Infrared Optical Studies of HgTe-CdTe Su

Infrared Optical Studies of HgTe-CdTe Superlattices and GaAs - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Infrared Optical Studies of HgTe-CdTe Superlattices and GaAs
Citation
Hetzler, Steven Robert
(1986)
Infrared Optical Studies of HgTe-CdTe Superlattices and GaAs.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/5ptm-z449.
Abstract
This thesis presents two different studies of the infrared optical properties of two different semiconductors. Chapter 2 describes the results of the first infrared photoluminescence (IRPL) measurements of a HgTe-CdTe superlattice. IRPL spectra of two different HgTe-CdTe superlattices from two different sources were measured from 100 to 270K. Sample 1 was grown on (111) Cd
0.96
Zn
0.04
Te, and was grown to have 250 repeats of 38 - 40 Å of HgTe followed by 18 - 20 Å of CdTe. Sample 2 was grown on (111) CdTe and was grown to have 75 repeats of 50 Å of HgTe followed by 50 Å of CdTe. Sample 1 exhibited a single asymmetrical luminescence line at all temperatures studied. (Low signal-to-noise ratio from sample 2 prevented detailed analysis of the lineshape.) The luminescence from both samples occured at significantly lower energies than that from Hg
1-x
Cd
Te alloys with the same Cd concentrations as the average Cd concentrations of the superlattices. At 240 K, the luminescence peak from sample 1 was near 148 meV, with a full width at half-maximum intensity of 42 meV, while the peak from sample 2 was near 242 meV, with a full width at half-maximum intensity of 69 meV. Analysis of the luminescence lineshape from sample 1 showed it to be consistent with wave-vector conserving band-to-band recombination. In this case, the band-gap energy of the superlattice would be near the low energy threshold of the luminescence peak. This study therefore represents the first direct determination of the band gap of an HgTe-CdTe superlattice. A comparison of the lineshapes from both samples with those measured in GaAs-Ga
1-x
Al
As super-lattices showed evidence for fluctuations in the layer thicknesses of both the HgTe-CdTe superlattice samples. A comparison was made between the data and a simple theory of the band gaps of HgTe-CdTe superlattices. The theory was shown to be consistent with the experiments, if there were small errors in the measurements of the superlattice layer thicknesses of each sample. The differences in the luminescence properties of the two samples show that it is possible to tailor the band gaps of HgTe-CdTe superlattices.
Chapter 3 describes the first observation of
-like excited states of a double acceptor in a semiconductor. Two experiments were performed to study the
-like excited states of the 78-meV acceptor in GaAs. The techniques used, selective excitation scattering (SEL) and electronic Raman scattering (ERS), are both sensitive to the detection of
-like excited states of single acceptors in semiconductors. Measurements on two different liquid encapsulated Czochralski GaAs samples showed two s-like excited state transitions of equal magnitude, separated by 4.0 meV. Only one
-like transition is expected in the energy range measured for a single acceptor. A simple effective mass-like model of a double acceptor was developed to account for the two
-like excited states. This model predicted a splitting of the 1
excited state of a double acceptor to be 2.6 meV, in good agreement with the observed value of 4.0meV. This proved that the 78-meV acceptor in GaAs is due to the first ionization of a double acceptor, the first such identification to be made based on the
-like excited state spectrum. It is therefore possible to identify the valency of an acceptor in a semiconductor by measuring the
-like excited state spectrum.
Appendix A describes a novel technique for performing infrared photoluminescence measurements using a Fourier transform infrared spectrophotometer. This technique was developed to perform the experiments described in Chapter 2.
Item Type:
Thesis (Dissertation (Ph.D.))
Subject Keywords:
Applied Physics
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:
McGill, Thomas C. (chair)
Goddard, William A., III
McCaldin, James Oeland
Johnson, William Lewis
Nicolet, Marc-Aurele
Defense Date:
28 October 1985
Funders:
Funding Agency
Grant Number
Caltech
UNSPECIFIED
IBM
UNSPECIFIED
Army Research Office (ARO)
UNSPECIFIED
Office of Naval Research (ONR)
UNSPECIFIED
Record Number:
CaltechETD:etd-03192008-084757
Persistent URL:
DOI:
10.7907/5ptm-z449
Related URLs:
URL
URL Type
Description
DOI
Article adapted for Chapter 2.
DOI
Article adapted for Chapter 3.
DOI
Article adapted for Chapter 3.
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
1011
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
21 Mar 2008
Last Modified:
24 May 2024 18:51
Thesis Files
Preview
PDF (Hetzler_sr_1986.pdf)
- Final Version
See Usage Policy.
4MB
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
INFRARED OPTICAL STUDIES

OF
HgTe-CdTe SUPERLATTICES AND GaAs

Thesis by
Steven Robert Hetzler

In Partial Fulfillment of the Requirements
for the Degree of

Doctor of Philosophy

California Institute of Technology

Pasadena, California

1986
(Submitted October 28, 1985)

To Roselynn

ii
Acknowledgments

I would like to thank Dr. T. C. McGill for his support and assitance during my
years of graduate study at Caltech. It has been both a pleasure and a privilege
to work with him.

I am indebted to Dr. A. T. Hunter of the Hughes Research Laboratories for
his continuing help on my various projects, and for setting a high standard of
experimental excellence. I have also profited from my many interactions with
Dr. D. L. Smith of the Los Alamos National Laboratories.

I owe a special debt of gratitude to O. J. Marsh, J. P. Baukus, R. Baron
and M. H. Young of the Semiconductor Crystal Physics Group at the Hughes
Research Laboratories for making their FTIR available for the experiment in
Chapter 2 and for providing the GaAs samples used for Chapter 3. I would
especially like to thank A. T. Hunter and J. P. Baukus for their ingenuity in
helping devise the FTIR experiment described in Appendix A. I would also like
to thank Dr. J. P. Faurie of the University of Illinois and Dr. P. P. Chow of
Honeywell for providing the HgTe-CdTe superlattice samples for the study in
Chapter 2.

I have also profited from many discussions with Drs. C. Mailhoit, R. M.
Feenstra, A. Zur and R. T. Collins, as well as with R. J. Hauenstein, G. Y. Wu,
T. E. Schlesinger, M. B. Johnson and R. H. Miles. I would also like to express my
thanks to Vere Snell, for her excellent secretarial work, and her always cheerful
disposition.

I would like to thank the following for financial support: the California In-
stitute of Technology, International Business Machines corporation, the Army
Research Office and the Office of Naval Research.

Above all, I am grateful to my wife, Roselynn Huong, for her patience, sup-

port and encouragement during my years of graduate study.

Abstract

This thesis presents two different studies of the infrared optical properties of
two different semiconductors. Chapter 2 describes the results of the first infrared
photoluminescence (IRPL) measurements of a HgTe-CdTe superlattice. IRPL
spectra of two different HgTe-CdTe superlattices from two different sources were
measured from 100 to 270K. Sample 1 was grown on (111) CdoeZno.osTe, and
was grown to have 250 repeats of 38 — 40A of HgTe followed by 18 — 20A of
CdTe. Sample 2 was grown on (111) CdTe and was grown to have 75 repeats of
50A of HgTe followed by 50A of CdTe. Sample 1 exhibited a single asymmetrical
luminescence line at all temperatures studied. (Low signal-to-noise ratio from
sample 2 prevented detailed analysis of the lineshape.) The luminescence from
both samples occured at significantly lower energies than that from Hg;_,Cd,Te
alloys with the same Cd concentrations as the average Cd concentrations of the
superlattices. At 240K, the luminescence peak from sample 1 was near 148 meV,
with a full width at half-maximum intensity of 42 meV, while the peak from sam-
ple 2 was near 242 meV, with a full width at half-maximum intensity of 69meV.
Analysis of the luminescence lineshape from sample 1 showed it to be consis-
tent with wave-vector conserving band-to-band recombination. In this case, the
band-gap energy of the superlattice would be near the low energy threshold of
the luminescence peak. This study therefore represents the first direct deter-
mination of the band gap of an HgTe-CdTe superlattice. A comparison of the
lineshapes from both samples with those measured in GaAs-Ga,_,Al,As super-
lattices showed evidence for fluctuations in the layer thicknesses of both the
HgTe-CdTe superlattice samples. A comparison was made between the data and
a simple theory of the band gaps of HgTe-CdTe superlattices. The theory was
shown to be consistent with the experiments, if there were small errors in the

measurements of the superlattice layer thicknesses of each sample. The differ-

iv
ences in the luminescence properties of the two samples show that it is possible
to tailor the band gaps of HgTe-CdTe superlattices.

Chapter 3 describes the first observation of s-like excited states of a double
acceptor in a semiconductor. Two experiments were performed to study the
s-like excited states of the 78-meV acceptor in GaAs. The techniques used,
selective excitation scattering (SEL) and electronic Raman scattering (ERS),
are both sensitive to the detection of s-like excited states of single acceptors in
semiconductors. Measurements on two different liquid encapsulated Czochralski
GaAs samples showed two s-like excited state transitions of equal magnitude,
separated by 4.0meV. Only one s-like transition is expected in the energy range
measured for a single acceptor. A simple effective mass-like model of a double
acceptor was developed to account for the two s-like excited states. This model
predicted a splitting of the 1s12s1 excited state of a double acceptor to be 2.6meV,
in good agreement with the observed value of 4.0meV. This proved that the 78-
meV acceptor in GaAs is due to the first ionization of a double acceptor, the first
such identification to be made based on the s-like excited state spectrum. It is
therefore possible to identify the valency of an acceptor in a semiconductor by
measuring the s-like excited state spectrum.

Appendix A describes a novel technique for performing infrared photolumi-
nescence measurements using a Fourier transform infrared spectrophotometer.
This technique was developed to perform the experiments described in Chap-

ter 2.

Parts of this thesis have been or will be published under the following titles:

Chapter 2:
Infrared Photoluminescence Spectra from HgTe-CdTe Superlat-
tices,
S. R. Hetzler, J. P. Baukus, A. T. Hunter, J. P. Faurie, P. P. Chow and
T. C. McGill, Applied Physics Letters 47, 260 (1985).

Infrared Photoluminescence Measurments of a HgTe-CdTe Su-
perlattice,

S. R. Hetzler, T. C. McGill, J. P. Baukus, A. T. Hunter and J. P. Faurie,
presented at the 1985 Electronic Materials Conference, Boulder, June 19—
21, 1985.

Infrared Photoluminescence Spectra from HgTe-CdTe Superlat-
tices,

S. R. Hetzler, J. P. Baukus, A. T. Hunter, J. P. Faurie, P. P. Chow and
T. C. McGill, presented at the Yamada (2nd International) Conference on

Modulated Semiconductor Structures, Kyoto, Japan, September, 1985.

Experimental and Theoretical Comparison of Optical Properties
of HgTe-CdTe Superlattices,

J. P. Baukus, A. T. Hunter, C. Jones, G. Y. Wu, S. R. Hetzler, T. C.
McGill and J. P. Faurie, to be presented at the 1985 U.S. Workshop on
the Physics and Chemistry of Mercury Cadmium Telluride, San Diego,
October 8-10, 1985, and to be published in the Journal of Vacuum Science

and Technology, July/August, 1986.

Chapter 3:
S-like Excited States of the 78-meV Acceptor in GaAs,
S. R. Hetzler, T. C. Mcgill and A. T. Hunter, Proceedings of the 17th
International Conference on the Physics of Semiconductors, San Francisco,

August 6-10, 1984.

Selective Excitation Luminescence and Electronic Raman Scat-
tering Study of the 78-meV Acceptor in GaAs,

S. R. Hetzler, T. C. McGill and A. T. Hunter, Applied Physics Letters
44, 793 (1984).

Selective Excitation Luminescence and Electronic Raman Scat-
tering Study of the 78-meV Acceptor in GaAs,

S. R. Hetzler, T. C. McGill and A. T. Hunter, Bulletin of the American
Physical Society 29, 290 (1984).

vii

Contents
Acknowledgments ii
Abstract lii
List of Publications Vv
1 Introduction 1
1.1 Background ........... 0... eee eee ee te ee ee 1
1.2 Superlattices 2.2... ee 3
1.2.1 The Kronig-Penney model.................. 6
1.2.2 Optical studies of superlattices ............022- 9
1.3 Impurities and defects ......... 2.0.2.0. e ee eee 9
1.3.1 Effective mass theory .............-0028005 9
1.4 Outline ofthesis ... 2.2... 0.2.0... eee ee ee eee 11
1.4.1 IRPL from HgTe-CdTe superlattices ............ 11
1.4.2 The 78-meV acceptor in GaAs .............26. 13
1.4.3 Photoluminescence using an FTIR ............. 14
2 IRPL from HgTe-CdTe Superlattices 16
2.1 Introduction... . 2... 2... ee ee ee ee es 16
2.1.1 Background... 1... .... eee ee 16

2.1.2 Results of thisstudy .............0.0000084 18

2.2
2.3

2.4

2.5

2.6

The
3.1

3.2

Sample descriptions ........ 2... ee eee ee ee ee 19
Description of experiment ........ 0.0. ee ee eee ee ees 20
2.3.1 General... 2... ee ees 20
2.3.2. Dispersive spectrometer experiment ............ 20
2.3.3 Limitations of the dispersive approach ........... 26
2.3.4 FTIR experiment............. 0.00 eee 28
2.3.5 Modulated reflectivity .............20 020004 33
Results. 2... ee 37
2.4.1 Hg;.,Cd,Te alloy IRPL ................... 37
2.4.2 HgTe-CdTe superlattice sample 1IRPL .......... 37
2.4.3 HgTe-CdTe superlattice sample2IRPL .......... 39
2.4.4 HgTe-CdTe sample 1 temperature dependence ...... 42
2.4.5 HgTe-CdTe sample 2 temperature dependence ...... 42
2.4.6 Signalorigins ........ 0... eee ee ee ee es 45
2.4.7 Substrate luminescence ............-02-02-0-- 45
Interpretation... 1... ee 48
2.5.1 Luminescence processes ......... 2.000 ee eee 48
2.5.2 Comparison with theory ...............200. 59
Conclusions .. 1... ee 67
78-meV acceptor in GaAs 75
Introduction. . 2... 2... ee ls 75
3.11 Background... ..........0 2.2.02. eee ene 75
3.1.2 Results of this work .............-..0.004. 76
Experimental techniques... . 1... ee eee ee eee 77
3.2.1 Electronic Raman scattering ..............0.0. 78
3.2.2 Selective excitation luminescence .............. 79

3.2.3 Applications 2... .. 0. ee ee ee 81

3.2.4 Experimental setup. ...........0.2. 000502 83

3.3 Sample descriptions .............-2. 0000+ +e eee 85
3.4 Results... 2... . ee 87
3.4.1 Photoluminescence...........-.----+-+2+--5 87
3.4.2 SELand ERS..................--.20--2, 90

3.5 Interpretation... . 2... . 0... ee ee ee ee 95
3.5.1 Double acceptor effective mass theory ........... 95
3.5.2 Single acceptor model .......-.........0+00- 98
3.5.3 Comparison of theories... 2... 1... ee et ee 99

3.6 Conclusion .. 1... 2... . ee ee en 99
A Photoluminescence using an FTIR 106
A.1 Introduction. . 2... 2... ee 106
A.2 FTIR principles... 2... ee ee 107
A.3 Photoluminescence measurements with an FTIR ......... 109
A.3.1 Double-modulation technique ................ 110
A.3.2 Actual experimental setup. .........0-2.00000. 113

A.4 Possible pitfalls... 2... ee ee ee ee es 115

List of Figures

1.1
1.2

1.3

2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14

2.15

Superlattice structure .. 2... ee ee ee es 4
Superlattice types 2... ee 5
Periodic square-well potential of a superlattice .......... 7
Dispersive spectrometer experimental setup ............ 23
Dispersive HgCdTe photoluminescence spectrum ......... 27
FTIR experimental setup ........0 202.00 eee eee 30
Theoretical output low-pass filter response curve ......... 32
IRPL from a Hgo,71Cdo.29Te alloy sample. ............. 34
Modulated reflectivity experiment ...........2.020084 35
IRPL spectra from a Hgo.713Cdog9Te alloy ............. 38
HgTe-CdTe superlattice sample 1 IRPL .............. 40
HgTe-CdTe superlattice sample2 IRPL .............. Al
HgTe-CdTe sample 1 temperature dependence. .......... 43
HgTe-CdTe sample 2 temperature dependence. .......... 44
Luminescence from the superlattice substrates. .......... 47
Band-to-band radiative transition. ..........0..0000.% 50
Lineshape fits to HgTe-CdTesamplel ............... 53

Comparison of the HgTe-CdTe superlattice sample 1 data with

2.16 Comparison of the HgTe-CdTe superlattice sample 2 data with

3.1 Selective excitation luminescence process. .............
3.2 Selective excitation luminescence experiment ...........
3.3 78-meV acceptor photoluminescence .............00-
3.4 Selective excitation luminescence spectra of the 78-meV acceptor
3.5 Electronic Raman scattering spectra of the 78-meV acceptor

3.6 Electronic Raman scattering spectra of 2samples.........

3.7 Double acceptor energy diagram ...........0+420005

A.1 Schematic diagram ofan FTIR ................002.
A.2 Schematic diagram of the FTIR setup ...............

A.3 Comparison of FTIR measurement techniques...........

80
84
88
91
92
93
97

xil

List of Tables

1.1

2.1
2.2
2.3
2.4

3.1

Binding energies of some shallow acceptors in GaAs ....... 11
Properties of selected infrared detectors ............0.. 21
Properties of selected infrared optical materials .......... 22
Available gratings for spectrometer ............0000. 24
HgTe-CdTe Superlattice Results ..........02.00000. 64

Observed line positions and interpretations ............ 100

Chapter 1

Introduction

1.1 Background

This thesis deals with infrared optical studies of semiconductors. The interest
in optical studies of semiconductors stems from the desire to understand both
the optical and electrical behavior of such materials. Semiconductors are useful
for fabricating both electrical and optical devices. An understanding of the prop-
erties of such materials, and how to control them, is necessary for the successful
design of useful devices. One of the studies in this thesis involves determining
the basic intrinsic properties of a new material, the HgTe-CdTe superlattice,
which, it is hoped, will be useful for fabricating infrared detectors. (The intrinsic
properties are defined here to be those material properties which are basic to the
ideal material, such as the band structure, lattice constant, effective masses and
phonon frequencies.) The other study involves extrinsic properties of GaAs, a
material used for the fabrication of high-speed devices. (The extrinsic proper-
ties are defined to be those properties which arise from non-ideality, such as the
presence of impurities, defects, and the crystal quality, all of which can be influ-

enced by external factors.) Optical studies are useful for these studies since they

can provide a wide variety of information about both the intrinsic and extrinsic

properties of semiconductors, and do not involve contacting the material.

The majority of infrared detectors today are based upon semiconductors as
the light-sensitive elements. In order to successfully fabricate a detector, it is im-
portant to understand how the material interacts with light. This interaction can
be affected not only by the intrinsic properties of the material, but by extrinsic
factors as well, such as the presence of impurities. Much effort has been spent in
trying improve the performance of infrared detectors by controling the extrinsic
properties of materials used, since most materials have few adjustable intrinsic
parameters. Alloy semiconductors, such as Hg;,Cd,Te, are extensively used
since it is possible to tailor the band gap of these materials (an intrinsic prop-
erty) by varying the concentrations of the constituents. However, once the de-
sired band gap is chosen, the only remaining adjustable parameters are extrinsic.
Recently it has become possible to tailor the intrinsic properties of semiconduc-
tors by controlling the structure of the material at near inter-atomic dimensions.
This has opened up a whole new field of materials characterization, since the
properties of many of these structures have only been investigated theoretically.
Optical studies, such as photoluminescence, can provide useful information on
the important intrinsic properties of these materials, such as determining the

band-gap energies.

Most semiconductor devices are fabricated by doping selective regions of ma-
terial on an insulating substrate to isolate the various regions. The production of
devices using certain materials is limited by the availablility of suitable substrate
material. It is important for the substrate to remain insulating during the entire
fabrication process. This requires good control of the electrical properties of the
substrate. For most semiconductor materials, it is the impurities that determine

the electrical properties. Therefore it is important to understand how impuri-

ties behave in the material, and the mechanisms for their introduction into the
material. Optical studies provide a means for both identifying and studying the

behavior of impurities in semiconductors.

