Development of wide-bandgap II-VI semico

Development of wide-bandgap II-VI semiconductor light-emitting device technology based on the graded injector design - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Development of wide-bandgap II-VI semiconductor light-emitting device technology based on the graded injector design
Citation
Swenberg, Johanes F. N.
(1995)
Development of wide-bandgap II-VI semiconductor light-emitting device technology based on the graded injector design.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/p2ey-5557.
Abstract
This thesis describes the technical development of a novel semiconductor device design aimed at realizing short wavelength visible light emitters. The device structure, called the graded injector, achieves minority carrier injection in a heterojunction system with unfavorable type-II band alignment. Band edge engineering with an alloy graded intermediary layer effectively reduces the conduction band offset and allows for efficient minority carrier injection. The basic device structure consists of a n-CdSe/Mg[subscript x]Cd[subscript 1-x]Se/p-ZnTe heterojunction, where the Mg[subscript x]Cd[subscript 1-x]Se region is graded.
The device design, materials growth, and characterization of II-VI green LEDs based on this structure are presented. Simulations demonstrate the operating principle of the graded injector. Early device development had been hindered by the lack of a p-type dopant for MBE ZnTe and the unavailability of high quality substrates. These restrictions have been overcome with the development of efficient nitrogen p-type doping of ZnTe and the growth capability of high quality heteroepitaxy on GaSb substrates. The materials characterization of the Mg-chalcogenides has enabled more accurate band edge engineering necessary for an operating device.
The advances in growth technology and materials characterization have been incorporated to grow and fabricate working graded injector LEDs. The operating characteristics of these devices unequivocally demonstrate the diode-like operation and efficient minority carrier injection. The electrical and optical performance of these devices will be presented and analyzed.
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:
14 April 1995
Record Number:
CaltechETD:etd-10122007-142152
Persistent URL:
DOI:
10.7907/p2ey-5557
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
4058
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
15 Oct 2007
Last Modified:
16 Apr 2021 22:58
Thesis Files
Preview
PDF (Swenberg_jfn_1995.pdf)
- Final Version
See Usage Policy.
3MB
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
DEVELOPMENT OF WIDE-BANDGAP II-VI
SEMICONDUCTOR LIGHT-EMITTING
DEVICE TECHNOLOGY BASED ON THE
GRADED INJECTOR DESIGN

Thesis by
Johanes F. N. Swenberg

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

California Institute of Technology

Pasadena, California

1995
(Defended April 14, 1995)

my family

ili

Acknowledgments

I will forever be grateful for the unique opportunity of being part of the Caltech
community. More importantly, I am thankful for having been a member of the
McGill group. I wish to express my gratitude for this unique opportunity. Tom
McGill has created a rich and rewarding environment for graduate research, and
personal and professional development. I have grown immeasurably these past
years. His creative teaching, mentoring and friendship have prepared me for an
exciting and uncertain future.

Any successful organization’s greatest asset is its people. This has clearly been
the rule with Tom McGill’s group of individuals he has brought together. It has
been a privilege and pleasure working and interacting with these talented people.
Perhaps the most valuable asset in Tom’s operation has been Marcia Hudson.
Without her the group would soon lose its momentum. Marcia has been a trusted
friend and accomplice.

My early interactions with former students gave me a good foundation for my
subsequent research. Mike Jackson, Dave Chow, and Ed Croke along with other
older students served as good examples to learn from. My earliest experimental
research exposure with Yasantha Rajakarunanayake benefited me in the following
years.

I owe a great deal to Mark Phillips for his guidance and instruction in the
cleanroom and with the II-VI MBE machine. If he succeeded in transferring a

small fraction of his skills, abilities and knowledge to me, I can call myself a great

experimentalist. Many thanks to Doug Collins for his patience trying to instruct
me in the ways of III-V MBE. I owe him many beers. Working the past years with
Mike Wang on the II-VI project has been very rewarding. I have enjoyed our times
in the cleanroom, but I had preferred the good dim sum.

The group as a whole and myself have benefited tremendously from our inter-
actions with senior researchers. I consider myself extremely lucky to have shared
a conversation or heard their stories. Dr. J. McCaldin has been the creative mind
behind the II-VI project. I am constantly amazed at his past and present ac-
complishments. Ogden Marsh has been a great source of ideas and reality checks.
Andy Hunter has helped directly with early Hall experiments.

My interactions with the members of the group have contributed to my accom-
plishments and happiness these past years. Naturally, all the individuals who have
participated in maintaining the computers and network have made this thesis and
daily life possible. I thank you all for your valuable expertise. Dave Ting has been
a great theorist and realist to share a conversation. He and the other theorists
in the group have changed my earlier preconceptions about theorists. Ron Mar-
quardt was a great roommate and labmate in spite of his magnet. He will long
be remembered in the group. Yixin Liu has added a valued cultural and culinary
experience to the group. I am impressed with his uncanny ability to answer all my
questions. Harold Levy has been way cool. Totally incredible mind, dude. Chris
Springfield has brought a complementary artistic creativity to the group each of
us has been enriched by. I am glad I will one day be able to say, “Yes, I know
the director of that film. I went to school with him.” Shaun Kirby was a good
officemate and fellow Republican to talk with. ’m sorry our times at Caltech did
not overlap more. Of course Rob Miles and I have much in common and many
debates for when we disagreed. It is always a pleasure to find a fellow conservative
and patriot. I will miss his comradery. I have truly enjoyed the turbo-charged
energy of Per-Olov Pettersson. I greatly admire and respect his creative ideas,

curiosity, and shared value in the all-mighty-dollar. All total, my friendship with

the members of the group will far outlast what I learned at Caltech.

Having watched many newer students join the group, I think I understand some
of Tom’s enthusiasm with teaching fresh young minds. Alicia Alonzo, Xiao-Chang
Cheng, Erik Daniels and Eric (Speakers) Piquette have been exciting to see grow
and contribute so quickly. I wish them much luck.

Outside of school, the faithful companionship of both Babe and Toddy (my
wife’s and my dogs), has enriched our happy and comforting home. They have
helped balance my life and maintain my sanity.

Without the support and encouragement of my friends and family I don’t know
where I would be today. My parents have and would sacrifice anything for my
betterment. I love them dearly and thank them for their wisdom and strength.
My big brother, Chuck, has been a valued mentor and best friend.

Finally, I thank my wife Linda Swenberg for her love, support, and patience.

She has given greater meaning for my struggles and hard work. Bless you.

vil

Abstract

This thesis describes the technical development of a novel semiconductor device
design aimed at realizing short wavelength visible light emitters. The device struc-
ture, called the graded injector, achieves minority carrier injection in a hetero-
junction system with unfavorable type-II band alignment. Band edge engineering
with an alloy graded intermediary layer effectively reduces the conduction band
offset and allows for efficient minority carrier injection. The basic device structure
consists of a n-CdSe/Mg,Cd,_..Se/p-ZnTe heterojunction, where the Mg,Cdy_,5e
region is graded.

The device design, materials growth, and characterization of II-VI green LEDs
based on this structure are presented. Simulations demonstrate the operating prin-
ciple of the graded injector. Early device development had been hindered by the
lack of a p-type dopant for MBE ZnTe and the unavailability of high quality sub-
strates. These restrictions have been overcome with the development of efficient
nitrogen p-type doping of ZnTe and the growth capability of high quality heteroepi-
taxy on GaSb substrates. The materials characterization of the Mg-chalcogenides
has enabled more accurate band edge engineering necessary for an operating device.

The advances in growth technology and materials characterization have been
incorporated to grow and fabricate working graded injector LEDs. The operating
characteristics of these devices unequivocally demonstrate the diode-like operation
and efficient minority carrier injection. The electrical and optical performance of

these devices will be presented and analyzed.

Vill

Contents
Acknowledgments iii
Abstract vii
List of Figures x
Glossary of Acronyms xl
List of Publications xiii
1 Introduction 1
1.1 Motivation... 2.2... ee 2
1.2 Present commercial LEDs .................0..04. 6
1.3 Development of short wavelength LEDs................ 6
1.3.1 ZnSe based devices ..................0000, 9
1.3.2 Nitride based devices... .............0.-004. 12
1.3.38 The gradedinjector...................0040, 15
14 Qutline.... 15
References... 16
2 The Graded Injector Light Emitter 23
2.1 Introduction... ... 2.0... 00.00.0000. eee eee 23
2.2 Heterojunction approach .................0000004 25

2.3 n-CdSe/p-ZnTe heterojunction ................0004 29

2.4 Mg,Cdy,Se graded junction. . 2... ee ee
95 Conclusion. ...... 0. ce ee ee
References... 00. ee

Molecular Beam Epitaxy, and Nitrogen Doping of ZnTe

3.1 Introduction... 1. ee
3.2 Overview and outline... ee
3.3 Substrate choice... 1... ee
3.3.1 ZnTe.. .
3.3.2 GaSb...
3.4 Nitrogen doping of ZnTe ......... ren
3.4.1 Nitrogen plasma source... . 2... ee eee
3.4.2 Nitrogen doping of ZnTe... 2... .. ee ee ee
3.4.3 Dilute nitrogen doping of ZnTe .............0-..
3.5 MBE growth of graded junction .............-.0000.
3.5.1 Thermal grading of Mg,Cdy_,5Se ............00..
3.6 Summary .. 1...
References...

Development of the Graded Injector Light Emitter

4.1 Introduction and outline .... 2... 20... 2.000. e eee ee ee
4.2 Device engineering .. 2... 2. ee
4.2.1 Band gap studies of Mg,Cd}_,Se .............-.
4.2.2 Valence band offset measurement of MgSe/Cds¢Zn.44Se . . .
4.2.3 Band edge engineering ...........-...00.00004
4.3 Device structure... 2... ee ee
4.3.1 Device fabrication. ... 2... ee ee ee
4.4 Material characterization... ......... 0.000.000 000-

AAL TEM... ee ee ee

41
41
43
45
45
46
47
49
49
53
54
50
56
57

4.5

4.6

44.2 X-ray oe 71
Device operation ©... 73
4.5.1 Electrical performance .......... 0.00.55 eee 73
4.5.2 Optical performance .. 1... 2 ee 78
Conclusion and areas for future development ............ 86

References... 0. ee 87

List of Figures

1.1
1.2
1.3

2.1
2.2
2.3
2.4
2.5
2.6
2.0
2.8

3.1
3.2
3.3

4.1
4.2
4.3
4.4
4.5

Evolution of commercial LED performance .............. 3
Luminous performance of commercial LEDs ............. 7
External quantum efficiencies of commercial LEDs. ......... 8
p-n homojunction LED... 2... eee 24
McCaldin diagram for various semiconductors ............ 26
Flatband diagram of an ideal type-I heterojunction ......... 28
Flatband diagram of CdSe/ZnTe heterojunction ........... 30
CdSe/ZnTe heterojunction under forward bias ............ 31
McCaldin diagram of II-VI compounds ............... . 33
Flatband diagram of Mg,Cd,_,Se graded device... ........ 34

Graded junction CdSe/Mg,Cd,_,Se/ZnTe device under forward bias 36

Hall measurement of ZnTe:N.. 2... 0.2.0.0... 02000 eae 51
X-ray diffraction of ZnTe/ZnTe:N modulated doped layer... ... 52
X-ray rocking curve of ZnTe:N epilayer ..............4. 54
Photoluminescence spectra of Mg,Cd,_,Se epilayer ......... 64
Estimated band gap Mg,Cd;_,Se alloy ..............0.0. 66
Band edge alignment of Mg,Cdj_,Se ..............0040- 68

X-ray rocking curve of a graded device grown on a GaSb substrate. 72

Comparison of J-V characteristics... ........0.. 2.2008, 74

4.6 Log J-V of graded Mg,Cd1_,Se and simulate device... ......
4.7 Log J-V of graded Mg,Cd,_,Se and commercial GaP:N LED

4.8 Room temperature electroluminescence of graded injector LED .
4.9 Device lifetime properties ©... . 0... ee ee ee

4.10 Light emission vs. bias voltage change .............4..

xii

Glossary of Acronyms

LED

CW
MBE
MOCVD
UHV
RGA
RHEED
CIE
FWHM
CBE
VBE
VBO
MIS

light-emitting diode

laser diode

double heterojunction

single heterojunction

continuous wave

molecular-beam epitaxy
(=OMVPE=MOVPE) metalorganic chemical vapor deposition
ultra-high vacuum

residual gas analyzer

reflection high energy electron diffraction
Commission Internationale de L’Eclairage
full width half maximum

conduction band edge

valence band edge

valence band offset
metal-insulator-semiconductor
photoluminescence

electroluminescence

RF
ECR
TEM
AFM

XPS

radio frequency

electron cyclotron resonance
transmission electron microscopy
atomic force microscopy

X-ray photoemission spectroscopy

quantum efficiency

List of Publications

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

Advances in the development of graded injector visible light emit-
ters , J. F. Swenberg, M. W. Wang, R. J. Miles, M. C. Phillips, A. T. Hunter,
J. O. McCaldin, and T. C. McGill, J. Cryst. Growth 138, 692 (1994).

X-ray photoelectron spectroscopy measurement of valence-band
offsets for Mg-based semiconductor compounds, M. W. Wang, J. F.
Swenberg, M. C. Phillips, E. T. Yu, J. O. McCaldin, R. W. Grant, and T.
C. McGill, Appl. Phys. Lett. 64, 3455 (1994).

Measurement of the MgSe/Cdo54Zn0.465e valence band offset by X-
ray photoelectron spectroscopy, M. W. Wang, J. F. Swenberg, R. J.
Miles, M. C. Phillips, E. T. Yu, J. O. McCaldin, R. W. Grant, and T. C.
McGill, J. Cryst. Growth 138, 508 (1994).

Investigation of crystal quality and surface morphology of ZnTe:N
epilayers grown on ZnTe and GaSb substrates, R. J. Miles, J. F.
Swenberg, M. W. Wang, M. C. Phillips, and T. C. McGill, J. Cryst. Growth
138, 523 (1994).

xvi

n-CdSe/p-ZuTe based wide band gap light emitters: numerical sim-
ulation and design,

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

A new approach to wide band gap visible light emitters,
M. C. Phillips, J. F. Swenberg, M. W. Wang, J. O. McCaldin and T. C.
McGill, Physica B 185, 485 (1993).

Proposal and verification of a new visible light emitter based on
wide band gap II-VI semiconductors,
M. C. Phillips, M. W. Wang, J. F. Swenberg, J. O. McCaldin and T. C.
McGill, Appl. Phys. Lett. 61, 1962 (1992).

Forming of Al-doped ZnTe epilayers grown by molecular beam epi-
taxy,

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

Chapter 1

Introduction

This thesis describes major advances in the development of a novel device design
proposed for wide band gap semiconductor visible light emitters. The standard
approach to wide band gap semiconductors has traditionally focused on addressing
the doping constraints which have plagued these materials. The long standing in-
ability to dope the wide band gap semiconductors both n and p-type in the same
material had frustrated the fabrication of diode structures. The device design
presented in this thesis, called the graded junction electron injector or “graded
injector” for short, offers an alternative approach to realizing diode behavior from
heterostructure wide band gap materials. This device avoids the major doping
obstacles associated with the wide band gap semiconductors. The operating prin-
ciples of this device provides a means for achieving minority carrier injection in a
heterojunction system with unfavorable band alignments.

The major contributions in this thesis have been the successful p-type doping of
ZnTe by MBE, the high quality heteroepitaxy of ZnTe on II-V (GaSb) substrates,
and the materials characterization of Mg,Cd,_,Se. The results of this materials
growth and characterization development have been successfully implemented to
realize working LEDs based on the graded injector design. The operating charac-

teristics of these devices have been dramatically improved over the first reported

working devices, and unequivocally demonstrate the diode operation of our het-
erostructure design. This first chapter offers motivation for interest in the wide
band gap materials, and discusses the relevance of our device to shorter wavelength

visible light emitters.

1.1 Motivation

The commercialization of visible light emitting devices (LEDs) began in the early
1960’s following the work of Holonyak and Bevacqua [1]. Since then the optical
performance of commercial visible LEDs has increased by over two orders of mag-
nitude (Figure 1.1). This dramatic performance gain has been achieved through
significant advancements in materials and growth technologies coupled with im-
proved device engineering. The net effect is an over $1 billion per year market
(1992 market) for visible LEDs with more than 20 billion diode chips produced
each year [2]. The visible LED market is in fact the largest segment of the com-
pound semiconductor device market comprising 37% of the overall market revenue
in 1992 [3]. By contrast, during the same year the laser diode (LD) market com-
manded only an 11% share of the compound semiconductor market with worldwide
production exceeding 49 millions units and revenues of roughly $287 million.

We expect the market and demand for visible LEDs to continue to grow. This
will result largely from the development of higher performance devices which have
enabled their introduction into new products and applications. High efficiency
LEDs are in fact the devices of choice over incandescents for most applications due
to their long lifetimes (approaching 10° hours), sharp linewidths for spectral purity,
fast switching times, and reduced power consumption. As visible LED performance
has improved, the range of applications has grown from small indicators used in low
light level environments to larger displays and indicators used in outdoor sunlight
environments.

The human eye’s response to visible light depends strongly on the wavelength.

