associated with forcing amphoteric doping,

the approach used in the graded electron injector was to form a heterojunction of

one n-type and one p-type dopable semiconductor, while satisfying the constraints
of close lattice match and carrier injection into a semiconductor with the desired
bandgap. One promising pair of semiconductors that satisfies both the dopability
and lattice match constraints is CdSe and ZnTe. CdSe can be easily doped n-
type, while ZnTe can be doped p-type, and the two semiconductor have a lattice
mismatch of only 0.4% (Figure 1.4). To obtain carrier injection into the wider
bandgap, ZnTe material requires the insertion of a graded Mg,Cd,_,Se layer to
facilitate electron injection into the ZnTe layer, while blocking holes from entering
the lower bandgap CdSe region. A schematic diagram of the operation of this
device is shown in Figure 1.6.

The main challenge in the graded electron injector approach lies in optimizing
the graded Mg,Cdj_,Se layer, not in obtaining amphoteric doping as was the case
initially in the ZnSe- and nitride-based systems. For ideal operation, the conduc-
tion band edge of Mg,Cdi_,Se must line up with the conduction band edge of
ZnTe at the Se/Te interface. The ideal composition of Mg,Cd,_,Se to accomplish
this depends on both the bandgap of MgSe and the band alignment of Mg,Cd,_,Se
to ZnTe. Thus, optimization of the graded injector design requires characteriza-
tion of the properties of Mg,Cd,_,Se ternaries, and careful design of the graded
injector layer itself. Characterization of materials properties is also required, using
such techniques as XRD, TEM, and SIMS. Current-voltage, electroluminescence,
photoluminescence, efficiency and device lifetime measurements are also needed in

order to evaluate LED device performance.

E 2.25 eV

| 0.64 ov Co

-Cds Graded -
n e MoGdse p-Znle

Figure 1.6: Schematic diagram showing the CBE and VBE for the graded in-
jector device at flatband. The graded Mg,Cd,_,Se region facilitates electron in-
jection into the higher bandgap material. The valence band offset between the
Mg,Cd,_,Se and the ZnTe epilayer remains abrupt, blocking hole injection into

the narrower bandgap CdSe material.

1.3. The Mixed Anion InAs/GaSb Heteroin-

terface

1.3.1 Motivation

The InAs/GaSb/AISb material system is of interest because it has a number
of technologically significant applications, and is also relevant to fundamental
studies of mixed anion interfaces. Applications of the this material system in-
clude: InAs/AISb oscillators operating at frequencies greater than 700GHz [61],
novel InAs/AlSb/GaSb [62, 63, 64] and InAs/GaSb [65] based tunnel structures,
InAs/Ga,_xIn,Sb IR SL detectors [66, 67], high power GaIlnAsSb/AlGaAsSb
LDs operating at wavelengths useful for fiber optics communication [68] and
InAs/AlSb/GaSb based quantum effect transistors [69]. One reason that the
InAs/GaSb/AISb material system is of such interest is because of the flexibility in
band alignments that can be obtained without having to sacrifice lattice-match.
This is shown in Figure 1.7(a), where we see that a variety of band alignments can
be formed in this system. Other material systems of technical interest could also
benefit from general studies of mixed anion interfaces. These include GalnP/GaAs
heterojunction bipolar transistors (HBTs) [70, 71] and lattice matched heterojunc-
tion InP/GalnAs/InP LDs [72, 73]. Interface characteristics are important in these
two material systems as well, and the effect of interface composition and growth or-
der on the valence band offset has been investigated [74]. The application directly
motivating our investigation of the InAs/GaSb interface is the IR SL detector,
which makes use of the type II band alignment for InAs/In,Ga,_,Sb. More details
concerning this device are given in the following section.

For many of the applications mentioned above, parameters such as interface
abruptness, interface composition and band alignments can have a large effect
on the performance of the device. For example, cross-incorporation of the an-

ion species can result in shorter carrier lifetimes and poor material quality, while

(a)
2.57
ay \ Eo
2.07 =» 2 #@ 8 of =» 8 ow ke
Ss
© 15,
& 2.22
2 lor |
oi 0.15 0.67
O57 f Y — A
cy ae A 0.40
ob e 0.36 | vy —y Ey
Vv
InAs GaSb AlSb
(b)

1.0 rr ae

> c
2 r Increases
d with strain HH
5 0.5- \J — E,
m E
LH
8) — Ey rn 4H Gag lng,4Sb
InAs InAs Gag glng 4S
(Tension) (Compression)

Figure 1.7: (a) Schematic of the conduction and valence band edges for the
InAs/GaSb/AISb material system. For the AlSb conduction band, dashed and
solid lines indicates the indirect and direct conduction band minima respectively.

(b) The band edges for strained and unstrained InAs and GalnSb.

the interfacial composition affects the strain configurations at the interface which
changes the type and level of background doping, as well as the SL bandgap [75].
Furthermore, the interface composition in the InAs/AlSb system can also affect
the carrier mobility, carrier concentration, and the InAs/AISb valence band offset
[76, 77]. Thus, for the purposes of device applications, it is critical to be able to
grow arsenide/antimonide structures with no anion cross-incorporation and with
abrupt, composition-controlled interfaces. However, atomic level control of the in-
terface properties is very difficult because of the range of bond strengths, surface
free energies and bond lengths in the InAs/GaSb system. In addition, III-V struc-
tures are typically grown with group V overpressures, since their vapor pressures
are much larger than those of group III elements which results in lower sticking
coefficients. For growth of mixed anion structures, this overflux of the group V
component becomes a problem and can lead to cross-incorporation of the anion
species as well as difficulty in controlling the interface composition. Details of these

issues are presented in section 1.3.3.

1.3.2 InAs/In,Ga,_,Sb IR SL Detectors

Uses for IR detectors include not only military and civilian surveillance appli-
cations, but also satellite monitoring of the earth’s temperature, IR astronomy,
chemically sensitive monitoring of the atmosphere for pollution control purposes
and medical imaging applications [78].

The strained layer nAs/In,Ga,_,Sb IR SL detector was first proposed by Smith
and Mailhiot in 1987 [66]. The basic operation of this device relies on the type II
nature of the InAs/GaSb band offset. Since the conduction band minimum of InAs
lies below the valence band maximum of GaSb (see Figure 1.7(a)), a superlattice
constructed from these semiconductors can potentially have an effective bandgap
that is much smaller than either of the two constituent semiconductors, which is

desirable for long wavelength detection applications. This is seen in Figure 1.8(a)

where the InAs and GaSb band edges and the hole and electron probability density
functions are shown schematically on the same plot. Note that the probability and
energy plots are superimposed in this figure for visualization purposes only, and
that the SL bandgap is determined by the energies of the electrons and holes whose
probability density is shown, not by the minima and maxima in the probability
density plots. We see that for sufficiently short SL periods, the wavefunction of an
electron in the InAs well will penetrate into the GaSb layers on either side of the
well. Similarly, holes in the GaSb will have a finite probability density in the InAs
layers. The effective SL bandgap then becomes the difference in energy between
the lowest-energy electron state in the InAs and the highest-energy hole state in
the GaSb. This is represented by the shaded region in Figure 1.8(a). Note that
the effective SL bandgap will increase with shorter SL periods due to quantum
confinement effects which increase the minimum electron energy and decrease the
maximum hole energy. However, to obtain the narrow bandgaps needed for long
wavelength (> 10 ym) detection, the layers become so thick that the wavefunction
overlap and the optical matrix elements become too small for detector applications
[79}.

Smith and Mailhiot’s proposal was to replace the GaSb in the SL with
In,Ga;_,Sb. Since GalnSb and InAs are not lattice matched, the individual layers
in the SL will be strained, splitting the valence bands and changing the bandgaps
as shown in Figure 1.8(b). We see in this figure that while the tensile strain in
InAs causes the light hole (LH) band to shift up and the heavy hole (HH) band to
shift down and reduces the bandgap, the compressive strain in In,Ga,_,Sb splits
the HH band up and the LH band down and increases the bandgap. The net effect
is an increase in the energy separation between the bottom of the InAs conduction
band and the top of the In,Ga,_,Sb HH band. This results in a decrease in the
effective SL bandgap for a given set of layer thicknesses, as shown in Figure 1.8(b).
Thus, the strained layer IR SL detector design proposed by Smith and Mailhiot

should, in theory, result in long wavelength detectors with short SL periods and

therefore reasonable optical matrix elements.

1.3.3. Challenges and Issues for Mixed Anion Interfaces

In addition to the reasons given in Section 1.3.1 for studying the InAs/GaSb
interface, motivation exists to determine the cause of the short lifetimes in
InAs/In,Ga,_,Sb IR SL detectors. In theory, these detectors have several advan-
tages over other long wavelength IR detector technologies, including the current in-
dustry standard, Hg,Cd,_,Te detectors. Benefits offered by the InAs/In,Ga,_,Sb
system include: (a) smaller leakage currents due to the decoupling of bandgap
and effective mass, allowing for detectors with small bandgaps but large effective
masses, (b) more precise control of uniformity, since the detection wavelength is
determined primarily by layer thicknesses rather than ternary composition, (c)
compatibility with more well developed III-V processing technologies and (d) sup-
pression of Auger processes which limits the upper limit on detectivity in detectors
[80].

Another competing technology is the GaAs/Al,Ga,_,As quantum well infrared
photodetector (QWIP) [81]. However, despite recent advances in this approach
[82], calculations have shown that the limiting performance of QWIPs will be
significantly below that of Hg,Cd,_,Te devices [83].

Despite the apparent advantages of the InAs/In,Ga,_,Sb IR SL design, detec-
tors competitive with Hg,Cd,_,Te technologies have not yet been fabricated in this
system [84]. This poor performance has been attributed to the shorter extrinsic
lifetime in InAs/In,Ga,_,Sb IR SL devices, due to the presence of Shockley-Read-
Hall (SRH) centers [80]. One possible explanation for this is that the generation
of SRH centers is somehow related to difficulties in controlling the mixed anion
As/Sb interface, which could result in antisite defects.

Figure 1.9 shows schematically why the control of this mixed anion interface

is so difficult. Details of the mechanisms of InAs/GaSb interface formation are

Unstrained InAs/GaSb Superlattice

Electron
Probability

Density X\

Effective
Superlatice —>
Bandgap

Hole 7
Probability
Density

Gasb Gasb Gasb Gasb
InAs InAs InAs InAs

Strained InAs/Ga,In,,Sb Superlattice

Electron
Probability

Density NX

Effective
Superlattice >

Bandgap
Hol a WH ed NY NS
ole
Probability
Density

GalnSb GainSb GalnSb GalnsSb
InAs InAs InAs InAs

Figure 1.8: Schematic of band edges and carrier probability densities for
InAs/GaSb (top) and InAs/GalInSb (bottom) SLs. Notice the decrease in effec-
tive superlattice bandgap (shaded region) for the strained SL. The probability and

energy plots are superimposed for visualization purposes only.

(a)

InSb-like
Interface

(b)

InAs-on-GaSb GaSb-on-inAs

Figure 1.9: (a) Schematic of two possible interface compositions that may be
formed in an InAs-on-GaSb heterojunction. Changing the interface composition
may change such quantities as carrier mobility, carrier concentration, and band
alignment. (b) Schematic of the two possible growth orders. Some of the effects
that may affect interface abruptness and cross-incorporation are bond strengths,

surface free energies, and bond lengths.

analyzed in Chapters 5 and 6. Figure 1.9(a) shows an obvious complication in-
troduced with mixed anion interfaces. Since there is no common element across
the interface, two possible interface compositions may be formed; in the case of
InAs/GaSb, both InSb-like and GaAs-like interfaces are possible. Figure 1.9(b)
shows the two possible growth orders available: InAs-on-GaSb and GaSb-on-InAs.

Other, somewhat more subtle mechanisms influencing interface formation are
differing bond strengths, bond lengths and surface free energies, all of which can
affect the dynamics and chemistry of interface formation. For example, in the
InAs/GaSb material system, the very strong GaAs bond could drive the exchange
of Sb with As in GaSb layers, the strong surfactant nature of Sb could influence
interface abruptness and the large lattice mismatch (~ 6-7%) between GaAs and
InSb with respect to GaSb could also have a large effect on interface formation. In
light of these factors, growth of the mixed anion InAs/GaSb interfaces is very chal-
lenging and requires systematic characterization to achieve proper control of the
interface properties. It should be noted that many of the difficulties in controlling

the InAs/GaSb interface are common to other mixed anion systems as well.

1.3.4 Characterization Techniques

Due to the very small length scales that are of interest in studies of interface abrupt-
ness, it is very difficult to obtain a good measure of interface quality using only one
characterization technique. In the studies presented in Chapters 5 and 6 a number
of experimental techniques (XPS, RHEED, cross-sectional scanning tunneling mi-
croscopy (STM), and SIMS) were utilized in order to characterize the nAs/GaSb
interface. Since results in this thesis deal primarily with the XPS and RHEED
studies, brief overviews of these two techniques are given here. Cross-sectional
STM and SIMS studies are presented only in Chapter 6; therefore, description
of the respective techniques are given there. It should be emphasized that the

RHEED, STM and SIMS studies were all performed by other researchers.

Figure 1.10 shows a schematic drawing of a typical XPS setup, with a represen-
tative XPS binding energy spectrum. In the X-ray source, electron bombardment
of a metal target, typically Mg or Al, produces X-rays including Kay, lines at
energies of 1253.6 eV and 1486.6 eV respectively. The Ka line is used because it
accounts for roughly one-half of the X-rays produced by the electron bombardment.
The shape of this line without a monochrometer is asymmetric with a FWHM of
~ 1.0 eV for Al and ~ 0.8 eV for Mg. If a monochrometer is used, this can be
reduced to 0.3-0.4 eV. The X-rays impinge on the sample, ejecting photoelectrons
from the core levels within the sample. The photoelectrons detected at the electron

spectrometer have a kinetic energy determined by

where hv is the incident X-ray energy, F is the binding energy of the core level from
which the photoelectron was ejected, ¢sp is the work function of the spectrometer
and T;, is the photoelectron kinetic energy measured at the spectrometer. Thus, for
a given X-ray source and spectrometer work function, measuring the photoelectron
kinetic energy determines the binding energy of its respective core level allowing
chemical identification as shown in Figure 1.10. Both the binding energy and
relative intensities of core level peaks are used in XPS data analysis. It should
be noted that for the X-ray energies typically used in XPS, the photoelectron
escape depths are tens of angstroms, thus XPS is a near-surface chemical analysis
technique.

A schematic diagram of a typical RHEED setup is shown in Figure 1.11. In
RHEED, high energy electrons (~ 10 keV) impinge on the sample surface at graz-
ing angles. Since the surface of crystalline samples have characteristic periodicities
in certain directions, the electrons will diffract from the sample surface with a
characteristic pattern which is captured by a phosphor screen. The RHEED pat-
tern on the phosphor screen can then be analyzed in real time, or captured and

digitized for later analysis. Since the electrons have very shallow incident angles

E X-ray source
ON
Electron photoelectron

Monochrometer

soectrometer
C Ga 3p As 3d || Ga 3d
In 4d

Photoelectron Binding Energy

Figure 1.10: Schematic of the X-ray Photoelectron Spectroscopy (XPS) experi-

mental setup, and a representative photoelectron spectrum.

Phosphor
screen

e-gun
Specular Streak
Reconstruction / Lovder streak
Phosphor
Screen
Digitize
Cc
co)
£ _\Measured Streak
3 Spacing

Position

Figure 1.11: Schematic of the Reflection High Energy Electron Diffraction
(RHEED) experimental setup. Quantities of interest are the reconstruction pat-

tern, the specular streak intensity and the streak separation.

(1-3°), they penetrate only ~5 A into the sample, so RHEED effectively measures
only very near surface structure.

Three main pieces of information can be extracted from a RHEED pattern:
the surface reconstruction pattern, the specular streak intensity and the streak
separation. The surface reconstruction pattern is due to the rearrangement of
the surface atoms to terminate dangling bonds. This surface reconstruction can
change significantly depending on the exact growth conditions, so it is useful in
monitoring the effect of parameters such as substrate temperature and flux ratios.
The specular streak intensity varies as each layer proceeds through various stages
of nucleation. Partial nucleation results in surface roughness which reduces the
specular streak intensity. This feature can be used to monitor growth rates and
changes in growth modes. While the steak separation can also be obtained from
RHEED patterns, it is more difficult to extract conclusive information from this

quantity.

1.4 Outline of Thesis

The thesis is divided into two parts. Part I (Chapters 2-4) describes work that has
been performed on the II-VI graded injector LED. Simulation results, materials
characterization and device fabrication issues will be discussed. Part II of this the-
sis (Chapters 5-6) describes the characterization of the InAs/GaSb heterointerface.
Most of the work presented is based on XPS analysis; however, results from other
characterization techniques will also be discussed.

Chapter 2 describes in detail the design and operation of the graded electron
injector LED, as well as other proposed devices. Simulations based on the Drift-
Diffusion model are used to determine the feasibility of a variety of device designs,
including one that does not perform optimally, two that have been experimen-
tally implemented and others that have not yet been implemented. Details of the

model, which was modified from the standard Drift-Diffusion model to account for

heterostructures, are presented.

In Chapter 3 of the thesis XPS measurements of the MgSe/Cdo54Zno.4gSe and
MgTe/Cdo.sgZno.12Te valence band offsets are presented. Background on the meth-
ods used to measure valence band offsets using XPS are given. Also specific details
of the sample growth and XPS measurements and data analysis are presented. The
implications of the measured band offsets on current II-VI light emitter design are
discussed.

In Chapter 4, we present materials characterization studies and the latest device
results. XRD, TEM and SIMS studies of the graded electron injector are presented.
Potential materials problems brought to light by these characterization techniques
are discussed. The most recent current-voltage, electroluminescence and efficiency
measurements of LED devices are given. Device degradation and device fabrication
issues are also addressed.

In Chapter 5 surface exchange reaction studies of the InAs/GaSb heteroin-
terface are presented. The exchange of As with Sb during Sb interrupts of InAs
surfaces, and the exchange of Sb with As during As interrupts of GaSb surfaces
are presented. The exchange reactions are studied using XPS and RHEED; details
of the technique are given. The implications of our results for IR SL detectors are
discussed.

Finally, Chapter 6 presents studies of the nAs/GaSb heterojunction. XPS
measurements of the InAs/GaSb valence band offset as a function of interface
type and growth order are presented. Implications of the results to both device
applications and interface studies of mixed anion systems are discussed. Lastly,
investigations of asymmetry in the abruptness of the InAs/GaSb interface using

XPS, RHEED, cross-sectional STM and SIMS are presented.

Bibliography

[1] M.C. Phillips, M.W. Wang, J.F. Swenberg, J.O. McCaldin, and T.C. McGill,
Appl. Phys. Lett. 61, 1962 (1992).

[2] L.W. Cook, M.D. Camras, S.L. Rudaz, and F.M. Steranka, in Proceedings
of the 14th International Symposium on GaAs and Related Compounds 1987
(Institute of Physics, Bristol, 1988), pp. 777-780.

[3] G.M. Craford, Circuits and Devices 8, 24 (1992).

[4] A red LED traffic signal lamp uses 25 W compared to 150 W for conventional
incandescent lamps. This results in a savings of 1095 kWh/year or $131/year
(at $0.12/kWh) for each signal head. (Econolite Control Products, Inc. 1993
Catalog.)

[5] For example, the energy savings for Pasadena, which has over 3000 traffic

signal heads, would be roughly $400,000/year.
[6] Business Week, page 82, July 25, 1994.

[7] Projected from number of signal heads in Pasadena and number of LEDs used

in Econolite LED lamps.
[8] M. Levitt, Proc. SPIE - Int. Soc. Opt. Eng. 1719, 1 (1992).

[9] K.H. Huang, J.G. Yu, C.P. Kuo, R.M. Fletcher, T.D. Osentowski, L.J. Stinson,
M.G. Craford, and A.S.H. Liao, Appl. Phys. Lett. 61, 1045 (1992).

[10] I. Hino, K. Kobayashi, and T. Suzuki, Jpn. J. Appl. Phys. 23 L746 (1984).

[11] H. Sugawara, M. Ishikawa, and G. Hatakoshi, Appl. Phys. Lett. 58, 1010
(1991).

[12] R.M. Fletcher, C.P Kuo, T.D. Osentowski, K.H. Huang, M.G. Craford, and
V.M. Robbins, J. Elect. Materials 20, 1125 (1991).

[13] F.A. Kish, F.M. Steranka, D.C. DeFevere, D.A. Vanderwater, K.G. Park,
C.P. Kuo, T.D Osentowski, M.J. Peanasky, J.G. Yu, R.M. Fletcher, D.A.
Steigerwald, M.G. Craford, and V.M. Robbins, Appl. Phys. Lett. 64, 2839
(1994).

[14] C.P. Kuo, R.M. Fletcher, T.D. Osentowski, M.C. Lardizabal, M.G. Craford,
and V.M. Robbins, Appl. Phys. Lett. 57, 2937 (1990).

[15] J.A. Edmond, H.-S. Kong, and C.H. Carter, Jr., Physica B 185 453 (1993).

[16] Y. Fan, J. Han, L. He, J. Saraie, R.L. Gunshor, M. Hagerott, H. Jeon, A.V.
Nurmikko, G.C. Hua, and N. Otsuka, Appl. Phys. Lett. 61, 3160 (1992).

[17] J.O. McCaldin, J. Vac. Sci. Technol. A 8, 1188 (1990).
[18] S. Nakamura, T. Mukai, M. Senoh, Appl. Phys. Lett. 64, 1687 (1994).

[19] H. Cheng, J.M. DePuydt, J.E. Potts, and T.L. Smith, Appl. Phys. Lett. 52,
147 (1988).

[20] J.M. DePuydt, M.A. Haase, H. Cheng, and J.E. Potts, Appl. Phys. Lett. 55,
1103 (1989).

[21] A. Taike, M. Migita, and H. Yamamoto, Appl. Phys. Lett. 56, 1989 (1990).

[22] K. Ohkawa, T. Karasawa, and T. Mitsuyu, J. Cryst. Growth 111, 797 (1991).

[23] R.M. Park, M.B. Troffer, C.M. Rouleau, J.M. DePuydt, and M.A. Haase,
Appl. Phys. Lett. 57 2127 (1990).

[24] J. Qiu, J.M. DePuydt, H. Cheng, and M.A. Haase, Appl. Phys. Lett. 59, 2992
(1991).

[25] M.A. Haase, J. Qiu, J.M. DePuydt, and H. Cheng, Appl. Phys. Lett. 59, 1272
(1991).

[26] H. Okuyama, K. Nakano, T. Miyajima, and K. Akimoto, Jpn. J. Appl. Phys.
30 L1620 (1991).

[27] H. Okuyama, T. Miyajima, Y. Morinaga, F. Hiei, M. Ozawa, and K. Akimoto,
Electronics Letters 28 1798 (1992).

[28] J.M. Gaines, R.R. Drenten, K.W. Haberern, T. Marshall, P. Mensz, and J.
Petruzzello, Appl. Phys. Lett. 62 2462 (1993).

[29] Y. Moringa, H. Okuyama, and K. Akimoto, Jpn. J. Appl. Phys. 32, 678
(1993).

[30] Y. Lansari, J. Ren, B. Sneed, K.A. Bowers, J.W. Cook, Jr., and J.F. Schetzina,
Appl. Phys. Lett. 61, 2554 (1992).

[31] F. Hiei, M. Ikeda, M. Ozawa, T. Miyajima, A. Ishibashi, and K. Akimoto,
Electronics Letters 29, 878 (1993).

[32] P.M. Mensz, Appl. Phys. Lett. 64, 2148 (1994).
[33] R.G. Dandrea and C.B. Duke Appl. Phys. Lett. 64, 2145 (1994).

[34] H. Jeon, J. Ding, A.V. Nurmikko, W. Xie, M. Kobayashi, and R.L. Gunshor,
Appl. Phys. Lett. 60, 892 (1992).

[35] W. Xie, D.C. Grillo, R.L. Gunshor, M. Kobayashi, H. Jeon, J. Ding, A.V.
Nurmikko, G.C. Hua, and N. Otsuka, Appl. Phys. Lett. 60, 1999 (1992).

[36] M. Hagerott, J. Ding, H. Jeon, A.V. Nurmikko, Y. Fan, L. He, J. Han, J.
Saraie, R.L. Gunshor, C.G. Hua, and N. Otsuka, Appl. Phys. 62, 2108 (1993).

[37] J. Ren, D.B. Eason, Y. Lansari, Z. Yu, K.A. Bowers, C. Boney, B. Sneed,
J.W. Cook, Jr., J.F. Schetzina, M.W. Koch, and G.W. Wicks, J. Vac. Sci.
Technol. B 11, 955 (1993).

[38] D.B. Eason, Z. Yu, C. Boney, J. Ren, L.E. Churchill, J.W. Cook, Jr., J.F.
Schetzina, and N.A. El-Masry, to be published in J. Cryst. Growth (1994).