1.2 Superlattices

Recently it has become possible to fabricate layered semiconductor materials
whose structure is controlled at near interatomic dimensions. Such structures
can have properties which differ dramatically from those of bulk materials. One
such structure, first proposed by Esaki and Tsu in 1969,! is composed of a
series of alternating thin layers of two different materials. Such a structure is
called a superlattice, since the period of the lattice along the direction normal to
the plane of layers is many atomic spacings long. Figure 1.1 shows the general
structure of a superlattice. The substrate is usually bulk material of one of the
constituents, although alternate materials are sometimes used. The superlattice
layers may be grown by a number of techniques, with molecular beam epitaxy
(MBE) and metalorganic chemical vapor deposition (MOCVD) being the most
prevalent techniques. Both of these methods grow superlattices by depositing a
number of monolayers of each material onto a substrate. A buffer layer of one
of the materials is sometimes grown before the superlattice, usually to guarantee
a high quality surface for growing the superlattice. The layer thicknesses are
typically anywhere from 20 - 500A.

Figure 1.2 shows the spatial dependence of the constituent material band
edges for three different types of superlattices.?, In each case, the upper line
represents the energy of the conduction band minimum as a function of position,
and the lower line represents the energy of the valence band maximum. The

different superlattice types reflect the differences in the relative alignments of

Figure 1.1: Structure of a semiconductor superlattice. The superlattice itself

is composed of alternating thin layers of two different semiconductor materials.

The layer thicknesses are typically in the range of 20 — 500A.

TYPE |
7 : Y Y CONDUCTION
aa, BAND
T Y
- YY Z Yj
° Egp7Ega
VALENCE
BAND
TYPE Il — STAGGERED
CONDUCTION
t ae YY Y BAND
YY ie ZT Y
4 Lf
VALENCE
TYPE Il — MISALIGNED “N°
t L
> Ege’ Y Y] CONDUCTION
or T l BAND
2 ZA WA Ea Z
VALENCE
BAND
TYPE Ill
Y bp} 4 OY Vii. CONDUCTION
fe Y 9p? Y Y BAND
DISTANCE z= BAND

Figure 1.2: Spatial variations of the band edges for three types of superlattices.
The horizontal direction is perpendicular to the plane of the layers, and the
vertical direction is energy. In each case, material A has a smaller band gap
than material B. The shaded regions indicate the band gaps of the consituent

materials. (Taken from Ref. 2.)

the band edges of the two materials. The energy offset between the valence
band maxima of the two materials is called the band offset. Type I superlattices
are characterized by having a band offset such that the electrons and holes are
both confined in the same layers. GaAs-Ga,_,Al,As is an example of a type I
superlattice. Type II superlattices have band offsets such that the electrons are
confined to the layers adjacent to the holes. This category has been subdivided
into two cases. In the staggered case, there is some overlap between the bands,
while in the misaligned case there is no overlap. InAs-GaSb is an example of a
type II superlattice. The type III superlattice is similar to the type I superlattice,
except one of the materials has zero band gap. HgTe-CdTe is an example of a
type III superlattice, since HgTe is a zero band gap semiconductor.

Type I and type III superlattices posess a number of advantages over bulk
semiconductor materials for certain applications. In such superlattices, as shown
in Section 1.2.1, the band gaps depend on the thicknesses of layers, as well as
on the band gaps of the constituent materials. The band-gap energies can be
adjusted to values between the band gaps of the constituent materials, and a
given band-gap energy can be obtained by a number of combinations of layer
thicknesses. Thus the superlattice system has two adjustable parameters. This
is an improvement over ternary semiconductor alloys, where the choice of a de-
sired band-gap energy determines all the other material properties. These and
other unique properties can be exploited for devices.?, (The GaAs-Ga,_,Al, As
superlattice system has been very successful for fabricating lasers.*) For some su-
perlattices, it is possible to determine the effects of the compositional modulation

on the band structure using a simple model.

1.2.1 The Kronig-Penney model

For some materials, it is possible to determine the band structure of a super-

Viz)

Ahke 0 FF Be zZ—

Figure 1.3: Periodic square-well potential for conduction band electrons in a
superlattice. The horizontal axis is distance perpendicular to the plane of the
layers. The vertical axis is energy. |, and lg are the thicknesses of the layers of

materials A and B, respectively.

lattice by considering the behavior of the conduction and valence bands sepa-
rately. (This study will be limited to the band structure in the direction normal
to the layers.) In this case, an electron in the conduction band would see a po-
tential like that in Fig. 1.3. Material B is assumed to have a larger band gap
than material A. The regions of material B present barriers to the electrons in
material A. The potential therefore looks like a periodic array of finite square
wells. The band structure for such a potential was solved by Kronig and Penney
in 1931.4 (A good description of the Kronig-Penney model is given in Ref. 5.)

In this model, the Schrédinger equation

hn? d?¥

is solved for one electron in the potential

0 n(latlp)V(z) = (la + In) Va + ln) +1 n = integer, (1.2)

where J, and lg are the thicknesses of the layers of materials A and B, respec-
tively. In the regions of material A, where the potential is zero, the solution will

be a linear combination of traveling waves,

WU = Acta? + Be that, (1.3)
where
2 2
pan ha (1.4)
2m

In the regions of material B, where the potential is Vo, the solution is

W = Ce*s* +. De *a*, (1.5)
where
h7kp?
E=V- . 1.6
0 (1.6)

The boundary conditions are the continuity of VW and dW/dz at z =O and z= Iza,

as well as the Bloch condition. The solution is then given by

kp” — ka?

sinhkglgsinkal, + coshkglgcoskgl, = cos qd, (1.7)
2kpka

where q is the electron momentum perpendicular to the layers, and d=I,4 +l,
is the superlattice period. A similar anaylsis can be performed for the valence
band.

Equation 1.7 gives the band structure of the superlattice normal to the plane
of the layers. The band gap will be given by the q = 0 solutions of the conduction
and valence band equations. This analysis is well suited to type I superlattices
when the two constituent materials have similar band structures. The GaAs-
Ga ,_,Al, As superlattice is an example of a type I superlattice where the Kronig-
Penney model can be used to describe the band structure. However, it may
not be appropriate for other types of superlattices. The Kronig-Penney model
is not well suited to type II or III structures since the interactions between the
conduction and valence bands of the two materials is important. However, it

does provide a basis for understanding the properties of superlattices.

1.2.2 Optical studies of superlattices

New superlattice materials are being proposed and grown at a rapid rate. Op-
tical techniques provide a non-destructive means for determining important prop-
erties of superlattice materials. For example, photoluminescence experiments can
provide information about the band gap of a superlattice. Non-destructive mea-
surements are required during the early stages of development of a particular

system, since the number of samples may be quite limited.

1.3 Impurities and defects

In bulk semiconductors, the presence of impurities or defects can have a
profound effect on the electrical properties. The most well known types of impu-
rities are the shallow donors and acceptors. For the majority of semiconductor
devices, the electrical characteristics are governed by controlling the concentra-
tions of shallow impurities. An important step in learning how to control the
properties of a semiconductor is the identification of the centers responsible for
the shallow levels. Optical measurements can be very useful here, since they can
provide positive identification of the presence of many specific impurities. One
method for identifying acceptors in semiconductors is by measuring the excited

state spectra.

1.3.1 Effective mass theory

One method for calculating the binding energy of an acceptor in a semicon-
ductor is the effective mass theory. Here the acceptor is treated as a negative
point charge in a uniform dielectric medium. The hole is treated as a positive
particle with mass equal to the heavy-hole effective mass. The resulting bind-

ing energies for the acceptor ground state and excited states are then given by

scaling the energies given by the Bohr model of the hydrogen atom by m,/me’,
where m, is the hole effective mass and e is the static dielectric constant for
the material. (Baldersechi and Lipari have modified the effective mass theory to
include the effects of crystal symmetry and interactions with the light-hole and

split-off hole valence bands.®*”)

The effective mass theory predicts that all acceptors will have the same bind-
ing energy, since only the coulombic part of the acceptor potential is considered.
This is typically a poor assumption for acceptors.As can be seen from Table 1.1,
the binding energy of an acceptor in GaAs depends upon the actual impurity.
The difference between the effective mass acceptor binding energy and the ac-
tual binding energy is called the central cell shift. A measurement of the central
cell shift provides a means of identifying the impurity producing an acceptor
in GaAs. The most accurate method for measuring the central cell shift of an
acceptor is to determine the energies of the excited states of the acceptor. The
p-like excited states of a semiconductor are not sensitive to the central cell shift
since they have little probability of being near the impurity center, so they expe-
rience only the average coulombic potential. Therefore, the p-like excited states
for all acceptors will be bound with the same energy with respect to the valence
band edge. (The excited state spectrum for effective mass acceptors in GaAs has
been calculated in Ref. 7.) The central cell shift is then given by the difference
between the measured binding energies of the p-like excited states relative to the
acceptor and those predicted by the theory. The s-like excited are also useful,
since they experience the non-coulombic part of the acceptor potential to a lesser
extent than the ground state does. This is useful for determining the behavior of
the central part of the potential, which can provide insight into the structure of
the acceptor. Since the relative amount of the central cell shift for the s-like ex-

cited states has been calculated, thay are also useful for determining the central

Table 1.1: Binding energies of some shallow acceptors in GaAs.*

Acceptor C Be Mg Zn Si Cd Ge Sn

Energy (meV) 26.0 28.0 28.4 30.7 34.5 34.7 404 171
Lattice site As Ga Ga Ga As Ga As As

a. These values were obtained from D. J. Ashen, P. J. Dean D. T. J. Hurle,
J. B. Mullin, A. M. White and P. D. Greene, J. Phys. Chem. Solids 36,
1041 (1975).

cell shift. There are a number of methods for measuring the binding energies of
acceptor excited states optically. Chapter 3 discusses two such techniques which

are useful for studying the s-like excited states of acceptors.

1.4 Outline of thesis

This thesis is divided into three parts, two devoted to the description of optical
studies of semiconductors, and the third to a description of a novel technique for

performing infrared photoluminescence measurements.

1.4.1 IRPL from HgTe-CdTe superlattices

Chapter 2 describes the first photoluminescence results from HgTe-CdTe su-
perlattices. This study is important because it represents the first direct determi-
nation of the band gap of a HgTe-CdTesuperlattice. The HgTe-CdTe superlattice
system was first proposed as an interesting infrared material in 1979. Since this

system is a type III superlattice, infrared band gaps should be easily attainable

for reasonable layer thicknesses. (Ignoring effects due to the discrete nature of
the layer thicknesses, it should be possible to obtain band gaps anywhere from
near OmeV to near the band gap of the barrier material in type III superlattices.)
There has been a great deal of interest in this system for use in infrared detectors,
since the superlattice structure should allow for the tailoring of some properties
at a chosen band-gap energy. It is hoped that the HgTe-CdTe superlattice would
therefore be superior to the Hg,_,Cd,Te alloy system, where all the material

properties are fixed for a given choice of the band gap.

The first HgTe-CdTe superlattices were grown in 1982. However, these sam-
ples were grown with very thick HgTe layers due to difficulties in controlling the
layer thicknesses. According to the theories, these first samples should have had
band-gap energies very near OmeV, and would not be suitable for optical studies
or as detector material. The degree of growth control has improved since then,
and the first samples with HgTe layers thin enough to have infrared band gaps
were grown in 1984. The photoluminescence study reported in Chapter 2 was

performed using two of these samples.

The results of this study showed that the HgTe-CdTe superlattice system is
a promising infrared material. Two superlattices with different layer thicknesses
and from different sources were shown to luminesce in the infrared. The lumi-
nescence was shown to be consistent with wave-vector conserving band-to-band
recombination, allowing the first direct determination of the band gap of a HgTe-
CdTe superlattice. Analysis of the luminescence lineshapes showed evidence for
the existence of fluctuations in the layer thicknesses. In each case the lumines-
cence was shown to occur at significantly lower energies than that of Hg;_,Cd,Te
alloys with the same Cd concentrations as the average Cd concentrations of the
superlattices. The temperature dependence of the luminescence was compared

with theoretical predictions. The theory predicted the band gaps of the two

structures fairly well, if allowance was made for fluctuations and small errors in
the measurements of the sample layer thicknesses. The differences in the behav-
ior of the two superlattices prove that it is possible to tailor the properties of

HgTe-CdTe superlattices.

1.4.2 The 78-meV acceptor in GaAs

Chapter 3 reports the first observation of the s-like excited states of a double
acceptor in a semiconductor. The center being studied was the 78-meV acceptor
in GaAs. It has been found that the presence of the 78-meV acceptor in liquid
encapsulated Czochralski (LEC) GaAs indicates the material is p-type, and not
suitable for use as a semi-insulating substrate material. It was suspected that this
level was due to a double acceptor, possibly involving the Gay, anti-site defect.
A determination of the nature of this acceptor will be necessary for controlling
the properties of LEC GaAs. The purpose of this study was to measure the
s-like excited state spectrum of this level, which should provide information on
the potential of this acceptor.

Two experiments were performed, and each showed a splitting of the first
s-like excited state of the 78-meV acceptor. The techniques used, selective exci-
tation luminescence (SEL) and electronic Raman scattering (ERS), were chosen
for their ability to measure the s-like excited states of single acceptors in semi-
conductors. An effective mass-like description for a double acceptor in a semicon-
ductor was proposed to explain the observed splitting. This theory was the first
to predict a splitting of the 1s!2s! state a double acceptor in a semiconductor. A
comparison of the data with the theoretical predictions for single acceptors and
the double acceptor model led to the conclusion that the 78-meV level in GaAs
is due to the first ionization of a double acceptor. The techniques used for this

study should be useful for identifying double acceptors in other materials.

14
1.4.3 Photoluminescence using an FTIR

Appendix A describes the method used to adapt an FTIR to infrared pho-
toluminescence measurements in the presence of a thermal background. The
experiment in Chapter 2 required the ability to detect a weak infrared signal
whose wavelength was unknown. Initial attempts to use a standard grating
spectrometer were unsuccessful due to the poor signal-to-noise ratio inherent
in such a system. The grating spectrometer was then replaced with a Fourier-
transform infrared spectrophotometer (FTIR). The FTIR had the ability to scan
a large wavelength region in a short time with a greater signal-to-noise ratio than
that achievable with the grating instrument. However, the small signal strength
and the presence of a 300K blackbody radiation background prevented the ob-
servation of the signal. This led to the development of a novel technique for

eliminating the background signal when using an FTIR.

References

1. L. Esaki and R. Tsu, IBM Research Note RC-2418 (1969).

2. L. Esaki, in Proceedings of the 17th International Conference on the Physics

of Semiconductors (Springer Veralg, Berlin, 1985) pp. 473-483.

3. The experiment in Chapter 2 was performed using a commercially available
GaAs-Ga,_,Al,As multi-quantum well laser diode. The experiment would
not have been possible with conventional laser diodes due to the high optical

power requirements.

4. R. De L. Kronig and W. J. Penney, Proc. Roy. Soc. (London) A130, 499
(1931).

5. C. Kittel, Introduction to Solid State Physics, 5th edition (John Wiley and
Sons, New York, 1976) pp. 191-193.

6. A. Baldereschi and N. O. Lipari, Phys. Rev. B8, 2679 (1973).

7. A. Baldereschi and N. O. Lipari, Phys. Rev. B9, 1525 (1974).

Chapter 2

IRPL from HgTe-CdTe

Superlattices

2.1 Introduction

2.1.1 Background

The Hg;_,Cd,Te alloy system has been an important material for use in
infrared detectors for many years. The usefulness of this system arises from the
basic properties of HgTe and CdTe. HgTe is a zero band gap semiconductor, while
CdTe has a band gap near 1.6eV.! As an alloy system, the band-gap energy of
Hg,_,Cd,Te can be tailored to any value between 0 and 1.6eV. This allows the
tailoring of detectors for specific regions in the infrared. Hg,_,Cd,Te detectors
have been used mostly in the regions where the atmosphere is transparent, usually

2 although there has been recent interest in

in the 3 — 5 and 8 — 13m ranges,
using Hg;_,Cd,Te detectors with fiber optics in the 1.3 - 1.5 um range. There are
a number of material-dependent parameters which can affect the performance of

an infrared detector.* In the alloy system, once the desired band-gap energy has

been chosen, all the other parameters, such as the carrier effective masses, are

determined. Given such fundamental limitations to Hg;_,Cd,Te detectors, new

infrared materials are needed to improve device performance.

The HgTe-CdTe superlattice system was proposed as an alternative to the

45 A HgTe-CdTe superlattice consists of alternating

Hg,_,Cd,Te alloy system.
thin layers of HgTe and CdTe. It is predicted that such a superlattice would
have markedly different properties from a HgCdTe alloy of the same average

45 With this system, it is possible to obtain a given band-gap en-

composition.
ergy with a number of different combinations of HgTe and CdTe layer thicknesses
(ignoring for the moment the discrete nature of the layer thicknesses). The choice
of the layer thicknesses used can then be made to optimize other material prop-
erties such as effective masses. These properties might be exploited to improve

the performance of devices.®

A number of groups have reported the successful fabrication of HgTe-CdTe

superlattices.’~9%

The samples have been characterized by Auger-electron spec-
troscopy, transmission-electron microscopy, secondary-ion-mass spectroscopy and
reflectance, all of which indicate the existence of spatial variations of the Hg to
Cd ratios. However, until this study, no direct determination of the band gaps
has been made. Infrared transmission measurements have been made on some

superlattices.®:1°

However, in each case the samples consist of a rather thin
layer of superlattice on a substrate. The small change in transmission along
with Fabry-Perot resonances make the interpretation of the data rather difficult,
leading to some question as to the precise value of the band gap. In contrast,
photoluminescence measurements are expected to give a less ambiguous value
for the band gap. The advantage here is that photoluminescence should produce
a peak near the band gap. Infrared transmission experiments usually involve

fitting the transmission curve to a theoretical model, and extrapolating to find

the band-gap energy. This can produce ambiguous results with thin samples

since the absorption may not become significant until the energy is well above

the band gap.

2.1.2 Results of this study

This chapter reports on the first measurements of photoluminescence from
HgTe-CdTe superlattices. Two superlattice samples of differing composition,
and from different sources, were shown to luminesce in the infrared. Analysis
of the luminescence lineshapes showed the luminescence to be consistent with
that expected for wave-vector conserving band-to-band recombination. These
measurements then represent the first direct determination of the band gaps
of HgTe-CdTe superlattices. The dependence of the luminescence on the sample
temperatures was also studied, from 100 to 270K. In each case, the luminescence
was shown to occur at significantly lower energies than that of Hg,_,Cd,Te al-
loys with Cd concentrations equivalent to the average Cd concentrations of the
respective superlattices. This result is consistent with the current theories, which
predict such an effect.4°11_ A comparison of the data with a simple temperature-
dependent model of HgTe-CdTe superlattices was made, showing good qualita-
tive agreement. This study proves that HgTe-CdTe superlattices have different
optical properties than Hg,_,Cd,Te alloys. The photoluminescence data seemed
to show a lack of impurity luminescence, indicating that the electrical properties
of the HgTe-CdTe superlattices may be superior to those of Hg,_,Cd,Te alloys.
The difference in the luminescence from the two samples shows that it is possible
to tailor the properties of HgTe-CdTe superlattices, an important requirement

for device fabrication.

2.2 Sample descriptions

Two different HgTe-CdTe superlattice samples from two different sources were
used for this experiment. Both samples were grown in molecular-beam epitaxy
(MBE) machines specially tailored to handle Hg. Sample 1 was grown to have
250 repeats of 38 to 40A of HgTe followed by 18 to 20A of CdTe.? The layer
thicknesses were obtained from the calibrated growth rates for the source fluxes
used. The total superlattice thickness was measured by a surface profilometer,
and confirmed by analysis of the interference fringes in the infrared transmission
data.!° The total superlattice thickness obtained by these methods (~1.5 um) is
approximately equal to that obtained by using the source fluxes. The substrate
material for sample 1 was (111)-oriented Cdo.9gZno.o4Te. The zinc concentration
was chosen such that the lattice constant of the substrate would be between that
of CdTe and that of HgTe. It has been shown that higher quality crystal growth of
both HgTe-CdTe superlattices and Hg;_,Cd,Te alloys can be obtained by using
such CdZnTe substrates.’ Sample 2 was grown to have 75 repeats of 50A HgTe
followed by 50A CdTe.® The total thickness as measured by a mechanical stylus
was ~0.75 um, in approximate agreement with the sum of the layer thicknesses.
The substrate material for sample 2 was (111)-oriented CdTe. A 2500A CdTe
buffer layer was grown before the superlattice. (No buffer layer was grown on
the substrate for sample 1 since the MBE machine was not equipped to grow
CdZnTe.) No direct measurement of the individual layer thicknesses of either
superlattice was made, therefore the accuracy of the quoted thicknesses is not
easily determined.

