Investigation of new devices and characterization techniques in the III-V semiconductor system - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Investigation of new devices and characterization techniques in the III-V semiconductor system
Citation
Cheng, Xiao-chang
(1999)
Investigation of new devices and characterization techniques in the III-V semiconductor system.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/4pmy-xb16.
Abstract
This thesis concerns the investigation of novel devices and material characterization techniques in the III-V semiconductor system. In the first part of the thesis, we demonstrate that novel devices, such as avalanche photodiodes and tunnel switch diodes, can be fabricated from InAs/GaSb/A1Sb heterostructures by molecular beam epitaxy (MBE). In the second part of the thesis, ballistic electron emission microscopy (BEEM) is employed to examine the local band offset in these heterostructures, which is often found to be crucial in device design.
In the avalanche photodiode study, devices with near infrared response out to 1.74 [mu]m were demonstrated. Two types of devices were investigated: those with a bulk Al0.04Ga0.96Sb multiplication region and those with a GaSb/A1Sb superlattice multiplication region. Both types of devices were implemented in a MBE grown p[superscript -]n[superscript +] structure that uses a selectively doped InAs/AlSb superlattice as the n-type layer. This particular structure was optimized through several design, fabrication, characterization cycles. It was found that the photodiode dark current depended critically on the InAs/A1Sb superlattice period and the resulting band offset at the p[superscript -]n[superscript +] heterojunction. The InAs/AlSb superlattice was henceforth optimized by using a three stage design. The ionization rates in bulk multiplication layer devices were measured and found to be consistent with hole impact ionization enhancement in Al0.04Ga0.96Sb. However, direct comparison with superlattice multiplication layer devices revealed the latter to be more promising due to more effective dark current suppression from the larger band gap of the superlattice multiplication layer.
The second device studied is the tunnel switch diode. We have fabricated the first such device in the antimonide material system and obtained characteristic "S" shaped I-V curves from these devices. The epilayer and barrier dependence of tunnel diode switching were studied and found to deviate significantly from the punch-through model of operation. In addition, the device I-V curve was observed to "hop" between two branches when subjected to high levels of stress. We speculate that this was due to instability associated with mobile charges in the A1Sb tunnel barrier. A computer model was used to simulate the device behavior and generated results consistent with the observed dependence of switching on tunnel barrier thickness.
In the second part of the thesis, III-V heterostructures were characterized by using ballistic electron emission microscopy (BEEM). BEEM images were shown to reveal sub surface features in AlxGal-xAs. epilayers, whereas BEEM spectroscopy was used to map out the shift in [capital gamma], X, and L band edges with material composition in AlxGal-xAs. BEEM spectroscopy was also applied to device relevant antimonide heterostructures such as A1Sb barriers and InAs/AlSb superlattices. It was found that electron transport in A1Sb was dictated by the conduction band minium near the X point, and there is large local variation in the AlSb Schottky barrier height. These results were in good correlation with the observed barrier characteristics of A1Sb. Due to the small bandgap of InAs/A1Sb superlattice and the associated high level of noise current, only the shortest period superlattice was examined by BEEM. The resulting band offset agreed with the calculated value and demonstrated that BEEM spectroscopy can be applied to structures with a large number of hetero-interfaces.
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:
4 May 1999
Record Number:
CaltechETD:etd-12122007-134645
Persistent URL:
DOI:
10.7907/4pmy-xb16
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
4974
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
12 Dec 2007
Last Modified:
19 Apr 2021 22:25
Thesis Files
Preview
PDF (Cheng_xc_1999.pdf)
- Final Version
See Usage Policy.
10MB
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
Investigation of New Devices and Characterization
Techniques in the III-V Semiconductor System
Thesis by
Xiao-chang Cheng
In Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy
California Institute of Technology
Pasadena, California
1999
(Submitted May 4, 1999)
il
Xiao-chang Cheng
il
To my parents
iv
Acknowledgements
I am grateful to my advisor, Tom McGill, for having faith in me and providing nu-
merous opportunities throughout my graduate career. Over the years, I have come
to appreciate Tom’s incredible sharp instincts for finding interesting research direc-
tions and his ability to accommodate a diverse group of talented individuals. I am
constantly amazed by his flexibility and yet hold great. admiration for the instances
when he stuck to the principles. If I showed up years ago at the door of graduate
school with scarcely any idea about “state of the art” research operations, Tom is the
person who changed that.
I have benefited tremendously from my associations with former and current mem-
bers of the McGill group. I am indebted to Doug Collins and David Chow for showing
me the art of molecular beam epitaxy. Doug has been my initiation to the world of
experimental semiconductor physics whereas David is the consummate professional
and a role model for any aspiring young researcher. I am also grateful to Andy Hunter
and Gerry Picus. Without their guidance, much of the avalanche photodiode work
in this thesis would not have been possible. I have to thank Erik Daniel for riding
out the frustration in antimonide processing together and laying down the foundation
work for the tunnel switch diode simulation. Erik is one of the most generous people
around and I have picked up many things from him, not the least of which is a taste
for Jackie Chan movies. I must also thank Rob Miles for introducing me to the world
of scanning probe microscopy. Rob built the ballistic electron emission microscopy
set-up described in this thesis, which enabled much of the ensuing characterization
work. I would also like to thank David Ting, who has been a constant influence in
the group and has reached out to me on numerous occasions. Ogden Marsh and Jim
McCaldin have also had a constant presence in the research background. Ogden has
provided many useful tips and his sunny disposition has been infectious. Over the
years, I have also enjoyed interactions with group members Harold Levy, Yixin Liu,
vi
Ron Marquardt, Johanes Swenberg, Mike Wang, Peo Petterson, and Chris Springfield.
Among current members of the McGill group, I am most grateful to Eric Piquette
for his gentlemanly manner, which I came to admire during our co-tenure in the clean
room. I have enjoyed my collaborations with Alicia Alonzo and appreciated very much
her defensive stand for the II-V MBE system. Zvonimir Bandic has been a reliable
source of enthusiasm and his openness has been refreshing. I am also grateful to fellow
Canadian Paul Bridger, who has been the optics guru in the group and provided
numerous experimental advices. I have also benefited from my associations with
the younger graduate students, including Joe Jones, Bob Beach, Cory Hill, Xavier
Cartoixa, Matthew Barton, and Rob Strittmatter. Cory has been very helpful as
the group system administrator and is always on the spot when computer difficulties
arise. Xavier is responsible for the simulation work in the tunnel switch diode part
of this thesis, while Matt has contributed towards the characterization of the TSD
devices. I hope greater things lay ahead for this talented bunch of individuals.
I would like to thank Marcia Hudson, Gloria Pendlay, and Tim Harris for their
excellent administrative work. I would also like to acknowledge the National Science
and Engineering Research Council of Canada for providing financial support during
my stay at Caltech. Many thanks go back to Bob Fedosejevs and Jim McMullin for
opening the door of research to me during my undergraduate days at the University
of Alberta.
I would like to thank Jiang Wen and Xu Weihua, whose friendship I will always
treasure. Special thanks must go to Song Yang, who has changed my life at Caltech.
Without her love and support, this thesis would not have been possible. Finally, I
would like to thank my parents and sister. Their unconditional, unbounded love and
support have been a true blessing.
vil
List of Publications
Work related to this thesis has been or will be presented in the following papers:
Tunnel Switch Diode Based on AlSb/GaSb Heterojunctions,
X-C. Cheng, X. Cartoixa, M. A. Barton, C. J. Hill, and T. C. McGill, submitted
to Applied Physics Letters.
Near Infrared Avalanche Photodiodes with Alpo,4GagogSb and
GaSb/AISb Superlattice Gain Layers,
X-C. Cheng and T.C. McGill, submitted to the Journal of Applied Physics.
Molecular Beam Epitaxy Growth of Antimonide Avalanche Photo
Detectors with InAs/AISb Superlattice as the n-type Layer,
X-C. Cheng and T.C. McGill, to be published in the Journal of Crystal Growth.
Avalanche Photo Detectors in the InAs/GaSb/AISb Material System
by Molecular Beam Epitaxy,
X-C. Cheng and T.C. McGill, SPIE International Symposium on Optoelectron-
ics, 1999.
Ballistic Electron Emission Microscopy Spectroscopy Study of AlSb
and InAs/AISb Superlattice Barriers,
X-C. Cheng and T.C. McGill, Journal of Vacuum Science and Technology B
16, pp 2291-2295, 1998.
Mapping of Al,Ga,_,As Band Edges by Ballistic Electron Emission
Spectroscopy,
X-C. Cheng, D. A. Collins, and T.C. McGill, Journal of Vacuum Science and
Technology A 15, pp 2063-2068, 1997.
Vill
Work not included in this thesis has been or will be presented in the following papers:
Strain in wet thermally oxidized square and circular mesas,
A.C. Alonzo, X-C. Cheng, and T.C. McGill, in preparation.
Effect of cylindrical geometry on the wet thermal oxidation of AlAs,
A.C, Alonzo, X-C. Cheng, and T.C. McGill, Journal of Applied Physics 84, pp.
6901-6905, 1998.
ix
Abstract
This thesis concerns the investigation of novel devices and material characterization
techniques in the III-V semiconductor system. In the first part of the thesis, we
demonstrate that novel devices, such as avalanche photodiodes and tunnel switch
diodes, can be fabricated from InAs/GaSb/AISb heterostructures by molecular beam
epitaxy (MBE). In the second part of the thesis, ballistic electron emission microscopy
(BEEM) is employed to examine the local band offset in these heterostructures, which
is often found to be crucial in device design.
In the avalanche photodiode study, devices with near infrared response out to
1.74 zm were demonstrated. Two types of devices were investigated: those with a
bulk Alo.o1Gao.ggSb multiplication region and those with a GaSb/AISb superlattice
multiplication region. Both types of devices were implemented in a MBE grown
pn structure that uses a selectively doped InAs/AISb superlattice as the n-type
layer. This particular structure was optimized through several design, fabrication,
characterization cycles. It was found that the photodiode dark current depended
critically on the InAs/AISb superlattice period and the resulting band offset at the
p-n* heterojunction. The InAs/AISb superlattice was henceforth optimized by using
a three stage design. The ionization rates in bulk multiplication layer devices were
measured and found to be consistent with hole impact ionization enhancement in
AloosGagogSb. However, direct comparison with superlattice multiplication layer
devices revealed the latter to be more promising due to more effective dark current
suppression from the larger band gap of the superlattice multiplication layer.
The second device studied is the tunnel switch diode. We have fabricated the first,
such device in the antimonide material system and obtained characteristic “S” shaped
I-V curves from these devices. The epilayer and barrier dependence of tunnel diode
switching were studied and found to deviate significantly from the punch-through
model of operation. In addition, the device I-V curve was observed to “hop” between
two branches when subjected to high levels of stress. We speculate that this was due
to instability associated with mobile charges in the AlSb tunnel barrier. A computer
model was used to simulate the device behavior and generated results consistent with
the observed dependence of switching on tunnel barrier thickness.
In the second part of the thesis, III-V heterostructures were characterized by
using ballistic electron emission microscopy (BEEM). BEEM images were shown to
reveal sub surface features in Al,Ga,_,As epilayers, whereas BEEM spectroscopy
was used to map out the shift in T, X, and L band edges with material composition
in Al,Ga,_,As. BEEM spectroscopy was also applied to device relevant antimonide
heterostructures such as AlSb barriers and InAs/AISb superlattices. It was found that
electron transport in AlSb was dictated by the conduction band minium near the X
point, and there is large local variation in the AlSb Schottky barrier height. These
results were in good correlation with the observed barrier characteristics of AlSb.
Due to the small bandgap of InAs/AISb superlattice and the associated high level
of noise current, only the shortest period superlattice was examined by BEEM. The
resulting band offset agreed with the calculated value and demonstrated that BEEM
spectroscopy can be applied to structures with a large number of hetero-interfaces.
Xl
Contents
Acknowledgements
List of Publications
Abstract
1 Introduction
1.1 Thesis Overview... 0.20.00 00
1.2 Motivation... 2... 2... 2
1.2.1 InAs/GaSb/AlSb system... 0. ..0.00.020.00.0.2000,
1.2.2 Antimonide Avalanche Photodiode ..........02..,
1.2.3 Antimonide Tunnel Switch Diode .........020.002..
1.2.4 Ballistic Electron Emission Microscopy ............
1.3 MBE crystal Growth ........0..0.0.00.00.. 0.002.000.
1.3.1 MBE Environment .............0....0..00.0,
1.3.2 UI-V Growth ...........0..0.0 2.000.000.0000.
Bibliography . 2... 2.
I Novel Devices Based on InAs/GaSb/AISb Het-
erostructures
2 Avalanche Photodiode Theory
2.1 Introduction to Chapter ..........0.0..0.0.0..000000,
2.2 Basic Operation... 2.0... 2000200000000 00 ee
2.3 Impact Ionization Ratio . 2... 0...
2.4 Impact Ionization Enhancement ........0....20.0.002,
vii
Ix
uo DW Wo +
10
11
12
13
16
19
21
21
21
23
26
2.9
xii
2.4.1 Hole Impact Ionization Enhancement from Spin-orbit Split-off
Band Resonance ....................2.00.2
2.4.2 Electron Impact Ionization Enhancement from Superlattice
Band Offset... 2...
Practical Considerations: Dark Current. .........2.2.2.2.2..
Bibliography ........0.0.0.0 00000000000 ee,
Design, Fabrication, and Characterization of Avalanche Photodi-
odes
3.1
Introduction to Chapter... .....0.0.00.020.0..0.0.0.000000.
3.2 Device Design... 2... 0.020.000.0200. 000000022 2,
3.2.1 Multiplication Layer .. 2.2... .0.0200202002020000042.
3.2.2 _InAs/AlSb n-type Superlattice. . 2.2...
3.3 Growth 2.0... 0...
3.3.1 Buffer and Multiplication Layers .........02002..
3.3.2 InAs/AlSb Superlattice. 2...
3.4 Processing... 2...
3.4.1 Photolithography ...........0.......0...0.2.
3.4.2 Etching ...........2..0.00 200.000.000.000.
3.4.3 Passivation . 2... 2...
3.0 I-V and Photo Response Characterization ...........202..
Bibliography ... 2... .00200 00000000000 ee
Results of Avalanche Photodiode Study
4.1
4.2
4.3
4.4
4.5
Introduction to Chapter .........0.00.0.0.0.002.00000,
Harly Results 2... 20...
Effect of InAs/AISb Superlattice Period on Dark Current... .. .
Optimization of InAs/AISb Superlattice Design ... 0.020000;
Results from Bulk Alp osGapggS5b Devices ........2.2.200202.
4.5.1 Photo Response Unity Gain Correction... ......002.
4.5.2 Photo Gain and Dark Current Characteristics .......~.
26
30
32
34
36
36
36
37
39
41
Al
42
46
46
AT
49
50
53
xi
4.5.3 Impact Ionization Rates ... 2... ..020022002200200.., 67
4.6 Results from Superlattice Devices... 2... 2.0. 71
4.6.1 Photo Gain and Dark Current Characteristics . 2... 2... 72
4.6.2 Quantum Efficiency at Low Bias... .......0.00202. 75
4.7 Summary and Conclusion ...........0.........004. 76
Bibliography . 2... 0. 78
5 Tunnel Switch Diodes Based on AlSb/GaSb Heterojunctions 79
5.1 Introduction to Chapter ........0..0.00.00.0..00.0000. 79
5.2 Motivation and Background ...................00. 79
3.3 Tunnel Switch Diode Theory and Design .............02.. 81
5.4 Growth and Fabrication ..........0.0.020.0....00.00., 85
5.5 Characterization Results... ....0.0.002.0.0..2.0.00028., 86
5.5.1 Effect of Barrier and Epilayer Thickness on Switching... . 87
5.5.2 Effect of Current Stress and Dual Mode Switching Behavior 90
5.6 Simulations .. 2... 20.00.0000 0000 ee 94
5.7 Summary and Conclusion .............0.....2.000., 97
Bibliography .... 2.0.0.0. 02000000000 ee 98
II Ballistic Electron Emission Microscopy Study of ITI-V
Heterostructures 101
6 Ballistic Electron Emission Microscopy Theory and Experiment 103
6.1 Introduction to Chapter .......0..0..0...0.0.0.0020, 103
6.2 BEEM Theory .............0. 00000000000 200, 103
6.2.1 Basic Operation of BEEM ..........020...0020.. 103
6.2.2 Parabolic Turn On Model ........0.00202000002. 105
6.2.3 Sample Requirements... ...............00.. 107
6.3 BEEM Experiment ................0...00..0000. 109
6.3.1 Apparatus... 2.2... 0000000000000. 000004 109
6.3.2 Experimental Issues... 2... ...0.020200.200..2..0.0. 110
Bibliography 2.2... 114
BEEM Study of Al,Ga,_,As 115
7.1 Introduction to Chapter .........0....0..0..0.0000. 115
7.2 Motivation... 2... 2. 115
7.3 Sample Description and Preparation ..............02... 117
7.4 BEEM Imaging Results... .........0.20.0.......04. 120
7.5 BEEM Spectroscopy Results... .....0.00.0.0.0.0.....00. 124
7.5.1 BEEM Turn on Threshold .................2. 124
7.5.2 Effect of Epilayer Thickness and Capping Layer ....... 126
7.5.3 Variation of Band Edge and Schottky Barrier Height with Al
Concentration... 0.0... 127
7.0.4 Mapping of the Relative Position of Band Edges in Al,Ga;_,As 130
7.6 Summary and Conclusion .........0........0.0002. 132
Bibliography .. 2... 0. 133
BEEM Study of AlSb and InAs/AISb Superlattice 135
8.1 Introduction to Chapter ........0.02.0..0.0.2.00.0000. 135
8.2 Motivation... 2... 135
8.3 Sample Description and Preparation ................. 138
8.4 Results from AlSb Study... ....20.20.2.0..0..0..0 200.0, 140
8.4.1 Effect of Sample Configuration ............0202.. 140
8.4.2. BEEM Characterization ..................2.. 142
8.5 Results from Superlattice Study... 2.0.0.0. 145
8.5.1 Effect of Superlattice Period... 2... .00200200002.. 145
8.5.2 Results from 12 A/12 A, InAs/AlSb Superlattice 2.0... . 146
8.6 Summary and Conclusion .........0..........000. 147
Bibliography 2.2... 200.0000. 000000 2 151
XV
List of Figures
1.1
1.2
1.3
1.4
1.5
1.6
1.7
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
InAs/GaSb/AISb lattice matched system. (a) Band alignment. (b)
Negative Schottky barrier between InAs and metal. ........~.,
Spectrum of lunar light. . 2... 0.0 ee
Transistorless static random access memory from a tunnel switch
diode. (a) Circuit schematic. (b) Load line analysis... . 2... .«.
Ballistic electron emission microscopy. ................
Schematic of molecular bean epitaxy set-up. ..........202.
Current voltage behavior of antimonide RIT grown on GaAs wafer.
AFM scan of dots formed from two monolayers of GaSb deposited
on GaAs substrate. Note that the large bright object on the left is a
dust particle. 2.2...
Band diagram and basic operation of an avalanche photodiode [1]. .
Schematic representation of avalanche multiplication process. (a)
a=G. (b) B=0. 2...
Field dependence of multiplication factor for different ionization ra-
Effect. of impact ionization ratio on excess noise factor [3]... . . .
Band structure view of an electron initiated impact ionization pro-
cess [8].
Hole initiated impact ionization process involving the spin-orbit split-
off band [9.22
Electron impact ionization enhancement in a GaAs/AlGaAs super-
lattice multiplication layer... 2... 0... 2
Noise equivalent power of an avalanche photodiode. .........,
11
12
14
22
23
24
26
27
29
31
33
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
3.10
3.11
3.12
Xvl
Band diagram of the antimonide avalanche photodiode grown
by MBE. The device had a p~n* configuration with a p~ bulk
Alo.oaGao.g6Sb or GaSb/AISb superlattice multiplication layer and
a selective doped n* InAs/AISb superlattice layer. The device is
shown under reverse bias. .. 2... ...0..0.20 20000200004
Variation of spin-orbit split-off band difference A and bandgap E,
with Al composition in Al,Gay_,Sb [4]... ..0.0000....02..
Calculated InAs/AISb superlattice bandgap energy and band overlap
with GaSb as a function of the superlattice period thickness [7]. The
InAs and AlSb layer thicknesses were assumed to be equal. .... .
X-ray diffraction scan of Al composition calibration sample. The Al
composition of the AlGaSb layer was varied by changing the Al cell
temperature during MBE growth. ................0...
X-ray diffraction scan of 27 A/27 A InAs/AISb superlattice (a) grown
at a high substrate temperature which resulted in excess As incorpo-
ration (b) grown under optimized conditions... ...........
Growth defects may form for short. period superlattices despite good
X-ray data. (a) X-ray scan of 10 A/20 A InAs/AISb superlattice. (b)
SEM scan of the same wafer... 2... ..0.0..002.0..2.2.2..002,4
Avalanche photodiode device mesas. (a) Simple mesa with thin metal
contact. (b) Mesa designed for direct injection of light through the
opening in contact metal... 2...
(a) Wet etched surface. (b) Dry etched surface with smoother surface
and fewer etch defects. ©... 2.0.2 2
Cross-sectional SEM micrograph of the device mesa... .......
Effect of post dry-etching processing steps on device dark current. .
Experimental setup for photo response characterization... .....
Electron and hole carrier injection by using light of different wave-
length, 2.
37
38
40
42
44
45
47
48
49
90
ol
52
Al
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
4.10
XVil
Current-voltage characteristics of first avalanche photodiode struc-
tures fabricated. (a) Scaling with device area. (b) Scaling with device
SIZ.
Low temperature I-V characteristics of first avalanche photodiodes
fabricated. 2 2.
Current-voltage characteristics of avalanche photodiodes with
Alo.osGao.ggSb gain layer and 27 A/ 27 A InAs/AISb superlattice
grown under different conditions. ...................
Current-voltage characteristics of avalanche photodiodes with
Alo.osGao.g6Sb gain layer and 10 A/ 20 A, InAs/AISb superlattice. .
Effect of superlattice period on the band alignment between n-type
InAs/AISb superlattice and the Alo.o1Gao.96Sb multiplication layer. .
(a) Avalanche photodiode structure with Aloo4Gao.9gSb gain layer
and optimized, three stage, n-type InAs/AISb superlattice. Device is
shown under reverse bias. (b) X-ray diffraction scan of the structure.
Current-voltage characteristics of avalanche photodiodes with
Alo.o1Gao.95b gain layer and optimized n-type InAs/AISb superlat-
tice,
Low temperature reverse break down characteristics of avalanche
photodiode with Alpo,GagogSb gain layer and optimized n-type
InAs/AlSb superlattice. 2... 0.00002... .2.22222004
Photo response of the Alp.o4Gao.9g5b gain layer device without. cor-
rection. The curves are fitted at low bias to correct for changes in
quantum efficiency with device bias... ..............2..
(a) Dark current and near infrared photo gain characteristics of
avalanche photodiodes with Alg.oqGag.9g5b as the multiplication layer.
The dashed line shows the un-multiplied dark current. (b) Device
dark current plotted as a function of the photo gain. The data was
fitted to a power law I=I,(M)?, where p=1 for constant un-multiplied
dark current. 2... 2 ee
56
oT
59
60
61
62
63
64
65
4.11
4.12
4.13
4.14
4.15
o.1
5.2
XVill
(a) Photo gain curves for hole and electron injection using 781 nm
and 1645 nm light. (b) Calculated hole and electron impact ionization
rates in Alo.osGao.9g5b. The device was assumed to have an abrupt
pn junction... 2...
Franz-Keldysh absorption of photons by semiconductors. Absorption
in the bandgap is possible due to overlapping of wave functions under
high field conditions [6). ..............2.....2.00..
(a) Dark current and photo gain characteristics of avalanche pho-
todiodes with a 10 period, 300 A/300 A, GaSb/AISb superlattice
multiplication layer. The dashed line shows the un-multiplied dark
current. (b) Device dark current plotted as a function of the photo
gain. The data was fitted to a power law I=I,(M)?, where p=1 for
constant un-multiplied dark current. ..........0.0...020..
Dark current scaling with device size for APD’s with (a)
AloosGao.965b multiplication layer and (b) GaSb/AISb superlat-
tice multiplication layer. The data is fitted to a power law I=I,(L)?,
where L is the device size. The curve fit should yield p=2 for perfect
scaling with device area and p=1 for perfect scaling with device
perimeter. 2...
Photo response of the GaSb/AISb superlattice avalanche photodiode
at. different light intensity levels. The data was obtained by using a
1645 nm laser light. Similar results were obtained from 781 nm and
1740 nm light sources. . 2... 2 el
Thyristor like, “S” shaped I-V curve of a tunnel switch diode.
Band diagrams of an antimonide TSD under forward bias. (a) High
impedance state with deep depletion in the p-GaSb epilayer. (b) Low
impedance state with the pn junction turned on and most of the bias
dropped across the AlSb barrier. The energy scale of the high and
low impedance states are shifted for clarity... ......00002.
70
71
73
74
76
80
82
5.3
5.4
5.9
5.6
oF
5.8
5.9
5.10
6.1
6.2
6.3
6.4
6.5
6.6
6.7
7.1
XIX
X-ray diffraction scan of an antimonide TSD structure. .......
Typical I-V characteristics of antimonide TSD’s.........202..
Effect of p-GaSb epilayer thickness on TSD switching voltage.
I-V characteristics and break down behavior of devices with thick
AlSb barriers... 2. 0
TSD break down from current stressing. .............02..
TSD I-V characteristics before and after current stressing. .... .
Clustering of switching voltages and currents. (a) 100 A AlSb barrier,
0.6 zm p-GaSb epilayer. (b) 200 A AlSb barrier, 0.6 um p-GaSb
epilayer, 2.
Hopping between stable I-V curves due to current stressing. (a) From
high branch to low branch. (b) From low branch to high branch. . .
Simulated characteristics for TSD devices with 0.6 ym p-GaSb epi-
layer and different AlSb thicknesses. Bistable states were obtained
for 100 A and 200 A barrier thicknesses. ...........0...
Basic operation of BEEM. ..........20.....0.....00.2.
Conservation of transverse momentum at base collector interface.
Sample requirement for BEEM [11]. (a) Band diagram. (b) Circuit
model... 2...
Experimental Setup for BEEM. ........0200020200002.
Artifact in BEEM image due to high scan speed... ......2.2..
STM images show surface modifications following BEEM spec-
troscopy at high tip voltages and currents. ..........0202..
Calibration measurements from Au/Si system. (a) Effect of STM
tip current on BEEM threshold. (b) Variation in the magnitude of
BEEM turn on with STM tip current. ......0.0000000,
(a) Measured shift in Al,Ga,_,As band edges with Al concentration.
(b) Band structure of GaAs... 2... 0.