[39] Presented by J.F. Schetzina at the 1994 Device Research Conference in Boul-
der, CO.

[40] A. Salokatve, H. Jeon, J. Ding, M. Hovinen, A.V. Nurmikko, D.C. Grillo, L.
He, J. Han, Y. Fan, M. Ringle, and R.L. Gunshor, Abstract from the 1994
Conference on Lasers and Electro-Optics (CLEO) in Anahiem, CA, pg. 202.

[41] Sony results in plenary talk by D. Olego at the 1994 Device Research Confer-

ence in Boulder, CO.

[42] S. Guha, J.M. DePuydt, M.A. Haase, J. Qiu, and H. Cheng, Appl. Phys. Lett.
63, 3107 (1993).

[43] L.H Kuo, L. Salamanca-Riba, J.M. DePuydt, H. Cheng, and J. Qiu, Appl.
Phys. Lett. 63, 3197 (1993).

[44] J. Ren, D.B. Eason, Z. Yu, B. Sneed, J.W. Cook, Jr., J.F. Schetzina, N.A.
El-Masry, X.H. Yang, J.J. Song, G. Cantwell, and W.C. Harsh, J. Vac. Sci.
Technol. B 12, 1262 (1994).

[45] S. Strite and H. Morkog, J. Vac. Sci. Technol. B 10, 1237 (1992).
[46] S. Strite, M.E. Lin, and H. Morkoc, Thin Solid Films 231, 197 (1993).

[47] T. Matsuoka, J. Cryst. Growth 124, 433 (1992).

45
[48] H. Amano, M. Kito, K. Hiramatsu, and I. Akasaki, Jpn. J. Appl. Phys. 28,
L2112 (1989).

[49] H. Amano, N. Sawaki, I. Akasaki, and Y. Toyada, Appl. Phys. Lett. 48, 353
(1986).

[50] S. Nakamura, Jpn. J. Appl. Phys. 30, L1705 (1991).
[51] S. Nakamura, T. Mukai, and M. Senoh, Jpn. J. Appl. Phys. 30, L1998 (1991).

[52] N. Yoshimoto, T. Matsuoka, T. Sasaki, and A. Katsui, Appl. Phys. Lett. 59
2251 (1991).

[53] S. Nakamura, T. Mukai, and M. Senoh, Jpn. J. Appl. Phys. 31, L16 (1993).
[54] S. Nakamura, M. Senoh, and T. Mukai, Jpn. J. Appl. Phys. 32, L8 (1993).
[55] S. Nakamura, M. Senoh, and T. Mukai, Appl. Phys. Lett. 62, 2390 (1993).

[56] I. Akasaki, H. Amano, N. Koide, M. Kotaki, and K. Manabe, Physica B 185,
428 (1993).

[57] S. Nakamura, T. Mukai, M. Senoh, and N. Iwasa, Jpn. J. Appl. Phys. 31,
L139 (1992).

[58] S. Nakamura, N. Iwasa, M. Senoh, and T. Mukai, Jpn. J. Appl. Phys. 31,
1258 (1992).

[59] H. Morkog and B. Sverdlov, Evaluation of Nichia Blue and Blue-Green InGaN
LEDs, unpublished.

[60] T. Detchprohm, K. Hiramatsu, H. Amano, and I. Akasaki, Appl. Phys. Lett.
61, 2688 (1992).

[61] E.R. Brown, J.R. Séderstrém, C.D. Parker, L.J. Mahoney, K.M. Molvar, and
T.C. McGill, Appl. Phys. Lett. 58, 2291 (1991).

[62] M. Sweeny and J. Xu, Appl. Phys. Lett. 54, 546 (1989).

[63] J.R. Séderstrém, D.H. Chow, and T.C. McGill, Appl. Phys. Lett. 55, 1094
(1989).

[64] L.F. Luo, R. Beresford, and W.I. Wang, Appl. Phys. Lett. 55, 2023 (1989).

[65] D.A. Collins, E.T. Yu, Y. Rajakarunanayake, J.R. Séderstrém, D.Z.-Y. Ting,
D.H. Chow, and T.C. McGill, Appl. Phys. Lett. 57, 683 (1990).

[66] D.L. Smith and C. Mailhiot, J. Appl. Phys. 62, 2545 (1987).

[67] R.H. Miles, D.H. Chow, J.N. Schulman, and T.C. McGill, Appl. Phys. Lett.
57, 801 (1990).

[68] H.K. Choi, and S.J. Eglash, Appl. Phys. Lett. 61, 1154 (1992).
[69] D.A. Collins, D.H. Chow, and T.C. McGill, Appl. Phys. Lett. 58, 1673 (1991).
[70] H. Kroemer, J. Vac. Sci. Technol. B 1, 126 (1983).

[71] S.L. Delage, M.A. DiForte-Poisson, H. Blanck, C. Bryliinski, E. Chartier, and
P. Collot, Electronics Letters 27, 254 (1991).

[72] B.I. Miller, J.H. McFee, R.J. Martin, and P.K. Tien, Appl. Phys. Lett. 31, 44
(1978).

[73] W.T. Tsang, F.S. Choa, M.C. Wu, Y.K. Chen, A.M. Sergent, and P.F.
Sciortino, Jr., Appl. Phys. Lett. 58, 2610 (1991).

[74] J.P. Landesman, J.C. Garcia, J. Massies, G. Jezequel, P. Maurel, J.P. Hirtz,
and P. Alnot, J. Vac. Sci. Technol. B 10, 1761 (1992).

[75] D.H. Chow, R.H. Miles, and A.T. Hunter, J. Vac. Sci. Technol. B 10, 888
(1992).

[76] G. Tuttle, H. Kroemer, and J.H. English, J. Appl. Phys. 67, 3032 (1990).

[77] J.R. Waldrop, G.J. Sullivan, R.W. Grant, E.A. Kraut, and W.A. Harrison, J.
Vac. Sci. Technol. B 10, 1773 (1992).

[78] T.C. McGill and D.A. Collins, Semicond. Sci. Technol. 8, S1 (1993).

[79] D.K. Arch, G. Wicks, T. Tonaue, and J.L Staudenmann, J. Appl. Phys. 58,
3933 (1985).

[80] E.R. Youngdale, J.R. Meyer, C.A. Hoffman, F.J. Bartoli, C.H. Grein, P.M.
Young, H. Ehrenreich, R.H. Miles, and D.H. Chow, Appl. Phys. Lett. 64,
3160 (1994).

[81] B.F. Levine, K.K. Choi, C.G. Bethea, J. Walker, and R.J. Malik, Appl. Phys.
Lett. 50, 1092 (1987).

[82] B.F. Levine, C.G. Bethea, G. Hasnain, V.O. Shen, E. Pelve, R.R. Abbott,
and S.J. Hsieh, Appl. Phys. Lett. 56, 851 (1990).

[83] M.A. Kinch and A. Yariv, Appl. Phys. Lett. 55, 2093 (1989).

[84] R.H. Miles, D-H. Chow, J. N. Schulman, D.A. Collins, M.W. Wang, R.W.
Grant, and T.C. McGill, submitted to J. Elect. Materials .

Part I

Graded Electron Injector, II-VI
Wide Bandgap Light Emitters

Chapter 2

Light Emitter Simulation and
Design

2.1 Introduction and Outline

A fundamental problem in fabricating wide gap binary II-VI light emitting diodes
(LEDs) has been the inability to obtain both n- and p-type doping in these materi-
als by conventional doping techniques. Recently, this limitation has been partially
overcome by nitrogen plasma doping to produce p-ZnSe with hole concentrations
up to 108 cm? [1]. Although such doping has led to the demonstration of the
first blue-green laser diodes [2], hole concentrations are still not high enough to
afford ohmic contacts to p-ZnSe. Researchers have developed sophisticated con-
tacting schemes such as pseudo-graded ZnSeTe layers [3], ZnTe/ZnSe superlattices
[4], and Hg-based contacts [5] in order to obtain a low resistance contact to p-ZnSe.
Thus, the realization of highly doped p-ZnSe is still a major obstacle to the further
development of ZnSe based LEDs and laser diodes. An alternative solution to the
dopability problem is to use a heterojunction formed between a naturally n-type
and a naturally p-type material.

In this chapter, we examine the simulation and design of I-VI heterostruc-

ture LEDs based on n-CdSe and p-ZnTe. We choose these two semiconductors
because of their favorable dopability and close lattice match. The drift-diffusion
model, modified to account for heterojunctions, is used in our simulations. To
determine LED design feasibility, bulk and interfacial recombination are included
in the model. Details of the model are given in Section 2.2. Some of the more
important parameters accounted for in the design process are band offsets, lattice
mismatch, dopability, and interfacial effects. Where unknown, band offsets used in
the simulations are estimated using the common anion rule [6]. We emphasize here
that the simulation results are not dependent on the validity of this rule, as demon-
strated in one simulation in which an undesirable deviation from the common anion
rule is used. Details of the design methodology are discussed in Sections 2.2 and
2.4. Simulation results are presented in Section 2.4. A simple CdSe/ZnTe hetero-
junction is first presented, followed by a CdSe/Mg,Cd,_,Se/ZnTe heterostructure
design utilizing a graded injector to facilitate electron injection [7]. Next, a design
incorporating a Mg,Zn,_,Te electron confining layer is presented. Finally, a de-
vice design that allows tunability of the wavelength of light emission from green to
blue is proposed and analyzed. Recent developments in the graded injector device
design and a proposal for a lattice matched quaternary-based LED structure are

presented in Section 2.5. Section 2.6 concludes the chapter.

2.2 Design Methodology

In the design of heterostructure devices, it is often useful to first examine a dia-
gram which simultaneously exhibits lattice parameters, band offsets, and dopabil-
ity for the materials of interest. The McCaldin diagram [8] of Figure 2.1 shows
the common wide bandgap binary II-VI compounds, along with other common
semiconductors for comparison purposes. The band offsets are those predicted by
Harrison’s linear combination of atomic orbitals (LCAO) method [9].

Examining Figure 2.1, we note first that none of the wide bandgap II-VI com-

AA
S IF
LY
> Prrrery avers 7s 68
S 0 Sj
Cc
Wet
—2 4 O

ZnS

—3 a a DO OS OP SS

5.4 9.6 9.8 6.0 6.2

Lattice constant (Angstroms)

Figure 2.1: McCaldin diagram for common semiconductors. Each material is repre-
sented by a vertical line with a square at the bottom, corresponding to the valence
band edge, and a triangle at the top, corresponding to the conduction band edge.
The horizontal position of each line is determined by the lattice parameter. The
length of each line scales with the bandgap of the material that it represents. The
vertical position of each line is determined by the respective valence band offset
calculated using Harrison’s LCAO method. Open triangles and squares represent
semiconductors that cannot be doped n or p type respectively using conventional

doping techniques. The zero of energy is chosen to be the valence band edge of

GaAs,

pounds can be conventionally doped both n- and p-type, and only one, ZnTe, can
be conventionally doped p-type. Thus, it is logical that a wide bandgap LED de-
sign should incorporate ZnTe. Also, since interface states caused by dislocations
can have severely detrimental effects on the electrical and optical properties of a
device, it is highly desirable to have a device design that uses only nearly lattice
matched H-VI compounds. From Figure 2.1, we see that CdSe can be doped n-
type and has only a 0.4% lattice mismatch to ZnTe. Thus, an n-CdSe/p-ZnTe
heterojunction shows some promising characteristics for a wide bandgap LED. No
other combination of II-VI compounds allows for both a small lattice mismatch
and formation of a conventionally doped p-n heterojunction. However, the exper-
imental valence band offset between CdSe and ZnTe is 0.64 + .07 eV [10], with
the ZnTe valence band edge higher than the CdSe valence band edge. Since the
CdSe bandgap is 1.74 eV, this gives a conduction band offset of roughly 1.15 eV,
resulting in a type IT heterojunction which favors carrier injection into the smaller
bandgap CdSe layer [11]. The simulations in Section 2.4.1 confirm that a simple n-
CdSe/p-ZnTe heterojunction device suffers from interface recombination and hole
injection.

To improve on the simple CdSe/ZnTe device, we need first to examine the
properties of some magnesium based II-VI compounds. The McCaldin diagram
in Figure 2.2 shows the familiar binary II-VI compounds, as well as MgS, MgSe,
and MgTe. Mg,Cd;_,Se and Mg,Zn,_,Te ternaries are represented by the shaded
regions in this diagram. Again, the LCAO method is used to determine the valence
band offset in this diagram. In Chapter 3, XPS measurements of Mg-calcogenide
band offsets show that the valence band maxima of MgSe and MgTe lie below the
respective valence band maxima of Cdo54Zno.4gSe and Cdg ggZng.12Te; however, the
simulations in Section 2.4 were performed prior to these measurements and so used
band alignments based on the common anion rule. This assumption does not affect
the conclusion drawn from the simulation results; which is that the graded injector

concept is feasible, but it does influence the details of the LED structure, as will

Energy (eV)

Ww”

MgTe CdTe

ZnSe MgSe CdSe
O 0 cas
ZnS Mgs 7
aot L Lid Liits L, LiL it Li | oe on oe oe ee oe i | on on oe oe ee ee I LELLELLL om | LeEpeeuEy se
5.4 5.6 5.8 6 6.2 6.4 6.6

Lattice constant (Angstroms)

Figure 2.2: McCaldin diagram for some commonly used II-VI materials with the

addition of MgS, MgSe and MgTe. Mg,Cd,_,Se and Mg,Zn,_,Te ternaries are

represented by the two shaded regions.

be discussed in Section 2.5.

The room temperature bandgap of MgTe is 3.57 eV [12]. The bandgap of
MgSe is more uncertain, with reported values ranging from 3.6 eV at 77K [13]
to 5.63 eV at 90K [14]. The first value gives a room temperature bandgap of
roughly 3.5 eV. In Chapter 4, the bandgap of MgSe is measured to be 4.54 eV
from studies of Mg,Cd,_,Se ternaries. The simulations in Sections 2.4.1 to 2.4.4
used the least favorable of the published values (3.5 eV) for the bandgap of MgSe.
Again, this does not affect the basic operation of the LED structures simulated,
but is important for the proper fabrication of actual devices.

With the knowledge of the bandgaps and band alignments for MgSe and
MgTe, we now observe that the MgSe to CdSe valence band offset is such that
a Mg,Cd,_,Se ternary of the proper composition can have a zero conduction
band offset to ZnTe while maintaining a valence band offset similar to that in
a CdSe/ZnTe heterojunction. From Figure 2.3, we see that by grading from CdSe
to Mg,Cd,_,Se we should be able to form a region that will facilitate electron
injection into the ZnTe layer, while the valence band offset between Mg,Cd _,Se
and ZnTe will continue to block hole injection. This will hold as long as the valence
band edge of Mg,Cd,_,Se is not more than a few tenths of an eV above the va-
lence band edge of CdSe, as confirmed by the measurements presented in Chapter
3. Thus, the operation of this device is not very sensitive to the validity of the
LCAO method or the common anion rule. This graded injector concept for wide
bandgap light emission was first proposed in Ref. [7]. Details of the operation of
this device and simulation results will be given in Section 2.4.2.

If we now study the Mg,Zn_,Te ternary in Figure 2.2, we see that it has a large
conduction band offset and a small valence band offset to ZnTe. This allows us
to use Mg,Zn,_,Te layers to confine electrons without significantly affecting hole
transport. Also, by forcing the bulk recombination to occur in a Mg,Zn,_,Te layer,
we can potentially extend the wavelength of light emitted from our device into the

blue. Sections 2.4.3 and 2.4.4 examine in more detail the use of Mg,Zn,_,Te layers

E, ZnTe p-ZnTe 2.25 eV
77 TT Ein Graded Epilayer Substrate
Mg,Cd,_,Se
1.74eV] n-CdSe Ep--~-----.- Y
0.64 eV |
E, Ib

Figure 2.3: Schematic diagram showing the conduction and valence band edges
for the graded injector device at flatband. The graded Mg,Cdj_,Se region facili-
tates electron injection into the higher bandgap material. The valence band offset
between the Mg,Cd,_,Se and the ZnTe epilayer remains abrupt, blocking hole in-
jection into the narrower bandgap CdSe material. Notice that the device operation
is fairly independent of the Mg,Cd,_,Se to CdSe valence band offset as long as it

does not deviate from the common anion rule by more that a few tenths of an eV.

in wide bandgap LEDs.

One major disadvantage of using these Mg-based ternary layers in the LED
designs is the large lattice mismatch that is introduced into the structure. One
possible approach which allows bandgap engineering without introducing signif-
icant strain is to use quaternary layers. Figure 2.4 shows some of the pos-
sible quaternaries available with lattice parameters near 6.1 A as well as the
Zn ;_xMg,S,Se;_y quaternary used in ZnSe-based LDs. Studies of devices incorpo-
rating Zn;_,Mg,SeyTe,_y and Cd;_,Mg,Se,Te;_, quaternaries and issues relating

to the growth of Se/'Te compounds will be presented in Section 2.5.

2.3 Model

Using the drift-diffusion transport equations, we can solve for the conduction and
valence band edges, as well as the charge and current densities as a function of
position. For a homostructure, the basic equations to be solved in the drift-diffusion
model are Poisson’s equation, the electron and hole current equations, and the
electron and hole continuity equations. In one dimension, assuming steady state

conditions and spatially constant temperatures, these are:

oe = (pnt. NF - Nz) (2.1)
Jp = CUppE —~ Dy (2.2)
In = CflynE + Dye (2.3)
te = ~—eU — eB(pn — n?) (2.4)
te = eU + eB(pn — nr?) (2.5)

where £ is the electric field, e is the electron charge, n; is the intrinsic carrier
concentration, n and p are the electron and hole concentrations, and ¢ is the

dielectric constant. D, yu, and j refer to the diffusion constant, mobility and current

Bandgap vs Lattice Constant

5.0 F 7

40, 7
S 3.0 + ;
())
ne)
Cc
aa)
@ 2.0 F 7
Lu

AISb CdTe
A GaAs InP
1.0 F Sj :
A AGaSb
Ge InAs, InSb
0.0 i: oun i

5.2 5.4 5.6 5.8 6.0 6.2 6.4 6.6
Lattice Constant, (Angstrom)
Figure 2.4: Diagram showing the bandgap and lattice parameter of II-VI (circles)

and other (triangles) semiconductor compounds. Lines are drawn to schematically

show some of the possible quaternary II-VI compounds.

density respectively, where the subscripts p and n refer to holes and electrons
respectively. B is the probability for radiative recombination and U is the indirect
recombination rate, described in detail below. Nj and Nz are the concentrations

of ionized donors and acceptors given by

+ Na
Na = 1+ 2exp ((Eyn — Ea)/ksT) (2.6)
- Na (2.7)

N- =
“1+ 4exp ((Ea — Epp) /keT)

where Ey, and E;, are the electron and hole quasi-Fermi levels, respectively. E,
and Ey are the acceptor and donor ionization energies respectively. The electron
and hole quasi-Fermi levels are determined from the carrier concentrations using
standard Fermi-Dirac statistics.

The van Roosbroeck-Shockley method of equating equilibrium band-to-band
recombination and absorption rates [15] is used to estimate B. The optical ab-
sorption coefficient required for this calculation is determined as detailed in Ref.

[16]. The final expression for the radiative recombination rate is

3/2 2
an heen, {1 1 1
B= 7 (—+4+—4-—_) fF? .
(ran + aay) 127 €y (= + Me + —) 9? (2:8)

where n, is the index of refraction, E, is the bandgap of the semiconductor, €p is

the permittivity of free space and c is the speed of light in vacuum.
The indirect recombination rate, U, for a single trapping level is given by

standard Shockley-Read-Hall statistics [17]:

np—n
U = i 2.9
(p + p1)Tr + (n +11) (2.9)

where 7, and 7, are the electron and hole lifetimes, and n; and p; are given by
ny, = N.exp((E: — E.)/kT) (2.10)

N, and N, are the effective densities of states for electrons and holes respectively,

and E, is the energy of the trap level.

To model heterostructures, equations (2.1) - (2.5) must be modified slightly
[18] [19] as follows:

dH e E de

o” = 2(N+ —~ NOK as 2.12
dx € (Na — Na —" +P) € dz (2.12)
Jp = €bppE — pyp Xt Fa) ~ eDpT + (2.13)
dy dn eD,ndQN.
— _ —*4 ep, —— ~ — 14
ju = Cllmn EB — [inn + €Dn = N. de (2.14)
Up _ 2
a —eU, — eU; — eB(pn — nz) (2.15)
xr
En y

where x and LH, are the electron affinity and bandgap of the semiconductor. U,
and U; refer to the bulk and interfacial recombination rates.

The physical interpretation of the additional heterostructure terms in Equa-
tions (2.12) - (2.16) are given below. More detailed derivations and discussions of
the above heterostructure drift-diffusion equations can be found in Refs. [18] and
[20]. The last term in Equation (2.12) simply imposes continuity of displacement
at the interface between two different materials. The second term on the right-
hand side of Equations (2.13) and (2.14) is a drift component due to a so-called
quasi-electric field for holes or electrons [20] [21] . The hole quasi-electric field,

Lad(x + E,)

2.
e de’ (2.17)

is an additional field influencing holes in graded regions and at interfaces, where
the valence-band has an additional slope due to valence-band offsets between the
materials in question. Similarly, electrons feel a quasi-electric field,
1 dy
-=, 2.18
edz (2.18)

due to conduction-band offsets. Known or estimated band offsets are used to de-

termine the positional variation of the electron affinity. The last term of Equation

(2.13) is a diffusion-like term resulting from a position-dependent effective density
of states (DOS). This term causes holes to diffuse from regions of low DOS to
regions of high DOS. A similar term appears in Equation (2.13) for electrons. The
bulk recombination, U;, is given by Equation (2.9). Finally, an additional inter-
facial recombination term has been added to Equations (2.15) and (2.16), since
interfacial recombination can have a significant effect on heterostructure device
operation. The interfacial recombination due to trap levels produced by dangling
bonds was implemented using a modified Shockley-Read-Hall model [19, 22], as

given below.

(np — n?)
U;,= + . 2.19
t(p + p1)/S, +t(n + n1)/S, (2.19)

where

Sr — UthOnNet (2.20)
Sp = Uh OpNst (2.21)

4(a3 — aj)
= 22
Not azaz (2 )

In Equation (2.19) S, and S, are the surface recombination velocities for electrons
and holes respectively and t is the thickness of the interface between the two
semiconductors. In Equations (2.20) - (2.22) u,, is the thermal velocity, Ng is
the trap density per unit area, 0, and o, are the capture cross sections for holes
and electrons respectively and a, and ag are the lattice parameters of the two
semiconductors comprising the interface. Note that Equation (2.19) assumes a
single trap level at mid gap, and Equation (2.22) assumes that the growth is in the
[100] direction, and that all of the lattice mismatch between the two semiconductors
comprising an interface is taken up by dislocations and not by strain. In regions
where part of the lattice mismatch is taken up by strain, Equation (2.22) gives the
upper limit for Ng.

In regions where the composition, x, varies, the dielectric constant and effective

mass are given by [23]

So =x(223) 40-9 (253) (2.28)

e+2 €y +2 €2 +2
1 1—
~*74-7% (2.24)
me mi m5

where x is the mole fraction of material 1. The bandgap in graded regions can be
modeled by [24]
E, = A+ Bx+Cx’, (2.25)

where A is the bandgap for material 1 and B and C are experimentally determined
parameters. The devices that we are modeling have graded regions where Equation
(2.25) is applicable; however, values for B and C are not known for the materials
in certain regions. In these regions, we have assumed a linear dependence of E,
on material composition. Also, we have neglected the effects of strain on band
structure in lattice mismatched graded regions.

In heterostructures, regions with high electric fields and carrier concentrations
are common, so the assumption of a constant mobility and the use of the non-
degenerate form of Hinstein’s equation is no longer valid. The mobility as a function

of both the field strength and the doping concentration is given by [25, 26]

* N. + Na Onpy~t
np
. . Veat ( E\4 E\‘*\7
Ln = [i + = (=) f + (=) | (2.27)
* —1
‘ bp
bp =o fi + (2.28)

where FE is the electric field, Ng and N, refer to donor and acceptor concentrations
respectively and subscripts n and p refer to electrons and holes respectively. Ny»,
Gn,p: Ho, Veat and vm are experimentally determined parameters. These parameters
are not generally known for the materials used in our devices, so as an initial

approximation, we have used N,,», Oy, and Bq as determined for GaAs. Ln0,p0

and {nip are determined by using the same ratio of Uno po tO nip: as for GaAs,
while constraining y* equal to experimental values of mobility in our materials for
zero field and low doping concentrations. The high field saturation velocities, V..,
and vu», are set to one-half the peak velocities estimated from Ey and yu*. While
not ideal, the method outlined above will give, to first order, trends in the change
in mobility with respect to variations in field strength and doping concentration. It
is necessary to include field dependent mobilities in the model, because in regions
with high fields the ratio of diffusivity to mobility becomes large. If the decrease
in mobility with increasing field is not accounted for, unrealistically large diffusion
constants will result.