A HgCdTe alloy sample was used for comparison with the superlattice pho-
toluminescence spectra. The sample was Hgo.71Cdo,29Te, grown by the solid state
recrystallization method, and was annealed in Hg vapor to reduce the number of

Hg vacancies.'”

2.3 Description of experiment

2.3.1 General

The photoluminescence experiment for this study required good response over
a wide range of infrared wavelengths. Early theoretical predictions for the band
gap values of the samples used in this study ranged from 5 to 124m at OK. Thus
there was no way of knowing where to expect the luminescence. Scanning this
large a wavelength region presents a few problems. First, the detectors available
for this range have rather limited detectivities (see Table 2.1). This necessitates
long integration times to observe even moderately strong sources. The type of
optics to be used is another important consideration (see Table 2.2). Many of
the materials which are transparent in the infrared have drawbacks such as water
solubility or large thermal expansion coefficients which limit their usefulness. The
experiment is further complicated by the large background of 300K blackbody
radiation, which peaks at 10m. All of these factors must be carefully considered

when designing the experiment.

2.3.2 Dispersive spectrometer experiment

The initial experimental setup used was based on a SPEX 1404 double grating
spectrometer (see Fig. 2.1). This setup was used for all the substrate lumines-
cence studies, as well as the initial search for the superlattice photoluminescence.
The spectrometer could cover a range of 0.4 —- 12m by using a series of gratings,
each blazed for a particular wavelength (see Table 2.3). The gratings had to be
changed to reach the various ranges, which added to the time required for the
experiment.

Three detectors were used to cover the region being studied. The detector

used for the 0.4 — 1.2m range was an S-1 curve photomultiplier tube cooled

Table 2.1: Properties of selected infrared detectors.*

Material Useful range’ D*(Amaz)° r oe Temp*
(um) (emHz/*4/w) (um) (K)
InAs 1 — 3.3 5 x 101 3 77
InSb 1—-5.5 8 x 101° 5 77
Ge:Cu 2-27 1.5 x 10° 20 4.2
Ge:Zn 2 - 43 1 x 107° 33 4.2
3-15 2 x 107° 10-14
HgCdTef 77
5 — 20 3 x 10° 14 - 16

. The values used in this table are from the SBRC Inc. Infrared Components

Brochure.

. This range represents typical wavelengths where the detector may be used,
although the performance can be much lower than the peak performance

when used far from Amaz-
. This value represents the maximum detectivity of the detector.
. The wavelength where the detectivity is greatest.
. The temperature at which the detector is operated.

. The parameters for HgCdTe detectors depend on the alloy composition

used, which determines Amaz.

Table 2.2: Properties of selected infrared optical materials.*

Material Useful range’ Transmission® Other Properties
(um) %

Quartz 2-3 94 hard, will temperature
cycle

Sapphire 2-5.5 94 hard, will temperature
cycle

BaF, 2-11 94 cleaves, hygroscopic

IRTRAN2 5 — 12 70 hard, will temperature
cycle

KCl 2-17 94 soft, water soluble

ZnSe 6 — 20 70 hard, good for CO,
laser optics

KRS5 6 — 40 70 toxic, soft, cold flows
under pressure

CsI 3 — 50 85 soft, highly water solu-

ble

a. The values used in this table are from the Janos Technology Inc. Precision

Optics and Components Catalogue.

6. This range represents the wavelengths where the material is transparent,

although the transmission can be much lower than the peak value near

the ends of the range.

c. This value represents the maximum transmission of the material. The

behavior away from the maximum depends on the material.

Double grating

Spectrometer filter luminescence
( dT BaF, optics
Lock—in

output ref. input

preamp

immersion
dewar

4j=>~+} sample

laser
filter

§ | chopper
a)

? 9
VCO
MCA

Computer

Kr
laser

Figure 2.1: Dispersive spectrometer experimental setup. The output windows

on the dewar were IRTRAN2 and ZnSe. The transfer lenses were BaF,. A Ge

filter was placed before the spectrometer entrance slit to prevent higher order

light from hitting the detector. The detector was Ge:Cu at 4.2K. The chopping

frequency was 1kHz.

Table 2.3: Available gratings for spectrometer.

blaze wavelength grooves /mm wavelength range
(um) (um)
1.6 600 0.4 — 1.2 (second order)
1.6 600 1.2 — 2.4 (first order)
3.0 300 2.4 -— 4.0
6.0 150 4.0 — 8.0
10.0 75 8.0 — 12.0

to 77K. This was used to study the photoluminescence properties of the super-
lattice substrates. The 1.2 — 3.0um range was covered by an InAs photovoltaic
detector cooled to 77K. A copper-doped germanium detector (Ge:Cu) was used
to cover the remaining 3 ~- 12m range. All the external optics used were BaF 2,
which is transparent from 0.3 - 13m and only slightly hygroscopic. The sam-
ples were mounted in a Janis DT-8 variable temperature immersion dewar. The
use of an immersion dewar presents a problem in the infrared. There are few
infrared materials which can be cemented to the aluminum sample chamber and
cycle from room temperature to 1.8K. IRTRAN2 was the best compromise for
this application, since it could tolerate the temperature range, and has a similar
coefficient of thermal expansion to that of aluminum. This choice limited the
long wavelength transmission to about 12m (see Table 2.2). The dewar also
had an internal cold shield between the sample chamber and the outer wall, but
no window was used here (the transmission of the available windows is not very
high, and their effectiveness as a cold shield is limited by their transparency to

infrared radiation) The outer window was ZnSe, used mainly due to its avail-

ablity (BaF, would have been preferable). The window was not anti-reflection
coated, which would have limited its range of usefulness. The tradeoff was 70%
transmission over the extended range. A Ge band-pass filter was placed in front
of the spectrometer entrance slit to filter out the laser light, as well as to prevent

higher order light from hitting the detector.

A Coherent CR-3000K krypton ion laser was used as the optical pump source.
Either the 6471A or the 6764A line was used, and was chosen by using a SPEX
1460 grating spectrometer to filter the beam. To minimize the effects of the 300K
blackbody radiation background, the standard method of synchronous detection
was employed. When using the Ge:Cu detector, the laser light was chopped
at a frequency of 1.02kHz, chosen to maximize the signal-to-noise ratio of the
detector. The Ge:Cu detector used was an SBRC 9145-2, with a matching model
A130 preamplifier. The ouput of the preamplifier was fed into a PAR model
124A lock-in amplifier using a model 117 preamplifier. The input filter on the
lock-in was set at band-pass mode at the chopping frequency, and the Q was set
to 100 (the maximum setting). This was designed to filter noise, since the signal
would be a slowly varying modulation at the chopping frequency. The output
time constant was typically 1 — 3 seconds with a 12db/octaveslope. This output
filtering was necessary due to the weak signal strength. The output from the
lock-in amplifier was sent to a VCO based on an Analog Devices 458J chip. The
VCO provided an output pulse rate proportional to the input signal, and had a
dynamic range of 10°, since the full scale output voltage of the lock-in matched
the full scale voltage of the VCO (10 volts). The gain setting on the lock-in
amplifier could then be chosen such that the RMS noise level corresponded to
about 100 counts out of the VCO. In practice, this meant the lock-in was always
set for the lowest gain (x 20). By using this technique, the maximum benefit

of the VCO dynamic range could be obtained, and it was possible to avoid

overloading the input amplifier of the lock-in with a strong signal. The ouput
pulses from the VCO were then fed to a multichannel analyzer for storage, and

on to a computer for analysis.

2.3.3 Limitations of the dispersive approach

A photoluminescence scan of the HgCdTe alloy sample is shown in Fig. 2.2.
The signal-to-noise ratio in this spectrum is not very good, considering that it
required an integration time of 15 minutes. Luminescence from this sample could
be detected only at temperatures below 20K using this setup. Using this setup,
no luminescence signal was seen from either superlattice in the 1.6 — 12 um range.
All the luminescence at wavelengths shorter than 1.6m could be attributed to
the substrates. The low signal-to-noise ratio observed here can be traced to
limitations in the setup. In a double grating spectrometer, the grating efficiency
drops very rapidly as it is tuned away from the blaze angle. For the spectrometer
used here, the throughput for gratings blazed at wavelength \ will be down to
25% of the maximum at 2/3 and 2. The gratings must be changed frequently
when scanning large wavelength ranges to achieve maximum throughput, and
there are still regions where the efficiency is quite low. Atmospheric absorptions
are quite strong in some regions in the infrared, therefore it is important to limit
the optical path length in air for the signal. Unfortunately, even though the
spectrometer was nitrogen purged, the flow rate was not great enough to lower
the absorption significantly (the optical path length within the spectrometer is
nearly 8 meters, making it by far the longest path in the system). Further, the
BaF, optics used were not optimized for either the detector or the spectrometer.
The field of view of the detector was 60°, which corresponds to f/0.58, and
was not matched by the optics. All of these factors lead to a lowering of the

signal-to-noise ratio.

Hgo.70Cdo.30Te PL

T = 6.0K \
200 300

INTENSITY (ARBITRARY UNITS)

T = 2.9K

0 100 400

ENERGY (meV)

Figure 2.2: Photoluminescence spectra of a Hgo,79Cdo.39Te alloy, taken with the
grating spectrometer setup. The gratings used were blazed for 64m (see Ta-
ble 2.3). A Ge:Cu detector was used, and the total scanning times were 15
minutes in each case. These scans represent the maximum signal-to-noise ratio

attainable in this wavelength range due to their proximity to the blaze wave-

length.

It was then necessary to devise a photoluminescence experiment with a greater
signal-to-noise ratio. The main disadvantage of the above setup is the use of a
grating spectrometer. Dispersive instruments are poorly suited to infrared work
since any increase in resolution is accompanied by a decrease in the signal-to-noise
ratio. Only a small fraction of the total luminescence intensity is incident on the
detector at any given time, and as the resolution increases, this fraction decreases.
In many photodetectors, the signal-to-noise ratio increases as the intensity of light
falling on the detector increases.° This is true of the photoconductive detectors
used for this study. A better signal-to-noise ratio would be achieved if all the
light could fall on the detector during the scan. This is precisely the situation in
a Fourier transform infrared spectrophotometer.

A Fourier transform infrared spectrophotometer (FTIR) is basically a Michel-
son interferometer with a moving mirror for scanning in wavelength (see Ap-
pendix A for a more complete description). Such a spectrometer posesses two
major advantages over the dispersive instrument: the ability to have a large per-
centage of the sample luminescence incident on the detector throughout the scan,
and the ability to scan large wavelength regions without the need to change grat-
ings. The end result is a much greater signal-to-noise ratio than that achieved by
dispersive instruments. An FTIR also doesn’t suffer from the loss of throughput
inherent in grating spectrometers off the blaze wavelength, however there are
losses due to the efficiency of the beamsplitter.

An FTIR is not a solution in itself, however. Appendix A discusses the
problems associated with doing infrared photoluminescence with an FTIR, and

presents a unique solution for overcoming them.

2.3.4 FTIR experiment

The final experimental setup was based on a Bomem DA3.01 Fourier trans-

form infrared spectrophotometer (see Fig. 2.3). This particular instrument
offered the advantages of an external viewport and easy access to the signal from
the detector before it entered the Fourier transform computer. The detector used
was zinc-doped germanium (Ge:Zn), cooled to 4.2K. The internal beamsplitter
and the external viewport were KCl (see Table 2.2). The entire spectrometer was
evacuated to 17mTorr during the measurements. The external infrared optics
were all KCl, and matched to the f/numbers of the spectrometer input (f/4) and
the dewar exit window (f/2). A long-pass Ge filter with a 2m cutoff wavelength

was placed before the FTIR external viewport to filter out stray laser light.

The optical pump source was an SDL 2410-C multi-quantum well laser diode
operated at 77K, with a typical average power of 200mW.1* The laser diode was
powered by an HP 214B pulser, operating at a frequency of 40kHz, with a 50%
duty cycle. The signal from the Ge:Zn detector was sent to an Ithaco 1211 current
mode preamplifier. The high frequency response limit of the 1211 was 56kHz;
therefore, the pulsing frequency was limited to 40kHz to prevent attenuation
and phase shifting of the signal. The output from the preamplifier was sent to
a PAR 124A lock-in amplifier for demodulation. The lock-in preamplifier was a
PAR 116 operated in direct mode. The lock-in gain was typically 10*. The 116
preamp was not optimized for this configuration, since the output impedance of
the 1211 preamplifier was 600, and the input impedance of the 116 was 10MN.
The noise figure for this arrangment was 6db (the amplifier added 6db to the
thermal noise at the input). The input filter on the lock-in was set to 20kHz,
high pass, with a Q of 1. The output time constant on the lock-in was set to
minimum, which corresponded to a value of about 600ys. The slope was set
to 6db/octave. The ouput from the lock-in was sent to the Fourier transform

computer. Appendix A describes the motivation behind the settings chosen.

The samples were mounted in a Janis Supertran variable temperature dewar,

. le
KCI optics samp
Bomem aaa e-—-
external -- ~~~,
past etme UE) [oe
FTIR poo oT vm
. i \
Ge filter / \
Ge:Zn
Detector =
| |
| |
Tf fT
bias | |
; |
} supply =
Preamp © pulser ° | x
[output] [trigger] [output]
2 t
laser
diode (77K)
(u)
[reference]
vector
3) Lock—in @ |
[in] amplifier [out] processor

Figure 2.3: FTIR experimental setup. The infrared optics were all KCl, including

the FTIR beamsplitter. A Ge filter was placed before the spectrometer viewport

to filter out stray laser light. The detector was Ge:Zn cooled to 4.2K. The laser

frequency was 40kHz.

with a KCl exit window. This dewar consists of a liquid-helium-cooled cold
finger which is mounted in a vacuum. This arrangement posed a sample heating
problem. The samples were originally mounted by the method used in immersion
dewars — masking tape was used to hold the samples to the cold finger. While
this is sufficient for immersion dewars, it did not work well in a vacuum; when
the samples were illuminated with 200 mW average laser power, the samples were
observed to heat up from 12K to over room temperature in a few minutes. The
mounting procedure was changed to the following: the samples were placed on
indium foil on the copper cold finger. A corner of the foil was folded over to the
front of the sample, and used as the contact point for a metal spring clamp. This
method worked fairly well for heat sinking the sample, but had the disadvantage
of straining the sample slightly. However, there was no indication that the strain

had any effect on the luminescence.

The wavelength range covered by this system was roughly 2-17 um. A system
response curve was not obtained since there was no broadband infrared source
available which could be modulated at 40kHz. Given the degree of electronic
filtering of the system, a meaningful response curve would require the source to
be modulated at the frequency used for the measurements. Given the output
time constant of the lock-in amplifier, and the 6db/octave slope, the 3db signal
strength point occured at 265Hz. Thus the signals from wavelengths modulated
by the FTIR to frequencies higher than 265 Hz would be attenuated by the lock-
in amplifier. Figure 2.4 shows the theoretical response curve of this filter is
plotted against the energy of the luminescence signal. All the spectra were
corrected for this filter response. The effect of the correction is not large, but
it is readily visible in the noise amplitude. Without the correction, the noise
amplitude decreases toward higher energy (higher frequency), while after the

correction the noise amplitude is fairly independent of the frequency.

Output low pass filter response

100

40 + 6 db/octave

RELATIVE RESPONSE (%)

TC = O0.6msec

20 F 7

Oo A j 4 1 1 1 1
0 100 200 300 400

ENERGY (meV)

Figure 2.4: Theoretical response curve for the output low-pass filter, plotted
against the energy of the luminescence entering the spectrometer. The curve was
determined from the filter response and the wavenumber-to-frequency conversion

for the mirror speed used in the experiments.

The improvement in the signal-to-noise ratio gained by using this technique
was impressive. Compare the signal-to-noise ratio of the 30K curve in Fig. 2.5
with that of Fig. 2.2. It was possible to measure the photoluminescence from the
alloy sample up to room temperature with this setup, whereas the upper limit for
the original setup was 20K. The broad feature in the upper two curves centered at
120meV is thought to be due to room temperature blackbody radiation reflected

off the sample.

2.3.5 Modulated reflectivity

If the reflectivity of the sample were modulated at the laser frequency, then
the detector would see specularly reflected raditation from the room temperature
entrance window on the dewar. This would produce a broad spectrum, peaking
at 120meV. Such an effect has been seen in Hg,_,Cd,Te alloys.45 In this
case, above band-gap radiation creates free carriers in the bands, which behave
as an electron-hole plasma. At sufficiently high carrier densities, the dielectric
function can be appreciably altered, changing the reflectivity. This effect is most
pronounced at long wavelengths.

An experiment was performed to test for modulated reflectivity of the alloy
sample used here. The setup is shown in Fig. 2.6. The sample was mounted on
an aluminum block at room temperature to provide some degree of heat sinking.
The 6471A line of a krypton ion laser was used as the pump source. The plasma
lines were filtered by a Spex 1460 grating spectrometer. The illuminating laser
power density was chosen to be 12W/cm’?, to match that used in the IRPL
experiment. The laser beam was chopped at 1kHz, and was incident normal
to the sample surface. A blackbody source at 1470K was placed at 45° to
the sample surface, such that the specularly reflected blackbody radiation was

incident on a Hg,_,Cd,Te detector. A long-wavelength pass Ge filter was placed

~Hgo.71Cdo.29Te IRPL
T = 240K
Scale: x110

F-

>»

= T = 160K

an) Scale: x45

ee”

Lil

cod

T = 30K
Scale: x1

0 100 200 300

ENERGY (meV)

400

Figure 2.5: This figure shows IRPL from a Hgo,71Cdo.29Te alloy sample. The tem-

perature for each curve is given at the left, along with a relative vertical scale

indication. The optical pump density was 12W/cm?. Note the improved sig-

nal-to-noise over the dispersive spectrum. The broad feature centered at 120meV

is due to room temperature blackbody radiation, and is explained in the text.

Lock—in

ref. in input
0) Q

preamp

HgCdTe BaF,
detector optics

cr
y, IN

Ge filter .“ |\ ‘sample
laser v

sample mount

ref. out ’

BaF. optics

! Y beam
laser Y chopper

| |

, ||

filter [LJ] S
——

mirror blackbody
source

Figure 2.6: Modulated reflectivity experimental setup. A Kr ion laser was used
as the pump source. The specularly reflected blackbody radiation was directed
onto a HgCdTe detector, cooled to 77K. A Ge filter was used to block stray laser
light. The infrared optics were all BaF,. The signal was observed visually on the

lock-in amplifier. The sample was at room temperature.

before the detector to prevent scattered laser light from hitting the detector.
The output from the detector was synchronously demodulated at the chopping
frequency with a lock-in amplifier. If either the laser or the blackbody were
blocked, no signal was seen at the detector. This proved that the signal was not
due to the laser or blackbody alone. With the laser on, a small signal could be
seen at the lock-in. This signal was due to the reflectivity of the sample being
modulated at the laser frequency. When the laser power was increased by a factor
of 5, a very small signal was observed at 180° out of phase with the stronger
signal. This signal was interpreted to be due to modulating the temperature of

the sample at the chopping frequency.

A final test for modulated reflectivity was performed using the FTIR ex-
periment by placing a glass window on the cold shield input port. This would
ensure that the specular reflection off the sample would be from a surface with
a temperature between room temperature and the sample temperature. The
signal strength of the 104m feature was observed to decrease in this case. The
distinguishing features of this signal are that it is very broad, relatively weak,
and the peak position is independent of the sample temperature. It seemed to
show up more dramatically at high temperatures, due to the lower luminescence
signal strength, and hence the expanded vertical scale. The luminescence signal
strength is expected to decrease with increasing temperature, and the output
low-pass filter rolloff will decrease the observed intensity of signals as they shift
to higher energies. Another factor is that the ambient temperature of the dewar

is higher; thus, the cold shield will be closer to room temperature.