100.0 T T T

c= Unfiltered Incandescent Lamp

= |__ Yellow-filtered Incandescent
10.0

3 Red-filtered Incandescent

3 SH AlGaAs/GaAs

(4)

ra) 1.0

DH AllnGaP

DH AlGaAs/AiGahs

DH AiGaAs/GaAs

0.1 1 ! n
1970 1975 1980 1985
Year

Figure 1.1: Evolution of commercial visible LED performance [2]. The luminous
performance (lumens/watt) is calculated by multiplying the power efficiency (op-
tical output power/electrical input power) by the human’s eye response function
defined by the Commission Internationale de L’Eclairage, or CIE. Thus, the perfor-
mance is measured relative to the eyes sensitivity. Double-heterostructure (DH)
devices outperform single-heterostructure (SH) devices. For the AlGaAs/GaAs
structure, a GaAs absorbing substrate is present, while for the AlGaAs/AlGaAs
structure the GaAs substrate is removed and a transparent AlGaAs “substrate” re-
mains. DH red AlGaAs LEDs perform better than red-filtered incandescent lamps.
AllnGaP promises to be brighter than AlGaAs in the yellow and possibly in the

red as well.

1990 1995 2000

It is most sensitive in the green region of the spectrum and less so into the red
and blue. A plot of the relative eye sensitivity is found in figure 1.3. Due to this
wavelength dependence, we need a measure other than optical power to quantify
the visual response to a generic light source. We do this by using photometric units
rather than their radiometric counterparts. Radiometric quantities deal solely with
the energy content of optical radiation. An example is the radiant flux, more com-
monly refered to as optical power, measured in units of watts. On the otherhand,
photometric units are defined to quantify the visual response produced by a light
source. The photometric analog to the radiant flux (power) is the luminous flux,
measured in lumens (lm). The luminous flux is essentially the integral of the spec-
tral radiant flux over the visible wavelengths weighted by the eye response [4].
Therefore, the lumen quantifies a visible light source by incorporating the eye’s
sensitivity to the radiation. This enables quantitative comparisons between differ-
ent light sources independent of spectral characteristics. Using photometric units,
we refer to a device’s luminous performance (lIm/W), or simply performance, as
the ratio of the light output in lumens to the electrical input power in watts.

Early red LEDs had luminous performances of 0.15 Im/W and were suitable
for many indoor applications such as indicator lamps and numeric displays. Al-
though, these devices were significantly outperformed by 3.5 lm/W red-filtered in-
candescent lamps. The double-heterostructure (DH) AlGaAs/AlGaAs red diodes
developed in the late 1980’s exceeded the performance of red-filtered incandescent
lamps. This opened a large and growing market for outdoor indicator lamps such
as automotive brake lights and message panels. The advancement of AllnGaP
technology beginning in 1990 [5], with performance roughly ten times better than
existing yellow LEDs and comparable to the highest performance red DH AlGaAs
LEDs, extended the high performance of LEDs into the orange and yellow regions
of the visible spectrum. Together these devices will find growing use in areas
demanding high flux and/or low power consumption.

Much of the commercial success for visible LEDs has occurred in spite of the fact

that LEDs operating in the green and blue have not shared in the remarkable per-
formance gains of the longer wavelength yellow-red LEDs. The present commercial
green (GaP) and blue (SiC) LEDs are not suitable for more advanced applications.
They presently are restricted to small and restrictive markets. Furthermore, these
materials can not be advanced technologically to any great degree in terms of opti-
cal performance or making LDs due to their indirect band gap. The availability of
high performance blue and green LEDs comparable to the commercial yellow and
red would not only enable the use of green and blue LEDs for additional outdoor
indicators but would open entirely new markets to LEDs. A high brightness green
LED, for example, would offer a totally solid-state solution to traffic lighting. This
would lead to faster, brighter, and spectrally sharper lights demanding less energy
and maintenance. The combination of blue and green high brightness LEDs along
with red would enable the creation of full-color large-area displays. Additionally,
high-brightness LEDs can replace other light sources for non-display applications
such as in copiers and printers. Therefore there is tremendous commercial and
technological interest in developing more advanced short wavelength LEDs.

It should be noted that an equal if not greater interest exists in developing
shorter wavelength laser diodes. This results from system’s demanding shorter
wavelength LDs, such as high capacity compact discs players and high performance
color laser printers. Unit sales and revenue of shorter wavelength LDs may fall
short of comparable wavelength LEDs, but the value added to systems would
be significant and possibly greater than for short wavelength LEDs. Therefore,
shorter wavelength LDs and LEDs are of equal interest commercially. In fact, the
solutions to realizing both commercially viable LDs and LEDs will share similar
technological developments. Both, for instance, will need to address similar types

of materials, growth and device engineering issues.

1.2 Present commercial LEDs

Figure 1.2 illustrates the trends and differences in device performance (lumens/watt)
of existing commercial LEDs as a function of their emission wavelength. Viewed
another way, figure 1.3 plots the external quantum efficiencies for various commer-
cial LEDs as a function of the peak emission wavelength. Both figures highlight the
dramatically reduced performance of LEDs as the emission wavelength is reduced.
This phenomena is due to the inherently poor radiative efficiency of indirect band
gap semiconductors. Both GaP and SiC are indirect gap materials and AllInGaP
approaches an indirect gap material as the band gap increases. The drop-off in
performance of the AlInGaP with increasing band gap is expected as the direct-
indirect gap transition is approached and carriers are lost to the higher energy

indirect minima.

1.3 Development of short wavelength LEDs

The need for shorter wavelength LEDs and the inadequacies of the standard HI-V
semiconductors used in visible LEDs has led to an interest in alternative wide band
gap materials. The most likely candidates have long been the wide band gap II-VI
and the III-V nitride (GaN, AIN, InN) semiconductors. These materials with their
wide band gap from the green through the blue and into the UV are most attractive
for short wavelength light emitters. However, two areas have consistently plagued
the development of light emitters based on the wide band gap materials. The
difficultly in achieving low resistivity amphoteric doping (doping both p and n-
type in the same material) required for electron and hole injection and the lack of
suitable lattice matched substrates for homoepitaxial growth have hindered these

material potentials from being realized.

100.0 erm at

AllnGaP/GaAs

Unfiltered Incandescent

_ _ Yellow-filtered
10.0 E7 Incandescent

DH AlGaAs/AlGaAs 4

T_ -Red-filtered

DH AlgaAs/GaAs ;
| “Incandescent

GaP:N/GaP ;
| SH AlgaasiG aAs |

GaAsP:N/GaP

Performance (Lumens/Watt)

F o
[ GaP/GaP

0.1 F 7
F = SIC 1
a‘

450.0 500.0 550.0 600.0 650.0 700.0

Peak Wavelength (nm)

Figure 1.2: LED luminous performance as a function of emission wavelength [2].
High efficiency red and yellow LEDs based on the III-V semiconductors outperform
filtered incandescent lamps. AlGaAs LEDs are only available in one color, but the
AlInGaP quaternary can span the colors from green to red. Unfortunately, the
AlInGaP system becomes less efficient as the emission wavelength becomes shorter.
This results from a transition from direct to indirect band gap with increasing Al
concentration. Indirect semiconductors, like GaP are inherently inefficient light

emitters and can not achieve the performance demanded of high brightness LEDs.

J bd UJ . w . wT Ld Ly Ld T ¥ Lj . LJ bd L
DH AlGaAs/AlGags
one + 1.0
10.0 ra ‘\ DH AIGaAs/Gans
<< / Alln@aP/GaAs
ry ’ 408 a
2 / , SHAIGaAs/GaAs >
= A \ ui 1.0F n \
= ' \ —<
a} / GaP:N/GaP \GaAsP:N/GaP oO
= ! o \ ep)
wo H ‘ ®
CG i <4 \ + 0.4 O.
a 4 GaAsP:N/GaP =
‘ \ <.
Cc O1F / Oo \ClECurvel &
£ ry GaP/GaP ome
ii ta? ‘ 40.2
LL . A,’ \ .
r 4¢ SIC/SIC . \
¢ x
. ‘
se
i a r] i . 4 i i rn I L

r" rn 2 2 ry 4 r" r 0.0
470 490 510 530 550 570 590 610 630 650 670
Peak Wavelength (nm)

Figure 1.3: External quantum efficiencies (photons emitted from device/electrons
passing through device) for various commercial LEDs as a function of peak emission

wavelength. Also plotted with a dashed line is the CIE eye-response curve.

1.3.1 ZnSe based devices

For decades the most popular wide band gap material for short wavelength visible
light emitters has been ZnSe, due to its large band gap in the blue, and close lattice
match to GaAs substrates. The major obstacles have been the inability to grow
p-type ZnSe. There had been many attempts to grow p-type ZnSe using various
growth and doping techniques [6]. Most approaches have been unsuccessful due to
compensation of the incorporated acceptor species, lack of solubility or the forma-
tion of deep levels. The dopant Lithium received attention following Nishizawa’s
et al. [7] reported growth of p-type ZnSe. They used the temperature difference
solution growth method, a bulk crystal growth technique, under controlled vapor
pressure to grow p-type ZnSe crystals. They also reported working p-n diodes
with blue light emission. Soon afterwards, in 1988, Cheng et al. [8] reported the
first p-type conversion of ZnSe grown by MBE using Li. Although they initially
only achieved free hole carrier concentrations of 3.6x10'%cm~, net acceptor con-
centrations as high as 8.0x10/%cm~? and resistivities as low as 2.9 Q-cm were later
reported [9], and p-n junction blue LEDs were successfully grown using Li as the
p dopant {9, 10, 11]. Unfortunately, two major obstacles limited the usefulness of
Li as a p-type dopant in ZnSe. First, the net acceptor concentration saturates at
8.0x10'cm~3, a level too low for suitable ohmic contacting, and with higher Li
incorporation compensation processes occur resulting in highly resistive ZnSe [9].
Second, Li is an unstable dopant, known to diffuse interstitially in ZnSe quite
rapidly [9, 12].

Nitrogen had also received considerable attention as a likely candidate p-type
dopant. Stutius [13] suggested that nitrogen is the most promising element for
a p-type dopant, taking into account problems with the other likely candidates
such as Li, Na, P, and As. Dean et al. [14] identified nitrogen impurities in ZnSe
resulted in the incorporation as a shallow acceptor state with an activation energy

of ~ 110 meV. Park et al. [15] demonstrated the incorporation of nitrogen accep-

tors in ZnSe during MBE and confirmed an activation energy of ~ 110 meV. In
this work they used neutral Np and NH3 and managed to incorporate only small
concentrations of nitrogen impurities due to the low sticking coefficient and solu-
bility of the nitrogen molecular species. Mitsuyu et al. [16] apparently succeeded
in enhancing incorporation of nitrogen impurities in MBE ZnSe by employing a
low-energy ionized beam of NH3. Unfortunately, in both these experiments p-type
conduction was not observed. Although N was believed to be a suitable p-type
dopant, an appropriate method of incorporating sufficient concentrations of the
impurity during MBE remained to be discovered.

A tremendous breakthrough occurred when Park et al. [17] and Ohkawa et
al. [18] working independently discovered that by employing an RF plasma dis-
charge doping incorporation of N in ZnSe was enhanced. N incorporation up to
1.0x10/%cm~? [18] and net acceptor concentrations as large as 3.4x10'"cm~? [17]
were achieved for the first time. p-n homojunction LEDs exhibiting blue lumines-
cence based on p-ZnSe:N /n-ZnSe:Cl were also reported at this time by Park [17].
The dramatic doping success was soon followed by the announcement from 3M [19]
of the first demonstrated operation of a blue-green (490nm) laser diode operating
under pulsed current injection at 77K.

In spite of the successes of N doping of ZnSe many obstacles associated with
this material system remain unsolved. The first noted problem was the satura-
tion in the maximum net acceptor concentration (N4 - Np), using N, at roughly
1x10'8cm~? [20]. These levels of doping are sufficient for low resistivity epilayers
for LEDs and LDs but are by themselves insufficient for ohmic contacts. In fact
this doping saturation problem is exacerbated as the band gap of ZnSe is increased
with the addition of Mg and S forming Mg,Znj_,5,5e;_,. This alloy allows for
tuning of the band gap of the layers for a given lattice constant. Okuyama et
al. [21, 22] demonstrated the use of the Mg,Zn,_.5,Sei_y alloy for electrical con-
finement and optical cladding in separate confinement heterostructure blue laser

diodes [23, 24, 25]. Unfortunately, the p-type doping of Mg,Zn;_,5,Se1_y is com-

plicated by a dramatic decrease in the maximum net acceptor concentration and
increase in the energy of the nitrogen acceptor level with increasing band gap [26].
Therefore laser threshold and operating voltages increase as the operating wave-
lengths decrease [27, 28].

The doping difficulties have been proposed to be fundamentally related to “self-
compensation” effects (through dopant lattice relaxation [29] or the generation of
donor-like centers during growth [30, 31, 32]) or due to acceptor solubility lim-
itations [33]. Nevertheless, considerable progress has been made in overcoming
these doping limitations by engineering novel ohmic contacting schemes. A pseu-
dograded band gap Zn(Se,Te) [34] epilayer, resonant tunneling multiquantum well
ZnTe/ZnSe [35] epilayer, and HgSe [36] contacting layer all succeeded in reducing
contacting resistances to levels acceptable for LEDs and LDs. Subsequently, low
voltage operation (< 4 Volts) of ZnSe based blue and green LEDs [37], and low volt-
age (Vi, ~ 4.4 V) room temperature operation of blue-green laser diodes [27, 28]
were demonstrated.

The continued advancement of ZnSe based light emitters has progressed to the
point where LED structures operating in the blue and green have performances
(1.6 Im/W 489nm, 17 Im/W 512nm) [40] as good as commercial red. The green
LEDs have maximum external quantum efficiencies of 5.3%. Despite these accom-
plishments, the ZnSe based devices face a major challenge relating to the issue
of stability and operating lifetime. The blue and green LEDs, while performing
orders of magnitude better than their commercial conterpart, are severly limited
in device lifetime. Improvements in bulk growth technology of ZnSe [38, 39] has
resulted in high quality ZnSe with x-ray full width at half-maximum (FWHM) of
11-16 arcsec, indicating crystal quality comparable to that of GaAs substrates. Ho-
moepitaxial devices grown on on these substrates have the best lifetimes, although
even operating under optimal conditions (J = 15 A/cm?) these LEDs’ light output
decay exponentially with a time constant of only 675 hours [40]. As for the laser

diode, the longest published continuous wave operation at room temperature is

one hour [41].

Guha et al. [42] studied the degradation of ZnSe light emitting devices, and
showed that the degradation was related to the formation of new crystal de-
fects from the vicinity of pre-existing defects such as stacking faults and these
defects propagated throughtout the active region and enhanced nonradiative re-
combination. They further found that these stacking faults originated at the epi-
layer/substrate(GaAs) interface during the initial stages of the growth [43]. Addi-
tional studies have shown that the incorporation of nitrogen affects the structural
properties of ZnSe, with the generation of point defects accompanying N dop-
ing [44], and an increase in the formation of stacking faults [45]. The success or
failure of the ZnSe devices will largerly depend on overcoming these degradation
problems. While this may be possible to achieve through improved growth tech-
nology, it could prove to be intractable if the material stability is fundamentally

associated with the nitrogen doping.

1.3.2 Nitride based devices

The III-V nitrides (InN, GaN, AIN) have also been viewed as promising materials
for short wavelength light emitters [46], because the wurtzite polytypes of InN,
GaN, and AIN form a continuous alloy sytem whose direct band gaps range from
1.9 eV for InN, to 3.4 eV for GaN, to 6.2 eV for AIN. Additionally, these materials
possess physical hardness, large band offsets, high thermal conductivity, and high
melting temperatures [47]. The major obstacles, similar to those of ZnSe, have
been the lack of a suitable substrate material that is lattice matched and thermally
compatible with the nitrides and the difficulty in achieving p-type conductivity.
Due to the lack of a suitable substrate, it has been difficult to grow epitaxial
GaN films of high quality. For lack of an ideal substrate, most nitride material
has been grown on sapphire substrates despite a huge lattice mismatch (~13.8%)

and large difference in thermal expansion coefficients. This choice of substrate is

attributed to its availability, hexagonal structure, ease in handling, and thermal
stability at elevated growth temperatures. Consequently, much work has been
devoted to the understanding and development of epitaxy of nitrides on lattice
mismatched substrates 48]. A major improvement in the structural quality of
GaN epilayers was accomplished by using an AIN buffer layer between the grown
GaN film and sapphire substrate in both MBE [49] and MOVPE [50] growth.
Nakamura et al., subsequently discovered that a thin buffer layer of GaN grown at
lower growth temperatures could also produce similar results.

The longstanding problem of p-type doping of GaN was frustrated by compen-
sation processes leaving only highly resistive material. Nevertheless, GaN LEDs
had been reported over 20 years ago [51], but these devices were high field, low effi-
ciency metal-insulator-semiconductor (MIS) structures. Recently however, Amano
et al. succeeded for the first time in producing p-type GaN [52] and AlGaN [53}.
As grown Mg-doped GaN layers had a hole concentration p = 2x10'°cm™3, and
resistivity p = 320 Q-cm. By using low energy electron beam irradiation (LEEBI)
the electrical properties of the GaN were observed to change dramatically to p =
3x10'8cm~*, and resistivity p = 0.2 Q-cm. Using this technique, Amano et al.
developed the first conventional p-n junction GaN LED [52].