As mentioned above, in regions of high carrier concentrations, the ratio of dif-
fusivity to mobility can deviate significantly from k,T'/e. If the original derivation
of the drift-diffusion equations from Boltzmann’s equation is done without assum-
ing Boltzmann statistics, then the following general form of Einstein’s relation is
obtained [20]:

D _ 2kyTF,/2(n)

Lh ef_yjo(n) ’ 228)

where F; is the Fermi-Dirac integral of order j, and 7 is (Es, — E.)/k,T for elec-
trons, and (E,, — E,)/k,T for holes. Analytic expressions approximating F, 2 and
F_1/2 [27] were used to evaluate Equation (2.29).

The numerical solution of Equations (2.12) - (2.16) was carried out using the
relaxation method [28]. Successful implementation relies on reasonable choices
of scaling factors and initial guesses [29]; the latter were determined as in Ref.
[30]. Variable mesh spacing was incorporated, with larger mesh densities used in
regions of rapid potential variation. Finally, abrupt junctions were graded over a
distance of a few angstroms to avoid numerical instabilities that are introduced by
singularities at an abrupt interface. The simulation results do not depend on the
grading distances of the interfaces as long as they are not unreasonably large. For

example, shortening the grading distance of an interface causes only the individual

derivative terms in Equation (2.13) to increase in magnitude, with no net effect on

the overall current density.

2.4 Simulation Results

2.4.1 CdSe/ZnTe

As discussed in Section 2.2, an n-CdSe/p-ZnTe heterojunction has favorable lattice
match and dopability characteristics for a wide bandgap LED; however, the band
alignments in this heterojunction favor interfacial recombination over thermionic
injection. In this section we examine results from numerical simulations of an n-
CdSe/p-ZnTe heterojunction in order to determine the feasibility of such a device.
All of the simulations presented below were done at room temperature. We as-
sumed a donor concentration of 1 x 10!° for n-CdSe and an acceptor concentration
of 1 x 10'’ cm? for p-ZnTe. Ionized donor and acceptor concentrations are given
by equations (2.6) and (2.7). For the interfacial traps we assumed a density of 1
x 10’! cm~? and cross sections of 1 x 107!* cm?.

Figure 2.5 shows the conduction and valence band edges for a CdSe/ZnTe
heterojunction at forward biases of 0.0 and 1.25 V. Figure 2.6 shows the electron
and hole concentrations at these same biases. At zero bias, we note that both the
CdSe and the ZnTe are depleted at the interface. As we apply a forward bias, the
depletion width decreases, and at a large enough bias, electrons and holes begin
to accumulate at the interface. Figure 2.7 shows the electron current densities at
0.75 and 1.25 V forward biases. Referring to Equations (2.15) and (2.16) we note
that the slope of a current density plot at a given point gives a direct measure of
the rate of recombination at that point. At a bias of 0.75 V, we see that all of
the recombination occurs at the CdSe/ZnTe interface. At a higher bias of 1.25 V,
all of the bulk recombination occurs in the CdSe region, indicating that holes are

being injected into the CdSe layer rather than electrons into the wider bandgap

37 CdSe (n+) ZnTe (p) :

Energy (eV)
ro)

x (Angstroms)

Figure 2.5: Conduction and valence band edges for an n-CdSe/p-ZnTe heterostruc-

ture at 0.0 V and 1.25 V forward bias. Zero on the horizontal axis corresponds to

the CdSe/ZnTe interface.

(a)
a IAL ELE Usenet ee
“~ V = 0.0 V 7

n -

ow O8 fF fo ee Pp 7]
“O pecorrcr tara
~ o P
rd J

x10 —>,

c a
bee a | aa a Lip [ee

500 1000 1500 2000

x (Angstroms)

(b)

20 rerrrrt rerrerrrrt Preeerrrrt pevrerrrry Trrerrrrry Teeerrrrrs Perens err
o~ iw6F V = 1.25 V :
1 -

3 n J
wo”
“> p |
= |
o. =
c 1
oh Redendenbecd Perea we de Die leeked BRendednt Dutt Snedeke k nd ee ba Lua

-100 0 100 200 300 400 500

x (Angstroms)

Figure 2.6: (a) Electron and hole charge densities for an n-CdSe/p-ZnTe het-
erostructure at 0.0 V forward bias. (b) At 1.25 V forward bias, high carrier con-

centrations occur at the heterojunction interface.

(a)

2.8 + ;
2.45 7

J. (107° A cm’)

-10 -5 O 5 10

(b)

J (107% A cm7”)
co

~0.2 Pree Pe er ee ee

-10 -5 0) 5 10
x (um)

Figure 2.7: Electron current densities for an n-CdSe/p-ZnTe heterostructure at (a)
0.75 V applied bias and (b) 1.25 V applied bias. Since the recombination rate is
proportional to the slope of the current density, this figure shows that interfacial
recombination dominates the device operation at lower applied biases, and that

holes are injected into the CdSe at higher applied biases.

ZnTe region. Thus, numerical simulations of an n-CdSe/p-ZnTe heterojunction
explicitly demonstrate that this design will not produce efficient light emission in
the wide bandgap ZnTe material. Instead, recombination in the CdSe and at the

CdSe/ZnTe interface will dominate.

2.4.2 Graded Injector

The next device that we present offers an improvement on the CdSe/ZnTe design.
By including a graded Mg,Cd1_,Se region between the CdSe and ZnTe layers, we
hope to facilitate electron injection while continuing to block hole injection into
the CdSe layer. The material parameters for the CdSe layer are the same as in the
previous device. Two types of ZnTe are used in this simulation: an undoped ZnTe
epilayer, and a lightly (1 x 10’’cm~%) doped p-ZnTe substrate with an oxygen
center 0.4 eV below the conduction band edge [31, 32]. The oxygen center is
needed to accurately model the substrates used in actual devices. Due to the poor
quality of the ZnTe substrates, recent devices have been grown on GaSb substrates.
Also, growth of p-type ZnTe layers has recently been achieved. These and other
recent developments to the graded electron injector structure are discussed in more
detail in Section 2.5.

The Mg,Cd,_,Se layer is graded from zero Mg concentration to a concentration
such that there is no conduction band offset to ZnTe (see Figure 2.3). Since Harri-
son’s LCAO method predicts a small valence band offset between CdSe and MgSe
and since small changes in the CdSe/MgSe valence band offset do not significantly
affect the operation of this device, we use the common anion rule in our calcu-
lations for this device. For a MgSe bandgap of 3.5 eV, this gives a value of 0.66
for x at the Mg,Cd,_,Se/ZnTe interface. The doping of the graded Mg,Cdj_,Se
region is n-type at the CdSe interface and undoped at the ZnTe interface. These
doping parameters account for the likelihood that Mg,Cd,_,Se will not be dopable

n-type for high Mg concentrations. The thickness of the graded region was chosen

to be 200 A. A thinner graded region decreases the accumulation of electrons near
the CdSe/Mg,Cd,_,Se interface, making electron injection more difficult, while
increasing the thickness might prevent the graded Mg,Cd,_,Se from being coher-
ently strained to the ZnTe.

Figure 2.8 shows the conduction and valence band edges for this device at 0.0
and 2.0 V forward bias. Figure 2.9 gives the hole and electron charge densities at
the same biases. At zero bias the ZnTe layer is depleted at its interface with the
graded Mg,Cdy_,Se region, while electron accumulation occurs at the interface be-
tween the CdSe and the graded Mg,Cd;_,Se region. As a forward bias is applied
to the device, the depletion width in the ZnTe layer decreases and eventually holes
accumulate at the Mg,Cd,_,Se/ZnTe interface. An important point to note here
is that in the CdSe/ZnTe device, electron and hole accumulation at the hetero-
junction interface led to high rates of non-radiative recombination. In the graded
device, however, electron and hole accumulations are spatially separated, leading
to a considerable reduction in the amount of non-radiative recombination occur-
ring at the interfaces. As we continue to increase the forward bias, Figure 2.10
shows that electron injection occurs prior to any significant hole injection. Recall
that the slope in the current density is a measure of the recombination rate, so
the right-hand side of Figure 2.10 shows that injected electrons are recombining
in the ZnTe material. Since the total current is constant throughout the device,
the slope in j, is simply the negative of the slope of j,. Notice that interfacial
recombination dominates the operation of the device only at lower voltages, and
that bulk recombination begins to dominate at higher voltages.

The graded Mg,Cd,_,Se layer facilitates electron injection as follows. At zero
bias, the barrier to electron injection is essentially the same in the graded device
as it is in the CdSe/ZnTe device. As we forward bias the CdSe/ZnTe device,
the barrier to electron injection remains essentially constant, determined by the
fixed conduction band offset between the two materials. In the graded device,

however, as we apply a forward bias, more and more electrons accumulate within

3F CdSe (n+) ZnTe ZnTe (p) :
Epilayer
2r q
Graded
Mg Cd Se
x 1-x
1F- 4
Fo’
~V~ E
5 of—
i.
c ;
Li
r Vv
—2 .
TSP ne V = 2.0 V :
-1000 -500 @) 500 1000 1500 2000

x (Angstroms)

Figure 2.8: Conduction and valence band edges for the graded injector device at
0.0 and 2.0 V forward bias. In this figure and in all succeeding figures, zero on
the horizontal axis corresponds to the right-hand side of the graded Mg,Cd,_,Se

layer.

(a)

7 oe Pes Terr rt Terr rrr reer Tr
o 6 fT V = 0.0 V ,
Ve 5 + n ;
Cae ee | P |
fo) BL
oS ee =~

2+L cect q
Qa L ¢

0 a Dae ow or eT —<—— Po a oe | a er

—500 0 500 1000 1500 2000 2500

x (Angstroms)

(b)

np (10'°cm7>)

Se oe oe [alleles llandllkantRanaallliaslliikr lt vullaastllen AD ssalllianstlandlicatllay lla tllatassdllaadien Eicsdllandtthstile

500 1000 1500 2000 2500

x (Angstroms)

Figure 2.9: Electron and hole charge densities for the graded injector device at (a)
0.0 V forward bias and (b) 2.0 V forward bias. Notice the spatial separation of

accumulated charge.

1.24 ;

0.6 L ;
0.4 | /

“-_

re) I ;
< 0.8 +

aw j
c 0.2 + 4
> I J

120 peepee a a es

100 + .
80 + Vz 20V 7

£ .
e 60 + 4
x ‘

20 | |

x (um)

Figure 2.10: Electron current density for the graded injector device at (a) 1.6 V
forward bias and (b) 2.0 V forward bias. The second flat portion in (a) corresponds

to the ZnTe epilayer.

the graded Mg,Cdj_,Se layer. This accumulation of electrons causes the bands to
bend downwards, as seen in Figure 2.8, thereby reducing the barrier to electron
injection.

Figure 2.10 shows the electron and hole current densities at 1.6 and 2.0 V
forward bias. Very few of the injected electrons actually recombine in the ZnTe
epilayer, since the diffusion length in the ZnTe epilayer is much greater than the
epilayer thickness used in our devices. Most of the electrons recombine in the ZnTe
substrate, which results in red light emission because of the oxygen center. Thus,
for the purposes of high efficiency green light emission, and for laser applications,
we need to be able to confine injected electrons to the ZnTe epilayer. In more
recent devices grown on GaSb substrates, the p-ZnTe is grown thicker, but is still
less than the diffusion length so the confinement of injected electrons is important

for these devices as well. The next design proposes a method to achieve this.

2.4.3. Graded Injector with Confining Layer

To confine injected electrons to the ZnTe epilayer, we refer again to Figure 2.2.
This figure shows that a Mg,Zn,_,Te ternary will have a larger bandgap than
ZnTe, with most of the difference going into the conduction-band. We propose
to insert a MgyZni_yTe confining layer between the ZnTe epilayer and the ZnTe
substrate. Again, there have been no reports in the literature of experimental
measurements of the valence band offset between MgTe and ZnTe (see Chapter 4);
however, the LCAO method predicts a small valence band offset, which is desirable
for our device operation. In our calculations, we assume a less favorable case with
two-thirds of the band offset going into the conduction-band and one-third in the
valence-band. We do this to demonstrate that the performance of our devices is
not highly dependent on the validity of the common anion rule.

Using the MgTe bandgap of 3.57 eV and choosing a conduction-band barrier
height of 0.5 eV gives a value y = 0.57 at the Mg,Zn,_yTe/ZnTe epilayer interface.

3, CdSe (n+) Graded ZnTe (p) 7
Mg Zn Te
y I-y
Graded
Mg Cd
x 1-
1 _
2 E
3 8)
LJ
—1 _
Vv
—2 +4
—3 fF ---- V=21V :
-1000 -500 0 500 1000 1500 2000

x (Angstroms)

Figure 2.11: Conduction and valence band edges for the graded injector device with
an electron confining layer at 0.0 and 2.1 V forward biases. Note the undesirable
deviation from the common anion rule used for the Mg,Zni-yTe to ZnTe band

offset.

oN
arr

Q pee a a a oe a

8 f 1

“oo 7 : V = 0.0 V
' 6 | n j
ad ) es P 7
"Oo 4 i
Aw, 3 f a
mn 2 I x40 ——> eee :

0 p= ---+ Lape Do ——--__ it—t_L_ id tk |

—500 0 500 1000 1500 2000 2500

x (Angstroms)

(b)

np (10'8cm7*)

-—500 6) 500 1000 1500 2000 2500

x (Angstroms)

Figure 2.12: Electron and hole charge densities for the graded injector device with

a confining layer at (a) 0.0 V forward bias and (b) 2.1 V forward bias.

30 a
a V= 2.1 V
Ec 20 + 7
re
<<
~ 10 + a

O +

x (um)

Figure 2.13: Electron current densities for the graded injector device with a
confining layer at a forward bias of 2.1 V. The two regions showing a very
large slope in the current density plot correspond to the Mg,Cd,_,Se/ZnTe and
ZnTe/Mg,Zn;_,yTe interfaces respectively.

The Mg, Zn,_,Te confining layer is graded to avoid blocking holes from entering the
ZnTe epilayer. If a more favorable valence band offset is used, then the Mg,Zn,_yTe
confining layer need not be graded and the Mg concentration will be less for the
same barrier height, resulting in less strain. The width of the graded Mg,Zn,_,Te
region only needs to be thick enough to prevent significant electron tunneling
through to the ZnTe substrate. We use 75 A in our calculations, but this can be
increased significantly, if required, without introducing too much strain.

Figure 2.11 shows the conduction and valence band edges for this device at
forward biases of 0.0 and 2.0 V. Figure 2.12 gives the hole and electron charge
densities at these same biases. Until the onset of electron injection, this device
behaves in much the same way as the previous one. After the onset of electron

injection, however, we see that the confining layer prevents electrons from entering

the ZnTe substrate. In Figure 2.13, the electron current densities for this device
at a forward bias of 2.0 V show that with the addition of the electron confining
layer, the bulk recombination in the device is confined to the ZnTe epilayer. Also
notice that interfacial recombination comprises a significant portion of the total
recombination in the device. In general, interfacial recombination plays a larger
role at small biases and becomes less significant as the bias is increased; however,
quantitative calculations of radiative recombination rates and quantum efficiencies

will depend largely on the specific properties of the interfaces in each device.

2.4.4 Tunable Bandgap LED

We next consider a device design that will allow the wavelength of emitted light
to be tuned continuously from the green to the blue wavelength regimes. This
design was first proposed in Ref. [7]. The basic idea behind the device is to
have the radiative recombination occur in a MgyZny_yTe layer, where y can be
adjusted to obtain the desired band-to-band recombination energy. Figure 2.14
shows a schematic of the proposed device at flatband. Note that we have used
the common anion rule in simulating this device, so there is no valence band offset
between ZnTe and MgZnTe ternaries. This allows us to forego the grading of the
Mg,Zn,_,Te confining layer that was needed in the previous device. We do this for
convenience and clarity of presentation only; the device concept does not depend
on the validity of the common anion rule. Figure 2.15 shows the conduction and
valence band edges of the device at 0.0 and 2.25 V forward bias. Figure 2.16 shows
the hole and electron charge densities at these same biases. From these figures
we again see that the graded Mg,Cdy_,Se layer enhances electron injection into
the wider bandgap material, while spatially separating regions of interfacial charge
accumulation. Figure 2.17 shows the electron current densities for this device at
a forward bias of 2.25 V. In this figure we see that all of the bulk recombination
occurs in the Mg,Zn,_,Te epilayer. Thus, the wavelength of light emitted from

Mg,Zn,.,Te

MgyZn,.yTe 7
WZ

Adjustable

E — Bandgap =. ante 2.25 eV
L Et, Graded Substrate
Mg,Cd,.,Se
1.74eV] n-CdSe y Epo. ------- Y
0.64 eV

Ey Y A Ip

Figure 2.14: Schematic diagram showing the conduction and valence band edges
for the tunable bandgap LED at flatband. Electrons are injected into the wide
bandgap Mg,Zn;_yTe but are blocked from entering the ZnTe by the Mg,Zn,_,Te

layer.

57 CdSe (n+) ZnTe (p) 7
Mg Zn Te
z 1—z
2} |
Graded
Mg Cd
1+ x 1 7
“oo
> }
ee cE.
o 0
Lid
—1 b
F OV
—2 +
—3 ‘ 4
-1000 -500 0 500 1000 1500 2000

x (Angstroms)

Figure 2.15: Conduction and valence band edges for the tunable bandgap LED at
0.0 and 2.5 V forward bias.

(a)
8 open ror res Torre a Terr Peers 1
— V = 0.0 V
E n :
i P ;
fo)
L 1
c x40 — aa ]
500 1000 1500 2000 2500
x (Angstroms)
(b)
8 pees roe reer rors reer reer
7 - a
m et V = 2.5 V
E 5 | n 5
0 4 <<— x0.5 om p J
fo) I
oS 3 1
a. 2 4
c 1 § q
0 b-- ~~~. ae oe awe a ee [a oe a oe | oe or a ee | a eT

—500 0 500 1000 1500 2000 2500

x (Angstroms)

Figure 2.16: Electron and hole charge densities for the tunable bandgap LED at
(a) 0.0 V forward bias and (b) 2.5 V forward bias. Notice the accumulation of
both holes and electrons in the Mg,Zn,_,Te layer at 2.5 V bias.

280 + 7
240 t ’
200 | V=z25VvV_ ,
160 | ;
120 } :
BO | 4

JA em™’)

~0.2 -0.1 0.0 0.1 0.2
x (um)

Figure 2.17: Electron current density for the tunable bandgap LED at 2.5 V for-
ward bias. The slope in the electron current density plot in the Mg,Zn,_,Te layer
indicates successful electron injection and confinement with this layer. The wave-

length of light emitted from this layer can be tuned into the blue.

this device can be varied from green to well into the blue wavelength regime,
depending on the the Mg concentration used in the Mg,Zn,_yTe epilayer. The
tunable bandgap LED operates in much the same manner as the previous device,

with the exception that the wavelength of light emission is tunable from green to

blue.

2.5 Recent Developments

2.5.1 Current LED Structure

As was mentioned, the four devices simulated in Section 2.4 did not incorporate
some recent developments to the graded electron injector design. These modifi-
cations do not affect the validity of the simulations, but are important factors in
the fabrication of actual devices. For completeness, these recent developments and
their influence on the device structure are discussed below.

Early LED structures using the graded injector scheme were grown on lightly
doped ZnTe substrates purchased from Eagle-Picher. These substrates were ini-
tially very useful, because p-type doping of MBE grown ZnTe had not yet been
developed, so only thin ZnTe epilayers could be used in the LEDs before series
resistance became too large. In effect, the ZnTe substrates were used for the active
regions wherein the radiative recombination took place. At roughly the same time
that the very efficient N-doping of ZnTe was developed, it was also discovered that
the ZnTe substrates being used had serious problems with surface and structural
quality, as well as reproducibility problems. These two developments led to the use
of GaSb substrates instead of ZnTe substrates, and thick p-type ZnTe epilayers for
the active regions. This made the processing somewhat more complicated since
both contacts to the LED had to be made from the top of the sample, but it did
not affect the basic operation of the device.

The other recent developments were measurement of the MgSe bandgap and
the MgSe and MgTe band alignments. The primary impact of these measurements
was to change the desired composition of the Mg,Cd,_,Se graded region at the
Se/Te interface. A larger MgSe bandgap means that less Mg is required to grade
the conduction band edge to match that of ZnTe, while a lowering of the valence
band edge with the addition of Mg necessitates more Mg at the Se/Te interface.
Accounting for both the bandgap and the band offset measurements, the desired

Mg,Cd,_,Se composition at the Se/Te interface is roughly 70% Mg and 30% Cd.

2.5.2 Possible Quaternary LED Designs

One of the primary disadvantages of the graded injector concept is the strain
induced in the graded region. Preliminary TEM studies (see Chapter 4) have
shown that this region can have a very high density of defects. Another problem
lies in the CdSe layer which, if too thick, will absorb a substantial fraction of
the light emitted from the ZnTe active region. To solve these problems, it is
necessary to use either ternary or quaternary layers which have a range of bandgaps
and lattice parameters. The added advantage in using quaternary semiconductors
is that the bandgap and lattice parameter can be independently varied. This
section proposes one possible LED design based on the use of quaternary II-VI
semiconductor compounds.

To gain insight into the design of wide bandgap LEDs using quaternary lay-
ers we first study the Zn,_,Mg,S,Sei_, quaternary already used in ZnSe-based
LDs and LEDs. This quaternary is indicated schematically in Figure 2.4 by
the lines connecting ZnSe, MgSe, MgS, and ZnS. We see that the bandgap of
Zni-xMg,S,Se,_, can be varied while still maintaining complete lattice match
to either ZnSe or GaAs. The difficulty currently being encountered is that Mg
and S$ both pull down the valence band edge when added to ZnSe. This makes
Zni-xMg,SySe;_y quaternaries layers even less p-type dopable than the already
difficult to dope ZnSe. Since the Zn,-xMg,S,Se;_y is being used for the thick
cladding layers, series resistance in these layers may eventually be the limiting fac-
tor for attaining shorter wavelength LEDs and LDs. Also, some of the reliability
problems in these devices may be related to the difficulties in p-type doping either
in the form of contacting problems or in the active layers.

One way to avoid this problem is to use two different quaternary layers, one of

which is easily dopable p-type and the other n-type, while maintaining the design

4 Mgse

Mgle

Want to work in Ce Cate

this region

Energy Bandgap (eV)

Sse-rich: easily dopable n-type
Te-rich: easily dopable p-type

Lattice Constant (Angstroms)

Figure 2.18: Schematic diagram of the Zn;_,Mg,SeyTe;_, and Cd _,Mg,SeyTei_,
quaternaries. Shaded areas denote the Te-rich and Se-rich composition which
should allow easy p-type and n-type dopability respectively. Where the two shaded

regions overlap, both n- and p-type dopability can be achieved without sacrificing

lattice matching.

constraints of close lattice match and substrate availability. An example of such a
combination is shown in Figure 2.18. The two quaternaries, Zny_,Mg,Se,Te,_y and
Cd,_xMg,SeyTe;_,, indicated by the solid lines, have regions of overlapping lattice
parameter which can also be lattice matched to GaSb, InAs or ZnTe substrates.
The advantage in using two quaternary layers can be noted by observing that Se-
based semiconductors such as ZnSe and CdSe tend to be easily dopable n-type,
while Te-based semiconductors such as ZnTe and CdTe tend to be easily dopable
p-type (see also Figures 2.1 and 2.2). Thus, Te-rich Zn1_xMg,SeyTei_, should be
p-type dopable, while Se-rich Cd,_,Mg,Se,Te;_, should be n-type dopable.

With the addition of Mg and small amounts of Te to CdSe, and small amounts
of Se and Mg to ZnTe, we should be able to obtain two semiconductors with the
following properties: 1) perfect lattice match to each other and to a suitable sub-
strate, 2) both p- and n-type dopability, 3) flexibility in bandgap engineering at
a single lattice parameter, and 4) both layers can have bandgaps larger that the
wavelength of emitted light. A schematic diagram of an LED structure incorpo-
rating such layers is shown in Figure 2.19. As shown in this figure, another critical
element to the successful implementation of an LED based on two quaternary lay-
ers is the graded injector layer which facilitates carrier injection into the desired
semiconductor layer.

To implement the design described above, a number of very important exper-
imental concerns must be addressed. The most critical of these is the existence
of miscibility gaps for certain ternary and quaternary compounds. In particular,
ZnTe;_ySey is known to have miscibility gaps under certain growth conditions;
however, researchers at Bellcore have been successful in growing ZnTey_ySey over
the entire range of composition using MBE [33]. Despite this encouraging re-
sult, it still remains to be determined whether or not the Zn,_xMg,Se,Te;_, and
Cdi_xMg,SeyTe;_y quaternaries can actually be grown with the desired composi-
tion. Another important, but not critical, factor is band bowing. Band bowing

in the quaternaries of interest will require higher Mg and Se or Te concentrations,

ee — EC

Graded MgZnSele MgZnSelTe

MgCdSele
MgCdSele —

(n-type) (p-type)

Figure 2.19: Schematic diagram of fully lattice matched graded injector structure.
The use of quaternaries allows flexibility in bandgap engineering while still allowing
lattice matched conditions. The use of two quaternaries also facilitates both n- and

p-type dopability.

and if great enough, could prevent the development of shorter wavelength blue and

violet LEDs and LDs.