In the subsequent experiments, care was taken to ensure that the FTIR viewed
a cold surface in the specular reflection off the samples. This was accomplished
either by placing a glass window on the cold shield input port, or turning the

sample such that the specular reflection along the line of sight of the FTIR came

from the cold shield wall. The FTIR experiment made it possible to observe the

photoluminescence from the two superlattice samples.

2.4 Results

2.4.1 Hg, ,Cd,Te alloy IRPL

Infrared photoluminescence measurements were made on a Hgo.71Cdo.29Te
sample to test the FTIR system and for comparison with the superlattice re-
sults. Figure 2.7 shows four IRPL spectra from the Hgo,.71Cdo.29Te alloy sample,
each taken at a different temperature. Alloy spectra were taken at a range of
temperatures, from 20 — 185K, although alloy luminescence was detected even at
room temperature. The lower scan in Fig. 2.7 shows a distinct peak at 199meV,
a shoulder on the high energy side at 213 meV, and a shoulder on the low energy
side at roughly 7meV. As the temperature increases to 50K, the low energy
shoulder disappears, and the high energy shoulder becomes a pronounced peak.
By 100K, the original peak is merely a slight shoulder on the high energy line,
and has completely vanished by 185K. The dip in the 185K spectrum at 290 meV
corresponds to the 4.2 — 4.3 4m atmospheric CO, absorption line. The broad fea-
ture in the 185K spectrum near 120meV is thought to be due to reflected room
temperature blackbody radiation, as described in Section 2.3.5. This sample was
included for comparison with HgTe-CdTe superlattice sample 1. If the superlat-
tice layers grown in sample 1 had completely interdiffused into a uniform alloy,
then the resulting Cd concentration would have been 33%. Thus the band gap

would be greater than that for the Hgo.71Cdo.29Te alloy sample.

2.4.2 HgTe-CdTe superlattice sample 1 IRPL

Figure 2.8 shows infrared photoluminescence data from HgTe-CdTe superlat-

INTENSITY (ARBITRARY UNITS)

~Hgg.71Cdp.o9Te IRPL

T = 185K
Scale: x45

T=
Scale: x8
T = 50K b

Scale: x1.5

T = 12K
Scale: x1
100 200 500 400

ENERGY (meV)

Figure 2.7: Infrared photoluminescence spectra from a Hgo.71Cdo.99Te alloy. The

sample temperature for each spectrum is indicated at the left. The scale refer-

ence is used to indicate the relative magnification on the vertical axis for each

spectrum, with the bottom spectrum taken as 1.

tice sample 1, taken at four different temperatures. At 180K, the peak position
of sample 1 is at 122 meV, with a full width at half-maximum intensity of 48 meV.
Contrast this with the Hgo.71;Cdo.29Te sample at 185K, where the peak position is
at 276meV. The luminescence peak from sample 1 occured at lower energy than
the peak from the alloy sample at all temperatures, consistent with the theoret-
ical predictions for superlattices. The signal intensity seen from sample 1 was
less intense than that of the alloy sample at all temperatures. No luminescence
signal was detected from sample 1 at temperatures below 80K. This is probably
due to the luminescence peak position shifting to a wavelength beyond the cutoff

of the optical system at 17 wm (73 meV).

2.4.3 HgTe-CdTe superlattice sample 2 IRPL

Figure 2.9 shows infrared photoluminescence data from HgTe-CdTe superlat-
tice sample 2, taken at four different temperatures. At 170K, the peak position
is near 205meV, with a full width at half-maximum intensity of 74meV, consid-
erably broader than the peak from sample 1. There is a second less-intense peak
near 87meV, which is present only at temperatures 170K and below. It is not
certain whether this signal is due to a luminescence process in the sample, or has
some other origin. There is a possibility that it is due to modulated reflectivity
of the sample, but this is not likely. First, the signal goes away as the temper-
ature is increased, contrary to the alloy case, and second, the peak position is
too low in energy to correspond to room temperature blackbody radiation. No
luminescence signal was seen from sample 2 at temperatures below 110K. The
peak intensity from sample 2 over most of the temperature range was lower than
that from sample 1 by a factor of approximately 3 at a pump power of 1.7 times
that used for sample 1. This intensity difference accounts for the additional noise

seen in the spectra from sample 2.

Hgte/CdTe SAMPLE 1 IRPL

T = 240K
“ns
Y)
za
=)
< T = 180K
aa)
oe
~~
> T = 140K
7)
Zz
LJ

0 100 200 300 400
ENERGY (meV)

Figure 2.8: Infrared photoluminescence spectra from HgTe-CdTe superlattice
sample 1. The sample temperature for each spectrum is at the right. The long
wavelength cutoff of the optical response of the system is near 73meV (17 um).

Each spectrum is plotted on a different vertical scale.

Hgte/CdTe SAMPLE 2 IRPL

T = 240K
7“_-
WY)
Zz
a)
Y T = 200K
aa)
ad
Zz
LJ

0 100 200 300 400

ENERGY (meV)

Figure 2.9: Infrared photoluminescence spectra from HgTe-CdTe superlattice
sample 2. The sample temperature for each spectrum is at the right. The long
wavelength cutoff of the optical response of the system is near 73meV (17 um).

Each spectrum is plotted on a different vertical scale.

42
2.4.4 HgTe-CdTe sample 1 temperature dependence

Figure 2.10 shows the temperature dependence of the IRPL from HgTe-CdTe
superlatticesample1. The circles indicate the energy of the peak in the photolu-
minescence intensity, while the bars indicate the energies of the points where the
luminescence intensity is one-half the peak intensity. We were unable to observe
luminescence at temperatures lower than 80K since the peak wavelength shifted
beyond the cutoff of the optical system at 73meV (17m). The solid line gives
the band-gap energy for a Hgo.¢7Cdo33Te alloy,® which would have the same
average Cd concentration as sample 1. It is clearly evident that the peak energy
for the luminescence from the superlattice is lower than that for the alloy. This
difference is strong evidence for sample 1 being a superlattice. A limited study
of the pump power dependence of the luminescence was made at high sample
temperatures, and no change in the lineshape was observed. However, given the
narrow range of pump powers available and the signal-to-noise ratio, a small

change might not have been discernible.

2.4.5 HgTe-CdTe sample 2 temperature dependence

Figure 2.11 is a similar plot of the temperature dependence of the lumines-
cence from HgTe-CdTe superlattice sample 2. Also included is a curve showing
the band-gap energy for a Hgo.59Cdo.59Te alloy,!® which would have the same
average Cd concentration as sample 2. The luminescence peak from sample 2 is
clearly at lower energy than that for the alloy band gap. There is also a consid-
erable difference in the temperature dependences — the band gap of a 50% Cd
Hgi-zCd,Te alloy shows almost no temperature dependence. This is evidence
that sample 2 is a superlattice. The position of the low energy peak is not plot-
ted here, nor did its peak position show any temperature dependence. It was

not possible to determine whether this peak was related to the superlattice. No

500 . Li . ' . ' . Ly . Lu " T . Ly . LI

IRPL from Sample 1 O

Hgo.67Cdo.331'e Band Gap

400 F
300

200 +

co aetts

ENERGY (meV)

6) 4 1 ‘ L 1 L ‘ 1 ‘ I . l 1 L 2 L 1
0 40 80 120 160 200 240 280 320

TEMPERATURE (K)

Figure 2.10: Temperature dependence of the infrared photoluminescence signal
from HgTe-CdTe superlattice sample 1. The circles indicate the energy of the
peak signal, while the bars indicate the energies of the half-intensity points. The

solid line gives the band-gap energy for a Hgo¢7Cdo.33Te, which has the same

average Cd concentration as sample 1.

600 5
IRPL from Sample 2 O
500 F
Hg9.50Cd9.59'e Band—Gap
0)
—“ 400 F
we
ul
Zz
Li
300 +
200 +
100 . H 1 I 4 ii ‘ J. ‘ L rn si 1 ] 4 l

0 40 80 120 160 200 240 280 320
TEMPERATURE (kK)

Figure 2.11: Temperature dependence of the infrared photoluminescence signal
from HgTe-CdTe superlattice sample 2. The circles indicate the energy of the
peak signal, while the bars indicate the energies of the half-intensity points. The
solid line gives the band-gap energy for a Hgo59Cdo59Te, which has the same

average Cd concentration as sample 2.

study of the pump power dependence of the luminescence was made due to the
low signal strength from the sample (the spectra were taken at the high power

limit of the laser).

2.4.6 Signal origins

To insure that the peaks in the IRPL spectra from the superlattices originated
from the superlattices and not the substrates, both samples were turned over
and the substrates irradiated with the same laser. The laser wavelength was
greater than the band-gap energies of either of the substrates, ensuring that the
superlattices would not be illuminated. Neither substrate showed a peak in the 73
— 500meV energy range. The luminescence observed at energies above 500 meV
could all be attributed to the substrates. Another way to be sure that the
signals are from the superlattices and not the substrates is via the temperature
dependence. The band-gap energies of both substrates decrease with increasing
temperature, while the luminescence peak energies from the superlattices increase

with increasing temperature.

2.4.7 Substrate luminescence

The luminescence from both substrates was measured to identify the signals
related to the substrates. The experimental setup used was similar to that in
Fig. 2.1, except an S-1 curve photomultiplier with the usual photon counting
electronics was used for the 0.5 — 1.2 um range, instead of the solid-state detector
using synchronous detection. In each case, the substrate side of the samples
was illuminated with the 6471A line of a Kr ion laser. The absorption depth
for this wavelength of light in the substrates was less than 1m in both cases,
avoiding the possibility of exciting the superlattices (the substrates were both

roughly 1mm thick). The substrate for sample 1 was Cd,_,Zn,Te, with a stated

Zn concentration of roughly 4%. The substrate for sample 2 was CdTe. The
luminescence data in the near-band gap range for both substrates is shown in

Fig. 2.12.

Both samples exhibit strong acceptor-bound exciton lines near their respec-
tive band gaps. The weak lines toward the high energy side of the large peaks
represent the free exciton lines and excited states of bound excitons. The lumi-
nescence from the substrate of sample 2 is typical of bulk-grown CdTe. The main
bound exciton peak occurs at 1589meV. The peaks on the low energy side of the
main bound exciton peak have all been observed before, and include LO phonon
replicas of the main bound exciton line, donor-acceptor luminescence and phonon
replicas of to donor-acceptor lines. The large peak near 1473 meV has been seen
before in CdTe,!”18 and has been attributed recombination of an exciton at an
isoelectronic oxygen trap.!® This line is usually accompanied by a series of LO
phonon replicas to lower energy, due to the strong LO phonon coupling of defects

in CdTe. This series is evident in the spectrum, showing 3 replicas.

The luminescence from the substrate of sample 1 is shown in the lower trace
of Fig. 2.12. The band gap for CdTe at 6K is 1605 meV, while the gap for ZnTe
at 6K is 2380meV. As expected, the band gap of the CdZnTe substrate is greater
than that of CdTe. The main bound exciton peak occurs at 1613meV, 24meV
higher in energy than that for the CdTe substrate. All the lines present in this
spectrum appear to be analogs of lines previously observed in CdTe. There is
a common series of photoluminescence lines in CdTe, with the no-phonon line
occurring att 1449meV, and up to seven phonon replicas to lower energy, com-
monly referred to as the 1.42eV band.!® The periodic structure in the 1470 —
1550 meV range corresponds to this 1.42eV band in CdTe. The difference in the
energies of the lines may be attributed to the larger band gap of Cd,_,Zn,Te,

and the accompanying change in the binding energies of the impurities involved.

SUBSTRATE LUMINESCENCE

Sample 2 substrate
(CdTe)
T = 6K

Sample 1 substrate
(CdZnTe)
T = 6K

INTENSITY (ARBITRARY UNITS)
LC

1300 1400 1500 1600

ENERGY (meV)

Figure 2.12: Luminescence from the superlattice substrates. The substrates were
illuminated directly with the 6471A line of a Kr ion laser. The upper curve is
from the substrate of sample 2, and shows typical CdTe luminescence. The lower
curve is from the substrate of sample 1, and shows luminescence typical of low-Zn

concentration CdZnTe.

The luminescence from this substrate was compared with a Cd,_,Zn,Te sample
with a 4.5% Zn concentration. The luminescence was very similar, and the differ-
ences could all be attributed to differences in the band-gap energies (caused by
differences in the Zn concentration), or to the presence of different defects. Lumi-
nescence from this substrate was also observed through through the superlattice
as well, and no difference was observed. For both samples, the luminescence in

the 1.65 — 3um range was all attributed to the substrates.

2.5 Interpretation

2.5.1 Luminescence processes

In order to gain a better understanding of the properties of the superlattices, it
is important to determine the process or processes producing the luminescence.
The luminescence spectra from sample 1 (see Fig. 2.9) exhibit asymmetrical
lineshapes which are narrower at lower sample temperatures. This behavior is
suggestive of a process involving a thermal distribution of the charge carriers
before recombination. One possibility for the superlattice luminescence signal is
the recombination of electrons and holes near the respective band edges. In this
case, it is possible to describe the resulting luminescence lineshapes theoretically,
and compare them to the superlattice luminescence. Two possible schemes for
band-to-band and near band-gap recombination are discussed below, and the
lineshapes are compared with spectra from HgTe-CdTe sample 1. The signal-to-
noise ratios of the spectra from sample 2 are considerably lower than those from

sample 1, limiting the usefulness of such fits in this case.

Wave-vector conserving recombination

In most semiconductors, the recombination of electrons and holes proceeds

via wave-vector conserving transitions. There are a number of possible sources
for such luminescence in a semiconductor at high temperatures. The two most
likely sources are band-to-band recombination and free exciton recombination. In
band-to-band recombination, an electron in the conduction band may recombine
with a hole in the valence band. When carriers are excited into the conduction
and valence bands, the coulomb attraction between the electron and hole can give
rise to a free exciton. The electron and hole in the exciton recombine by conserv-
ing wave-vector. The HgTe-CdTe superlattice luminescence is expected to be due
to one of these processes. In GaAs-Ga,_,Al,As superlattices, free exciton recom-
bination is the dominant source of luminescence at low temperatures,?%?! while
at higher temperatures band-to-band luminescence is the dominant process.??
In either case, the luminescence lineshape will be similar.

The lineshape for such wave-vector conserving transitions is described below.
Since a photon has negligible wave-vector, the energy of the electron and the
hole are both fixed for a given emitted photon energy (see Fig. 2.13). Thus the
density of states at a given photon energy may be expressed as a joint density
of states for both bands, g(hv). The intensity of emitted radiation at an energy

hv is given by

I(hv) = P.(E.)Pr(Ey)g(hv) E> E, 0.1
0 E

where P.(£,) is the probability of an electron occupying a state with energy E;,
and P;,(E,) is the probability of a hole occupying a state with energy E,. If the

bands are assumed to be parabolic, then the joint density of states is given by
g(hv) = A(hv — E,)*/?, (2.2)

where A is independent of the photon energy.”? g(hv) behaves like the density

of states for a band whose effective mass is the reduced mass of the electron and

Figure 2.13: Band-to-band photoluminescence diagram. hv is the energy of
the photon emitted by the recombination. The energy is measured using the

conduction band minimum as the origin.

hole effective masses. The occupation probabilities are given by the Fermi-Dirac

functions

P. (£.) 1+ e(Ee—He)/keT (2 3)

waa

where p, and pa, are the electron and hole chemical potentials, respectively. kg
is the Boltzmann constant, and T is the sample temperature. The electron and

hole energies are known functions of the photon energy
Ee = me (hy — E;)

E, = wtlE,— hv),

(2.4)

where m, and my, are the electron and hole effective masses respectively, and m,
is the reduced mass of m, and Mh.

This distribution is fairly complicated, but it has two interesting limits. In the
limit of low carrier concentration and high temperature (non-degenerate limit),

the occupation probabilites can be replaced by simple Boltzmann factors, leaving
I(hv) = C(hv — E,)V?e(Fs~hy)/ka (2.5)

The constant C will be a function of the carrier concentration, but it will only
affect the amplitude, not the lineshape. The peak position as a function of

temperature is given by
hVpear(T) = E, (T) + ghat, (2.6)

where E,(T) is band-gap energy at the temperature T. Thus the peak position of
such luminescence will shift to higher energy at a rate of skp faster than the band
gap. If the carrier concentration generated by the laser is small compared to the
thermal occupation of the bands, then this approximation will hold. In this limit,
changing the laser power by a small amount will not affect the lineshape. The

other limit where the intensity expression simplifies is in the limit of high carrier

concentration (degenerate limit). In this limit, the electron and hole chemical
potentials can be treated as a free-electron fermi gas, and are proportional to n2/3,
The luminescence lineshape is then a strong function of the carrier concentration.
Thus a small change in the laser power will result in a significant change in the
lineshape. In the intermediate region, the chemical potentials are described by
Fermi integrals, complicating the analysis.

Figure 2.14 shows the least-squares fit of the non-degenerate limit of wave-
vector conserving recombination (long-dashed line) to the 270K spectrum from
HgTe-CdTe sample 1. The temperature was fixed for the fit, and the free
parameters used were the band-gap energy, E,, and the amplitude. Thus the
amplitude and the width are coupled at a given temperature. The band-gap
value determined from this fit was 143meV. The use of the non-degenerate limit
is justified by the lack of pump-power dependence of the luminescence lineshape,
and the high sample temperatures. Such fits were made to the spectra at other
temperatures, and the results were similar in each case. The general shape of
the luminescence is suggestive of a wave-vector conserving process, however the
luminescence linewidth is wider than that predicted by this model. This width
may be attributable to fluctuations in the superlattice layer thicknesses. The
result of such fluctuations would be a superposition of a range of band gaps,

broadening the observed luminescence line width.

Superlattice layer fluctuations

It is quite likely that there are fluctuations in the superlattice layer thick-
nesses, across the sample horizontally or vertically. There has been much research
into this phenomenon in GaAs-Ga,_,Al,As superlattices, where the luminescence
linewidths have been ascribed to fluctuations in the layer thicknesses.2°- 2224

Each of these references attributes the widths to thickness fluctuations within

HgTe/CdTe Sample 1 Lineshape

T = 270K

INTENSITY (ARBITRARY UNITS)

400

ENERGY (meV)

Figure 2.14: IRPL spectrum from HgTe-CdTe sample 1, with two luminescence
lineshape fits. The solid curve is a 270K IRPL spectrum from HgTe-CdTe sample
1. The long-dashed curve is the least-squares fit to the case of wave-vector
conserving recombination in the non-degenerate limit. The short-dashed curve
is the least-squares fit to the case of non-wave-vector conserving recombination

in the non-degenerate limit.

the layers, as opposed to thickness fluctuations along the direction of growth.
This conclusion was based on the degree of control of the layer-to-layer growth
rates, and the lack of control over the growth in the plane of the layers, as well
as by analysis of the resulting linewidths. In the high temperature photolumi-
nescence study, the luminescence linewidths were observed to be larger for the

22, This increase was attributed to the

samples with thinner layer thicknesses.
increased effect of small layer thickness fluctuations for such samples. The width
was shown to be up to a factor of four wider than that due to the thermal
distribution of the carriers.

The observed linewidth of sample 1 at 270K is a 1.5 times as wide as that
predicted for wave-vector conserving recombination. This compares favorably
with the GaAs-Ga,_,Al,As results. Thus it seems quite likely that the lumines-

cence from sample 1 is due to wave-vector conserving recombination, and that

there is some degree of fluctuation in the layer thicknesses.

Non-wave-vector conserving recombination

In some semiconductors, electrons in the conduction band can recombine
with holes in the valence band without conserving wave-vector. There are a
number of mechanisms which can relax the wave-vector conservation. If the
carrier concentration is sufficiently high, carrier-carrier scattering can relax the
conservation. In alloy materials like Hg;_,Cd,Te, scattering from disorder in
the crystal can also cause relaxation of wave-vector conservation. However, it
is difficult to find a mechanism to relax wave-vector conservation in HgTe-CdTe
superlattices. Such recombination is not common in other superlattices, and
the superlattice should not exhibit the large degree of disorder present in the
alloys. Also, the luminescence lineshape showed no dependence on pump power,

indicating the carrier concentration is too low to relax wave-vector conservation.