85
87
88
89
90
91
92
93
96
104
106
108
109
110
lil
112
116
7.2
7.3
7.4
7.0
7.6
7.7
7.8
7.9
7.10
7.11
8.1
8.2
8.3
8.4
8.5
8.6
XX
(a) Structure of Al,Ga,;_,As BEEM sample. (b) Band diagram of
BEEM sample with AlAs epilayer..... 2... 0000000002.
BEEM images of buried Alo 11Gao.g9As interface. (a) 100 nm by 100
nm scan. (b) 500 nm by 500 nm scan... 2... 0...
BEEM images at different tip bias. The tip bias varied from 0.8 V in
the first picture to 2.2 V in the last picture. Scan area was 100 nm
by 100nm. 2...
Average signal from BEEM images at different tip bias... 2... .
(a) BEEM spectroscopy curve from an Aloi;Gaog9gAs sample. (b)
Extraction of BEEM threshold from differentiated curve. ......
Variation of BEEM threshold with Al,Ga,_,As layer thickness.
Effect of capping layer on BEEM threshold. (a) Alo.50Gao.50As sam-
ple. (b) AlAs sample... 2... 0.00000... 000000000,
Variation of Al,,Ga,_,As band edges with Al composition x. Multiple
data points at 7 = 0.50 and x = 1.0 are slightly offset for clarity. . .
P-type Schottky barrier height inferred from BEEM data. Multiple
data points at x = 0.50 and x = 1.0 are slightly offset for clarity. . .
Relative positions of the higher lying band edges as measured by
BEEM. Multiple data points at « = 0.50 and « = 1.0 are slightly
offset for clarity... 2... en
AlSb Schottky gate in dual channel mobility modulated transistor. .
Calculated variation of InAs/AISb superlattice bandgap and Schot-
tky barrier height with InAs layer thickness [11]... .........
Structure of antimonide BEEM sample... ...........002..
(a) LV characteristics of two types of AlISb BEEM samples. Mesa
size was 1 mm. (b) Corresponding band diagrams... ........
(a) BEEM L-V curves for AlSb samples. The tunneling current was
held constant at 10 nA. (b) BEEM threshold statistics. . 2... .
Band structure of AISb. .. 2... 0.000200. 0.200.0 004.
118
121
122
123
125
127
128
129
130
131
136
137
139
141
143
145
8.7
8.8
8.9
xxi
High resolution X-ray diffraction scan from the 12 A/12 A, InAs /AISb
superlattice BEEM sample. .............0...0.000.,
Current-voltage characteristics of 12 A/12 A, InAs/AISb superlattice
BEEM samples. (a) Liner plot. (b) Extraction of Schottky barrier
height from Log plot... 2... 0. 2
BEEM spectroscopy curves from 12 A/12 A, InAs /A\Sb superlattice
sample. ‘Tunnel current was held constant at 10 nA... 2.2.2.2...
147
148
XXll
List of Tables
1.1 Intensity of lunar light, star light, and night air glow in the different
bands [10]... 2 2 2 8
3.1 Summary of etch results... 2... .00.00.00.. 22.2.0 20048., 48
5.1 Antimonide TSD structures fabricated and the observed switching
characteristics. Device size was 67 pm. . 2... ee 86
7.1 List of Al,Ga,_,As BEEM structures studied. . 2... ...000020. 119
Chapter 1 Introduction
1.1 Thesis Overview
This thesis concerns the investigation of novel devices and material characterization
techniques in the III-V semiconductor system. The bulk of the work is carried out
in the lattice matched InAs/GaSb/AISb system, where new device possibilities arise
from the unique broken gap band alignment of InAs/GaSb and the small bandgaps
of these materials. In the first part of the thesis, we demonstrate that novel devices,
such as avalanche photodiodes and tunnel switch diodes, can be fabricated from
InAs/GaSb/AISb heterostructures by molecular beam epitaxy (MBE). In the second
part of the thesis, ballistic electron emission microscopy (BEEM) is employed to
examine the local band offset in these heterostructures, which is often found to be a
crucial factor in device design.
The antimonide avalanche photodiode forms a major part of the device study.
This is an interesting device due to its near infrared responsivity, ionization ratio
enhancement possibilities and integration potential. The design, growth, fabrication,
and characterization procedures are covered in detail in Chapters 2, 3, and 4, which
also serve to illustrate the background work common to subsequent heterostructure
studies. Two types of avalanche photodiodes will be described: those with a bulk
Alo.osGao.o65b multiplication region and those with a GaSb/AISb superlattice mul-
tiplication region. Both types of devices were implemented in a MBE grown p-nt
structure that used a selectively doped InAs/AISb superlattice as the n-type layer.
This particular structure was refined through several design, fabrication, character-
ization cycles. It was found that the photodiode dark current depended critically
on the InAs/AISb superlattice period and the resulting band offset at the p~n* het-
erojunction. The InAs/AISb superlattice was henceforth optimized by using a three
stage design.
The ionization rates in bulk Alo.o,4Gao965b multiplication layer devices were mea-
sured by using a two wavelength photo response scheme. The results were consistent
with hole impact ionization enhancement in Alo.o4Gap.9g5b. However, the ionization
enhancement advantage was largely compromised by the high level of dark current.
found in the bulk device. In comparison, the superlattice multiplication devices was
deemed more promising because the dark current was more effectively suppressed by
barriers in the superlattice multiplication region while impact ionization may still be
enhanced by separate band offset adjustment.
The tunnel switch diode is the second antimonide device studied. Compared to
the avalanche photodiode, this is a much simpler device from a fabrication stand-
point. Yet it’s no less interesting due to its unique “S” shaped current-voltage (I-V)
characteristics. In Chapter 5, we demonstrate first time measurement of such I-V
behavior from an antimonide heterostructure. The epilayer and barrier dependence
of tunnel diode switching were studied and found to deviate significantly from the
punch-through model of operation. In addition, the devices exhibited “hopping”
between two current voltage branches when subjected to high levels of stress. We
speculate that this is due to instability associated with mobile charges in the AlSb
tunnel barrier. A computer model was used to simulate the device behavior and
generated results consistent with experimental findings about the barrier thickness
dependence of switching.
The rest of the thesis (Chapters 6, 7 and 8) concerns the characterization of III-V
heterostructures by BEEM, which is a scanning tunneling microscopy (STM) based
technique capable of imaging the buried surface and yielding local band structure
information. The Al,Ga,_,As system was studied first as a testing ground for the
technique (Chapter 7). BEEM imaging was shown to reveal sub surface features in
Al,,Ga,_,As samples, while BEEM spectroscopy was used to study the local variation
of [, X, and L band edges in Al,Ga,_,As. The shift in these band edges with material
composition was also mapped out by using BEEM spectroscopy.
Following the Al,Ga,_,As study, BEEM was applied to device relevant antimonide
heterostructures such as AlSb barriers and InAs/AISb superlattices (Chapter 8). The
Al on AlSb Schottky barrier height was measured and found to be dictated by the
conduction band minimum near the X point. The barrier height exhibited a large
local variation, in correlation with the observed barrier characteristics of AlSb. Due
to the smaller bandgap of InAs/AISb superlattices and the associated high level of
noise current, only the shortest period superlattice was examined by BEEM. The
resulting band offset agreed with the calculated value and demonstrated that BEEM
spectroscopy can be applied to structures with a large number of hetero-interfaces.
1.2. Motivation
1.2.1 InAs/GaSb/AISb system
The motivation for this thesis study grew out of the versatility of III-V semiconductor
heterojunction systems. Compared to the Si system, which has been extensively
developed and enjoys a much more mature technology base, the III-V system derives
its advantages from the compound nature of its constituent material. The unique
properties of these materials, i.e., direct bandgap, high carrier mobility, can be tailored
to specific applications through bandgap engineering techniques.
To date, most of the III-V research efforts have focused on the lattice matched
Al,Ga,_,As system, which has yielded commercial injection lasers [1] and high speed
transistors [2]. As part of this thesis study, the band structure of the Al,Ga,_,As sys-
tem was characterized on a local scale by using ballistic electron emission microscopy
(BEEM). The major part of the thesis, however, is devoted to the less well known
system of InAs, GaSb and AISb, which share a common lattice constant at around
6.1 A.
The versatility of this system can be seen by examining Fig. 1.1(a), which shows
the band alignment of various component materials. The most striking feature in
this diagram is the presence of broken gap type II band alignment in addition to the
type I band alignment found in the Al,,Ga,_,As system. The unique heterojunction
band offset between InAs and GaSb has resulted in a resonant inter band tunneling
(a) |
0.58 eV
Px |,
| AlSb
0.72 eV GaSb 1.62 eV
0.20 eV
r 2 0.40 eV
0.36 eV INAS :
Figure 1.1: InAs/GaSb/AISb lattice matched system. (a) Band alignment. (b) Neg-
ative Schottky barrier between InAs and metal.
i)
(RIT) diode with oscillations up to 712 GHz [3]. In addition to the wide variety of
band offsets available, these materials have a number of advantages: InAs has the
unusual property of forming a negative Schottky barrier when brought in contact with
most metals (Fig. 1.1(b)) [4]. Hence it makes an ideal contacting material. It also
has a small effective mass (0.025mMetectron) [5], which implies that a stronger quantum
confinement effect can be achieved in RIT type applications. Moreover, AlSb has a
large bandgap (2.2 eV at the I point and 1.6 eV at the conduction minimum near
the X point [5]) and can be used as a barrier in many device structures.
Besides the aforementioned quantum effect devices, the single biggest application
for the antimonide material system is in the area of infrared sources and detectors.
This is in large part due to the small bandgaps of InAs, GaSb, their alloys and super-
lattices. The most significant developments in this area include InAs/GaSb quantum
cascade lasers, which achieve high quantum efficiency because non-radiative recom-
bination due to phonon scattering is suppressed from the type II band alignment [6].
There are also reports of injection lasers [7] in the mid infrared wavelength range and
far infrared detectors based on InAs/InGaSb superlattices [8].
While the antimonide material system have distinct advantages in many of these
applications, it is also plagued by the immaturity of the technology and a relative lack
of understanding on certain materials issues. For example, carrier transport in AlSb
is not well understood. While AlSb is an adequate barrier in RIT type structures, it
is too leaky as the gate insulator in a three terminal transistor configuration [9]. The
challenges in antimonide research thus lie in two directions:
1. Study of new devices that exploit the unique band offset and intrinsic material
property of the system.
2. Characterization of the basic material properties in a setting relevant to ad-
vancement of device research.
The thesis study therefore is a reflection of attempts to address both of these issues.
The main body of the thesis is accordingly split into two part. The first part of the
thesis describes the design, fabrication, and characterization of two new devices in the
antimonide system: an avalanche photodiode with response in the near infrared for
night vision, and a tunnel switch diode based on AlSb barriers and GaSb pn junctions.
Both devices rely on molecular beam epitaxy (MBE) as the fabrication method and
employ heterostructures for advantages. It will be seen that basic material issues such
as heterojunction band offset and transport mechanism through barriers are keys to
the operation of these devices. These issues are examined in great detail in the second
part of the thesis, where ballistic electron emission microscopy is employed to study
local transport in device-like heterostructures.
1.2.2 Antimonide Avalanche Photodiode
There is much interest in making a near infrared avalanche photodiode in the anti-
monide system. This is in part due to the narrow bandgap of GaSb, which at 0.72
eV corresponds to a long wavelength cut off of 1.7 wm. By incorporating indium in
the GaSb light absorption region, the sensitivity range of the antimonide photodiode
can be easily extended to beyond 2.0 zm. Aside from obvious communication appli-
cations at the 1.55 zm wavelength, such extended response in the near infrared fills
the special niche of night vision.
This can be seen by examining the night spectrum. In a simplified view, there are
three natural contributions to the night glow aside from the highly variable human
light source: lunar light, star light, and air glow due to transitions of atmospheric
ions [10]. The lunar light spectrum is shown in Fig. 1.2. Since it is in fact reflected
sunlight, the spectrum resembles that of a 3000 K black body. The star light can also
be thought of as agglomerated black body spectrums with temperatures that range
from 2000 K to 23000 K. Due to absorption of the atmosphere, the available radiation
can be divided roughly into several bands [10]. The absolute intensity of these bands
and the relative contributions from the different sources are highly variable depending
on the seasonal and weather ambient. An estimate of the average numbers is shown
in Table 1.1.
0.0052 mr T ory
0.0048
0.0044 ; q
Lunar Spectral Irradiance Outside Atmosphere
0.0040
0.0036 4
E,W m-2 pm)
oo
8B
fe
=.
7,
0.0000! L fata t pif lh ft tel L :
03 05 07 09 1 18°45 17 19 21 23 25 27 29 34 393
Wavelength (um)
Figure 1.2: Spectrum of lunar light.
It can be seen that there are two bands centered around the wavelength of 1.6 wm
and 2.1 wm, which would be captured by an antimonide near infrared photodiode [10].
Further examination of Table 1.1 reveals that the signal in these bands are rather faint.
For a detection pixel with a size of 30 wm by 30 ym, an integration time of 30 ys, and
a F number of 1, the incident optical power on the pixel front end is on the order of
10-'* W. The miniscule amount of power necessitates use of photodetectors of high
sensitivity, such as a photo multiplier or its solid state equivalent - the avalanche
photodiode.
The antimonide is a good candidate for such a device due to a number of system
and materials advantages. Compared to conventional InP based avalanche devices,
which fall in the same wavelength range and have been extensively developed for
telecommunication applications, the antimonide avalanche photodiode has a distinct
advantage in terms of integration potential: by building the antimonide avalanche
photodiode on the same chip as an InAs/InGaSb far infrared photodetector [8], it
will be possible to create a compact, robust system with multi-color response ideally
suited for military type night vision applications.
On the materials level, GaSb has been shown to have larger ionization coefficients
Table 1.1: Intensity of lunar light, star light, and night air glow in the different
bands [10].
Band Peak Ax Moon Star Night
Airglow
[pera] [Photons/cm?-sec]
1 0.58 0.49 5.83x10" 1.21x10° 4.12107
2 1.00 0.16 1.83x 10"? 1.24 108
3 1.24 0.21 2.09101" 1.15x108
4 1.59 0.29 1.9910" 8.67x 10" 4.48x 10°
5 2.10 0.43 1.37x 10" 6.02107 2.24x 101°
than InP [11], which will allow the device to have a thinner multiplication region for
faster response. More importantly, there has been some experimental evidence and
theoretical conjecture that hole ionization in AlGaSb is enhanced when the spin-orbit
split-off band difference A equals the energy bandgap E, [12]. This is complemented
by the possibility that electron ionization may be enhanced in GaSb/AlGaSb super-
lattices due to band offset differences [13]. As will be discussed in detail in Chapter
2, the enhanced ionization rate is the single most important property for avalanche
operation. It leads to better gain-bandwidth product, smaller excess noise factors,
and reduction of microplasmas which are detrimental to the stability of the avalanche
action [14].
To date, there have been a number of studies on antimonide avalanche pho-
todiodes, all of which have relied on liquid phase epitaxy (LPE) as the crystal
growth method and focused on hole impact ionization enhancement in bulk Al-
GaSb [12, 16, 15]. The results of these studies indicate a lack of consensus about
the resonant hole ionization effect. Hence there is much incentive for a new study
with molecular beam epitaxy (MBE) as the device fabrication method. The flexibil-
ity of the MBE technique will allow exploration of antimonide avalanche photodiodes
with both bulk and superlattice gain mediums. This should further clarify the ioniza-
tion enhancement issue and maximize the potential of the device through comparison
of the bulk and superlattice approaches.
1.2.3. Antimonide Tunnel Switch Diode
The motivation for the antimonide tunnel switch diode grew out of the need for a two
terminal device that has an “S” shaped I-V curve. The usefulness of antimonide elec-
tronics depend on high speed operation and possible reduction of circuit complexity
from non-linear elements. Since there already exists an antimonide RIT diode with
“N” shaped I-V characteristics [17], a new “S” shaped I-V device would complement
the RIT device as its circuit dual and enable a wider variety of circuit applications.
Quiescent Sensing
R wtp
Vout , High Current
> 109 ' State
1 x
@)
SZ TSD 5 ios
107 '
= 10°
0 1 2 3 4 5
Voltage (V)
(a) ()
Figure 1.3: Transistorless static random access memory from a tunnel switch diode.
(a) Circuit schematic. (b) Load line analysis.
Of particular interest is a transistorless static random access memory (SRAM)
element that can be fabricated from the tunnel switch diode alone [18]. Fig. 1.3
shows the basic implementation of this scheme. The memory effect derives from the
existence of a high and a low current branch in the I-V characteristics. The logic
of the state is read out from the current disparity as the Load line shifts to high
voltages. Writing is accomplished by a large voltage swing that switches the TSD to
a particular state. Such a memory element would be very fast due to the tunneling
nature of the switching. Compared to conventional implementations which requires
six transistors per memory cell, the tunnel switch diode element is extremely compact
due to its simplicity.
10
The antimonide tunnel switch diode can also act as the testing ground for tunnel
switch diode theory and provide an opportunity to study AlSb barriers in a device
setting. While the basic principle of tunnel switch diode is known, its actual operation
is highly dependent on the role of the tunnel barrier, which is not well understood.
This has been a long standing problem in the silicon implementation of the device [19].
By replacing the SiOz tunnel barrier with AlSb, which has a different barrier height
and may have deep traps within its bandgap [20], it is hoped that more experimental
evidence will be collected to shed light on the role of the tunneling barrier in TSD
operation.
1.2.4 Ballistic Electron Emission Microscopy
From the device section of this thesis, it will be seen that basic heterojunction prop-
erties such as band offset and Schottky barrier height play prominent roles in device
design and often dictate whether or not the device is feasible. While these proper-
ties can be ascertained from conventional characterization techniques such as X-ray
photoelectron spectroscopy (XPS), photo electric, I-V, and C-V measurements, the
results are usually laterally averaged over the whole interface or at least over macro-
scopic dimensions of more than a few ym. In contrast, ballistic electron emission
microscopy (BEEM) provides local mapping of these properties with a theoretical
resolution limit as low as 10 A [21, 22].
The BEEM technique is illustrated in Fig. 1.4. The high resolution of BEEM
derives from its scanning tunneling microscopy (STM) origin. By placing a third
terminal at the back of the semiconductor sample to collect hot electrons that pass
through the buried interface, local electronic properties of the interface can be studied
as the STM tip moves to different parts of the surface.
Thus in the second part of this thesis, BEEM is used to investigate semiconductor
heterostructures similar to those employed in the device studies. The Al,Ga,_,As
system was examined first to clarify ambiguities with regard to its band structure,
especially at high Al concentrations. It also served as the testing ground for the
Ll
ZOICEZO esse F eed back eto
Collector
Ohmic contact
Figure 1.4: Ballistic electron emission microscopy.
BEEM technique so that it can be readily applied to less conventional structures
such as the AlSb Schottky barrier or the InAs/AISb superlattice. The band offset
and Schottky barrier information derived from the antimonide study can be applied
to device applications, whereas the local variation of these properties are also directly
useful as a feed back to the crystal growth process.
1.3 MBE crystal Growth
The underlying link between the device and characterization studies in this thesis is
the molecular beam epitaxy (MBE) growth of HI-V heterostructures. The flexibility
and control inherent to the MBE crystal growth technique is what makes all these
studies possible. Since MBE is a well known technique and the detailed growth
sequence for each particular structure will be covered in subsequent sections of the
thesis, only a few general principles will be outlined below as an introduction.
12
1.3.1 MBE Environment
Figure 1.5 shows a schematic representation of a MBE machine. The substrate is
placed in a fully enclosed steel chamber and surrounded by a number of evaporation
crucibles known as Knudsen cells. During growth, the Knudsen cells are heated to
specific temperatures so that its elemental content evaporates off at a controlled rate.
These elements recombine at the heated, clean substrate surface and fall into the
existing crystal template.
Chamber
Wall
Substate Heater RHEED
Wafer | Gun
Knudsen
Effusion Cells
Figure 1.5: Schematic of molecular bean epitaxy set-up.
The distinguishing feature of MBE growth is the stringent requirement it places
on the ambient pressure. The background pressure in a well maintained MBE cham-
13
ber is typically below 107° Torr. At such a low pressure, it takes several hours for
a monolayer of impurities to accumulate on a clean substrate surface [23]. Since the
growth rate is on the order of one monolayer per second, MBE allows exquisite con-
trol of epilayer thickness. By shuttering the source cells on and off in the correct
sequence, atomically sharp interfaces can be achieved. The low background pressure
also implies that the mean free path of materials in the growth beam is much larger
than the separation between source cells and the substrate. As a result, liquid or
gaseous flow patterns do not complicate MBE growth as can be the case in chemical
vapor deposition or liquid phase epitaxy. Because of the UHV environment, a number
a diagnostic tools can be applied for in situ monitoring of the crystal growth process.
The most important and commonly available of these are reflection high energy elec-
tron diffraction (RHEED) and the residual gas analyzer. As shown in Fig. 1.5, a high
energy electron beam (10 keV) is directed at the substrate at a grazing angle in the
RHEED set-up. The resulting diffraction pattern provides information on the growth
rate, surface reconstruction and morphology of the substrate. This is complemented
by the residual gas analyzer which yields information about the chemical species in
the growth chamber and can be used to adjust. beam fluxes and identify background
impurities.
1.3.2 III-V Growth
The MBE growth of semiconductor crystal is optimized by adjusting two main param-
eters: substrate temperature and source flux. Since III-V compounds preferentially
desorb group V elements, the group V flux must be maintained between three to
ten times higher than the group III] flux to grow a stoichiometric crystal. This ratio
increases as the substrate temperature is raised, provided that the temperature does
not exceed the “congruent sublimation temperature,” at which point the IIJ]-V com-
pound becomes unstable [23]. Since the excess V element desorbs, the growth rate is
controlled by the III flux and the associated cell temperature. It is generally known
that the crystal quality can be improved by using a higher substrate temperature and
14
minimizing the V flux while still maintaining a V element stabilized growth front [23].
The structures studied in this thesis consisted of GaAs/AlGaAs and
InAs/GaSb/AISb. All of these structures were grown in a Perkin-Elmer 430
MBE chamber equipped with cracked As and Sb cells. The AlGaAs samples were
grown on epi-ready (100) GaAs substrate at a substrate temperature of 570 °C,
which is slightly below the oxide desorption temperature of GaAs. An As stabilized
front was always maintained by keeping a 4 x 2 RHEED pattern.
Most of antimonide heterostructures were grown on (100) GaSb substrate, which
were etched prior to introduction to the UHV environment. The growth temperature
for the bulk materials was slightly below the GaSb oxide desorption point at 520
°C, whereas a lower temperature was required for InAs/AISb superlattice growth in
order to reduce cross anion contamination. The RHEED reconstruction patterns for
GaSb/AISb and InAs are 1 x 3 and 4 x 2, respectively.
AlSb AlSb
4 =
GaSb
x 2 InAs InAs
= L_] [
~ O
Cc
@ .
, a
S .2
AL
i i i 1 1 | 1 L
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
Applied Bias [V]
Figure 1.6: Current voltage behavior of antimonide RIT grown on GaAs wafer.
The crystal growth is divided into the arsenide and antimonide groups due to the
8% lattice match between GaAs and GaSb. It is possible, however, to overcome the
15
lattice match and the resulting misfit dislocations by using a buffer layer [24]. This is
demonstrated in Fig. 1.6 which shows the L-V curve of an antimonide RIT diode grown
on GaAs substrate. One can also take advantage of the lattice mismatch between these
two material systems. As illustrated by the atomic force microscope (AFM) scan in
Fig. 1.7, GaSb dots are formed as a result of the surface free energy effect when several
monolayers of GaSb are grown on GaAs substrate. Though lacking uniformity, these
dots may be a viable way to realize zero dimensional quantum structures. These
examples clearly demonstrate the versatility and flexibility of the ITI-V MBE process.
| NanoScope AFM
Scan size : 1.000 pm
Setpoint Z =1.000 VU .
Scan rate oo 3.815 Hz.
Number of samples 512
X 0.200 um/7diy |.
4 100.000 nm/div
09191447 .001
Figure 1.7: AFM scan of dots formed from two monolayers of GaSb deposited on
GaAs substrate. Note that the large bright object on the left is a dust particle.
16
Bibliography
[1] A. Y. Cho and H. C. Casey, Jr., Appl. Phys. Lett. 25, 288 (1974).
[2] M. Abe, T. Mimura, N. Yokoyama, and H. Ishikawa, IEEE Trans. Electron
Devices 29, 1088 (1982).
[3] E. R. Brown, J. R. Séderstrom, C. D. Parker, L. J. Mahoney, K. M. Molvar, and
T. C. McGill, Appl. Phys. Lett. 58, 2291 (1991).
[4] A. G. Milnes and A. Y. Polyakov, Mater. Sci. Engineering 18, 237 (1993).
[5] Semiconductors: Group IV elements and III-V Compounds, edited by O.
Madelung, Spriner-Verlag, Berlin, 1991.
[6] R. Q. Yang and S. S. Pei, J. Appl. Phys. 79, 8197 (1996).
[7] R. H. Miles, D. H. Chow, T. C. Hasenberg, A. R. Kost, and Y-H. Zhang, Appl.
Phys. Lett. 67, 3700 (1995).
[8] D. L. Smith and C. Mailhiot, J. Appl. Phys. 62, 2545 (1987).
(9] E. S. Daniel, Thesis, California Institute of Technology, 1997.
[10] G. S. Picus, unpublished, California Institute of Technology, 1997.
(11] F. Osaka, T. Mikawa, and T. Kaneda, IEEE J. Quantum Electron. 21, 1326
(1985).
[12] O. Hildebrand, W. Kuebart, K. W. Benz, and M. H. Pilkuhn, IEEE J. Quantum
Electron. 17, 284 (1981).
[13] G. F. Williams, F. Capasso, and W. T. Tsang, IEEE Trans. Electron Devices 3,
71 (1982).
17
[14] G. E. Stillman and C. M. Wolfe, in Semiconductors and Semimetals, Volume 12,
edited by R. K. Willard and A. C. Beer, Academic, New York, 1977.
[15] L. Gouskov, B. Orsal, M. Perotin, M. Karin, A. Sabir, P. Coudray, S. Kibeya,
and H. Luquet, Appl. Phys. Lett. 60, 3030 (1992).
[16] H. Kuwatsuka, T. Mikawa, S. Miura, N. Yasuoka, Y. Kito, T. Tanahashi, and O.
Wada, Appl. Phys. Lett. 57, 249 (1990).
[17] J. R. Sdderstrom, D. H. Chow, and T. C. McGill, Appl. Phys. Lett. 55, 1094
(1989).
[18] H. J. Levy, Thesis, California Institute of Technology, 1995.
[19] P. O. Pettersson, A. Zur, E. S. Daniel, H. J. Levy, O. J. Marsh, and T. C. McGill,
IEEE Trans. Electron Devices 45, 286 (1998).
[20] A. Nakagawa, J. J. Pekarik, H. Kroemer, and J. H. English, Appl. Phys. Lett.
57, 1551 (1990).
[21] W. J. Kaiser and L. D. Bell, Phys. Rev. Lett. 10, 1406 (1988).