2.6 Summary

We present the numerical simulation and design of novel wide bandgap II-VI LEDs.
The drift-diffusion model, modified to account for heterojunctions, is used in the
simulations. All of the designs we present incorporate both n-CdSe and p-ZnTe.
Zn'Te is chosen because it is the only wide bandgap II-VI compound that can
be doped p-type using conventional techniques. CdSe, with its small lattice mis-
match to ZnTe and its n-type dopability, is then the best compound to use for
the formation of a II-VI p-n heterojunction with ZnTe. A simple n-CdSe /p-ZnTe

heterostructure is first analyzed. Simulation results for this design show high inter-
face recombination rates and hole injection into the smaller bandgap CdSe layer.
The CdSe/Mg,,Cdi_,Se/ZnTe device proposed by Phillips et al. improves on the
CdSe/ZnTe heterojunction design by incorporating a graded Mg,Cdj_,Se electron
injector. The addition of this layer significantly reduces interface recombination
and facilitates electron injection into the wide bandgap ZnTe layer. A further im-
provement on this design utilizes a Mg, Zn,_,Te confining layer which restricts bulk
recombination to the ZnTe epilayer. A tunable bandgap LED is demonstrated, in
which the bulk recombination occurs in a Mg,Zn,_,Te epilayer. The Mg concen-
tration then determines the wavelength of the light emitted from the device, which
can be varied from green to well into the blue wavelength regime. Simulations
of a device incorporating recent materials characterization results are presented
and, finally, a design that potentially allows fully lattice-matched structures with

flexible bandgap engineering is proposed.

Bibliography

[1] J. Qiu, J.M. DePuydt, H. Cheng, and M.A. Haase, Appl. Phys. Lett. 59, 2992
(1991).

[2] M.A. Haase, J. Qiu, J.M. DePuydt, and H. Cheng, Appl. Phys. Lett. 59, 1272
(1991).

[3] Y. Fan, J. Han, L. He, J. Saraie, R.L. Gunshor, M. Hagerott, H. Jeon, A.V.
Nurmikko, G.C. Hua, and N. Otsuka, Appl. Phys. Lett. 61, 3160 (1992).

[4] F. Hiei, M. Ikeda, M. Ozawa, T. Miyajima, A. Ishibashi, and K. Akimoto,
Electronics Letters 29, 878 (1993).

[5] Y. Lansari, J. Ren, B. Sneed, K.A. Bowers, J.W. Cook, Jr., and J.F. Schetzina,
Appl. Phys. Lett. 61, 2554 (1992).

[6] J.O. McCaldin, T.C. McGill, and C.A. Mead, Phys. Rev. Lett. 36, 56 (1976).

[7] M.C. Phillips, M.W. Wang, J.F. Swenberg, J.O. McCaldin, and T.C. McGill,
Appl. Phys. Lett. 61, 1962 (1992).

[8] J.O. McCaldin, J. Vac. Sci. Technol. A 8, 1188 (1990).
[9] W.A. Harrison, J. Vac. Sci. Technol. 14, 1016 (1977).

[10] E.T. Yu, M.C. Phillips, J.O. McCaldin, and T.C. McGill, J. Vac. Sci. Technol.
B 9, 2233 (1991).

[11] J.O. McCaldin, Growth and Optical Properties of Wide-Gap II-VI Low-
Dimensional Semiconductors, edited by T.C. McGill, C.M. Sotomayor Torres,
and W.B. Gebhardt (Plenum, New York, 1989), pp. 39-51.

[12] J. Marine, T. Ternisien D’ouville, B. Schaub, A. Laugier, D. Barbier, J.C.
Guillaume, J.F. Rommeluere, and J. Chevallier, Journal of Electronic Mate-

rials, 7, 17 (1978).

[13] H. Okuyama, K. Nakano, T. Miyajima, and K. Akimoto, Jpn. J. Appl. Phys.
Part 2-Letters 30, L1620 (1991).

[14] H. Mittendorf, Z. Physik 183, 113 (1965).
[15] W. van Roosbroeck and W. Shockley, Phys. Rev. 94, 1558 (1954).

[16] R.N. Hall, Proc. Institution of Electrical Engineering 106B, Suppl. No. 17,
923 (1960).

[17] W. Shockley and W.T. Read, Phys. Rev. 87, 835 (1952).
[18] A.H. Marshak, IEEE Trans. Electron Devices 36, 1764 (1989).

[19] J.E. Sutherland and J.R. Hauser, IEEE Trans. Electron Devices 24, 363
(1977).

[20] A.H. Marshak and K.M. van Vliet, Solid-State Electron. 21, 419 (1978).

[21] M.S. Lundstrom, Fundamentals of Carrier Transport (Addison-Wesley, Read-
ing, 1990).

[22] P.E. Dodd, T.B. Stellwag, M.R. Melloch and M.S. Lundstrom, IEEE Trans.
Electron Devices 38, 1253 (1991).

(23| J.W. Harrison and J.R. Hauser, J. Appl. Phys. 47, 292 (1976).

[24] A.G. Thompson and J.C. Woolley, Canadian J. of Phys. 45, 255 (1967).

[25] M. Kurata and J. Yoshida, IEEE Trans. Electron Devices 31, 467 (1984).
[26] H.C. Chan and T.J. Shieh, IEEE Trans. Electron Devices 38, 2427 (1991).

[27] X. Aymerich-Humet, F. Serra-Mestres, and J. Millan, Solid-State Electron.
24, 981 (1981).

[28] W.H. Press, B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling, Numerical
Recipes (Cambridge, New York, 1986).

[29] G.B. Tait, Solid-State Electron. 32, 369 (1989).
[30] L.J. Turgeon and D.H. Navon, IEEE Trans. Electron Devices 25, 837 (1978).

[31] J.L. Merz and R.T. Lynch, II-VI Semiconducting Compounds 1967 Interna-
tional Conference, 731 (1967).

[32] J.J. Hopfield, D.G. Thomas, and R.T. Lynch, Phys. Rev. Lett. 17, 312 (1966).

[33] R.S. Turco-Sandroff, R.E. Nahory, M.J.S.P. Brazil, R.J. Martin, and H.L.
Gilchrist, Appl. Phys. Lett. 58, 1611 (1991).

Chapter 3

XPS Measurements of Band
Offsets for Mg-based

Semiconductor Compounds

3.1 Introduction and Outline

The importance of Mg-based semiconductor compounds for II-VI wide bandgap
light emitters has recently been demonstrated. In particular, the blue-green laser
diodes (LDs) initially demonstrated by Haase et al. [1] are now fabricated with
Zn,_,Mg,Sy5e1_y cladding layers [2], resulting in the demonstration of the first
blue LD [3] and the pulsed operation of pseudomorphic blue-green LDs up to
temperatures as high as 394 K [4]. The latest results report blue (489.9 nm)
and green (523.5 nm) LDs incorporating multiquantum well ZnSe/ZnTe contacts
showing thresholds of < 4 V and 400 A/cm?, and room temperature CW operation
for greater than 9 minutes [5].

In addition, compositionally graded Mg,Cd,_,Se used as an electron injector in
the LED design proposed by Phillips et al. [6] results in devices with nearly ideal

current-voltage characteristics and promising device lifetimes [7]. There is also

interest in Mg,Cd,_,Te for light emitter applications, and growth of Mg,Cd,_,Te
by molecular-beam epitaxy (MBE) has recently been achieved [8]. Finally, simula-
tions show that Mg,Zn,_yTe layers can be used to shorten the wavelength of light
emission in the graded electron injector devices [9].

As of yet there has been very little characterization of Mg-based semiconduc-
tors. In particular their valence band offset, or band alignment, to other semicon-
ductor compounds has not been studied, except for a recent calculation reported
by Nakayama [10]. To optimize the performance of these devices, an accurate
measurement of the band offset between Mg-based semiconductors and other ma-
terials will be required. The importance of the valence band offset to LED and
LD design is considerable: the appropriate Mg concentrations needed in both
the graded Mg,Cd,_,Se region and the Zn,_.Mg,S,Se,_, cladding layers are de-
pendent on the respective valence band offsets, especially as the wavelength of
light emission is extended into the blue. Currently, the LD designs incorporating
Zn;_xMg,SySe;_y layers assume a type I band lineup between Zn1_-xMg,SySey_y
and ZnS;_,Se,, based on the common anion rule and experimental data in Ref.
[2] and on a semi-empirical tight binding method in Ref. [11]. The devices using
Mg, Cd,_,Se for a graded injector were also designed using the common anion rule.

The original common anion rule [12] was observed not to apply to the one
cation, Al, then studied from the third row of the periodic table, so we should not
expect it to necessarily apply to Mg, another third row cation. Similarly, Wei and
Zunger [13] predict that the common anion rule applies only when the d orbitals of
the cations on both sides of a heterojunction, e.g., MgSe /Cdo.54Zng.4gSe, interact
with their respective valence bands in a comparable manner. Since the 3d orbitals
in Mg are unoccupied, as in the case of Al, we might expect a deviation from the
common anion rule similar to that observed in the AlAs/GaAs system.

In this chapter, we report the measurement of AE, in the lattice matched
MgSe/Cdos4Zno.4gSe (5.89 A, zincblende [2]) and MgTe/Cdo.sZno.12Te (6.435
A, zincblende [8]) heterojunctions by X-ray photoelectron spectroscopy (XPS).

Cdo.54Zno.ag5e and Cdo9gZnoi2Te were chosen to avoid difficulties associated
with measuring band offsets in lattice mismatched systems, and to test the va-
lidity of the common anion rule for Mg based compounds. Using our mea-
sured value for the MgSe/Cdo.54Zno,.4gSe valence band offset, we can estimate the
Zn;_xMg,SySe;_,/ZnS,_,Se, and Mg,Cd,_,Se/CdSe valence band offsets by linear
interpolation.

Section 3.2 of this chapter describes the sample growth, and the XPS experi-
mental setup. Section 3.3 outlines the XPS data analysis. The band offset results

are presented in Section 3.4, and the chapter is concluded with Section 3.5.

3.2 Experiment

3.2.1 Sample Growth

The structures studied here were grown in two Perkin-Elmer 430P molecular-beam
epitaxy (MBE) systems, one devoted to III-V and the other to II-VI semiconduc-
tor growth. GaSb buffer layers were grown on GaSb (100) substrates to provide
a smooth growth surface. After the GaSb growth, the samples were transferred
via an ultrahigh vacuum (UHV) transfer tube to the II-VI growth chamber. Thick
(> 3000 A), relaxed Cdo.54Zno.4g5e (Cdo,sgZno.12Te) layers, followed by the band
offset structures, were grown at a substrate temperature of 270°C (300°C) and
a growth rate of approximately 53 A/ min. The Cdo54Zng.4g5e and Cdo.ggZno.12Te
composition were calibrated using XPS and X-ray diffraction, to keep the lattice
mismatch to a minimum. The measured Cdo.54Zno,4gSe and Cdo.sgZno 12Te lattice
parameters were 5.93A and 6.43A allowing us to neglect their small lattice mis-
matches (0.687% and 0.06%) to MgSe and MgTe respectively. All layers were grown
in the cubic zincblende structure as indicated by reflection high energy electron

diffraction patterns.

3.2.2 XPS Measurements

Following the II-VI growth, the samples were transferred via a UHV transfer tube
to a Perkin-Elmer Model 5500 analysis system equipped with a monochromatic
Al Ka X-ray source (hy = 1486.6 eV) and a spherical capacitor electron-energy
analyzer with multichannel detection capability. The base pressure in the XPS
chamber was typically in the low 107!° Torr range. Care was taken to ensure
that the escape orientation of the photoelectrons remained constant from sample
to sample to minimize any electron diffraction effects due to the single crystalline
nature of the samples. The election binding energy scale was routinely calibrated
using binding energy peaks from a sputter cleaned calibration sample consisting
of Au, Cu and Al. Details of the determination of the instrumental resolution
function from acquired Au spectra are given in Section 3.3. XPS energy sepa-
rations measured on the same sample under identical conditions were typically
reproducible to better than + 0.02 eV, and all samples yielded energy separations
that were reproducible to within experimental error.

For the Cdo54Zno.4g5e/MgSe band offset measurement, the core level to va-
lence band edge energy separations were measured on two bulk Cdo54Znop a4gSe and
two bulk MgSe samples. The core level binding energy separations were mea-
sured on two thin (~ 20 A) Cdo.54Zno.4g9e on MgSe samples, and one thin MgSe
on Cdo.54Zno.4gSe sample. Thin overlayer thicknesses are necessary to acquire a
reasonable photoelectron signal from both sides of the heterojunction, since pho-
toelectron escape depths are ~ 10-20 A. For the Cdg ggZno.12Te /MgTe band offset
measurement, the core level to valence band edge energy separations were measured
on one each of bulk Cdo.ggZno.12Te and bulk MgSe samples. The core level binding
energy separations were measured on one Cdg.ggZno.42Te on MgTe sample. Com-
mutativity of the valence band offset was not verified for the Cdo.sgZno.12Te/MgTe
band offset measurement. Figures 3.1 and 3.2 show representative scans required

for the Cdo.54Zng.46Se/MgSe and CdoggZno.12Te/MgTe band offset measurements

respectively.

3.3 XPS Data Analysis

To determine the MgSe/Cdo.54Zmo.46Se and Cdo ggZno.12Te/MgTe valence band off-

sets using XPS, the following relation was used:
AE, = (Ezn 3p 3/2 — EC%X) _ (Bagg 26 — EM9%*) — (Ezn 3p 3/2 ~ Eq 2s); (3-1)

where the X in CdZnX and MexX refers to either Te or Se. The core level to
valence band maximum, E,, binding energy separations were measured on bulk
CdZnX and bulk MgX samples, and the core level binding energy separations were
measured on thin (~ 20A) CdZnX on MgX samples and thin MgX (~ 20 A) on
CdZnX samples as shown in Figures 3.1 and 3.2. For MgSe and MgTe, the use of
either the Mg 2s or the Mg 2p core level peaks in the analysis did not affect the
results. For Cdo.54Zng.4gSe and Cdo.gZno.12Te, the only resolvable core level peak
was Zn 3p, since Se and Te are common to both sides of the junction and both the
Cd 4d and Zn 3d peaks overlap with states from the valence band.

To determine the core level peak positions, we first performed an integrated
background subtraction and then fit the peaks to Voigt functions, allowing the
binding energy, HWHM, Lorentzian fraction and intensity to vary. In the case of
the MgX/CdZnX heterojunction samples, the Zn 3p 3/2 and Mg 2s core levels
overlap, so it was necessary to constrain the peak shapes to those obtained from
bulk MgX and CdZnX samples, so as to reduce the number of fit parameters from
eight to four. A representative fit for the Zn 3p 3/2 and Mg 2s core levels for a
sample consisting of 20 A MgSe on Cdo54Zno.4g5e is shown in Figure 3.3. In the
fitting procedure, only the intensity and binding energy are allowed to vary for
each of the two core levels (Zn 3p 3/2 and Mg 2s ).

The determination of the precise energy of the valence band edge for the bulk

semiconductors is complicated by the fact that instrumental broadening smears out

a)
(a)
EZn ap /2 - Ey
CdZnSe
—~ | (b)
= - 3
GJ ad Emg as ~ Ezn 3p! Caznse
ra
aT, M
2 gSe
2 wn
(Cc)
—_ Eng 2s - Ev
MgSe
oN
100 80 60 40 20 0

Binding Energy (eV)

Figure 3.1: XPS spectra for (a) bulk Cdos54Znmp.46Se, (b) thin (~ 20 A)
Cdo.54Zno.4gSe on MgSe, and (c) bulk MgSe. Energy separations used in the band

offset measurement are labeled.

(a)
zn apo - Ey >
CdZnTe
= (b) ;
a —|— Eng 287 Ezn ap~/2 Capnto
ran
= MgTe
Cc
(Cc)
Emg 2s” Ey
MgTe
_A
100 80 60 40 20 0

Binding Energy (eV)

Figure 3.2: XPS spectra for (a) bulk CdoggZno12Te, (b) thin (~ 20A)
Cdo.egZno.12Te on MgTe, and (c) bulk MgTe. Energy separations used in the band

offset measurement are labeled.

° data

Intensity (a.u.)

1 i L 4 A i L i L \ 1 i n
97.0 950 930 910 890 870 85.0 83.0
Binding Energy (eV)
Figure 3.3: Least squares fit of Zn 3p 3/2 and Mg 2s core level peaks for a structure
consisting of 20 A MgSe grown on Cdo54Znp.469e. Four parameters (the intensity
and binding energy of each peak) were used in the fit. Background subtraction

was performed prior to the fit.

the XPS spectra, so that simple linear extrapolation is not sufficient to accurately
position the valence band edge. To overcome this difficulty, we use, following the
procedure developed by Kraut et al. [14]. Basically, this involves calculating the
band structure for each material and then integrating over the Brillouin zone to
determine the theoretical valence band density of states (VBDOS). This VBDOS
is then convolved with the instrumental resolution function as determined from
experimentally acquired Au spectra. Finally, the instrumentally broadened the-
oretical VBDOS is fit to the experimental VBDOS to obtain the position of the
valence band edge.

For the heterojunctions studied in this chapter, the calculation of the band
structure was made more difficult by the fact that relatively little is know about
the ternary Cdo54Zng 4g5e and CdoggZng.12Te compounds, and even less is known
about MgSe and MgTe. Thus, the empirical pseudopotential method cannot be
used to directly calculate the band structure for these compounds. The methods
used to determine the band structure for Cdo.54Zno 4¢Se, Cdo.ggZno.12Te, MgSe and
MgTe are given below.

For Cdo.54Zno.469e (Cdo.ggZno.12Te) the band structure was calculated for a num-
ber of different semiconductors, but the best fit for the VBDOS was obtained using
the band structure of ZnSe (ZnTe) calculated by the pseudopotential method [15],
while allowing one additional energy scaling parameter to account for differences
in the band structure of ZnSe (ZnTe) versus Cdo.54Zno.46Se (Cdo.ggZno,12Te). Spin-
orbit interactions [16, 17] and an electron effective mass parameter to incorporate
the nonlocality of the pseudopotential [18] were included in the calculations. The
critical point energies needed for the pseudopotential calculations, and parameters
needed to determine the spin-orbit interactions, were obtained from Ref. [17] and
[19, 20] respectively.

For MgSe and MgTe the empirical pseudopotential method could not be used
since there have been almost no studies of these binary compounds because of

their reactivity, so very little is known of their band structure. Instead of the

pseudopotential method, we calculated the MgSe band structure using the semi-
empirical linear combination of atomic orbitals method [21], including spin orbit
interactions [22], and allowing for the additional energy scaling parameter men-
tioned above. This resulted in theoretical VBDOSs in reasonably good agreement
with the experimentally measured VBDOS.

The final step in determining the position of the valence band maximum is
to determine the instrumental resolution function and fit the VBDOS, broadened
with this function, to the experimentally measured VBDOS. Using measured XPS
Au 4f core level peaks (HWHM=0.37eV), the instrumental resolution function can
be determined, since the intrinsic linewidth of the Au 4f core level peak is well
known. For this experiment, the instrumental resolution function of the system
was determined to be a Voigt function with a Lorentzian fraction of 0.06, and a
HWHM of 0.28eV.

Figure 3.4 shows the fitted VBDOS for MgTe, CdoggZno12Te, MgSe and
Cdo54Zno.4g5e. The use of an energy scaling parameter requires verification of
the validity of the calculated band structures. To do this we followed the proce-
dure in Ref. [14] and varied the valence band edge fit region as shown in Figures 3.5
and 3.6. In these figures, the core level to valence band edge energy separation
is plotted as a function of the upper limit of the interval over which the VBDOS
is fitted. The reasoning here is that a theoretical VBDOS that accurately models
the experimental VBDOS will result in the same fitted valence band edge position
regardless of the fitting interval. The small variation of the fitted valence band
edge position as a function of fit region indicates that the theoretical VBDOS used

is in reasonable agreement with the experimental VBDOS.

100

Intensity (a.u.)

Cdo gly jo 1 @

Cd 542Mp ggS@

L 1 n . i]

15 10 O5 OO -O5
Binding Energy (eV)

Figure 3.4: Fit of experimental VBDOS (circles), and instrumentally broadened
theoretical VBDOS (solid lines) for (from top to bottom) MgTe, Cdo54Zno.46Se,
MegSe, and Cdo.8gZno.12Te.

101

(a) f
87.95

87.90 F _ +7 Tt
87.85

87.70 F Lo
87.65 F

87.60 § ! !
0.5 1.0 15 2.0

b)
88.25 [ TT TTT
88.20 |
88.15 f eT a
88.10
88.05 |
88.00 F

87.95 ¢ - ! !
0.5 1.0 15 2.0

E ax (eV)

Ezn3p/2 - Ey (eV)

ToT
—_

Eniges"Ey (eV)

Figure 3.5: Core level to VBM separation as function of the maximum binding
energy of the fitting interval for (a) Cdo ggZno.42Teand (b) MgTe. Low variance in

the energy separations indicates accurate modeling of the experimental VBDOS.

102

s765¢ | [J tpr7
87.60 } T)vtrttrrrrtett
87.55 F
s750f 99
87.45 |
87.40 | jlte >

87.35 | = ! —
0.5 1.0 15 2.0

Ez 3p /2- Eve (eV)

(b)
87.95

87.90
87.85 5 Ti dd7
87.80
87.75 £

i Li+ |
87.70 § Lio :

87.65 | — !
05 1.0 15 2.0

EF (eV)

Enig2s"Evem (eV)

max (

Figure 3.6: Core level to VBM separation as function of the maximum binding
energy of the fitting interval for (a) Cdo.54Zno.4gSeand (b) MgSe. Low variance in

the energy separations indicates accurate modeling of the experimental VBDOS.

103

E, Ey

0.56 +/- 0.07eV 0.43 +/- 0.1leV

Figure 3.7: MgSe/Cdo54Zno.4gSe and MgTe/Cdo.sgZno,42Te valence band offsets

measured using XPS. Bandgaps and offsets are not drawn to scale.

3.4 Results and Discussion

3.4.1 MgSe/Cdo54Zno4g5e Valence Band Offset

The average results from the XPS measurements of the MgSe/Cdo.54Zno,46Se va-
lence band offset are (Ln 3p 3/2 — Eucaznse)) = 87.52 + 0.04 eV, (Evy 2s —
Ev(ugse)) = 87.76 + 0.04 eV, and (Ezn 3p 3/2 — Eug 2s) = 0.80 + 0.04 eV. From
eq. (1) we obtain an average MgSe/Cdo54Zno.4gSe valence band offset of 0.56 +
0.07 eV with the valence band edge of MgSe below that of Cdo.54Zno.4gSe as shown
schematically in Figure 3.7. This result has the same sign for the valence band
offset as predicted by recent calculations [10]. This also deviates from the common
anion rule in the same direction and with the same order of magnitude as the
AlAs/GaAs system. Since both Mg and Al are third row elements in the periodic
table and have unoccupied d orbitals, this supports the importance of including
the repelling effects of cation d orbitals on the valence band edge in band offset
predictions for common anion systems.

To apply the results of this band offset measurement to the design of current
II-VI light emitters requires that we estimate the Zn1~-xMg,SySe;_y/ZnS1_,Se,
and Mg,Cd;_,Se/CdSe band offsets based on our new results. Using the

104

ZnSe/ZnS band offset obtained from Ref. [23], assuming a linear dependence of
the band offset on composition, and neglecting strain effects, we estimate the
Zno.79Mgo.2150.35€0.7/ZNSp,965€0.94 valence band offset to be roughly 0.29 eV versus
0.19 eV obtained using the common anion rule. The corresponding conduction
band offsets are 0.04 eV and 0.14 eV respectively. These quantities are approxi-
mate; however, the qualitative trend indicates that the Mg and S concentrations
in the Zn;_,Mg,SySe;_, layers may need to be increased in the LD design for
adequate electron confinement, especially if the wavelength of light emission is ex-
tended further into the blue. Another important impact of this measurement is
that we now know that both Mg and S pull down the valence band edge when added
to ZnSe. Since p-type dopability tends to decrease for semiconductors with lower
valence band edges, this would suggest that p-type doping of Zny-xMg,SySe1_, will
be increasingly difficult as more Mg and S are added to increase the bandgap. This
effect has already been observed in the form of compensation in nitrogen doped
Zn1-xMg,SySe1_y [24, 25], although the mechanism responsible for the compen-
sation was attributed to impurities in the first of these studies. Our results also
indicate that the Mg concentration in the Mg,Cd,_,Se layer for the graded electron
injector device should be increased for efficient carrier injection; however, the ex-
act composition will depend on the MgSe bandgap. Using the results presented in
Chapter 4, the desired Mg,Cd,_,Se composition at the Se/Te interface is roughly
70% Mg and 30% Cd.