It was also discovered that compensated Mg-doped GaN could be converted
equally well to p-type material by thermal annealing under an Nz ambient [54].
This technique was an improvement over electron irradiation since it could convert
the entire layer rather than only the irradiated regions. The compensation mech-
anism was identified as resulting from an Mg-H neutral complex formed during
growth [55]. Soon afterwards, Moustakas et al. [56] reported the successful p-type
doping of GaN grown by MBE using a solid source Mg cell as the dopant. They
were able to achieved p-type material without post-growth processing due to the
absence of H in the UHV environment of MBE.

With improvements in epitaxial material and amphoteric doping capabilities,

it was not long before higher performance nitride LEDs were fabricated. Since

GaN has a band gap in the UV, the ternary alloy InGaN was the likely candi-
date for realizing visible light emitters since its band gap varies from 1.9 to 3.4
eV depending on the In concentration. The growth of high quality InGaN films
obtained by Yoshimoto et al. [57], and the growth of Si-doped n-InGaN [58] lead
to longer wavelength p-GaN/n-InGaN/n-GaN LEDs. A dramatic improvement
in optical performance was realized with the successful Zn doping of InGaN in a
InGaN/AlGaN DH blue LED [59]. Zn doping had been used in GaN [60, 61, 62]
for strong below band gap emission. Incoporation of Zn in InGaN resulted in re-
comination emission 0.5 eV lower than the band edge emission of InGaN. This
enabled the devices to operate in the blue. Quantum efficiencies as high as 2.7%
were reported for these blue (450nm) LEDs, and they have been announced as a
commercial product. Recently, external quantum efficiencies as high as 5.4% for
blue (450nm) and 2.4% for blue-green (500nm) were reported by researchers from
Nichia [63] using the InGaN/AlGaN DH structure. The optical performance of
these LEDs and lifetimes of ten of thousands of hours make these devices com-
mercially viable. In addition to the success of the nitrides in realizing commer-
cial quality LEDs, recent reports of optically pumped near ultraviolet lasing from
single-crystal GaN [64] has given encouragement for the propects of nitride based
diode lasers in the near future.

Despite the remarkable results achieved with the nitrides in the past few years,
there are still many issues which need to be addressed for commercial success.
The present LED’s optical characteristics suffer from very broad linewidths in the
emission spectra, FWHM ~ 70nm, due to the deep Zn recombination center. This
broad emission seriously degrades the spectral purity. Ideally band to band recom-
bination would be prefered over the deep Zn center (especially for lasing). This
will require alloys with higher In concentrations. Generally, improved substrate
material, ideally GaN, must be pursued to further improved material quality and
enable the fabrication of laser diodes. Additionally, doping (primarily p-type),

and processing and ohmic contacting technologies must also be improved for laser

development.

1.3.3. The graded injector

An alternative approach to realizing wide band gap light emitters based on the
“sraded injector” is discussed in this thesis. This novel device concept offers a
unique approach to achieving minority carrier injection in a p-n heterojunction
with unfavorable band offsets. Additionally this device avoids problems with am-
photeric doping constraints and ohmic contacting by using materials which are
capable of being readily doped either p-type or n-type. In spite of the above men-
tioned successes with ZnSe and GaN, the graded injector still remains of extreme
technical interest due to its unique operation and potential application in the area

of wide band gap semiconductors.

1.4 Outline

The remainder of the thesis is arranged as follows: Chapter 2 introduces the con-
cept and basic design of the graded injector. The relevant issues necessary for a
heterojunction diode are presented and incorporated in the device structure. De-
vice simulations demonstrate the operating principle behind the graded junction.

Chapter 3 discusses the issues and details of the growth technology, molecular
beam epitaxy (MBE), used to fabricate the graded injector LED. The choice and
availability of appropriate substrates are presented. The p-type doping of ZnTe
and the epitaxial growth of high quality ZnTe on GaSb substrates are presented.

Chapter 4 reports on the major advances in the development of the graded
injector. Materials characterization and device engineering necessary for device
development are present. Characterization of the device’s material quality, electri-

cal behavior and optical performance are discussed.

Bibliography

[9)

N. Holonyak and S.F. Bevacqua, Appl. Phys. Lett. 1, 82 (1962).
M.G. Craford, IEEE Circuits and Devices 8, 24, (1992).

From data of M.G. Craford adapted from Strategies Unlimited report on Com-
pound Semiconductor Device Market 1992. Net revenues totaled $2.8 billion.

The lumen is defined as the luminous flux of monochromatic radiation at 555
nm whose radiant flux is equal to 1/683 W. For a discussion of photometric
quantities and units see R.W. Boyd, Radiometry and the Detection of Optical
Radiation, (John Wiley, New York, 1983), Chapter 6.

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

For a review of this work see R.N. Bhargava, J. Cryst. Growth 59, 15 (1982).

J. Nishizawa, K. Itoh, Y. Okuno and F. Sakurai, J. Appl. Phys. 57, 2210
(1985).

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

M.A. Haase, H. Cheng, J.M. DePuydt and J.E. Potts, J. Appl. Phys. 67, 448
(1990).

[10] T. Yasuda, I. Mitsuishi and H. Kukimoto, Appl. Phys. Lett. 52, 57 (1988).

17
[11] J. Ren, B. Sneed, K.A. Bowers, D.L. Dreifus, J.W. Cook, J.F. Schetzina and
R.M. Kolbas, Appl. Phys. Lett. 57, 1901 (1990).
[12] G.F. Neumark and S.P. Herko, J. Cryst. Growth 59, 189 (1982).
[13] W. Stutius, J. Cryst. Growth 59, 1 (1982).

[14] P.J. Dean, W. Stutius, G.F. Neumark, B.J. Fitzpatrick and R.N. Bhargava,
Phys. Rev. B 27, 2419 (1983).

[15] R.M. Park, H.A. Mar and N.M. Salansky, J. Appl. Phys. 58, 1047 (1985).
[16] T. Mitsuyu, K. Ohkawa and O. Yamazaki, Appl. Phys. Lett. 49, 1348 (1986).

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

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

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

[20] J.Qui, JM. DePuydt, H. CHeng and M.A. Haase, Appl. Phys. Lett. 59, 2993
(1991).

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

(22| H. Okuyama, K. Nakano, T. Miyajima and K. Akimoto, J. Cryst. Growth
117, 139 (1992).

[23] K. Akimoto, H. Okuyama, T. Miyajima, T. Morinaga, F. Hiei and M. Ozawa,
Tech. Dig. 2, 306 (1993).

[24] J. Gaines, R. Drenton, K. Haberern, T. Marshall, P. Mensz and J. Petruzzello,
Appl. Phys. Lett. 62, 2462 (1993).

[25] J. Petruzzello, J. Gaines and P. van der Sluis, Appl. Phys. Lett. 75, 63 (1994).

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

[27] A. Salokatve, H. Jeon, J. Ding, M. Hovinen, A.V. Nurmikko, D.C. Grillo, J.
Han, L. He, Y. Fan, M.D. Ringle, R.L. Gunshor, G.C. Hua and N. Otsuka,
Electronics Letters 29, 2192 (1993).

[28] N. Nakayama, S. Itoh, H. Okuyama, M. Ozawa, T. Ohata, K. Nakano, M.
Ikeda, A. Ishibashi and Y. Mori, Electronics Letters 29, 2194 (1993).

[29] D.J. Chadi and K.J. Chang, Appl. Phys. Lett. 55, 575 (1989).

(30] A. Ohki, Y. Kawaguchi, K. Ando and A. Katsui, Appl. Phys. Lett. 59, 671
(1991).

(31| B. Hu, G. Karczewski, H. Luo, N. Samarth and J.K. Furdyna, Appl. Phys.
Lett. 63, 358 (1993).

[32] K.A. Prior, B. Murdin, C.R. Pidgeon, S.Y. Wang, I. Hauksson, J.T. Mullins,
G. Horsburgh and B.C. Cavenett, J. Cryst. Growth 138, 94 (1994).

[33] Y. Fan, J. Han, L. He, R.L. Gunshor, M.S. Brandt, J. Walker, N.M. Johnson
and A.V. Nurmikko, Appl. Phys. Lett. 65, 1001 (1994).

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

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

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

19
[37] J. Ren, Y. Lansari, Z. Yu, J.W. Cook and J.P. Schetzina, J. Elec. Mater. 22,
973 (1993).

[38] G. Cantwell, W.C. Harsch, H.C. Cotal, B.G. Markey, S.W.S. McKeever and
J.E. Thomas, J. Appl. Phys. 71, 2931 (1992).

[39] H. Cotal, B.G. Markey, S.W.S. McKeever, G. Cantwell and H.C. Harsch,
Physica B 185, 103 (1993).

[40] D.B. Eason, Z. Yu, W.C. Hughes, W.H. Roland, C. Boney, J.W. Cook, J.F.
Schetzina, Appl. Phys. Lett. 66, 115 (1995).

[41] S. Itoh and A. Ishibashi, Proc. SPIE Conference on II-VI Blue/Green Laser
Diodes, R.L. Gunshor and A.V. Nurmikko, eds., SPIE (1994).

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

[43] S. Guha, J.M. DePuydt, J. Qiu, G.E. Hofler, M.A. Haase, B.J. Wu and H.
Cheng, Appl. Phys. Lett. 63, 3024 (1993).

[44] J. Petruzzello, J. Gaines, P. van der Sluis, D. Olego and C. Ponzoni, Appl.
Phys. Lett. 62, 1496 (1993).

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

(46] For a review of the III-V nitrides see, S. Strite, M.E. Lin and H. Morkog, Thin
Solid Films 231, 197 (1993)

[47] S. Strite and H. Morkog, J. Vac. Sci. Technol. B 10, 1237 (1992).

[48] For a review of substrates and nitride epitaxy see Ref. [47]

[49] S. Yoshida, S. Misawa and S. Gonda, Appl. Phys. Lett. 42, 427 (1983).

20
[50] H. Amano, N. Sawaki, I. Akasaki and Y. Toyoda, Appl. Phys. Lett. 48, 353
(1986).
(51) J.I. Pankove, E.A. Miller and J.E. Berkeyheiser, RCA Rev. 32, 383 (1971).

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

[53] H. Amano, M. Kito, K. Hiramatsu and I. Akasaki, Inst. Phys. Conf. Ser. 106,
725 (1990).

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

(55| S. Nakamura, N. Iwasa, M. Senoh and T. Mukai, Jpn. J. Appl. Phys. 31, 191
(1992).

(56] T.D. Moustakas and R.J. Molnar, Materials Research Society Symp., Boston,
MA, 1992.

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

[58] S. Nakamura, T. Mukai and M. Senoh, Jpn. J. Appl. Phys. 31, L16 (1992).
[59] S. Nakamura, T. Mukai and M. Senoh, Appl. Phys. Lett. 64, 1687 (1994).
[60] J.L. Pankove and J.A. Hutchby, Appl. Phys. Lett. 24, 281 (1974).

[61] H. Monemar, O. Lagerstadt and H.P. Gislason, J. Appl. Phys. 51, 625 (1980).

[62] P. Bergman, G. Ying, B. Monemar and P.O. Holz, J. Appl. Phys. 61, 4589
(1987).

[63] S. Nakamura, IEEE IEDM, 567 (1994).

[64] X.H. Yang, T.J. Schmidt, W. Shan, J.J. Song and B. Goldenberg, Appl. Phys.
Lett. 66, 1 (1995).

Chapter 2

The Graded Injector Light

Emitter

2.1 Introduction

An LED’s basic structure is a p-n junction diode. Figure 2.1 illustrates the band
diagram for a simple p-n homojunction. Under a forward bias close to the band gap
of the device material, minority carriers are injected into the recombination region
of the device where they can recombine and emit light [1]. Commercial devices
have been based on simple p-n homojunctions, where the p and n regions are the
same binary or alloy material. More sophisticated structures, such as single- and
double-heterostructures improve the recombination and extraction efficiencies but
they are still based on the homojunction design [2].

This standard approach to LEDs has been sufficiently successful for narrow
band gap materials, but has seen less success in the development of high efficiency
green and blue LEDs. The fundamental problem in the development of wide band
gap light emitters based on the II-VI semiconductors has been the inability of
obtaining both n- and p-type doping in the same material. To some degree this

restriction has been mitigated with the p-type doping of ZnSe by MBE [3, 4], but

Figure 2.1: Schematic band diagrams for p-n homojunction at (a) zero bias and
(b) under a forward bias Vgias. The dashed line indicates the Fermi level in
the unbiased diagram and the quasi-Fermi levels in the biased diagram. Under a
forward bias, electrons are injected into the p-type material and holes are injected
into the n-type material. The injected carriers then recombine with the majority

charge carriers and emit light.

there remain several challenges ahead in the development of ZnSe based LEDs
and laser diodes as previously discussed in chapter 1. An obvious yet non-trivial
solution to avoid the doping problem is to use two different materials, one easily
doped n-type and the other easily doped p-type, and form a heterojunction.

In this chapter, we examine a novel device concept based on a unique p-n
heterojunction structure. This LED design is based on the only closely lattice

matched II-VI binary compounds, namely n-CdSe and p-ZnTe. By taking advan-

tage of the natural doping types of these materials we avoid contacting and doping
problems. An abrupt n-CdSe/p-ZnTe junction is found to be incapable of carrier
injection. Using a properly graded alloy of Mg,Cdy_,Se placed between the CdSe
and ZnTe junction we create a heterostructure capable of efficient minority car-
rier injection into the wider band gap ZnTe. Computer simulations illustrate the

operating priciple of this graded injector LED.

2.2 Heterojunction approach

The basic paradigm to form a p-n heterojunction diode is to join a p-type material
to a different n-type material. Forming a structure with these materials would by
definition contain a p-n junction. In addition to using suitably dopable materials,
there are strict constraints on lattice matching and band offsets necessary for a
realistic device. In order to maintain material quality, lattice matched materials
must be used, although small amounts of strain can be accomodated. As will be
demonstrated below, whether a heterojunction device will be able to inject carri-
ers will depend completely on the band offsets for that particular heterojunction
structure.

To design a heterostructure device we need information on materials’ band
gaps, band offsets, dopabilities and lattice parameters. McCaldin [5] introduced
a graphical technique to simultaneously present these parameters in a single dia-
gram. By displaying all the relevant materials properties, these so called McCaldin
diagrams allow one to easily choose suitable materials for a heterostructure device.
Figure 2.2 shows a McCaldin diagram for some of the common II-VI binary com-
pounds and other popular semiconductors. The band edge positions are plotted
relative to the valence band of GaAs and are from the theoretical prediction of
Harrison’s linear combination of atomic orbitals (LCAO) theory [6]. The figure
plots as a function of lattice constant the band edges of a semiconductor. The

symbol for a material’s band edge is filled/empty if the semiconductor can/cannot

Energy (eV)

4 rerryper rrr rrp rrp rr rere rrp rere PE epee
2 r ‘ 4
TF A GoSb :
OF Si GoAs aZnTe CdTe|
CdSe UOMgte
—1 + Bi ZnSe Mase 4
_9 | Ua2ns C) MgS 0 cas |
—3 ped pe po fe a ee
3.4 5.6 9.8 6 6.2 6.4 6.6

Lattice constant (Angstroms)

Figure 2.2: McCaldin diagram for various semiconductors. As a function of lattice

parameter, the band edges of a semiconductor are plotted relative to the valence

band of GaAs.

This plot is from the valence band edge offsets calculated by

Harrision [6]. The conduction band is indicated by a triangle and the valence band

is indicated by a square. The band gap is given by the vetical line connecting the

triangle and square. If the triangle/square is filled it indicates that the material

can be conventionally doped n-type/p-type. If the shape is empty it means the

material cannot be doped that type. Band offsets can be read from the diagram

by the energy difference between band edges.

be doped that conductivity type. Yu et al. [7] provide a comprehensive review on
the subject of band offsets and the salient issues associated with the theoretical
and experimental results. We rely on the obvious trends displayed by the diagrams
and make use of accepted experimental results for a given band offset to construct
a heterojunction.

The use of heterojunctions to avoid the doping trends in the I-VI is an old idea.
The most intensively investigated heterostructure among the II-VI semiconductors
has been n-ZnSe/p-ZnTe [8, 9, 10]. Unfortunately, as can be seen in the Mccaldin
diagram the extremely large 7% lattice mismatch between these materials proves to
be disastrous for LED operation. McCaldin and McGill [11] recently proposed the
n-AlSb/p-ZnTe junction as an electron injector into Zn'Te based on the theoretical
band alignment of Harrison [6]. Yu et al. [12] experimentally measured the band
offset of AlSb/ZnTe and found the theroretical values to incorrectly predict the sign
of the conduction band offset. Instead of the conduction band of AlSb lying above
ZuTe, it was measured to be 0.21 eV below. Attempts to fabricate n-AlSb/p-ZnTe
diodes by Han et al. (13] demonstrated the inability of obtaining electron injection
for this heterojunction.

