Optical processes in cadmium sulfide thin films


Material Information

Optical processes in cadmium sulfide thin films
Physical Description:
viii, 240 leaves : ill. ; 28 cm.
Clausen, Edward M., 1958-
Publication Date:


Subjects / Keywords:
Thin films   ( lcsh )
Optical data processing   ( lcsh )
Semiconductors -- Optical properties   ( lcsh )
Cadmium sulfide   ( lcsh )
Materials Science and Engineering thesis Ph. D
Dissertations, Academic -- Materials Science and Engineering -- UF
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1987.
Includes bibliographical references (leaves 235-239).
Additional Physical Form:
Also available online.
Statement of Responsibility:
by Edward M. Clausen.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 022282147
oclc - 18476908
System ID:

Full Text







Copyright 1987


Edward M. Clausen, Jr.


Not enough can be said for the people who have contributed to the

start and particular the completion of this dissertation. It would

have been very difficult for me to get through this project without

their help.

I would like to express my deepest appreciation and gratitude to

Dr. Joseph H. Simmons, my academic adviser, who supplied much

encouragement, direction, assistance and the opportunity to work on

this project. His insight and ability to help me solve problems proved

to be an extremely valuable asset, although his confidence in my

abilities to "get the job done" was invaluable. I would also like to

express my appreciation to the members of my committee, Dr. Paul

Holloway, Dr. Robert Dehoff, Dr. Stan Bates, Dr. Tim Anderson and Dr.

Ramakant Srivastava for their suggestions, comments and guidance. This

is certainly one of the best combinations of abilities and expertise

for a committee and I thank the members for their help and enthusiasm.

My closest friends Guy Latorre, Richard Robinson and B.G. Potter

deserve special thanks for their personal support and reassurance. My

roommate and friend, Steve Wallace deserves the most thanks and credit

for having to endure the more trying times of this project. I would

especially like to thank my mother and father for their support

throughout all of my endeavors, and my closest and dearest friend Laura

Harmsen for her inspiration and emotional support.

Finally, I would like to express my graditude to the Air Force

Office of Scientific Research (AFOSR 84-0395) for funding this research




ACKNOWLEDGMENTS . . . . . . . . . . . iii

ABSTRACT . . . . . . . . . . . . . vii


I INTRODUCTION . . . . . . . . .. 1

Optical Signal Processing . . . . . .. 1
Nonlinear Optical Materials . . . . . . 2
Thin Films . . . . . . . . . 5


Nonlinear Optical Susceptibility . . . . 8
Optical Bistability . . . . . . . .. 19


Bulk Properties of Cadmium Sulfide . . . .. .28
Photoluminescence of CdS . . . . . .. .32
Resonant Raman Scattering in CdS . . . .. 37
Nonlinear Susceptibility . . . . . .. .38
CdS Thin Films . . . . . . . .. 39
Vacuum Deposition . . . . . . .. 39
RF Sputtering . . . . . . . .. 45
Semiconductor Doped Filter Glasses . . . .. .50

IV EXPERIMENTAL METHOD . . . . . . . .. .52

Vacuum Deposition System . . . . . .. .52
RF Magnetron Sputtering . . . . . . .. .60
Thin Film Deposition . . . . . . .. 63
Substrate materials . . . . . .. 64
Substrate temperatures . . . . .. .66
Deposition rates . . . . . . .. 68
Co-Sputter Alternating Deposition (COSAD) . 68
Post Deposition Treatments . . . . . .. .70
Thin Film Characterization . . . . . .. .71
Microstructure . . . . . . . .. 71
Chemical Analysis . . . . . . .. 73

Optical Measurements . . . . . . .. 76
Index of Refraction . . . . . . .. 77
UV-VIS Absorption . . . . . . .. 86
Photoluminescence and Raman . . . . .. .88