3.4.2 MegTe/CdossZno.12Te Valence Band Offset

The average results from the XPS measurements of the MgTe/Cdo 9gZng49Te va-
lence band offset are (Ezn 3p 3/2 — Eycaznte)) = 87.80 + 0.04 eV, (Eg 2s —
E.(ugte)) = 88.16 + 0.05 eV, and (Ezn 3p 3/2 — Emg 28) = 0.79 + 0.09 eV. From
eq. (1) we obtain an average MgTe/Cdo.9gZno.19Te valence band offset of 0.43 +
0.11 eV with the valence band edge of MgTe below that of Cdy ggZng.1Te as shown

105

schematically in Figure 3.7. Again, this value deviates from the common anion rule
in the same direction as in the AlAs/GaAs system, which supports the hypothesis
of Wei and Zunger [13].

Since the addition of Mg to ZnTe pulls down the valence band edge, this result
is consistent with experimental observations of enhanced compensation in phos-
phorus doped Mg,Zn;_yTe [26]. The impact of this on the graded electron injector
devices is minimal as long as the deviation from the common anion rule is ac-
counted for in the design. As was shown in Chapter 2, a small valence band offset
between Mg,Zn;_yTe and ZnTe will not affect device performance as long as the
confining Mg,Zn,_,Te layers are graded to allow holes to flow over the MgyZn,_yTe

barriers.

3.5 Summary

We have used XPS to measure the valence band offset of the lattice matched
MgSe/Cdo.54Zno.465e and MgTe/Cdo ggZng,.12Te heterojunctions. The measured va-
lence band offsets were 0.56 + 0.07 eV and 0.43 + 0.11 eV respectively, with the
valence band edge of MgSe below that of Cdo.54Zno,4gSe, and the valence band edge
of MgTe below that of Cdo.9gZno12Te. Both of these values deviate significantly
from the common anion rule and support the importance of cation d orbitals
in valence band offset predictions for common anion heterojunctions. Also, the
measured MgSe/Cdo54Zno.4g5e valence band offset agrees favorably with trends
predicted in recent calculations [10].

Our measured valence band offset indicates that the Mg and S concentrations in
Zni-xMg,SySe;_y cladding layers may need to be higher than previously thought
for adequate electron carrier confinement, particularly as the emission wavelength
is shortened. The p-type dopability of these layers is also expected to decrease
due to the pulling down of the valence band edge with the addition of Mg and
S. This may be a limiting factor on the wavelength of light emission for the LD

106

structures incorporating Zn,_,Mg,S,Se,_, layers due to high series resistance. For
graded electron injector devices, the Mg concentration in the Mg,Cd,_,Se grading
layers may need to be increased for efficient electron injection into the Te-based
layers, depending on the bandgap of MgSe assumed. Using the value from Chap-
ter 4, ~ 70% Mg is required at the Se/Te interface in these structures. Finally,
MgyZn;_yTe confining layers will need to be graded to facilitate hole transport
over the Mg,Zn,_yTe layers; however, this should not significantly affect device

performance.

107

Bibliography

[1] M.A. Haase, J. Qiu, J.M. DePuydt, and H. Cheng, Appl. Phys. Lett. 59, 1272
(1991).

[2] H. Okuyama, K. Nakano, T. Miyajima, and K. Akimoto, Japan J. Appl. Phys.
30, L1620 (1991).

[3] H. Okuyama, T. Miyajima, Y. Morinaga, F. Hiei, M. Ozawa, and K. Akimoto,
Electron. Letters 28, 1798 (1992).

[4] J.M. Gaines, R.R. Drenten, K.W. Haberern, T. Marshall, P. Mensz, and J.
Petruzzello, Appl. Phys. Lett. 62, 2462 (1993).

[5] Sony results in plenary talk by D. Olego at the 1994 Device Research Confer-

ence in Boulder, CO.

[6] M.C. Phillips, M.W. Wang, J.F. Swenberg, J.O. McCaldin, and T.C. McGill,
Appl. Phys. Lett. 61, 1962 (1992).

[7] J.F. Swenberg, M.C. Phillips, M.W. Wang, J.O. McCaldin, and T.C. McGill,
J. Cryst. Growth 138, 692 (1994).

[8] A. Waag, H. Heinke, S. Scholl, C.R. Becker, and G. Landwehr, J. Cryst.
Growth 131, 607 (1993).

[9] M.W. Wang, M.C. Phillips, J.F. Swenberg, E.T. Yu, J.O. McCaldin, and T.C.
McGill, J. Appl. Phys. 73, 4660 (1993).

108

[10] T. Nakayama, Jpn. J. Appl. Phys. 33, L211 (1994).

[11] Y. Morinaga, H. Okuyama, and K. Akimoto, Japan J. Appl. Phys. 32, 678
(1993).

[12] J.O. McCaldin, T.C. McGill, and C.A. Mead, Phys. Rev. Lett. 36, 56 (1976).
[13] S. Wei and A. Zunger, Phys. Rev. Lett. 59, 144 (1987).

[14] E.A. Kraut, R.W. Grant, J.R. Waldrop, and $.P. Kowalczyk, Phys. Rev. B
28, 1965 (1983).

[15] M.L. Cohen and T.K. Bergstresser, Phys. Rev. 141, 789 (1966).

[16] L.R. Saravia and D. Brust, Phys. Rev. 176, 915 (1968).

[17] J.R. Chelikowsky and M.L. Cohen, Phys. Rev. B 14, 556 (1976).

[18] J. Chelikowsky, D.J. Chadi, and M.L. Cohen, Phys. Rev. B 8, 2786 (1973).

{19] F. Herman and S. Skillman, Atomic Structure Calculations (Prentice-Hall,
Englewood Cliffs, NJ, 1963).

[20] F. Herman, C.D. Kuglin, K.F. Cuff, and R.L. Kortum, Phys. Rev. Lett. 11,
541 (1963).

[21] W.A. Harrison, J. Vac. Sci. Technol. 14, 1016 (1977).
[22] M.D. Jaffe and J. Singh, Solid State Commun. 62, 399 (1987).

[23] K. Shahzad, D.J. Olego, and C.G. Van de Walle, Phys. Rev. B 38, 1417
(1988).

[24] P.M. Mensz, S. Herko, K.W. Haberern, J. Gaines, and C. Ponzoni, Appl.
Phys. Lett. 63, 2800 (1993).

109

[25] H. Okuyama, Y. Kishita, T. Miyajima, A. Ishibashi, and K. Akimoto, Appl.
Phys. Lett. 64, 904 (1994).

[26] K. Somogyi, J. Chevallier, J.F. Rommeluere, J. Marine, and B. Schaub, IEEE
Trans. Electron Devices 26, 1198 (1979).

110

Chapter 4

Status of the Graded Electron

Injector Project

4.1 Introduction and Outline

In this chapter, brief descriptions of materials characterization results and recent
graded electron injector device performance characteristics are presented. These
results are presented because of their importance to understanding both the current
status of the graded electron injector project as well as some of the important issues
that still need to be addressed.

XRD is used to study the long range structural quality of the devices, while
TEM probes the structural quality on a more local scale. SIMS, a chemical analysis
technique, is utilized to investigate chemical impurities in the epilayers. Current-
voltage and electroluminescence (EL) characteristics of the graded electron injector
LED structures are presented. The external quantum efficiency of recent devices
and possible approaches to improving light extraction and internal quantum efhi-
ciencies are given. Finally, the reliability of the LED structures is discussed. It
should be emphasized here that the TEM, SIMS and EL studies were all performed

by other researchers [1].

lil

4.2 Material Characterization

4.2.1 XRD

XRD is a standard tool for characterizing the long range structural quality of semi-
conductor layers. The XRD data shown in Figure 4.1 was acquired with a Philips’
Materials Research Diffractometer equipped with a four-crystal Ge-monochrometer
in the (220) reflection setting. In this setting, the Cu Ka, (A = 1.54056 A) X-ray
line has a divergence of ~ 12 arc seconds.

Figure 4.1(a) shows a typical XRD (2/26 scan for the graded electron injector
structure consisting of a GaSb substrate, a ~1 ym ZnTe buffer layer, a 300-600
A graded Mg,Cd,_,Se layer and a ~ 300 A CdSe layer at the sample surface.
The XRD FWHM for the GaSb and the ZnTe layers are 25 and 39 arc seconds
respectively. The CdSe layer has an expectedly very broad peak (~6 arc minutes)
due to the thinness of the layer, and the peak corresponding to the thin graded
Mg,Cd,_,5e layer is too broad and weak to be discernible above the noise. The
0/26 plot shown in Figure 4.1(a) is essentially what one would expect for a graded
injector device with good structural characteristics.

In contrast, Figure 4.1(b) shows a XRD 22/20 scan of a very similar structure
grown under nominally the same growth conditions. It can be seen that the FWHM
of all of the layers are increased significantly, with a corresponding loss of signal
intensity. In fact, the CdSe peak is no longer discernible. The structural quality
of this device is very poor as characterized by XRD.

Interestingly, the device with the poor structural quality as measured by KRD
had the highest external quantum efficiency measured to date, while the struc-
tural with the excellent structural quality had relatively poor external quantum
efficiency. In fact, initial analysis of the XRD data compiled on the graded injector
devices shows no correlation between structural quality LED performance. These

XRD results imply that other factors unrelated to the long range structural quality

112

GaSb
10 z Znte |

bene cE

2 F CdSe
OD

arnt

0! I ima | t

30.0 30.2 30.4 30.6

30.8

a)
Ty

ZnTe GaSb

3 | } CdSe

Intensity (cps)

Le)

10’ tt
30.0 30.2 30.4

Omega/2 Theta

30.6

30.8

Figure 4.1: XRD data for graded electron injector device. Representative data for a

structure with (a) desirable XRD characteristics and (b) poor XRD characteristics.

No correlation between XRD linewidths and LED performance was observed.

113

of the films are presently limiting the performance of the LEDs.

Graded injector LEDs with good long range structural quality as measured by
XRD have been reproduced, although not consistently. More importantly, LEDs
that simultaneously exhibit both excellent XRD characteristics and device prop-

erties have not been achieved.

4.2.2 TEM

TEM is used to study the structural quality of the graded electron injector struc-
tures on a more local scale. The TEM image shown in Figure 4.2 was acquired with
a Philips’ EM430 microscope using 300 keV electrons. Ion milling of the samples
was performed at low incident angles using 3-4 keV Ar ions. The low ion energy
was used in order to minimize damage to the fragile II-VI materials.

Figure 4.2 shows a representative TEM image of the top ~ 1500 A of a graded
electron injector LED structure. The growth is in the [001] direction and the [110]
direction is perpendicular to plane of the page. The ZnTe/Mg,Cd,_,Se interface
is readily apparent, while the exact location of the CdSe/Mg,Cd,_,Se interface
is not clear, since the Mg concentration is graded down to less than 10% at that
interface.

The two main features relevant in Figure 4.2 are the high density of defects
and the total thickness of the Se layers. Two defects are indicated by the arrows
labeled A and B. The defects originate from the ZnTe /Mg,Cd,_,Se interface and
often extend all the way through to the sample surface (arrow B). The intersection
of the defects with the (110) milled surface forms an angle of roughly 55 degrees,
indicating that the defects are likely to be stacking faults in the [111] direction
originating from point defects at the ZnTe/Mg,Cd,_,Se interface. Although it is
difficult to properly determine defect densities from cross-sectional TEM images,
a very rough estimate using 4 defects in a 1500 A by ~500 A (thickness of milled

region) gives a defect density of greater than 5 x 10!° cm-?. This is well above the

114

001
© [001]
[110]

Growth
Direction

CdSe

Znle

50 nm

Figure 4.2: TEM image of a graded electron injector device. Growth direction is
indicated by an arrow in the [001] direction. The [110] direction is perpendicular
to plane of the page. Two of the defects are indicated by arrows A and B.

115

desired defect density of < 10* — 105 cm~”.

In addition to the high defect density, the total thickness of the CdSe and
Mg,Cd,_,Se layers is significantly larger than expected (1000 A instead of 600
A): this is most likely due to a faster Mg,Cd,_,Se growth rate compared to
the well-calibrated growth rates for CdSe and ZnTe. The additional 300-400 A
of thickness contributes significantly to the defect density, since Mg,Cd,_,Se has
a lattice mismatch to ZnTe of 0.4% to 3.5% depending on the Mg concentration.
Ideally, the thickness of the graded region should remain below ~300 A to keep the
structure below the critical thickness. Thus, reducing the thickness of the lattice-
mismatched Mg,Cd,_,Se region could significantly reduce the defect density in the
LED structures. Currently, the grading in the Mg,Cd,_,Se region is implemented
by shutting off power to the Mg cell shortly before growth of the Se layers, so the
thickness of the graded region is determined by the cooling time of the Mg source
cell. To reduce the thickness of the graded region requires implementation of a
valved Mg source, a shuttering scheme for the Mg source, or a small Mg cell with

fast thermal time constant.

4.2.3 SIMS

Secondary ion mass spectrometry (SIMS) is often used for studying impurities
because of its excellent chemical sensitivity (parts per million to parts per billion)
and reasonably good depth resolution (> few nm). The underlying principle of
SIMS is to chemically profile a sample by performing mass spectrometry on ions
that have been sputtered from the sample surface. More details on the SIMS
technique are given in Chapter 6.

The SIMS data shown in Figure 4.3 was taken by researchers at Charles Evans
and Associates using 14.5 keV Cs* ion bombardment for the C, O and Sb profiles,
and 8.0 keV Of ion bombardment for the As profile. It should be noted that Al,

Ga, In and N were also profiled with roughly the same results, but were not shown

116

a) II-294
{ 0”

“ee

20 ee
S10 ‘.

S 140"

0.0 ) 500.0 1000.0 1500.0 2000.0
Position (angstroms)

Il-328

210

9 10°

wz 17

0.0 500.0 1000.0 1500.0 2000.0
Position (angstroms)

Figure 4.3: Representative SIMS impurity profiles for an LED structure with (a)
high impurity concentrations, and (b) low impurity concentrations. No correlation

between impurity concentration and LED performance was observed.

117

for clarity purposes. The sputtering was performed over a 350 wm x 350 um area,
while the detected secondary ions were extracted from a 85 sum diameter area in
the center of the sputtered crater.

The first thing to notice in Figure 4.3 is that the impurity concentration is
significantly reduced in the sample shown in Figure 4.3(b) compared to sample
shown in Figure 4.3(a). Both samples are graded injector structures as described
in Section 4.2.1. The sample surface corresponds to a position of 0 A on the hor-
izontal axis. Similar to what occurred for the XRD data, no correlation between
impurity concentration and electroluminescence characteristics was observed. In
fact, despite the high impurity concentration in the sample shown in Figure 4.3(a),
it had much better electroluminescence properties than the sample shown in Fig-
ure 4.3(b), which luminesced primarily in the red. Thus, at present it is believed
that some mechanism other than impurity concentration is the limiting factor in
determining the electroluminescence properties of the LEDs.

A final observation regarding Figure 4.3 is that the impurity profiles for both
plots peak near the graded region (~ 1000 A). This could be due either to a
matrix effect in the graded region or to the gettering properties of Mg. So far,
no correlation has been observed between the impurity concentration within the
graded region and LED performance; however, this could be because other factors

are still limiting device performance.

4.3 Device Performance

4.3.1 Current-Voltage Characteristics

Figure 4.4 shows the current-voltage characteristics for a typical graded electron
injector LED using both linear and logarithmic scales for the current density. The
current-voltage characteristics for the devices are excellent, with almost no reverse

bias leakage current and a low forward bias turn-on voltage, indicative of good

118

40 F

20 F

Current Density (A/cm’)

10° F L 1 ! F ! ‘ ! au

Voltage (V)

Figure 4.4: Experimental (solid) and calculated (dashed) current-voltage charac-
teristics for a graded injector device, plotted on (a) linear and (b) logarithmic
scales. Deviation of the data from the ideal, calculated values is due to leakiness

in the LED structure.

119

ohmic contacts and a reasonably low series resistance. This is expected, given that
the underlying principle of the graded electron injector concept is to avoid doping
and contacting problems by using a heterojunction design.

The current-voltage characteristics shown in Figure 4.4 are comparable to the
best obtained in the ZnSe-based material system [2], and the only deviation from
an ideal current-voltage curve is that the current onset occurs at a slightly lower
voltage than in an ideal structure, as indicated by the dotted lines. This is most
likely due to leakage current caused by non-optimal grading of the Mg,Cd,_,Se
layer and could probably be remedied by better control of the graded region, as

will be discussed in more detail elsewhere [3].

4.3.2 Electroluminescence

Figure 4.5 shows a representative electroluminescence (EL) spectra from a graded
electron injector structure biased at 2.8 V. The dominant peak at ~ 2.23 eV and
the absence of a peak at ~ 1.74 eV indicate that electrons have been successfully
injected into the ZnTe layer while holes have been blocked from entering the CdSe
layer. This is further evidence for the successful implementation of the graded
electron injector scheme. The feature at ~ 1.95 eV is present to some extent in
all of the graded injector devices fabricated thus far. The intensity of this peak
relative to the ZnTe peak varies from sample to sample even when grown under
nominally identical conditions. Depending on the relative intensities of the 1.95
eV and 2.23 eV peaks, the EL can appear yellow or even red, when the intensity
of the 1.95 eV peak is large enough. It should be noted that the low-energy peak
is also observed in photoluminescence (PL) studies. The source of this low energy

EL/PL peak is currently not known and is still under investigation.

120

Bias = 2.8 volts

Intensity (a.u.)

1.5 1.6 17 #148 749 20 21 22 23 2.4
Energy (eV)

Figure 4.5: Electroluminescence data from a graded electron injector LED struc-

ture.

4.3.3 External Quantum Efficiency

The external quantum efficiency of the graded electron injector LEDs is still not
very high, with the best efficiency to date being 0.007%. There are a number of
reasons for this rather low value, the first of which is related to extraction of the
light radiated from the ZnTe active region. It has been shown that the external
quantum efficiency of an LED can be drastically affected by the efficiency of the
light extraction [4, 5]. Figure 4.6 shows a schematic view of the graded injector
device and some possible contacting schemes to the CdSe. Since the CdSe cannot
be grown very thick, due to strain and absorption considerations, a contacting
scheme must be devised which spreads current over the entire surface of the mesa
while still allowing radiated light to escape.

The first contacting scheme used was a thick Zn cap. During the etch step of
the processing, the Zn cap etched at a much faster rate than the II-VI materials,
resulting in a narrow region around the edge of the device where light could escape.
This procedure was not ideal, as only a small fraction of the radiated light was

extracted from the device. Another approach used a thick Au/Ge pad to contact

121

Need transparent contact

—_—

CdSe
MgCdse

Znie

L Au
Pa

Zn Au/Ge | TO
CdSe CdSe CdSe
MgCdSe MgCadSe MgCdSe
zZnle zZnle z2nle
Zn cap Thin Au with indium tin oxide
thick Au/Ge pad (ITO)

Figure 4.6: Schematic diagrams of the graded electron injector device (top) and
some possible contacting schemes (bottom). The Zn cap (left) allows light ex-
traction from only a thin region at the edge of the mesa. The thin Au and thick
Au/Ge contacting scheme (center) improves on the Zn cap, and the ITO contacting

scheme (right) should result in even better light extraction.

122

the CdSe layer and a thin Au layer (a few 100 A) to uniformly spread the current
over the surface of the mesa. This approach is better than the Zn cap in that
less of the mesa surface is light-absorbing; however, it is still not ideal since the
Au/Ge is completely opaque and the thin Au layer is partially absorbing and not
very robust. This is the method used in present devices. A better approach being
investigated would be to use indium tin oxide (ITO), which is both transparent
and conductive. This scheme would significantly improve external quantum effi-
ciencies by allowing light to escape over the entire top surface of the mesa without
significant absorption.

Another light extraction issue is the substrate. The GaSb substrates used in
present devices completely absorb any light emitted in the downward direction. If
ZnTe substrates of sufficient quality were available, the external quantum efficiency
would be further improved since light emitted into the substrate would not be
absorbed.

Another component of the external quantum efficiency is the actual efficiency
of radiative recombination at the desired wavelength. Mechanisms such as non-
radiative recombination via leakage current through the graded region could con-
siderably lower the overall external quantum efficiency. Thus, it is important to
optimize the Mg,Cd,_,Se graded region so as to reduce leakage current to a min-
imum. The thickness of the ZnTe active region also plays an important role in
the radiative recombination efficiency. If this layer is thinner than the electron
diffusion length, a significant fraction of the carriers will reach the GaSb buffer
layer before recombining, lowering the external quantum efficiency. One way to
alleviate this problem is to use QWs or confining layers (see Chapter 2) to trap
the carriers and increase the radiative recombination efficiency. This is currently
being investigated.

Thus, the external quantum efficiency of the graded electron injector LEDs is
currently not very high. However, a number of approaches are available to increase

both the extraction and the radiative recombination efficiencies, thereby increasing

123

the overall external quantum efficiency.

4.3.4 Device Degradation Studies

The most important issue in determining the feasibility of LEDs and LDs based on
II-VI semiconductors is that of device reliability. II-VI semiconductors materials
are well known for their softness, especially when compared to the very rugged
nitrides. Preliminary studies of LEDs based on the graded electron injector design
have shown evidence for long device lifetimes, but in general the device lifetimes
have been poor.

Figure 4.7 shows plots of normalized photodiode current versus time for two
separate devices. Figure 4.7(a) shows the lifetime characteristics for an early de-
vice grown on a ZnTe substrate with undoped ZnTe epilayers. The current-voltage
characteristics and efficiencies for this device were very poor; however, the device
reliability was relatively good, with less than 50% degradation after ~ 1500 hours.
After subsequent implementation of p-type doping of ZnTe, higher purity Mg and
better quality substrates (GaSb), the device lifetimes are as shown in Figure 4.7(b).
Despite much improved external quantum efficiencies and current-voltage charac-
teristics, the device lifetimes are much poorer than in earlier devices. Notice the
change in scale for the time-axes in the two plots.

The cause for the reduction in device lifetimes is still uncertain. One possibility
is that it is related to the N-doping of the ZnTe. Another possibility is that growing
on GaSb instead of ZnTe substrates results in a higher defect density which will

reduce device lifetimes. These and other explanations are still under investigation.

4.4 Summary

In conclusion, successful implementation of the graded electron injector scheme has

been demonstrated. The current-voltage characteristics show diode-like behavior

124

Current Density = 100 Alem?

hk.
T T T

oO
oo

la (Normalized)
o 9°
LL OD
I J T

Oo
ho

Oo
ro)

0 — ~ 500 1000 1500

kt
ro)
—-—

Current Density = 52 A/cm?

lia (Normalized)
>) oO
NR ®
y | u i

oO
Po
™ FT

! — A le

0 50 100 150
Time (hours)

oO
ro)

Figure 4.7: Device lifetime measurements for two devices: (a) an early device grown
on a ZnTe substrate without p-type doping of ZnTe and (b) a more recent device
grown on a GaSb substrate with N-doping of the p-ZnTe. Note the difference in

scale for the time axes.

125

with low turn-on voltages and low series resistances. Also, green luminescence from
radiative recombination within the ZnTe layer shows that the graded Mg,Cd,_,Se
layer is functioning properly.

However, a number of issues still need to be addressed in order to achieve
high-performance LEDs based on the graded electron injector design. These in-
clude structural quality, spectral purity, external quantum efficiency and device
lifetimes. Poor structural quality of the devices was observed in the form of high
defect densities in the TEM data and large FWHMs in some of the XRD data.
SIMS analysis shows high impurity concentrations in some of the devices. Although
no correlation was observed between LED performance and defect densities, XRD
FWHM and impurity concentrations, reduction of these quantities should eventu-
ally improve overall device performance.

Spectral purity problems were indicated by the presence of a variable intensity,
low energy luminescence peak observed in the EL spectra. This peak must be
eliminated in order to consistently achieve high-performance LEDs. The external
quantum efficiency is still not optimal, and a number of improvements on the
present LED design should allow for much higher efficiencies. The issue of device
lifetimes is another issue critical to the long term feasibility of the graded electron
injector device. Present device lifetimes are poor, and it remains to be determined
whether this is a consequence of the N-doping of ZnTe, due to lack of control of the
graded region, or a problem intrinsic to the relatively soft II-VI semiconductors.

In conclusion, the graded electron injector LED design shows promising results,
but a number of important issues must be addressed before the long term feasibility

of this approach can be demonstrated.

126

Bibliography

[1] TEM studies performed by Carol Garland, SIMS studies performed by re-
searchers at Charles Evans and Associates and EL studies performed by Jo-

hanes Swenberg.