We can notice various trends in doping when a larger collection of materials are
plotted using different theoretical calculations and experimental data. It seems to
hold, as suggested [14, 15], that the positions of the valence and conduction band
edges with respect to some absolute energy scale are an important factor in doping.
Generally we see, the lower the position of the valence band edge, the more difficult
the material is to be doped p-type and, similarly, the higher the conduction band
edge, the more difficult it is to be dope n-type [5]. This appears to hold true for the
selenides, sulfides and ZnTe. We should note that ZnTe is the only wide band gap
II-VI compound which is conventionally doped p-type. Unfortunately, the doping
trends evident in the McCaldin diagrams counter the need for either an n-type
material with a high conduction band or a p-type material with a low valence

band. This is the same impediment with wide band gap II-VI homojunctions,

n-type material p-type material

Figure 2.3: Flat band diagram of an ideal type-I heterojunction for efficient mi-
nority carrier injection. In this case a wider band gap n-type material achieves
efficient electron injection into the recombination p-type material. The holes from
the p-type material are blocked by the valence band offset, and the electrons see

no barrier to injection.

since a large band gap material will have either a high conduction band or low
valence band.

An ideal heterojunction for efficient minority carrier injection would have a
type-I band offset for a given recombination material as shown in figure 2.3. We
notice that in this ideal case there would be no barrier for electron injection into
the p-region, but the valence band offset would block holes. This band alignment is
the same principle used in long wavelength single and double heterostructure LEDs
and LD where amphoteric lattice-matched ternery and quaternary alloys are used
for both p and n materials. Unfortunately a natural heterojunction system for
a type-I band alignment doesn’t exist in the II-VI materials under the physical
constraints present.

To realize a II-VI LED device based on a heterostructure, we must design a

structure given the material constraints of the II-VI semiconductors (doping and

lattice constant) and engineer a solution to the unfavorable band offsets.

2.3 n-CdSe/p-ZnTe heterojunction

On close examination of figure 2.2, we note that CdSe and ZnTe are the only
closely lattice matched (0.4% mismatch) heterojunction pair among the II-VI bi-
nary compounds. Furthermore, the two compounds can form a p-n heterojunction
relying only on the natural doping proclivities for these materials. Although CdSe
does not exhibit a stable zinc blende phase when grown in the bulk, prefering in-
stead the wurzite structure, high quality epitaxial layers of zinc blende CdSe can
be grown by MBE on (100) zinc blende substrates [16] and epilayers of ZnTe [17].
With the slight lattice mismatch between the cubic lattice constants, several hun-
dred angstroms of CdSe can easily be grown on ZnTe without exceeding the critical
thickness. Given this we might expect the n-CdSe/p-ZnTe heterojunction to form
the basis of a wide band gap II-VI heterojunction LED.

Figure 2.4 shows a flatband diagram of an abrupt n-CdSe/p-ZnTe heterojuc-
tion. Unfortunately this heterojunction has a type-II band alignment, with the
valence band of CdSe lying 0.64 eV below that of ZnTe and the conduction band
of CdSe lying 1.22 eV below that of ZnTe [18]. The valence band offset provides
a desired barrier to hole injection into the CdSe, but the larger conduction band
offset serves as a greater barrier to electron injection. Under forward bias, both
electrons and holes see large energy barriers at the heterojunction interface. As a
result of this we would expect insignificant thermionic emission and mostly non-
radiative recombination.

Computer simulations and attempts at fabricating a n-CdSe/p-ZnTe hetero-
junction confirm this suspicion. Figure 2.5 shows the band edge profile and carrier
concentrations for an abrupt heterojunction under a forward bias of 1.25 volts

calculated using the drift-diffusion model of Wang et al. [19].

Conduction Band Offset
1.22 eV

225 ev p-Znle

n-CdSe 1.67 eV 4

0.64 eV
Valence Band Offset

Figure 2.4: Flat band diagram of the only closely lattice-matched bmary I-VI
compounds. The resulting p-n heterojunction has a type-II band alignment. There
exists the favorable condition that the valence band offset blocks the extraction of
holes from the ZnTe. Unfortunately, the larger conduction band offset serves as a

barrier to electron injection into the ZnTe.

As expected, the device simulations show carrier accumulation at the interface
and virtually no carrier injection across the junction. The resulting overlapping
carrier accumulation leads to predominantly non-radiative interfacial recombina-
tion. At larger biases some hole injection into the CdSe may occur but negligible
electron injection is seen due to the larger conduction band barrier. Therefore, the
simulations corroborate our intuition and demonstrate the futility of an abrubt
n-CdSe/p-ZnTe heterojunction as a light emitter.

"We could use these two binaries to form a quaternary alloy, (CdSe);_,(ZnTe).,
which could be used as an intermediary layer between the n-CdSe and p-ZnTe.
In principle, we could grade the band edges continuously from ZnTe to CdSe by
varying x from 1 to 0 in the quaternary alloy (CdSe);_,(ZnTe),. This would

eliminate the abrupt interface and carrier accumulations but the resulting current

2.0
1.0 F -
S 5
2 0.0 Ec Bias = 1.25 Volts
> ,
o)
® -1.0 F
5 L _|~
-2.0 r
3.0 hacen tonne nn
20.0
oO a
E 16.0} Electrons Bias = 1.25 Volts
© | ——-— Holes
2 120
So 80F {
c {
= 4of |
rab) L
2 a
S 0.0 P------------
Oo Poa ee |

-500.0 -300.0 ~—--100.0 100.0 300.0 500.0
Position (Angstroms)

Figure 2.5: Calculated band bending and carrier concentrations for an abrupt
n-CdSe/p-ZnTe heterojunction. The computer simulation is based on the drift-
diffusion model of Wang et al. [19]. Both valence and conduction band barriers
lead to overlapping carrier accumulation at the interface and non-radiative recom-

bination.

would be predominantly hole injection into the CdSe due to the much smaller

effective barrier in the valence band.

2.4 Mg,Cd,_,Se graded junction

Although an abrupt heterojunction will not suffice for a light emitter, the n-CdSe
and p-ZnTe system can still be used to realize a p-n heterojunction LED. The
graded injector device is based on the favorable properties of these two compounds,
namely the natural doping type for these materials and the valence band offset.
To facilitate electron injection from the n-CdSe into the p-ZnTe, while prevent-
ing hole extraction, we must grade the conduction band (CB) continuously from
the CdSe CB to the ZnTe CB in an intermediary junction region, while main-
taining the abrupt valence band offset. Since the anion of the II-VI materials
predominantly determines the valence band edge position, this junction grading
can be accomplished by alloying CdSe with an appropriately chosen wider band
gap selenide.

The McCaldin diagram in figure 2.6 shows the properties of an expanded
group of II-VI materials including the magnesium chalcogenides. We see the Mg-
chalcogenides have very large band gaps and can be alloyed with the Zn and Cd
compounds to increase the band gap. Unfortunately, there is no lattice matched
ternary II-VI analogous to the III-V Al,Ga,_,As system. Therefore any introduc-
tion of Mg in a II-VI ternary will be associated with a lattice strain.

We observe in the McCaldin diagram of figure 2.6 that an alloy of Mg,Cd,_,Se
can be formed which will have a zero conduction band offset and a large valence
band offset with ZnTe. The graded injector device first proposed by Phillips et
al. {20| is based on using an alloy of Mg,Cd,_,Se to continously grade the con-
duction band from ZnTe to CdSe, at the same time maintaining the valence band
offset. The schematic in figure 2.7 illustrates a flatband diagram for the idealized

graded injector device resulting from this band edge engineering. Starting from

oer eee er ee peo
4r aA |
3 F 4
S 7 |
e TT
2 of oY
lu ZnTe MgTe J
—-1 +F |
} ZnSe MgSe CoSe 1
-2 P Cds 7
mt [ae ne a ee ee PT

5.6 5.8 6.0 6.2 6.4

Lattice constant (Angstroms)

Figure 2.6: McCaldin diagram of common II-VI materials and the Mg-
chalcogenides. The Mg,Cd,_,Se ternary alloy is represented by the shaded region.
The band gap of MgSe is the estimated value derived in Chapter 4 of this thesis.
The band offsets of MgSe and MgTe are from Wang et al. [21]. The band gap of
MegTe is from Parker et al. [22] and the lattice constant is from Waag et al. [23].
The lattice constant of MgSe is from Okuyama et al. [24], as are the band gap
and lattice constant of MgS. The band offset for MgS assumes the common anion
rule, which states that the position of the valence band edge is determined by the
constituent anion. The band gaps and lattice constants for the Mg-Chalcogenides
are extrapolated values for the zinc blende structure. We notice that for a specific

alloy of Mg,Cd ,_,Se there can a zero conduction band offset with ZnTe.

Figure 2.7: Flatband diagram of an n-CdSe/Mg,Cd;_,Se/p-ZnTe heterojunction.
This “graded junction” device facilitates electron injection while preserving the
valence band offset to confine the holes. Beginning at the Mg,Cd,_,Se/ZnTe
interface we grade the Mg concentration until we have only CdSe. This schematic
assumes the common anion rule for the valence band offset, although the valence
band of Mg,Cd,_,Se is lowered with increasing Mg concentration. The initial
Mg concentration will depend on the band gap and valence band position of the

Mg,Cdj_,Se alloy.

the Mg,Cd,_,Se/ZnTe interface we grade the Mg concentration from some specific
amount down to CdSe over a few hundred angstroms.

To properly grade the alloy to the conduction band of ZnTe we must know
the band gap and band offset for Mg,Cd,_,Se as a function of Mg concentration.
There is a large variation in the estimates and extrapolations of the band gap of
MgSe ranging from 3.6 to 5.6 eV [24, 25]. Furthermore, the effect on the valence
band edge with the addition of Mg was previously unknown. In chapter 4, we
investigate these materials’ properties to best determine the alloy concentration

needed for a zero conduction band offset between Mg,Cd _,Se and ZnTe. We find

a good estimate for the maximum Mg concentration to be about 69%.

The lattice constant of zinc blende MgSe has an extrapolated value of 5.89 A [24],
therefore the introduction of Mg into CdSe will introduce a fairly large lattice
strain with respect to ZnTe. For the estimated maximum Mg concentration, the
mismatch to ZnTe will be 2.4%. This is a large lattice strain to accommodate,
but since the Mg concentration is graded down to CdSe it is only the peak strain.
Therefore if we can keep the alloy junction region thin enough we should be able
to stay below the critical thickness. We lack any accurate ability to calculate the
critical thickness for this given structure but we estimate that the thickness of the
whole graded region must be less than 300 A (20]. Device simulations demonstrate
that the grading can be very thin and still operate as expected [19].

To better understand the role the graded junction plays in the device operation
and the mechanism by which the graded device works we can look at the drift-
diffusion simulation results of Wang et al. [19]. Figure 2.8 shows the resulting band
edge profile and carrier concentrations for a Mg,Cdy_,Se graded junction device
under a 2.0 Volt forward bias. These results where calculated for a 200 A graded
region. As the thickness becomes much thinner the injection efficiency begins to
decrease, but if the thickness becomes too large we will have to be concerned
with exceeding the critical thickness and defect formation. Simulation results also
demonstrate that the operation of the graded junction does not require the graded
region to be doped.

We note two significant differences in this device operation compared to the
abrupt heterojunction in figure 2.5. First, we see the graded region effectively
reducing the barrier in the conduction band and preserving the barrier in the
valence band. And secondly, we notice that the graded junction spatially separates
the accumulation of electrons and holes. As long as this separation distance is
roughly larger than the Bohr radius of an exciton, there should be no significant
overlap of the electron and hole wavefunctions and therefore a dramatic reduction

in the non-radiative recombination due to the accumulation. The net effect of the

15 -
<— O5OF _—_
S -0.5
5 r Bias = 2.0 Volts
CO +45 E
Lu } NP
-2.5 &
3 5 a a a a a —y a i i +
' 7.0
E Electrons Bias = 2.0 Volts
00 —-—-— Holes
"oOo 50F
Cc
Oo 3.0 F
g X 200
5 10F ‘
pe Le ee ee ee ee
je) A 2 $ . + mn a af . . as
C1. a : tte
-1000.0 -500.0 0.0 500.0 1000.0

Position (Angstroms)

Figure 2.8: Calculated band edge profile and carrier concentrations under a forward
bias of 2.0 Volts. The simulated results are based on the drift-diffusion model as
calculated by Wang et al. [19]. This model used a graded region 200 A thick.
Notice the reduced barrier for electrons compared to the abrupt heterojunction in
figure 2.5, as well as the spatial separation of the charge carriers. We are beginning

to see the onset of minority carrier injection into the ZnTe.

graded region is to facilitate electron injection into the wider band gap ZnTe. At
low bias voltage, most of the current is non-radiative interfacial recombination.
But as the bias is increased, we begin to see significant electron injection into the
ZnTe where they can recombine radiatively. Since the large valence band offset
remains, there is no hole extraction into the CdSe.

The early device results were severly limited by the lack of a p-type dopant for
MBE grown ZnTe and the use of poor quality ZnTe substrates [20]. In spite of
these restrictions, early devices worked as light emitting devices with light emission
originating from the ZnTe substrate. Unfortunately though, these devices operated
at high bias voltages and therefore the injection mechanism was not thoroughly
understood. It was unclear whether the devices operated as high field effect devices
rather than normal thermionic p-n junction diodes. Work on materials and device
development covered in the following chapters of this thesis led to the fabrication
of graded junction LEDs which unequivocally demonstrate the diode like operation

as predicted by the simulations.

2.5 Conclusion

A novel p-n heterojunction LED based on the only closely lattice matched II-VI bi-
nary pair (n-CdSe/p-ZnTe) is demonstrated to be capable of minority carrier injec-
tion. Although an abrupt n-CdSe/p-ZnTe heterojunction results in non-radiative
currents, an efficient electron injector is created by engineering the band edges

with an intermediary Mg,Cd,_,5e alloy junction region.

Bibliography

[1] For a discussion of the physics of p-n junctions, see A. S$. Grove, Physics and

Technology of Semiconductor Devices (Wiley, New York, 1967), pp. 149-207.

[2] H. Kressel, Semiconductor Lasers and Heterojunction LEDs (Academic Press,

New York, 1977).
[3] K. Ohkawa, T. Mitsuyu and O. Yamazaki, J. Cryst. Growth 86, 329, (1988).

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

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

[7] E.T. Yu, J.O. McCaldin and T.C. McGill, Solid State Physics,v. 46, (Aca-
demic Press, 1992).

[8] M. Aven and W. Garwacki, Appl. Phys. Lett. 5, 160, (1964).

[9] M.V. Kot, L.M. Panasyuk, A.V. Simashkevich, A.E. Tsurkan and D.A.
Sherban, Sov. Phys.-Solid State 7, 1001, (1965).

[10] S. Fujita, S. Arai, F. Itoh and T. Sakaguchi, J. Appl. Phys. 46, 3070, (1975).

]11] J.O. McCaldin and T.C. McGill, J. Vac. Sci. Technol. B 6, 1360, (1988).

39
[12] E.T. Yu, E.T. Crooke, D. H. Chow, D. A. Collins, M.C. Phillips and T.C.
McGill, J. Vac. Sci. Technol. B 8, 908, (1990).

[13] J. Han, T.S. Stravrinides, M. Kobayashi, R.L. Gunshor, M.M. Hagerott and
A.V. Nurmikko, J. Elec. Mater. 22, 485, (1993).

[14] S.Y. Ren, J.D. Dow and J. Shen, Phys. Rev. B 38, 10677, (1988).

[15] J.D. Dow, R.D. Hong, S. Klemm, S.Y. Ren, M.H. Tsai, O.F. Sankey and R.V.
Kasowski, Phys. Rev. B 43, 4396, (1991).

[16] N. Samarth, H. Luo, J. Furdyna, S. Qadri, Y. Lee, A. Ramdas and N. Otsuka,
J. Elec. Mater. 19, 543, (1989).

[17] H. Luo, N. Samarth, F. Zhang, A. Pareek, M. Dobrowolska, J. Furdyna, K.
Mahalingam, N. Otsuka, W. Chou, A. Petrou and 8. Qadri, Appl. Phys. Lett.
58, 1783, (1991).

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

[19] M. Wang, M. Phillips, J. Swenberg, E. Yu, J. McCaldin and T. McGill, J.
Appl. Phys. 73, 4460, (1993).

[20] M. Phillips, M. Wang, J. Swenberg, J. McCaldin and T. McGill, Appl. Phys.
Lett. 61, 1962, (1992).

[21] M. Wang, J. Swenberg, M. Phillips, E. Yu, J. McCaldin R. Grant and T.
McGill, Appl. Phys. Lett. 64, 3455, (1994).

[22] S. Parker, A. Reinberg, J. Pinnell and W. Holton, J. Electrochem. Soc. 118,
979, (1971).

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

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

[25] W. Strehlow and E. Cook, J. Phys. C 2, 163, (1973).

Chapter 3

Molecular Beam Epitaxy, and

Nitrogen Doping of ZnTe

3.1 Introduction

Molecular beam epitaxy (MBE) offers a number of advantages as a crystal growth
technology for the growth of II-VI compounds and the graded injector device struc-
ture. Elemental and/or compound source materials with purities of 99.99999% or
greater are available. The ultra-high vacuum (UHV) growth condition of MBE
enables the growth of layers with very high purity as well as allowing for zn situ
erowth analysis. A residual gas analyzer (RGA) detects the constituent atomic
and/or molecular species in the UHV environment. Reflection high energy elec-
tron diffraction (RHEED) provides surface analysis of epitaxial films during MBE
growth. The low growth temperatures, as low as 150 °C for ZnSe, and the pre-
cise monolayer control of beam fluxes enable the formation and control of abrupt
heterojunction interfaces with interdiffusion and reactions between III-V/II-VI or
IJ-VI/II-VI junctions limited to a few atomic layers. MBE is well suited for growth
of the graded injector device not only for its material purity and low temperature

growth, but also for its thickness and compositional control which are necessary

for our device structure. Other techniques, such as metal organic chemical vapor
deposition (MOCVD), are capable of the low temperature epitaxial growth but
they require more understanding and control of the chemical reactions and gas
flows. Furthermore MBE is considerably safe when compared to MOCVD, which
can require the use of toxic gas sources.