V RESULTS AND DISCUSSION . . . . . . .. .96

Thin Film Physical Properties . . . . .. .96
Microstructure Characterization . . . .. 96
Transmission electron microscopy . . .. .96
Scanning electron microscopy . . . .. .109
X-ray diffraction experiments . . . .. .123
Compositional Analysis . . . . . .. .132
Stoichiometry determination of target material 132
Analytical techniques . . . . . .. .134
Co-Sputter Alternating Deposition . . .. .149
Microstructure . . . . . . .. 149
Composition . . . . . . . .. 163
Thin Film Optical Properties . . . . .. .165
Absorption Spectra . . . . . . .. 167
Room temperature UV-VIS absorption ..... .168
Low temperature UV-VIS absorption . . .. .174
Index of refraction . . . . . .. 185
Photoluminescence and Raman Spectroscopy . 190
Photoluminescence spectra . . . . .. .193
Raman spectroscopy . . . . . .. .207
COSAD Optical Properties . . . . . .. .211

VI CONCLUSIONS . . . . . . . . . .. 218

VII FUTURE WORK . . . . . . . . .. 224



B ELLIPSOMETRY . . . . . . . . .. 230

BIBLIOGRAPHY . . . . . . . . . . . .. 235

BIOGRAPHICAL SKETCH . . . . . . . . . .. 240

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy



Edward M. Clausen, Jr.

December, 1987

Chairman: Joseph H. Simmons
Major Department: Materials Science and Engineering

The semiconductor, cadmium sulfide, has received much attention

for its optical, electronic and piezoelectric properties. Recently,

several investigators have demonstrated that its Wannier excitons hold

great promise for applications in nonlinear optics. However, the

results showed, as expected by the current theoretical thinking, that

the nonlinear optical behavior and the exciton and band gap energy

configurations are dependent upon the microstructure of the samples.

Since most applications in computer logic, or communications require

waveguide geometries, the object of this dissertation has been the

study of thin films.

The investigation presented here, therefore concerns the study of

the effect of microstructure and preparation conditions on the optical

properties of CdS thin films. High optical quality films were produced

by RF magnetron sputtering with a variety of microstructures and

crystallographic characteristics controlled by the deposition process


and subsequent heat treatments. An in-depth study of the thin film

microstructure revealed the relationship between crystallographic

defects and band gap defects which lead to band tailing absorption.

Optical characterization related exciton photoluminescence, absorption,

resonant Raman scattering and band edge photoluminescence and

absorption to the microstructure of films. For the first time,

absorption bands and photoluminescence associated with exciton states

were observed in a polycrystalline thin film.

A supplementary study investigated composite films, consisting of

CdS crystals in a glass matrix, formed by a novel co-sputter deposition

process, in which alternating layers of CdS and a borosilicate glass

were deposited to form a thin film. A wide variation in the structure

of the deposited film was obtained by changing the amount of deposited

CdS and by post deposition heat treatments. Low concentrations produced

CdS microcrystallites as small as 70 A, small enough for quantum

confinement processes to affect the band energy. For all films

produced, a shift in the band absorption edge to higher energies was

observed; however, it was determined that this shift must be partially

associated with a chemical shift. Possible formation of CdO could

raise the band gap energy, although this alone could not produce the

shifts observed in the films with the smallest particles, and therefore

a confinement process might exist.



Optical Signal Processing

With the very intense development of optical communications in

recent years, the need for integrated optical systems for the

processing of optical signals has greatly increased. To take full

advantage of the speed and information bandwidth available at optical

frequencies, an all optical system is desired. All of today's systems

operate by the high bandwidth transmission of optical signals in

optical fibers, followed by conversion to lower bandwidth electrical

signals before any processing, such as amplification and multiplexing,

can take place. The limiting drift velocity of charge carriers in a

semiconductor and the capacitive coupling between adjacent elements

present the fundamental limit for processing speed in these systems.