[2] D.B. Eason, Z. Yu, C. Boney, J. Ren, L.E. Churchill, J.W. Cook, Jr., J.F.
Schetzina, and N.A. El-Masry, to be published in J. Cryst. Growth (1994).

[3] J.F. Swenberg, Ph.D. thesis dissertation (1994).

[4] K.H. Huang, J.G. Yu, C.P. Kuo, R.M. Fletcher, T.D. Osentowski, L.J. Stinson,
M.G. Craford, and A.S.H. Liao, Appl. Phys. Lett. 61, 1045 (1992).

[5] F.A. Kish, F.M. Steranka, D.C. DeFevere, D.A. Vanderwater, K.G. Park,
C.P. Kuo, T.D, Osentowski, M.J. Peanasky, J.G. Yu, R.M. Fletcher, D.A.
Steigerwald, M.G. Craford, and V.M. Robbins, Appl. Phys. Lett. 64, 2839
(1994).

127

Part II

XPS Studies of the Mixed Anion
InAs/GaSb Heterointerface

128

Chapter 5

Surface Exchange Reaction

Studies

5.1 Introduction and Outline

The mixed anion arsenide/antimonide system has a number of technologically in-
teresting applications, including InAs/AISb oscillators operating at frequencies
greater than 700 GHz [1], novel nAs/AISb/GaSb based [2, 3, 4] and InAs/GaSb
based [5] tunnel structures and InAs/Ga,_,In,Sb infrared (IR) superlattice (SL)
detectors [6, 7]. One of the main problems in growing such devices is control-
ling the structural and chemical properties of the interface. However, this is very
difficult because of the different bond strengths, surface free energies and bond
lengths at the interfaces in these systems. As well, IIJ-V structures are typically
grown with group V overpressures, since their vapor pressures are much larger
than those of the group III elements. For growth of mixed anion structures,
this overflux of the group V element becomes a problem and can lead to cross-
incorporation of the anion species and variation in interface composition [8]. In IR
SL structures cross-incorporation of the anion species can result in shorter carrier

lifetimes and poor quality material, while interfacial composition affects the type

129

and level of background doping as well as the SL bandgap, due to changes in the
strain configurations at the interface [9]. Furthermore, the interface composition
in the InAs/AISb system affects the carrier mobility, carrier concentration and
the InAs/AISb valence band offset [10, 11]. Thus, it is important to be able to
grow arsenide/antimonide structures with no anion cross-incorporation and with
abrupt, composition controllable interfaces. To do this, the detailed mechanisms
of the interface formation must be understood.

In this chapter, we use X-ray photoelectron spectroscopy (XPS) and reflection
high energy electron diffraction (RHEED) to study interface formation during
molecular beam epitaxy (MBE) growth of GaSb/InAs structures. The GaSb/InAs
system is more amenable to XPS study than In, Ga;_,Sb/InAs due to the presence
of unique elements on both sides of the heterojunction. Furthermore, since the
problems in controlling arsenide/antimonide interfaces are associated with their
mixed anion nature, the results of our study should be extendible to any of the
arsenide/antimonide structures described above.

The XPS experiments study the core level peak intensities and binding energy
separations as a function of cracker power and soak time. The soaks, or interrupts
as they are sometimes called, consist of exposing an InAs or GaSb surface to an Sb
or As flux, respectively, and are intended to mimic the soaks used in GaSb/InAs
heterostructure growths. The XPS core level peak intensities and binding energy
separations are used to determine the extent of the exchange reactions. Results
from both As soaks of GaSb surfaces and Sb soaks of InAs surfaces are presented.

RHEED specular streak intensity analysis was also used to monitor the ex-
change reactions during growth. The advantage of RHEED as a method to char-
acterize these reactions is that it is a real-time, in situ measurement. However, it
provides only near surface structural information, hence it is necessary to corre-
late the RHEED data with a chemical analysis technique such as XPS in order to
accurately characterize exchange reactions.

In Section 5.2, we describe the MBE growth and the details of the XPS and

130

RHEED experiments. Section 5.3 contains brief descriptions of the data analysis
procedures used. Section 5.4 presents the results from the XPS and RHEED
studies. In Section 5.5 we discuss the results and their significance to interface

abruptness studies, and Section 5.5 concludes the chapter.

5.2 Experiment

5.2.1 Sample Growth

All of the structures studied here were grown in a Perkin-Elmer Model 430 MBE
system equipped with cracked Sb and As sources. Two cracker powers, 80% and
40%, were used to emulate growth systems with and without cracked sources re-
spectively. We refer to the source fluxes by their primary component; Sb, and
Asq (uncracked) at 40% cracker power, and Sb and Asp (cracked) at 80% cracker
power. The structures were grown on a number of different (100) substrates (GaAs,
n-GaSb, p-GaSb), with no variation in results. Samples were grown on thick stress
relaxed GaSb buffer layers at a substrate temperature of ~ 380°C. Multiple sam-
ples grown on the same substrate were separated by buffer layers thick enough to
prevent detection of underlying layers with XPS. More details of the growth may
be found in Ref. [12].

5.2.2 XPS and RHEED Measurements

The XPS measurements were obtained using a Perkin-Elmer Model 5100 analysis
system with a monochromatic Al Ka source (hu = 1486.6 eV). All of the samples
studied were transferred from the growth chamber to the XPS chamber via an
ultrahigh vacuum transfer tube. The base pressure in the XPS chamber was typi-
cally ~1 to 5 x 107! Torr. Care was taken to ensure that the escape orientation
of the photoelectrons remained constant from sample to sample to minimize any

electron diffraction effects due to the single crystalline nature of the samples. More

131

details of the XPS experimental setup are given in Section 1.3.4.

The RHEED studies presented here were performed by another researcher.
Some details of the RHEED experimental setup are given in Section 1.3.4. More
details of the RHEED experiment can be found in Ref. [13]. Briefly, the RHEED
pattern on the phosphor screen is acquired using a CCD camera and digitized
into a 640 x 480 array of single-byte data. From this, the streak separation and
specular streak intensity can be extracted. In this study, only the specular streak

intensity was used to characterize the exchange reactions.

5.3 Data Analysis

Figure 5.1 shows two sample XPS spectra. Figure 5.1 (a) shows a spectrum from
an InAs epilayer after exposure to a 15 second Sbz soak, and Figure 5.1 (b) shows
a spectrum from a GaSb epilayer after exposure to a 240 second Asg soak. The
data analysis for the Sb soaks of InAs surfaces consisted of isolating the Sb 4d, In
4d and As 3d core level peaks, performing an integrated background subtraction
on each peak, fitting these peaks to Voigt functions and finally determining the
corresponding integrated intensities and core level binding energies. From the peak
intensities, we were able to estimate the Sb coverage on InAs. A similar procedure
was used for analysis of the As soaks of GaSb surfaces, except that the As 3d,
Ga 3d and Sb 4d peaks were used. To estimate the group V coverage, we used
a simple attenuation model based on published effective photoelectron mean free
paths (MFPs) [14, 15, 16]. Due to the uncertainty in these MFPs, this procedure
cannot be used to determine precise compositions; however, comparisons between
samples and qualitative estimates of surface coverage are valid.

Some care must be taken in the data analysis described above. As seen in
Figure 5.1(a), some of the XPS features overlap. For example, the In 4d plasmon
loss tail interferes with the Sb 4d peak. To isolate the Sb 4d peak, we first fit the
In 4d peak while constraining the peak shape to that obtained from a bulk InAs

132

(a)

_ Sb, In 4d

5 NX

‘O

40 30 20
Binding Energy (eV)

(b)
— As,

3 Sb 4d
& N\

3 GaSb Ga 3d
c As 3d

40 30 20

Binding Energy (eV)

Figure 5.1: XPS binding energy spectra of (a) 15 sec. Sb soak of an InAs surface
and (b) 240 sec. Asg soak of a GaSb surface.

133

standard. Then, using the same InAs standard, we perform spectrum stripping to
remove the In 4d loss tail from the Sb 4d signal. Figure 5.2 shows a representative
Sb 4d peak after background subtraction (dots), and the resulting fitted Voigt
functions (solid). The dotted lines denoting the individual Sb 4d components are
discussed below. The Sb 4d loss tail was stripped from the As 3d peak following
the same procedure, using GaSb as the standard from which to obtain the shape
of the plasmon loss tail. An analogous procedure was used to isolate the Ga 3d,
Sb 4d and As 3d peaks for the spectra obtained from As soaks of GaSb surfaces.

In addition to interferences from overlapping features, chemically shifted com-
ponents must also be properly accounted for in the analysis. Chemically shifted
XPS peaks are important because they are an indication of multiple bonding con-
figurations, and because they can skew measured peak intensities if not properly
analyzed. Figure 5.2 shows the Sb 4d signal from a sample consisting of a 15
second Sbz soak of an InAs surface. The two spin split doublets comprising the
Sb 4d peak were obtained by stripping off the In 4d loss tail, subtracting the in-
tegrated background function and fitting the resulting spectra to two Sb 4d peaks
with their peak shapes constrained to that of Sb 4d from a GaSb standard. Only a
vertical scale and a binding energy (BE) parameter were allowed to vary for each
of these doublets for a total of four parameters. Using this technique, core level
binding energy separations and peak intensities on identical samples were typically
reproducible to better than + 0.03 eV and +5%.

The RHEED data analysis was performed by another researcher, and a de-
tailed explanation can be found in Ref. [13]. Briefly, the RHEED specular streak
intensity was monitored in real time as a function of soak time and cracker power.
Figure 1.11 in Chapter 1 shows a schematic diagram of the experimental setup used
to extract the RHEED specular streak intensity. Upon initiation of an Sb soak,
the specular streak intensity decreases and then recovers with a characteristic time
dependent on the Sb source cracker power. By studying the time-dependency of

the RHEED intensity and comparing it with the XPS results, we hope to infer a

134

Intensity (a.u.)

36 35 34 33 32 31 30 29
Binding Energy (eV)
Figure 5.2: Sb 4d peak from 15 sec. Sb soak of an InAs surface. Data is fitted

with two chemically shifted peaks with lineshapes constrained to that of Sb 4d
from bulk GaSb.

135

correlation between the RHEED data and the chemical exchange reaction at the

surface of the samples.

5.4 Results

5.4.1 Sb Soaks of InAs Surfaces

The experiments on Sb soaks of InAs surfaces were done in two parts. In the first
experiment, the Sb cracker power and soak time were chosen to duplicate the con-
ditions during growth of actual devices. For the 5 second Sb soaks of InAs surfaces,
the As 3d to In 4d peak intensity ratio relative to that of bulk InAs decreased by
18% and 1% for Sbz and Sb, soak species respectively. We also observed a much
larger Sb component for the Sbz soak as shown in Figure 5.3. Using the simple
attenuation model mentioned above, this corresponds to a majority of the termi-
nating As layer being exchanged during the Sb2 soak, and very minimal exchange
during the Sby soak. Due to the uncertainty in published MFPs, we can only
estimate relative changes in coverage, and cannot quote exact values.

In the second Sb soak experiment, Sb coverage on InAs as a function of both
cracker power and soak time was studied. Since the 5 second Sb soak at growth
conditions resulted in roughly complete exchange of the As terminated surface, the
bulk evaporator temperature was lowered to reduce the total Sb flux impinging on
the sample surface by roughly 80%. This increases the exchange time, allowing
better temporal resolution of the exchange process. Figure 5.4 shows the ratio of
the Sb 4d to In 4d peak intensities as a function of soak time for (a) 40% cracker
power, and (b) 80% cracker power. As expected, the Sb coverage increases with
time of soak and, at low soak times, is significantly greater for cracked soaks than
for uncracked soaks. The dotted and dashed lines in Figure 5.4 and the chemically
shifted Sb 4d component shown in Figure 5.2 are discussed below.

It should be noted that some concern has been raised over why the exchange

136

morn Sb, on InAs Asga "4d
As3p ---- Sb, on InAs ; i
i InAs | Sb4d

Intensity (a.u.)

seseesseaneessonesees As, on GaSb
~-----. As, on GaSb .° *e
> —— GaSb :
rary
150 | 100 50

Binding Energy (eV)

Figure 5.3: XPS spectra for Sb soaks of InAs surfaces (top) and As soaks of GaSb
surfaces (bottom). Spectra from bulk samples (solid), uncracked sources (dashed)

and cracked sources (dotted) are shown.

(a)

0.4

0.3

Sb 4d to In 4d Peak Intensity Ratio

(b)

0.4

0.0

Sb 4d to In 4d Peak Intensity Ratio

137

| @—O Total Sb signal bos Sb,
G----G Low BE component
& --AHigh BE component Poo
ae _°
er ia]
[oe B.... ae? se
~pecor- 9 ne Oe ee A
200 400 600
Sb, Soak Time (sec)
O----& Low BE component 25250
-A Hi eM =
4& - -A High BE component ecco
0 200 400 600

Sb, Soak Time (sec)

Figure 5.4: Sb 4d to In 4d peak intensity ratio as a function of soak time for (a)

cracked, and (b) uncracked Sb sources. Components of the total peak are also

shown (dotted and dashed lines). Lines are drawn to guide the eye. In plot (b)

lines extend to additional data points at 1800 s.

138

1.5 T T ™ T 1.5 | T T — T

Pa | spb Cracked Sb flux | Sb, Uncracked Sb flux

77) 2 on InAs surface | } InAs surf

Cc _——-— exposure on InAs surface

2 | xposure |

= 1.0}!

a 1:9 f Q----@----- oF

Ld l

< 9 |

) Ii

> il:

505+ ||! 4 0544

3 |

fo) | '

5 | I: ] +1 ® @-OSb Surface ¢
' - , - urface Coverage

B |! OO RREED ‘Spat ‘tera I — RHEED Spot intensity
: L

0.0 o—_—_——_—_ 0.0 -@—-——.
0 10 20 0 20 40 60
Sb, exposure Time (sec) Sb, exposure Time (sec)

Figure 5.5: RHEED (solid) and XPS measured Sb coverage (dashed with open cir-
cles) versus exposure, or soak, time for Sb (left) and Sb, (right). Commencement

of exposure is indicated by vertical dashed line.

reaction should proceed at all, given that the InAs bond strength is greater than
that of InSb. The explanation is that even though the displacement of As with
Sb to form InSb from InAs is unlikely, due to the differing bond strengths, the
constant flux of Sb coupled with the absence of a source of As allows the exchange
reaction to proceed.

Figure 5.5 shows a comparison of RHEED and XPS data for Sb soaks of InAs
surfaces. The vertical axis in each plot shows both the normalized RHEED specular
streak intensity (solid lines) and the normalized Sb coverage as determined by
XPS (dashed lines with open circles). The horizontal axis shows the exposure, or
equivalently, soak time for a cracked (left plot) or an uncracked (right plot) Sb

flux incident on the InAs surface. The vertical dashed line in each plot represents

139

the start of the exposure. In these plots we see the characteristic drop off and
subsequent recovery in the RHEED specular streak intensity, but more importantly
we see that the recover time of the RHEED profile is similar to the rise time of

the XPS Sb coverage profile. The significance of this is discussed in Section 5.5.

5.4.2 As Soaks of GaSb Surfaces

The studies of As soaks of GaSb surfaces consisted of two parts. In the first
experiment, the soak time was held constant at 5 seconds, while the cracker power
was varied. The bottom plot of Figure 5.3 shows the results of this experiment.
Examining the As 3d peak, we immediately observe a qualitative increase in the
extent of the exchange reaction when using Asz as compared to As4. In addition,
for the 5 second As2 soak of a GaSb surface, the Sb 4d to Ga 3d peak intensity ratio
decreased by 26% relative to bulk GaSb. The decrease in the same peak intensity
ratio for an As4 soak is 7%. This corresponds to roughly complete exchange of
the terminating Sb layer for the Asp soak and partial exchange for the As soak.
Again, exact values for the extent of the exchange reaction cannot be determined
due to uncertainties in the photoelectron MFPs.

In the second soak experiment, As coverage on GaSb as a function of As, soak
time was studied. Figure 5.6 shows the ratio of the As 3d to Ga 3d peak intensities
as a function of soak time. Similar to the Sb soaks of InAs surfaces, the occurrence
of an exchange reaction is readily observed. However, for the As soak of a GaSb
surface we observe no saturation in the As 3d for longer soak times. This indicates
that the exchange reaction of As for Sb proceeds past the terminating Sb layer,
down into the underlying GaSb. More justification for this conclusion is given in

the next section.

140

0.50

om 0.40 b 4

£ 0.30 + 4

xo ]

im)

0.10 £ J
(ep)

0.00 ,; ;
0.0 100-0 200.0

As, Soak Time (sec)

Figure 5.6: Ratio of As 3d to Ga 3d core level peak intensity ratios as a function
of Asg soak time on a GaSb surface. Notice that the As 3d peak intensity does not

saturate for longer soak times as in the case for Sb soaks of InAs surface.

141

5.5 Discussion

5.5.1 Sb Soaks of InAs Surfaces

Since there is so much uncertainty in the published electron MFPs [14, 15, 16], we
are not able to accurately calculate surface coverages for any given group V soak
experiment. However, we can estimate these surface coverages by studying the
time dependent soak experiments. To do this for the Sb soaks of InAs surfaces,
we need to determine the source of the chemically shifted Sb 4d component shown
in Figure 5.2. A detailed analysis of this peak shows a variation in the magnitude
of the chemical shift of more than 0.1 eV. This is a large variation considering
that the chemical shift is only ~ 0.5 eV; however, as can be seen in Figure 5.7,
analysis of the higher BE Sb 3ds/2 core level reveals a nearly identical variation in
the separation between its constituent peaks. This demonstrates that the variation
in peak separation is a real effect and not an artifact of the XPS data acquisition.
A representative, fitted Sb 3d5/2 peak is shown in the inset in Figure 5.7, and a
representative, fitted Sb 4d5/2 peak is show in Figure 5.2. In both cases, the peak
shapes were constrained to those obtained from bulk GaSb.

One possible explanation for the second Sb component might be a surface
reconstruction related shift; however, the lack of a shifted component at very low
soak times argues against this theory. Another possible explanation for the two
Sb components is the formation of metallic Sb islands. If true, then based on the
direction of the observed chemical shift, the higher BE Sb peak should be due to
the islands [17]. To check this, we measured the BE of metallic Sb 4d, from a
thin metallic Sb layer grown on InAs. The measured Sb 4d BE was less than the
BE of the shifted Sb 4d peak, which is consistent with the shifted peak being due
to metallic Sb islands, since islanding can shift peaks to higher binding energy.
Island formation also provides an explanation of the variation in the position of

the chemically shifted peak, since changes in average island size can cause a change

142

0.8

fo)
fon)

Ol
O]

OSb 3d,, |
O Sb 4d

oO
iN

ie)

Binding Energy Difference, E,-E, (eV)
Intensity (a.u.)

531-520 527. ~«525
Binding Energy (eV)

0.0 l l ! _L !
0 1 2 3 4 5 6 7 8

Sample Number

Figure 5.7: Energy separation between the two component peaks for the Sb 3d
and Sb 4d core levels. The same fluctuation in separation is observed for both
peaks. Inset shows components comprising Sb 3d peak. Sb 4d peak is shown in

Figure 5.2.

143

in the magnitude of the chemical shift.

In addition, the energy separation between the lower BE Sb 4d peak and the
In 4d peak is stable, making it reasonable to believe that this peak is due to an
exchanged layer of Sb bonded to In which should be stable from sample to sample,
while the higher BE peak, due to islanding, is more apt to vary from sample to
sample. Also, referring to Figure 5.4, we see that since the low BE Sb peak in both
the cracked and uncracked soaks stabilize at roughly the same value, it is quite
possible that the steady state value corresponds to exactly one monolayer of Sb
atoms exchanging with the As terminated InAs surface. If this is the case, then the
coverage obtained with the Sb2 device-like soaks is slightly less than one monolayer,
as indicated by the arrow in Figure 5.4 (b). It should be noted that even if the
islanding explanation proposed above is incorrect, the fact that the soaks currently
used in heterojunction growth occur in a region where the coverage is dependent
on the time of soak indicates that further optimization of growth parameters may
be possible.

The conclusion that the Sb exchange with As is self-terminating at one mono-
layer is significant in that it predicts that a GaSb-on-InAs interface should be able
to be grown with abrupt InSb interfaces. This can be done by terminating an
InAs layer with As, soaking the sample with Sb until the exchange has proceeded
to completion, leaving InSb bonds, and finally, proceeding with GaSb growth. The
excess metallic Sb produced by the soak will be consumed during the GaSb growth
and will not interfere with the interface abruptness. Chapter 6 examines in more
detail the issue of interface abruptness for GaSb/InAs interfaces.

Another important result of the Sb on InAs soak studies is the correlation be-
tween the RHEED and XPS results shown in Figure 5.5. The agreement between
the recovery time in the RHEED data and the rise time in the XPS data is signifi-
cant, especially considering that the two analysis techniques measure very different
quantities: the RHEED profile is a measure of the structural properties of the sam-

ple surface during the Sb soak of a single InAs surface, while the XPS profile is a

144

direct chemical measurement of the exchange reaction for a number of InAs sam-
ples, each exposed to an Sb soak of differing durations. The similarity between the
RHEED and XPS profiles, coupled with the observation that the minima in the
RHEED profiles correspond to roughly 50% exchange as determined from XPS,
indicate that the drop in the RHEED streak intensity may be due to differences in
the Sb and As form factors for electron scattering. This would explain the recovery
in the RHEED streak intensity as the exchange proceeds towards full exchange of
the terminating As layer with 5b. Another possible explanation is that partial
exchange of the As layer with Sb causes surface roughening which will also reduce
the specular streak intensity.

Regardless of the exact mechanism for the behavior of the RHEED specular
streak intensity, the close correlation between the RHEED and XPS recovery and
rise times indicates that RHEED can be used to monitor surface exchange reac-
tions. This is very important, since RHEED is a real-time analysis technique which

can be used during the growth of actual devices.

5.5.2 As Soaks of GaSb Surfaces

It was shown in Section 5.4.2 that for As soaks of GaSb surfaces, the As signal
as measured by XPS is not self terminating (see Figure 5.6). Without further
investigation, it cannot be concluded whether the continued increase in the As
signal is due to accumulation of As at the sample surface, or continued exchange
of As for Sb down into the underlying GaSb.

Figure 5.8 shows Sb 4d to Ga 3d and Ga 3p to Ga 3d peak intensity ratios
plotted as a function of As soak time. The peak intensity ratios are normalized
to peak intensity ratios for bulk GaSb. We see a significant drop in the Sb 4d to
Ga 3d peak intensity ratio for longer soak times, which strongly indicates that the
exchange of As for Sb is proceeding past the top monolayer. However, since the

binding energy of the Sb 4d core level is larger than that for Ga 3d, part of the

145

1.20 | | | - | A:
© 1.00 + —O- on
=>
® 0.80 - 7
@ 0.60; <—~< Sb 4d to Ga 3d Intensity Ratio 7
ou O-——© Ga 3p to Ga 3d Intensity Ratio

0.40 . ! . !

0.0 100.0 200.0

As, Soak Time (sec)

Figure 5.8: Sb 4d to Ga 3d and Ga 3p to Ga 3d peak intensity ratios for Asz soaks
of GaSb surfaces. Peak intensity ratios are normalized to those obtained from bulk

GaSb.

decrease in the Sb to Ga peak intensity ratio could be attributed to the stronger
attenuation of the Sb 4d peak if there existed an overlayer of As on the sample
surface. Recall that a larger binding energy corresponds to a smaller photoelectron
kinetic energy, which translates into a shorter MFP.

If this were true then we should observe an even stronger decrease in the Ga
3p to Ga 3d peak intensity ratio, because the binding energy of the Ga 3p peak
is larger than that of the Sb 4d peak. Since the Ga 3p to Ga 3d peak intensity
ratio is independent of the Aso soak time, we can safely conclude that both the
continued increase in the As 3d peak intensity and the continued decrease in the
Sb 4d to Ga 3d peak intensity ratio for longer soak times is due to exchange of As
for Sb past the terminating monolayer of Sb, and into the underlying GaSb.

The exchange of Sb with As was somewhat expected, given that the GaAs bond

146

Sb on InAs As on Gasb

Soak Time Soak Time

Figure 5.9: Schematic diagrams of the surface exchange reactions for Sb soaks
of InAs surfaces (left), and As soaks of GaSb surfaces (right). Note the self-
terminating nature of the exchange shown on the left as compared to the exchange

shown on the right.

strength is greater than that of GaSb. However, the exchange into the underlying
GaSb was unexpected and is very significant to studies of interface abruptness.
What the XPS results show is that an InAs-on-GaSb interface grown using an
As soak at the interface will result in an extended GaAs-like interface unless the
As soak time is very precisely monitored. Even with precise monitoring it may
be that the interface will not be abrupt, with As exchanging deep into the GaSb
in some regions, and less than complete exchange in other regions. The issue of
interface abruptness for InAs/GaSb heterojunctions is discussed in more detail in

Chapter 6.