Even though this mechanism is not justifiable for the HgTe-CdTe superlattices
studied here, it is instructive to compare this case with the luminescence, since
there is evidence for this process in Hg;_,Cd,Te alloys.”

In this non-wave-vector conservating scheme, an electron in the conduction
band can recombine with any hole in the valence band. The intensity of the

luminescence can be written as
I(hv) = A | P.(©)Pa(e)De(€)D. (ede, (2.7)

where D,(e), Dn(e) are the density of states in the conduction and valence bands,
respectively. If the energy origin is chosen occording to Fig. 2.13, the result is

e/2(hy — € — E,)'/?
[1 + elhye“Eyt;)/koT | [2 + ele) /kaT]

(hv) =f de, (2.8)

where kg is the Boltzmann constant, and T is the sample temperature.”> Again,
fe and pp, are the electron and hole chemical potentials, respectively. It has been
assumed that the conduction and valence bands are parabolic. This lineshape
simplifies greatly in the limit of low carrier density and high temperatures (non-
degenerate limit). The Fermi-Dirac functions in the denominator then come out

of the integral, resulting in
I(hv) = C(hv — E,)*e(Fo-h)/kat (2.9)

Again, the carrier concentration affects only the magnitude, not the width of the
lineshape. The luminescence peak position as a function of temperature is given
by

hUpeak = Ey(T) + 2kaT, (2.10)
where E,(T) is the band-gap energy at the temperature T. Thus, the lumines-

cence peak position will shift to higher energy at a rate of 2kg greater than the

band-gap energy.

The least-squares fit of the non-degenerate limit lineshape to the 270K spec-
trum from HgTe-CdTe sample 1 is shown in Fig. 2.14 on page 53 as the short-
dashed line. The band-gap value determined from the fit was 116meV. Similar
fits were obtained from the spectra at other temperatures. The agreement be-
tween the data and the fit might look good, however it is too wide. The full-width
at half-maximum intensity of the observed luminescence at 270 K is 65 meV, while
the width of the fit is 79meV. This extra width is significant, and it occurs for
fits to spectra at other temperatures. It is very difficult to find a mechanism
for narrowing the observed superlattice luminescence linewidth, especially since
any fluctuations in the layer thicknesses would increase the observed linewidth.
Therefore it is very unlikely that the luminescence from HgTe-CdTe superlattice

sample 1 is due to non-wave-vector conserving recombination.

Wave-vector conserving processes

As mentioned above, there are two important wave-vector conserving pro-
cesses which have been observed in GaAs-Ga,;_,Al,As superlattices. Given the
limited range of temperatures for which luminescence spectra are available, it
is difficult to make a definitive determination of the luminescence process. It is
much easier to determine luminescence process with data at low temperatures,
where thermal effects are minimized. However, there is much to be learned from
the high temperature luminescence data. The temperature dependence of the
peak position of the luminescence is quite linear, and only a single peak is visible
at all the temperatures studied. This indicates that only one process is responsi-
ble for the luminescence in this temperature range. If either the electron or hole
is in a bound state before recombination, the binding energies of the states must
be at least 23meV for it to produce luminescence at 270K. Given the absence

of a peak due to a second process, the binding energy would have to be even

greater. (If it is assumed that a second process would be visible if it were 10%
as probable as the main process, then the binding energy of bound state would

have to be at least 46meV.)

In most ITJ-V compounds the free exciton binding energy is usually a few
millivolts. It has been shown that the free exciton binding energy in III-V su-
perlattices can be increased by two-dimensional confinement effects, and free
excitons have been observed in room temperature absorption data.?® This in-
crease in the binding energy is caused by an increase in the electron effective
mass in the superlattices. The electron effective mass for HgTe-CdTe superlat-
tice sample 1 is 0.02 m, parallel to the layers and 0.009 m, normal to the layers,
while the heavy hole effective mass is expected to be about 0.5 m,.?” This sug-
gests a binding energy of only a few meV, which would not be large enough to
allow free excitons to be observed at these high temperatures. Therefore, free
exciton luminescence is probably not the source of the luminescence from sample
1. This conclusion could be confirmed if luminescence data were available at low

temperatures, which should show the presence of free exciton recombination.

There are other processes leading to near-band-gap luminescence. It is possi-
ble for excitons to bind to neutral impurities (bound excitons). The binding en-
ergy for bound excitons is typically 2-3 meV in low Cd concentration Hg;_,Cd,Te

alloys.?5

This luminescence would disappear at temperatures above 30-40K.
There is no evidence to suggest that the binding energies of bound excitons
in superlattices are significantly larger than those in the constituent materials.

Therefore, it is unlikely that the luminescence from HgTe-CdTe sample 1 is due

to bound excitons.

In bulk semiconductors, it is possible to have recombination where either
the electron or hole, or both, are bound to impurities. Such impurity lumines-

cence has been reported in GaAs-Ga,_,Al,As superlattices.22 The luminescence

was observed only at low temperatures, and exhibited different behavior than
impurity luminescence in bulk semiconductors. This effect is expected for su-
perlattices, since the binding energies for impurities depend on the position of
the impurities within the layers.29 A simple model for impurity luminescence is
given in Ref. 30. Given the high temperatures studied for sample 1, it is unlikely
that the luminescence is impurity related.

The luminescence spectra from sample 1, taken over a range of 100 to 200K,
show only a single peak, whose position is a linear function of the temperature.
This indicates that a single process is responsible for the luminescence over the
entire temperature range. The high sample temperatures imply that neither the
electron nor hole are in bound states proir to the recombination. This is com-
patible with high temperature luminescence data from GaAs-Ga,_,Al,As super-
lattices, where processes involving bound electrons or holes are observed only at
lower temperatures. Analysis of the lineshape fits agrees with this conclusion,
and indicates that the luminescence process conserves wave-vector. Therefore it
may be concluded that the luminescence from HgTe-CdTe sample 1 is due to

wave-vector conserving band-to-band recombination.

HgTe-CdTe sample 2 luminescence

As can be seen from Figs. 2.8 and 2.9, the luminescence linewidths of sample
2 are much broader than those from sample 1 at all temperatures. The observed
signal strength from sample 2 was lower than that from sample 1 as well. The
temperature dependence of the peak positions indicates that this sample is a
superlattice. The luminescence peak position occurs at significantly lower en-
ergy than that of a 50% Cd Hgi_,Cd,Te alloy, and shifts to higher energy with
increasing temperature at a rate of 0.59meV/K. This shift is much larger than

that of Hgi_,Cd,Te alloys, even those whose band-gap energies are similar to

those of sample 2.

There are a number of possibilities for the increased linewidth and lower signal
strength from sample 2. As mentioned above for sample 1, it is possible that there
are spatial fluctuations in the layer thicknesses of this sample. Given the large
widths of the luminescence peaks (74meV at 170K), this possibility seems quite
likely. Although the linewidth fluctuates from scan to scan, a comparison of the
observed linewidths with those expected for wave-vector conserving recombina-
tion shows the luminescence to be roughly a factor of 2 to 3 wider than predicted
by the theory. These widths compare favorably with those observed in some
GaAs-Ga,_,Al,As superlattices, indicating that layer thickness fluctuations are
the probable source of the increased widths. There are other possibilities which
may account for the lower signal strength. The small thickness of the super-
lattice (~ 7500 A) might be expected to produce less luminescence than sample
1. If the carrier lifetime were sufficiently long to allow the carriers to diffuse
a significant fraction of the superlattice thickness before recombining, then the
integrated luminescence intensity would depend upon the total thickness of the
sample. Also, it is possible that the layer thickness fluctuations play a greater
role in the luminescence from this sample, perhaps lowering the luminescence

efficiency as well as increasing the linewidth.

2.5.2 Comparison with theory

A simple theory for calculating the band-gap energies of HgTe-CdTe super-
lattices is presented in Refs. 11 and 30. This theory is more complicated than
the Kronig-Penney model discussed in Chapter 1. The Kronig-Penney model is
essentially a one-band approximation, where the superlattice conduction band
energy is determined by considering only the spatial variations in the conduction

band energy from one material to the other. This method works satisfactorily

for GaAs-Ga,.,Al,As superlattices, but the situation for HgTe-CdTe superlat-
tices is more complicated. In CdTe, the conduction band has s-like symmetry
and the valence band has p-like symmetry, which is common for most wide-gap
semiconductors. In HgTe, however, both the conduction and valence bands have
p-like symmetry. It is this difference in the symmetries of the HgTe and CdTe
conduction bands that precludes the use of a simple one-band approximation for
the superlattice states. The model developed by G. Bastard in Refs. 11 and 30 is
a two-band approximation, which includes effects due to mixing of the s-like and
p-like states which make up the bands. This model makes simple calculations of
the band-gap energies possible for a wide range of superlattice parameters, and
agrees well with more involved calculations.?”. However, this model only predicts

the band-gap energies of superlattices at OK.

Guldner et al. introduced temperature dependence into the model by us-
ing the temperature-dependent band-gap energies for HgTe and CdTe.*! The
band-gap energies for HgTe and CdTe at the chosen temperature are used as
the paramters for use in the model developed by Bastard. The end result is
an approximation to the temperature dependence of the superlattice band gap.
This model is sensitive to the band-gap values used for the constituent materi-
als. Unfortunately, there are discrepancies in the literature on the temperature-
dependent band gaps of HgTe and CdTe. These differences can change the pre-
dicted superlattice band-gap energies by up to 10meV, as well as changing the
temperature dependence. For this study, the temperature-dependent band-gap
values used for HgTe and CdTe were obtained from the empirical Hg,_,Cd,Te
temperature dependent model of Ref. 16, applied for the cases of x = 0 and
x =1. While the model was stated to be valid for these values, it predicted the
OK band gap of CdTe to be 1.645eV, while the more commonly accepted value

is 1.608eV.*:33 The use of these values was not ideal, but it provided a good

enough approximation for comparison with the data.

Theoretical predictions for sample 1

Figure 2.15 shows how the luminescence data from sample 1 compare with the
predictions of this theoretical model. An offset of 40meV between the valence-
band maxima of HgTe and CdTe, with HgTe higher, was used for the calculations.
This seems to be the most commonly accepted value,**** although the evidence is
not conclusive. Since both superlattices were grown on (111)-oriented substrates,
the actual layer thicknesses should be discrete multiples of the lattice constant
along this direction (3.73A). For both samples, it was assumed that the total
layer thickness and the total number of layers were known. Given the stated
layer thicknesses for sample 1 (404 HgTe and 20A CdTe), the nearest layer
numbers are 11 layers of HgTe (41A) and 5 (18.7A) layers of CdTe. If the
luminescence were due to a wave-vector conserving band-to-band process, then
the actual band gap would be near the low energy threshold of the luminescence
peak. However, as Table 2.4 and Fig. 2.15 show, the temperature-dependent
Bastard theory for these layer thicknesses predicts too large a band gap. The
luminescence lies between the energies predicted for 12/4 layers of HgTe/CdTe
and for 13/3 layers of HgTe/CdTe. Assuming the luminescence is due to band-
to-band recombination, the best fit to the data from sample 1 is provided by 13
layers of HgTe (48.5 A) and 3 layers of CdTe (11.2 A), and is plotted in Fig. 2.15.
For these layer thicknesses, the theory gives values of 68.3 meV for the band gap
at 100K, and 114meV at 240K. Experimentally, the low energy threshold at
100K is about 75 meV, and at 240K the threshold is at 110 meV, so the observed
band-gap energy is somewhat greater than that predicted by the theory for these
layer thicknesses. If a linear fit is made to the peak positions for the IRPL

from sample 1, the slope obtained is 0.42meV/K. The theory predicts a slope of

300 | Sample 1 IRPL O
Theory 11/5 — — -
280 + Theory 12/4 ------
Theory 13/3 —-— B

240 b _-
S Loo
© — — -
2 200 } Le _
> _ _ _ — Le 7
° 160 bE — ae
ul coy —

120-0 ee 4 ae

| eon _—
, -_
_—
80 TF b “oa
-_—
-_—
-_—
40-7
@) L 1 rn Ll ‘ ] 1 j L L nl I r j 1 |

fe) 40 80 120 160 200 240 280 320
TEMPERATURE (k)

Figure 2.15: Comparison of the HgTe-CdTe superlattice sample 1 data with the
temperature-dependent Bastard model. The circles represent the IRPL peak
positions for sample 1. The bars represent the full-widths at half-maximum in-
tensity. The dashed lines are theoretical curves, where the number of HgTe/CdTe

atomic layers is given in each case.

0.33meV/K, which is 1kg less than the shift of the peak. This is greater than
the value expected for wave-vector conserving band-to-band recombination. The
agreement is fairly good given that the temperature-dependent band gaps used
for HgTe and CdTe are not precise. Using different estimates of the HgTe and
CdTe band gaps from the literature, the predicted superlattice band gaps can
be changed by up to 10meV. Therefore, it is best to conclude only that the
observed superlattice band gap falls between those predicted by the theory for
12/4 and 13/3 layers of HgTe/CdTe. If the theory is accurate, this implies an
error of 24 to 40% in the measurment of the CdTe concentration. Such an error
is possible given that the individual layer thicknesses were measured only by

calibrated values of the source fluxes.

Theoretical predictions for sample 2

Figure 2.16 shows how the luminescence data from sample 2 compare with
the predictions of this theoretical model. Given the stated layer thicknesses
for sample 2 (50A HgTe and 50A CdTe), the closest number of atomic layers
are 13 each of HgTe and CdTe (48.5A). The theoretical predicition for this
combination of layer thicknesses is plotted in Fig. 2.16 as the long-dashed line.
Note that the slope of the luminescence data is much steeper than that predicted
by the theory. Theoretical curves for two other compositions, 15/11 (56A/41A)
and 17/9 (63.4A/33.6A) layers of HgTe/CdTe, are also plotted in Fig. 2.16.
None of the curves seems to match the data very well, although they are all
in the right energy range at some temperatures. In each case, the slope of the
luminescence peak position is 4kg greater than that predicted by the theory. (It is
possible that this discrepancy is caused by the HgTe and CdTe band-gap values
used in the theory.) The 0K band gap for 17/9 layers of HgTe/CdTe is very

close to the extrapolation of the luminescence peak position to OK. However,

Table 2.4: HgTe-CdTe Superlattice Results.

Sample 1 Sample2 Units
stated comp." 40/20 50/50 A
ave. Cd conc.? 33% 50%
IRPL peak OK intercept 46 102 meV
position vs. T° slope 0.42 0.59 meV/K
equivalent OK band gap 296 592 meV
HgCdTe alloy? slope 0.17 0 meV/K
— theory — { comp. 11/5 13/13 # layers
stated layer 4 OK band gap 154 172 meV
thicknesses® slope 0.27 0.24 meV/K
—theory— ( comp. 13/3 17/9 = # layers
best fit OK band gap 38 101 meV
to the dataf | slope 0.33 0.24 meV/K

a. The HgTe/CdTe layer thicknesses given by the crystal growers.

b. The average cadmium concentration in the superlattice.

c. These values are from linear fits to the IRPL data.

d. An alloy with the same average Cd concentration as the superlattice.

e. The values for the HgTe/CdTe layer numbers closest to the stated thick-

nesses, from the Bastard model discussed in the text.

f. These values are the best fit of the Bastard model to the data.

320 +
280 + _ |
Ty C - 7
240 + -- bu- -—
S _ _a@ 7 [4+ ee
& 200 } _~4-7 7 weer
war _— — 1-7 -_—™
25 Le a wer “- 7 aa
ra 160F ee - L eo
GOR aan
120 } aa
a Sample 2 IRPL oO
80 + Theory 13/13 — — -
Theory 15/11 ------
40 7 Theory 17/9 —-—
O i I 1 l 1 1 4 I 1 | 1 Ll 1 1 1 I

0 40 80 120 160 200 240 280 320
TEMPERATURE. (K)

Figure 2.16: Comparison of the HgTe-CdTe superlattice sample 2 data with the
temperature-dependent Bastard model. The squares represent the IRPL peak
positions for sample 2. The bars represent the full-widths at half-maximum in-
tensity. The dashed lines are theoretical curves, where the number of HgTe/CdTe

atomic layers is given in each case.

it seems that any curve between 15/11 and 17/9 layers of HgTe/CdTe would
provide a reasonable fit to the data. This would imply an error of 16 to 30%
in the measurement of the average Cd concentration. It is also possible that
associating the superlattice band-gap energy with the low energy threshold for
the luminescence is not valid for this sample, since the lineshapes do not really

resemble band-to-band recombination.

Limitations of the temperature-dependent Bastard model

There are a number of parameters which can affect the predictions of the
temperature-dependent Bastard model. As discussed above, the band-gap values
used for HgTe and CdTe can significantly affect the predictions of this model.
Unfortunately, thare are no universally accepted values for either material. Given
the range of data present in the literature, the choice of values can lead to a
slight change in both the predicted band gaps, and the temperature dependence.
Therefore, care should be exercised when comparing actual band-gap values and
temperature dependences.

Another parameter which can change the theoretical predictions is the value
of the energy offset between the valence band of HgTe and the valence band of
CdTe. The band offset for this system has widely been assumed to be small,
usually 40 meV.*4°> Schottky barrier height measurements of HgTe-CdTe het-
erojunctions place an upper limit of roughly 500meV on the value of the band
offset. The band-gap energies predicted by the theory decrease with increasing
band offset. In order to draw any conclusions about the band offset from IRPL
measurements of HgTe-CdTe superlattices, it will be necessary to perform a sys-
tematic study of a number of superlattices configurations whose compositional

properties are more well known.

It is also possible that strain effects play a major role in determining the

band structure. The lattice constant of HgTe is 0.3% larger than that of CdTe,
which means that HgTe-CdTe superlattices are always strained. Sample 1 was
grown on a CdZnTe substrate whose lattice constant was between HgTe and
CdTe, spreading the strain out between the CdTe and HgTe layers. Therefore,
strain should not be very important in sample 1. Sample 2 was grown on CdTe;
therefore, the strain would be contained in the HgTe layers. The effects of strain
on HgTe-CdTe superlattices grown on CdTe have been investigated, and it does

change the band structure.*®

The calculations have been performed only for a
limited number superlattice configurations at OK, making it difficult to compare
the results with the samples used here. However, the results indicate that strain

may play an important role in the behavior of sample 2.

Finally, it is possible that there is some interdiffusion between the HgTe
and CdTe layers. The effects of interdiffusion on the band gaps of HgTe-CdTe
superlattices have been investigated, and the general result is an increase in the
band gap over the value expected without interdiffusion.*® It is not clear from
the data that any interdiffusion has occured. Such a determination might be
made possible by annealing a sample in stages, and observing the changes in the

luminescence.

2.6 Conclusions

The results of the first infrared photoluminescence measurements of HgTe-
CdTe superlattices were presented here. The temperature dependence of the
IRPL spectra of two different superlattices from different sources was measured.
In each case the luminescence was shown to occur at significantly lower ener-
gies than that for Hg;_,Cd,Te alloys with Cd concentrations equivalent to the

average Cd concentration of the respective superlattices. The luminescence line-

shapes from one of the samples were asymmetrical, and were also narrower at
lower temperatures. Analysis of the lineshape led to the conclusion that the
process producing the luminescence was wave-vector conserving band-to-band
recombination. The band-gap energy of the superlattice would then be near the
low energy threshold of the luminescence peak. A comparison of the data with
photoluminescence measurements on III-V superlatices indicated the presence of

small fluctuations in the superlattice layer thicknesses.

A comparison of the data with a temperature-dependent model of the band-
gap energies showed good qualitative agreement with the theory. As predicted
by the theory, the sample with the thicker CdTe layers had the large band gap.
The direction of the temperature dependence was also predicted; however, in
one case the magnitude of the change predicted was too small. This discrepancy
may not be significant given the uncertainties in the parameters used in the
theory. Good quantitative agreement between the theory and experiment was
possible if the assumption of errors in the layer thickness measurements was
made. Given the uncertainties in the parameters used in the theory and in the
sample characterizations, the agreement between the theory and experiment was

actually quite good.