Also, the serial nature by which the electronic data must be

manipulated presents another speed barrier. An alternate approach

would utilize optical bistability and optical switching demonstrated in

certain materials for all-optical modulation, detection and

multiplexing. Optical switching is achievable with materials which

display a third order nonlinear susceptibility. A third order

susceptibility leads to an intensity dependent index of refraction or

absorption. This nonlinear refractive index can exhibit onset and

decay on a very fast time scale, which makes it an attractive


phenomenon for optical signal processing. With the proper material,

all optical operations could conceivably be carried out in a monolithic

thin film, which would act as a guiding medium with both active and

passive regions. An optical communications fiber could be coupled

directly to this thin film, thereby fully utilizing the speed and

bandwidth of the optical signal.

A complementary application of an integrated optics technology

which utilizes optical switching would be in the area of high speed

logic operations for the next generation of computers. Logic gate

operation has been demonstrated with several materials which exhibit a

nonlinear optical susceptibility and optical bistability. A few of

these gates have been shown to switch on a subnanosecond time frame,

which is competitive with present day high speed electronic systems.

The primary advantage of the optical gate is the possibility of

parallel processing on the fundamental logic cell level, which adds

tremendous speed advantages over electronic systems. Most optical

computer designs today are based on integrated optics, in which the

active regions consist of arrays of bistable devices arranged with a

high spatial density. A high density of gates is possible because

optical gates are not subject to capacitive coupling, thus making

possible massive parallel processing without the connection problems

encountered in today's electronic systems.

Nonlinear Optical Materials

A large number of architectures have been proposed for both

computer systems and multiplexing circuits based on optical switching.

While this field has greatly advanced, there is, however, a very great

need for suitable materials and systems. Many materials exhibit a

third order nonlinear susceptibility,1 but very few have the

characteristics necessary for an integrated optics system. A few of

the semiconductors which have been investigated include InSb, GaAs,

GaP, CdS, and CdTe. Most of these semiconductor materials which show

optical switching have been investigated in bulk form. Only the

multiple quantum well (MQW) structures made from gallium arsenide are

the notable exception and are made in thin film form.2 This material

is perhaps the most promising today for use in optical signal

processing systems, primarily because its nonlinearity occurs at the

same wavelength as the semiconductor lasers which are currently

available. For an integrated optical system this is an essential

consideration because most of the processes which produce the nonlinear

susceptibility require laser light of energy near the band gap of the


Equally important considerations, however, include the value of

the nonlinear coefficient, the switching speed, and the absorption

coefficient, since these values determine how much power is required to

switch the device, and how fast recovery will be. Low power operation

is essential for any large scale integration, although the figure of

merit (FOM) most often quoted is given by

FOM = n2 (1.1)
t a

where n2 is the nonlinear index, T is the switch-off time, and a is the

absorption coefficient. Obviously the larger the FOM the more


attractive a system is. MQW gallium arsenide has an additional

advantage of room temperature operation, but suffers from a slow

switch-off time and very large absorption. There are other materials

which have been shown to exhibit a much larger nonlinear effect than

MQW gallium arsenide. A material which has been demonstrated to exhibit

one of the largest nonlinear coefficients is cadmium sulfide (CdS).3

The large coefficient was obtained by saturating a bound exciton level

in the band gap.4 The mechanism which leads to the large exciton

saturation effect in CdS is central to this dissertation and will be

described in a subsequent section. However, only a very few groups

have looked at CdS, and no one has either investigated the bound

exciton saturation mechanism in thin films or explored the possible

applications in integrated optics.

The most important consideration for any practical nonlinear

application is the temperature at which the nonlinear process

predominates. To the present day the largest nonlinear coefficients are

only measured at very low temperatures. For example, the large

coefficient obtained by saturating the bound exciton level in CdS

occurs at 2 K. Since the binding energy of the bound exciton only

corresponds to a few millielectron volts, the state is thermally

annihilated at higher temperatures, and the large nonlinear effect


The electronic structure of the band gap, however, may be altered

to permit access of exciton levels at higher temperatures by

controlling the physical size of the material. The phenomenon by which

this occurs is known as quantum confinement. When the crystal size of


a material is on the order of the radius of the exciton state, new

boundary conditions can distort the translational motion of the

exciton, its binding energy and the individual orbits of the electron

and hole. At this size the electron and hole interactions with the

crystal surface begin to govern the electronic properties of the

semiconductor. Quantum confinement also occurs in the MQW structures

of GaAs; however the exciton is only confined in one direction. A much

stronger effect occurs when the confinement is in two or three


Thin Films

A study of CdS thin films was chosen for a number of reasons: 1)