5.6 Summary

147

In conclusion, we have used XPS and RHEED to study in situ, the dynamics of
the formation of the mixed anion InAs/GaSb interface as a function of cracker
power and soak time. For the studies of exchange as a function of cracker power,
it was found that anion exchange occurs more readily when using cracked versus
uncracked sources during the soak. This was true for both Sb soaks of InAs
surfaces, and As soaks of GaSb surfaces.

Varying the soak time for Sb soaks of InAs surfaces showed a saturation of the
Sb coverage for longer soak times. Based on detailed analysis of the XPS core level
peaks, we see evidence for island formation and conclude that the steady state
coverage may correspond to monolayer exchange of the As terminated surface
with Sb and the formation of metallic Sb islands. This is shown schematically
in Figure 5.9. Since soaks used in actual device growths presently lie in a region
where variations in soak time may vary the extent of the group V exchange, further
optimization of interface characteristics may be obtained by modifying these soak
times. The saturation of the exchange reaction after one monolayer for Sb soaks
of InAs surfaces suggests that the growth of GaSb-on-InAs interfaces can be made
abrupt with InSb interfaces.

Parallel studies, during the surface exchange reaction, of the RHEED specular
streak intensities correlate well with the XPS data. The recovery times for the
RHEED intensities profiles were similar to the soak times for saturation of the XPS
Sb signals. It was concluded that the partial exchange of the terminating monolayer
of As causes the RHEED intensity to decrease through differing form factors for
As and Sb, or through surface roughening, and that full exchange restores the
specular streak intensity. This is significant in that the RHEED pattern can be
used to monitor the surface exchange reaction in real-time, during the growth of
device structures, which cannot be accomplished using XPS alone.

For As soaks of GaSb surfaces it was found that the measured XPS As 3d to
Ga 3d peak intensity ratios did not saturate for longer soak times. This was in

contrast to the results obtained for the Sb soaks of InAs surfaces. By studying the

148

Sb 4d to Ga 3d and Ga 3p to Ga 3d peak intensity ratios, it was determined that
As was exchanging with not only the terminating Sb layer at the surface of the
GaSb, but was also exchanging with Sb in the underlying GaSb. This is shown
schematically in Figure 5.9. This result is very significant, because it shows that
an InAs-on-GaSb heterojunction grown using an As interrupt will result in an
extended GaAs-like interface, and that very precise shuttering would be required

to obtain an abrupt GaAs interface for the InAs-on-GaSb growth order.

149

Bibliography

[1] E.R. Brown, J.R. Sdderstrém, C.D. Parker, L.J. Mahoney, K.M. Molvar, and
T.C. McGill, Appl. Phys. Lett. 58, 2291 (1991).

[2] M. Sweeny and J. Xu, Appl. Phys. Lett. 54, 546 (1989).

[3] J.R. Sdéderstrém, D.H. Chow, and T.C. McGill, Appl. Phys. Lett. 55, 1094
(1989).

[4] L.F. Luo, R. Beresford, and W.I. Wang, Appl. Phys. Lett. 55, 2023 (1989).

[5] D.A. Collins, E.T. Yu, Y. Rajakarunanayake, J.R. Sdderstrém, D.Z.-Y. Ting,
D.H. Chow, and T.C. McGill, Appl. Phys. Lett. 57, 683 (1990).

[6] D.L. Smith and C. Mailhiot, J. Appl. Phys. 62, 2545 (1987).

[7] R.H. Miles, D-H. Chow, J.N. Schulman, and T.C. McGill, Appl. Phys. Lett.
57, 801 (1990).

[8] T.D. Golding, H.D. Shih, J.T. Zborowski, W.C. Fan, C.C. Horton, P.C. Chow,
A. Vigliante, B.C. Covington, A. Chi, J.M. Anthony, and H.F. Schaake, J.
Vac. Sci. Technol. B 10, 880 (1992).

[9] D.H. Chow, R.H. Miles, and A.T. Hunter, J. Vac. Sci. Technol. B 10, 888
(1992).

[10] G. Tuttle, H. Kroemer, and J.H. English, J. Appl. Phys. 67, 3032 (1990).

150

[11] J.R. Waldrop, G.J. Sullivan, R.W. Grant, E.A. Kraut, and W.A. Harrison, J.
Vac. Sci. Technol. B 10, 1773 (1992).

[12] D.H. Chow, R.H. Miles, C.W. Nieh, and T.C. McGill, J. Cryst. Growth 111
683 (1991).

[13] D.A. Collins, M.W. Wang, R.W. Grant, and T.C. McGill, J. Appl. Phys. 75
259 (1994).

[14] M.P. Seah and W.A. Dench, Surf. Interface Anal. 1, 2 (1979).
[15] C.M. Kwei and L.W. Chen, Surf. Interface Anal. 11, 60 (1988).
[16] S. Tanuma, C.J. Powell, and D.R. Penn, Surf. Interface Anal. 17, 927 (1991).

[17] M.K. Wagner, J.C. Hansen, R. deSouza-Machado, S. Liang, J.G. Tobin, M.G.
Mason, S. Brandt, Y.T. Tan, A.-B. Yang, and F.C. Brown, Phys. Rev. B 43,
6405 (1991).

151

Chapter 6

InAs/GaSb Heterojunction
Studies

6.1 Introduction and Outline

This chapter presents two studies of the InAs/GaSb heterointerface. The first set
of experiments examine the InAs/GaSb band alignment using XPS. The valence
band offset is measured as a function of both the nominal interface composition and
the growth order. In the second part of the chapter, we present a detailed study
of the abruptness of the InAs/GaSb interface, again as a function of both nominal
interface composition and growth order. A number of characterization techniques
(XPS, RHEED, cross-sectional STM and SIMS) were used in this study. It should
be noted that the RHEED, cross-sectional STM and SIMS studies were performed
by other researchers, but are included here for completeness [1].

Section 6.2 presents the XPS measurements of the InAs/GaSb valence band
offset. The interface abruptness studies are presented in Section 6.3, and the

results from both studies are summarized in Section 6.4.

152

6.2. InAs/GaSb Band Alignment Studies

6.2.1 Motivation

The InAs/GaSb valence band offset is an important parameter for a number
of device applications, and is relevant to fundamental studies of band align-
ments in mixed anion/cation material systems. The device applications in the
InAs/GaSb/AISb material system include: InAs/AISb oscillators operating at fre-
quencies greater than 700GHz [2], novel InAs/AlSb/GaSb based [3, 4, 5] and
InAs/GaSb [6] tunnel structures, and InAs/Ga,_,In,Sb infrared (IR) superlattice
(SL) detectors [7, 8]. For many of these applications, the band alignment plays a
critical role in the device performance; however, determination of band alignments
in mixed anion/cation systems is complicated by the fact that two different types
of interfaces can be formed. In the case of an InAs/GaSb heterojunction, both
InSb-like and GaAs-like interfaces can be formed. It is very important to deter-
mine what effect changing the interface composition has on band alignments in the
InAs/GaSb material system as well as in other mixed anion/cation systems.

The InAs/GaSb valence band offset has been experimentally measured by a
variety of techniques [9, 10, 11, 12]; however, none of these studies addressed the
issue of interface composition. A number of theoretical studies have predicted that
the InAs/GaSb valence band offset should depend on the interface composition
[13, 14, 15]. These theories predict that the InAs/GaSb valence band offset should
be larger for an InSb-like interface compared to a GaAs-like interface; however, the
magnitude of this effect ranges considerably, from 31 to 150 meV, depending on
which theory is used. Additionally, Gualtieri et al. [12] observed a 50 meV larger
valence band offset for InAs grown on GaSb compared to GaSb grown on InAs;
however, this was deemed to be within the experimental uncertainty.

The effect of interface composition and growth order on the valence band offset

has also been studied in other material systems. For example, the InP/GalInAs

153

valence band offset shows a 180 meV asymmetry with growth order due to a
change in the interface composition, whereas the GalnP/GaAs valence band offset
shows no such asymmetry [16]. The InP/GalnAs result is in contradiction with
Hybertsen’s calculations [17], which predict no dependency of the valence band
offset on interface composition or growth order. The InAs /AlSb interface has also
been studied for both InSb-like and AlAs-like interface compositions [18, 19]. In
those studies, a slight increase in the valence band offset was observed for the AlAs-
like interface compared to the InSb-like interface, but no change in the valence band
offset was noted for changes in the growth order.

Thus, there is significant motivation for studying the InAs /GaSb band align-
ment, both for device applications and as a fundamental study of band align-
ments to be applied to other material systems. We report measurements on the
InAs/GaSb valence band offset as a function of both interface composition and

growth order using X-ray photoelectron spectroscopy (XPS).

6.2.2 Experiment

All of the structures used in this study were grown in a Perkin-Elmer 430 molecular
beam epitaxy (MBE) system. A number of different (100) substrates (GaAs, n-
GaSb, p-GaSb) were used with no variation in results. Samples were grown on thick
stress relaxed GaSb buffer layers at a substrate temperature of ~ 380°C. Multiple
samples grown on the same substrate were separated by buffer layers thick enough
to prevent detection of underlying layers with XPS. Hetero junction samples were
grown with thin (~ 20 A) overlayers, such that an XPS signal is detected from
both sides of the interface. X-ray diffraction of superlattice structures was used
to verify that the growth schemes used produced the correct nominal interface
compositions. Growth details may be found in Ref. [20].

The XPS measurements were obtained using a Perkin-Elmer Model 5500 anal-

ysis system with a monochromatic Al Ka source (hv = 1486.6 eV). All of the

154

samples studied were transferred from the growth chamber to the XPS chamber
via an ultrahigh vacuum transfer tube. The base pressure in the XPS chamber was
typically ~3 x 10~!° Torr. Care was taken to ensure that the escape orientation
of the photoelectrons remained constant from sample to sample to minimize any
electron diffraction effects due to the single crystalline nature of the samples.
The standard approach to determine valence band offsets using XPS is to mea-
sure core level to valence band maximum separations in bulk samples and core
level to core level separations in heterojunction structures. For example, referring
to Fig. 6.1, BGiA’ — Eins, BGe8>_ GaSb and Bgasb_ Ein As would be measured in
bulk InAs, bulk GaSb, and InAs/GaSb heterojunction samples respectively, where
the Ec, and E, denote core level and valence band maximum binding energies for

the indicated material. The valence band offset is determined using
AE, = Bins _ EGasb — (BGe5>_ EGa5b) _ (Barts — Binds) _ (EGes?— dn As) (6. 1)

Since we are only interested in changes in the valence band offset, and since core
level to valence band maximum separations are constant, variations in the valence

band offset are given by
6(A E,) = _ 6(BaAs — Ege) (6.2)

Thus, to determine changes in the valence band offset as a function of interface
composition, we need only to measure changes in core level separations in hetero-
junctions with varying interface compositions. For the InAs/GaSb interface, four
different core level separations may be used to monitor changes in the valence band
offset: Ga 3d- As 3d, Ga 3d - In 4d, Sb 4d- As 3d, and Sb 4d- In 4d.

To determine core level energy separations, we use the following procedure,
given in more detail in Ref. [21]. For each core level we remove any interfering
loss tails, implement an integrated background subtraction, and finally perform a
least square fit to Voigt functions. Since the In 4d and Ga 3d peaks overlap, we
constrain their shapes to be those obtained from InAs and GaSb bulk standards,

155

E Gasp
E InAs _ E GaSb Cc
V V
Gasb
Ey
GaSb GaSb
E cL E V
InAs InAs
E CL E V
| Vv E Gasb
E InAs y CL
CL i
CL CL

Figure 6.1: Schematic diagram showing XPS binding energy separations used in

valence band offset measurements.

156

in order to reduce the number of fitting parameters. The Ga 3d and In 4d loss tails
which interfere with the Sb 4d core level were stripped using InAs and GaAs bulk
standards respectively. Figure 6.2 (a) shows a representative unprocessed XPS
spectrum for a GaSb-on-InAs heterojunction. Figure 6.2 (b) shows the resulting
normalized fit of the As 3d, Sb 4d, Ga 3d and In 4d core level peaks. Core level

separations for a given sample are typically reproducible to better than 0.02 eV.

6.2.3 Results and Discussion

The resulting average core level energy separations for both InAs-on-GaSb and
GaSb-on-InAs growth orders, and for both InSb-like and GaAs-like interface com-
positions are shown in Table 6.1. | Multiple XPS spectra and structures were
analyzed for each of the growth orders and interface compositions as indicated by
x:y where x and y represent the number of samples and the total number of XPS
spectra acquired, respectively. From Equation 2, changes in the peak separations
reflect changes in the valence band offset, 6(AEy). For the GaSb-on-InAs growth
order, the average change in the valence band offset with change in interface com-
position is 0.003 eV. For the InAs-on-GaSb growth order, this value is 0.014 eV.
Both of these values are less than our experimental uncertainty (~ 0.02eV). Thus,
we observe no dependence of ABy on the interface composition for both growth
orders to within our experimental uncertainty.

Since the core level energy separations for a given growth order are independent
of interface composition, we can average these values for each growth order to
determine if the valence band offset changes in going from one growth order to
another. This is shown in Table 6.2. The first row in this table shows the average of
the core level separations for the GaSb-on-InAs growth direction for both interface
compositions. The second row in Table 6.2 shows the corresponding average core
level separations for the InAs-on-GaSb growth direction. The change in the valence

band offset in going from the GaSb-on-InAs to the InAs-on-GaSb growth direction

157

a) Sb 4d
2 - Data
2 Fit
: ¥ : Ga 3d
As 3d io: fp in4d
= |) asad Sb 4d Ga 3d
7 In 4d
40 30 20

Binding Energy (eV)

Figure 6.2: Figure (a) shows an example of an unprocessed XPS scan of a GaSb-

on-InAs heterojunction. Figure (b) shows the fitted As 3d, Sb 4d, Ga 3d, and

In 4dcore level peaks after removal of interfering peaks, background subtraction,

normalization, and least squares fitting to Voigt functions.

158

Core Level Separations (eV)

Sample Ein 447 Eqn 4a- Eas 347 Eas 34-
Egy 4d Ega 3d Egp 4a Ega 3d
GaSb-on-InAs
GaAs-like -14.698 -1.753 8.822 21.766
2:3 (0.003) (0.010) (0.003) (0.006)
InSb-like -14.685 -1.728 8.802 21.759
2:4 (0.003) (0.004) (0.013) (0.009)
5 (AEy) 0.013 0.025 -0.019 -0.008
InAs-on-GaSb
GaAs-like -14.572 ~1.680 8.917 21.810
3:6 (0.007) (0.019) (0.016) (0.016)
InSb-like ~14.570 ~1.646 8.912 21.836
5:6 (0.011) (0.037) (0.015) (0.026)
} (AEy) 0.002 0.034 ~0.005 0.026

Table 6.1: The measured core level separations for InSb-like and GaAs-like inter-

face compositions. The notation 2:4 represents two growths with a total of four

scans acquired. Values in parenthesis are the standard deviations for the respective

core level splittings. All spin-orbit split peak positions are measured with respect

to the J = 5/2 component.

159

Core Level Separations (eV)
Sample Em 4a- Ein 4d Bas 3d Eas 3d-

Esp 44 EGa 3d Esp 4a Ega 34

GaSb-on-InAs

-14.691 -1.741 8.812 21.763

InAs-on-GaSb
-14.571 -1.663 8.915 21.823
5 (AEy) 0.120 0.078 0.103 0.060

extended GaSb-on-InAs
-14.635 -1.695 8.869 21.809

6 (AEy) 0.056 0.046 0.057 0.046

Table 6.2: The measured core level separations for both InAs-on-GaSb and GaSb-
on-InAs growth orders. Also included are the core level separations for an inten-
tionally extended GaSb-on-InAs interface. All spin-orbit split peak positions are

measured with respect to the J = 5/2 component.

is shown in the third row of this table. The average change is 0.090 eV with a
standard deviation among the four values of 0.027 eV, where the InAs-on-GaSb
interface has the larger valence band offset. This value is significantly larger than
the experimental uncertainty in the core level energy separations.

The exact reason why the valence band offset depends on the growth order is
still under investigation; however, it is most likely related to the asymmetry in the

InAs/GaSb interface. STM studies and recent growth studies provide a possible

160

explanation for the difference in valence band offsets in terms of the interface
structure [22, 23]. The GaSb-on-InAs interface is found to be more abrupt than
the InAs-on-GaSb interface, with the latter exhibiting signs of Sb incorporation in
the InAs overlayer and an As/Sb exchange reaction in the underlying GaSb. More
details of the interface abruptness studies and the exchange reaction studies are
given in Section 6.3 and Chapter 5 respectively.

To verify this dependence of the valence band offset on interface structure, an
intentionally extended GaSb-on-InAs heterojunction was studied. The structure
was grown by opening the Sb shutter during the last three monolayers of the
InAs growth, producing an extended InSb-like interface. The measured core level
separations for this structure are shown in Table 6.2. From the core level binding
energy separations for this structure, we obtain a valence band offset that is 51 meV
greater than that for the nominally abrupt GaSb-on-InAs growth direction. This
value lies between that of the nominally abrupt GaSb-on-InAs and the extended
InAs-on-GaSb structures.

This is consistent with the hypothesis that the extended nature of the InAs-
on-GaSb interface is responsible for the difference in the InAs/GaSb valence band
offset for the two growth orders. Also, a detailed analysis of core level peak shapes
shows no evidence for chemical shifts or broadening. This result strongly indicates
that the change in valence band offset is not due to either band-bending effects
from surface fermi level pinning of the InAs surface or surface-induced chemical
shifts. Attempts to test these conclusions by altering the composition of the InAs-
on-GaSb interface produced no significant change in the InAs/GaSb valence band
offset, probably due to the inherent difficulty in growing abrupt InAs-on-GaSb
interfaces [23].

We still have not explained how an extended InAs/GaSb interface would pro-
duce a change in the valence band offset for this system. One possible explanation
is that an anti-structure defect [24], related to Sb cross-incorporation or As ex-

change, may produce a dipole moment, resulting in the observed change in the

161

valence band offset. Initial estimates of the magnitude of this effect are consistent
with the observed change in the valence band offset and realistic concentrations (~
10 % of a monolayer) of anti-structure defects. The sign of the measured change
in the valence band offset is consistent with an anti-structure in the InAs-on-GaSb
samples, with group III atoms on group V sites followed by group V atoms on
group III sites as viewed from the sample surface down. Further investigations are
currently underway to determine the validity of this explanation.

To summarize, we have used XPS to measure the variation in the nAs/GaSb
valence band offset as a function of both interface composition (InSb-like and
GaAs-like) and growth order (InAs-on-GaSb and GaSb-on-InAs). The InAs/GaSb
valence band offset was observed to be independent of interface composition irre-
spective of growth order; however, between the two different growth orders, a 90
meV difference in the InAs/GaSb valence band offset is observed, with the InAs-
on-GaSb interface showing the larger valence band offset. The difference in valence

band offset is attributed to the extended nature of the InAs-on-GaSb interface.

6.3 Abruptness of the InAs/GaSb Heteroin-

terface

6.3.1 Motivation

The technological and scientific importance of arsenide/antimonide heterointer-
faces has already been well motivated in Chapter 1 and in Sections 6.2.1 and 5.1.
Specific studies of the interface abruptness in the InAs/GaSb material system are
motivated primarily by InAs/In,Ga,_,Sb IR SL detector applications, and as a
fundamental study of mixed anion interface formation. Despite the advantages of
the InAs/In,Gai_,5b IR SL design, detectors competitive with Hg,Cd,_,Te tech-
nologies have yet to be fabricated [25]. This poor performance has been attributed

to the shorter extrinsic lifetime in InAs/In,Ga,_,Sb IR SL devices, due to the

162

presence of Shockley-Read-Hall (SRH) centers [26]. One possible mechanism for
this is that the generation of SRH centers is related to interdiffusion of the group
V sublattice at the mixed-anion As/Sb interface, which could result in antisite
defects, leading to SRH centers.

Since the length scales for interdiffusion are potentially quite short, it was nec-
essary to draw on a number of different characterization techniques in order to
properly assess the existence of interdiffusion at the InAs/GaSb interface. The
techniques used were XPS, RHEED, cross-sectional STM and SIMS, each of which
provided a unique piece of information about the InAs/GaSb interface: XPS pro-
vided in situ, near surface chemical analysis; RHEED provided real-time, in situ,
near surface structural characterization; cross-sectional STM provided very high
resolution imaging, electronic structure information, and interface roughness anal-
ysis for a SL structure; and SIMS provided chemical analysis with excellent sensi-

tivity.

6.3.2 Experiment

Details of the growth and transfer of the samples used in this experiment are given
in Sections 5.2.1 and 6.2.2. In addition, brief descriptions of the XPS experimental
setup have already been presented in Chapters 1, 3, 5 and earlier in this chapter.
In the remainder of this chapter, XPS is used primarily to study core level peak
intensity ratios as opposed to the core level energy separations. A number of InAs-
on-GaSb and GaSb-on-InAs samples were studied, each grown with both InSb-like
and GaAs-like interfaces. For each of these samples, the ratio of the As 3d to In 4d
peak intensities were measured and normalized to the same ratio obtained from an
As terminated (2 x 4 reconstruction) InAs standard. Similarly, the Ga 3d to Sb 4d
peak intensity ratios were normalized to an Sb terminated (1 x 3 reconstruction)
GaSb standard. From the variation in these peak intensity ratios, we are able

to infer relative interface abruptness and to study anion cross-incorporation. The

163

UHV chamber

P~ 10" Tor

in-situ Cleave
STM tip

InAs/GaSb
superlattice

Figure 6.3: Schematic Diagram of Cross-sectional STM setup used to study
InAs/GaSb SLs.

spectrum stripping and peak fitting procedures for the XPS samples are the same
as those presented in Section 6.2.2.

Brief descriptions of the RHEED experimental setup have already been pre-
sented in Chapters 1 and 5. In this chapter RHEED is used to study interface
asymmetry through analysis of the RHEED reconstruction patterns and specu-
lar streak intensity oscillations. Figure 1.11 shows a schematic diagram of the
experimental setup used to extract the RHEED data.

Figure 6.3 shows a schematic diagram of the cross-sectional STM setup used
in our studies. Briefly, the cross-sectional STM procedure consisted of cleaving a
sample in situ at a pressure of ~4 x 107! Torr. STM imaging and spectroscopy
can then be performed on the cleaved surface without exposure to atmospheric
contaminants. Some of the advantages of this characterization technique are high
resolution imaging, contamination free surface, and the ability to perform cross-
sectional imaging which is not possible with standard STM.

The STM images shown in this chapter were acquired with a constant current

of 0.1 nA. Carefully prepared and characterized single crystal tungsten probe tips

164

Energy
Filter

lon Beam

Na
Mass

Secondary “a Spectrometer
lons

Secondary
lon
Detection Detector
Region
Figure 6.4: Schematic Diagram of the SIMS setup used to profile InAs-on-GaSb

interfaces.

were used in the STM data acquisition. A detailed description of the STM experi-
mental technique can be found in Ref. [22]. It should be emphasized that the STM
studies were performed by another researcher, but are included in this chapter for
completeness and clarity [1].

Figure 6.4 shows a schematic diagram of a standard SIMS experimental setup.
Briefly, SIMS analysis consists of sputtering a sample over a region of a few 100 um
on aside. Secondary ions are then detected over a much smaller region (< 100 um
diameter) in the center of the sputtered crater. Before reaching the detector, the
secondary ions pass through first an energy filter and then a mass spectrometer.

Some of the advantages of SIMS are excellent sensitivity down to parts per
million or even parts per billion, profiling capability and good chemical speci-
ficity. Disadvantages include the destructive nature of the measurement and rela-

tively poor depth resolution (a few nm at best). The SIMS analysis presented in

165

this chapter were performed by researchers at Charles Evans and Associates on a
CAMECA IMS-4f machine using 3.0 keV OF bombardment. Cs ion bombardment

was also attempted, but resulted in poorer depth resolution.

6.3.3 Observation of Interfacial Asymmetry

In this section, we report on observations of interfacial asymmetry in InAs/GaSb
heterointerfaces using XPS, RHEED, and cross-sectional STM. The asymme-
try was observed in the form of an increased broadening, or roughness, of the
InAs/GaSb interface for the InAs-on-GaSb growth direction compared to the

GaSb-on-InAs growth direction.

XPS Results

Tables 6.3 (a) and (b) show summaries of XPS results which demonstrate this
asymmetry. In these tables we tabulate the Sb 4d to Ga 3d and As 3d to In 4d peak
intensity ratios after normalization to bulk GaSb and InAs standards. The peak
intensity ratios are tabulated for both (a) GaSb-on-InAs and (b) InAs-on-GaSb
growth directions, and for both InSb-like and GaAs-like interface compositions.
Each tabulated entry specifies the range of peak intensity ratios compiled from a
number of samples and XPS spectra.