[22] L. D. Bell and W. J. Kaiser, Phys. Rev. Lett. 61, 2368 (1988).
[23] The Technology and Physics of Molecular Beam Epitaxy, edited by E. H. C.
Parker, Plenum Press, New York and London, 1985.
[24] J. R. Soderstrom, D. H. Chow, and T. C. McGill, Mat. Res. Soc. Symp. Proc.
145, 409 (1989).
18
19
Part I
Novel Devices Based on
InAs/GaSb/AISb Heterostructures
20
21
Chapter 2 Avalanche Photodiode Theory
2.1 Introduction to Chapter
The first device investigated in this thesis is the antimonide avalanche photodiode.
This chapter describes the basic theory of avalanche photodiode operation and pro-
vides the necessary background for the following two chapters. Much of the chapter
will be devoted to the importance of impact ionization enhancement, which is the
critical factor in avalanche photodiode design and one of the initial driving forces
for antimonide avalanche photodiode research. The equally important, and perhaps
more practical, subject of dark current suppression is also discussed.
2.2 Basic Operation
The avalanche photodiode consists of a reversed biased pn junction. As shown in
Fig. 2.1, incoming photons are absorbed by the semiconductor and generate electron
hole pairs. Under high field conditions, these electrons and holes may gain large
amount of energy from drift motions in the depletion region. In the so called “impact
ionization process,” the hot carriers knock off bound electrons upon collision with
atoms in the crystal. The end result is that each hot carrier gives up its energy
towards the creation of a new electron hole pair. The process repeats itself and the
final multiplied current can be several hundred times larger than the initial incoming
signal [1].
The signal to noise (S/N) ratio of an avalanche photodiode is given by:
5(qnPo/hv)? M?
a (2.1)
N 2q(IpFp + IpFp)M2B + 4kTB/R
where q is the electronic charge, 7 is the quantum efficiency, Pp is the incoming signal
22
p side Electron
energy
Egy --- nap ea ee ;
F2 Position
| Kone)
Ve y
ee ee J se edeaenaicameatianametineanienmedtimmmetiommntticescaimmedticeatinaneetinende Sn -cterelieriare Go, a a ee aaa nee Epy
©)a free electron
G) a free hole
Figure 2.1: Band diagram and basic operation of an avalanche photodiode [1].
power, hv is the photo energy, M is the multiplication factor, Ip and Ip are the signal
current and dark current, Fp and Fp are the associated excess noise factors, B is the
detector band width, kT’ is the thermal energy, and R is the detector impedance [1].
The larger multiplied signal implies a better S/N ratio and greater detector sensitivity:
as the multiplication factor is increased, the Johnson noise term 4kTB/R plays a
diminished role and the S/N ratio approaches the shot noise limit. However, this is
not without penalties. The statistical nature of the avalanche process means that
additional noise is introduced as the current is multiplied. This is accounted for
by the excess noise factor F(M), which becomes increasingly significant at higher
multiplication factors. Thus the avalanche photodiode researcher is concerned about
finding ways to reduce the excess noise factor F'(/), especially in night vision type
applications where high detector sensitivity is desired.
23
2.3 Impact Ionization Ratio
The side effect due to the statistical nature of the avalanche process is minimized
when only one type of carrier dominate the impact ionization process [2]. Hence
the most important parameter for the multiplication material is the ratio of electron
and hole impact ionization rate (a/@). A large impact ionization ratio will result
in more gradual increase of F'(/) with multiplication factor M, a higher gain band
width product, and reduction of microplama formation, which is detrimental to the
stability of avalanche photodiode operation.
fe) Position re) Position
Electron __,J Electron __.J Impact . }
Injection aK ionization {
3 9: = i
’ Impoct
lonization
Injection
ee ee eo
(a) (b)
Figure 2.2: Schematic representation of avalanche multiplication process. (a) a=(.
(b) G=0.
Intuitively, this can be understood by examining Fig. 2.2, which depicts electrons
and holes drifting in opposite directions in the high field region of an avalanche
diode. In Fig. 2.2(a), the ionization rates of electrons and holes are equal, whereas in
Fig. 2.2(b) only the electrons can impact ionize. Note that the total number of output
carriers is the same in both pictures because the gains are the same. To compensate
for the lack of hole ionization, the electric field in Fig. 2.2(b) is much higher, which
results in higher electron ionization rates and a shorter average ionization path. The
process in Fig. 2.2(a) is inherently more noisy and slower because the ionization path
has long segments and zig-zags across the depletion region many more times. In
24
comparison, the ionizations in Fig. 2.2(b) only goes toward one direction and there
is much less statistical variation due to the shorter average ionization path.
The mathematics of the avalanche process and resulting dependence of excess
noise factor on ionization ratio was first worked out by McIntyre in 1965 [8], whereas
the gain bandwidth product dependence was shown to depend critically on impact
ionization ratio by Emmons and Lucovsky [4]. The main results of their findings are
summarized as follows.
The multiplication factor M is given by
1 = fe aexp[— f(a — B)da'|dx
Mn, (2.2)
where w is the width of the depletion region. Note that the electron and hole impact
ionization rates a and @ are functions of electric field and electron injection from the
p side edge (« = 0) of the junction is assumed. For hole injection from the n side of
the reverse biased pn junction, simply exchange the roles of a and @ and change the
inner integration limits from [0,2] to [x, w].
Ww
a,W = f Gp dx
01 02 04060810 15 20 30 40 50 60
1000——F FT -TTT TT TI || | rn |r
E B,=1|05| 02| olloostoo2} foo: o
5 &E
mt c
o L
5 €
= We
i l l l
20 25 30 35 40 45
ELECTRIC FIELD (V/cm)x10+5
Figure 2.3: Field dependence of multiplication factor for different ionization ratios [5].
25
The electron multiplication factor is plotted as a function of electric field for var-
ious ionization ratios in Fig. 2.3. It can be seen that the multiplication is much more
sensitive to the electric field when the ionization rates are nearly equal. Despite high
gains at lower field, this is an undesirable scenario because the device is very unstable
with respect to field fluctuations which may result from crystal imperfections [5]. In
fact, local breakdowns known as “microplasmas” of electrons and holes do form and
are detrimental to the noise characteristics of the device [6].
The gain bandwidth product of the avalanche photodiode is given by
Me = Tey G/a) (2.3)
where w is the cut off frequency, W the depletion width and v, the saturation velocity
of the electron [4]. There is a trade off between gain and bandwidth because the higher
gains require more ionization segments (see Fig. 2.2(a)). The resulting extra passage
through the depletion region slows down the avalanche process. For the same gain, the
device with the higher ionization rate ratio is faster because of the shorter ionization
path.
The excess noise factors are given by
Fy = M,((1 — (1 — 1/k)[(Mp — 1)/Mp]’)) (2.5)
where k = 8/a, and M, and M, denote electron and hole multiplication factors,
respectively [3]. These expressions are plotted in Fig. 2.4 for various values of a/@.
Because of the symmetry of these equations in n, p, k, and 1/k, only one set of curve
is required. In this figure, it can be seen that for a low excess-noise factor the electron
and hole ionization rates must be greatly different. In addition, the device structure
must be designed so that the carrier with the highest ionization rate is injected into
the high field region. If the reverse is true, the excess-noise factor will actually be
26
worse than that for a device with equal ionization rates.
Ub |
© 100F
LS F
5 ;
Ow
LL. k=1
@®
YX
S job k=0.1
nn E
Y)
<>)
x<
Lu
1 1 i Po oe oer ee ee | pil
1 10 100
Multiplication Factor M
Figure 2.4: Effect of impact ionization ratio on excess noise factor [3].
2.4 Impact Ionization Enhancement
2.4.1 Hole Impact Ionization Enhancement from Spin-orbit
Split-off Band Resonance
Much of the interest in antimonide avalanche photodiode is due to the possibility
of hole impact ionization enhancement from the spin-orbit split-off band resonance.
Such an effect would result in an improved hole to electron impact ionization ratio
and better avalanche noise characteristics.
In order to understand this effect, one must take a closer look at the impact
ionization process. Figure 2.5 shows an electron initiated impact ionization transition
in a band structure exhibiting general features of a zinc blende semiconductor. In
the event depicted, an initiating electron makes a transition from state i(E;,k;) to
27
ENERGY
WAVE VECTOR
Figure 2.5: Band structure view of an electron initiated impact ionization process [8].
state 1(£,,k,) promoting an electron to state 2(£, ke) and resulting in a hole in the
valence band state 3(£3, kg). A central requirement for this process is conservation of
energy. A brief reference to Fig. 2.5 shows that this means that the threshold energy
must be at least as large as the bandgap, and one should expect the threshold energy
to depend strongly on E,. Secondly, conservation of momentum must be maintained.
Thus the necessary conditions for impact ionization are
E(ki) = E(ky) + Elke) ~ Elks) (2.6)
The threshold energy for the process is determined by minimizing the energy of the
initial particle with respect to arbitrary variations in the states of the final particles,
28
subject to the restrictions of conservation of energy and momentum [7]:
dk; - Vy E(k) = 0 = dky - Vy E(k) + dky - VE (ke) — dks - VE (ks)
(2.8)
dk; = 0 = dk, + dky — dks (2.9)
Substituting (2.9) into (2.8) and recognizing V, E(k) as the group velocity v, we get
For the above relation to be satisfied for arbitrary dk; and dk» implies:
Vi =V2=V3 (2.11)
That is, the group velocities, or slopes of the £ —k diagram, must be equal for all
final states.
For a parabolic conduction band with effective mass m, and a valence band with
effective mass mpp, the conditions above yield a threshold energy [8]
Me
Ey, = E,(1 + ——— for electrons 2.12
th (1 + Tmo Tn or electrons (2.12)
En, = E,(i+ moh Tm for holes (2.13)
The hole initiated process is further complicated by the presence of the spin-
orbit split-off band. The three band hole impact ionization process is depicted in
Fig. 2.6. The spin-orbit split-off hole initiated process dominated over the light hole
or heavy hole initiated process because its threshold energy is much lower. Following
derivations similar to the electron process, the threshold energy can be shown to be
29
Light Hole,
Heavy Hole
Band f
/ f 2
fea
Split-off
Hole Band
Figure 2.6: Hole initiated impact ionization process involving the spin-orbit split-off
band [9].
given by [8]
Mso(1 — A/E,)
2Mhh + Me — Ms
Ee, = Ej(lt+ ) for A < E, (2.14)
En=A for A > E, (2.15)
where A is the spin-orbit split-off band offset and m,, is the effective mass of the split
off band.
These equations make it clear that the spin-orbit splitting can have a pronounced
effect on the hole ionization threshold energy by reducing it, while leaving the electron
ionization threshold energy unchanged. Since the ionization rate depend exponen-
tially on the threshold energy [2], a strong enhancement of hole/electron ionization
ratio is expected in a material where the spin-orbit split-off band difference A matches
the bandgap energy E,.
30
In the Al,Ga,_,Sb system, A matches E, at x = 0.04 ~ 0.065, where Ey = 0.75
eV. This effect was first proposed for avalanche multiplication with limited amount
of experimental evidence by Hildebrand et al. [9]. However, other experimental and
theoretical studies have since generated contradictory results. In particular, Hidel-
brand’s findings were supported by Gouskov et al. [10], whereas Kuwatsuka et al. [11]
showed that 8/a in Alp 9gGag.94Sb was lower than previously measured and there was
no enhancement of 3/a at the split-off band resonant condition. Alternative theoret-
ical explanations for the observed effect. based on composition disorder has also been
proposed [12]. Thus hole ionization enhancement from spin-orbit split-off band reso-
nance in AlGasSb is still an unsolved problem with large technological and scientific
consequences.
It should be pointed out that previous studies on the subject have all used liq-
uid phase epitaxy (LPE) as the crystal growth method. Since the effect is possibly
material dependent [12], there is clear incentive to study similar device structures
fabricated from molecular beam epitaxy.
2.4.2 Electron Impact Ionization Enhancement from Super-
lattice Band Offset
An alternative way to enhance the impact ionization ratio is to use superlattice struc-
tures with large differences in conduction and valence band offset [13]. This effect is
illustrated by Fig. 2.7.
Because of the very low doping, the electric field can be regarded as constant across
the barrier and well layers in the superlattice. Consider a hot electron accelerating
in the large bandgap barrier layer. Upon entering the well it abruptly gains an
energy equal to the conduction band edge discontinuity AF,. The effect is that the
electron “sees” an ionization energy reduced by AEF,. Since the impact ionization
rate q@ increases exponentially with decreasing threshold energy, a large increase in
the effective a is expected. When the electron enter the next barrier region, the
threshold energy in this material is increased by AE,, decreasing a in the barrier
w |
aa €=2.7 x10" V/em
Oo;
\e
Figure 2.7: Electron impact ionization enhancement in a GaAs/AlGaAs superlattice
multiplication layer.
layer. However, since Qe > Qbarrier, the exponential dependence on the threshold
energy ensure the average a given by
a= (QwettLwell + Qbarrier barrier) /(Lweit + Loarrier) (2.16)
is increased (L denotes layer thicknesses).
In contrast, the hole ionization rate 6 is not substantially increased because the
reduction in hole ionization energy is only the valence band discontinuity AF,, which
is made much smaller than AE,. The net result is a large enhancement of a/{.
This approach to ionization ratio enhancement has been successfully demonstrated
in AlGaAs [14] and InGaAIAs systems [15]. The scheme has not been adequately
exploited in the GaSb/Al,Ga,_,Sb system despite the large band offset differences
available. The success of the approach in the AlGaSb system will depend on the
transport mechanism in AlSb. If point transport dominates, the effective conduction
band offset between AlSb and GaSb will be 1.15 eV, which is much larger than their
32
valence band offset. at 0.45 eV. This would result in a large enhancement of electron
to hole ionization ratio in a superlattice structure. However, if electron transports
via the X point valley in AlSb, the effective conduction band offset is only 0.55 eV,
and the ionization enhancement effect would be diminished.
2.5 Practical Considerations: Dark Current
As mentioned earlier, much of the initial driving force behind the antimonide
avalanche photodiode is due to the possibility of ionization ratio enhancement, which
results in a better S/N ratio and detector sensitivity. A closer look at the noise
contributions and minimum detectivity of the device, however, reveals that the dark
current plays just as important a role. In fact, low dark current is the prerequisite
for optimal operation of avalanche photodiode.
This can be understood by examining the noise-equivalent-power (NEP) of an
avalanche photodiode, which is defined as the incident optical power required to
produce a power signal to noise ratio of one in a 1 Hz bandwidth. In essence, the
NEP expresses the signal to noise relationship such that the detector sensitivity is
better characterized. It’s a good indicator of how well the device can be adapted for
low signal level applications such as night vision. From the signal to noise relation of
an avalanche photodiode, the NEP is derived as
hv 2 I OKT
NEP = al (InpFp + 28 + ——
q M2" GRM?2 yy? (2.17)
where hy is the photon energy, 77 is the quantum efficiency, q is the electronic charge,
Ipp is the bulk dark current, Fp is the excess noise factor associated with Ipz, Ips
is the surface dark current, M is the multiplication factor, kT is the thermal energy,
and R is the detector impedance [2].
The various terms in the expression is plotted as a function of the multiplication
factor in Fig. 2.8. It can be seen that the Johnson noise contribution decreases with
avalanche gain whereas the contribution from the shot noise term increases with gain.
33
Johnson Noise
1kQ Load
a 10°?
"N
= [=0.1HA K=1
Oo.
LL
Zz [=0.1uA K=0.1
it
Noise from
Multiplied
Dark Current
1 10
Multiplication Factor M
100
Figure 2.8: Noise equivalent power of an avalanche photodiode.
The detectivity is optimized near the region where these two terms cross each other.
Since the Johnson noise comes from the amplifier circuit it is considered part of the
system constraint. From the viewpoint of the device designer, a smaller NEP and
better sensitivity must be achieved by lowering the shot noise term. This can be done
in two ways:
1. Minimize the dark current so that the shot noise contribution starts at a lower
level (curves c and d in Fig. 2.8).
2. Improve the ionization ratio so that the excess noise factor increases slower with
avalanche gain (curves b and d in Fig. 2.8).
Thus enhancement in ionization ratio alone will only have a secondary effect on
the sensitivity of the device. It is crucial that the device dark current is kept as low
as possible.
34
Bibliography
[1]
[2]
[12]
[13]
A. Yariv, Optical Electronics, Saunders College Publishing, 1991.
G. E. Stillman and C. M. Wolfe, in Semiconductors and Semimetals, Volume 12,
edited by R. K. Willard and A. C. Beer, Academic, New York, 1977.
R. J. McIntyre, IEEE Trans. Electron Devices 13, 164 (1966).
R. B. Emmons and G. Lucovsky, Proc. IEEE 52, 869 (1964).
P. P. Webb, R. J. McIntyre, and J. Conradi, RCA Rev. 35, 234 (1974).
D. J. Rose, Phys. Rev. 105, 413 (1957).
C. L. Anderson and C. R. Crowell, Phys. Rev. B 5, 2267 (1972).
T. P. Pearsal, R. E. Nahory, and J. R. Chelikowsky, Symposium on GaAs and
Related Compounds: St Louis 1976, Institute of Physics, London, 1977.
O. Hildebrand, W. Kuebart, K. W. Benz, and M. H. Pilkuhn, IEEE J. Quantum
Electron. 17, 284 (1981).
L. Gouskov, B. Orsal, M. Perotin, M. Karin, A. Sabir, P. Coudray, S. Kibeya,
and H. Luquet, Appl. Phys. Lett. 60, 3030 (1992).
H. Kuwatsuka, T. Mikawa, 5. Miura, N. Yasuoka, Y. Kito, T. Tanahashi, and O.
Wada, Appl. Phys. Lett. 57, 249 (1990).
Y. Jiang, M. C. Teich, and W. I. Wang, J. Appl. Phys. 67, 2488 (1990).
G. F. Williams, F. Capasso, and W. T. Tsang, IEEE Trans. Electron Devices 3,
71 (1982).
30
[14] F. Capasso, W. T. Tsang, A. L. Hutchinson, and G. F. Williams, Appl. Phys.
Lett. 40, 38 (1982).
[15] T. Kagawa, Y. Kawamura, H. Asai, M. Naganuma, and O. Mikami, Appl. Phys.
Lett. 66, 993 (1989).
36
Chapter 3 Design, Fabrication, and
Characterization of Avalanche
Photodiodes
3.1 Introduction to Chapter
The avalanche photodiode work can be viewed as a closed feedback loop consisting
of growth, processing, characterization and design. Due to the relative immaturity
of the antimonide system, much of the device research is devoted to overcoming
materials/fabrication issues and establishing the feedback loop. The experimental
procedures and methodologies critical to the realization of antimonide avalanche pho-
todiodes are established in this process and are described in detail in this chapter.
3.2 Device Design
Avalanche photodiodes typically have a PIN configuration with most of the electric
field dropped across the intrinsic multiplication region under reverse bias. A p7nt
configuration is adapted here because it is difficult to grow AlGaSb layers with low
background impurity levels. Due to the low vapor pressure of antimony, Ga tend
to occupy Sb vacancies in a AlGaSb crystal and form an anti-site defect which is a
double acceptor. Thus unintentionally doped AlGaSb is always p-type [1].
The basic device structure is shown in Fig. 3.1 and consists of three sections: a
heavily doped p+ (p=2x10'8/cm*) GaSb contact /absorption layer, an unintentionally
doped p~ multiplication layer, and a selectively doped nt InAs/AISb superlattice
layer.
37
Bulk Al, .,6@Q,.,90 or GaSb/AlSb
superlattice Multiplication
Layer (Unintentionally Doped,
p=5x10'%/em")
InAs/AiSb Superlattice
(InAs Layer Selectively
Doped with Si,
n=1x10"%/cem’)
GaSb Bottom
Contact Layer
(Si doped, m4 mo ASD 7
p=2x10"%/em’)
a a yi
i a aaa ca E,
' Metal
SHE TO
Growth Direction
SRE al
Figure 3.1: Band diagram of the antimonide avalanche photodiode grown by MBE.
The device had a p~n* configuration with a p~ bulk Alg.o«Gao.9gSb or GaSb/AISb
superlattice multiplication layer and a selective doped n* InAs/AI\Sb superlattice
layer. The device is shown under reverse bias.
3.2.1 Multiplication Layer
The multiplication layer consists of either bulk AlGaSb or GaSb/AISb superlattices.
For bulk AlGaSb multiplication layers, the Al concentration must be adjusted to
match the spin-orbit split-off band offset A with the bandgap E, to possibly lower
the hole ionization threshold energy [2]. Fig. 3.2 shows the variation of A and E, with
Al composition x. It can be seen that the resonant condition occurs approximately
for x between 0.04 and 0.065. The composition of 0.04 is chosen by consulting the
latest literature [3].
From Hall measurement, the background doping in the AlGaSb layer was found to
be p=5x10'*/cm3. For an one-sided, abrupt pn junction, the avalanche break down
38
1.45
1.25
Bandgap E,
fo)
Spin Orbit Split Off a
Energy [eV]
0.4 -
0.2-
0.0 rl J rl ! L i] n j
0.0 0.1 0.2 0.3 0.4
Al Concentration in Al Ga, Sb
Figure 3.2: Variation of spin-orbit split-off band difference A and bandgap E, with
Al composition in Al,Ga,_,Sb [4].
voltage is given by
Vp = 60(E,/1.1)°/*(Ng/10!%)~3/4 (3.1)
where E, is the room temperature bandgap in eV, and Nz is the background doping
incm® [5]. Given the bandgap of Alo.o4Gag96Sb at 0.75 eV and the measured doping
level, the bulk device is estimated to have an avalanche breakdown voltage of 14 V
and a breakdown depletion width of 0.6 um on the lightly doped side. Thus the
Alo.osGao.965b layer was kept to be at least 0.6 4am thick to maximize the length of
the multiplication region. Multiplication layers much thicker than the depletion width
will degrade device performance because photo generated carriers in the underlying
p-layer will have to traverse a longer path before being swept into the multiplication
region.
39
The superlattice gain layer consists of ten periods of alternating GaSb and AlSb
layers. The large conduction band offset (1.15 eV to the [ point of AlSb, 0.55 eV to
the conduction minimum near the X point of AlSb) and comparatively smaller valence
band offset (0.45 eV) between these materials indicate potential for electron ioniza-
tion enhancement. For comparison purposes, the overall thickness of the GaSb/AISb
superlattice gain layer was kept the same as its bulk counterpart at 0.6 um. This re-
sults in a GaSb or AlSb single layer thickness of 300 A which enables ionizing carriers
to gain enough energy at high field conditions (E>10°/cm) to get out of the well.
3.2.2. InAs/AISb n-type Superlattice
The InAs/AISb superlattice n layer is the distinguishing feature of the MBE grown
device structure. This approach to n-layer fabrication was first implemented in an-
timonide mid-infrared lasers [6] and has a number of advantages over conventional
ternary or quaternary material. By incorporating Si only in the InAs layer, heavy
n-type doping can be achieved without using tellurium, which is highly toxic and a
known contaminant to III-V growth. Since the lattice constant of GaSb is between
that of InAs and AlSb, the superlattice can be precisely lattice matched to the GaSb
substrate by adjusting the InAs and AlSb constituent layer thickness. Variation of
the InAs and AlSb layer thicknesses also results in separate tuning of the superlattice
conduction and valence band edges, which can be used to optimize the band offset
within the device structure. Due to the deeper electron quantum well, the conduction
band edge can be tuned over a much wider range, hence the superlattice bandgap is
largely dependent on the InAs layer thickness. As illustrated in Fig. 3.3, decreasing
the InAs layer thickness from 27 A to 5 A results in a 0.4 eV shift in the conduction
band edge and an increase of the superlattice bandgap from 0.8 to 1.2 eV [7].
To first order, design constraints for the InAs/AISb superlattice are due to carrier
transport and light absorption considerations. The superlattice is doped at a high
level (n=1x10'8/cm*) to form the desired p~n* junction and minimize contact resis-
tance at the surface. When the junction electric field reaches avalanche breakdown
40
2.0
1.54 InAS/AISB 4° E
SLS \"
S ~-e.
= 40b Superlattice
S Bandgap E,
Shee
ay ~
0.5 Band Overlap a7 ~ — _
between Superlattice ~ Ts
and GaSb .
0.0 ] ‘ i 1 l L L i i 1 ]
5 10 15 20 25 30
InAs Layer Thickness [A]
Figure 3.3: Calculated InAs/AISb superlattice bandgap energy and band overlap with
GaSb as a function of the superlattice period thickness [7]. The InAs and AlSb layer
thicknesses were assumed to be equal.
levels (2x10° V/cm ), the depletion width on the n-side is on the order of several
hundred A, which is small compared to the overall thickness of the superlattice layer
at, several thousand A. Hence bias induced variation in depletion width has only a
slight effect on the quantum efficiency of the light absorption process. The superlat-
tice bandgap must also be kept larger than the underlying multiplication layer and
the GaSb substrate so that it is transparent to long wavelength photons designed
to be absorbed in the p-layer. This will result in electron injection as required in
subsequent two wavelength photo characterization experiments (Section 3.5). Given
the bandgap of GaSb at 0.72 eV, figure 3.3 indicates that the InAs layer thickness
must be kept below 30 A.
41
3.3 Growth
3.3.1 Buffer and Multiplication Layers
The avalanche photodiode structures were grown on (100) GaSb wafers, which were
etched [8] prior to indium bonding with a solid growth block. Radiatively heated
growth blocks were also tried but temperature regulation was a problem due to the
small thermal mass of these blocks.
Following oxide desorption under Sb over pressure, a 1 zm thick GaSb buffer
layer was deposited at a substrate temperature of 520 °C. The buffer layer was heav-
ily doped with Si and acted as the p* bottom contact to the device. The Si cell
was subsequently shuttered off during growth of the unintentionally doped multi-
plication layer. The same substrate temperature was used since the multiplication
layer consisted of bulk AlposGaggeSb. It was raised slightly to 535 °C for growth of
GaSb/AISb superlattice multiplication layers because AlSb required a higher growth
temperature [9].
The Sb to Al/Ga flux ratio was typically kept at 3 to 1 as indicated by the residual
gas analyzer (RGA). The growth front was smoothed at each hetero interface by a 10
to 30 second Sb soak. A two dimensional growth front was maintained by monitoring
the 1 x 3 reflection high energy electron diffraction (RHEED) pattern characteristic
of reconstructed AlGaSb surface.
The composition of the AlGaSb bulk multiplication layer was adjusted by varying
the Al and Ga cell temperatures and confirmed by X-ray diffraction studies. As an
example, Fig. 3.4 shows a high resolution X-ray scan of a calibration sample. The
high quality of the crystal was evidenced by the narrow and symmetric diffraction
peaks of the buffer and multiplication layers. Typical full width at half maximum
(FWHM) for these peaks were below 30 arc seconds and only slightly larger than
that of the GaSb substrate. Such narrow FWHM’s allowed easy identification of the
closely spaced peaks, and allowed accurate determination of the Al content of these
peaks from their positions relative to the substrate peak.