The differences between the optical properties of the two superlattices prove
that it is possible to tailor the properties of HgTe-CdTe superlattices. This
is an important result, since it shows that the properties of superlattices are
more adjustable than those of Hg;_,Cd,Te alloys. The luminescence from the
superlattice samples did not seem to show the impurity-related luminescence
common to Hg;-,Cd,Te alloys. This is typical of III-V superlattices as well,
where the strength of impurity-related luminescence is much weaker than in
the alloy compounds. This could mean that HgTe-CdTe superlattices will have

electrical characteristics superior to those of Hg,_,Cd,Te alloys. These reuslts

indicate that HgTe-CdTe superlattices will become a very important infrared

material.

After the IRPL experiments involving the two original superlattice samples
were completed, similar experiments were performed on five other superlattice
samples. No luminescence was observed from any of these samples, from 12 to
300K. Some of the structures were grown with thick HgTe layers and thin CdTe
layers (e.g. 27 layers of HgTe and 13 layers of CdTe). The temperature-dependent
Bastard model predicted very small band gaps for these particular configurations.
Therefore, it is possible that the luminescence from these samples occured at
wavelengths longer than 17 wm, and were beyond the cutoff of the optical system.
It would be useful to extend the range of the optical system to longer wavelengths,
but as Table 2.2 shows, such materials are very difficult to work with and hard
to obtain. It is possible that the other samples were not of sufficient quality to
luminesce, although one of the samples showed signs of damage due to heating

by the laser (probably the result of poor sample mounting.)

IR transmission and photoconductivty measurements were made on sample 1
using the FTIR. These experiments confirmed the soft absorption edge predicted
for this sample. The absorption and photoconductivity did not become significant
until well above the band gap obtained from the IRPL experiment, which is

reasonable given the small thickness of this sample.

In order to obtain a better understanding of the photoluminescence and ab-
sorption in HgTe-CdTe superlattices, a systematic study of a number of super-
lattice configurations is necessary. This should help resolve questions concerning
the theory as well. It would be better if the total superlattice thicknesses were
greater than 2m, since this would increase the IR absorption to more man-
agable levels. The thin samples studied here presented problems due to the low

total absorption, and possibly also due to carrier diffusion. While superlattice

thicknesses aproaching 2m have been reported,® greater thicknesses may not
be easily attained. At the growth rates used for the superlattices studied here,
the growth time for 2m of material is about 2 hours. At the growth tempera-
tures used, about 180°C, interdiffusion of the bottom layers during growth may
present a limitation on the total thickness which can be obtained.

The current experimental results on HgTe-CdTe superlattices indicate that
they should be well suited for work as infrared detectors. Aside from actually
fabricating detectors, there are a number of useful optical experiments yet to
be performed on the superlattices. Time resolved IRPL would be helpful in
indentifying the luminescence process, and measuring the free carrier lifetimes.
Raman scattering experiments should provide information on band offsets, layer
thicknesses and carrier effective masses. HgTe-CdTe superlattices should prove

to be rich field for research in the near future.

References

1. R. Dornhaus and G. Nimtz, in Narrow Band Gap Semiconductors, Springer

Tracts in Modern Physics Vol. 98, (Springer-Verlag, Berlin, 1983).

2. M.B. Reine, A. K. Sood and T. J. Tredwell, in Semiconductors and Semimet-
als, edited by R. K. Willardson and A. C. Beer, Vol. 18 (Academic Press,
New York, 1981).

3. See for example the articles in Semiconductors and Semimetals, edited by

R. K. Willadrson and A. C. Beer, Vol. 18 (Academic Press, New York, 1981).
4, J. N. Schulman and T. C. McGill, Appl. Phys. Lett. 34, 663 (1979).

5. D. L. Smith, T. C. McGill and J. N. Schulman, Appl. Phys. Lett. 43, 180
(1983).

6. T. C. McGill, Proceedings of the 17th International Conference on the Phys-
ics of Semiconductors, San Francisco, 1984 (Springer-Verlag, Berlin, 1985)

p. 375.

7. J. P. Faurie, M. Boukereche, J. Reno, S. Sivananthan and C. Hsu, J. Vac.
Sci. Technol. A3, 55 (1985).

8. P. P. Chow and D. Johnson, J. Vac. Sci. Technol. A3, 67 (1985).

10.

1l.

12.

13.

14.

15.

16.

17.

18.

. J. M. Ballingall, F. A. Ponce, G. B. Anderson, B. J. Feldman and W. J.

Takei, 27th Electronic Materials Conference, Boulder, Co. 1985.

C.E. Jones, T. N. Casselman, J. P. Faurie, S. Perkowitz and J. N. Schulman,
Appl. Phys. Lett. 47, 141 (1985).

G. Bastard, Phys. Rev. B25, 7584 (1982).

This sample was provided by Peter Bratt of the Santa Barbara Research

Center. The anneal was also performed at SBRC.

For a good explanation of signal to noise in solid state detectors, see: S. M.
Sze, Physics of Semiconductor Devices, (John Wiley and Sons, New York,
1981) pp. 743-748, or A. Yariv, Introduction to Optical Electronics (Holt,
Rinehart and Winston, New York, 1976) pp. 309-314.

The SDL 2410-C laser diodes were chosen for their high CW output power,
and their ability to be pulsed at high frequencies at a 50% duty cycle. They
were designed to operate at room temperature, with a maximum CW output
power of 100mW. It was found that by cooling them to 77K that the CW

output power could be increased to near 500mW.

A. V. Nurmikko and B. D. Schwartz, J. Vac. Sci. Technol. 21, 229 (1982).
J. Chu, 8. Xu and D. Tang, Appl. Phys. Lett. 43, 1064 (1983).

J. P. Noblanc and G. Duraffourg, Phys. Stat. Sol. (b) 46, 705 (1971).

N. V. Agrinskaya, G. I. Aleksandrova, E. N. Arkad’eve, B. A. Atabekov,
O. A. Matveev, G. B. Perepelova, S. V. Prokof’ev and G. I. Shmanenkova,
Sov. Phys. Semicond. 8, 202 (1974).

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

There have been a large number of studies on this luminescence feature.
See for example: T. Taguchi, J. Shirafuji and Y. Inuishi, Osaka University
Technical Report 23, 195 (1973).

C. Weisbuch, R. Dingle, P. M. Petroff, A. C. Gossard and W. Weigmann,
Appl. Phys. Lett. 38, 840 (1981).

D. C. Reynolds, K. K. Bajaj, C. W. Litton, P. W. Yu, W. T. Masselink, R.
Fischer and H. Morkog, J. Vac. Sci. Technol. B 3, 694 (1985).

L. Goldstein, Y. Horikoshi, 8. Tarucha and H. Okamoto, Jpn. J. Appl. Phys.
22, 1489 (1983).

O. Madelung, Introduction to Solid State Theory, (Springer, Berlin, 1978)
p. 269.

J. Singh, K. K. Bajaj and S. Chaudhuri, Appl. Phys. Lett. 44, 805 (1984).
A. T. Hunter and T. C. McGill, J. Appl. Phys. 52, 5779 (1982).

See for example: D. S. Chemla, D. A. B. Miller, P. W. Smith, A. C. Gossard
and W. Wiegmann, IEEE J. Q. E., QE-20, 265 (1984), or R. C. Miller, D. A.
Kleiman, W. T. Tsang and A. C. Gossard, Phys. Rev. B24, 1134 (1981).

G. Y. Wu and T. C. McGill, private communication.

B. Lambert, B. Deveaud, A. Regreny and G. Talalaeff, Solid State Commun.
43, 443 (1982).

. C. Mailhoit, Y. C. Chang and T. C. McGill, Phys Rev. B26, 4449 (1982).
. G. Bastard, Phys. Rev. B24, 4714 (1981).

. Y. Guldner, G. Bastard and M. Voos, J. Appl. Phys. 57, 1403 (1985).

33.

34.

35.

36.

37.

. M.H. Weiler, in Semiconductors and Semimetals, edited by R. K. Willardson
and A. C. Beer, Vol. 16 (Academic Press, New York, 1981).

The value of 1608meV agrees quite well with the value obtained from pho-
toluminescence measurements on the substrate of sample 2 in this study.
The temperature dependence given for CdTe in Ref. 32 was not used since it
predicted the 300K band gap of CdTe to be 1425meV, while the majority of
other references give a value closer to 1500 meV. This illustrates the problems

associated with determining the band gaps of HgTe-CdTe superlattices.

Y. Guldner, G. Bastard, J. P. Vieren, M. Voos, J. P. Faurie and A. Million,
Phys. Rev. Lett. 51, 907 (1983).

J. N. Schulman and Y. C. Chang, Appl. Phys. Lett. 46, 571 (1985).
T. F. Kuech and J. O. McCaldin, J. Appl. Phys. 53, 3125 (1982).

M. B. Johnson, A. Zur, J. O. McCaldin and T. C. McGill, J. Vac. Sci.
Technol. B 3, 1260 (1985).

. G. Y. Wu and T. C. McGill, Appl. Phys. Lett. 47, 635 (1985).

Chapter 3

The 78-meV acceptor in GaAs

3.1 Introduction

3.1.1 Background

Recently, there has been considerable interest in using semi-insulating (SI)
GaAs as a substrate for device fabrication. Devices are usually fabricated on
GaAs by ion-implantation of dopant layers into a substrate. Ion-implantation
produces damage in the material which must be annealed out before defining
the device structures. Therefore it is necessary for the substrate to remain semi-
insulating during the anneal.

In the past, GaAs was grown by the horizontal Bridgman technique in quartz
crucibles. During growth, the material would incorporate silicon atoms from
the crucible. Silicon is usually a donor in GaAs, and the material was made
SI by compensating the silicon with chromium, which is a deep acceptor. Such
material is not thermally stable, however, since the chromium can diffuse out of
the substrate during annealing.

This problem led to the development of the liquid encapsulated Czochralski
(LEC) growth method, which produces thermally stable, SI material.! The LEC

method uses pyrolytic boron-nitride crucibles to eliminate the problem caused
by silicon incorporation during growth. The semi-insulating behavior is due to
a near mid-gap donor state which compensates the shallow acceptors (carbon
is the most common) usually present in GaAs. However, it had been observed
that if the melt stoichiometry became gallium rich during the growth process,

2 This material lacked the deep donor level, so

the material turned out p-type.
the conductivity was dominated by the shallow acceptors. Understanding the
process producing this change is an important step in learning how to control
the properties of LEC GaAs.

Studies of the p-type material have revealed that it posesses an acceptor
level at 78meV above the valence band, which is not present in the SI materi-
al.24 Temperature-dependent Hall effect and photoluminescence measurements
showed similar concentrations of the 78-meV acceptor and another acceptor at
200meV above the valence band in the p-type material, while neither acceptor
was present in appreciable quantities in the SI material.® While these mea-
surements do not prove that a relationship exsists between the two levels, it
has been suggested that this center is a double acceptor, with first and second
ionization energies of 78 and 200meV, respectively.6-7 Given the dependence
on stoichiometry, the anti-site Gay, (a gallium atom occupying an arsenic site),
or a complex involving this center and boron have been advanced as possible

explanations.

3.1.2 Results of this work

This study is the first observation of the s-like excited states of a double
acceptor in a semiconductor. The excited states of the 78-meV acceptor were
measured using two different techniques: selective excitation luminescence (SEL)

and electronic Raman scattering (ERS). Both of these techniques are sensitive to

the detection of s-like excited states of acceptors, and are useful for identifying
acceptors in bulk-grown semiconductors. The measurements made in this study
showed that the 1s!2s! excited state of the 78-meV acceptor in GaAs is split
into two levels separated by 4.0meV. An effective mass-like theory for a double
acceptor in a semiconductor was proposed to account for the observed splitting.
A comparison of the experiments with the double acceptor theory developed here
and theoretical predictions for a single acceptor led to the conclusion that the
78-meV acceptor in GaAs is due to the first ionization of a double acceptor. This
is the first direct evidence that the 78-meV level in GaAs is caused by a double

acceptor.

The theory presented here is the first to predict the s-like energy levels of
a double acceptor, including the splitting of the first s-like excited state. The
theory predicts the s-like states of a double acceptor to be highly degenerate. It
should then be possible to distinguish low-symmetry centers producing double
acceptors by their s-like excited state spectra. The experimental techniques em-
ployed in this study are useful for identifying double acceptors in semiconductors,
and should be applicable to many materials with band gaps in the near-infrared
and visible. The application of novel tunable optical sources such as F-center
lasers should extend the usefulness of these techniques to smaller-band gap ma-

terials like Si, and some IJI-V and II-VI compounds.

3.2 Experimental techniques

The p-like excited states of the 78-meV acceptor have been measured by in-
frared absorption.* However, p-like excited states do not give a strong indication
of the valency of an acceptor. The binding energies of the p-like excited states

relative to the valence band edge are nearly identical for simple single and dou-

ble acceptors, since p-like excited states have little probability of being near the
acceptor core. Thus, the hole binding energies for p-like excited states of single
and double acceptors are nearly identical, as is the case in germanium.® The
s-like excited states are expected to provide a better indication of the nature
of an acceptor, since they experience the central part of the potential. The ex-
perimental techniques used in this study were chosen for their ability to detect

transitions between s-like initial states and s-like final states.

3.2.1 Electronic Raman scattering

Electronic Raman scattering is useful for studying the excited states of ac-
ceptors in p-type material. It is therefore well suited to studying the 78-meV
acceptor, which occurs predominately in p-type material. In such material, the
acceptors are occupied by holes, and the donors are ionized (empty of electrons).

In the ERS experiment, a below band-gap photon in the crystal can decay
into a virtual electron-hole pair. Either the electron or hole can interact with a
hole bound to an acceptor. The bound hole can then be raised to an excited state
of the acceptor. The virtual electron-hole pair can then recombine, emitting a

photon. The energy of the scattered photon is given by
Aw scattered = hwtaser _— (Ea _ Ex:), (3.1)

where hwtaser is the incident photon energy. Ey, is the binding energy of the
acceptor ground state relative to the valence band, and E’,. is the binding energy
of the acceptor excited state relative to the valence band. Thus, the difference
between the incident and emitted photon energies is the ground state to excited
state energy difference of the acceptor. The donors are all ionized in p-type
material; therefore, the ERS experiment will not detect donor excited states.

The process of raising the bound hole to an excited state does not involve either

the emission or absorption of a photon; therefore, transitions between s-like intial

statesa and s-like final states are allowed.

3.2.2 Selective excitation luminescence

Selective excitation luminescence is useful for studying the excited states of
acceptors in compensated material.®1° Compensated material has roughly equal
numbers of donors and acceptors. In this case, the donors and acceptors are
ionized — i.e., the donor states are occupied by holes and the acceptor states
are empty of holes (Fig. 3.1a). It is easiest to understand this experiment by
considering the behavior of the holes.

The SEL experiment involves using below band-gap radiation to promote a
hole from a donor state to one of the excited states of an acceptor (Fig. 3.1b).
The hole can then relax to the acceptor ground state (Fig. 3.1c). The relaxation
process does not necessarily involve the emission of a photon. It could also be
done via phonon emission or other non-radiative decay. It is therefore possible
for the hole to relax from a 2S excited state to the 1S ground state. Finally, the
hole relaxes back to the donor ground state, emitting a photon (Fig. 3.1d).

The energy required to excite a hole from a donor to one of the excited states

of an acceptor is given by

hwrtaser = Egap — Ep ~~ Ea. + = + J* (R), (3.2)

where huraser is the energy of the incident photon, and Eyq, is the band-gap
energy. E’p is the donor binding energy relative to the conduction band edge and
Ea» is the binding energy of the acceptor excited state relative to the valence
band edge. FR is the distance between the donor and the acceptor, and ¢€ is the

static dielectric constant. The term

Figure 3.1: Schematic diagram of the processes involved in the SEL experiment.
The locations of the holes are indicated by the circles. Only one acceptor is

shown for clarity. The coulombic interaction between the donor and acceptor is

os: | ¢—6hl 6c OlUDOSHlUmUCUCSOCCSS

a.)

ground state —
excited states —_

ground state —
excited stotes

5 i = 2 = 2 = roa
¢ 6 68c—lCUC OUWC OElUmUC—-C(“‘SPFéC#SS

ground state Oy;

excited states __

ground stote —
excited states __

o¢ 6 686lCOC OB LT OCS
ne

not shown here.

a.)

The initial configuration of the compensated material. All the donors are

Conduction Band
Donors

Acceptor
Valence Band

Conduction Band
Donors

Acceptor
Valence Band

Conduction Band
Donors

Acceptor
Valence Band

Conduction Band
Donors

Acceptor
Valence Band

occupied with holes, and the acceptor is empty.

Here an incident photon of the proper energy is absorbed, promoting the

hole from a donor at distance R from the acceptor to an excited state of

the acceptor.

The hole relaxes to the acceptor ground state.

The hole falls back to the donor, emitting a photon.

represents the coulomb interaction between the donor and the acceptor, screened
by the static dielectric constant. The term J*(R) represents the non-coulombic
interaction caused by overlap of the donor ground state and the acceptor excited
state wave functions. When the hole relaxes back to the donor ground state, the
energy of the emitted photon is given by

hwemitted = Egap _ Ep _ Ea + = + J(R). (3.3)

Egap, Ep, € and R are the same as for eqn. 3.2. Ey, is the binding energy
of the acceptor ground state relative to the valence band edge. The term J(R)
represents the non-coulombic interaction between the donor and acceptor ground
state wave functions.

If we ignore the non-coulombic interactions — these terms should be small at
large pair separations — then the difference between the energies of the incident

and emitted photons is
Ahw = hwtaser _ hwemitted = Es — Eas. (3.4)

This is just the acceptor ground state to excited state energy difference. If a
number of different laser energies are used for the spectra, then the photons from
SEL processes will occur at fixed energy loss from the laser photon energy. This
provides a means of distinguishing SEL processes from background luminescence

processes, which would occur at fixed energy.

3.2.3. Applications

Both ERS and SEL have proved to be useful for identifying single acceptors
in bulk-grown semiconductors.14!?_ In many semiconductors, the binding energy
of an acceptor depends on the center producing it. For example, in GaAs the

hole binding energy for a carbon acceptor is 26.0meV, while that for germanium

is 40.4meV.!° This breakdown of the effective mass approximation is called the

central-cell shift.

In the effective mass approximation, the acceptor core is treated as a point
charge in a dielectric medium. All acceptors would therefore have the same
binding energy. In GaAs, the effective mass binding energy for an acceptor is
25.7meV. In an actual semiconductor, this approximation is only valid far from
the impurity center — the local potential near the center can deviate appreciably
from coulombic. The heavy hole effective mass in GaAs is roughly 0.5m,, giving
a Bohr radius for an acceptor of 13A in the effective mass approximation. The
lattice constant of GaAs is 5.65 A; therefore, a hole bound to an acceptor is easily
affected by the short range perturbations of the lattice. The short range terms
in the potential depend upon the actual center involved. In GaAs, for example,
both carbon and germanium form acceptors if they occupy an arsenic site. One
major difference between these two atoms is their size. A carbon atom is substan-
tially smaller than a germanium atom. The difference in size will cause different
local perturbations of the lattice, giving rise to different short range potentials.
The short range potentials also depend on the site occupied by an acceptor. Zinc
acceptors in GaAs on a gallium site have a binding energy of 30.7meV,!* which
is quite different from the binding energy for a germanium acceptor on an ar-
senic site even though the atomic sizes are similar. The difference between the
observed binding energy of an acceptor and that predicted by the effective mass

approximation is called the central cell shift.

It is possible to identify the center responsible for producing an acceptor
level by measuring the central cell shift. The higher-lying excited states of an
acceptor have a much lower probability of being at the impurity center, therefore
experiencing only the screened coulombic potential. The central cell shift can

be determined experimentally by measuring the binding energies of the excited

states relative to the acceptor ground state. Techniques such as the two-hole shift
can be used to identify acceptors in selectively-doped, high-purity material.
The values from ERS and SEL experiments in bulk-grown material can then be

compared to these results.