One common problem with the growth of II-VI semiconductors has been the
unavailability of high quality substrates. This problem is partially mitigated by
using lattice matched III-V substrates, which are commercially available and of
high structural quality. For instance, for the growth of ZnTe we can use GaSb
substrates with a lattice mismatch of 0.128%. There are two subtle, yet noticeable
problems associated with this approach. First, there is measureable cross contam-
ination due to the out-diffusion of group III and V elements. Second, there are
interfacial chemical reactions at the III-V/II-VI heterojunction.

Since the constituent atoms of a III-V compound are dopants in II-VI materials
the substrate elements can diffuse out of the HI-V layer and into the II-VI material
during growth. This will result in doping compensation and the formation of im-
purity states. Rajakarunanayake et al. [1] found the out-diffusion of group II and
group V impurities in the growth of epilayers of ZnTe on various III-V substrates.
Photoluminescence (PL) from the ZnTe exhibited characteristic impurity bound
excitons due to the cation out-diffusion. Secondary-ion mass spectroscopy and
electron microprobe analysis showed substantial out-diffusion of the III-V cations
and anions into the ZnTe with higher concentrations at the surface of the ZnTe
than in the epilayer. It was also seen that there was no significant back diffusion
of Te into the III-V layers. In our studies of ZnTe growth on GaSb substrates we
observed significant Sb out-diffusion, detected by XPS, with roughly a monolayer
of Sb riding up along the growth surface. This isn’t surprising since Sb is known to
be and is used as a surfactant during MBE growth [2, 3]. Although out-diffusion is
not ideal, it can be somewhat reduced and tolerated so long as it doesn’t prevent

doping or hinder the II-VI epilayer structural quality.

A more severe consequence of II-VI/III-V heteroepitaxy is the formation of
interface compounds. Tu and Kahn [4] observed the formation of an interfacial
layer (~5 A) between ZnSe and GaAs during MBE, which was later determined by
XPS to be GaySe [5]. A similar interfacial reaction was found for the ZnTe/GaSb
heterojunction. XPS [6], raman scattering [7] and high resolution electron mi-
croscopy [8] confirmed the formation of GagTe3 at the ZnTe/GaSb interface during
MBE growth. Duddles et al. [9] found an anomalous blue shift of the heavy-hole
exciton that could not be explained by the lattice mismatch between ZnTe and
GaSb. They suggested that the interfacial GagTeg contributed to additional strain.
We find the presence of this interfacial IIJ-VI layer degrades the subsequent layer
quality.

Petruzzello et al. [10] found that the structural quality of ZnSe grown on GaAs
was strongly dependent on the specific growth initiation. With an initial exposure
of Zn-only prior to growth initiation, the interface layer and epitaxial film was of
high structural quality. On the other hand, an initial exposure of Se resulted in
films of significantly lower quality. This demonstrated not only that the interface
compound layer affects the epitaxial growth quality, but also that it may be possible

to control the degree of its effect.

3.2 Overview and outline

Our MBE system consists of three separate chambers, interconnected via UHV
transfer tubes. Two of the chambers are Perkin-Elmer 430P MBE systems, one
dedicated to III-V only and the other to H-VI only growths. The third chamber
is a Perkin-Elmer Model 5500 analysis system used for x-ray photoelectron spec-
troscopy (XPS). Elemental source materials are used in conventional MBE ovens.
The base pressures in the chambers during growth or analysis are typically in the
low 10~'° Torr range, although during nitrogen doping the pressures are as high as

10-6 Torr. Typical substrate growth temperatures during II-VI growth are 300 °C,

and fluxes are set for growth rates of ~ 1 A per second. Optimal MBE growth of
ZnTe was found to occur for a flux ratio of Zn to Te which is slightly “Te-rich” and
exhibiting the characteristic (2 x 1) RHEED reconstruction. The transfer tubes
allow for sample manipulation between the separate chambers without remov-
ing from the UHV (107! Torr) environment. By interconnecting these different
chambers we are capable of growing II-VI/III-V heteroepitaxial samples as well as
offering additional in situ (XPS) materials characterization, while avoiding cross
chamber contamination.

Early attempts at fabricating a graded injector LED based on the graded in-
jector heterojunction were severely limited by the lack of high quality ZnTe sub-
strates. Before the successful p-type doping of MBE ZnTe, the early devices had
to be grown on poor quality p-ZnTe, and we had to rely on the p-typeness of the
substrate with only a thin buffer layer for device operation. With the success of
p-type doping of ZnTe, detailed in this chapter, we could then grow high conduc-.
tivity layers of p-Zn'Te and use the closely lattice matched GaSb as a substrate.
Although GaSb substrates offer a partial solution to the substrate deficiency, we
found it necessary to control the II-VI/III-V interfacial reaction. By using a thin
intermediary AlSb buffer we could suppress the II-VI/III-V interfacial reaction and
grow considerably higher quality ZnTe epilayers.

In this chapter we discuss the issues and details of the growth technologies
used to fabricate the graded injector LED. A discussion of the substrates used for
MBE growth is presented with the emphasis on the ability to use commercially
available GaSb. In addition, nitrogen doping of ZnTe is presented along with
the technique used to achieve high quality epilayers with adequate doping levels
necessary for device operations. Lastly the growth technique employed to construct

the Mg,Cd,_,Se graded region is discussed and evaluated.

3.3. Substrate choice

3.3.1 ZnTe

Ideally, we would prefer homoepitaxy over heteroepitaxy to avoid interfacial reac-
tions and any residual strain. Unfortunately, high quality ZnTe substrates have not
been available. Single-crystal ZnTe substrate were obtained from Eagle-Picher [11].
A proprietary seeding technique was used to nucleate crystal growth on a substrate
and bulk substrate growth proceeded via a physical-vapor-transport. Crystalline
quality has varied, with x-ray FWHM ranging from 70-110 arcsecs. Polishing and
surface preparation technology has not been as advanced as for the III-V semi-
conductors, and polish damage was visually apparent on the ZnTe substrates.
Furthermore, substrate preparation and deoxidation of ZnTe substrates has been
a considerable problem. We have been unsuccessful in desorbing oxygen bound
to Zn on the surface. This residual surface oxide nucleates twins during growth.
Ar ion sputtering can remove the surface oxide, but the sputtering damage also
leads to growth defects nucleating at the defect sites on the surface. Annealing
the substrates at 460 °C prior to growth improves the surface structure, as seen
in the RHEED pattern. This annealing step was found to lead to better growth
nucleation than the unsputtered substrates so all growths on ZnTe are sputtered
and annealed. In spite of this procedure, the surface morphology of thick epi-
layers of ZnTe grown on these ZnTe substrates is generally poor compared to
high quality heteroepitaxy on GaSb. These thick layers on ZnTe substrates have
much higher surface roughness than growths on GaSb, and exhibit hillock forma-
tion nucleating at the substrate/epilayer interface [12]. The electrical (J-V) and
optical (electroluminescence and quantum efficiency) properties of devices grown
on ZnTe substrates were greatly outperformed by comparable devices grown on
GaSb. Therefore, ZnTe substrates were used primarily for growths of characteri-

zation epilayers rather than for devices, and GaSb substrates were used for device

growths.

3.3.2 GaSb

For lack of a homoepitaxial technique to ZnTe growth, GaSb offers an improved
alternative for a substrate material. GaSb is a closely latticed match II-V binary
substrate commercially available with reasonable quality. The ZnTe/GaSb lattice
mismatch is 0.128%, which is better than that of ZnSe/GaAs (0.254%) and that
of AlAs/GaAs (0.132%). GaSb substrates used have typical x-ray FWHM of 18
arcsec. High quality MBE GaSb is well established, so epilayers of GaSb can
be grown for later IJ-VI epitaxy. These features make GaSb an ideal candidate
substrate for II-VI ZnTe epitaxy. Additionally since we can now dope ZnTe p-
type, GaSb can be a substrate for device applications. The valence band of ZnTe
is 0.60 eV lower than the valence band of GaSb [13], so a p-type contact through
the substrate is difficult. Fortunately, ohmic contacts to p-ZnTe can be made from
the top side laterally to a device.and serve as a back contact.

Unfortunately, the interfacial reaction between ZnTe and GaSb forming GagTes
can result in epilayers with high defect densities and poor structural and surface
quality. Typical x-ray FWHM for epilayers of ZnTe on GaSb where > 100 arcsecs.
The interface reaction appears to be slightly supressed by initiating growth with
an initial Zn flux similar to the technique of Petruzzello et al. [10]. Since the Ga-Sb
and Zn-Te bonds are respectively weaker than the Ga-As and Zn-Se bonds, this
technique is less effective than with ZnSe growth on GaAs. A better technique
employed is to grow a thin (~ 20 A) buffer of AlSb on the GaSb epilayer. The
AlSb has a stronger chemical bond and the formation of AlyTe3 at the AlSb/ZnTe
interface is reduced. This result was observed in the RHEED pattern during the
initial growth nucleation. For samples without the AlSb, the RHEED pattern
becomes diffused immediately upon growth of ZnTe and then becomes slightly

spotty and streaky after roughly 4-6 seconds. For samples with the AlSb, the

RHEED pattern is less diffused and slightly spotty at the onset of ZnTe growth.
This at least qualitatively demonstrates a reduced interfacial reaction with the
AlSb buffer. In addition, higher quality ZnTe epilayers where grown using an
AlSb buffer. Best x-ray FWHM of 39 arcsec and typical x-ray FWHM of 40-50
arcsec were achieved using the AlSb buffer. Therefore, all II-VI growths on GaSb
substrates had the thin AlSb buffer layer.

3.4 Nitrogen doping of ZnTe

Bulk grown ZnTe, grown with an excess of Te will be p-type due to the formation
of the double acceptor Zn vacancy. We would expect ZnTe to be easily doped
p-type by MBE, but there has been considerable difficulty achieving high quality
p-type ZnTe. Undoped MBE-grown ZnTe is naturally p-type but highly resistive.
During MBE growth, excess Zn or Te can re-evaporate from the surface resulting
in stoichiometric material with high resistivity. A number of early attempts to
dope ZnTe during MBE growth have had only limited success. Antimony dop-
ing has been limited both by low solubility and low activation efficiency, with
maximum p-type carrier concentrations of 3x10'°cm~? achieved without degrad-
ing the films [14]. Elemental phosphorus doping with beam equivalent pressures
comparable to the Zn and Te sources has attained carrier concentrations up to
4x10'"cm~%, but with deteriorated crystalline quality [15]. More efficient p dop-
ing has been achieved using Zn3Asg for arsenic doping with hole concentrations
up to 1x10'8cm~* [16]. But as with the phosphorus doping the beam equiva-
lent pressure was greater than for the Zn and Te. More limiting, though, is the
formation of deep levels for p > 1x10'"cm~3, and severely degraded band edge
photoluminescence in favor of deep emission due to arsenic incorporation.
Recently, nitrogen has shown to be an effective p-type dopant in ZnSe [17]. Re-
sistivities below 1 Q-cm [18] have been grown, however net acceptor concentrations

saturate at 2x10'8cm73 under optimal growth conditions for ZnSe [19]. Further

nitrogen incorporation can be increased to as high as 1.3x10'%cm~%, but the net
acceptor concentration does not increase due to compensation. The incorporation

3 was further shown to affect the structural

of nitrogen higher than 1x10'8cm7
properties of ZnSe attributed to the formation of point defects accompanying the
N doping [20].

The successful nitrogen doping has been accomplished by forming a nitrogen
plasma. Most researchers have employed an Oxford Applied Research (OAR) [21]
radio frequency (RF) plasma source. Another possible source studied has been an
electron cyclotron resonance (ECR) plasma source [22]. Vaudo et al. [23] have in-
vestigated the optical emission spectra of both RF and ECR sources and identified
atomic species in the nitrogen plasma as the most likely species responsible for
p-type doping. Furthermore, it was shown that the RF source produces a larger
percentage of nitrogen atoms for a given set of operating condition as well as dop-
ing to higher free hole concentrations than the ECR source. Therefore we suspect
that an RF plasma source will be a preferred efficient atomic nitrogen source for
p-type doping of H-VI compounds.

Soon after the accomplishment of p-type doping of ZnSe with a nitrogen plasma
source, p-type doping of ZnTe was demonstrated. Han et al. [24] grew ZnTe:N
epilayers on (100) GaAs susbstrates with acceptor concentrations approximately
1x10'%cm7? and x-ray FWHM of 270 arcsec. We successfully doped ZnTe grown
on ZnTe substrates using a nitrogen plasma source with hall carrier concentrations
up to 1x10%®cm~? [25]. Tao et al. [26] also acheived doping levels of 1x10?°em™?
in Zn'Te by growing on tilted (311)B GaAs substrates. On (100) GaAs substrates
they measured hole concentrations of 8.9x10'%cm~? under identical growth condi-
tions. The successful doping of ZnTe to these levels has facilitated the fabrication
of low resistance “pseudograded band gap” contacts to p-ZnSe formed with a
ohmic contact to ZnTe:N graded with Zn(Se,Te):N to ZnSe:N [27]. Additionally,
p-type ZnTe has led to significant advancements in the development of the “graded

injector” [25] visible light emitter discussed later in this thesis.

3.4.1 Nitrogen plasma source

Nitrogen doping is achieved by using an OAR Atom/Radical-Beam Source, Model
MPD21. This source fits in the standard effusion cell of the MBE machine. A
Nanochem Series L-60 gas purifier with Nanochem 1400 resin is used to reduce
any impurities of 02, H,0, and C0, in the gas source to the ppb (parts per billion)
levels. The gas flow rate is set with a precision needle valve. A plasma is struck in
the PBN (pyrolytic boron nitride) discharge tube by an electrical discharge created
from inductively-coupled RF excitation at 13.56 MHz. Atomic and radical species
produced in the discharge can escape into the vacuum chamber via an array of
fine holes in the beam aperture plate. Two apertures with different number of
pin holes and pin hole diameters have been used in our system: a 21 x 0.3mm (
number x diameter) array with an effective area of 1.48 mm?, and a 4 x 0.2mm
array with an effective area of 0.13 mm?. The size and number of holes in the
aperture sets the aperture conductance and therefore controls the beam flux and
flow rates necessary for plasma discharge. For instance, with the larger array a Ne
flow > 0.05 sccm is necessary for plasma operation, while with the smaller array a

No flow > 0.01 sccm is needed [28]. As will be shown below the smaller array was

used to reduce the dopant flux.

3.4.2 Nitrogen doping of ZnTe

An ultra high purity (99.9995%) nitrogen gas source was used in our early doping
experiments. The major impurity Ar, which is inert, had levels < 5 ppm (parts
per million). Other contaminants are effectively reduced by the Nanochem purifier.
The plasma source had the 21 x 0.3mm aperture installed. The nitrogen flow rate
was adjusted for these experiments to yield a background chamber pressure of
1x10~° Torr, and the plasma was operated at 125W with no reflected power.
Epilayers of nitrogen doped ZnTe were grown on (100) ZnTe substrates to study

the electrical characteristics of ZnTe:N. Growth conditions were typical for MBE

ZnTe epilayers previously discussed. The nitrogen doped layers were each grown
to approximately 2 um after the growth of roughly 1 wm of undoped ZnTe. The
electrical properties of the ZnTe epilayers were evaluated by Hall measurements.
Dopant concentrations in all experiments were sufficiently high to form ohmic
contacts with all metals and contacts used (Au, Au/Ge, silver paint and indium
solder).

Doped layers of ZnTe:N grown under these conditions were extremely conduc-
tive with resistivities of 107? Q-cm, and hole carrier concentrations of 1x107°cm7?.
We were unable to significantly reduce the nitrogen beam flux, while maintaining
a plasma, for the given nitrogen source aperture. Therefore, to reduce the aver-
age doping concentration, we modulated the doping in a superlattice type fashion
by repeatedly growing periods of undoped ZnTe followed by doped ZnTe:N. Fig-
ure 3.1 shows the results of Hall measurements of three different epilayers all grown
to roughly the same thickness: one continously doped ZnTe:N, and two different
modulated doped layers. One modulated doped layer has 40 periods of 500A
ZnTe/ 50A ZnTe:N. The other modulated doped layer has 40 periods of 500A
ZnTe/25 A ZnTe:N. The lack of any temperature dependence is characteristic of
fully degenerately doped material. In this regime, wave functions of N acceptor
holes overlap and an impurity band is formed.