very few semiconductors which display a nonlinear susceptibility have

been tested in a thin film form, despite the potentially dominating

role in applications. 2) Cadmium sulfide displays one of the largest

excitonic saturation effects, and thus the influence of film structure

due to deposition conditions or subsequent treatments could be

investigated. 3) Quantum confinement effects and their interaction with

the perturbation of the exciton absorption process are of great

interest to the future of nonlinear optical developments and

applications. Cadmium sulfide promises to offer a means of studying

both the effects and their interplay with the development of the thin

film structure. 4) Since the energies of the exciton states correspond

to wavelengths in the visible part of the spectrum, the experimental

optics for the measurement of these states is simplified, and many


investigative tools become available for following the underlying


The study of any thin film for this application should start with

examining the properties of the bulk material which contribute to the

specific origin for the nonlinear effect. In the case of CdS this

means looking at the electronic states of the material which lead to

the presence of excitons. These states have been thoroughly examined

for more than 30 years and they are probably the best understood in

this material. Exciton states have been shown to occur in thin

epitaxial films, but no one has investigated the presence of these

states in polycrystalline thin films, nor have the effects of

preparation conditions on the excitonic transitions of a thin films

been investigated. The objective of this study therefore is to

determine how exciton states would occur in polycrystalline thin films

of the material, how they are affected by structure and formation

conditions, and how size variations and grain boundary structures might

affect their energy level structure.

A supplementary part of this study will investigate the

possibility of producing thin film structures consisting of a glass

matrix with small isolated crystals whose sizes matched those needed to

develop quantum confinement effects. The investigation will use

spectroscopic techniques to determine if quantum confinement effects

can be induced on the exciton states in the material. The objective is

to produce a thin film of semiconductor doped filter glass.

The process of quantum confinement has been and still is under

investigation in bulk semiconductor doped filter glasses, which have


received considerable attention lately for nonlinear optics

application. These are crown base glasses which contain one or two

percent of CdS or mixtures of CdS and CdSe. It has been postulated

that the semiconductor crystals exist in the glass matrix as finely

dispersed microcrystallites, which are small enough to permit quantum

confinement effects to occur. This effect is still not well understood

and has not been clearly demonstrated. Many authors have observed

energy shifts that may be due to compositional effects rather than

microstructure size. However, it appears theoretically that with

sufficient confinement, the exciton state will be accessible at room



Nonlinear Optical Susceptibility

The nonlinear refractive index which is observed in certain

semiconductor materials is a result of an electronic polarization which

is induced by the interaction with a monochromatic radiation field.

The susceptibilities of the polarization determine the values of the

experimentally measured optical properties. To understand the

relationships between the susceptibilities and the optical properties

we must first consider the electro-magnetic field, E, which is given by

E(t) = E(w)e-iwt + E*(w)eiwt (2.1)

The resulting polarization has frequency components at all multiples of

+/-w, but considering only those that occur at w

P(W) = X(1)E(w) + x(2)[E]E(w) + X(3)[E]2E(w) + . (2.2)

The first term in the series, X(1) is the linear susceptibility and by

using first order perturbation theory,5 the linear dispersion of the

refractive index below the band gap can be calculated. The higher


order terms in the series are the nonlinear susceptibilities, and

although their magnitudes are much smaller than the first term, under

very high field intensities a number of different effects can be

observed. For non-centrosymmetric crystals (i.e. crystals without an

inversion center), under appropriate conditions, the second order term

is manifested as two different effects. The first is a quadratic

variation of the refractive index with applied voltage, which is known

as the Kerr effect. The second is the generation of a second harmonic

radiation field. Second harmonic generation is a very useful effect

for doubling the frequency of a laser beam. Both of these effects have

many important applications; however, the primary interest of this

study is the effects which lead to the third order term X(3). One

consequence of the third order term is the generation of a third

harmonic radiation field. For the current topic of this study,

however, the manifestation of this term as a nonlinear refractive index

is of primary interest. The relationship between the third order

susceptibility and the nonlinear refractive index can be understood by

first considering basic dielectric theory for the displacement of a

charge in response to an applied electric field:

D(w) = E(w) + 4iP(w) = c E(w) (2.3)

where D(w) is the frequency dependence of the displacement and e is the

complex dielectric constant. The variable, e can be defined as

C (n + ica ) 2

where ca/2w is the extinction coefficient. By combining these two

equations a relation between the polarization and the index of

refraction can be written.

(n + ica )2= 1 + 4m P(w)/E(w)


The nonlinear susceptibility can be defined by expansion of the

refractive index in terms of the intensity I, of the radiation inside

the sample:5

n = nI + n21 + n312 +


Next we assume that the extinction coefficient is very small compared

to n. Then by using equations 2.5 and 2.6 and expanding the terms,

2 2 (21)2 (i) X(3) 2
n = n + 2n 1 n21 + (n21) = 1 + 4 X( + 47X [E]2. (2.7)

Finally we assume n2 is much smaller than nI and by comparing

coefficients of [E]2 and I


n2 1 + 4n X (2.8)


n 2 2Tr (2.9)

There are essentially four electronic processes which can produce

a reactive nonlinear susceptibility in semiconductors. These are known

as 1) the induced free-carrier plasma, 2) the dynamic Burstein-Moss

effect, 3) the direct saturation of interband excitations, and 4) the

saturation of exciton absorption.5 Of these 1 and 2 are the most

commonly studied, 3 occurs in most direct band gap semiconductors, and

4 is the most promising for high speed operations. All four processes

can occur in some materials. The most dominating process which is

observed depends somewhat on the material, but mostly on the particular

experimental setup and measurement temperature. As shown in Table 1,

different processes result in widely different values of the reported

nonlinear index and saturation intensity.

The four processes listed above basically describe how the

transitions between different levels in a semiconductor and the

saturation of those levels result in the observed nonlinear

susceptibility. Depending on the band structure of a material and the

particular wavelength of light used for the analysis, one of the four

processes will dominate. The one process that can occur in nearly

every semiconductor, however, is the induced free-carrier plasma.

Assuming that photo induced transitions produce electron-hole pairs,

the number of these free carriers will be intensity dependent.


Listing of

Nonlinear Optical Values for Certain
Semiconductor Materials







IS (W/cm2)

n2 (cm2/W)

toff (sec)


& a a a
















Key for Electronic Processes:


Free Exciton Saturation
Dynamic Burstein-Moss
Direct Interband Saturation
Bound Exciton Saturation
Free Carrier Plasma







Depending on the recombination time TR, and the absorption coefficient

a, the steady state density of free carriers will be given by

a I R

h w


Once these carriers are formed, they are allowed to diffuse and form an

electron-hole plasma. The plasma will respond to an applied electric

field and the resulting undamped oscillations will produce a

polarization which can be related to the index of refraction through

the dielectric constant6

2 = ( -
n = ( c -

4mNe )
m w


By use of equation 1.10, the nonlinear refractive index for a plasma

can be written as

n2(P) =e a R

n m W



where n is the refractive index of the material without the plasma and
m is the effective mass. The transient susceptibility is determined

by the time it takes for the free carriers to build up;7 however, this

can be a very short time. This process will also occur at room

temperature. The disadvantage of this process for nonlinear optical

applications is that the effect is very small because there is no

coupling between the states, and therefore no resonance effects take

place. The process also can suffer from a very slow recovery time

because the recombination time in certain semiconductors is on the

order of usec.

Another process that is based on an intensity dependent free

carrier concentration is the dynamic Burstien-Moss or blocking effect.