The purpose of this method of tabulation is to infer information about relative
interface abruptness by simultaneously displaying the results from a number of
different samples. Due to uncertainties in photoelectron MFPs [27, 28, 29], it
is impossible to unambiguously evaluate interface abruptness on a single sample
using a simple attenuation model for the XPS signal. Essentially, the difficulty is
that a single XPS spectrum can be accurately modeled with a number of different
interface configurations depending on the photoelectron MFPs used in the fitting.
By displaying variations in peak intensity ratios over a number of samples, as

shown in Tables 6.3 (a) and (b), it is possible to infer relative interface abruptness.

166

Minimum and Maximum
Peak Intensity Ratios
(a) (Normalized)

Sb4d/Ga3d } As3d/ Indd

InSb-like 1,.26-1.31 0.63-0.74

GaAs-like 1.20-1.25 0.74-0.86

Range 1.20-1.31 | 0.63-0.86

More
Abrupt

Minimum and Maximum
(b) Peak Intensity Ratios
(Normalized)

$b4d/Ga3d | As3d/ In4d

InSb-like 1.06-1.24 0.81-0.96
GaAs-like 0.55-0.80 1.28-1.37
Extended

Table 6.3: XPS peak intensity ratios, for both growth orders and both interface
compositions. Each entry shows the range (minimum-maximum) of values for a
given interface composition and growth order. The total range for each growth

order is also given.

167

Specifically, if we examine the bottom row in each table, we see that the range
of values for a given peak intensity ratio (e.g., Sb 4d to Ga 3d) is much larger for the
InAs-on-GaSb growth direction compared to the GaSb-on-InAs growth direction.
In fact, for the InAs-on-GaSb growth direction, the range in the values is much
greater than the experimental uncertainty for a given peak intensity ratio, and is
inconsistent with the range of values expected for a simple change in the interface
composition [23]. That is, even allowing for variation in the photoelectron MFPs,
the data in Table 6.3 (b) cannot be modeled as an abrupt interface due to the large
range in the values for a given peak intensity ratio. This is true even though any
one of the XPS spectra used to compile the data in this table could be modeled
with an abrupt interface.

For the GaSb-on-InAs growth direction, the range in values is consistent with
an abrupt interface, but could also be modeled with a non-abrupt interface, so it
is not possible to conclude with certainty whether or not this interface is abrupt.
However, since the range in the peak intensity ratios is much smaller than for
the InAs-on-GaSb growth direction, it can be concluded that the GaSb-on-InAs
growth direction is the more abrupt of the two growth directions.

Individual peak intensity ratios shown in Table 6.2 are studied in more detail
in Section 6.3.4, in order to extract the mechanisms underlying the asymmetry in

the abruptness of the InAs/GaSb interface.

RHEED Results

Another characterization technique that was used to observe the asymmetry
in the InAs/GaSb interface was RHEED. Both the RHEED reconstruction pat-
terns and the RHEED intensity oscillations showed abnormal behavior for the
InAs-on-GaSb growth direction. Figure 6.5(a) shows a schematic diagram of the
progression of the RHEED reconstruction pattern during the growth of an InAs-
on-GaSb heterojunction. During the GaSb growth the RHEED pattern showed

a1 x 3 reconstruction pattern which switched to a 1 x 5 reconstruction pattern

168

(Q)
<«<——- |x]
_—— 1x3, 2x3
«———._ 1x5 (Sb interrupt)
—_—— 1x3
(b) 120 ~
“oO;
3 InAs grown on InAs )
Lo]
al
= 80 + .
a7)
Cc
= |
= al — 4
D120
7)
uJ
tid
a 80 | J
5 q
cS}
3 InAs grown
© on GaSb
a.
Y 40 eo ee areas Ar ee | n
0 25 50 75 100

Time (seconds)

Figure 6.5: RHEED studies of interface asymmetry. Unusual 1 x 3, 2 x 3 re-
construction is observed in the InAs reconstruction patterns shown schematically
in (a). RHEED intensity oscillation profiles also show unusual behavior for the

InAs-on-GaSb growth direction.

169

during the Sb interrupt at the interface. This is typical behavior for a Sb termi-
nated GaSb layer. At the onset of the InAs growth, however, unusual behavior
is observed in the reconstruction pattern. Instead of the usual 1 x 1 reconstruc-
tion pattern observed for InAs at the growth temperatures and fluxes used, we
observed a 1 x 3 or faint 2 x 3 reconstruction for the first few monolayers of InAs
growth. This type of unusual behavior in the RHEED reconstruction pattern was
not observed for the GaSb-on-InAs growth direction.

One possible cause for the 1 x 3, 2 x 3 reconstruction pattern is Sb incorpo-
ration into InAs from the underlying GaSb. This explanation is consistent with
RHEED studies of InAs,Sb,_, layers which also show a1 x 3 reconstruction pat-
tern, and also offers an mechanism for the transition from the 1 x 3 to the 1 x 1
reconstruction in the form of decreasing Sb concentration in the InAs layer after
the first few monolayers of growth. Another explanation for the 1 x 3, 2 x 3
reconstruction pattern is that GaAs forms at the interface causing the subsequent
InAs layer to have a 3-D growth mode. This is consistent with the reconstruc-
tion patterns observed for the growth of InAs-on-GaAs. The transition from the
1 x 3 to the 1 x 1 reconstruction is then explained by the coalescing of the InAs
islands. Because of the qualitative and structural nature of the information ob-
tained from RHEED reconstruction patterns, it is not possible to determine with
certainty which of the two explanations offered above is valid. In Section 6.3.4,
more quantitative chemical characterization of the interface formation is presented.

Figure 6.5(b) shows the RHEED specular spot (or streak) intensity oscillations
for (top) InAs grown on InAs and (bottom) InAs grown on GaSb. Multiple profiles
are plotted for both cases to demonstrate the reproducibility of the measurement.
In both plots, the top InAs growth is initiated at roughly t = 10 seconds, after a
growth interrupt. For the InAs grown on InAs we observe the standard RHEED
intensity oscillations indicative of smooth 2-D growth. For the InAs grown on
GaSb, however, we observe very unusual behavior during the first few monolayers

of growth. This type of behavior is not observed during the growth of GaSb-on-

170

InAs interfaces. Another study which also reported this type of behavior in the
RHEED intensity profile proposed Sb incorporation into the InAs as a possible
cause, but no quantitative justification was given [30]. Although the RHEED data
presented above cannot, by itself, be used to determine the exact mechanisms of
the InAs/GaSb interface formation, it does provide clear evidence for the asym-

metry between the InAs-on-GaSb and the GaSb-on-InAs growth directions.

Cross-sectional STM Results

Figure 6.6 shows two representative cross-sectional STM images of InAs/GaSb
SLs: one grown with (a) InSb-like interfaces and the other with (b) GaAs-like
interfaces. Both images were acquired using a sample voltage of -1.3 V with the
InAs layers shown in dark and the GaSb layers in bright. The growth directions
and scales are indicated on the images.

The main feature to be observed from these images is the asymmetry in interface
roughness for the two growth directions. For both interface types, the InAs-on-
GaSb growth direction is observed to have an increased interface roughness when
compared to the GaSb-on-InAs growth direction, although this is difficult to dis-
cern visually for the sample grown with InSb-like interfaces. Detailed analyses of
the interface roughness by taking the Fourier transform of the horizontal derivative
of the images shown in Figure 6.6 can be found in Refs. [22, 31].

The interfacial asymmetry can also be observed in cross-sectional STM spec-
troscopy studies. In these studies, STM spectra were acquired every 1.06 A along
the growth direction over a distance greater than one SL period. Subsequent analy-
sis of the spatially resolved spectroscopy results showed that for the InAs-on-GaSb
growth direction, the interface spectra are smeared out over a range of ~6 A.
These results have been reproduced over a number of data sets using different
samples and probe tips. Thus, the cross-sectional STM imaging and spectroscopy
results both show direct evidence of the asymmetry in the InAs/GaSb interface.

However, since STM doesn’t have direct chemical sensitivity, it was not possible

171

(a) InSb-like Interfaces

[001] growth —— 2 nm

(b) GAaAs-like Interfaces

efits «= L001] growth — 2 nn

Figure 6.6: Cross-sectional STM images of the InAs/GaSb SLs grown with (a)
InSb-like interfaces and (b) GaAs-like interfaces.

172

to determine the exact mechanism for the asymmetry.

6.3.4 Underlying Causes of Interfacial Asymmetry

In order to determine the underlying causes for the observed asymmetry in the
InAs/GaSb interface, chemically sensitive characterization techniques were re-
quired. Two techniques were used in this study: XPS and SIMS. As mentioned
previously, XPS has the advantages of being a near surface, in situ characteri-
zation technique, while SIMS offers high chemical sensitivity and depth profiling
capabilities.

Two possible mechanisms for non-abruptness of the InAs-on-GaSb interface
were proposed but not proved in the previous section: GaAs growth during the
interface formation and Sb cross-incorporation into the InAs from the underlying
GaSb. The first of the possibilities has already been demonstrated. The exchange
reaction studies in Chapter 5 showed conclusively that an Sb to As exchange can
occur during the formation of an InAs-on-GaSb interface, resulting in an extended
GaAs-like interface. Also, referring to the GaAs-like interface entries in Table 6.3
(b) we see that the Sb 4d to Ga 3d and As 3d to In 4d peak intensity ratios are
much smaller and larger than the respective values for GaSb and InAs standards.
While by itself this data is not conclusive, comparing it with the data for the other
interface types and growth directions leads to the conclusion that at least some of
the GaAs-like interfaces for the InAs-on-GaSb growth direction are extended into
the underlying GaSb.

Confirmation of cross-incorporation of Sb from underlying GaSb layers into
InAs overlayers was much more difficult. While incorporation of Sb from back-
ground Sb in the chamber has been observed, cross-incorporation of Sb from un-
derlying layers has never been confirmed. Jn situ XPS and ion sputtering results
are shown in Figure 6.7. The top profile in this figure shows an XPS spectrum for

an InAs-on-GaSb heterojunction grown with a nominal InSb-like interface and a ~

173

So 4d In 4d
Ga 3d

Unsputtered

Sputtered

ane

Ae
. S\—
eb L\
: wae
SM
SJ\

a 5 A

50 40 30
Binding Energy (eV)

Figure 6.7: XPS/ion sputtering studies. Top profile shows XPS spectra for an
unsputtered InAs-on-GaSb heterojunction. Profiles below were acquired after suc-

cessive ion sputtering sequences.

174

20 A InAs overlayer. After acquisition of the top spectrum, the sample was lightly
sputtered using 2.5 keV Ar* ions at an incident angle of 61 degrees to the surface
normal, and another XPS spectrum was acquired. This procedure was repeated
until a clear decrease in the In and As core level peak intensities was observed.
The duration of each sputter, except for the final one, was 12 seconds. The bottom
profile was acquired after almost all of the InAs overlayer had been removed. The
system did not have a Faraday cup installed so the ion beam current is unknown.

The effect of the ion sputtering is clearly seen in the decrease in the As 3d
and In 4d peak intensities. Detailed analysis of the peak intensity ratios show two
important features. First, the Sb 4d to Ga 3d peak intensity ratio decreases after
the first sputter. This is very significant in that, based on the photoelectron MFPs
for the two peaks, this ratio should increase as the InAs layer is thinned. The most
likely explanation for the observed decrease in the Sb to Ga peak intensity ratio,
given the well known surfactant properties of Sb, is that Sb is riding up on the
InAs growth surface.

To verify that Sb can ride up on the InAs growth surface, a second sample
consisting of 0.25 um of InAs grown on AlSb was studied. The results of this
study are shown in Figure 6.8. This figure shows the XPS core level peaks for Sb
3d just after growth. To ensure that the 5b signal was not due to incorporation
from background Sb within the growth chamber, the sample was ion sputtered
to a depth of a few tens of angstroms. The absence of an XPS Sb signal after
the sputtering indicates that the Sb signal was originating from the surface of the
sample and was due to Sb riding up on the InAs growth front.

The other important feature in Figure 6.7 is that the Sb 4d to Ga 3d peak
intensity ratio continues to decrease even after the first few sputter /XPS sequences.
This suggests the possible existence of Sb incorporation within the InAs near the
interface. Unfortunately, it is not possible to determine with certainty whether or
not this is true from the XPS/sputter results. This is due to the fact that sputter

induced drive-in of Sb, that was originally residing on the surface of the sample

175

(a)

Before Sputtering
7 Sb 3d 5/2
aS)

Sb 3d 3/2
£&
540 635 530 - 525
(bb)

After Sputtering
w”
Cc
ia Hep ANS ys waynes ASA nnn WA ena id

540 535 530 525

Binding Energy (eV)

Figure 6.8: XPS studies of Sb riding on the InAs growth front. (a) shows 0.25 um

InAs layer grown on AlSb. (b) shows same sample after Ar ion sputtering to a

depth of a few nm.

176

prior to the first sputter sequence, could result in an anomalous Sb signal from
within the unsputtered InAs. This would have the same effect on the sputtering
results as Sb cross-incorporation.

Since one of the main purposes of this study was to determine the mechanisms
for non-abruptness of the InAs-on-GaSb interface, it was very important to deter-
mine with certainty whether or not Sb incorporates within the InAs layer. The
difficulty in determining this using XPS is that any Sb signal within the InAs layer
must be distinguished from the very strong Sb signal originating from the GaSb
layer and the signal originating from the Sb at the surface of the InAs. The latter
signal is also relatively strong, since it originates from the surface of the sample
and is therefore not attenuated. Angle resolved XPS studies and analysis of higher
binding energy peaks was performed; however, it was not possible to determine
whether or not Sb was cross-incorporating into the InAs.

The characterization technique that eventually succeeded in determining with
certainty that Sb was incorporating into the InAs was SIMS. The first attempt at
this was basically to SIMS depth profile a sample consisting of a series of InAs-on-
GaSb heterointerfaces. The hope was that, by overlaying the Sb and Ga profiles
from the SIMS data, any cross-incorporated Sb would show itself in the form of
a broader onset for the Sb signal compared to the Ga signal. That is, if cross-
incorporation was occurring, then as the depth profiling proceeded through the
InAs to the GaSb, the Sb signal would increase before the Ga signal. Unfortu-
nately, the depth resolution of the SIMS was not fine enough to conclude either
the existence or non-existence of cross-incorporation. It was estimated that an ideal
abrupt GaSb interface would have an onset slope of 25 A/decade due to sputter
roughening and drive-in effects. Since the concentration of any Sb incorporation
would be small and would decrease rapidly away from the InAs/GaSb interface, it
would not be distinguishable from the inherent broadening of the signal.

The results from a second SIMS profile experiment are shown in Figure 6.9. In

this experiment, the SIMS sample consisted of only two InAs on GaSb interfaces:

177
(a)
107 . :

Low T, (380 °C)
---+ High T, (475 °C)

107 }
InAs GaSb

Sb Concentration (ctoms/cc)

107" 5 7
Low T,
(380 °C)
10” ~~ — High T, 7
po ~se (475 °C)
1 0” l L
500 600 700 800
Depth (angstroms)
(b)
1075
~ Low T, (380 °C)
910% - ~~~. High T, (475 °C) ;
~~
5 cat
2 10? InAs GoSb 3
ed <_—_— | ——
3 10”
rs
a ——
2 10" 77 Low T, —»
8 (380 °C)
° High T,
° 10% - (A475 °C) me |
10” .
500 600 700 800

Depth (angstroms)

Figure 6.9: SIMS analysis of InAs-on-GaSb interface abruptness. (a) shows Sb
concentration profiles for both a high and a low temperature grown InAs-on-GaSb
interface. (b) shows corresponding Ga concentration profiles. The results shows

cross-incorporation of Sb into InAs from the underlying GaSb.

178

one grown at the usual growth temperature (~ 380°C), and one grown at a higher
growth temperature (~ 475°C), where cross-incorporation is expected to be less of
an effect. At the higher growth temperatures, background As incorporates more
readily into the GaSb, which is an undesirable effect for devices, but this does not
affect the SIMS analysis. The hope was that by a comparison of the Sb profiles for
the two growth temperatures, existence or non-existence of cross-incorporated Sb
could be confirmed. The order in which the two interfaces were grown was critical
in this experiment. Since the bottom interface suffers additional broadening due
to sputtering effects, it was necessary to have the low temperature grown interface
closer to the sample surface, so that any broadening in that interface compared to
the high temperature grown interface could not be attributed to artifacts of the
ion sputtering.

In Figure 6.9(a) we see that the low temperature grown interface does have a
broader onset profile in the region of ~ 580-640 A, despite being closer to the sam-
ple surface. Notice that we are interested in the shape of the onset profile, not in
the absolute concentrations at the left-hand side of each plot. This effect in the Sb
profiles was duplicated in two separate SIMS data acquisitions. To verify that this
effect was indeed due to Sb cross-incorporation and not a result of anomalously
high sputter induced roughness in the low temperature grown interface, the Ga
profiles were also analyzed. In Figure 6.9(b) we see that the high temperature in-
terface has a broader onset profile than the low temperature grown interface. This
is consistent with increased sputter induced roughness as the SIMS profiling pro-
ceeds, and verifies that the effect seen in Figure 6.9(a) is due to cross-incorporation

of Sb into the InAs from the underlying GaSb.

6.4 Summary

In conclusion, we have measured the variation in the InAs/GaSb valence band offset

as a function of both interface composition (InSb-like and GaAs-like) and growth

179

Extended [ Abrupt

b Sie)

Ves,

Figure 6.10: Schematic diagram of the interface asymmetry observed in the

InAs/GaSb material system.

order (InAs-on-GaSb and GaSb-on-InAs). The InAs/GaSb valence band offset
was observed to be independent of interface composition for both growth orders;
however, upon comparison between the two growth orders, a 90 meV difference in
the InAs/GaSb valence band offset was observed, with the InAs-on-GaSb interface
showing the larger valence band offset. The difference in valence band offset is
attributed to the extended nature of the InAs-on-GaSb interface. To test this, a
GaSb-on-InAs sample was grow with an intentionally extended InSb-like interface.
This resulted in a valence band offset between that obtained for the nominally
abrupt GaSb-on-InAs samples and the extended InAs-on-GaSb samples. This
confirmed our hypothesis that the change in the valence band offset was due to
the non-abrupt nature of the InAs-on-GaSb interface.

Initial studies of interface abruptness for the InAs/GaSb heterointerface re-
vealed an asymmetry in the abruptness for the two growth orders: InAs-on-GaSb
and GaSb-on-InAs. XPS, RHEED and cross-sectional STM results showed that
the InAs-on-GaSb interface was more extended than the GaSb-on-InAs interface.

This is shown schematically in Figure 6.10. XPS analysis of core level peak in-

180

tensity ratios in heterojunction samples and studies of surface exchange reactions
showed that one mode of broadening was As exchanging into the GaSb, forming
an extended GaAs-like interface. The driving force for this is the large GaAs bond
strength compared to that of GaSb. XPS analysis was also used to show that Sb
can ride up on the InAs growth front, but cross-incorporation of Sb into the InAs
from the underlying GaSb could not be confirmed with certainty. SIMS analysis
was used to clearly determine that Sb cross-incorporation was occurring. The rel-
ative abruptness of the GaSb-on-InAs interface can be explained first by the fact
that As is much less of a surfactant than Sb, so As cross-incorporation into GaSb
from the underlying InAs is less likely, and secondly by the fact that the InSb
bond strength is much less than that of InAs. Thus, as was shown in the surface
exchange reactions in Chapter 5, Sb does not exchange for As into the underlying
InAs.

These basic forces causing the InAs-on-GaSb interface to be extended can po-
tentially cause problems in devices implementing As/Sb interfaces. However, in
some of these devices, abruptness of the interfaces may not be crucial. In those
devices where interfaces comprise a large fraction of the device and are critical to
device performance (e.g., InAs/In,Ga,_,Sb IR SLs), it will be very important to
determine first of all what effect the broadening of the InAs-on-GaSb interface has
on device performance. Secondly, if the extended interface is detrimental to device
performance, it must be determined whether or not the extended InAs-on-GaSb
interface is fundamental or if different growth conditions and procedures can be

used to grow an abrupt InAs-on-GaSb interface.

181

Bibliography

[1] RHEED studies performed by Dr. D.A. Collins, STM studies performed by
Dr. R.M. Feenstra at IBM Watson, and SIMS studies performed by researchers

at Charles Evans and Associates.

[2] E.R. Brown, J.R. Séderstrém, C.D. Parker, L.J. Mahoney, K.M. Molvar, and
T.C. McGill, Appl. Phys. Lett. 58, 2291 (1991).

[3] M. Sweeny and J. Xu, Appl. Phys. Lett. 54, 546 (1989).

[4] J.R. Séderstrém, D.H. Chow, and T.C. McGill, Appl. Phys. Lett. 55, 1094
(1989).

[5] L.F. Luo, R. Beresford, and W.I. Wang, Appl. Phys. Lett. 55, 2023 (1989).

[6] D.A. Collins, E.T. Yu, Y. Rajakarunanayake, J.R. Séderstrém, D.Z.-Y. Ting,
D.H. Chow, and T.C. McGill, Appl. Phys. Lett. 57, 683 (1990).

[7] D.L. Smith and C. Mailhiot, J. Appl. Phys. 62, 2545 (1987).

[8] R.H. Miles, D.H. Chow, J.N. Schulman, and T.C. McGill, Appl. Phys. Lett.
57, 801 (1990).

[9] H. Sakaki, L.L. Chang, R. Ludeke, C.-A. Chang, G.A. Sai-Halasz, and L.
Esaki, Appl. Phys. Lett. 31, 211 (1977).

[10] G.A. Sai-Halasz, L.L. Chang, J.-M. Welter, C.-A. Chang, and L. Esaki, Solid
State Commun. 27, 935 (1978).

182

[11] L.M. Claessen, J.C. Maan, M. Altarilli, and P. Wyder, Phys. Rev. Lett. 57,
2556 (1986).

[12] G.J. Gualtieri, G.P. Schwartz, R.G. Nuzzo, R.J. Malik, and J.F. Walker, J.
Appl. Phys. 61, 5337 (1987).

[13] Y. Foulon and C. Priester Phys. Rev. B 45, 6259 (1992).

[14] R.G. Dandrea, C.B. Duke, and A. Zunger, J. Vac. Sci. Technol. B 10, 1744
(1992).

[15] L.A. Hemstreet, C.Y. Fong, and J.S. Nelson, J. Vac. Sci. Technol. B 11, 1693
(1993).

[16] J.P. Landesman, J.C. Garcia, J. Massies, G. Jezequel, P. Maurel, J.P. Hirtz,
and P. Alnot, J. Vac. Sci. Technol. B 10, 1761 (1992).

[17] M.S. Hybertsen, Materials Science and Engineering B14, 254 (1992).

[18] G.J. Gualtieri, G.P. Schwartz, R.G. Nuzzo, and W.A. Sunder, Appl. Phys.
Lett. 49, 1037 (1986).

[19] J.R. Waldrop, G.J. Sullivan, R.W. Grant, E.A. Kraut, and W.A. Harrison, J.
Vac. Sci. Technol. B 10, 1773 (1992).

[20] D.H. Chow, R.H. Miles, C.W. Nieh, and T.C. McGill, J. Cryst. Growth 111,
683 (1991).

[21] M.W. Wang, D.A. Collins, T.C. McGill, and R.W. Grant, J. Vac. Sci. Technol.
B 11, 1418 (1993).

[22] R.M. Feenstra, D.A. Collins, D.Z.-Y. Ting, M.W. Wang, and T.C. McGill,
Phys. Rev. Lett. 72, 2749 (1994).

[23] D.A. Collins, M.W. Wang, R.W. Grant, R.M. Feenstra, and T.C. McGill,
submitted to Appl. Phys. Lett. (1994).

183
[24] W.A. Harrison, E.A. Kraut, J.R. Waldrop, and R.W. Grant, Phys. Rev. B
18, 4402 (1978).

[25] R.H. Miles, D.H. Chow, J. N. Schulman, D.A. Collins, M.W. Wang, R.W.
Grant, and T.C. McGill, submitted to J. Elect. Materials.

[26] E.R. Youngdale, J.R. Meyer, C.A. Hoffman, F.J. Bartoli, C.H. Grein, P.M.
Young, H. Ehrenreich, R.H. Miles, and D.H. Chow, Appl. Phys. Lett. 64,
3160 (1994).

[27] M.P. Seah and W.A. Dench, Surf. Interface Anal. 1, 2 (1979).
[28] C.M. Kwei and L.W. Chen, Surf. Interface Anal. 11, 60 (1988).
[29] S. Tanuma, C.J. Powell, and D.R. Penn, Surf. Interface Anal. 17, 927 (1991).

[30] W.C. Fan, J.T. Zborowski, T.D. Golding, and H.D. Shih, J. Appl. Phys. 71,
2249 (1992).

[31] R.M. Feenstra, D.A. Collins, D.Z.-Y. Ting, M.W. Wang, and T.C. McGill, to
be published in J. Vac. Sci. Technol. B (1994).