42
AlGaSb Layer
Al @ 1025 C
10° - FWHM=29 arc “ GaSb Substrate and
E Buffer Layer
AlGaSb Layer / FWHM=21 arc sec
40° E Al @ 1045 C
FWHM=27 arc sec
X-ray Counts
ra
1 r 1 ‘ L n i
C. r 1 rn i ‘
30.2 30.3 30.4 30.5 30.6 30.7 30.8
Omega Angle [Degree]
Figure 3.4: X-ray diffraction scan of Al composition calibration sample. The Al
composition of the AlGaSb layer was varied by changing the Al cell temperature
during MBE growth.
3.3.2 InAs/AISb Superlattice
The selectively doped InAs/AISb superlattice was grown following the multiplication
layer. N-type doping was easily achieved by opening the Si shutter during growth
of the InAs constituent layer. Since this is the last layer in the growth sequence,
it is pertinent to maintain a high crystal quality in the layers prior to superlattice
growth. Compared to the bulk layers, the superlattice structural quality was much
more difficult to maintain due to the short period and mixed anion nature of material.
The difficulties to InAs/AISb superlattice growth originate from cross incorpora-
tion of As and Sb spieces and exchange of these spieces at the InAs/AISb interface [10].
Arsenic incorporation in the antimonide layers is especially severe because As has a
much higher vapor pressure than Sb at a given temperature. During growth of the
AlSb layer, there is considerable As background pressure even though the As shut-
ter is closed. For short period superlattices, the situation is worsened because the
43
shutter times are shorter and there are more individual layers through which crystal
imperfections can accumulate.
The X-ray diffraction pattern of an InAs/AISb superlattice with severe excess
As incorporation is shown in Fig. 3.5(a), whereas a well grown crystal is shown in
Fig. 3.5(b) for comparison. Note that the superlattice central peak is on the large
angle side of the GaSb substrate peak in Fig. 3.5(a) and it is on the small angle
side in Fig. 3.5(b). This is the tell-tale sign of arsenic incorporation because the
arsenides have smaller lattice constants and larger X-ray diffraction angles. Compar-
ison between the two X-ray scans indicate that less arsenic incorporation results in
much better crystal quality as evidenced by the narrower superlattice central peak
and existence of higher order satellite peaks in Fig. 3.5(b).
There are two important factors to preventing arsenic incorporation and achieving
the result in Fig. 3.5(b): minimize the arsenic flux and lower the substrate temper-
ature as much as possible. The arsenic flux is reduced by using a valved cracker
and using the least amount of arsenic flux that still results in an arsenic stabilized
InAs growth front (4 x 2 RHEED pattern). This requires careful calibration since
further reduction in arsenic flux results in an indium stabilized growth front (2 x 4
RHEED pattern) which is detrimental to the crystal quality. Substrate temperature
reduction works because the disparity between As and Sb background pressure drops
off rapidly with temperature. This strategy is limited by the fact that atoms on
the crystal surface lose mobility and three dimensional islanding will result when the
growth temperature becomes too low.
Empirically, the arsenic flux was calibrated from trial runs and the ideal growth
temperature for the InAs/AISb superlattice was found to be slightly above the 1 x 3
to 1 x 5 transition point of GaSb. The 1 x 5 RHEED pattern signifies excess amount
of antimony on the surface and the transition point is dependent on the antimony
flux used. For the antimony flux required in the superlattice growth, the transition
typically occured at 420 °C. Hence the substrate was cooled by nearly 100 °C at the
start of InAs/AISb superlattice growth.
The As/Sb exchange reaction at the InAs/AISb interface was controlled by using
44
AlGaSb Layer,
Substrate and
4.
10'E GaSb Buffer Layer Superlattice
r FWHM=84 arc sec Central
2 10° 7 FWHM=144 arc sec
5 E
(e)
> +1
ran}
TT
x<
FE F i : i 1 1 F | 1 ! 1 !
27 28 29 30 31 32 33
(a) Omega Angle [Degree]
AiGaSb Layer
Superiattice \ FWHM=61 arc sec
Central
4b
10°F EWHM=89 arc seo Substrate and GaSb
Buffer Layer
10° k FWHM=43 arc sec
F +1
X-ray Counts
So 6
por
fa)
27 28 29 30 31 32 33
(b) Omega Angle [Degree]
Figure 3.5: X-ray diffraction scan of 27 A/27 A InAs/AISb superlattice (a) grown
at a high substrate temperature which resulted in excess As incorporation (b) grown
under optimized conditions.
A5
ro neo *. Substrate, GaSb Buff
4|. InAs/AISb ubstraie, Ga uffer,
10 Superiattice and AlGaSb Layer
: Superlattice
10° Central
X-ray Counts
; +1
10°F
10°
26 2°~*~Sé] ;]FtC(<‘«a!!S”*~*«
(a) Omega Angle [Degree]
(b)
Figure 3.6: Growth defects may form for short period superlattices despite good X-
ray data. (a) X-ray scan of 10 A/20 A InAs/AISb superlattice. (b) SEM scan of the
same wafer.
46
a Sb soak at each hetero interface. X-ray photo electron spectroscopy (XPS) study
has indicated that this will result in an InSb like interface [11], which was known to
have a lower defect density level [12]. A soaking time of 5 seconds was used because
the exchange reaction saturates after a few seconds [11].
With these measures, high crystal quality was consistently achieved. The RHEED
pattern remained streaky even for short period superlattices (5 A, 10 A) and exhibited
sharp 2 x 4 and 1 x 3 reconstructions for the InAs and AlSb layers, respectively.
X-ray diffraction scans typically yielded second and third order satellite peaks for
superlattices with relatively long periods (Fig. 3.5(b)) whereas only the second order
peak is visible for short period superlattices (Fig. 3.6(a)). Much of the broadening
in the X-ray diffraction peaks are due to drift in substrate temperature and can be
further improved with better temperature control. It should be mentioned that good
RHEED and X-ray data do not always guarantee defect free wafers. As shown in
Fig. 3.6, scanning electron microscopy (SEM) may reveal defect like features even
though the corresponding X-ray data looks promising. These defects arise from Ga
and In spitting or agglomeration of In, and are highly conductive and detrimental to
device performance.
3.4 Processing
3.4.1 Photolithography
To prevent oxidation of the AlSb, all growth runs ended with a 50 A GaSb capping
layer. The wafers were then taken out of the growth chamber and metallized ex situ by
using a sputter deposition tool. Photolithography was used to define rectangular and
circular device mesas that ranged in size from 37 zm to 200 wm. As shown in Fig. 3.7,
two types of devices were fabricated with different contact metal configurations. In
Fig. 3.7(a), the metal layer was kept at 50 A so that it was semi-transparent to light.
These mesa required only one mask step and was used for preliminary current-voltage
and photo response characterization. The device in Fig. 3.7(b) was designed for direct
AT
Au ———__,
<«—___ n+ Superlattice Layer —__,
*——~ p Multiplication Layer ——”
p+ GaSb Buffer Layer
«—— Unintentionally Doped —» _
p-type GaSb Substrate
«———— In Back Contact --—__,
(a) (b)
Figure 3.7: Avalanche photodiode device mesas. (a) Simple mesa with thin metal
contact. (b) Mesa designed for direct injection of light through the opening in contact
metal.
injection of light into the semiconductor required a two mask process. The mesa had
a light sensitive opening surrounded by a ring of contact metal 2000 A in thickness.
3.4.2 Etching
The etch-down of the device mesas presented a special problem because of the mixed
anion nature of the InAs/AISb superlattice. As shown in Table 3.1, standard wet etch
recipes for arsenides and antimonides did not work well. The arsenide etches were
stopped by the antimonide layers whereas the antimonide etches did not go through
the arsenides. Mixing of these etches was not recommenced due to unforseen chemical
reactions that may take place. For example, the sulfuric acid in the arsenic etch may
react with the methanol in the antimonide etch and form dimethylene sulfate which
is highly toxic [14]. After much experimentation, a satisfactory solution was found
by using Cly assisted dry etching. In this process, the sample was immersed in Cl»
flow while subjected to bombardment by accelerated Ar ions. Etching took place
as the high energy ions milled away the material weakened by reactions with Cl.
48
Table 3.1: Summary of etch results.
Etch Type Result
Brg:Methanol Etch very uneven, surface extremely rough
Brg:HBr:Methanol Stopped by InAs
Bro:HNO3:HCl:Acetic [8] Stopped by InAs
H2SO04:H209:DI water Stopped by AlSb
HF:H2O9:Tartaric acid [13] | Etches all layers, but results in rough sur-
face
Xe/Cly dry etch Etches all layers (including metal) with
smooth surface
The combination of chemical and physical etching makes the process applicable to
arsenides, antimonides, and even metal. The process was also highly anisotropic due
to the directional bias of the high energy ions. Typical etch rates were on the order
of 1 pm/min for GaSb and 0.1 m/min for the InAs/AISb superlattice. As shown
in Fig. 3.8, the dry etched sample had a much smoother surface than its wet etched
counterpart.
(a) (b)
Figure 3.8: (a) Wet etched surface. (b) Dry etched surface with smoother surface and
fewer etch defects.
A9
3.4.3 Passivation
An inherent problem to the dry etching process is the damage on the side walls of
the device mesa. This is shown in the cross-sectional SEM scan of the device mesa
in Fig. 3.9. The fine striations in the micrograph were created from bombardment of
high energy Ar ions and may result in additional surface leakage current.
InAs/AlISb
Superlattice AlGaSb GaSb
Au Absorption Gain Buffer GaSb
Layer Layer Layer Substrate
1 J | |
Figure 3.9: Cross-sectional SEM micrograph of the device mesa.
As a remedy, the devices mesas were immersed in HF:H 20O2:Tartaric acid [18] for
2 minutes following the dry etch. Sulfur passivation was also tried since it was known
to passivate III-V surfaces [15]. The procedure involved exposing the device mesas
to (NH4)25 solutions for up to 5 minutes. The results of these additional processing
steps on device dark current are illustrated in Fig. 3.10. It can be seen that the wet
etch anneal did not have a noticeable effect on dark current. Sulfur passivation did
reduce surface leakage at low reverse bias, but the effect was diminished at high bias.
50
10° £
F \ Dry Etched
10°F
< 10°F
~ a
ce) af
<= 10 F
5 E
O 4
10°
67 um Device
10°7 F i] L : 1 ' 1 7 1 : i r 1 ' l
“42 10 8 6 4 #2 QO 2
Voltage [V]
Figure 3.10: Effect of post dry-etching processing steps on device dark current.
These additional processing step did cause deterioration on the chip surface and
more variation in individual device I-V characteristics. The effect was especially
severe for the (NH4)25 soak. Hence the wet etch anneal and sulfur passivation were
not included as part of the standard device processing procedure.
3.5 I-V and Photo Response Characterization
The current-voltage (I-V) characteristics of the devices were examined by using a
HP 4156 semiconductor parameter analyzer. The I-V curves were taken at room
temperature and liquid nitrogen temperatures. The latter required wire bonding
onto a device header and dunking the set-up into a liquid nitrogen dewer.
Following I-V characterization, devices with low dark current and good avalanche
characteristics were studied for their photo response characteristics. Fig. 3.11 shows
a schematic representation of the photo response set-up. In this experiment, the
51
avalanche photodiodes were excited by using semiconductor lasers. A lock-in scheme
at 10 KHz was employed to pick out the multiplied photo signal against the dark
current background. The semiconductor laser light was injected via a single mode
fiber butt coupled to the top surface of the device mesa. Since the core diameter of
the fiber (9 jum) was much less than the mesa diameter (> 37 ym), coupling loss can
be prevented as long as the fiber was brought sufficiently close to the device mesa (<
50 ym). This simple light coupling scheme was favored because no re-focusing was
needed when a laser of different wavelength was hooked up at the input end of the
fiber.
Fiber
Laser Light
Modulated HP 4185
Voltage Source
Photo-
po Response
Signal
Lock in
APD Device Mesa Amplifier
= Trans Impedance
Amplifier
Figure 3.11: Experimental setup for photo response characterization.
In order to measure the electron and hole ionization coefficients of multiplication
material, it is necessary to have light absorption in both the p and n regions of the
device [16]. This results in electron and hole initiated photo multiplications and
different photo gain curves when the electron and hole ionization coefficients are
different. The ionization coefficients and their ratio can then be calculated from the
coupled pair of photo gain curves if the field profile of the multiplication region is
known [16].
92
Three Stage n-InAs/
AISb Superlattice Figure 3.12: Electron and hole carrier injection by using light of different wavelength. Experimentally, this was accomplished by using semiconductor lasers of different 53 Bibliography [1] [11] [12] D. Effer and P. J. Etter, J. Phys. Chem. Solids 25, 451 (1964). O. Hildebrand, W. Kuebart, K. W. Benz, and M. H. Pilkuhn, JEEE J. Quantum L. Gouskov, B. Orsal, M. Perotin, M. Karin, A. Sabir, P. Coudray, S. Kibeya, Semiconductors: Group IV elements and HI-V Compounds, edited by O. S. M. Sze, Physics of Semiconductor Devices, John Wiley Sons. Inc., New York, D. H. Chow, Y. H. Zhang, R. H. Miles, and H. L. Dunlap, J. Cryst. Growth 150, J. N. Schulman and R. H. Miles, unpublished. F. W. O. Da Silva, M. Silga, C. Raisin, and L. Lassabatere, J. Vac. Sci. Technol. D. A. Collins, Thesis, California Institute of Technology, 1993. D. H. Chow, R. H. Milesand, and A. T. Hunter, J. Vac. Sci. Technol. B 10, 888 M. W. Wang, D. A. Collins, R. W. Grant, and T. C. McGill, J. Vac. Sci. Technol. G. Tuttle, H. Kroemer, and J. H. English, J. Appl. Phys. 67, 3032 (1990). 04 [14] G. Petzow, Metallographic Etching, American Society for Metals, 1976. [15] M. Perotin, P. Coudray, A. Etcheberry, L. Gouskov, C. Debiemme-Couvy, and [16] G. E. Stillman and C. M. Wolfe, in Semiconductors and Semimetals, Volume 12, 90 Chapter 4 Results of Avalanche 4.1 Introduction to Chapter In this chapter the results of antimonide avalanche photodiode research are presented. comparison between the two approaches. 4.2 Early Results The very first avalanche photodiode structure examined had a bulk Alpo4Gag.9gSb The I-V characteristics of this device is shown in Fig. 4.1 and exhibited diode- bias and reached break down levels (10 mA for a 67 wm device) before avalanche 56 2[ O 1 x 10 E ray 10° i Cc OQ 10'— Device Mesa dpe Cc 38 um oO. + 10°F ~~ 53pm : = - ++ -67 um ; O 10° | rn | 1 | i 1 rn l rn l rn | -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 (a) Voltage [V] Cc ® P Seb > 10 E @ Device Mesa G 5/777 767mm : @ 10 E i 1 1 rn | 1 | 1 | 1 L rn | Voltage [V] oO Figure 4.1: Current-voltage characteristics of first avalanche photodiode structures o7 the expected avalanche break down voltage of 14 V. Room Temperature -8 | 1 | 1 J 1 ] 1 i Voltage [V] Figure 4.2: Low temperature I-V characteristics of first avalanche photodiodes fabri- The origin of the high leakage current was addressed by studying I-V scaling with tic of tunneling. The break down could not be due to avalanche mechanisms because 58 the breakdown voltage varied inversely with temperature. At low temperatures, the hot carriers are not scattered as much by phonons [1]. 4.3 Effect of InAs/AISb Superlattice Period on Dark Current The bulk nature of the leakage current indicated that improvements in crystal growth function of superlattice period thickness. 59 ooh 2, TTT TT Current [A] Dry Etched -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 Figure 4.3: Current-voltage characteristics of avalanche photodiodes with As illustrated in Fig. 4.5, a decrease in the InAs layer thickness leads to narrowing tunnel current I «x exp(—E/kT) (4.1) 60 27 Al27 A . i tot = 10 A204 Superlattice 10° Dry Etched Alo.o4Gao.9g5b gain layer and 10 A/ 20 A, InAs/AISb superlattice. where £ is the tunnel barrier (band overlap) and kT is the room temperature thermal Note that the InAs layer thickness has a much stronger effect on the conduction the superlattice bandgap. 61 Ne Alp .o4G ap 9656 : Band 27 A/27 A SLS 10 A/10 A SLS Figure 4.5: Effect of superlattice period on the band alignment between n-type 4.4 Optimization of InAs/AISb Superlattice De- Despite the reduced dark current, the 10 A/20A, InAs /A\Sb superlattice retained the To further reduce the bulk tunneling current and improve the Schottky contact and reduce the tunneling current. Due to the extreme short superlattice period and 62 1 um GaSb Three Stage, InAs/AlSi 0.6 uM Alp o¢@Gy 5,90 : 1 1 1! 28 29 30 31 32 33 Figure 4.6: (a) Avalanche photodiode structure with Alo.o4Gag.9gSb gain layer and 63 third InAs/AISb superlattice stages. 10°F Alo.osGaog65b gain layer and optimized n-type InAs/AISb superlattice. The I-V characteristic of the structure is shown in Fig. 4.7, where previous results order of magnitude over the previous iteration due to the 0.12 eV increase in band 64 obtained and very close to the predicted avalanche break down value of 14 V. 10°F Voltage [V] Figure 4.8: Low temperature reverse break down characteristics of avalanche photo- To verify the avalanche nature of the break down, I-V data were taken at liquid due to reduced phonon scattering at low temperatures. 4.5 Results from Bulk Alp 94GaoqogSb Devices By adapting the optimized, three stage design for the InAs/AISb superlattice, the direct study of avalanche characteristics. The photo gain, associated dark current, 65 using the optical fiber setup described in Section 3.5. 4.5.1 Photo Response Unity Gain Correction Figure 4.9 shows the unprocessed photo gain data. The 1740 nm photo gain curve with bias must be carefully corrected if the true avalanche gain is to be measured. 3.5 - + - - Fit to I=1_,/cosh(L,(1-(V/V,)"”)/D) 67 um Device Mesa Normalized Photo Current 0.5 1 i 1 i rn i n ] 1 i n Reverse Bias [V] Figure 4.9: Photo response of the Alg.ogGao.9gSb gain layer device without correction. device bias. Since the 781 nm light was absorbed in the heavily doped n-type InAs/AISb 66 superlattice, the increase of quantum efficiency with bias could not be attributed to to _ qGo where Ly, is the diffusion length for holes in Alp o4Gag.9gSb, Go is the clearing rate Fig. 4.9 and can be seen to differ only slightly from the straight line fit. For simplicity 67 and consistency, the straight line fit was again adapted here. 4.5.2 Photo Gain and Dark Current Characteristics The corrected near-infrared photo gain curve for the device is shown in Fig. 4.10 At low bias, the dark current can be seen to increase exponentially with voltage, of the Alo.o,Gaoog5b multiplication layer and were inherent to the bulk device. 4.5.3 Impact Ionization Rates The ionization rates of the electrons and holes in the bulk gain layer device were ionization rates can be derived from these curves by using the formulas: 68 E i . LU ¥ T r i ’ T \ T ¥ i] 7 100 Figure 4.10: (a) Dark current and near infrared photo gain characteristics of avalanche 69 BE) = BG pa — a aq) to) (4.4) where a(£) and @(£) are the electron and hole impact ionization rates, M,,(V) and The calculated ionization rates are shown in Fig. 4.11(b) and can be seen to follow There are two explanations for this result. The first possibility is that the long photo gain curves in Fig. 4.11(a) lied on top of each other before reaching a nominal 70 126 Figure 4.11: (a) Photo gain curves for hole and electron injection using 781 nm 71 efficiency. Low Field, hv Figure 4.12: Franz-Keldysh absorption of photons by semiconductors. Absorption Despite these short comings, the photo gain curves in Fig. 4.11(a) were consistent. enhancement in Alp .94Gag.ggSb. 4.6 Results from Superlattice Devices Following study of avalanche photodiodes with bulk Alg94Gao.9gSb gain layer, devices the two types of devices. The optimized, three stage InAs/AISb superlattice n-type 72 device was also adapted in the superlattice device. 4.6.1 Photo Gain and Dark Current Characteristics The near infrared photo gain and dark current characteristics of the superlattice gain Further examination of the superlattice gain layer characteristics reveals more The comparative merits of the superlattice and bulk gain layer devices are best illustrated in Fig. 4.14, which plots on a log-log scale the leakage currents for both 73 aL! T T T "0° F 1645 nm Light 6 38 um Device Mesa ra (b) Photo Gain M Figure 4.13: (a) Dark current and photo gain characteristics of avalanche photodiodes 74 f Curve Fitted to I=1(L)” Figure 4.14: Dark current scaling with device size for APD’s with (a) Alg.o4Gao.9gSb perimeter. 79 types devices against device mesa size at different voltages. The slope of the curve superlattice device performance further improved. 4.6.2 Quantum Efficiency at Low Bias The superlattice gain layer exhibited quantum efficiency characteristics significantly The carrier trapping effect also resulted in variations in quantum efficiency and the applied electric field and reduce the carrier ionization rates, resulting in smaller 76 10¢ Reverse Bias [V] Figure 4.15: Photo response of the GaSb/AISb superlattice avalanche photodiode at photo gains at higher input light intensity [8). It should also be mentioned that a small negative resistance region was observed phenomenon. 4.7 Summary and Conclusion In summary, we have demonstrated antimonide avalanche photodiodes with sensi- GaSb/AISb superlattice gain layers were both realized in a MBE grown structure 77 with a selectively doped, InAs/AISb superlattice as the n-type layer. The avalanche The ionization coefficients of the Alo.o4Gao.9gSb gain layers were measured by using Impact ionization rates in GaSb/AISb superlattice gain layer devices were not ionization may still be enhanced by separate band offset adjustment. 78 Bibliography [3] M. H. Woods, W. C. Johnson, and M. A. Lampert, Solid-State Electron. 16, [4] G. E. Stillman and C. M. Wolfe, in Semiconductors and Semimetals, Volume 12, [5] P. A. Wolff, Phys. Rev. 95, 1415 (1954). [6] S. Wang, Fundamentals of Semiconductor Theory and Device Physics, Prentice- [7] T. Kagawa, Y. Kawamura, H. Asai, M. Naganuma, and O. Mikami, Appl. Phys. [8] R. E. Cavicchi, D. V. Lang, D. Gerhsoni, A. M. Sergent, H. Temkin, and M. B. 79 Chapter 5 Tunnel Switch Diodes Based 5.1 Introduction to Chapter The second device investigated in this thesis is a tunnel switch diode (TSD) based the presence of a tunneling contact. 5.2 Motivation and Background The tunnel switch diode, also known as the metal insulator semiconductor switch Following the initial discovery, silicon TSD devices with tunnel barriers consisting inherent speed, and integration capability which arises from its vertical structure. For 80 10-13 Voltage [V] Figure 5.1: Thyristor like, “S” shaped I-V curve of a tunnel switch diode. example, a static random access memory (SRAM) cell has been constructed from a should advance understanding of TSD operation through another case study. The 81 role of the barrier in TSD switching. 5.3. Tunnel Switch Diode Theory and Design The tunnel switch diode structure generally consists of a tunnel barrier in series with When a small positive voltage is applied to the AlSb barrier, the device enters As the voltage further increases, the device can be switched to the low impedance p-GaSb and turns on the pn junction even more. This positive feedback mechanism 82 (a) High impedance 3 n-GaSb “3F#— - - - Electron quasi-Fermi level FP es Hole quasi-Fermi level Position [um] Figure 5.2: Band diagrams of an antimonide TSD under forward bias. (a) High continues until the device is switched into the low impedance state (band diagram required to deplete the entire p-GaSb layer: where q is the electronic charge, w, is the zero bias depletion width of the pn junction, GaSb epilayer, respectively [1]. 83 given by where E, is the room temperature bandgap in eV, and Nz is the background doping A third possibility for switching exists when the metal electrode Fermi level goes As shown by the band diagram (b) in Fig. 5.2, once in the low impedance state, From the above discussion, it should be apparent that the tunnel barrier plays a low or the barrier is too thin, electrons will leak through and not accumulate at the 84 For Si devices, switching was obtained for oxide barriers with thicknesses that The typical structure of the antimonide TSD is shown in Fig. 5.2. Different samples to confirm this effect. 85 The antimonide tunnel switch diodes were grown by molecular beam epitaxy on Te structure was capped by a 50 A GaSb layer to prevent oxidation of the AlSb barrier. Experiment AISb Layer X-ray Counts C i 4 Omega Angle [Degree] Figure 5.3 shows the high resolution X-ray diffraction scan of a typical structure. mum (FWHM) of 18 arc seconds for the GaSb epitaxial layer and the close agreement 86 between the experimental data and the simulated curve. Following growth, the wafers were metallized ex situ with Au in a sputter deposi- reproducible surfaces with few etch defects. 5.5 Characterization Results The TSD devices were characterized by using the HP 4156 semiconductor parameter switching. Table 5.1: Antimonide TSD structures fabricated and the observed switching char- Device configuration | Switching voltage and current AlSb | p-GaSb Initial values After stressing barrier | epilayer 100 A | 0.2 wm 2.3 V/2.0 mA | 1.2 V/1.0 mA 100 A | 0.4 jum 3.5 V/3.0 mA | 1.2 V/1.0 mA 100 A | 0.6 ym 3.0 V/2.1mA | 1.3 V/0.2 mA 200 A | 0.6 pam 3.1V/1.8mA_ | 1.2 V/0.3 mA 300 A | 0.6 jam no switching observed 500 A | 0.6 wm no switching observed 87 As listed in Table 5.1, switching behavior was obtained for all devices with 100 A and 4 100 A AlSb, Voltage [V] It can be seen that the reverse current of the device was large and varied expo- In the forward direction, switching from the high impedance state to the low of switching current to barrier thickness has also been observed in Si devices [16]. 