3.2.4 Experimental setup

The setup for both experiments is nearly identical (see Fig. 3.2). The 6471 A
line of a Coherent model CR-3000K krypton ion laser was used to optically pump
a Coherent model 590 tunable dye laser. The below band-gap light was produced
by either one of two carbo-cyanine dyes. DOTC was used for working near the
GaAs band gap in the SEL experiment, and HITC was used for working beyond
8500A in the ERS experiment. The dyes were dissolved in DMSO (dimethyl
sulfoxide), and ethylene glycol was added to increase the viscosity of the dye

stream.1}5

Tuning the dye laser was accomplished via a three-plate quartz bire-
fringent filter which had been optimized for use in the 8500 — 9500A range. The
tuning range for DOTC is given in the literature as 7500 — 8700A,!*° although
only 7800 ~ 8400 A was achieved in practice. The tuning range for HITC is given
as 8300 — 9100A, but only 8400 — 8900A was obtained. One factor limiting the
attainable tuning range was the selectivity of the birefringent filter, which had
transmission orders roughly 700A apart. This, coupled with the gain curves for
the dyes, limited the tuning ranges to about 600A. A different set of optimized
optics was used for each dye. A resolution of 1.2A or better was usually achieved
within the tuning ranges. The maximum tuned output power from the dye laser
was frequently only 100mW with an input pump power of 5 W.

The laser light was then directed onto the sample, which was mounted in a

Janis model DT-8 cryogenic dewar. The sample was immersed in liquid helium

pumped below the lambda point. There are a number of advantages to this: it

Double grating sample
Spectrometer filter luminescence dewar

i AG <2
i -
TW ab

(( wo |

—1

\ ‘] S PMT tunable
dye

laser

amplifier

- t

disc. :

Computer Ff MCA Kr*
laser

Figure 3.2: Schematic diagram of the setup for the SEL and ERS experiments.
The infrared laser dyes used were DOTC for SEL, and HITC for ERS.

achieves a very low temperature (typically 1.8K), the temperature is easily regu-
lated by controlling the pumping speed, and there is very little optical scattering

since the fluid does not boil.

The luminesence from the sample was directed into a Spex 1404 double grat-
ing spectrometer for spectral analysis. The gratings were blazed for 1.6um, and
used in second order for maximum effeciency at 8000A. A long-wavelength pass
filter was sometimes inserted before the spectrometer to eliminate stray room
light. The collection optics consisted of a short focal length lens matched to
the f-number of the dewar, and a second lens matched to the f-number of the

spectrometer.

An S-1 curve photomultiplier tube cooled to 77K was used as the detector.
The long-wavelength response limit of the photomultiplier was about 1.2 um,
eliminating the need to filter out first-order light entering the spectrometer. The
output from the photomultiplier was sent through a PAR model 1120 amplifier-
discriminator and to a multi-channel analyzer for photon counting. The data

from the multi-channel analyzer was then sent to a digital computer for analysis.

3.3 Sample descriptions

Two different liquid encapsulated Czochralski GaAs samples were used for
this study. Both samples were grown p-type from As-deficient melts. Temper-
ature-dependent Hall effect (TDH) measurements on sample 1 indicated 1.1 x
10°cm™* donor, 5.3 x 10!5em7? carbon and 8.1 x 10!5cm7~? 78-meV acceptor

concentrations, making it ideally suited for the ERS experiment.

The SEL experiment required compensated material, which was achieved by

neutron transmutation doping sample 2. In neutron transmutation doping the

sample is bombarded with thermal neutrons, causing the reactions:

n+Ga— Get+yt+e
n+As — Set +¥7t+e.

Selenium will always be a donor, and germanium will usually be a donor unless
it sits on an arsenic site (Ge,,). The neutron filtering during the transmutation
doping was not perfect, and it allowed some hot neutrons to hit the sample. This,

along with the emission, caused damage to the sample.

Neutron transmutation doping of sample 2 added 3.8 x 1015cm7~? Se and Ge
donors, plus additional damage related donors. The concentrations of the trans-
muted atoms were calculated from the neutron flux incident on the sample and
the respective neutron capture cross sections for gallium and arsenic. The sample
was given a partial anneal at 580°C for 1 hour. Temperature-dependent Hall
effect measurements on a non-irradiated sample adjacent to sample 2 and given
the same anneal gave concentrations of 2.6 x 10!5cm~? donors, 7.4 x 10!5cm™?
carbon, and 3.6 x 10'*cm7* 78-meV acceptors. The measurements on the non-
irradiated sample showed the slope for carbon below 4.2K, indicating that the
78-meV level was fully occupied at this temperature.!”? TDH measurements on
a sample transmutation doped and annealed in a manner identical to sample 2
showed an 0.07eV slope at 4.2K, indicating that the 78-meV acceptor was par-
tially compensated. There was a tradeoff here between annealing out the damage
and compensating the 78-meV level. The damage lowers the luminescence effi-
ciency, but annealing makes the material more p-type, which hinders the SEL

experiment.

3.4 Results

3.4.1 Photoluminescence

The presence of the 78-meV acceptor was identified in samples 1 and 2 by pho-
toluminescence and by temperature-dependent Hall effect measurements. Pho-
toluminescence curves for both samples are shown in Fig. 3.3. The lower curve
shows the photoluminescence from sample 1. At 1.9K, the band gap of GaAs is
1520meV.!® The near-gap region around 1510meV shows quite a bit of struc-
ture. Most of the lines in this region are believed to be excitonic in nature. The
line at 1512.4meV is due to an exciton bound to a carbon acceptor.!3 The line
at 1507meV is thought to be due to an exciton bound to the 78-meV acceptor,

3 This assign-

since it only occurs in material containing the 78-meV aceptor.
ment is in agreement with the rough estimate of the expected acceptor bound
excition binding energy, which can be made as follows. The ratio of the acceptor
bound exciton binding energy to the acceptor binding energy should be roughly
constant, since they both should experience the central cell shift, although the
exciton is affected to a lesser degree. The free exciton luminescence in GaAs
occurs at 1515.5meV at 4.2K.1% Thus, the binding energy of the carbon bound
exciton is 3.1meV. The acceptor binding energy for carbon in GaAs is 26meV,
thus the 78-meV acceptor is bound 3 times as strongly as carbon. A rough
estimate of the 78-meV exciton binding energy is then 9.3 meV, giving a lumines-

cence peak position of 1506.2 meV, which is close the the 1507 meV line observed

in the photoluminescence.

The large feature at 1490 meV is related to the presence of carbon. The main
peak is due to donor-acceptor luminescence, where a hole bound to a carbon ac-
ceptor recombines with an electron bound to a donor. The high energy shoulder is

band-acceptor luminescence, where a hole bound to a carbon acceptor recombines

PHOTOLUMINESCENCE SPECTRA

SAMPLE 2
T=1.9K
2.5 W/cm?

SAMPLE 1
T=1.9K
0.25 W/cm2

INTENSITY (ARBITRARY UNITS)

1380 1400 1420 1440 1460 1480 1500 1520
ENERGY (meV)

Figure 3.3: Photoluminescence spectra from samples 1 and 2. The photon energy
for the luminescence is plotted against the intensity of the luminescence. The
lower curve is from sample 1, and the upper is from sample 2. The two curves
are not plotted on the same vertical scale. The excitation source was the 6471 A

line of a Kr* ion laser.

with an electron in the conduction band. The low energy shoulder has not been
conclusively identified, although there is evidence suggesting it is due to donor-
acceptor luminescence, where the acceptor is zinc. The photoluminescence peak
energy for this shoulder is near that expected for zinc — 1488.8meV.'? There
is also evidence in SEL and ERS data from sample 1 indicating the presence of

zinc.

The weak feature at about 1457meV is an LO phonon replica of the carbon
donor-acceptor luminescence. In GaAs, the LO phonon energy is 36.5meV;19
thus, the LO phonon replica of a luminescence feature occurs at 36.5meV to-
ward lower energy than the main feature. The peak near 1441meV is the
donor-acceptor luminescence involving the 78-meV acceptor. The weak line
near 1405meV is the LO phonon replica of the 78-meV donor-acceptor lumi-
nescence. The relative intensities of the 1 LO phonon replicas of the donor-
acceptor luminescence for the 78-meV acceptor and carbon, and their respective
no-phonon lines indicate that the electronic transition of the 78-meV acceptor is

more strongly coupled to the LO phonons.?

The upper curve in Fig. 3.3 shows the photoluminescence from sample 2.
There is no evidence of the near-gap luminescence seen in sample 1, even though
the optical pump power was 10 times that used for sample 1. This reflects the
damage done to the crystal by neutron transmutation doping, which lowers the
luminescence efficiency by providing paths for non-radiative transitions. The
peak near 1494meV is due to carbon donor-acceptor luminescence. The weak
peak at 1480meV may be donor-acceptor luminescence involving Ge acceptors —
the expected peak photoluminescence peak energy is 1497 meV'* — which could
be produced by transmuting gallium atoms, with the resulting germanium atom
occupying an arsenic site. This could be caused either by the transmutation of Ga

anti-site defects (a suspected source of the 78-meV acceptor), or by displacement

of the Ge atoms during transmutation doping. The peak at 1442meV is due to
donor-acceptor luminescence from the 78-meV acceptor. The peak near 1405meV
is the LO phonon replica of the 78-meV donor-acceptor luminescence.

The peak position of the 78-meV donor-acceptor luminescence occurs at
slightly different energies in the two samples. This is indicative of the prob-
lems associated with identifying acceptors by photoluminescence in bulk-grown
GaAs. A measurement of the excited state binding energies provides a more

precise identification.

3.4.2 SEL and ERS

The SEL results for sample 2 for three different laser excitation energies
are shown in Fig. 3.4. Two transitions are evident in the SEL data, one at
62.5meV (line A) and one at 66.5meV (line B). The broad background in the
spectra is due to donor-acceptor luminescence from the 78-meV acceptor. The
ERS results for sample 1 for two different exciting laser energies are shown in
Fig 3.5. Three peaks are visible in the ERS data: line A at 62.9meV, line B at
66.9meV and line C at 72.9meV. Lines A and B appear shifted about 0.4meV
toward greater energy loss in the ERS data than in the SEL data. This may be
expected since any non-coulombic interaction between the donor and acceptor
wave functions in the SEL experiment would tend to shift the measured energies
from the actual binding energies. The energies obtained in ERS; however, are a
direct measurement of the actual binding energies. Line C was not seen in the
SEL experiment.

The two lines, A and B, occur only in material containing the 78-meV accep-
tor. Figure 3.6 compares ERS spectra from sample 3, which does not contain the
78-meV acceptor, with that from sample 1. There is no in indication of lines

A or B in sample 3, which does not contain the 78-meV acceptor. This suggests

SELECTIVE EXCITATION LUMINESCENCE SPECTRA

SAMPLE 2
T=1.9K

1500.9 meV

1502.9 meV

INTENSITY (ARBITRARY UNITS)

1504.9 meV

pe ee a a eb ee we em ee eee he wee =

80 76 72 68 64 60 56 52
ENERGY LOSS (meV)

Figure 3.4: Selective excitation luminescence spectra from sample 2. Intensity
of the emitted light is plotted against the difference between the emitted photon
energy and that of the laser. Three different laser energies were used to distin-
guish between SEL features and background processes. The laser photon energy

for each spectrum is at the right.

INTENSITY (ARBITRARY UNITS)

ELECTRONIC RAMAN SCATTERING SPECTRA.
Cc B A

SAMPLE 1
T=1.9K

1399.8 meV

1419.5 meV

L ‘ 4 l 1 1 4 1 I 1 1 | . 1

76 72 68 64 #60 56. 52
ENERGY LOSS (meV)

Figure 3.5: Electronic Raman scattering spectra from sample 1. Intensity of

the emitted light is plotted against the difference between the emitted photon

energy and that of the laser. Two different laser energies were used to distinguish

between ERS features and background processes. The laser photon energy for

each spectrum is at the right.

INTENSITY (ARBITRARY UNITS)

ope Og

ELECTRONIC RAMAN SCATTERING SPECTRA

SAMPLE 3
T=1.9K
LASER = 1435.1 meV |

mm a er rr re - - - - - 0

SAMPLE 1
T=1.9K
LASER = 1428.0 meV

=see eee ewe mn ew ew ee ee

x20

a mm a een D>

70 60 50 40
ENERGY LOSS (meV)

Figure 3.6: Electronic Raman scattering spectra from samples 1 and 3. Intensity

of the emitted light is plotted against the difference between the emitted photon

energy and that of the laser.

Narrow lines at 36.5 and 33.6meV are due to

Raman scattering from LO (If) and TO (IL) phonons, respectively. The laser

photon energy for each spectrum is given in the center.

that they both are due to the presence of the 78-meV acceptor. Although they
lie in the energy region for two-phonon Raman scattering, we can be sure that
they are electronic in nature since the excitation intensity in the SEL experiment
is too low to observe even single-phonon Raman scattering. There is some evi-
dence for a line at 73meV in sample 3, which may be related to line C in sample
1. Since we do not observe line C in the SEL experiment, and it appears to be
present in sample 3, it may be due to Raman scattering from 2 LO (I) phonons
(room temperature Raman scattering measurments give a value of 72.9meV for
this transition?®), and not related to the 78-meV acceptor. However, given the
intensity of line C relative to the other two lines, it probably would not be visible
above the background in the SEL experiment. Thus, the relationship between

this line and the 78-meV acceptor is not firmly established by these experiments.

95
3.5 Interpretation

The results of both the SEL and ERS experiments indicate that lines A and
B are both due to transitions from s-like excited states. The relative intensities
of the two lines are nearly identical, while those due to s-like states are usually
a factor of 5-10 times more intense than those of p-like excited states when

11,12 A single acceptor would not produce such

observed by these techniques.
an excited state spectrum. However, the 1s'2s! (using atomic single particle
notation) excited state of a double acceptor will be split into two s-like states by

the exchange interaction between the two holes. This splitting might give rise to

such a spectrum.

3.5.1 Double acceptor effective mass theory

A simple model of a double acceptor in a semiconductor can be obtained by
analogy to the effective mass approximation. The holes are considered to be
j = 3/2 particles, with mass equal to the heavy hole effective mass, m,;. The
electric fields will be screened by the static dielectric constant. The acceptor
can then be treated like a helium atom in a uniform dielectric medium. This
will result in a two-fold splitting of the 1s!2s! state to first order: a 10-fold
degenerate (75,75) state, which is spatially anti-symmetric and has symmetric
spins, and a 6-fold degenerate (15,°S) state, which is spatially symmetric and

has anti-symmetric spins. The energy of the splitting is given by

AE = 2(1 : 2s;2 : ls |—_—--_.
( e|R2 — Rj |

1:18;2: 2s), (3.6)

where ¢ is the static dielectric constant. This is just the exchange splitting due
to the coulomb interaction between the two holes. For the purposes of this
approximation, single acceptor hydrogenic wave functions can be used as a first-

order approximation to this integral. In this case, the value for the splitting can

be obtained by scaling the observed exchange splitting in helium. This is done
by multiplying the energy for helium by the ratio of the Bohr radii divided by

the static dielectric constant:
Mh
A Evsemiconductor = met eH (3.7)

The observed splitting in He is 0.79eV,”! which scales to 2.6meV in GaAs. The
data shows a splitting of 4.0meV, which is in good agreement with the data.
Including the effects of the crystal field and the spin-orbit interaction with the
valence band will not change the splitting to first order, since they will affect the
wave functions used, not the potential.

The energy level diagram for a double acceptor in GaAs is plotted in Fig. 3.7.
The state assignments are given on the right, and the theoretical estimates of the
hole binding energies relative to the valence band are given in the center column.
The experimentally observed values for the binding energies are given in the left
column. These values were obtained by subtracting the observed transition en-
ergies from the ground state energy (78meV). Both of the experimental values
for the 1s? and 1s* (the first and second ionization energies, respectively) states
were determined by photoluminescence. While there has been some discussion as
to whether the second ionization energy was 200 or 230meV,®” recent time re-
solved photoluminescence measurements favor 200meV.??_ The values predicted
by the simple effective mass model agree quite well with the observed values for
the ionization energies, as well as with the 1s'2s! states. This strongly indicates
that the 78-meV acceptor is an effective mass double acceptor.

It is interesting to compare the calculated binding energies for the 1s'3s!
states to the observed value for line C. Line C occurs at 72.9meV, while the
average energy of the 1s'3s! states is 72.5meV. The theory predicts a splitting
of only 0.6meV between the two states, which would not be resolved in the ERS

experiment. Thus it is possible that line C corresponds to the combination of

EXPT. | THEORY STATE
(meV) (meV)
200°) 181 1s"
78%) 1s?
15.1%) 15.5 1s'2s1 (33,73)
11.1 12.9 1s'25!(15,9S)
1 1/367
(d.)§ 5.8 1s'3s' (°S,’S)
(5.1) 5.2 1s'3s!('s,9s)
0 0 MOEA EES Ls VALEN CE
BAND

Figure 3.7: Energy level diagram for a double acceptor in GaAs (not drawn to
scale). The state assignments are given in the column on the right. The values
obtained from the double acceptor effective mass calculation are given under
the theory column. The energies are binding energies measured relative to the

valence band. The column on the left gives the experimental values.

a.) See Ref. 22.
b.) From photoluminescence measurements.
c.) Calculated by subtracting observed energies from the 1s? energy (78 meV).

d.) This assignment is tentative — see the text.

both 1s'3s! states. However, given the proximity of line C to the 2 LO (I)
phonon Raman scattering line, and its absence in the SEL data, it is likely that

line C is not related to the 78-meV acceptor.

3.5.2 Single acceptor model

The experimental results should also be compared with those expected for
a single acceptor with a 78-meV ground state binding energy. In this case, the
78-meV line observed in photoluminescence would indicate a central cell shift
of 52.3meV (the effective mass binding energy for a single acceptor in GaAs
is 25.7meV).2*_ The theory for shallow acceptors in semiconductors has been
worked out in great detail by Baldereschi and Lipari.25?4 The effect of the
central cell shift must be included to correct the energy estimate for the 2S3/2
binding energy. The probability of finding the hole at the impurity site has been
calculated in Ref. 24. Using the value » = 0.767 given there, the 2S3/2 state
would experience a central-cell shift of 0.133 times that of the 153 /2 State. Here
that yields a shift of 7.0meV for the 253/2 state binding energy. Reference 24
gives a value of 70.4meV for the 153/2 — 25372 transition in GaAs, using 78meV
for the 153/2 binding energy. After adding the central cell correction, a value of
63.4meV is obtained. The 153/. — 2P3/2 transition energy would be 66.6 meV.
This would lead to the following assignments: line A could be the 153 /2 > 2837/2
transition, and line B could be the 1583/2 —> 2P3/2 transition of such a single
acceptor. Line C could then correspond to the 153;. > 2P5/2(I'7) transition,
which should occur at 72.2meV. While the observed energies seem to agree
fairly well with this model, other evidence strongly favors the double acceptor

model.

3.5.3 Comparison of theories

A comparison of the possible assignments is given in Table 3.1. As mentioned
previously, both SEL and ERS are sensitive to transitions between s-like states.
Given the nearly identical intensities for lines A and B, they must both be due
to s-like states. If line C were due to the 1583/2 — 2Ps5 p2(Tz ) transition, then
the 1583/2 — 2P5/2(T'g) transition should be visible at 70.95meV, where it was
observed in the IR absorption experiment,* since both SEL and ERS show p-
like states with roughly the same intensity. Finally, IR absorption measurements
heavily favor p-like states over s-like states, since the latter are forbidden to
first order. Thus one would not expect to see s-like states in IR absorption. If
line B were actually p-like, it should have been observed in the IR absorption
experiment of Ref. 4. This leads to the conclusion that lines A and B are both

s-like, and therefore that the 78-meV acceptor is a double acceptor.

3.6 Conclusion

This study represents the first observation of the s-like excited states of a
double acceptor in a semiconductor. The observation of these states was made
possible by the experimental techniques of selective excitation luminescence and
electronic Raman scattering. These techniques should prove useful for identify-
ing other double acceptors in semiconductors. This work also shows that the
1s'2s1 state of a double acceptor in a semiconductor is split by the exchange in-
teraction between the two holes, and that the magnitude of the splitting is large
enough to be observed by current techniques. This splitting provides a means of
distinguishing single and double acceptors. The key to the identification is the
determination of the symmetries of the excited states. SEL and ERS are both

good techniques for observing s-like states, while infrared absorption can be used

100

Table 3.1: Observed line positions and interpretations.