The modulated doping effectively reduces the average carrier concentration
and increases the resistivities of the epilayers by up to 2 orders of magnitude.
More importantly, the modulated doping improved the surface morphology of the
ZnTe epilayers. The surface crystallinity was examined during growth by in situ
RHEED. The RHEED shows the distinct (2 x 1) streaky “Te-rich” pattern char-
acteristic of normal two-dimensional ZnTe growth. In the case of the continous
doping, an elonged (2 x 1) spotty pattern developes as the layer grows indicating
three-dimensional growth and degraded surface morphology. On the other hand, in
the case of the modulated doping, the RHEED stays streaky and the surface mor-

phology is improved over the continous doped case. This improvement is quantified

10 —— Continuously doped ZnTe:N
on (a) —— Modulation doped 50/5 nm
, Ee —--= Modulation doped 50/2.5 nm
(s)
aad 1 Qo”
2Q
rs petra
= -—-
5 10°
Oo poorer mm ame Ow ee me ~~ —_
10° . i i ry i — |
0 50 100 150 200 250 300
q — Continuously doped ZnTe:N
(b) —— Modulation doped 50/5 nm
Cant —-+- Modulation doped 50/2.5 nm
E10) - —---—- eee e
2? no cone ay emt om _
= one
'S 10° | ~-—--— nr re
5 6 COE
rc
10°
T fs.

0 50 100 150 200 250 300
Temperature (K)

Figure 3.1: Hall measurement of nitrogen doped ZnTe epilayers as a function of
temperature for (a) carrier concentration and (b) resistivity [29]. The 40 period
modulated doped period thicknesses are given as undoped/doped in nanometers.
All epilayers are roughly 2 microns of doped material grown on 1 micron undoped

insulating ZnTe buffer layers.

by atomic force microscopy (AFM) and average root mean square (RMS) surface
roughness [12]. These structural improvement results are very similar to planar
doping techniques employed with chlorine doping of ZnSe by MBE [30].

The structural and growth problems associated with these degenerate doping
levels may in large part be due to the large lattice strain associated with nitrogen
incoporation. Nitrogen p-type doping is substitutional on a Te site. At the doping
incorporation levels in these experiments, 1x 10?°cm~3, roughly 0.5% of the Te sites

are substituted with N. The tetrahedral radii of N is 0.70 A while that of Te is 1.32 A

Experimental
---- Calculated
] Substrate. pprteznten Superlattic
w . ;
wee” q
> al .
74 y
Cc 5 ut ]
of . NA
i= | ings
= P é fl } a
j a i 4 4
| vl i t f i be
y 4 A 4 i H
a ‘ .@ i Hh
7 it rie P oa t
M1 t \r ‘ | . {
-! f att a
tals i ' at t
29.7 299 30.1 303 305 30.7 309

Omega-2Theta

Figure 3.2: 0/2 @ measured and calculated x-ray diffraction patterns for a 40
period ZnTe(240 A )/ZnTe:N(15 A ) layer grown on ZnTe. Calculated pattern is
generated by Philips PC-HRS software [32]. The lattice parameter for ZnTe:N is

estimated using the tetrahedral radii of nitrogen and the doping concentration.

[31]. Therefore the doping will lead to large local strains. Figure 3.2 illustrates this
effect. The X-ray diffraction for a modulated doped layer with 40 periods of 240 A
ZnTe/15 A ZnTe:N doped to 1x10?°cm-? grown on ZnTe substrates is compared
to a calculated diffraction pattern. A superlattice-like x-ray pattern is observed
due to the periodic lattice strain due to the nitrogen doping.

Graded device structures were grown using the modulated N-doped ZnTe on
GaSb substrates. These devices operated successfully as discussed in the following
chapter, but due to the degenerate doping the optical performance was degraded.

Photoluminescence from modulated doped ZnTe:N was less efficient than for un-

doped ZnTe. Therefore, it was necessary to reduce the doping levels significantly
to improve the structural and optical quality of the ZnTe:N.

To reduce the doping levels the aperture on the nitrogen source was changed
to the 4 x 0.2mm aperture. The effective area of this aperture is one order of mag-
nitude smaller than the 21 x 0.3mm aperture, thus reducing the flux equivalently.
Continuously doped epilayers of ZnTe:N, grown under similar growth conditions as
mentioned above, had room temperature Hall hole carrier concentrations reduced
to 2x10'%cm-*. These layers showed improved photoluminesce over the higher

doped layers but were still degraded as compared to undoped ZnTe layers.

3.4.38 Dilute nitrogen doping of ZnTe

To further improve the structural quality of the ZnTe:N we reduced the doping
levels by changing the nitrogen gas source to an ultra high purity (99.9995%) dilute
mixture of 10% Nz / 90% Ar. The plasma could be struck with a similar gas flow
rate and background chamber pressure. Therefore the nitrogen flux is effectively
reduced by one order of magnitude. Continously doped epilayers of ZnTe:N grown
using the dilute nitrogen and 4 x 0.2mm aperture under the above growth condi-
tions had room temperature Hall hole carrier concentrations of 1x10!%cm~3. These
levels proved sufficient for ohmic contacting and epilayer conductivity. Further-
more, these ZnTe:N layers exhibited greatly improved photoluminescence over the
higher doped layers and had PL comparable to undoped ZnTe layers. In addition,
the structural quality of the ZnTe:N films grown under these conditions were as
good as undoped epilayers. Figure 3.3 shows the x-ray rocking curve for an epilayer
of ZnTe:N grown on a GaSb substrate with an intermediary 20 A AlSb layer. This
ZnTe:N epilayer had the best x-ray FWHM of 39 arcsec for all ZnTe growths on
GaSb substrates. Therefore for optimal doping and materials quality we used the

dilute Ng source with the smallest aperture for subsequent device growths.

Intensity (a.u

Pe ee ee ee ae ee a ee

30,2 — 30.3 30.4 30.5
Omega (degrees)

Figure 3.3: X-ray rocking curve of 1.0um ZnTe:N epilayer doped with an RF
plasma source using a dilute gas mixture of 10% Nz / 90% Ar grown on GaSb
substrate. The room temperature free hole carrier concentration is 1x 10'8cm7?.
The growth prior to the II-VI layer consists of 0.6m GaSb:Si buffer layer followed
by a 20A AlSb blocking layer.

3.5 MBE growth of graded junction

A brief description of the technique employed to grow the graded junction region
(i.e. the Mg,Cdi_,Se region) is presented now so as to better understand some
of the complications discussed in the following chapter. A standard effusion cell
loaded with elemental ingots was used for the Mg source. The Mg concentration
in epilayers of Mg,Cdj_,5e was measured by x-ray photoemission spectroscopy

(XPS). These measurements were initially checked and calibrated with electron

microprobe analysis. It was found that under identical growth conditions (sub-
strate temperature, and source fluxes) the Mg concentration in Mg,Cdi_,Se was
consistently reproducible. Nevertheless, prior to growing a device we would grow
an epilayer of Mg,Cd,_,Se on a ZnTe substrate to measure the Mg concentration.
This measured value was then taken to be the peak Mg concentration in the graded

Mg,Cd,_,Se region.

3.5.1 Thermal grading of Mg,Cd,_,Se

Presently, the graded Mg,Cd,_,Se region of the device is grown by simply shut-
ting off the power to the Mg source cell, while keeping other sources static. The
subsequent thermal cooling reduces the Mg flux and thus decreases the concentra-
tion in the Mg,Cd,_,Se region as the layer grows. The resulting grading profile
is thus determined by the flux transient of the Mg. The Mg flux transient was
characterized by monitoring the Mg peak in the RGA spectrum. The transient
behavior was the same throughout all measurements. The CdSe growth rate was
adjusted so that the whole grading would occur over ~ 300A, and so that the
Mg concentration would decrease by roughly one order of magnitude. For our
particular cell and typical source material mass we noticed a consistent 1.5 minute
delay after shutting off the power, before the Mg flux would begin to decrease.
Once the Mg flux would start to decrease it would drop exponentially with time.
Therefore, the Mg cell’s power was shut off 1.5 minutes prior to the beginning of
the Mg,Cd ,_,Se graded layer. Overall, we assume that the Mg concentration in
the Mg,Cd,_,Se layer starts at the initial measured calibration maximum, and
ended up one order of magnitude lower over a total thickness of 300 A. This tech-
nique is fundamentally limited and constrained by the thermal transient properties
of the Mg source material, cell and oven. A discussion of alternative approaches
will be presented in chapter 4. Despite its crude approach, this grading technique
has worked sufficiently to grow graded Mg,Cd _,Se layer devices capable of light

emission.

3.6 Summary

Successful p-type doping of MBE ZnTe using an RF plasma nitrogen source has
enabled the use of GaSb substrates for the graded injector device discussed in
the following chapter. To realize high quality ZnTe:N we employed the smallest
aperture for the N-source and used a dilute 10% N2/ 90% Ar mixture to achieve

doping levels of 1x10'8cm73,

Epilayer of ZnTe:N grown under these conditions
exhibited room temperature PL comparable to undoped ZnTe. II-VI epitaxy on
GaSb substrates is greatly improved by growing an intermediary (~ 20 A) AlSb
buffer to suppress the II-VI/III-V interfacial chemical reaction. This procedure
has led to the highest quality epitaxial ZnTe:N layers grown on GaSb with x-ray
FWHM of 39 arcsec. This heteroepitaxial technique and I-VI doping have been

implemented to realize working graded injector LEDs.

Bibliography

[1]

Y. Rajakarunanayake, B.H. Cole, J.O. McCaldin, D.H. Chow, J.R.
Sdderstrom, T.C. McGill, C.M. Jones, Appl. Phys. Lett. 55, 1217 (1989).

B. Voigtlander and A. Zinner, J. Vac. Sci. Technol. A 12, 1932 (1994).

H.J. Osten, G. Lippert and J. Klatt, J. Vac. Sci. Technol. B 10, 1151 (1992).
D.W. Tu and A. Kahn, J. Vac. Sci. Technol. A 3, 922 (1985).

D.R. Menke, J. Qiu and R.L. Gunshor, J. Vac. Sci. Technol. B 9, 2171 (1991).
W.G. Wilke and K.Horn, J. Vac. Sci. Technol. B 6, 1211 (1988).

M.P. Halsall, D. Wolverson, J.J. Davies, B. Lunn and D.E. Ashenford, Appl.
Phys. Lett. 60, 2129 (1992).

C.T. Chou, J.L. Hutchison, D.Cherns, M.J. Casanove, J.W. Steeds, R. Vin-
cent, B. Lunn and D.A. Ashenford, J. Appl. Phys. 74, 6566 (1993).

N.J. Duddles, J.E. Nicholls, T.J. Gregory, W.E. Hagston, B. Lunn and D.A.
Ashenford, J. Vac. Sci. Technol. B 10, 912 (1992).

[10] J. Petruzzello, B. Greenberg and J. Gaines, Bull. Am. Phys. Soc. 35, 237

(1990).

[11] Eagle-Picher Research Laboratory, Miami, Oklahoma.

58
[12] R. Miles, J. Swenberg, M. Wang, M. Phillips and T. McGill, J. Cryst. Growth
138, 523, (1994).

[13] E. Yu, M. Phillips, D. Chow, D. Collins, M. Wang, J. McCaldin and T. McGill,
Phys. Rev. B 46, 13379 (1992).

[14] R. Feldman, R. Austin, A. Sher, M. Schnoes, S. Downey, A. Emerson, T.
Harris, R. Spitzer, G. Gualtieri and G. Schwartz, J. Cryst. Growth 118, 295

(1992).
(15| Y. Hishida, H. Ishii, T. Toda and T. Niina, J. Cryst. Growth 95, 517, (1989).

(16] F. Turco-Sandroff, M. Brasil, R. Nahory, R. Martin, Y. Zhang and B.
Skromme, Appl. Phys. Lett. 59, 688, (1991).

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

[18] R. Park, M. Troffer, E. Yablonovitch and T. Gmitter, Appl. Phys. Lett. 59,
1896, (1991).

[19] J. Qiu, H. Cheng, J. Depuydt and M. Haase, J. Cryst. Growth 127, 279,
(1993).

[20] J. Petruzzello, J. Gaines, P. van der Sluis, D. Olego and C. Ponzoni, Appl.
Phys. Lett. 62, 1496, (1993).

[21] Oxford Applied Research, Crawley Mill, Witney, Oxfordshire OX8 5TJ, UK
[22] T. Ohtsuka and K. Horie,Jpn. J. Appl. Phys. 32, L233, (1993).
[23] R. Vaudo, J. Cook and J. Schetzina, J. Cryst. Growth 138, 430, (1994).

[24] J. Han, T. Stravrinides, M. Kobayashi, R. Gunshor, M. Hagerott and A.
Nurmikko, Appl. Phys. Lett. 62, 840, (1993).

59
[25] J. Swenberg, M. Wang, R. Miles, M. Phillips, A. Hunter, J. McCaldin and T.
McGill, J. Cryst. Growth 138, 692, (1994).
[26] I. Tao, M. Jurkovic and W. Wang, Appl. Phys. Lett. 64, 1848, (1994).

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

[28] Specification sheet for Model MPD21 Atom/Radical Beam Source, Oxford
Applied Research, Crawley Mill, Witney, Oxfordshire OX8 5TJ, UK

[29] Temperature Hall measurements were performed by Dr. A.T. Hunter of
Hughes Research Laboratories, Malibu, California 90265. Room temperature
and liquid nitrogen temperature measurements were performed at Caltech by

author.
[30] Z. Zhu,H. Mori and T. Yao, Appl. Phys. Lett. 61, 2811, (1992).
(31] Gerald Burns, Solid State Physics (Academic Press, Inc., 1985), pp. 163.

[32] PC-HRS Version 1.0, High resolution simulation program for PC, Dr. P. Few-
ster, Philips Research Laboratories, Redhill U.K.

Chapter 4

Development of the Graded
Injector Light Emitter

4.1 Introduction and outline

The early graded injector devices were severely limited by the lack of a MBE p-
type dopant for ZnTe, a high quality substrate, and the lack of sufficient materials
characterization of Mg,Cd,_,Se necessary for a realistic approach to device engi-
neering. In spite of these shortcomings, the early devices fabricated using ZnTe
substrates appeared to work as LEDs but only under high biases and a restrictive
device design [1].

In this chapter, we report on advances in the development of the graded junc-
tion electron injector device, i.e. the graded injector. Materials characterizations
of the Mg,Cd,_,5e alloy necessary for device engineering issues are presented.
Success in p-type doping of ZnTe, discussed in the previous chapter, has subse-
quently enabled the use of higher quality GaSb substrates. Devices grown with
GaSb substrates show the benefits of high quality substrates. Present electrical
characteristics exhibiting normal diode-like behavior more clearly demonstrate the

operating principle of the graded injector over the previous studies of Phillips et

al. [2]. Electroluminescence (EL) characteristics of the graded injector LED are
presented along with a discussion of the the optical performance. These results
lay the foundation for and highlight the salient issues for further studies needed to

advance the graded injector device development to a point of commercial interest.

4.2 Device engineering

There have been very few studies of the epitaxial growth of Mg chalcogenides.
Early studies of Itoh [3] synthesized p-n diodes based on Mg,Cdj_,Te crystals
grown by the Bridgman-Stockbarger method. Okuyama et al. [4] grew and char-
acterized epilayers of Zn,Mg;_,5,Se;_, and proposed and subsequently demon-
strated [5] the use of this quaternary as lattice matched cladding layers in blue-
green laser diodes. More recently, MgTe and Mg,Cd,_,Te thin films grown by
MBE have been reported [6]. But there have been no reported studies of Mg,Cdi_,Se
alloy epilayers. The original design of the graded Mg,Cdi_,Se region was made
based on approximations of the material parameters due to this lack of character-
ization.

For efficient operation of the graded injector, the conduction band must be
properly graded to avoid any energy barrier to electrons at the Se/Te interface.
Ideally we would want to continuously grade the conduction band from that of
ZnTe down to CdSe. Device simulations suggest that small discontinuities in the
conduction band may be acceptable for device operation, but we would like to
reduce and eliminate any electron barrier at the Mg,Cd _,Se/ZnTe interface. In
fact it would be prefered that the conduction band of Mg,Cd ,_,Se be graded
above that of ZnTe rather than below. This has the unfortunate consequence
of introducing more strain due to the higher Mg concentration. So we would
like to be as close as possible to a continuous grading at the Mg,Cdi_,Se/ZnTe
interface. To accomplish this, it is necessary that we know the position of the

band edges of Mg,Cd,_,Se as a function of Mg concentration. To do this, we

need the band gap as a function of Mg concentration and the relative valence band
offset. The estimates of the band gap of MgSe range from 3.6 to 5.6 eV (4, 7].
Therefore we need an accurate measure of the Mg,Cd,_,Se band gap as a function
of Mg concentration. Furthermore, there had been no previous studies of the
valence band offsets for the Mg chalcogenides, and we would expect a deviation
from the common anion rule similar to that observed in the AlAs/GaAs system,
since Mg like Al has unoccupied d orbitals (8, 9]. Therefore it becomes critical to

experimentally measure the valence band offset of the Mg,Cd,_,Se/CdSe system.

4.2.1 Band gap studies of Mg,Cd,_,Se

Pseudomorphic epilayers of cubic Mg,Cd,_,Se, with x ranging from 0 to 63%,
were grown on (100) ZnTe substrates with thicknesses of at least 1000 A. The Mg
concentration was reproducibly grown and measured by XPS, after calibrating with
electron microprobe analysis. The RHEED pattern during the Mg,Cd,_,Se growth
exhibited a Se-rich (2 x 1) streaky reconstructed surface. The RHEED pattern
showed no indications of any deviation from cubic material even for layers grown
slightly beyond the critical thickness. X-ray diffraction patterns of Mg,Cd,_,Se
epilayers also indicated a cubic crystalline structure, although with large FWHMs
and a high degree of crystalline mosaicity.