88 5 7 p-GaSb Epilayer Thickness [um] Figure 5.5 shows the observed dependence of switching voltage on p-GaSb epilayer layer. This is a plausible scenario for antimonide TSD’s because of two factors: 1. The AlSb barriers were relatively thick and may support strong accumulation barrier can be fairly large and a significant amount of bias was dropped across 89 the barrier in the high impedance state, pulling the electrode Fermi level down with respect to the p-GaSb valence band edge. 2. The GaSb bandgap is smaller than the Si bandgap and it takes a smaller bias below the p-GaSb valence band edge. The experimental evidence which may counter this hypothesis is that the switching across the thicker barrier due to stronger electron accumulation at the AlSb/GaSb interface. 0 1 2 3 4 5 6 7 Figure 5.6: I-V characteristics and break down behavior of devices with thick AlSb When the AlSb barrier was increase to beyond 300 A, no switching was observed. it physically broke down at a voltage of 5 to 7 volts. The post break down I-V 90 barrier in the high impedance state of thick barrier devices. 5.5.2 Effect of Current Stress and Dual Mode Switching Be- havior In Si TSD studies, it was observed that the low impedance branch of the I-V curve the barrier in the low impedance state of the switching device. 60- ------ 3 mA 100 A AlSb 0 1 2 3 Voltage [V] Figure 5.7: TSD break down from current stressing. 91 49 | 100 A AlSb ; Voltage [V] For some devices, the switching voltages and currents were drastically modified one low impedance branch to another by increasing the current level in the device. 92 + Mesa Diameter 100 A AlSb Barrier oy Switching Voltage [V] Figure 5.9: Clustering of switching voltages and currents. (a) 100 A AlSb barrier, 0.6 93 3.0 200A AlSb 2.5+ 67 um circular mesa I switch = 0-15 MA (a) Voltage [V] 10} 100 A AlSb | 0.6 um p-GaSb gb 67 um circular mesa O _ j L L ] L L j (b) 0.5 1.0 1.5 Voltage [V] Figure 5.10: Hopping between stable I-V curves due to current stressing. (a) From 94 The “hopping” action could not be precisely controlled but was reversible, i.e. device The stress induced dual mode switching behavior again pointed towards a TSD the switching voltage. 5.6 Simulations Since the TSD switching was strongly dependent on the AlSb barrier thickness, a the tunneling contact, i.e., V- Jn —GR(o,n,p) = 0 (5.4) V°¢—-=(n—-p-—C) = 0 (5.3) V-J,+qR(6,n,p) = 0 (5.5) 95 with = gD,Vn = gftnnVo (5.6) ~qDyVp — gtppVo (5.7) so where ¢ is the electric potential, n and p are the electron and hole concentrations, recombination component and the tunneling current according to Ey, . Jn = ~—qRs + In calc (5.8) where €, is a unit vector perpendicular to the AlSb/GaSb interface, R, is the To reduce the computational load, the equations were first discretized by using diffusion of electrons from the substrate through the pn junction up to the barrier. 96 10°F L fi i n 1 1 i n i 1 it 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Forward Voltage [V] Figure 5.11: Simulated characteristics for TSD devices with 0.6 zm p-GaSb epilayer The carrier life time was relaxed back to its original value once the solution converged findings. 97 The main achievement in this section is the first time fabrication of a tunnel switch as well as the need for high quality antimonide heterostructures. 98 Bibliography [1] J. G. Simmons and A. El-Badry, Solid-State Electron. 20, 955 (1977). [4] T. Yamamoto, K. Kawamura, and H. Shimizu, Solid-State Electron. 19, 701 [5] J. D. Hwang, Y. K. Fang, K. H. Chen, and H. Y. Chiu, IEEE Trans. Electron [6] N. T. Ali and R. J. Green, IEEE Trans. Electron Devices 42, 1978 (1995). [7] Y. K. Fang, K.-H. Wu, and C.-Y. Tsao, IEEE Trans. Electron Devices 44, 34 [8] H. Kroger and H. A. R. Wegener, Solid-State Electron. 21, 643 (1978). [11] J. R. Sdderstrom, D. H. Chow, and T. C. McGill, Appl. Phys. Lett. 55, 1094 (12) D. H. Chow, H. L. Dunlap, W. Williamson, III, S. Enquist, B. K. Gilbert, S. [13] E. S. Daniel, Thesis, California Institute of Technology, 1997. 99 [15] S. E-D. Habib and J. G. Simmons, Solid-State Electron. 22, 181 (1979). [16] P. O. Pettersson, A. Zur, E. S. Daniel, H. J. Levy, O. J. Marsh and T. C. McGill, [17] E. S. Daniel, X. Cartoixa and T. C. McGill, to be published. (18] S. Selberherr, Analysis and Simulation of Semiconductor Devices, Springer- Verlag, Wien, Austria, 1984. [20] C. M. Snowden, Introduction to Semiconductor Device Modeling, World Scien- tific, Singapore, 1986. 100 101 Part Il Ballistic Electron Emission Heterostructures 102 103 Chapter 6 Ballistic Electron Emission 6.1 Introduction to Chapter It should be apparent from Part I of this thesis that band offsets and the inter- used for subsequent study is also described. 6.2. BEEM Theory 6.2.1 Basic Operation of BEEM The basic operation of BEEM is illustrated in Fig. 6.1. It is a three terminal technique typically consist of a thin conductive layer, known as the base layer, on top of a 104 Tip Base Collector Fermi Level (a) No Bias (b) Under Bias Figure 6.1: Basic operation of BEEM. semiconductor of interest. The tip and base terminals support STM tunnel current the semiconductor unimpeded, giving rise to a BEEM collector signal. 105 The BEEM technique can be used to characterize samples in two ways: if the tip carrier injection into the semiconductor and not fluctuations in the large base current. 6.2.2. Parabolic Turn On Model A model for the basic operation of BEEM has been proposed by Bell and Kaiser [2], accounts for all the electron energy in the semiconductor, the maximum transverse momentum is given by ki" = V 2m;(eV — eVo)/h?. This results in a critical angle of incidence: (kre)? ams eV — eVa Due to the small effective mass m, of the semiconductor and the large Fermi en- 106 imaging [2, 3}. Tip Base Collector Figure 6.2: Conservation of transverse momentum at base collector interface. The BEEM collector current at a given bias V can be found by integrating over BEEM collector current becomes eV+Ep Epes VotEr where C is a proportionality constant, Ey is the transverse component. of the electron . hk Mak — Ms (EB — Ep — eVo) from ky conservation. This 107 simplified integral is easily evaluated and indicates that J, behaves as I, =—=(V —V}* (6.3) The parabolic turn on model can be used to extract multiple band edges from hot electrons are scattered into the off axis direction. 6.2.3. Sample Requirements BEEM is a very powerful characterization technique due to its band structure sen- The first constraint is that the base to collector resistance R; must be large, which bias resistance that matters because no bias is applied across the junction during 108 sample metal . film (a) Ry Ip Figure 6.3: Sample requirement for BEEM [11]. (a) Band diagram. (b) Circuit model. BEEM operation. In practice, BEEM device mesas are made as small as possible to The second constraint is that the back contact must be a good ohmic contact. These requirement are readily met in Al,Ga,;_,As BEEM structures, which are BEEM signal and more limitations on sample configuration. 109 6.3.1 Apparatus Z-piezo Tunnel Voltage | ' Cote a BEEM Current Figure 6.4: Experimental Setup for BEEM. The samples in this study were examined by using a BEEM apparatus constructed furnished by the indium on the back side of the sample left over from growth. During 110 STM BEEM 0 100 nw 0 100 nm 04212116.001 Figure 6.5: Artifact in BEEM image due to high scan speed. operation, the BEEM collector current was first converted to voltage by a Keithley head. 6.3.2 Experimental Issues When using the setup for BEEM imaging, it was critical to use a very low scan with a relatively smooth surface and lower the scan rate as much as possible. Lil AM Re (a) Dent (b) Bump Figure 6.6: STM images show surface modifications following BEEM spectroscopy at During BEEM spectroscopy, the STM tip is parked at one location on the sample Since the experiment was carried out in air at room temperature, there was some Prior to the Al,Ga,_,,As experiment, the BEEM spectroscopy set-up was tested and calibrated with the well known Au/Si(100) system, which is one of the first 112 0.8 F Tunnel Current [nA] (b) Figure 6.7: Calibration measurements from Au/Si system. (a) Effect of STM tip 113 were free of artifacts and the system had worked as expected. 114 Bibliography [1] W. J. Kaiser and L. D. Bell, Phys. Rev. Lett. 10, 1406 (1988). [3] M. H. Hecht, L. D. Bell, W. J. Kaiser, and F. J. Grunthaner, Appl. Phys. Lett. [4] E. Y. Lee and L. J. Schowalter, Phys. Rev. B 45, 6345 (1992). [5] M. T. Cuberes, A. Bauer, H. J. Wen, D. Vandre, M. Pretsch and G. Kaindl, J. [6] A. E. Fowell, R. H. Williams, B. E. Richardson, A. Cafolla, D. I. Westwood and [7] T.-H. Shen, M. Elliott, A. E. Fowell, A. Cafolla, B. E. Richardson, D. Westwood, [8] M. Prietsch and R. Ludeke, Phys. Rev. Lett. 66, 2511 (1991). [12] R. J. Miles, Thesis, California Institute of Technology, 1995. 115 Chapter 7 BEEM Study of Al,Ga,_,As 7.1 Introduction to Chapter In this chapter, the capabilities of the BEEM technique are demonstrated through addition, BEEM imaging of the buried interface is demonstrated. 7.2. Motivation Because of the technological importance of Al,Ga,_,As, its various properties have These issues are ideally addressed by BEEM, which not only provides imaging of 116 3.0 7 on Nm 1.5 Band Edge [eV] Direct Band Gap Indirect Band Gap 1.0 1 ! 1 L. 1 it 1 1 1 (a) Al Concentration r Gaas / SE=03! (b) r [100] x Figure 7.1: (a) Measured shift in Al,Ga;_,As band edges with Al concentration. (b) 117 to verify the validity of the BEEM technique. 7.3 Sample Description and Preparation The Al,Ga,;_,As BEEM samples were grown by molecular beam epitaxy and metal- Table 7.1 lists the exact configuration of the epilayer for various Al concentrations. 118 100 A Au Layer 50 A GaAs or 70 AInAs 200 AAI.Ga,,As Layer 0.2 um GaAs Buffer Layer (n = 2x 10° cm?) 0.1 um GaAs Buffer Layer 0.2 um GaAs Buffer Layer (n = 1 x 10" cm®) GaAs Substrate (n = 1x 10" cm*) In Back Contact (a) \ T-pt 2.05 eV X-pt 1.56 eV T’-pt 1.06 eV PHT Ar AU Figure 7.2: (a) Structure of Al,Ga;_,As BEEM sample. (b) Band diagram of BEEM 119 A GaAs layer or a 70 A InAs layer to prevent oxidation of the Al,Ga,_,As layer. Table 7.1: List of Al,Ga,_,As BEEM structures studied. Epilayer Epilayer Cap Layer GaAs none A sputter-etch deposition system was used for post growth metallization. Gold * was the practical limit due to sample placement and tip that an area of 1 mm most samples, the surface morphology was smooth and appeared suitable for BEEM studies. The sputter deposition rate was varied for early samples and was found 120 Prior to metallization, samples were taken out of the UHV growth environment a long time. 7.4 BEEM Imaging Results The samples were examined by using the BEEM set-up described in the previous Figure 7.4 shows a series of BEEM scans over the same area at different tip bias. 121 STM BEEM it) 100 nm O 100 ne (a) 100 nm Area Scan STM BEEM it} 500 nm O 500 nm (b) 500 nm Area Scan Figure 7.3: BEEM images of buried Alp ;;Gao.g9As interface. (a) 100 nm by 100 nm 03192307.001 03192311.001 03192313.001 03192315.001 03192318.001 03192320. 001 03192322.001 03192325.001 03192327.001 03192330,001 03192333.001 03192336.001 me z 4 Ee ue hee. 03192354,001 03192356.001 03192359,001 03200001.001 03200004.001 03200006, 001 Figure 7.4: BEEM images at different tip bias. The tip bias varied from 0.8 V in the 123 The tunneling current was held constant at 5 nA. As the tip bias was gradually in- BEEM Spectroscopy Results from Au/100A AI(0.50)Ga(0.50)As 50.0 1.0 1.5 2.0 25 Figure 7.5: Average signal from BEEM images at different tip bias. Such sophisticated analysis was not developed due to the limited scope of this image as the tip bias was increased. Figure 7.5 shows three such plots from scans of 124 samples over scales greater than 10 nm. 7.5 BEEM Spectroscopy Results While direct BEEM imaging can reveal non uniformities in the buried interface, the surface and the collector current is monitored as a funtion of tip voltage. 7.5.1 BEEM Turn on Threshold Figure 7.6 shows a typical BEEM spectroscopy I-V curve from a sample. The data The parabolic turn on model was applied to analyze the BEEM I-V curve due to its simplicity [14]. Hence the BEEM I-V curve was assumed to take on the form I, = S0(V — Vj)? (7.1) i=] Where J, is the BEEM collector current, V is the tunnel voltage and V; is the thresh- old voltage. Note that each term in the sum only came in when the tunnel voltage 125 < 80 Tunnel Current = 5 nA 150+ O & (b) Tunnel Voltage [V] Figure 7.6: (a) BEEM spectroscopy curve from an Alo 1,;GaogoAs sample. (b) Ex- 126 due to increased scattering of energetic carriers in the metal layer [9]. 7.5.2 Effect of Epilayer Thickness and Capping Layer The Al,Ga,_,As layer thickness was varied for « = 0.25. As shown in Fig. 7.7, The effect of a capping layer is shown in Fig. 7.8. It can be seen that for both tween InAs and Al,,Ga,_,As (8 %), the critical layer thickness for InAs on Al,Ga,_,As 127 1.5/4 100 200 300 400 500 Figure 7.7: Variation of BEEM threshold with Al,Ga,_,As layer thickness. is only a few monolayers. Hence the 100 A thick InAs capping layer must have fully were stable for up to several months. 7.5.3 Variation of Band Edge and Schottky Barrier Height with Al Concentration The BEEM turn on thresholds were interpreted as band edges in the semiconductor. indicate that the capping layer did not significantly affect BEEM thresholds, we may 128 oy Second BEEM Threshold Figure 7.8: Effect of capping layer on BEEM threshold. (a) Alo59Gaos30As sample. 129 barrier height were consistent with the data reported in the literature {11, 10]. Limit of Parabolic Al Concentration Figure 7.9: Variation of Al,Ga,_,As band edges with Al composition x. Multiple To more easily compare the Schottky barrier result with previous findings, it is indirect, bandgap at higher Al concentrations, the surface Fermi level moved towards 130 Indirect 0.0 0.2 0.4 0.6 0.8 1.0 Figure 7.10: P-type Schottky barrier height inferred from BEEM data. Multiple data and stayed close to the middle of the indirect bandgap. This may be due to Fermi samples. 7.5.4 Mapping of the Relative Position of Band Edges in In general, Schottky barrier height depends on surface treatment and other details from the second threshold. The subtraction and error analysis was done for each 131 intrinsic property of the material and had less variation form sample to sample. — Hollow Symbol: No Capping Layer 0.0 0.2 0.4 0.6 0.8 1.0 Figure 7.11: Relative positions of the higher lying band edges as measured by BEEM. These relative energy positions were compared with Al,Ga,..As band structure parabolic model remained valid is also plotted on the same figure. It can be seen that 132 The spatial variation of the band edges are represented by error bars on the data since BEEM spectroscopy was averaged over the larger area. 7.6 Summary and Conclusion BEEM techniques have been successfully applied to the Au/Al,Ga,_,As system. band structure. 133 Bibliography [1] H. C. Casey and M. B. Panish, J. Appl. Phys. 40, 4910 (1969). [4] H. J. Lee, L. Y. Juravel, J. C. Woolley, and A. J. SpringThorpe, Phys. Rev. B [5] J. S. Best, Appl. Phys. Lett. 34, 522 (1979). [6] Yu. A. Gol’dberg, T. Yu. Rafiev, B. V. Tsarenkov, and Yu, P. Yakovlev, Sov. [7] A. A. Talin, D. A. Ohlberg, R. S. Williams, P. Sullivan, I. Koutselas, B. Williams, [8] Semiconductors: Group IV elements and III-V Compounds, edited by O. Madelung, Spriner-Verlag, Berlin, 1991 [9] C. R. Crowell and S. M. Sze, Physics of Thin Films, eds. G. Hass and R. F. [10] W. J. Kaiser, M. H. Hecht, L. D. Bell, F. J. Grunthaner, J. K. Liu, and L. C. [11] J. J. O’Shea, T. Sajoto, S. Bhargava, D. Leonard, M. A. Chin, and V. Narayana- [12] E. H. Rhoderick and R. H. Williams, Metal-Semiconductor Contacts, Clarendon 134 [13] J. O. McCaldin, T. C. McGill, and C. A. Mead, J. Vac. Sci. Technol. 13, 802 [14] L. D. Bell and W. J. Kaiser, Phys. Rev. Lett. 61, 2368 (1988). [15] A. E. Fowell, R. H. Williams, B. E. Richardson, A. Cafolla, D. I. Westwood and [16] L. J. Schowalter and E. Y. Lee, Phys. Rev. B 43, 9308 (1991). 135 Chapter 8 BEEM Study of AlSb and 8.1 Introduction to Chapter This chapter describes the application of BEEM technique to the InAs/GaSb/AISb also addressed. 8.2 Motivation There is much interest in applying BEEM to the antimonides due to the tech- We have selected AlSb barriers and InAs/AISb superlattices for this BEEM study Figure 8.1: AlSb Schottky gate in dual channel mobility modulated transistor. because these structures are highly relevant in current antimonide device research. main reason is that the AlSb Schottky gate is very leaky [6, 7, 8]. It appears that the 137 This issue is ideally addressed by BEEM. The local Schottky barrier data from ment is critically dependent on the band offset. differences within the superlattice [9]. 2.0 5 10 15 20 25 30 Figure 8.2: Calculated variation of InAs/AlSb superlattice bandgap and Schottky The selectively doped InAs/AISb superlattice is an interesting subject for BEEM Schottky barrier height of the superlattice is readily tunable by adjusting superlattice 138 different periods are examined. It is hoped that the resulting shift in band structure will be reflected in the BEEM data. 8.3. Sample Description and Preparation The antimonide BEEM heterostructures examined in this study were all grown by Following oxide desorption under Sb over pressure, an unintentionally doped GaSb For AlSb studies, a 500 A layer of unintentionally doped AlSb was grown over the 139 100 A Al or Au Layer 50 A GaSb or 100 A InAs Capping Layer 500 A AlSb Layer (Unintentionally Doped) or 2400 A InAs/AlSb Selectively 1000 A GaSb Buffer Layer Te Doped GaSb Substrate (n = 5 x 10’ cm”) or Unintentionally Doped GaSb in Back Contact Figure 8.3: Structure of antimonide BEEM sample. InAs capping layer. For InAs/AISb superlattice studies, the superlattice epilayer were grown following of 17A, 24 A, and 48 A. The period thickness was split between the InAs and AlSb 140 layer to better balance the compressive and tensile strain in these layers. The total Similar to the Al,Ga,_,As study, the sputter-etch deposition tool was used for 2 in area, and up to 100 A in thickness. Indium left in arrays of metal dots 1 mm ultra sonic rinse in acetone, isopropanol and de-ionized water before being introduced to the metalization chamber, as in the Al,Ga,_,As BEEM study. 8.4 Results from AlSb Study 8.4.1 Effect of Sample Configuration As discussed in Section 6.2.3, in order to read the minute BEEM signal, it is necessary samples with different configurations and examine their suitability for subsequent 141 0.2+F ‘ Sample A (b) Figure 8.4: (a) I-V characteristics of two types of AlSb BEEM samples. Mesa size ams. 142 Figure 8.4(a) shows the results from two types of AlSb samples. Sample A was at the origin. 8.4.2 BEEM Characterization Despite the improvement in BEEM noise current, sample B was still too noise and The BEEM I-V curve was analyzed by using the parabolic turn on model [13], 143 Nh 100 + BEEM Collector Current [pA] 1 i 1 it n it rn (a) Tunnel Voltage [V] Figure 8.5: (a) BEEM I-V curves for AlSb samples. The tunneling current was held 144 which assumes that the BEEM threshold behavior takes on the form Where J, is the BEEM collector current, V is the tunnel voltage and V; is the thresh- As shown in Fig. 8.5(b), there was significant variation among the individual It should be noted that the BEEM current background noise in the AlSb sample was on the order of 5 pA, which was higher than similarly prepared AlAs samples 145 10 vit AlLSb Xs Figure 8.6: Band structure of AlSb. even through the barrier height in both systems was about 1.2 eV. We attribute this of hole current was evident in the I-V response of the sample to ambient light. 8.5 Results from Superlattice Study 8.5.1 Effect of Superlattice Period Background noise was also a significant problem in BEEM spectroscopy of InAs/AISb adequately suppressed due to the small zero-bias resistance of junction. 146 The superlattice structures fabricated for this study had InAs/AISb period thick- experiments. 8.5.2 Results from 12 A/12 A, InAs/AISb Superlattice Figure 8.7 shows a high resolution X-ray diffraction scan of the 12 A/ 12 A period When these samples were inserted in the BEEM set-up, the background BEEM layer was tenuous. When the STM tip was placed over these regions, thresholds 147 10° E X-Ray Counts "26 27 28 29 30 31 32 33 34 Figure 8.7: High resolution X-ray diffraction scan from the 12 A/12 A, InAs/AISb could be observed in the BEEM spectroscopy curve. Figure 8.9 shows some typical using it to stress a new region. 8.6 Summary and Conclusion We have applied BEEM techniques to the InAs/GaSb/AISb material system. Due indicates that the low lying, indirect band edge of AlSb contributes significantly to 148 2.0 1mm Circular Mesa 1.0- 0.5- Current [mA] -2 -1 0 1 2 (a) Voltage [V] —_ 2, Exponential Fit -, nk 1mm Circular Mesa Current Density [A/cm‘] 0.0 0.1 0.2 0.3 0.4 0.5 Voltage [V] ey Figure 8.8: Current-voltage characteristics of 12 A/12 A, InAs/AISb superlattice 149 80 ~ 607 Threshold = 0.8 V 5 Run 1 E ol | O 407 pL a L g | 207 Run 3 8 | | O ok LU LiL} -20 1 1 I ! Ll 1 1 1 ! ' i ' | A fan) 0.0 O02 04 O06 O08 10 12 1.4 Tunnel Voltage [V] Figure 8.9: BEEM spectroscopy curves from 12 A/12 A, InAs/AISb superlattice electron transport and must be accounted for in device design. A large spread in the In the case of selectively doped n-type superlattice, BEEM spectroscopy was ham- The important issue in BEEM study of antimonides appear to be the large back- these materials. The problem may be partially rectified through careful design of 150 Lol Bibliography [2] H. Lee, P.k. York, R. J. Menna, R. U. Martinelli, D. Z. Garbuzov, S. Y. Narayan, [3] R. H. Miles, D. H. Chow, Y-H. Zhang, P. D. Brewer, and R. G. Wilson, Appl. [4] D. H. Chow, R. H. Miles, C. W. Nieh, and T. C. McGill, J. Cryst. Growth 111, [5] D. H. Chow, H. L. Dunlap, W. Williamson, III, S. Enquist, B. K. Gilbert, S. [6] E. S. Daniel, Thesis, California Institute of Technology, 1997. [7] M. Drndic, M. P. Grimshaw, L. J. Cooper, and D. A. Ritchie, Appl. Phys. Lett. [8] M. J. Yang, F.-C. Wang, C. H. Yang, and B. R. Bennett, and T. Q. Do, Appl. [9] G. F. Williams, F. Capasso, and W. T. Tsang, IEEE Electron Device Lett. 3, [10] D. H. Chow, Y. H. Zhang, R. H. Miles, and H. L. Dunlap, J. Cryst. Growth 150, [11] J. N. Schulman and R. H. Miles, unpublished. 152 [13] L. D. Bell and W. J. Kaiser, Phys. Rev. Lett. 61, 2368 (1988). [14] S. M. Sze, Physics of Semiconductor Devices, John Wiley Sons. Inc., 1981.
Deas VAVAVAVAVAVAVAVAVAVAU
Alp 94Gap,9656
p-GaSb
ae aaa
hv,
0.72 eV 0.75 eV 2
E,(GaSb)
wavelengths. As shown in Fig. 3.12, the 781 nm photons were absorbed in the n-
type InAs/AISb superlattice for hole injection due to its short wavelength, whereas
1645 nm and 1740 nm laser light resulted in electron injection for Alg.94Gapo.9gSb gain
layer devices because the energies of the long wavelength photons were below the
multiplication layer bandgap and the absorption took place in the underlying p-type
GaSb layer.
[2]
Electron. 17, 284 (1981).
and H. Luquet, Appl. Phys. Lett. 60, 3030 (1992).
Madelung, Spriner-Verlag, Berlin, 1991.
1981.
879 (1989).
B 8, 75 (1990).
(1992).
B11, 1418 (1993).
[13] P. S. Gladkov, Ts. Marinova, V. Krastev, Sh. Dinkov, J. Electrochem. Soc. 142,
2413 (1995).
H. Luquet, Mater. Sci. Engineering 28, 374 (1994).
edited by R. K. Willard and A. C. Beer, Academic, New York, 1977.
Photodiode Study
The material can be roughly divided into two parts. The first part of the chapter
describes early effort in dark current reduction and realization of a working device,
where the focal point of the research was on the design of the InAs/AISb superlattice
n-type layer. The second part of the chapter describes the photo response charac-
teristics of avalanche photodiodes with bulk and superlattice gain layers and draws
multiplication layer and a single stage InAs/AISb superlattice. The multiplication
layer thickness was 1 jzm, which ensured that the full depletion width contributed
to the avalanche process. The n-type superlattice was 0.5 jm thick and consisted of
100 periods of 27 A/27 A, InAs/AISb layers. The superlattice bandgap was slightly
greater than that of GaSb as required in subsequent two wavelength photo response
measurements. The relatively long period of the superlattice resulted in good crystal
quality as evidenced by the presence of second and third order satellite peaks in the
corresponding X-ray diffraction scan (Fig. 3.5 (b)).
like behavior. However, the reverse leakage current increased exponentially with
© 10°F
er oa a
a 10
© "
OD a3
@® 10°F
Y) 107k 38 um
= —- — 53um
7 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0
~~
fabricated. (a) Scaling with device area. (b) Scaling with device size.
characteristics could be observed. The high leakage current limited the maximum
reverse bias across the device to 2 V, which was an order of magnitude smaller than
10 -2 -1 0 1 2
cated.
device size and measuring I-V characteristics at low temperatures. From Fig. 4.1,
it can be seen that area scaling dominated the reverse behavior whereas perimeter
scaling was more important under forward bias. This indicated that the reverse leak-
age current had a significant bulk contribution. Given the exponential nature of the
reverse current, tunneling across the reverse biased pn junction was suspected as the
underlying dark current mechanism. This was confirmed by the low temperature I-V
data shown in Fig. 4.2. The reverse current at liquid nitrogen temperature was much
smaller than at room temperature but retained the exponential behavior characteris-
avalanche breakdown would have happened at a lower voltage due to lack of phonon
scattering, i.e. the ionization process is enhanced at low temperatures because the
or structural design were necessary. To address the first possibility, the growth con-
ditions were systematically varied to reduce interface defects and improve the general
quality of the crystal. Fig. 4.3 shows the I-V curves from two other attempts in com-
parison with the original result. Curve b was obtained from a wafer grown with a
higher Sb/Ga ratio for the avalanche multiplication layer. The higher Sb over pres-
sure should minimize Ga on Sb anti-site defects and reduce the background doping
level [2]. Curve C was obtained from a device with the InAs/AISb superlattice grown
at a slightly higher temperature and with a lower Sb/Ga ratio to reduce Sb incorpo-
ration. It can be seen that both strategies have produced devices with higher dark
currents. In general, deviations from the optimized growth conditions described in
Chapter 3 have resulted in worse crystal quality and poorer device characteristics.