Again a steady state density of charge carriers is given by equation

2.10; however, the origin of the absorption is not considered

explicitly. At low temperatures the carriers are assumed to thermalize

by a phonon scattering process so that they fill the bottom of the

conduction band. The top of the valance band becomes empty and the

shift of the effective band gap to higher energy becomes intensity

dependent. In association with this shift there must be an intensity

dependent contribution to the refractive index.5 This is because the

filled conduction band effectively blocks absorptive transitions, and

the blocked transitions no longer contribute to polarization and

refraction. Also, some type of unspecified coupling takes place

between excited states, so that the effect is enhanced somewhat. The

nonlinear refractive index for this process is given by

n2(BM) = 2 r ( e P )2 N (2.13)
3 n h w h (w G W) I

where N is the free carrier density given by equation 2.10, P is a

momentum matrix element, and wG is the effective band gap. For

nonlinear optical applications the advantage of this process is that

the occupied states are closer to resonance so that a larger n2

results. There also is the possibility that this process can occur at

energies below the band gap, as saturation of excited carriers can be

produced not by direct optical absorption, but by scattering from other

excited states.6 The one disadvantage of the process that is similar

to the induced plasma process is that the interband relaxations

required for decay of the state can be very slow. In some materials a

faster decay process can occur by scattering to intraband transitions,

which effectively relaxes the system by transferring the population to

other states.6

Another mechanism for the nonlinearity observed in some materials

is by a direct interband saturation process. This process assumes that

the band structure in a direct gap semiconductor can be modeled as a

set of uncoupled two level systems which are homogeneously broadened by

a dephasing time T2. Homogeneous broadening means that the individual

transitions are indistinguishable. The T2-Lorentzian broadening

results in absorption below the band gap and excitation into the T2-

broadened "band-tail" is assumed to be responsible for the nonlinear

refraction.6 At some high level of intensity the two level system

should become saturated, and associated with this saturation is a


nonlinear contribution to the refractive index.5 The nonlinear index

of refraction for this process is expressed by

n2(S) = 1 ( eP )4 T 2 X
h3 h w T2 15 n2 c

( 2 m ) G W ) 3/2 (2.14)

where T is the relaxation time, and c is the speed of light. The

advantage of this process for nonlinear applications is that the

effective nonlinear refractive index is inversely proportional to the

band gap energy, so for small band gap materials this process leads to

a very large effect. As indicated above, the broadening results in an

effect which occurs below the band gap energy, so absorption losses are

reduced. The disadvantage of utilizing this process is the same as for

the other two processes; in some materials there is no fast mechanism

for decay of the excited state.

The final process that will lead to an electronic nonlinear

refractive index in certain semiconductors is the saturation of bound

exciton levels. These particular defect states are characterized by a

very narrow transition linewidth, which is comparable to atomic

resonances. A bound exciton is an associated electron-hole pair that

is bound to an impurity site. The oscillator strength of the bound

exciton, which is related to the polarizability, is extremely large in

comparison to oscillator strengths of molecules.' In addition, there

is a very high density of oscillators, which contributes to a very

large nonlinear effect. The absorption transition of the bound exciton


is modeled as a saturable two level system, inhomogeneously broadened

by a T2 dephasing time. The transition linewidth is inhomogeneously

broadened because excitons bound at different locations see different

environments.4 An inhomogeneously broadened system is described as a

distribution of groups or classes of transitions, and within each class

the transitions are assumed to be identical homogeneouslyy broadened).

The saturation of the inhomogeneously broadened system does not depend

on the homogeneous lineshape function, but rather on the linewidth of

these "homogeneous packets".8 This means that the saturation intensity

is inversely dependent upon the dephasing time T2 as shown by

2 2
I= 2 n hvAv (2.15)

s X2

where Av = (i T2)-~ (2.16)