A characteristic photoluminescence (PL) spectra from a Mg,Cdy_,Se epilayer
is represented in Figure 4.1. A He-Cd laser (325nm) with 1mW power was used
for the excitation source. The broad high energy peak is thought to be a donor-to-
acceptor pair (DAP) emission from the Mg.4gCd.sgSe epilayer. Since the epilayer
is grown beyond the critical thickness and has poor structural quality, there is no
excitonic emission. Additionally, there is increased deep emission associated with
the Mg 4¢Cd_ 56Se layer.

The band gap energy of the epilayers was estimated from the high energy peak

band edge emission in the PL spectrum at 5K. We could not use the room temper-

eo}
aS
?)
nes
an
ro
WM
i)

2 L A
_~ 4
ZnTe
Exciton

1.5 1.7 1.9 5 2.7 2.9

2.1 2.3 2
Energy (eV)

Figure 4.1: Low temperature (5K) photoluminescence (PL) spectra of 1000 A
Mg4gCds5g5e epilayer grown on ZnTe substrate. 1ImW excitation from HeCd
(325nm) laser. An excitonic peak from the underlying ZnTe is indicated. The
high energy broad peak is from the Mg 4gCd_56Se layer. This peak is thought to be
a donor-to-acceptor pair (DAP) emission. No excitonic emission is observed from
the Mg 4g¢Cd_56Se epilayer due to the poor structural quality of the layer grown past
the critical thickness. This may also be responsible for the deep emission which

would be reduced under more optimal growth conditions.

ature PL spectra due to severely degraded emission for the high Mg concentration
layers. The DAP emission is a lower bound for the band gap at 5K. We expect
the band gap to be greater than 100 meV above the DAP emission based on the
deep acceptor states of the wide band gap II-VIs associated with the heavy hole
mass in the valence band. At room temperature, the band gap would be from
100-140 meV less than the 5K band gap, which is typical of the H-VIs. So we can
conservatively regard the band edge emission as the band gap for these alloys. For
the low concentration Mg layers this estimate was found to be consistent. The
band gap energy as a function of Mg concentration is plotted in Figure 4.2. We
notice a deviation from a linear dependence on Mg concentration, Fitting the data
to a quadratic function we extrapolate a band gap for MgSe to be 4.54 eV. The
specific band gap of MgSe is not necessarily important, but what is relevant is the
position of the Mg,Cd ,_,Se conduction band relative to that of CdSe and ZnTe
as a function of Mg concentration. To do determine this we need a measure of the

valence band offset of Mg,Cd,_,Se.

4.2.2 Valence band offset measurement of MgSe/Cd 5¢Zn448e

The valence band offset (VBO), AE,, of MgSe/Cd5¢Zn.44Se was measured by
Wang et al. [10] using X-ray photoelectron spectroscopy (XPS). This heterojunc-
tion is used so as to avoid difficulties associated with lattice strain. Bulk MgSe
naturally occurs in the NaCl structure, but the zinc blende lattice constant has
been extrapolated by Okuyama et al. [4] to be about 5.89 A. The lattice matched
Cd5gZn.445e alloy composition was measured and calibrated using X-ray diffrac-
tion for lattice constant and XPS for alloy concentration. We used (100) GaSb
substrates with GaSb buffer layers to provide a high quality smooth growth sur-
face. A thick buffer layer of ZnTe was grown on the GaSb substrates since this
II-VI/III-V heteroepitaxial growth technology was well advanced. Next, thick re-
laxed layer of Cd.5gZn.44Se with a lattice constant of 5.93 A provide closely lattice

5.0

E sap(Mg,Cd,.,Se) = 1.735 + .956x + 1.85x° eV

O Data
— Fit

—-— Extrapolation

0.2

0.4

0.6

0.8

1.0

Mg Concentration

Figure 4.2: Estimated band gap of Mg,Cdj_,Se alloy as a function of Mg con-
centration. Data taken from low temperature PL spectra is used to approximate
the room temperature band gap. Experimental data is best fit by a quadratic
equation. Extrapolation of the fit estimates the band gap of zinc blende MgSe to
be 4.54 eV.

matched layers for the XPS experiment. The MgSe epilayers were grown up to a
maximum of about 150 A. During growth these layers exhibited a streaky (2 x
1) RHEED diffraction pattern indicating zinc blende epitaxial growth.

The value obtained for the VBO is [10]:

4.2.3 Band edge engineering

We can construct the band edge (BE) positions, valence band edge (VBE) and
conduction band edge (CBE), relative to the BE’s of CdSe and ZnTe for the
Mg,Cd;_,Se alloy as a function of Mg concentration. Using the above measured
value for the MgSe/Cd_5gZn.44Se VBO and CdSe/ZnTe VBO of 0.64 eV [11] we plot
the position of the valence band of Mg,Cdj_,,Se relative to the VBEs of CdSe and
ZnTe. Here we assume no band bowing and a linear dependence in the VBE going
from CdSe to MgSe. Adding to this the band gap information for Mg,Cd,_,Se, we
plot the CBE of Mg,Cd,_,Se relative to the CBE of ZnTe. From this plot we ex-
tract an approximation for the Mg,Cd,_,Se alloy which will have its CBE aligned
with the CBE of ZnTe. We estimate this is accomplished with a Mg concentration
of 69 %. Therefore, for the graded device we use this estimate for the initial Mg

concentration in the Mg,Cd,_,Se alloy junction.

4.3. Device structure

Light emitting devices based on the graded injector design have been grown on
both (100) GaSb substrates and (100) p-ZnTe substrates. Devices grown on GaSb
showed significantly better structural, electrical and optical properties than those
grown on the ZnTe, so we concentrated on those with GaSb substrates. The poor
performance of the ZnTe substrates appeared to be due to the presence of Te
precipates in the substrates as well as the formation of dislocations due to poor
surface quality. Log plots of the J-V characteristics of devices on ZnTe substrates
showed large leakage currents and reduced optical performance resulting from the
structural defects. Ideally, we would prefer ZnTe substrates rather than GaSb, but
lacking a suitable commercial source for ZnTe the best available choice was GaSb.

A typical device structure is grown on a GaSb substrate indium bonded for

thermal conduction to a molybdenum block. In the III-V MBE chamber a 3000 A

4.0 r T ev Tr ¥ T — rT bd A
yo
af
a7
3.0 F Wa 4
_~ Pat CB aire
> CBugcase co “
i?) -_—— =
S 5
Li
xe) 1.0 F
& VBo Te 1
faa]
0.0 Fe, VB yigcuse a
| ~~ 4
-1.0 mn a + i 2 i mn a | r—_
0.0 0.2 0.4 0.6 0.8 1.0

Mg Concentration

Figure 4.3: Band edge alignment of Mg,Cd,_,Se alloy as a function of Mg concen-
tration. Conduction (CBE) and valence band edge (VBE) positions are relative to
the VBE of CdSe. Using the band offsets of CdSe/ZnTe and MgSe/Cd_5¢Zn.44Se
and assuming a linear variation in the VBE of Mg,Cd,_,Se together with the
band gap dependence of Mg,Cd,_,Se we construct the band edge dependence.
With these assumptions we estimate a 69% concentration of Mg in Mg,Cd,_,Se

is needed to align the conduction bands of Mg,Cd,_,Se and ZnTe.

GaSb:Si buffer layer followed by a 20A AlSb cap is grown. The sample is trans-
ferred via UHV to the I-VI chamber, where growth of the device proceeds at a
growth temperature of 300 °C. The device consists of ~ 1 um of ZnTe:N doped
to roughly 1x10!8cm~% hole carrier concentration. Next the graded Mg,Cdi_,Se
layer with thickness varying from 300-600 A is grown . Mg,Cd;_,Se layers with
maximum concentrations from 60-70% Mg at the ZnTe interface were graded down
to about one order of magnitude less Mg ~ 6-7 %. The final n-junction layer con-

sists of ~ 300 A CdSe:Al doped to about 1x10'%cm7.

4.3.1 Device fabrication

Device fabrication relies on standard lithographic and wet etch techniques. After
pulling the samples from UHV, we heat the moly block in an inert atmosphere glove
box to remove the substrate’s indium bonding. We use a controlled environment
to avoid any potential oxidation. The diodes are then lithographically defined into
octagon mesas with areas of 1.8x107* cm?. Electrical isolation is formed with a
0.5% bromine:ethylene-glycol wet etch 4000 A into the p-ZnTe layer, well beyond
the p-n junction. Lateral ohmic contacts to the p-ZnTe and top contacts to the
CdSe:Al are made with evaporated Au/Ge and defined by standard photolitho-
graphic lift-off procedures. An ohmic back contact to the GaSb substrate can also
be used but there is a 0.82 eV valence band offset for GaSb/ZnTe which lead to
resistance across this interface. This effect is slightly mitigated due to the large
area of the interface and the high doping concentrations but it still makes the
back contact through the GaSb more resistive than the front lateral contacts. The
devices are then placed on a TO-5 header, wire bonded and available for electrical

and optical characterization.

4.4 Material characterization

4.41 TEM

Transmission electron microscopy (TEM) studies [12] of the junction region of the
devices revealed a number of interesting features. We discovered that the actual
thickness of the Mg,Cd,_,Se region in devices grown were significantly thicker
than expected. As previously described in chapter 3, we adjusted the growth of
the Mg,Cd,_,Se by calibrating the growth of CdSe to the thermal transient of the
Mg flux. The growth of the alloy appears to be significantly faster than anticipated
given the growth rate of CdSe and the additional Mg. We expected only 300 Afor
the graded region but were actually growing almost twice as thick. The presence
of Mg, with a sticking coefficient of 1, seems to increase the sticking coefficient of
Cd thereby increasing the overall growth rate.

We believe a thinner grading is essential but will be very sensitive to the growth
technique used. This is apparent in the TEM images of the Mg,Cd,_,Se/ZnTe in-
terface. This interface is characterized by the appearance of an interfacial reaction
and point defect formation leading to stacking faults in the {111] direction. We
estimate well over 1x10!" cm~? defect densities in the graded regions.

Due to the fixed constraints of this method of grading the Mg,Cd _,Se region,
we have been unsuccessful in fully compensating for this growth anomaly. Attemps
to shutter the Mg cell or slow down the overall growth rate helped to reduce the
graded region thickness but to the detriment of device performance compared to
the devices with thicker grading. The reason for this phenomena still needs to be
addressed.

We believe that by properly controlling the Mg source we can both decrease
the grading thickness and reduce defect formations. We have investigated the
use of two potential alternative source cells to better control the flux. One likely

candidate is a valved cell, where a mechanical valve can adjust the beam flux

precisely without changes in source temperature. The other is a water cooled
Knudsen cell, where the thermal transients are reduced with very efficient cooling.
Both cells would allow for extremely quick flux decreases necessary for a thinner
grading. Ideally though, the valved cell is prefered because of its ability to variably
control fluxes independent of thermal transients. A full range of valve positions
from open to completely closed can be reached in seconds. Therefore, a grading
with precise Mg concentration can be achieved in well under 300 A. Unfortunately,
this type of cell has not yet been used with Mg as the source material and issues

of reliability and materials compatibility have not yet been determined.

4.4.2 X-ray

Figure 4.4 shows the X-ray rocking curve data from the (004) reflection of a graded
device structure grown on a GaSb substrate. Included in the plot is a computer
simulated rocking curve generating using the PC-HRS software of Fewster [13].
The FWHM of the GaSb and ZnTe layers are 25 and 39 arc seconds respectively.
The ZnTe epilayer of this device is coherently strained to the GaSb substrate.
This is confirmed by analyzing the asymetric (115) diffraction peaks relative to
the GaSb peak. The CdSe layer has a broad FWHM and weak amplitude close to
the theoretically expected values given the finite thickness (~ 300 A) of the layer.
Most striking is the presence of the zinc blende diffraction peak of the CdSe. This
confirms the growth of cubic material through the graded Mg,Cd,_,Se region.
Surprisingly, it was noticed that the structural quality of the device epilayers,as
measured by the ZnTe FWHM, was not correlated with the electrical and/or opti-
cal performance of the LED. In fact, the best performing device with the highest
external quantum efficiency had a much worse ZnTe FWHM than the device in
Figure 4.4. These results imply that it is not the structural quality of the ZnTe
material that is presently limiting the LED performance. The luminescence quality

of all the epilayers (excluding the fully relaxed thick layers) appear to be equiv-

alent as measured by PL intensities. Therefore it appears that the devices are
limited by the injection efficiency rather than the recombination efficiency. The
injection efficiency is primarily dependent on the Mg,Cd,_,Se graded region and
Mg,Cd,_,Se/ZnTe interface. This issue of device efficiency will be discussed later

in the chapter.

Experimenta
seme Calculated

Intensity (a.u.)

30.0 30.1 EE 30.4 305 30.6 30.7 30.8
Omega (degrees)
Figure 4.4: The (004) reflection X-ray rocking curve of a graded device grown
on a GaSb substrate. Included is a calculated rocking curve for the same device
structure. The simulated curve is generated using the PC-HRS software from
Philips Research Laboratories. We notice the zinc blende (004) peak of the CdSe
layer indicating cubic growth through the grading. The slight shoulder on the high
angle side of the simulated curve of the CdSe peak originates from the Mg,Cd;_,Se

grading. Data was aquired using a four-crystal Ge-monochrometer diffractometer.

4.5 Device operation

4.5.1 Electrical performance

Figure 4.5 shows a comparison of the room temperature electrical characteristics
of the original and the present graded devices. The original device’s electrical
properties shortcomings have been eliminated by being able to grow doped layers
of p-ZnTe. The high series resistance of the early devices was due to the low
conductivity of the ZnTe substrates, and the high turn-on voltage was probably
due to the charging of traps at or near the substrate/buffer layer interface and the
thin undoped ZnTe epilayer all together resulting in a large voltage drop across
this region. The present devices’ current-voltage characteristics are very good.
These devices turn on more like an ordinary diode and operate at a voltage near
the band gap of ZnTe. There is almost no reverse bias leakage current and a sharp
forward bias turn-on. Most of the series resistance in the present device (25 12)
comes from current spreading through the thin ZnTe:N epilayer and not from the
contact or junction region. This has been substantiated by the linear dependence
of the resistance through the p-ZnTe epilayer as a function of contact to contact
seperation. This small series resistance can be virtually eliminated with the use of
high quality p-ZnTe substrates and/or thicker epilayers of p-ZnTe on GaSb.

To better understand the current injection mechanisms in the LED, we plot the
current density on a log scale as seen in Figure 4.6. Plotted in this manner we can
extract the ideality factor (n) for a diode. n is used and given by a modification to
the basic diode current-voltage equation to account for space-charge recombination
current. A simple empirical representation of the current-voltage relation is thus

given by the following formula:

I Diode ~ iV /nk? (4.2)

An ideality n=1 would be the case for a pure diffusion current, while an ideal-

200 ——— . . >
L A]
5 i
—~..| 49 cere riginal device 1
> 150 L Original de Ice | J
= ——. Present device ; |
2 ; ‘
a ! {
Q t
= i q
= t00b / 4
wo U
Cc 5 ‘
o /
Q t
~~ L /
c ?
® - /
5 L ! 4
© 50 i
Ul
5 a |
weer
0 ro Ll ee ee Be
0.0 2.0 4.0 6.0 8.0 10.0

Voltage

Figure 4.5: Room temperature electrical characteristics of the original and present
graded devices. The original device is a 100 um diameter mesa with back contacts
made to the ZnTe substrate. The present device grown on GaSb is an octagon

mesa with an area of 1.8x 1074 cm? with front lateral contacts made to the p-ZnTe

epilayer.

ity n=2 would be for pure recombination current based on the Shockley-Read-Hall
model of recombination through a trap [17]. When both currents are comparable,
the ideality will vary between 1 and 2. A more accurate current-voltage relation
would equate the total diode current to the sum of a diffusion current plus vari-
ous space-charge (non-radiative) recombination currents. This would account for
deviations from pure diffusion current, but would be an intractable J-V relation.
Nevertheless, the simple model using an ideality factor gives insight into the pro-

cesses associated with the diode current mechanisms.

Experimental device
----- Ideal device
10 Ff ——— Non-ideal device rs oor

fo}

Current Density (A/cm’)
3.

2.5

Figure 4.6: Log plots of current density vs. voltage for an experimental device, and
simulated “ideal” and “non-ideal” graded devices. The ideal device has a continu-
ous grading in the conduction band and negligible space-charge recombination at
the Mg,Cd,_,Se/ZnTe interface. By construction the ideality of this ideal device
is n=1, which indicates pure diffusion current. The non-ideal device has a 300 meV
discontinuity at the Mg,Cd,_,Se/ZnTe interface and a significant density of inter-
face traps. The minimum ideality of the experimental device is n=1.8 indicating

a combination of diffusion and space-charge recombination currents.

The minimum ideality for the experimental data is n=1.8. This indicates a
combination of diffusion and interfacial-recombination currents. With higher in-
jection currents, the series resistance distorts the ideality factor making it appear
larger. Plotted along with the experimental data are simulated results using the
model of Wang et al. [18] based on the drift-diffusion transport equations. The
computer model used to generate the J-V behavior of the device can be used to
intuitively explain the non-ideal behavior of the experimental device.