These growth experiments indicated that the high dark current was not associated
with the crystal quality of the growth and must be due to the inherent property of
the structure itself. This hypothesis was confirmed by examining structures with dif-
ferent InAs/AISb superlattice designs. As shown in Fig. 4.4, changing the constituent
layer thicknesses of the superlattice from 27 A/27 A to 10 A/20 A led to markedly
different I-V characteristics. The improvement from the short period superlattice can
be understood by examining Fig. 3.3, which shows the calculated variation of the
superlattice bandgap and the band overlap between the superlattice and GaSb as a
10’ 38um Device
Voltage [V]
Alo.oaGao.9gSb gain layer and 27 A / 27 A InAs /A\Sb superlattice grown under differ-
ent conditions.
of the electron quantum well and stronger carrier confinement, causing an upward
shift in the electron energy level and the conduction band edge. This results in a
larger energy gap for the InAs/AISb superlattice and more band overlap between the
n-type superlattice and the multiplication layer. The latter is especially significant.
since much of the tunneling current originates from the pn heterojunction where
the effective bandgap is the smallest in the entire structure. For the 27 A/27 A
superlattice, the band overlap at the heterojunction is only 0.3 eV, whereas the 10
A/ 20 A structure has a band overlap of 0.58 eV. Using a simplistic model for the
INAS/AISB >.
Superlattice ‘\
ra | InAs/AISb
=)
38 um Device
| 1 1 rn | 1 L 1 i rn i rt ! rn H 1
-10 =-8 -6 “4 -2 0 2 4 6
Voltage [V]
Figure 4.4: Current-voltage characteristics of avalanche photodiodes with
energy (25 meV), one concludes that a 0.28 eV change in band overlap should reduce
the tunnel current by a factor of 10* to 10°. As shown in Fig. 4.4, this is roughly
what was observed.
band edge and the superlattice bandgap than the AlSb layer thickness. This is due
to the much deeper electron quantum wells from the larger conduction band offset
between AlSb and InAs (see Fig. 1.1). The shallow hole confinement means that
variations in AlSb layer thickness do not significantly alter the valence band edge and
Overlap
InAs/AISb superlattice and the Alp.o4Gao,9gSb multiplication layer.
sign
exponential reverse I-V behavior characteristic of tunneling. The bulk nature of the
dark current was confirmed by scaling studies and the fact that additional processing
only has a weak effect on dark current. Moreover, it can be seen from Fig. 4.4 that
the forward conduction of the short period superlattice sample was reduced due to
the larger superlattice bandgap and Schottky barrier height at the surface.
to the n-type superlattice, an optimized design employing three stages of superlattice
was adapted. As shown in Fig. 4.6(a), the multiplication layer was interfaced to a 5
A/10 A, InAs/AISb superlattice to maximize the band overlap at the heterojunction
Bottom Contact
Layer (Si doped,
p=2x10'/cm’)
Superlattice (InAs Layer
Selectively Doped with Si,
n=1x10"/em’)
Multiplication Layer 4 ad
(Unintentionally Doped, pos aye E,
p=5x10"/cm’) } Metal
} N* ?
300 A 5 cis * 0.1 pm,
» ONY 27N27A,
SA/1OA, OH nAS/AIS
INAS/AISb TON20M Superlattice
Growth Direction Superlattice — INAs/AlSb
——___ > Superlattice
(a)
Al, pgGao.9¢20 Gain layer, GaSb Substrate
10° & Three Stage InAs/AlSb Buffer Layer
F Superlattice Central Alp o4 Gao gg Layer
3 SLS2 SLS3
c i
a L
2b
O 105 +1 (SLS3)
> "
© | -1(Sts2) -1 (SLS3)
x< 10
Omega Angle [Degree]
optimized, three stage, n-type InAs/AISb superlattice. Device is shown under reverse
bias. (b) X-ray diffraction scan of the structure.
the large number of interfaces, the overall thickness of the stage was limited to 300
A so that good crystal quality was maintained for erowth of subsequent layers. The
second stage consisted of 0.2 um of 10 A/20 A, InAs /A\Sb superlattice and served as
the transition layer. This was followed by a 0.1 jm thick, 27 A/27 A layer with the
narrow bandgap necessary for improving contact characteristics at the surface. The
X-ray scan of one such structure is shown in Fig. 4.6(b). The high crystal quality of
growth was evidenced by the presence of satellite diffraction peaks of the second and
10°F
a 10°F ;
Cc InAs/AlSb Superlattice
© | ~=Configuration
5S 10°; ..... 27N/27A
O r —-- 10A/20A
40°! - —— Three Stages: 38 LM
5A/10A, 10A/20A, 27A/27A | Device Mesa
1 -13 L 1 i] 1 fi 4 | 1 i
° “15 -10 “5 0 5
Voltage [V]
Figure 4.7: Current-voltage characteristics of avalanche photodiodes with
are also plotted for comparison. It can be seen that the device was more conductive
under forward bias due to the smaller bandgap of the 27 A / 27 A superlattice contact
layer. The short period superlattice at the interface resulted in a 0.7 eV band overlap
with the multiplication region. And the dark current was reduced by another two
overlap, At a reverse bias greater than 10 V, rapid current. increase characteristic of
avalanching action could be observed. A breakdown voltage of 13 V was consistently
10°F
< 10
1)
Se
S 10°}
Three Stage
10°24 InAs/AlSb Superlattice
67 um Device t
] 4 i 1 i 1 i
-10 -5 0 5
diode with Alpo4Gao.9gSb gain layer and optimized n-type InAs/AISb superlattice.
nitrogen temperatures. As shown in Fig. 4.8, the avalanche onset voltage, i.e. the
voltage at which the slope of the I-V curve began to steepen, occurred at a lower bias
dark current in bulk Alpo4Gao.9g5b devices was reduced to a low enough level for
and ionization characteristics of the working device were henceforth measured by
was nearly identical to the 1645 nm curve and is not shown here to reduce clutter.
The quantum efficiencies for 781 nm, 1645 nm and 1740 nm light were 16%, 10%, and
5% respectively. It can be seen that the quantum efficiency rose slowly as the bias
was increased. The effect was most pronounced for 781 nm light and resulted in an
inflexion point in the photo gain curve at about 2 V. The drift in quantum efficiency
[Fit for Unity Gain
0 2 4 6 8 10 12
The curves are fitted at low bias to correct for changes in quantum efficiency with
the bias dependence of the depletion width. The junction depletion width on the
n side was no more than several hundred A even at break down voltages. Given
an overall thickness of 0.3 ym for the n-type layer and an absorption length on the
order of 0.1 zm, most of the 781 nm light was absorbed near the surface and the
slight. variation in junction depletion width should have negligible effect. on collection
efficiency. Instead we attribute this effect to band discontinuity in the InAs/AISb
superlattice. As shown in Fig. 4.6(a), band bending from heavy doping and the multi-
stage design of the n-type superlattice resulted in hole barriers in the valance band
even though most of the band offset occurred in the conduction band. Such barrier
induced light injection inefficiency was even more severe in devices with GaSb/AISb
superlattice multiplication layers as will be discussed in Section 4.6.2. To correct for
this effect, the photo gain curve around the inflexion point was fitted to a straight line,
which was extended to higher bias regions and taken as the unity gain background.
For 1645 nm and 1740 nm light, the drift in quantum efficiency entailed a differ-
ent mechanism. Since the long wavelength photons were absorbed in the underlying
p-type GaSb buffer layer, the photo-generated electrons must. diffuse across the un-
depleted region of the multiplication layer to be collected. The quantum efficiency
varied with bias because the depletion width (and hence the length of the undepleted
section) in the lightly doped multiplication region was a strong function of voltage.
According to Woods et. al. [3],the collected current should vary with voltage according
f= cosh(Ly — W)/Lpn) (4.2)
of holes at the hetero interface, Lo is the width of the Alp.o4Gao.96Sb multiplication
layer and W is the depletion layer width. The fitted unity gain curve is illustrated in
along with the device dark current characteristics. Maximum gains as high as 30
were observed. The dark current density was typically 6 A/cm? at a more moderate
gain of 10.
indicating that tunneling mechanism was at work. At high bias, the dark current rose
faster and deviated from the exponential curve. However, the exponential behavior
was recovered when the unmultiplied dark current (dark current divided by the photo
gain, dashed line in Fig. 4.10(a)) was plotted. This indicates that the reverse leakage
at high bias underwent multiplication and must be due to bulk tunneling rather than
surface leakage. As shown in Fig. 4.10(b), the additional exponential contribution
from tunneling caused the dark current to rise at a faster rate than the photo gain.
These observations were supported by scaling I-V scaling studies where the dark
current was found to scale with the device area (Fig. 4.14(a)). Thus we conclude that
the relative high levels of dark current were due to tunneling from the small bandgap
measured by using the two wavelength photo injection scheme described in Section 3.5.
Figure 4.11(a) shows the experimental photo gains curves for 781 nm and 1645 nm
light. The data have been corrected for bias-induced variation in unity gain. If
we assume that pure holes were injected from 781 nm light illumination and pure
electrons were injected from 1645 nm light illumination, the electron and hole impact
af 1740 nm Light
10 37 um Device Mesa
< :
e +f -
© 40%. >
& O
> ot
O L o)
x F G@)
= seb)
A 10°F 5
-10 r rn I 1 l 1 l i i 1 ] i i 0.1
10 ) 2 4 6 8 10 12
(a) Reverse Bias [V]
10°
Curve Fitted to I=1,(M)? -
— 10°F
ww
Swe
A 10°F
10’ 5 a
1 10
(b) Photo Gain M
photodiodes with Alpo4Gao.9g5b as the multiplication layer. The dashed line shows
the un-multiplied dark current. (b) Device dark current plotted as a function of the
photo gain. The data was fitted to a power law I=I,(M)?, where p=1 for constant
un-multiplied dark current.
M,(V) the photo gain at bias V for electron and hole injection, and E the maximum
electric field in the abrupt pn junction at bias V [4]. These equations were derived
by assuming the field profile of a one sided, abrupt, p™n? junction without punch
through, which implies that the depletion width is always smaller than the p~ layer
thickness and does not extend into the p* contact layer at high reverse bias.
the general a, 8 = exp(—a/bE) behavior. This is the expected field dependence for
impact ionization rates at high field conditions [5]. However, the opening between
the electron and hole ionizations curves was smaller than expected and the measured
hole ionization rates were only slightly higher than those of electrons.
wavelength photons were partially absorbed by the Alg.o4Gao,9g5b multiplication layer
or the InAs layers in the n-type superlattice and pure electron injection was not
achieved. As shown in Fig. 4.12, under high field conditions, a semiconductor can
absorb photons that fall within its bandgap due to the Franz-Keldysh effect [6]. Since
the 1645 nm (0.74 eV) and 1740 nm (0.72 eV) photons have energies just below the
bandgap of the Alpo4Gao9g5b multiplication layer (0.75 eV), partial absorption in
the multiplication layer was likely to be important. A second possibility is that
quantum efficiency increased more rapidly with bias than was accounted for by the
linear correction for unity gain. This was evidenced by the fact that the two corrected
Al, o4Gp ggSP .
Multiplication Layer 781 nm Light
10F- 67 um Device Mesa Hole Injection
& 8Fr
fa) 1645 nm Light
O Electron Injection
o §&F
_—_
Oo 47
2 s
4 | 1 J 1 i n | 1 | 1 |
0 2 4 6 8 10 12
(a) Reverse Bias [V]
10
Electron
~ 1
cab) E
a Ot oot
or “tele
Cc Bo “¢
S a
w a # a
o4F a
ro) a
~~ | i 1 rn 1 1 L
0.035 0.040 0.045 0.050
(b) 1V/E [um/V]
and 1645 nm light. (b) Calculated hole and electron impact ionization rates in
Alp.o1Gao.965b. The device was assumed to have an abrupt. pn junction.
gain of 2. The observed gain before this point may be due to an increase in quantum
in the bandgap is possible due to overlapping of wave functions under high field
conditions [6].
with hole ionization enhancement since hole injection always yielded higher photo
gains. This result was obtained from a large number of devices under different ex-
perimental conditions and can be regarded as partial confirmation of hole ionization
with superlattice gain regions were fabricated and characterized by using the same
methodology. The superlattice multiplication region consisted of 10 periods of alter-
nate GaSb and AlSb layers 300 A in thickness. The overall thickness of 0.6 jum for the
gain layer was the same as the bulk device and enabled direct. comparisons between
layer which resulted in low dark current and improved contact resistance for the bulk
layer device are shown in Fig. 4.13. The device yielded an avalanche break down
voltage of 18.5 V, which was higher than its bulk counterpart due to the presence
of additional AlSb barriers in the gain region. Since long wavelength photons were
absorbed by GaSb layers in the superlattice gain region, the two wavelength scheme
for measuring impact ionization rates could not be applied. Illuminations by the
781 nm, 1645 nm, and 1740 nm lasers all resulted in hole injection and there was
little difference between the photo gain curves except for the quantum efficiency
achieved (20%, 5% and 3% at unity gain, respectively). Comparing Fig. 4.13(a) to
Fig. 4.10(a) reveals that the avalanche characteristics were much more pronounced for
the superlattice device. As shown in Fig. 4.13, gain factors up to 300 were observed
in the near infrared for the superlattice device. At a gain factor of 10, the dark
current for the 37 ym device was 8 4zA, which was an order of magnitude lower
than in bulk Alp o4Gao.9g5b devices and comparable to InGaAs/InAlAs superlattice
avalanche photodiodes of similar design [7].
subtle differences. In contrast to the bulk device, Figure 4.13(b) shows that the
superlattice dark current increased slower than the photo gain in the avalanche region.
In fact, the un-multiplied dark current stayed constant or decreased with voltage at
high bias (dashed line in Fig. 4.13(a)). This indicated that much of the dark current
did not undergo multiplication and must be from surface leakage. This suggested
that bulk tunneling current had been suppressed due to the presence of AlSb barriers
in the superlattice gain region.
x 10 4 100
c U0
© >
S 410 9
e) o
a @
= of
‘a qi}
] H i 1 0.1
0 5 10 15 20
(a) Reverse Bias [V]
10°
oO pA
a”
with a 10 period, 300 A/300 A, GaSb/AISb superlattice multiplication layer. The
dashed line shows the un-multiplied dark current. (b) Device dark current plotted as
a function of the photo gain. The data was fitted to a power law I=I,(M)?, where
p=1 for constant un-multiplied dark current.
VOR = 12.2V, 0.97 V,
Bw an
q P=1.B Qn
ne woot *411.5V, 0.9 V,
<= Fo 0 op=2.2T%°7>
oO BO ween neo * 8 .5V, 0.67 V,
© Fe
— 4 p=1.8e8
=:
O + = ee = 4V, 0.31 V,
we 10F p=1.60---------00007777
ar F
Oo c
QQ 105 coneneeenene eer #1.5V, 0.12 V,
F p=1.58—~
10°
30 40 50 60 70 80
(a) Device Size L [um]
10°F Curve Fitted to I=I,(L)"
= 10¢
S 3 ogee aa ween ee #17V,0.9V,
‘ Pp=1 02M re
= 106 p=0.858°0 ~~“ * 42V, 0.64 Ve
O '
¥ 1OF piiiiinn-------- aoe enone * 6V,0.32 V,
= F 06 p=1.00
Q 10 a eeneeee _--------- ™2V,0.12V,
EF op=1.08"
10° , , ,
30 40 50 60 70 80
(b) Device Size L [um]
multiplication layer and (b) GaSb/AISb superlattice multiplication layer. The data
is fitted to a power law I=I,(L)?, where L is the device size. The curve fit should
yield p=2 for perfect scaling with device area and p=1 for perfect scaling with device
indicated whether the leakage current. scaled with device area or perimeter, and can
be viewed as an index on the relative importance of bulk and surface leakage current.
It can be seen that bulk tunneling current is significant in Alo.o4Gag.9gSb devices due
to the smaller bandgap of the multiplication layer. The superlattice dark current
was much lower and varied linearly with device size until the very onset of avalanche
break down. The surface character of the observed dark current implies that the
fundamental limit in leakage suppression was not yet reached. With better process-
ing and passivation techniques, the surface leakage can be readily reduced and the
different from its bulk counterpart. There was a strong dependence of collection ef-
ficiency on device bias at low voltages due to the presence of AlSb barriers in the
multiplication region. As shown in Fig. 4.15, a bias as high as 10 V was needed to
overcome the barrier and reach unity gain. This effect was similar to the heterostruc-
ture induced injection inefficiency for 781 nm light as discussed in Section 4.5.1. It
is much more pronounced here because the carriers can become trapped in GaSb
quantum wells and the band alignment in the multiplication region has a stronger
dependence on voltage due to its low background doping level.
photo gain with light intensity. As illustrated in Fig. 4.15, a higher reverse bias was
needed to achieve the same quantum efficiency as the input light intensity was in-
creased. The apparent photo gain also decreased when the device was under stronger
illumination. This is because the photo-generated electrons and holes drifted in differ-
ent directions and can become partially trapped in GaSb quantum wells at opposite
ends of the multiplication region. The accumulation of these carriers tended to screen
t Input Light Intensity
> 22° 1.4 uW
—~ — 23uW .
—— 159 pW . 4
c 4,
ay
O it
Oo E
——
{e)
OW
/ 1645 nm Light Injection
" 67 um Device Mesa
0.1 ul \ I , ! ‘ I
0 5 10 15
different light intensity levels. The data was obtained by using a 1645 nm laser light.
Similar results were obtained from 781 nm and 1740 nm light sources.
in the photo gain curve at low levels of light injection. The resulting peak in photo
gain curve became less prominent and shifted to a higher voltage when the input
light intensity was increased. An explanation does not yet exist for this interesting
tivity in the near infrared out to 1.74 wm. Devices with bulk Alp 94GagogSb and
photodiode dark current was found to critically depend on the n-type superlattice
design and the resulting band offset at the p-n* heterojunction. Early efforts were
plagued by large tunneling current due the small band overlap between the n-type
superlattice and the multiplication layer. This was remedied by using a InAs/AISb
superlattice with a three stage design which led to substantial improvements in device
dark current and contact characteristics.
a two wavelength photo response setup. The result was consistent with hole impact
ionization enhancement as was predicted by the spin-orbit split-off band resonance
argument. However, the ionization measurement was to a certain extent. compromised
due to possible mixing of carrier injection and bias-induced quantum efficiency effect.
Moreover, the Alp.oaGaog.gg5b gain layer device exhibited a relative high level of dark
current which tended to negate the detector sensitivity advantage gained from impact
ionization enhancement. The results of the ionization and dark current study leads to
the conclusion that hole ionization enhancement from spin-orbit. split-off band reso-
nance in Alpo4Gao,965b will be of only limited use because a high level of dark current
is inherently associated with the narrow bandgap of the Alp.o4Gao.9gSb multiplication
layer.
measured due to absorption of long wavelength light by GaSb layers in the gain region
and variation of photo gain with illumination intensity. However, the superlattice
gain layers resulted in devices with much lower dark current and more pronounced
avalanche characteristics. The observed dark current was due to surface leakage and
can be readily improved from better processing and passivation. The superlattice
multiplication layer was found to be more promising than its bulk counterpart because
tunneling current is readily suppressed by barriers in the gain region while impact
[1] S. M. Sze, Physics of Semiconductor Devices, John Wiley Sons. Inc., New York,
1981.
[2] K. F. Longenbach and W. I. Wang, Appl. Phys. Lett. 59, 26 (1991).
381 (1973).
edited by R. K. Willard and A. C. Beer, Academic, New York, 1977.
Hall, Englewood Cliffs, NJ, 1989.
Lett. 66, 993 (1989).
Panish, Phys. Rev. B 38, 13474 (1988).
on AlSb/GaSb Heterojunctions
on AlSb/GaSb heterojunctions. This chapter describes the design, growth, and char-
acterization process which resulted in the first such device in the antimonide system.
The device current-voltage characteristics and the effect of current stressing are ex-
amined in detail in order to deduce the TSD switching mechanism. These results are
correlated with drift diffusion simulations which have been modified to account for
(MISS) [1] or controlled inversion device (CID) [2], was first discovered by Yamamoto
while studying the properties of metal/SiO2/n-Silicon/p-Silicon structures [3]. As
shown in Fig 5.1, the device is characterized by a “S” shaped current-voltage (I-
V) curve with a negative differential resistance (NDR) region much like that of a
thyristor. It was found that. the TSD can be switched between the high and low
impedance branches of the I-V curve very rapidly. Switching times of less than 2 ns
were obtained for large area devices [4].
of SiC [5], plolysilicon [6], amorphous silicon [7] and Ge [8] were also fabricated. The
device has found unique applications in circuit design due to its large non-linearity,
Low
1024 Impedance
103 3
x : NDR Region
€ |
@ 1074
& :
6 1
105 3
High Impedance
1064
10-7 )
0 1 2 3
single TSD element in the Si system [9]. Compared to rival conventional designs which
require six transistors, the TSD-SRAM is both more compact due to its structural
simplicity and faster because of the tunneling nature of the switching process.
There is much interest in an all antimonide TSD because such a device forms
the much desired complement to the existing resonant interband tunneling diode
(RTD) [11]. The “S” shaped I-V curve of the TSD is symmetrically opposed to the
“N” shaped I-V curve of the RIT, and the two devices are in fact circuit duals of each
other. The addition of a device with thyristor-like negative differential resistance will
undoubtly bring more functionality to the burgeoning RIT-based antimonide high
speed circuits [12]. From a scientific standpoint, the antimonide version of the device
role of the oxide tunneling barrier is not well understood in the Si device [13]. By
replacing it with AlSb, which has a different barrier height and may have deep levels
within its bandgap, more experimental evidence will be collected to shed light on the
a pn junction. As shown in Fig. 5.2, the antimonide implementation employs an
AlSb barrier and a GaSb pn junction. For the pn junction polarity shown, the device
is reverse biased when a negative voltage is applied to the AlSb barrier (not shown
here). In this state, most of the bias is dropped across the pn junction and the device
I-V characteristics follows that of a reverse biased pn diode.
a high impedance state with most of the voltage dropped across the p-type GaSb
epilayer. If the AlSb tunnel barrier is of the right thickness, electrons will partially
accumulate in the p-GaSb layer and cause it to enter deep depletion (band diagram
(a) in Fig. 5.2). The current in this high impedance state is due to generation in the
p-GaSb layer. Hence it is proportional to the width of the depletion region and the
square root of the applied bias.
state in a number of ways. In the so called “punch through” mode of operation,
the depletion region in p-GaSb extends deeper into the epilayer until it reaches the
buried pn junction. Further bias causes the pn junction to be turned on, flooding the
p-GaSb layer with electrons. If the AlSb tunnel barrier is not too thin, the electrons
will accumulate near the interface between the p-GaSb and AISb layers and create an
inversion layer. The resulting high charge density at the interface is able to support a
large electric field across the AlSb barrier, causing the tunneling current to increase.
This results in high levels of hole injection which pulls down the hole Fermi level in
= |
Ot apa) [ooo-- eee —>
es
Cc —
Lu (b) Low impedance
-4 ‘ ! ' i 1 1 , ! 1
-0.2 0.0 0.2 0.4 0.6 0.8
impedance state with deep depletion in the p-GaSb epilayer. (b) Low impedance
state with the pn junction turned on and most of the bias dropped across the AlSb
barrier. The energy scale of the high and low impedance states are shifted for clarity.
(b) in Fig. 5.2). Since this process is initiate by “punching through” of the p-GaSb
depletion layer to the buried pn junction, the switching voltage is given by the voltage
Vewitch = qN.(w _ Wo)? /2€Gasb (5.1)
and N.,w, and égasp, are the doping density, width, and dielectric constant of the p
If the doping level in the p-GaSb layer is sufficiently high, the electric field in the
p-GaSb depletion layer may become large enough for avalanche processes to become
significant before “punch through” occurs [1]. The avalanche action will cause elec-
trons to accumulate at the p-GaSb/AISb interface and start the same chain of events
that lead to switching. The switching voltage for this avalanche induced process is
Vewiteh = 60(Eq/1.1)°?(Np/10"°)*/* (5.2)
in cm~? [14].
below the p-GaSb valence band edge, allowing large hole currents to tunnel into the
p-GasSb layer and turn on the buried pn junction [9]. This tends to occur for thick
barriers with relatively strong electron accumulation at the AlSb/p-GaSb interface.
In this case, a significant portion of the bias is dropped across the barrier in the high
impedance state.
most of the bias is dropped across the AlSb barrier and the device I-V characteristics
assume the exponential behavior of a tunnel barrier. The low impedance state is
sustained as long as there is enough electron current to support the inversion layer at
the GaSb/AISb interface and maintain the large bias drop across the tunnel barrier,
or as long as there is enough hole tunneling into the p-GaSb layer to keep the pn
junction forward biased. If these currents become low enough to be consumed by
recombination at the pn junction, the device switches back to the high impedance
state.
critical role in the TSD switching action. In order for switching to occur, the tunnel
barrier must have the right barrier height and thickness. If the barrier height is too
p-layer/barrier interface. Without a depletion region in the p-layer, the potential
energy of the p-GaSb layer will follow that of the surface electrode. This means that
the device will assume the highly conductive characteristics of a forward biased pn
junction and will always be in a low impedance state (the pn junction is the conduction
bottle neck here due to the ineffectiveness of the tunnel barrier). If the barrier height
is too large or the barrier is too thick, the carriers are effectively blocked and there is
no place for the accumulated electrons to escape. This will cause the surface of the
p-type layer to invert and screen the field from the metal electrode. In this case, most
of the bias falls across the barrier instead of the depletion region in the p-type layer.
Since the barrier is not very conductive, the device is stuck in a high impedance state.
ranged from 15 A to 40 A [15, 16]. The AlSb barrier in the antimonide device is
typically much thicker (greater than 100 A) due to its smaller barrier height. To
maximize the AlSb barrier height, the polarity of the pn junction is chosen such that
the AlSb layer partially blocks the electron flow in the high impedance state. This is
the desired configuration because AlSb is a more effective barrier for electrons than
holes, i.e. the conduction band offset between AlSb and GaSb is 1.15 eV at the P
point and 0.55 eV at the X point valley, which is larger than the valance band offset
of 0.4 eV between these materials.
devices were fabricated to study the structural dependence of device characteristics
and deduce the TSD switching mechanism. As listed in Table 5.1, the AlSb barrier
was varied from 100 A to 500 A to delineate its effect on switching. Since the p-type
GaSb epilayer was unintentionally doped, the devices were expected to operate in
the punch through mode with a switching voltage strongly dependent on the epilayer
thickness. Hence the p-type GaSb layer thickness was varied among 100 A AlSb
5.4 Growth and Fabrication
doped (n=1x10!8/cm?) wafers. Due to the lack of a suitable n-type dopant, the GaSb
pn junction was formed at the substrate surface by depositing an unintentionally
doped p-type GaSb buffer layer (p=5x10'6/cm*). To improve the quality of the pn
junction, the substrate was etched and thoroughly heated for oxide desorption prior to
buffer layer growth. Note that the superlattice doping scheme described in Chapter 3
could not be applied here because the n-type superlattice could not be grown directly
onto the substrate. The GaSb buffer layer constituted the p-type epilayer in the
device and was grown at a substrate temperature of 520 °C. The AlSb barriers were
grown at a slightly higher temperature, and a 30 second Sb soak was applied at the
GaSb/AISb interface to smooth out the growth front. At the end of the growth, the
10° — — Simulation | GaSb Layers
with 415 A AlSb layer FWHM =
71 arc sec
FWHM =
354 arc sec
8,
29.6 29.8 30.0 30.2 30.4 30.6
Figure 5.3: X-ray diffraction scan of an antimonide TSD structure.