A B Ce AE?
SEL 62.5meV 66.5 meV see 4.0meV
ERS 62.9meV 66.9meV 72.9meV 4.0meV
Double 62.5meV 65.1meV 72.5meV4 2.6meV
Acceptor® 1s?-+1s!2s! 1s? + 1s!2s! 1s? > 1813s!
(35,78) (7S,°S)
Single 63.4meVS 66.6 meV 72.7meV 3.2meV

Acceptor® 18372 _ 2837/2 18372 _ 2P3/2 13/2 —_ 2P5/2(T'7)

a. The evidence suggests that line C is actually due to 2 LO (I) phonon Ra-
man scattering and not related to the 78-meV acceptor. The assignments

are included for reference to the text.
b. The energy separation of lines A and B.
c. Based on the effective mass like calculation discussed in the text.

d. This was obtained assuming that the experiment could not resolve the
splitting between the two 1s'3s! states. The value here is the average

energy of the two states.
e. Based on the work done in Refs. 23 and 24.

f. This estimate was obtained by including the observed central cell shift of

78-meV.

101

to study p-like excited states. This study proved that 78-meV acceptor in GaAs
is produced by the first ionization of an effective mass double acceptor. The line
observed at 62.9meV is due to the 1s? — 1s!2s! (35,75) transition, and the line
at 66.9meV line is due to the 1s? — 1s'2s! (1S,°S) transition. While the line
at 72.9meV may be due to the ls? — 1813s! transition, there is also evidence
supporting the assignment of this line to 2 LO ([) Raman scattering. It was
difficult to make a positive identification of this line due to its low intensity and
proximity to the multiple phonon Raman scattering processes in GaAs.

While it was shown that the center responsible for the 78-meV acceptor in
GaAs is a double acceptor, the results of this study do not identify the source.
Given the similarity of the observed levels to those predicted by the double
acceptor effective mass theory, it may be reasonable to assume that the center is
not very complex. If the center had very low symmetry, it probably would lead
to further splitting of the highly degenerate 1s'2s1 states. Thus, the anti-site

defect Gaz, is still a possible source for this acceptor.

Note added in proof

Recent work by Bishop et al.?? has shown a that there is a connection between
the 200-meV level observed in GaAs and the 78-meV level. They conclude that
the 200-meV level is the second ionization of the 78-meV double acceptor. They
believe that the 230-meV acceptor is due to a different center, and is a single
acceptor. However, they were not able to identify the center responsible for the
78-meV acceptor.

Another interesting experiment has been performed on Ga,_,In,As by A. T.
Hunter.” An acceptor at ~ 78-meV was observed in low (a few percent) In
concentration Ga,_,In,As alloys. An ERS study of this acceptor showed two

lines corresponding to lines A and B of this study. Alloy broadening of the lines

102

was observed, indicating that they are indeed related to the 78-meV acceptor.
(This is expected since there will be a distribution of possible sites for the 78-meV
acceptor in the alloy, causing slight variations in the ground state binding energy.)
A line similar to line C was also observed, but it did not exhibit broadening. This

favors the identification of line C as 2 LO (I) phonon Raman scattering.

103

References

. An excellent overview of GaAs growth techniques may be found in: H.

Winston, Solid State Technology 26, 145 (1983).

. D. E. Holmes, R. T. Chen, K. R. Elliott and C. G. Kirkpatrick, Appl. Phys.

Lett. 40, 46 (1982).

. P. W. Yu and D. C. Reynolds, J. Appl. Phys. 53, 1263 (1982).

. K.R. Elliott, D. E. Holmes, R. T. Chen and C. G. Kirkpatrick, Appl. Phys.

Lett. 40, 898 (1982).

. A. T. Hunter, R. Baron, J. P. Baukus, H. Kimura, M. H. Young, H. Winston

and O. J. Marsh, in Semt-tnsulating III-V Materials, Evain 1982, edited by
S. Makram-Ebeid and B. Tuck (Shiva, Cheshire, England, 1982), p. 396.

P. W. Yu, W. C. Mitchel, M. C. Meir, S.S. Liand W. L. Wang, Appl. Phys.
Lett. 41, 532 (1982).

K. R. Elliott, Appl. Phys. Lett. 42, 274 (1983).

. R. A. Chapman, W. G. Hutchinson and T. L. Estle, Phys. Rev. Lett. 17,

132 (1966).

H. Tews, H. Venghaus and P. J. Dean, Phys. Rev. B19, 5178 (1979).

14.

15.

16.

17.

18.

19.

20.

21.

104

. P. J. Dean, D. J. Robbins and S. G. Bishop, J. Phys. C12, 5567 (1979).
. A. T. Hunter, Ph. D. Thesis, California Institute of Technology, 1981.
. A. T. Hunter and T. C. McGill, Appl. Phys. Lett. 40, 169 (1982).

. D. J. Ashen, P. J. Dean, D. T. J. Hurle, J. B. Mullin and A. M. White, J.
Phys. Chem. Solids 36, 1041 (1975).

W. Schairer and. T. O. Yep, Solid State Commun. 9, 421 (1971).

The dye recipes used were:

DOTC - 250ml DMSO, 750ml ethylene glycol, 500mg DOTC

HITC - 250ml DMSO, 750ml ethylene glycol, 500mg HITC

Both of these dyes can now be replaced by Oxazine 750 using the following
recipe:

625mg Oxazine 750, 53ml propylene glycol, 947ml ethylene glycol.

See the Exciton Chemical Co. (Overlook Station, Dayton OH 45431) laser

dye catalog and the references contained therein.

A. T. Hunter, M. H. Young and R. Baron, Hughes Research Laboratories
(unpublished).

J. I. Pankove, Optical Processes in Semiconductors (Dover, New York, 1971),

Appendix II, p. 412.
J. 5S. Blamkemore, J. Appl. Phys. 53, R123 (1982).

T. Sekine, K. Uchinokura, and E. Matsuura, J. Phys. Chem. Solids 38,
1091 (1977).

A. R. Strigonov and N. 8. Sventitskii, Tables of Spectral Lines of Neutral and
Ionized Atoms (IFI/Plenum, New York, 1968), Vol. I., p. 79.

105

22. 8. G. Bishop, B. V. Shanabrook, and W. J. Moore, J. Appl. Phys. 56, 1785
(1984).

23. A. Baldereschi and N. O. Lipari, Phys. Rev. B9, 1525 (1974).
24. A. Baldereschi and N. O. Lipari, Phys. Rev. B8, 2697 (1973).

25. A. T. Hunter, Hughes Research Laboratories, private communication.

106

Appendix A

Photoluminescence using an

FTIR

A.1 Introduction

The majority of photoluminescence measurements on semiconductors are per-
formed using dispersive spectrometers to analyze the wavelength dependence of
the signal. Dispersive spectrometers are best suited to operation in the visible
and near infrared regions. There is a wide variety of sensitive detectors available
in this wavelength range, allowing good signal-to-noise ratios to be obtained in
fairly short accumulation times. However, dispersive spectrometers are not well
suited to studies of broad wavelength ranges in the mid- to far-infrared. The de-
tectors available for the range beyond 10 um are not very sensitive (see Table 2.1
on page 21), necessitating long integration times. Grating efficiency also poses a
problem. While gratings can approach 90% efficiency near the blaze wavelength,
the efficiency can drop dramatically toward shorter and longer wavelengths. The
usual remedy for this problem is to have a number of different gratings for each

wavelength range to be studied. This problem is exacerbated in optical studies

107

of semiconductors by the inverse dependence of the wavelength of light on en-
ergy. Thus at small energies, a small shift in absolute energy results in a large
shift in wavelength in the infrared. The study of even a small energy range re-
quires the ability to cover a large wavelength range. Finally, there is a tradeoff
between throughput and resolution in a dispersive spectrometer. This makes
high-resolution studies over large wavelength ranges difficult with a dispersive
spectrometer. A Fourier transform infrared spectrophotometer (FTIR) is an in-
strument which can overcome many of these problems. However, a standard
FTIR is not well suited for studying weak externally generated signals in the
10m range, due to the presence of a large background of 300K blackbody ra-
diation. This appendix is a description of a novel technique for performing pho-
toluminescence with an FTIR, which overcomes the problems associated with

background blackbody radiation.

A.2 FTIR principles

There is a wealth of information in the literature describing the theory and
operation of FTIRs,! and the details will not be discussed here. An FTIR
basically operates as a Michelson interferometer (see Fig. A.1). Either light
from an internal source, or external light entering through a viewport, is directed
onto a beam splitter. One beam is directed onto a fixed mirror, which returns
it to the beamsplitter and onto the detector. The second beam strikes a moving
mirror, and is directed back to the beamsplitter and onto the detector as well.
The intensity of the light falling on the detector is modulated at a frequency
dependent upon both the wavelength of the incident light and the speed of the
moving mirror. The signal out of the detector represents the Fourier transform

of the input signal. The signal is then amplified, filtered, and sent to an analog-

108

ZZZZA MOVING Mirror
1 ot
vv

: <— <—
yy —_ mh | luminescence
Geen <— 7 —_ Hv
fixed mirror ¥
beamsplitter
14
vv

detector

focussing mirror

Figure A.1: Schematic diagram of a Fourier transform infrared spectrophotome-

ter.

109

to-digital converter and on to a vector processor. The vector processor takes the
Fourier transform of the signal, and sends it on to a computer for analysis.

The advantage of using an FTIR instrument for studying signals in the in-
frared comes from its ability to collect light over a broad spectral region in a short
time period, and allowing nearly all the light from the source to be incident on
the detector throughout the scan. As discussed in Chapter 2, the signal-to-noise
ratios of most solid state detectors used in the infrared increase as the signal
strength increases. Since dispersive instruments trade off the signal strength
falling on the detector for resolution, they suffer from lower signal to noise ra-
tios than FTIR instruments. These advantages are usually used for performing
infrared transmission measurements. For such measurements, an internal broad-
band source is directed onto the beam splitter. The sample is placed just before
the detector, transmitting the modulated light as a function of wavelength. A
high-pass filter on the detector output removes any d.c. components of the sig-
nal. This eliminates any unmodulated light from appearing as a signal after the
Fourier transform is performed. The output will then show the transmission of
the sample as a function of wavelength, but no signal from the blackbody radia-
tion of the spectrometer housing or the sample. While this makes the FTIR well
suited for transmission experiments, experiments involving external sources are

more complicated.

A.3 Photoluminescence measurements with an
FTIR

Blackbody background

When analyzing photoluminescence, or any an external signal, the presence

of 300K blackbody radiation poses a real problem, since the FTIR will modulate

110

all the light incident on the beamsplitter. When using the external viewport, the
beamsplitter will “see” the entire optical path, including the external viewport.
While the optics may be transparent in the infrared, they are also sources of
infrared radiation since they are at room temperature. This large thermal signal
can easily mask the weak signals generated by photoluminescence measurements.
While cooling the optics might limit this problem, it would be very impractical.
In regions where the background blackbody signal is fairly weak, <3 um in the
near infrared and % 30m in the far infrared, simple digital subtraction of the
background signal may be sufficient to observe external signals. Digital sub-
traction involves subtracting a background scan (where the signal of interest is
absent) from the signal scan. This subtraction is performed after digitization,
but before the Fourier transform is performed. This technique is less than ideal
— the background signal may change during a long scan; it is limited by the res-
olution of the digitization, and a large change in the signal strength falling on the
detector between the scans might change the detector response characteristics.
These problems confine this technique to regions where the signal is fairly strong
relative to the background. The solution arrived at for the HgTe-CdTe study in
Chapter 2 of this thesis allows the detection of very weak signals obscured by a

large background, and is much more sensitive than digital subtraction.

A.3.1 Double-modulation technique

This section describes a double-modulation technique used to convert the de-
sired emission signal from d.c. to a relatively high frequency a.c. signal. When
using a dispersive spectrometer in the presence of a large background, the usual
method for the detection of weak signals is to modulate the signal of interest
to an a.c. frequency. This allows for synchronous detection of the emission sig-

nal, and lowers the 1/f-noise added to the signal by the detector. There are a

111

number of considerations when adapting this technique to FTIR spectroscopy.
The FTIR itself modulates the external signal to a set of a.c. frequencies, where
the frequency depends on the wavelength of the light. This waveform must be
preserved if the vector processor is to produce the correct Fourier transform of
the signal. This requires modulating the signal before it enters the spectrome-
ter, and demodulating the signal just before the analog-to-digital converter. The
basic setup is shown in Fig. A.2. The system must be designed to block all the
fundamental FTIR frequencies, since these will correspond to the unwanted un-
modulated external signal. Therefore, the external modulation frequency should
be greater than the highest frequency produced by the FTIR to allow for fil-
tering. It is best to use as high a modulation frequency as the system will
allow, since this will ensure differentiation between the doubly-modulated signal
and the singly-modulated background signal, and narrow the bandwidth of the

doubly-modulated signal.

As shown in Fig. A.2, the photoluminescence is modulated by a pulsed laser
source. It is also possible to use a mechanical beam chopper to modulate the
luminescence, but pulsed lasers can operate at much higher frequencies. The
luminescence is collected by a set of infrared optics. The collection lens is chosen
to match the f-number of the output window on the dewar, and the transfer lens
is chosen to match the input f-number of the FTIR. A Ge filter may be placed
before the FTIR external viewport to prevent specularly scattered laser light
from entering the spectrometer. It is possible that the intense laser light could
change the properties of the detector; therefore, it should be blocked. A solid
state infrared detector is used (Ge:Zn was used in Chapter 2). The output from
the detector is amplified, and sent to a filter. The filter is designed to remove the
strong, singly-modulated background signal which could overload the following

amplifier stages or the mixer. This filter consists of a high-pass or band-pass

112

sample
FTIR external aes “a “+
viewport ae l (> aN
+ Tan rome me can <-—-—- am
; / \
Ge filter /
Detector |
| |
| |
} |
* Tt T
bias | |
preamp | |
supply =
| \
ref. [output]
filter osc. pulser
[outputs] [trigger] ‘oder
diode
input [ref. in] output vector
| | a |
amp mixer amp processor

Figure A.2: Schematic diagram of the FTIR experimental setup.

113

filter, set to pass the bandwidth of the doubly-modulated signal. The filtered
signal is then amplified more, since it now consists of only the weak doubly-
modulated signal, and sent to a mixer for demodulation. The demodulated
signal is amplified so that the maximum signal strength corresponds to the full
scale input of the analog-to-digital converter. The signal now appears as if it had
only the spectrophotometer modulation and can then be fed into the standard
signal processing electronics of the FTIR. The end result is the elimination of the
background blackbody signal. Figure A.3 shows two photoluminescence spectra
of a Hg;_,Cd,Te alloy sample, one taken using the double-modulation technique,
and one taken using only the instrument’s standard electronics. The lower trace
(double-modulation setup) shows a broad doublet structure, corresponding to
luminescence from near the band gap of this sample. The upper trace (non-
modulated) shows a wide continuum characteristic of 300K blackbody radiation
dominating the signal from the sample. The peak intensity of the upper trace
is about 200 times that of the lower trace. The photoluminescence signal could
not be observed, even by digitally subtracting a background scan from the upper
scan. This figure illustrates that a large improvement in the signal-to-noise ratio

can be obtained even for very small signals by using this technique.

A.3.2 Actual experimental setup

The actual experimental setup was based on a PAR model 124A lock-in am-
plifier, which contained the mixer, an input preamplifier, an input filter and an
output amplifier. This system worked quite well, and the full-scale output of
the lock-in matched the full-scale input of the analog-to-digital converter. A
schematic diagram of the actual system setup is shown in Fig. 2.3 on page 30.
The actual settings used for the experiment are described in Section 2.3.4. The

input high-pass filter frequency and Q were chosen to provide the maximum fil-

114

Hg,7,Cd,.,4e Lumi

nescence

Unmodulated
T = 12K

Modulated
T = 12K

INTENSITY (ARBITRARY UNITS)

sale n te i doe J. ry J... .

0 100 200 300 400 #4500
ENERGY (meV)

Figure A.3: A comparison of the two FTIR measurement techniques. Both
scans are measurements of the photoluminesence spectrum of a Hg;_,Cd,Te al-
loy sample. The conditions of each measurement are identical, except for the
measurement scheme. The upper scan is a conventional FTIR scan, while the
lower scan was obtained using the double-modulation scheme described in the
text. The vertical signal on the lower trace is roughly 200 times that of the upper

trace.

115

tering of the singly-modulated signal without introducing phase shifts into the
signal. Unfortunately, this necessitated using a fairly low filter frequency, and a
low Q, which provided a relatively shallow rolloff. A filter with a sharper turnoff
would probably have improved the signal-to-noise ratio. The lock-in amplifier
mixer had an output time constant, which was set to the minimum value of
about 600 ys, causing attenuation of freqencies above 265Hz. This output low-
pass filtering turned out to be useful. While the time constant limited the data
collection speed (the FTIR modulation frequencies are determined by the speed
of the moving mirror), using the direct mixer output and higher mirror speeds
resulted in a lower signal-to-noise ratio. This effect was not expected, but may

be due to high-frequency noise in the electronics.

A.4 Possible pitfalls

There are a few possible problems associated with the FTIR double modula-
tion technique. Given the high modulation frequencies used (40kHz was used for
the study in Chapter 2), obtaining a system response curve is fairly difficult due
to the shortage of broadband infrared sources which can be modulated to such
high frequencies. The attenuation due to the electronic filtering can be measured
or estimated, but the optical response is difficult to determine. The response
of components such as the beam splitter and optics can be measured without
resorting to double modulation, but the response of solid-state infrared detectors
is a function of the modulation frequency of the incident radiation. This effect
can not be measured in an easy manner using this setup. Therefore, one must
take care analyzing the output, since there may be some system response effects.
The photoluminescence spectra of Hg,_,Cd,Te alloys taken with the doubly-

modulated FTIR setup was compared with luminescence of similar samples taken

116
with a dispersive spectrometer, and no large system response effects were noticed.

This system is very sensitive to any signal modulated at the reference fre-
quency. When performing photoluminescence experiments in the infrared, the
sample is frequently placed on a cold finger in a vacuum, since many infrared op-
tical materials can not withstand the thermal expansion and temperature shock
present in immersion dewars. Vacuum is a very poor heat conductor, therefore it
is possible to heat up a sample very rapidly with only modest laser powers, if the
sample is not mounted well. With poor mounting, samples were observed to heat
up from 12 K to over room temperature in less than 2 minutes for 200mW average
incident power. However, even if the sample is mounted fairly well, it is possible
to modulate the temperature of the sample slightly at the laser frequency. This
temperature change could show up either as a change in the blackbody radiation,
or a change in the luminescence of the sample. It is also possible to modulate the
reflectivity of a sample (see Section 2.3.5). In this case, the sample will reflect
room temperature blackbody radiation into the FTIR, modulated at the laser
frequency. This effect was observed, although the actual process responsible for
the change in reflectivity was not identified; it was probably due to changes in

the carrier concentration.

In conclusion, a novel double-modulation technique has been developed for
performing photoluminescence using a Fourier transform infrared spectropho-
tometer in the presence of a large background signal. This technique produces a
dramatic improvement in the signal-to-noise ratio for weak signals over standard
FTIR experiments. The technique allows the advantages of FTIR spectrometers
over dispersive spectrometers for emission experiments to be used in previously
inaccessible wavelength regions. The resulting system is much more sensitive
than conventional dispersive spectrometers over a wide wavelength range, so

sensitive that photoluminescence from Hg;_,Cd,Te alloys with Cd concentra-

117

tions of ~ 30% was observed at room temperature. A standard system based
on a dispersive spectrometer was limited to sample temperatures below 60K.
This technique should find wide applications studying infrared emissions from a

variety of sources.

118

References

1. There are many good references on the principles of Fourier transform in-
frared spectroscopy. The following is a partial list of sources, but it should

provide a good starting point.

e W. Demtréder, in Laser Spectroscopy (Springer-Verlag, Berlin, 1982)
pp. 139-190.

e R. C. Milward, in Far-infrared Properties of Solids (Plenum, New York,
1970) pp. 1-34.

e R. G. Bell, Introductory Fourier Transform Spectroscopy (Academic
Press, New York, 1972).

e J. Chamberlain, The Principles of Interferometric Spectroscopy (John
Wiley and Sons, New York, 1979).