The “ideal” case corresponds to a graded device with a continuous conduc-
tion band and a negligible number of non-radiative recombination centers at the
Mg,Cd,_,Se/ZnTe interface. The effective surface recombination velocity at this
interface was assumed to be 200 cm/sec. Not surprisingly, the ideality for this case
is n=1. We can modify the qualitative and quantitative behavior of the J-V by
adjusting the interface trap concentration and the grading discontinuity. As the
density of traps at the Mg,Cd,_,Se/ZnTe interface is increased the non-radiative
space-charge current will increase. A discontinous grading in the conduction band
resulting in an electron barrier at the Mg,Cd,_,Se/ZnTe interface leads to an
accumulation of electrons at this interface and subsequently enhances any non- ra-
diative recombination processes. Increased space-charge current will both increase
the ideality and shift the log J-V curve upwards towards the experimental curve.

As an example, we plot in Figure 4.6 the simulation result for a “non-ideal”
device. This device had a discontinuous grading, with a 300 meV discontinuity
at the Mg,Cd,_,Se/ZnTe interface, and a high density of interface traps with an
effective surface recombination velocity of 1x10° cm/sec. The actual value of the
discontinuity isn’t the important parameter in this calculation. Physically what
matters is the presence of non-radiative channels at the interface, and an overlap
of carrier accumulation at this interface. We should not take the parameters as
physically meaningful but as illustrative of the trend and behavior of the interface
properties. From this behavior we conclude the interfacial non-radiative current

is non-negligible in our present graded devices. Furthermore, it is most likely the

10 il | v we LJ ill Bll 7 Tr qT Ly
10° r ; ona
10 B ottee HP GaP:N LED A

— CIT Graded LED

“e
5 10°
210° F
Q 10° F
= 10° -
am ]
6)
10° r
10°

20 18 1.0 05 09
Voltage

Figure 4.7: Log plot of graded injector LED compared to Hewlett Packard commer-

cial GaP:N green LED. Both devices exhibit small reverse bias saturation currents

and obvious indications of the onset of forward bias current injection. The GaP:N

LED begins to emit light at 1.75 Volts forward bias and the Caltech graded LED

begin at 1.85 Volts.

grading in the device is not optimized and there exists an electron barrier at the
interface. In spite of the interfacial problems the device operates as a normal LED
as will be discussed later.

In Figure 4.7 a log plot of the room temperature J-V characteristics of a graded
device is compared with a commercial GaP:N green LED. Both devices show simi-
lar J-V behavior, with the graded device having slightly more space-charge recom-
bination current and larger series resistance. Light emission turn-on, as defined

by when you can begin to see light with the eye, is coincidentally very similar in

the two devices as well. The commercial device begins to turn-on at about 1.75
Volts, and the graded device begins to turn-on at about 1.85 Volts. Light emission
can begin at voltages below the band gap due to the minority carrier injection
process [19]. Although the graded junction LED has a drastically different device
design and operating concept compared to the standard commercial LEDs, we see

remarkably similar operating characteristics.

4.5.2 Optical performance
Electroluminescence

Figure 4.8 shows a characteristic electroluminescence (EL) spectra from a graded
device operating under a 2.8 Volts DC bias at room temperature. The EL spec-
tra clearly demonstrates the electron injection into the p-ZnTe. The emission is
dominated by band to band recombination in the ZnTe at 2.25 eV, with no deep
emission present. The absence of any luminescence from CdSe (Eyap = 1.74 eV) in-
dicates that holes are blocked by the valence band offset barrier. The lower energy
emission centered at 1.96 eV is currently not conclusively known. This red emis-
sion peak is present in all graded devices fabricated to date. The relative intensity
of this peak compared to the band edge emission varies from sample to sample and
appears to be independent of substrate (ZnTe or GaSb) and growth conditions.
This 1.96 eV red peak is not present in the photoluminescence of ZnTe:N epilayers.
Although, it is observed in the photoluminescence of graded device structures as
well as epilayers of CdSe:Al on ZnTe:N (the same structure of a graded device
without the Mg,Cd,_,Se layer). In the case of the CdSe:Al/ZnTe:N structure the
luminescence of the red peak is weaker than for the comparable graded device
structure. Therefore, we believe the red luminescence to be related to the Se/Te
interface, and the magnitude of its effect is increased with the presence of Mg. The

exact nature of this phenomena is still under investigation.

Intensity (a.u.)

1.8 2.0

Energy (eV)

Figure 4.8: Room temperature electroluminescence spectra of a graded injector
device. Device is a hexagon mesa with an area of 1.8x10~* cm? operating under
a 2.8 Volt DC bias with 30mA current. Emission is dominated by band to band
recombination of ZnTe. No luminescence is observed from CdSe or from deep
emission. The low energy emission at 1.96 eV is believed to be from the Se/Te

interface.

Quantum efficiency

The external quantum efficiency 7..;, defined as the ratio of the number of photons
emitted from a device to the number of charge carriers passing through the diode,
was measured using an “Integrating Sphere” [20]. A Si photodetector is fit to the
sphere to detect the radiant flux. Geometric and reflectivity effects of the sphere
were calibrated using commercial LEDs with known quantum efficiencies. The best
measured external quantum efficiency of the graded injector LED grown on GaSb
substrates was 0.007%. This is close to an order of magnitude greater than the
best of the early devices grown on ZnTe substrates. In spite of the improvement in
performance, the present graded devices in there current state are not competitive
with commercial green LEDs, although we have made little effort in optimizing
light extraction which could drastically improve overall 7.2;. We have focused
instead on developing the technology to clearly demonstrate graded injection in an
LED.

To examine the reasons for the devices’ lower efficiencies and offer suggestions
for improvements we break the external quantum efficiency into three indepen-
dent terms, each which comprises a separate physical component for the overall

efficiency. These elements of the external quantum efficiency et are,

Neat = Th’ Vr * Ne (4.3)

where ni is the injection efficiency, 7, is the recombination efficiency, and 7, is the
light extraction efficiency.

Since the luminescence is generated in the p-region, the injection efficiency, 7;,
is the ratio of the electron injection current to the total current. The total current
is composed of the electron injection, hole injection and space-charge recombina-
tion. We saw previously there is negligible hole current, but significant interfacial
recombination current. To improve the injection efficiency we must reduce the
space-charge recombination current. To do this will require better control and

understanding of the Mg,Cd,_,Se/ZnTe interface and graded Mg,Cd;_,Se region.

The recombination efficiency, 7,, is the ratio of the radiative recombination
to the total recombination in the p-region of the device. Here there are a few
factors which limit the efficiency. The thickness and quality of the ZnTe active
region will strongly effect the recombination efficiency. In our device structure, if
the electron diffusion length is comparable to the thickness of the ZnTe ( ~ lum)
then 7, would be reduced by the carriers recombining in the GaSb. Mathine et
al. [21] found for thick (up to 24m, which is far greater than the absorption length
of the laser excitation, 4400 A) ZnTe epilayers grown on GaSb buffer layers an
appreciable PL recombination from the underlying GaSb, indicating the diffusion
of carriers through the ZnTe. Attempts at growing thicker ZnTe epilayers are
limited by the critical thickness of these epilayers grown on GaSb. On the other
hand the luminescent quality of the ZnTe material is very good, as seen in the
room temperature PL and the absence of deep emission. One way to mitigate the
thickness limitations and improve the recombination efficiency is to incorporate
quantum wells and quantum well structures to capture carriers and provide higher
radiative recombination efficiencies. We estimate that quantum well structures
could improve radiative efficiencies by at least one order of magnitude. Epilayers
of ZnTe with Cd,Zn,_,Te quantum wells have been seen to improve the integrated
PL by over an order of magnitude as compared to ZnTe epilayers with no quantum
wells. Working graded injectors with quantum wells have not to date been grown
but are currently being investigated.

The extraction efficiency, 7,, is the ratio of the number of photons which escape
the LED to the total number generated in the active region. This factor is prici-
pally governed by the device design and fabrication, i.e. the presence of absorbing
materials, the type and position of contacts, the fabrication of cubes rather than
mesas, and encapsulation techniques. The extraction of light is fundamentally lim-
ited by the critical angle for total internal reflection (Snell’s Law), and absorbtion.
The extraction of light generated by an LED has in fact been a major focus of com-

mercial interest [22, 23, 24]. By using thick transparent conducting window layers

to allow for current spreading and moving the junction away from the top contact,
Huang et al. [22] increased external quantum efficiencies by a factor of twofold.
Kish et al. [23] used a wafer-bonding technique to replace an absorbing substrate
with a transparent substrate to increase the external efficiency by a factor of 2
as well. Encapsulation and packaging are standardly used in commercial LEDs
to increase extraction by up to a factor of 8 [24]. Since we are using absorbing
GaSb substrates, all the light emitted down into the device is lost. Furthermore,
since the p-n junction is just below the top contact (< 1000 A) there is little cur-
rent spreading and all the light generation is underneath the contact. The top
contact, Au/Ge, is not transparent and prevents any extraction through the top
of the device. In our devices the observed light escapes only from the perimeter
of the contact and edges of the mesa. As a result the extraction efficiency is ex-
tremely poor. We have made no serious effort to improve the extraction efficiency
although Wang [25] has suggested some possible top contacting schemes to im-
prove the light extraction. In addition if ZnTe substrates were available we would
then have transparent substrates. Bulk growth of ZnTe substrates has not been
very advanced. One way to overcome this problem, would be to use thick layers (>
100 um) of ZnTe grown by MOCVD and later etch off the NI-V substrate. These
thick layers would then serve as a pseudo-substrate for MBE.

All together, it comes as no surprise that our measured external quantum ef-
ficiencies are low compared to commercial green LEDs. However, there are clear
approaches available to increase each of 7;, 7, and n,. By addressing the improve-
ments in just 7, and n, only we expect the graded injector to be competitive with

commercial green LEDs.

Device lifetime

Early graded injector devices grown on ZnTe exhibited evidence of potential long

lifetimes. These devices operating under high injection of 100 Amps/cm” degraded

by 50% after ~1500 hours of continuous operation. The recent devices, grown on
GaSb substrates however have much poorer operating lifetimes. Under current
densities of 52 Amps/cm? these devices degrade, as measured by the light emission,
by 50% after only ~10 hours. Figure 4.9 illustrates the lifetime characteristics for

a typical graded device grown on a GaSb substrate.

100.0 a — 2.75
OVoltage {2.70
53 10040 oO Current
& w
S }O a
2p {265 $
QD th | o
E D, o .—CV*oltage a
= Q) _—_— J @
=> 10} Oo g
—_— o 4 2.60
Current Bo 3 fa
re)

0.1 : : 7" . : 2.55
0.0 1000.0 2000.0 3000.0 4000.0 5000.0 6000.0
Time (min)
Figure 4.9: Device lifetime measurement of a graded injector LED. Device is oper-
ated under constant bias current of 52 Amps/cm?. Light emission is measured by
the photodiode current of a silicon photodetector in an integrating sphere. Bias

voltage is the voltage measured across the LED.

The reason for the poor device lifetime is presently unknown. The possibilities
being investigated include defect formation in the ZnTe epilayer due to nitrogen
doping, or from the ZnTe growth on GaSb, either of which could degrade re-

combination efficiency. Another possibility is that defect formation at the graded

junction Se/Te interface leads to greater interfacial recombination and therefore
reduced injection efficiency. This last reason may explain the difference in lifetimes
between the early and present devices due to a difference in device operation. The
earlier devices operated under high bias and may have been a high field effect de-
vice, while the present devices clearly achieve minority carrier injection as a normal
p-n diode. The present devices’ current voltage characteristics degrade consider-
ably along with the optical output, which wasn’t the case for the early devices.
This is observed in the reduced bias voltage with time under constant bias current
observed in Figure 4.9.

The present device’s J-V becomes much leakier (i.e. greater bias current for a
given bias voltage, or lower bias voltage for a given bias current) indicating greater
interface space-charge recombination. Figure 4.10 plots the relation between the
light emission current and bias voltage change operating under a constant bias
current. This relation exhibits a thermionic behavior between the light emission
and bias voltage. This suggests a dependence for the light emission (lifetime) on the
injection current efficiency rather than the recombination efficiency. Therefore the
lifetime degradation may be due to reduced injection efficiency rather than reduced
recombination efficiency. This may result from defect formation at the graded
junction Se/Te interface enhanced by the extra strain from the thick grading. The
early device’s lifetime may have been a measure of the active region’s recombination
rather than of the injection process. Therefore improvements in reducing the
interfacial defects may greatly improve the device lifetime characteristics.

The precise nature of the device lifetime is still under investigation. This may
turn out to be a fundamental problem associated with II-VIs. In general, the II-
VI semiconductors are soft materials and may not have the robustness needed for
long lifetime LEDs and laser diodes. This could prove to be an inherent physical
attribute which will frustrate device development. However, strong evidence points
to the lifetime problem being related to the poor Se/Te interface. This should be

a tractable problem with improvements in growth technology.

10 5

Light emission I,, (a-u.)

Data Fit

lop = 1, exp(AV

I, = 10.8
V,= 0.040

N,)

Bias’

0 1
-0.2

-0.1

Bias Voltage Difference (AV,,,. Volts

0.0

Figure 4.10: Light emission, as measured by the photodiode current of a sili-

con photodetector in an integrating sphere, vs. the change in bias voltage for a

graded device operating under constant current. Data is taken from that of Fig-

ure 4.9. The straight line fit exhibits a thermionic related process to the lifetime

degradation. This suggests a reduced injection efficienty rather than a decreased

recombination efficiency in the temporal behavior of the device operation.

4.6 Conclusion and areas for future develop-
ment

We have clearly demonstrated operation of the graded junction electron injector
LED based on the n-CdSe/Mg,Cd_,Se/p-ZnTe structure. Success with the p-
type doping of MBE ZnTe and the use of GaSb substrates to grow high quality
ZnTe has lead to major improvements in device performance. Normal p-n diode-
like LEDs with low turn-on voltages and efficient election injection into the wider
band gap material have been demonstrated. Device simulations and lifetime studies
suggest the graded junction Se/Te interface may play a role in the reduced and
degrading injection efficiency.

A number of areas are open for additional research. Further development of
the growth technology necessary to optimize the device structure should lead to
increasing performance gains. Improved control of the Mg flux, via a valved Mg
cell, will allow for precise thickness control of the graded Mg,Cd,_,Se junction and
better Se/Te interface control. The incorporation of quantum structures will lead
to improved recombination efficiencies. The development of substrate technology
for homoepitaxial growth will be critical to improving light extraction character-
istics. In addition, more sophisticated processing technology will enable greater
extraction efficiencies. Most importantly, the nature of the device lifetime and

failure modes should be studied.

Bibliography

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

M.C. Phillips, sl Graded Injection: A New Approach to Wide-bandgap Light
Emitters, (Ph.D. thesis dissertation, CalTech, 1993).

K. Itoh, J. Phys. Soc. Japan 22, 1119 (1967).

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

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

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

W. Strehlow and E. Cook, J. Phys. C 2, 163 (1973).
J.O. McCaldin, T.C. McGill and C.A. Mead, Phys. Rev. Lett. 36, 56 (1976).
S. Wei and A. Zunger, Phys. Rev. Lett. 59, 144 (1987).

M.W. Wang, J.F. Swenberg, M.C. Phillips, E.T. Yu, J.O. McCaldin, R.W.
Grant and T.C. McGill, Appl. Phys. Lett. 64, 3455 (1994).

88
{11] E. Yu, M. Phillips, J. McCaldin and T. McGill, J. Vac. Sci. Technol. B 9,
2233, (1991).

(12] TEM studies were performed by Carol Garland, Department of Material Sci-

ence, California Institute of Technology.

[13] P.F. Fewster, PC-HRS Version 1.0 Software, Philips Research Laboratories,
Redhill, UK (Copyright 1990). The model is based on the Tagaki-Taupin [14,
15] formula adapted by P.F. Fewster et al. [16]

[14] S. Tagaki, Acta. Cryst. 15, 1311 (1962).

[15] P. Taupin, Bull. Soc. Fran. Miner. Cryst. 87, 469 (1964).
[16] P.F. Fewster, C.J. Curling, J. Appl. Phys. 62, 4154, (1987).
[17] W. Shockley and W.T. Read, Phys. Rev. 87, 835 (1952).

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

[19] For a discussion of minority carrier injection in p-n junction diodes see Chap-
ter 8 of S. Wang, Fundamentals of Semiconductor Theory and Device Physics
(Prentice Hall, New Jersey, 1989).

[20] Integrating sphere model IS-040-SF from Lasphere, Inc. P.O. Box 70 Shaker
Street North Sutton, NH 03260.

(21] D.L. Mathine, $.M. Durbin, R.L. Gunshor, M. Kobayashi, D.R. Menke, Z.Pei,
J.Gonsalves, N. Otsuka, Q.Fu, M. Hagerott and A.V. Nurmikko, Appl. Phys.
Lett. 55, 268, (1989).

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

89
(23|] F.A. Kish, F.M. Steranka, D.C. Defevere, D.A. Vanderwater, K.G. Park, C.P.
Kuo, M.G. Craford and V.M. Robbins, Appl. Phys. Lett. 64, 2839 (1994).

[24] Stan Gage, Optoelectronics/Fiber-optics Application Manual, Hewlett-
Packard Company Optoelectronics Division Applications Engineering Staff,

(McGraw-Hill, New York, 1981).

[25] M.W. Wang, Chapter 4, Ph.D. thesis dissertation (1994).