The high quality of the growth was evidenced by the narrow full width at half maxi-
tion tool. Indium left over from the growth served as the back contact. Device mesas
were defined by standard photolithography and ranged in size from 38 pm to 200
jum. Chlorine assisted dry etching was used as the final etch down and yielded highly
analyzer. To accommodate the “S” shaped negative differential resistance region in
the I-V curve, the analyzer was run in the current sweep mode instead of the usual
voltage sweep mode. This prevented dramatic swings in device current from the
switching action and allowed recording of the full I-V curve in one sweep. The main
results of the characterization study are summarized in Table 5.1, which shows the
structural dependence of TSD operation and the effect of current stressing on TSD
acteristics. Device size was 67 fm.
(unstable)
5.5.1 Effect of Barrier and Epilayer Thickness on Switching
200 A AlSb barrier thicknesses. The typical I-V curve of a switching device is shown
in Fig. 5.4.
0.6 um p-GaSb
| 67 um mesa
= 2}
~~
5 be
= /
+ O
O /
V witch = 2.95 V
| ewiteh = 1.4 mA
-2 &
l l 1 l
-2 -1 0 1 2 3
Figure 5.4: Typical I-V characteristics of antimonide TSD’s.
nentially with voltage, indicating that tunneling through the buried pn junction was
significant. The less than ideal quality of the pn junction was expected since it was
formed at the substrate surface.
impedance state occurred at voltages from 2.3 V to 3.5 V. The switching current
densities ranged from 10 A/cm? to 300 A/cm?, and did not change appreciably when
the AlSb barrier thickness was increased from 100 A to 200 A for 0.6 um epilayer
samples. This did not agree with the model established by Simmons et al., which
predicts lower switching currents for thick tunneling barriers [15]. Similar insensitivity
: /
+ §=6. 100 A AlSb , /
S al Barrier ’ ay
Kaneawel /
@ - /
o>) /
© 3r / t
5 /
L /
> t y
Dm 2r 4
Cc 4
Ss r a , 4 Punch Through Model
= 1+ L7 - - - p=5x10"/cem*
dp) Foe -7 Cee p=1x10'/em*
¢) ei - 1 l 1 1 1 J A I
0.2 0.3 0.4 0.5 0.6
Figure 5.5: Effect of p-GaSb epilayer thickness on TSD switching voltage.
thickness for 100 A AISb barrier samples. The predicted results from the punch-
through model are also plotted in the same figure for comparison. It can be seen
that the switching voltage was relatively independent of epilayer thickness and did
not follow the predicted trend of the punch-through model. This is a surprising result
because the device was expected to work in the punch-through mode due to the low
background doping level of the p-GaSb epilayer. The switching could not be from
the avalanche initiated process because it would have required a switching voltage
of at least 10 V according to equation 5.2. This leaves the distinct possibility that
the switching action was initiated by holes that tunnel across the AlSb barrier as the
Fermi level in the metal electrode drops below the valence band edge of the p-GaSb
of elections at the GaSb/AISb interface. This means that the field in the AlSb
drop across the AlSb barrier to shift the Fermi level of the metal electrode to
voltage did not change much as the AlSb barrier thickness was increased from 100 A
to 200 A. However, this may be explained by the fact that a larger bias was dropped
3.0F 500 A AlSb
After 0.6 um p-GaSb
2.5 > Break- 150 um square mesa
Fr §=6down
20+
=|
aw t5F
Cc .
c= 1.0 V preakdown =6.2V
ue)
O r Breakdown | nreakdown = 0-4 MA
0.5 -
0.0 F Before Breakdown
Voltage [V]
barriers.
As shown in Fig. 5.6, the device instead remained in a high impedance state until
characteristics were pn diode like, indicating that the break down occurred in the
AlSb barrier. This was as expected since most the bias was dropped across the
shifted slightly upward when the device was current stressed [16]. As shown in Fig. 5.7,
similar behavior was observed for the antimonide device. There was a visible shift in
the low impedance branch as the device current was increased. The antimonide TSD
physically broke down when the current density reached 10° A/cm?. Similar to thick
barrier devices, post breakdown I-V characteristic was pn diode like, indicating that
the break down occured in the AlSb layer and most of the bias was dropped across
— 7-5 mA 0.6 um p-GaSb
SOP --o-° 10 mA 67 um Mesa
— 40+ Breakdown
<=
c 30F
®O
5 207
10F
0 _ -
; _Before Breakdown,
| 0.6 um p-GaSb
Bb 67 um circular mesa
< Before Stress }
£ ©) _ ~~ after Stress '
= 4
® 47 f
Sew
0 a7 V witch =1.2V ] V witch =2.95V
i witch =0.2mA H | ewitch =1.1mA
OF a
| , | ; . |
0 { 5 ;
Figure 5.8: TSD I-V characteristics before and after current stressing.
following significant current stressing. As listed in Table. 5.1, this was observed
for all device configurations that yielded switching. Figure 5.8 shows the typical
I-V characteristics before and after current stressing. Unlike the gradual “burn in”
process observed in Si devices, the switching voltages of the antimonide device tended
to change abruptly during current stress and clustered around a fixed value (1.2 V).
This is illustrated Fig. 5.9 where the virgin and after stress switching points of a
number of devices are plotted. It is as if there are two low impedance branches
in the I-V curve with two sets of switching voltages and currents. Most switching
devices exhibited the branch with the higher switching voltage and current when first
examined and moved to the other branch following current stressing. However, a
few devices started out on the branch with the lower set. of switching voltage and
current. What is most interesting is that it was possible to “hop” the I-V curve from
x 5+ @% 38um 0.6 um p-GaSb Epilayer
= ® 53um
— 44 ® 67um *K
Cc
= 3p
= 2K @
1+ 3
F ob oKK ®
l 1 i 1 | ry l 1 l i l rl
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
(a) Switching Voltage [V]
Mesa Diameter 200 A AlSb Barrier
a 5b * 38um 0.6 um p-GaSb Barrier
tt @ 67um
pat
Cc
6 @
m 2P
6 iF
= 83% tx,
OF
1 i 1 i t L n l i i h l ri l
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
~ee”
yum p-GaSb epilayer. (b) 200 A AlSb barrier, 0.6 zm p-GaSb epilayer.
0.6 um p-GaSb
— 20+
aw t5F
Cc .
co)
<= 1.0Fr
=)
O 0.5 — V switch = 1.2 Vv
0.0F
it iT i rl i 4 l rn H i 1 j
05 00 O85 1.0 1.5 2.0 2.5 3.0 3.5
= 6F V witch =2.7V
_ | witen = 4° MA
5 4t
Senn
a)
O 2+
high branch to low branch. (b) From low branch to high branch.
I-V curves were observed to move back and forth between the two low impedance
branches. This is illustrated by the double I-V sweeps in Fig. 5.10 where the current
was swept up to a maximum value and then swept down back to the origin. In
Fig. 5.10(a), the device was in the high branch of the I-V curve on the upward sweep
and was induced to the low branch at a current of 1.5 mA. In Fig. 5.10(b), the device
started out in the low branch on the upward sweep and abruptly moved over to the
high branch when the current reached 7 mA.
switching process initiated by holes that tunnel across the AlSb barrier. We speculate
that a deep level in AlSb was activated and deactivated by the large current and acted
as a sink for the accumulation charge at the AlSb/GaSb interface. When the deep
level was activated, charges can build up in the AlSb barrier layer, causing the electric
field in the AlSb barrier to assume a sharper profile and a larger portion of the applied
bias to drop across the barrier. This will result in a greater shift downward of the
electrode Fermi level relative to the p-GaSb valence band edge, causing a decrease in
computer model was used to simulate the I-V characteristics of device structures with
different AlSb barriers. The simulation was previously developed by E. S. Daniel et
al. for study of Si TSD devices [13, 17]. It was based on the Poisson and the
carrier continuity equations [18] with boundary conditions modified to account for
{|
q is the electron charge, € is the semiconductor dielectric constant, C is the net
concentration of ionized dopants (Nj — Nz), J, and Jp are the electron and hole
current densities, R is the net recombination rate, D, and D, are the electron and
hole diffusion constants, and jt, and ji, are the electron and hole nobilities. The
boundary conditions at the surface of the barrier are determined by the generation-
en . J, = qhs + Jp calc (5.9)
generation-recombination rate and Jp caic and Jp cate are the calculated electron and
hole tunneling currents respectively [13].
the Scharfetter-Gummel scheme [18, 19, 20] and the resulting set of nonlinear alge-
braic equations were solved iteratively by using the Newton-Raphson method. The
multi-valued nature of the TSD I-V curve contributed much to the complexity of
the simulation. In the negative differential resistance region, the device presented
two stable current values for a fixed voltage, hence the convergence of the simulation
to a particular solution was strongly dependent on the initial condition. The high
impedance branch was obtained by adapting the conventional method where the pre-
vious voltage solution was the starting iteration point for the present voltage. The
low impedance state was accessed by using a higher carrier life time, which favored
A e
i 10 aa ak o° a”
zt 10°F a“ ‘"
> p Sogood000000000
= 10° pA 100A 8 aoe"
Cc au
/ Q00 |
GQ 10' 00° hs mnnnnnsoce
~ I y
S 10°F m = 500A AlSb Barrier
g A
= 20A/" ©
10°F
O =
and different AlSb thicknesses. Bistable states were obtained for 100 A and 200 A
barrier thicknesses.
towards a point on the low impedance branch. This point was subsequently used as
the starting point to iteratively generate the rest of the low impedance branch.
Figure 5.11 shows the simulated I-V characteristics of TSD devices with AlSb
barriers ranging from 50 A to 500 A. It can be seen that only the low impedance
state existed for the 50 A AlSb barrier device whereas the 500 A barrier device
yielded only the high impedance state. Dual impedance state behavior was obtained
for AlSb barrier thicknesses of 100 A and 200 A, which agreed with experimental
5.7 Summary and Conclusion
diode in the antimonide system. Successful TSD operation was found to depend
critically on the AlSb barrier thickness. Thyristor like, “S” shaped I-V curves were
obtained for structures with AlSb barriers less than 300 A thick, as was predicted
by the drift diffusion simulation model. The switching voltage and currents of the
working devices had a weak dependence on the barrier and epilayer thickness and
did not agree with the punch-through or avalanche model of operation. Based on
the experimental observations, we propose that switching in the antimonide device is
initiated by holes that tunnel across the AlSb barrier as the Fermi level in the metal
electrode drops below the valence band edge of the p-GaSb epilayer. This switching
mechanism differs from that of the Si device and is a result of the thicker tunnel
barrier and narrower bandgap of the epilayer employed in the antimonide device.
Due to the relative novelty of the TSD device and the immaturity of the anti-
monide materials, the new device exhibited a number of unexpected behaviors. It
was found that the I-V curves can be significantly altered by current stressing. There
appeared to be two branches of the low impedance states with well defined switching
voltages. The device can “hop” between the two sets of -V curves when subjected to
high levels of current stress. We speculate that this was due to charging and discharg-
ing of a deep level in the AlSb barrier. Such non-idealities indicate the importance of
developing a better understanding of the AlSb barrier and the AlSb/GaSb interface
(2) H. Kroger and H. A. R. Wegener, Appl. Phys. Lett. 27, 303 (1975).
[3] T. Yamamoto and M. Morimoto, Appl. Phys. Lett. 20, 269 (1972).
(1976).
Devices 42, 2246 (1995).
(1997).
[9] H. J. Levy, Thesis, California Institute of Technology, 1995.
[10] H. J. Levy and T. C. McGill, U.S. Patent # 5,535,156, 1996.
(1989).
Subramaniam, P.-M. Lei, and G. H. Bernstein, INEE Electron Device Lett. 17,
69 (1996).
[14] S. M. Sze, Physics of Semiconductor Devices, John Wiley Sons. Inc., New York,
1981.
IEEE Trans. Electron Devices 45, 286 1998.
[19] W. L. Engl and H. K. Dirks and B. Meinerzhagen, Proc. IEEE 71, 10 (1983).
Microscopy Study of III-V
Microscopy Theory and Experiment
face properties of heterostructures play important roles in device research. In the
avalanche photodiode work, the band offset at the pn interface was critical to dark
current suppression. In the antimonide TSD work, device switching was strongly in-
fluenced by the band structure of AlSb barrier and surface states at the AlSb/GaSb
interface. These critical properties are usually obtained from characterization tech-
niques such as X-ray photo electron spectroscopy (XPS), photo electric, I-V, and C-V
measurements. The drawback of these conventional approaches is that the results are
laterally averaged over the whole interface or at least over macroscopic dimensions
of more than a few zm. In Part II of this thesis, we describe local probing of inter-
face properties by using ballistic electron emission microscopy (BEEM). This chapter
serves as a short introduction and covers the basic theory and some important exper-
imental issues relevant to the BEEM technique. The BEEM experimental apparatus
based on the scanning tunneling microscopy (STM) setup [1, 2}. BEEM samples
while a new collector terminal at the back of the sample collects elections that leak
into the semiconductor. At low tip bias, no BEEM collector current is observed
because electrons injected from the tip do not have enough energy to overcome the
potential barrier at the base collector interface. As the tunneling voltage increases
and the STM tip potential rises above the conduction band edge of the underlying
semiconductor, electrons can travel ballistically across the thin base region and enter
is held at a high enough bias for hot electron injection into the semiconductor, an
image of the buried interface can be acquired from the BEEM collector signal while
the STM image of the surface topography is being generated. Since the collector
signal is dependent on hot electron transport into the underlying semiconductor,
transport non-uniformities at the buried interface is directly imaged. Alternatively,
the transport properties of a local area can be examined in detail by holding the STM
tip stationary and observing the variation in BEEM collector current as the STM tip
voltage is increased. This is known as BEEM spectroscopy. During this process, no
external bias is applied between the collector and base terminals. The STM base
current is also held constant by changing the tip sample separation as the tip bias
is increased, which ensures that any change in the small collector signal reflects hot
who are the original inventors of the technique. They argued that the transverse
momentum ky, of the electron should be conserved at the base collector interface
in the absence of scattering. As shown in Fig. 6.2, this causes refraction of the
electrons at the base collector interface, which gives rise to a critical angle of entry
for propagation into the semiconductor. At bias voltage V, the maximum kinetic
energy of the incident electron is eV + Ep in the base layer and reduces to eV — eVo
in the collector layer (Fig. 6.1). Since ky is the largest when the transverse component
n’Q, — I .
a k? meV+Epr (6-1)
ergy Ep of the metal base layer, ©. is usually quite small and only electrons with
small transverse momenta in the base may be collected. Thus any scattering in the
base merely reduces the number of electrons collected and does not affect resolution.
This focusing effect is the basis for the high lateral resolution capability of BEEM
(Metal) (Semiconductor)
electrons with energy higher than the band edge of the semiconductor and transverse
momentum within the maximum allowed by ky conservation [2]. If we assume that
these “hot” electrons follow the Fermi distribution with the Fermi level at V + Ep,
and use a step function to approximate for the Fermi function, the expression for the
=cf ae" ae (6.2)
Je J0
energy given by Ey = ae and E
I m
for voltages just above the turn on threshold Vo.
BEEM spectroscopy data. The approach has proved to be successful in a number of
systems including metal on Si [1, 4, 5], GaAs [2, 6, 7], and GaP [8]. Note that for
conduction band minima oriented about an off axis such as the L point in GaAs grown
on (100) substrate, or the X point in Si grown on (111) substrate, the acceptance cone
is centered at a large angle from the normal. Since the tunneling current is sharply
peaked in the forward direction [9] and the acceptance angle is small, no BEEM turn
on should result from these band edges unless there is significant scattering in the
base layer. In reality, parabolic turn on’s due to these off axis band edges can be
observed [5], which indicates that the orientations of the hot electrons in the base
layer are sufficiently randomized due to inelastic scattering [10]. The magnitude of
these turn on’s are reduced compared to the on axis conduction minima since fewer
sitivity and local probing nature. However, In order to measure the minute BEEM
collector signal, the sample must satisfy a number of stringent requirements [11]. This
is illustrated in Fig. 6.3 which shows the band diagram and circuit model of a typical
BEEM sample.
is often a problem for narrow bandgap semiconductors with small Schottky barrier
heights or leaky heterostructures. The large resistance is needed to suppress collector
noise current from micro volt fluctuations across the junction. Note that it is the zero
semiconductor holder
increase Ry.
As shown in Fig. 6.3, the incoming collector current J, may flow to ground via the
base/collector interface (R,) or the back contact (Rez). Since only those electrons
flowing through the back contact are detected, the BEEM signal will be reduced or
not even detectable if Ry is too large compared to Ry.
examined in detail in the next chapter to clarify Al,Ga,_,As band structure ambigu-
ities and verify the effectiveness of the BEEM technique. In Chapter 8, we attempt
to apply BEEM techniques to antimonide structures which only partially satisfy the
constraints outlined above. As will be shown later, this resulted in large noise in the
6.3 BEEM Experiment
van i
‘ nA ) nae '
Base Contact | “ye
8 |
Tunnel Curent ' V7
Ground
(oa)
by Rob Miles [12]. As shown in Fig. 6.4, the set-up was based on a Digital Instruments
scanning tunneling microscope unit (Nanoscope IIT) and configured to operate in air
at room temperature. The original microscope head was modified to ground the
connections differently. The STM pre-amplifier had to be rebuilt so that the sample
instead of the tip was grounded. The stock sample mount was also replaced by a
custom made unit. In the new configuration, a fine Au wire was spring mounted
against the top of the sample as the STM base contact. The sample was proxied
to a copper plate via conductive silver paint. Since all BEEM samples were grown
by molecular beam epitaxy on indium bonded blocks, the BEEM back contact was
Data tupe Height Data type Aux A
Z range 20.0 nm Z range 0.0500 U
427 picoammeter before being fed to the stock digital signal processing unit. Digital
Instrument software was used to analyze the data and maintain control of the STM
speed. If the scan speed is too high, the STM tip will be momentarily too close to
the sample when the sample surface changes abruptly, resulting in a large BEEM
signal. ‘This is illustrated in Fig. 6.5, where BEEM image can be seen to echo the
grain boundaries in the corresponding STM image. Such close correlation between
the surface morphology and the measured signal is a tell tale sign of imaging artifact
in scanning probe microscopy. To reduce this effect, it is important to have samples
oo Oo oOo | i,
high tip voltages and currents.
surface for an extended period of time. As shown in Fig. 6.6, this sometimes resulted
in modification of the surface, which rendered the data unusable. We speculate that
this was due to tip heating and increased tip/sample interaction under high current
and voltage conditions. To prevent this from happening, it was found that the STM
current and voltage must be kept below 10 nA and 2.5 V, respectively.
tip drift even after the system had been given hours to equilibrate. To preserve spa-
tial resolution, it is necessary to use a voltage ramp as fast as possible during BEEM
spectroscopy. However, the system response time is limited by the junction capaci-
tance of the BEEM device (see Fig. 6.3) and the integration time of the picoammeter.
In practice, the ramp speed was increased as much as possible without inducing any
change in the observed BEEM threshold. |
=——_.—__* “——n,
2,
O 0.6 -
© 04F-
on
LL 0.25F
LL
mM
0.0 1 1 i 1 ! 1. i 1 it
2 4 6 8 10
(a) Tunnel Current [nA]
s 200
c=" |
o 150 -
se)
5 |
ey 100 - _
= Linear Fit
5 n Slope = 18
| 50F Y intercept = 0.28
2 0 ry 1 rn l 1 l 1 | n H
Lu 6) 2 4 6 8 10
faa)
current on BEEM threshold. (b) Variation in the magnitude of BEEM turn on with
STM tip current.
BEEM structures studied. We obtained a value of 0.77 eV for the Au on Si Schottky
barrier height, in agreement with the established value [1, 4]. As shown in Fig. 6.7,
the BEEM threshold was independent of the STM tip current, and the magnitude of
the collector turn on was proportional to the tip current, indicating that the results
[2] L. D. Bell and W. J. Kaiser, Phys. Rev. Lett. 61, 2368 (1988).
55, 780 (1989).
Vac. Sci. Technol. B 12, 2422 (1994).
D. A. Woolf, J. Vac. Sci. Technol. B 9, 581 (1991).
and R. H. Williams, J. Vac. Sci. Technol. B 9, 2219 (1991).
[9] J. G. Simmons, J. Appl. Phys. 34, 1793 (1963).
[10] L. J. Schowalter and E. Y. Lee, Phys. Rev. B 43, 9308 (1991).
[11] M. Prietsch, Physics Reports 253 163 (1995).
the case study of Al,Ga,_,As heterostructures. In particular, the study focuses on
extraction of local Al,Ga,_,As Schottky barrier heights and band edges from BEEM
spectroscopy measurement. The results are compared with existing data from other
measurement techniques, which serves to clarify uncertainties in previous findings. In
been extensively studied. In particular, parameters of the Al,Ga,_,As band structure
have been determined from a variety of measurements, including photo response [1],
optical transmission and photo luminescence [2], variation of Hall electron concen-
tration with temperature [3], and variation of electrical conductivity with tempera-
ture [4]. However, there is some uncertainty about the exact positions of [, L, and X
band edges, especially at high Al concentrations where the bandgap of the material
becomes indirect. Figure 7.1(a) shows the measured band edge shift with Al concen-
tration from these studies. The band structure of GaAs is also shown in Fig. 7.1(b)
for reference. It can be seen that band edge positions obtained from different mea-
surement techniques can differ by as much as 150 meV. In addition, there is a lack
of consensus about the Au/Al,Ga,_,As Schottky barrier height at different Al con-
centrations [5, 6], which can be partly attributed to the various sample preparation
procedures available.
—---- Photo Response Results
Conductivity Results
fo)
0.0 0.2 0.4 0.6 0.8 1.0
s 7 \ cower
VALLEY
Band structure of GaAs.
the buried interface, but also allow extraction of the local Schottky barrier height and
band edge positions from spectroscopy data. By varying the Al concentration in the
Al,Ga,_,As BEEM sample, the Schottky barrier height and higher lying band edges
of the material can be mapped out in both the direct and indirect bandgap regime. Of
particular interest are the higher lying conduction band edges in Al,Ga,_,As. Unlike
the metal on semiconductor Schottky barrier height, the position of the higher lying
band edge relative to the bottom of the conduction band is an inherent property of
the material. It should not depend on interface chemistry and should also exhibit less
local variation. Such information as measured by BEEM can be directly compared
with results obtained from other techniques. Hence the Al,Ga,_,As BEEM study
should clarify previous uncertainties about the various band edge positions and serve
lized ex situ with a sputter deposition tool. Figure 7.2 shows the structure and band
diagram of a typical sample. Highly doped (n=1 x 10'8/cm?) epi-ready GaAs (100)
wafers were used as substrates to ensure that the back contact is conductive enough
to support BEEM current. Following oxide desorption, a buffer layer was grown to
grade the doping profile down to the level of the unintentionally doped Al,Ga,_,As
epilayer. The doping in the buffer layer started out at n=1 x 10'°/cm? for the first
0.2 wm and was gradually tapered down to n=2 x 10!°/cm? over the next 0.1 jum.
At the end of buffer growth, samples were soaked in As for 30 seconds, yielding the
(2 x 4) reflection high energy electron diffraction (RHEED) pattern characteristic of
reconstructed GaAs surface. An unintentionally doped Al,Ga,_,As layer was grown
on top of the smoothed GaAs surface. The epilayers were kept thin to support trans-
port of BEEM current, and the doping level was kept low in the epilayer and the
buffer layer immediately below it to reduce effect of band bending.
Capping Layer (Optional)
(Unintentionally Doped)
(n=1x10"cm® to 2x10" cm*)
L-pt 1.36 eV
p 120A
sample with AlAs epilayer.
The Al,Ga;_,As layer was 200 A thick for samples with low Al content (x < 0.5) and
100 A thick for samples with high Al content. At z = 0.25, samples with epilayers of
both thicknesses were grown to examine its effect on BEEM turn on threshold. At
the end of the growth, samples with high Al content were capped off by either a 50
Composition Thickness [A]
Alo.11Gag.g9As 200 none
Alop.19Gao.g1 As 200 none
Alo.o5Gag.75.As 100 hone
Alo.25Gag.75As 200 none
Alo.25Gag.75.As 900 hone
Alo.59Gao.50As 100 none
Alo s9Gao.50As 100 GaAs
Alp s0Gao5oAs 500 none
Alo.s50Gao.50As 500 InAs
Alo.g90Gao.29 As 100 GaAs
AlAs 100 GaAs
AlAs 100 InAs
was sputtered off a solid target by Ar plasma and deposited onto the sample at a
rate of 0.7 A/sec. Samples were placed behind a mask and patterned with arrays
of Au dots 1 mm in diameter. As discussed in Section 6.2.3, the BEEM device
should be as small as possible to reduce the background noise current. It was found
engagement restrictions. The Au layer thickness was monitored by a crystal oscillator
and maintained at 100 A for all samples. Atomic force microscopy (AFM) studies
showed that the typical metal layer had a rms roughness on the order of 5 A. For
to have little effect on BEEM results, indicating that sputter damage at the base
collector layer interface was not a significant factor.
and exposed to the ambient. Hence a 20 - 30 A thick native oxide was present between
the metal and semiconductor layer. However, it has been shown that the oxide layer
does not affect BEEM results for Au/GaAs structures [7]. In fact, samples with native
oxide layers support BEEM current more consistently over a longer period of time [7].
In our study, it was found that samples with native oxide layers were stable for up to
several months. To minimize contamination from handling, a degreasing procedure
was followed before the sample was introduced to the metallization chamber. It
consisted of sequential ultra sonic rinse in trichloromethane, acetone, isopropanol and
de-ionized water, with each rinse lasting 2 minutes. The procedure helped generate
more consistent BEEM results, especially for samples that have been stored in air for
chapter (Section 6.3.1). Figure 7.3 shows BEEM images of the Alp.1;Gag.g9As sample
on two different scales. In both cases, the tip bias was at 1.2 V, which was above
the BEEM turn on threshold. It can be seen that the BEEM images bear little
resemblance to the corresponding STM topography, which was clear evidence that
the BEEM images were decoupled from their STM counterparts and were free of
scan parameter induced artifacts. Figure 7.3(a) was typical in that the BEEM image
of the buried surface appeared to be smooth and relatively featureless, which was a
reflection of the high growth quality of the sample and the uniformity of the buried
interface. In Fig. 7.3(b), the BEEM current over a large patch of area appeared to be
suppressed even though the surface topography was relatively uniform, which clearly
illustrates the effectiveness of using BEEM to image the buried interface.
Data type Height Data type Aux A
Z range 10.0 nw zZ range 0.100 ¥
Data type Height Data type Aux A
Z range 10.0 nea 2 range 0.100 Y
scan. (b) 500 nm by 500 nm sean.
first picture to 2.2 V in the last picture. Scan area was 100 nm by 100 nm.
creased to beyond the BEEM turn on threshold, the BEEM signal picked up and the
image became brighter. In essence, this was similar to BEEM spectroscopy measure-
ment except that an entire area instead of a single spot was examined. Due to tip
drift, the images are slightly shifted from frame to frame. In the series of images
shown, features can be seen to move to the right. If the data from different frames
can be linked to each other through feature tracking software, then the BEEM I-V
characteristics of a large number of positions can be simultaneously analyzed. This is
potentially a very powerful technique as it will result in a mapping of BEEM threshold
(Schottky barrier height) over a large area.
(BEEM turn on at different scoles/Tunnel current = 4 nA)
© 100x100 scan area
A 10x10 nm scon area
40.0 | # 2x2 nm scan oreo TH
Sl r RITE ea|
5 qt poet
3 yoit at
3 20.0} Fat zat qiytt
8 + $4 ze 4
cXUXXr? get +t
voters Et pelt
=a
+e eee eet tf
0.0 1
Tunnel Voltege [V]
study. A shot gun alternative was to plot the average BEEM current over the whole
different sizes. The error bars in the plot represented the spread in BEEM signal over
the whole image area. It can be seen that the results from the 10 nm and 100 nm scan
area were similar to each other, whereas the 2 nm scan area yielded a significantly
higher BEEM threshold. The greater variation in BEEM threshold for small scan
areas may be genuine, but it may also be due to more severe tip drift in the smaller
area scan. Regardless the origin of this difference, we can draw the conclusion that
the local BEEM threshold (Schottky barrier height) appeared to be uniform for these
local band structure of the sample is more accurately determined by using BEEM
spectroscopy. In this process, the STM tip is held at a fixed position on the sample
were averaged over 50 voltage ramps to improve the signal to noise ratio. Typically,
the tip drifted 5 to 10 nm during the 10 minutes it took to complete the 50 ramps.
Thus the spatial resolution of BEEM spectroscopy was drift limited and about an
order of magnitude higher than the theoretical limit [14]. The resulting BEEM I-V
curve should be considered an average over the same area.
® 60+
- 40F
iS}
ro 20-
Ti of
Lu L rl 1 rn L 1 1 L
co 0.0 0.5 1.0 1.5 2.0
(a) Tunnel Voltage [V]
> 100}
2;
= 50+
OD
0.0 0.5 1.0 1.5 2.0
traction of BEEM threshold from differentiated curve.
was above the corresponding threshold. As shown in Fig. 7.6(b), differentiating the
BEEM I-V curve generated a piece wise linear curve that clearly revealed the multi-
threshold nature of the turn on. The I-V curve shown in Fig. 7.6 was obtained from
an AlgiiGapg9As sample. Three thresholds were extracted. According to Fig. 7.1,
they corresponded to the [, L and X points in the Alg11;Gag.g9As layer. Other sam-
ples with low Al concentrations (x = 0,0.11,0.19, 0.25) produced similar BEEM I-V
curves. However, the third threshold, attributed to the X point, was not always
evident in every run. This may be due to the comparatively large effective mass
of the X valley [8], which tended to weaken the corresponding BEEM turn on (see
equation 6.3). Samples with high Al concentrations (2 = 0.50, 0.80, 1) also produced
BEEM I-V curves that had robust two threshold fits. The thresholds, however, were
attributed to the X and L points. For all samples, it was found that the parabolic turn
on model broke down at about 0.5 V above the first threshold, which was expected
epilayer thickness variation beyond 100 A did not significantly affect the measured
BEEM thresholds. ‘Thus the measured band edges may be considered bulk properties
of Al,Ga,_,As. This result agreed with findings from the Au/AlAs study by Kaiser
et al., who showed that most of the thickness induced threshold shift occured over
the first few monolayers of the semiconductor [10].
Al concentrations, the capping layer had only a slight effect on the BEEM thresh-
olds. One may expect that a InAs capping layer will lower the apparent Schottky
barrier height due to the negative Schottky barrier of InAs and band bending at the
InAs/Al,Ga,_,As interface. The absence of this effect in our sample is attributed to
relaxation at the InAs/Al,Ga,_,As interface. Due to the large lattice mismatch be-
Secnd BEEM Threshold
g——# ~s
2 1.07 First BEEM Threshold (SB Height)
io)
Sawn
‘i O.5+
I 1 i 1 I 1 l 1 l
Al, o5G@,7,As Layer Thickness [A]
relaxed to its natural lattice template, resulting in a large number of dislocations and
dangling bonds at the capping layer/Al,Ga,_,As interface, pinning the Fermi level
to the middle of the indirect bandgap. It should be noted that the capping layer
did help to prevent deterioration of the Al,Ga,_,As layer. The uncapped sample in
Fig. 7.8 supported BEEM current for only a few days whereas the capped samples
The variation of these band edges with Al concentration x is shown in Fig. 7.9.
The extent to which the parabolic model remained valid is also plotted in the same
figure. Each data point represents 20 to 30 runs. Since results for Alp s9Gao 5 As
a }
Ss 1.0 7
@ First BEEM Threshold (SB Height)
©)
Sa
Lo 0.55F
(a) Aly so Gag oS
0.0 L i |
nocap 50 A GaAs cap 70 A InAs cap
(a) Capping Layer
1.55 Second BEEM Threshold
— First BEEM Threshold (SB Height)
1.0 F
2,
Pa
©)
Sewee
Lu 0.5 -
(b) AlAs
i |
50 A GaAs 70 A InAs
(b) Capping Layer
(b) AlAs sample.
view the energy position curves as continua. It can be seen that the Au Schottky
barrier height increased with Al concentration x until the semiconductor changed
from direct bandgap to indirect bandgap (x > 0.45). At higher Al concentrations,
the Au Schottky barrier height stayed almost constant. The measured Schottky
Turn on Model i _
ad ae on eer SO
ee ee ee
> —_ ; se nn ~
® 1.0+ . .
— First Band Edge (SB Height)
Pa)
©)
Sam
i OST
Hollow Symbol: No Capping Layer
+ Symbol: 50 A GaAs Cap
Solid Symbol: 70 A InAs Cap
0.0 i] 1 i L 1 n ! 1 i rt L
0.0 0.2 0.4 0.6 0.8 1.0
data points at x = 0.50 and x = 1.0 are slightly offset for clarity.
helpful to plot the implied p-type Schottky barrier height, which is obtained by sub-
tracting the n-type Schottky barrier height from the semiconductor bandgap. Plotted
on the same graph as the bandgap, it reveals the position of the surface Fermi level
relative to the valence band edge [12]. As shown in Fig. 7.10, the surface Fermi level
stayed nearly constant at about 0.6 eV from the top of the valence band for x < 0.4.
This agreed with the common anion rule [13] which states that the position of the
Fermi level relative to the valence band edge in III-V and TI-VI compound should
only depend on the anion involved. As the material changes from direct bandgap to
2.0 -
Al.Ga, As Bandgap
Direct
1.55
eW""""
oO Mid Gap Position
> 1.0 eer r gt te
o eco se
@ x p-Type SB Height (Fermi Level
Lu O.5F ~ Relative to the Top of Valence Band)
| Hollow Symbol: No Capping Layer
+ Symbol: 50 A GaAs Cap
0.0 - Solid Symbol: 70 A InAs Cap
L a l n | i 1 4 1 n H
Al Concentration
points at 2 = 0.50 and x = 1.0 are slightly offset for clarity.
level pinning from surface states created by the additional capping layer in these
Al,.Ga,_,ASs
of sample preparation. For example, BEEM workers have obtained Au on GaAs
Schottky barrier heights that range from 0.82 eV to 0.90 eV [14, 15]. The position
of the higher lying band edge relative to the first band edge, however, is an intrinsic
property of the semiconductor and should be independent of processing. Figure 7.11
shows the variation of this band edge difference with Al composition x7. The plot is
derived from the BEEM threshold data in Fig. 7.9 by subtracting the first threshold
individual run. Note that the error bars in Fig. 7.11 are smaller than the sum of
threshold error bars in Fig. 7.9, which shows that the band edge difference was an
> ost t+ Symbol: 50 A GaAs Cap Solid Curves:
oO. ° Solid Symbol: 70 A InAs Cap Results Obtained
rad) by Lee et al.
oO r
Cc 0.6 F a Dashed Curves:
@ \ Limit of Parabolic
Oo Turn on Model
= 044
Q .
0.2 5+
LL
ze)
aa] 0.07 i l i : n ! i
Al Concentration
Multiple data points at 2 = 0.50 and x = 1.0 are slightly offset for clarity.
data obtained from other techniques [1, 2, 3, 4]. It was found that the BEEM results
agreed best with the conductivity findings by Lee et al., which are plotted in Fig. 7.11
for reference. For 7 < 0.45, where Al,Ga,_,As is a direct bandgap material, it can
be seen that the BEEM threshold difference tracked well with the difference between
the L and [ points as obtained by Lee et al. For 2 > 0.45, where Al,Ga,_,As is an
indirect bandgap material, there was more scatter in the data but BEEM threshold
difference agreed well with the difference between the L and X points. Since the
L point lies at an off angle from the (100) normal growth direction, its presence in
BEEM threshold analysis indicates that there was significant scattering before the
electrons reached the metal semiconductor interface [16]. The range over which the
the [’ point was out of range at high Al concentrations but the X point should have
been observed as the third BEEM threshold at low Al concentrations. In fact, the X
point turn on was present in some runs. However, it was a weak turn on due to the
large effective mass of the X valley, and the results were not consistent enough for
systematic analysis.
points. Each error bar was the result of 20 to 30 local measurements. The size of the
error bar ranged from 30 to 50 meV and was consistent with the level of uniformity
observed in corresponding BEEM images (Section 7.4). Tip drift limited our spec-
troscopy resolution to about 5 to 10 nm, which was an order of magnitude higher
than the theoretical limit [14]. This may have resulted in less measured variation
BEEM images were readily obtained and its capability for revealing sub-surface non-
uniformities demonstrated. In addition, BEEM spectroscopy was used to map out
the Schottky barrier height and the higher lying band edges in Al,Ga,_,As as the Al
concentration x was varied. It was found that the indirect band edge (L point) con-
tributed significantly towards the BEEM signal, which indicates that scattering was
significant in the base region of the sample. Moreover, the relative positions of the
higher lying band edges in Al,Ga,_,As were extracted from the BEEM spectroscopy
data. Since these relative energy positions are intrinsic to the material and indepen-
dent of sample preparation detail, the measurements represented direct probing of
the semiconductor band structure. Comparison with the existing data showed good
agreement between the BEEM measured values and the conductivity findings by Lee
et al. [4]. The study clarifies previous uncertainties about the Al,Ga,_,As energy
positions and demonstrates that BEEM is a effective tool for probing semiconductor
[2] B. Monemar, K. K. Shih, and G. D. Pettit, J. Appl. Phys. 47, 2604 (1976).
[3] A. K. Saxena, Phys. Stat. Solid. (B) 105, 777 (1981).
21, 659 (1980).
Phys. Semicond. 6, 398 (1972).
and K. L. Kavanagh, Appl. Phys. Lett. 62, 2965 (1993).
Thun, Academic Press, New York, 1967.
Davis, Phys. Rev. B 48, 18 324 (1993).
murti, J. Vac. Sci. Technol. B 12, 2625 (1994).
Press, Oxford, 1988.
(1976).
D. A. Woolf, J. Vac. Sci. Technol. B 9, 581 (1991).
InAs/AISb Superlattice
system. ‘T'wo device relevant structures are studied: AlSb barrier and selectively
dope InAs/AISb superlattice. Due to the large background noise in these structures,
BEEM images are not obtained. Instead, the study focuses on extraction of band
structure and transport characteristics from BEEM spectroscopy measurement. In
the AlSb case study, the impact of sample structure on BEEM background current is
nological importance of the system [2, 3, 4, 5] and the unique properties of the
various constituent materials. Compared to well known systems such as Au/Si
and Au/Al,Ga;_,As, the antimonides are distinctly under characterized by BEEM.
To date, there have been few published BEEM results on antimonide heterostruc-
tures [1] Much of this is due to the experimental difficulty associated with the large
BEEM background noise which arises from the type II band alignment and the small
bandgaps of these materials. Because of this, the system remains largely unexplored
and represents a stringent test ground for BEEM. A success here should yield a wealth
of information and leave no doubt about the versatility and capability of the BEEM
technique.
Due to the lack of insulating oxide, AlSb is often used as the barrier in antimonide
device structures. This has been demonstrated in the superlattice avalanche photo-
diode in Chapter 4 and the antimonide tunnel switch diode in Chapter 5. Another
important example of this is the mobility modulated transistor [6] shown in Fig. 8.1,
where AJISb is used as the gate insulator. This device is similar to a regular field effect
transistor but employs a dual channel for conduction: InAs for electrons and GaSb
for holes. Because the mobilities of the channels are vastly different (33000 cm?/V-s
for electrons in InAs and 850 cm?/V-s for holes in GaSb), the conduction between
the source and drain can be rapidly modulated by varying the field in the channel
and changing the coupling between the channel wave functions. The device is in the
on state when the wavefuctions are decoupled and the conduction is dominated by
the fast InAs channel. It’s turned off when the wavefuctions are coupled and the slow
GaSb channel dominates the conduction. While this device is potentially interesting
due to its fast switching speed, real world implementation has been problematic. One
Schottky barrier height of AlSb is lower than its bandgap would suggest.
BEEM spectroscopy should make clear the dominating mode of transport in AlSb,
i.e, whether the indirect X band edge contributes significantly to electron transport.
As discussed in Chapter 2, such information is also highly relevant to the design of
GaSb/AISb superlattice avalanche photodiodes where electron ionization enhance-
t Lg
| SLS
on
oO
— 40 - Superlattice
S Bandgap E,
See
) ~~
c ~~ ~~
LW ost yee
Superlattice Schottky ~ ~~~. _ _
barrier height ® on Au ~~
0.0 i ‘ } L | 1 | 2 ii 1 i
InAs Layer Thickness [A]
barrier height with InAs layer thickness [11].
study because of its heterostructure nature. To date, BEEM has not been used
extensively to probe epilayers with a large number of hetero-interfaces. The properties
of the superlattice and its application in antimonide avalanche photodiode have been
described in detail in Chapters 3 and 4. As shown in Fig. 8.2, the bandgap and
constituent layer thickness [10, 11]. In this study, InAs/AISb superlattices of several
molecular beam epitaxy. As shown in Fig. 8.3, the structure of the antimonide BEEM
sample was similar to its Al,Ga,_,As counterpart. In the early phase of the study, a
few AlSb epilayer samples were grown on highly doped p-type GaSb wafers. As will
be discussed later, this resulted in unacceptably large background BEEM current.
Hence subsequent samples were all grown on highly doped n-type (n=5 x 10'7/cm$
from Te doping) GaSb wafers. The high doping level ensured that the substrate would
be conductive enough to support the collector current in the BEEM experiment (see
Section 6.2.3).
buffer layer was grown. Since the substrate was n-type and the background doping in
the GaSb buffer layer was slightly p-type, the buffer layer was kept as thin as possible
without compromising the growth quality of subsequent layers. At low growth rate,
a 1000 A thick buffer layer was found to be adequate. At the end of the buffer
growth, samples were soaked in Sb, yielding the (1 x 3) reflection high energy electron
diffraction (RHEED) pattern characteristic of reconstructed GaSb surface.
smoothed GaSb surface. The thickness was selected so that bulk properties would
be examined while at the same time the layer was thin enough to support transport
of BEEM current. Because the AlSb layer was relatively thin, substrate temperature
was kept at 520 °C, the same as for GaSb growth. RHEED for the AlSb layer was less
streaky but still exhibited the characteristic 1 x 3 pattern. To prevent AlSb oxidation,
samples were capped off at the end of the growth by either a 50 A GaSb layer or a
100 A InAs layer. Substrate temperature was lowered to 470 °C during growth of the
Doped N-type Superlattice
(Unintentionally Doped)
Substrate (p = 1 x 107 cm* )
the recipe outlined in Chapter 3. The substrate temperature had to be lowered to
prevent excessive As incorporation in the antimonide layers. The structural quality of
the superlattice was significantly improved when the growth temperature was lowered
to 420 °C, at which point the GaSb surface turned Sb rich and the RHEED pattern
changed from 1 x 3 to 1 x 5. During growth of the InAs constituent layer, the
51 dopant cell shutter was opened, and As flux was minimized by using the valved
cracker while maintaining an As stabilized growth front. A 10 second Sb soak was
applied between each InAs and AlSb interface to ensure a InSb like interface, which is
known to produce material of superior quality [12], RHEED pattern remained streaky
throughout the growth and exhibited sharp 2 x 4 and 1 x 3 reconstructions for the
InAs and AlSb layers, respectively. Samples were grown with superlattice periods
thickness of the superlattice was kept constant for all samples at 2400 A. To prevent
oxidation, the superlattice was capped with 50 A of GaSb following completion of the
last AlSb layer.
post growth metalization. Aluminum and gold were sputtered off solid targets by
Ar plasma and deposited onto the sample at rates up to 0.4 A/ sec, which resulted
over from growth served as the back contact. The surface of the front contact metal
layer was comparable to that of Al,Ga,_,As samples, and appeared to be smooth
and suitable for BEEM studies. Prior to metalization, samples were taken out of the
UHV growth environment and exposed to the ambient. Hence the top 20 - 30 A of
the cap layer was oxidized. The native oxide may have stabilized the surface since
the samples yielded the same characteristics after storage of up to several weeks. To
minimize contamination from handling, the samples were subjected to a sequential
to have a large junction resistance across the metal to semiconductor interface at zero
bias so that noise current due to micro-volt fluctuation in the system is suppressed.
This was not a problem in the Al,Ga,_,As study because the Schottky barrier height
of Au/Al,Ga;_,As was relatively large (greater than 0.8 eV). In antimonide BEEM
samples, the junction may be too conductive due to the type II band alignment
and the narrow bandgap of GaSb. To determine the conductive characteristics of
0.1 - /
/ Sample B
—= 0.0 7
fa | -0.1 E , .
~ / -~-- InAs Cap/AlSb Barrier/
5 -0.2- ; Undoped GaSb Substrate
= '
6 03b / GaSb Cap/AISb Barrier/
! n-GaSb Substrate
0.44 |
-0.5 i t | l ! 1
-1.0 -0.5 0.0 0.5 1.0 1.5 2.0
(a) Voltage [V]
500 A 500 A
AlSb
AISb Unintentionally Unintentionally
Doped P-type GaSb Doped P-type GaSb
Buffer Layer and Substrate Buffer Layer
C1 a 50A
GaSb
. Te Doped N-type
100 A rapping GaSb Substrate
InAs ayer
Capping
Layer
Sample A Sample B
was 1 mm. (b) Corresponding band diagr
BEEM experiment, I-V measurements were taken from these samples prior to the
BEEM study.
from an early growth on unintentionally doped p-type GaSb substrate and capped by
an InAs layer. Sample B was grown on n-type GaSb substrate and capped by GaSb,
which was the standard configuration for most AlSb samples and all superlattice
samples. The band diagrams of these samples are shown in Fig. 8.4(b). It can
be seen from Fig. 8.4(a) that sample A was much more conductive than sample B.
This is because the AlSb barrier did not effectively block the tunneling current that
resulted from the type II band alignment between InAs and GaSb. In sample B,
tunneling was reduced due to the absence of InAs layer and the blocking action of
the underlying n-type substrate. When sample A was inserted in the BEEM setup, a
large background BEEM current was observed, completely overwhelming the collector
signal and rendering the device unsuitable for BEEM study. This was not surprising
considering the steep slope of the sample A I-V curve at zero bias. By contrast, the
BEEM background current was much smaller in sample B due to its larger resistance
too unstable for BEEM imaging. However, BEEM spectroscopy was readily obtained
from AlSb samples with the B type configuration. Figure. 8.5(a) shows two such
BEEM I-V curves. Each curve was taken from a different place on the sample surface
and took approximately 10 seconds to generate. Tip drift rates were about a few
nm per minute after the system was given time to equilibrate. Since there was large
variation from one run to another, BEEM scans were not averaged in order to preserve
spatial resolution in the experiment.
[o)
oO
0.0 0.5 1.0 1.5 2.0
5 + Average=1.17 V Y,
Stand. Dev.=0.15 V UY
2 “ry YY MAY
é 7
3p YOU
8 YUU
5 YY UY
Be 7)
: 77/7
5 it YOY
2 UUG7
Q000000
06 08 10 12 14 16 41.8
(b) BEEM Threshold Voltage [V]
constant at 10 nA. (b) BEEM threshold statistics.
I, = S-(V -V;) (8.1)
old voltage. By examining a large number of runs, it was found that the turn on
voltage centered around 1.17 eV with a standard deviation of 0.15 eV. This was
in fair agreement with the result obtained by Walachova et al [1] in their study of
InAs/AISb double barrier heterostructures. The band structure of AlSb is shown in
Fig. 8.6. It can be seen that the BEEM turn on threshold should be attributed to
the conduction band minimum near the AlSb X point, whereas the L and I point
of AlSb lie higher and could not be delineated from the BEEM data. This indicates
that electron transport in AlSb was dominated by the indirect band minima. Hence
the electron barrier height of AlSb was much lower than that given by the I’ point.
This is an important result and has many implications in antimonide device research.
In particular, electron ionization enhancement in GaSb/AISb superlattices depend
critically on the conduction and valance band offset difference between GaSb and
AlSb. The smaller conduction band offset implied by the BEEM result means that
the ionization enhancement effect will be weaker than previously thought.
BEEM I-V curves. The large variation in individual BEEM threshold indicated un-
evenness at the metal semiconductor interface. This is in contrast with BEEM study
of AlAs, where the BEEM turn on voltage exhibited minimum variation across the
wafer (see Section 7.5.4). Since transport across a barrier varies exponentially with
the barrier height, the I-V characteristics over a large area is dominated by regions
with small local barriers. Hence the large spread in the AlSb BEEM threshold was
consistent with the fact that the barrier height in AlSb Schottky structures was often
lower than expected.
Ki
discrepancy to the fact that the background doping was p-type for AlSb and n-type
for AlAs. The increased background BEEM current was accounted for by additional
hole thermionic emission over the smaller hole barrier height of AlSb. The dominance
superlattices. This is because the effective superlattice bandgap is substantially
smaller than that of AlSb, even for samples with a very short period. The smaller
bandgap of the superlattice epilayer led to reduced Schottky barrier height at the
metal to semiconductor interface. As the result, the BEEM noise current was not
nesses of 8 A/9 A, 12 A/12 A, and 24 A/24 A. As shown in Fig. 8.2, the corresponding
bandgaps were 1.2 eV, 1.15 eV, and 0.88 eV, and the expected Schottky barrier height
of Au/superlattice structure were 0.62 eV, 0.56 eV, and 0.32 eV, respectively. The
bandgaps and Schottky barrier heights could be made larger by growing structures
with shorter superlattice period, but the structural quality of the material deterio-
rated rapidly as the superlattice period was decreased. In fact, X-ray rocking curves
for samples with the 8 A/9 A configuration showed multiple splits at the superlattice
peak, indicating that the layer had relaxed from too much strain. The inferior quality
of these samples rendered them unsuitable for BEEM studies. The 24 A/24 A longer
period sample exhibited the best structural integrity but its bandgap and Schottky
barrier heights were too small to keep background BEEM current at a reasonable
level. Thus only the 12 A/ 12 A period samples was deemed suitable for BEEM
superlattice. The sharp X-ray diffraction satellites were indicative of the good struc-
tural quality achieved. The I-V curve of the metalized device is shown in Fig. 8.8
and indicated that the underlying superlattice was n-type. The curve deviated sig-
nificantly from ideal Schottky diode behavior at high voltages. But the low voltage
portion of the curve yielded a Schottky barrier height of 0.6 eV [14].
current noise was on the order of 100 pA, which overwhelmed any conventional BEEM
signal that would be present. Due to the large noise and the associated instability,
BEEM images could not be obtained. However, it was found that after the surface
was stressed by running a high voltage and current (-3 V and 50 nA) through the
STM tip, the metal layer could be deformed resulting in regions where the metal
F 100 Period, 12 A/12 A, GaSb Substrate
40° | InAs/AlSb Superlattice
0)
10°F
Omega Angle [Degree]
superlattice BEEM sample.
BEEM scans after the stress treatment. The threshold occured at around 0.8 eV for
the Au/superlattice system and could be reproduced by retracting the STM tip and
to the large background noise, BEEM images were not obtained. However, BEEM
spectroscopy was applied with various degrees of success to analyze the Schottky
barrier height and band structure of AlSb barriers and InAs/AISb superlattices.
The Al/AISb system yielded a BEEM threshold of 1.17 eV, which was attributed
to transport through the conduction band minimum near the AlSb X point. This
ie)
2,
wo
LJ
al
BEEM samples. (a) Liner plot. (b) Extraction of Schottky barrier height from Log
plot.
sample. Tunnel current was held constant at 10 nA.
AlSb BEEM threshold (0.2 eV) was also observed, indicating degradation of the AlSb
barrier due to local fluctuations at the metal semiconductor interface. The finding is
consistent with the fact that the observed Schottky barrier height is often lower than
expected.
pered by considerable background BEEM current due to the small bandgap and low
Schottky barrier height of the superlattice. The expected shift in BEEM thresh-
old from superlattices of different period could not be observed, and BEEM scans
yielded a threshold of 0.8 eV for the Au/24 A period superlattice system only after
considerable stressing of the metal layer.
ground current associated with the type I] band alignment and small bandgaps of
the surface capping layer and the underlying substrate in the antimonide BEEM
structure. In particular, the junctions formed by InAs cap/thin AlSb barrier/p-GaSb
were found to be especially leaky, whereas the GaSb cap/thin AlSb barrier/n-GaSb
configuration yielded working BEEM samples.
(1] J. Walachova, J. Zelinka, J. Vanis, D. H. H. Chow, J. N. Schulman, S. Karamazov,
M. Cukr, P. Zich, J. Kral, and T. C. McGill, Appl. Phys. Lett. 70, 3588 (1997).
and J. C. Connolly, Appl. Phys. Lett. 66, 1942 (1995).
Phys. Lett. 60, 1921 (1995).
683 (1991).
Subramaniam, P.-M. Lei, and G. H. Berstein, IEEE Electron Device Lett. 17,
69, (1996).
70, 481 (1997).
Phys. Lett. 69, 85 (1996).
71 (1982).
879 (1995).
[12] G. Tuttle, H. Kroemer, and J. H. English, J. Appl. Phys. 67, 3032 (1990).