and 0 is the ratio of the radiative lifetime to the spontaneous decay

time, which is usually taken as equal to one. In semiconductor systems

the dephasing time is on the order of 0.1 psec,4 which means that very

low saturation intensities are required to saturate bound exciton

transitions. The importance of the saturation intensity will be

described in a following section; however, a small value indicates that

the nonlinear refractive index is very large, as the two are inversely

proportional (see equation 2.25). As shown in Table 1, in CdS a

saturation intensity as small as 26 W/cm2 has been measured, which

corresponds to n2 value of 1.3 X 10-2 cm2/W.10 This is the largest n2


value ever reported. Additional advantages of bound exciton saturation

are that the state decays by a radiative transition and the lifetime is

on the order of 500 psec. This would make for a very fast, low power

switch. Also, since the excitons are bound to defect sites, there are

no carrier diffusion problems, which in the other three processes tend

to wash out the effect. The one primary disadvantage of utilizing this

process is that the strongest exciton resonance occurs at 2 K. As the

temperature is increased, the transition broadens, thereby requiring a

larger saturation intensity and hence a smaller n2 is observed. In

addition, since the exciton binding energy is only a few millielectron

volts, the state is thermally annihilated at higher temperatures.

As previously described the process of quantum confinement could

be used to access exciton levels at higher temperatures if the physical

size of the material could be made small enough. Multiple quantum well

structures produce quantum confinement in one direction because the

structure is made up of alternating layers in which the layers act as

infinite potential wells and the layer thickness is smaller than the

exciton radius. Although the exciton is not confined in the other two

directions, the effect is strong enough that nonlinearity can be

observed at room temperature. The effect would be larger if

confinement was made in the other two directions.


Optical Bistability

The primary means for measuring nonlinearity in materials is

through an internal feedback device known as a Fabry-Perot

interferometer. The saturation intensity of such a system is the

intensity at which the gain of the feedback saturates. This is the

point at which optical bistability occurs, and from the saturation

intensity and the interferometer parameters the nonlinear index of

refraction can be determined.

The Fabry-Perot consists of a cavity formed by two plane parallel,

highly reflecting mirrors. The transmission of monchromatic light

through the device is determined by the optical path length of the

cavity. If the cavity is not tuned to the wavelength of the light,

then a transmission of 1 % results. When the optical path length is

exactly equal to an integer number of wavelengths, then a resonance

effect occurs and the output intensity from the device reaches nearly

100 % of the input intensity. A diagram of this process is shown in

Figure 1. The optical path length is determined by the physical length

d, times the refractive index n of the material within the cavity. The

condition for resonance therefore is given by

2 n d = m X (2.17)

where m is the integer order number.

Beam = 100

Beam = 90

Beam =1

Figure 1 Schematic diagram of a Fabry-Perot interferometer showing
how interference of the forward and reverse beams changes
the output intensity.9

Beam = 10

Beam = 9


When the cavity does not satisfy the above requirement, then a

linear relationship will exist between the incident and transmitted

intensity. If a material with a nonlinear refractive index is placed

in the cavity, then a positive feedback loop will occur where the

refractive index and the light intensity become mutually reinforcing.

As the incident intensity is increased, a change in refractive index

occurs which brings the device closer to resonance, which further

increases the intensity inside the cavity, which further changes the

index, etc. This continues until a saturation intensity is reached, at

which point a phase shift to resonance occurs within the cavity and the

transmitted intensity suddenly increases.

The ratio of the incident intensity to the transmitted intensity

as a function of the phase shift 6 is given by the Airy function A(8)

t = A(Q) (2.18)


A(e) = 1 (2.19)
1 + F sin2(6/2)

and F is related to the reflectivities of the mirrors R by

F = (2.20)
I -R )

\I \ I '--r2 = 0.04


I -0.8
-4- r .
0 I I--F=200
-21r 0 2 4x 6


Figure 2 Descriptions of the operation of a Fabry-Perot. a) plot of
the Airy function as a function of the phase shift 6,
reflectivities r, and the finesse F; b) transmitted image
from a high finesse Fabry-Perot.53


Figure 2a displays how the relative transmitted intensity and the

sharpness of the transition is related to the reflectivities and

finesse of the cavity. As the reflectivity of the mirrors is

decreased, the finesse is decreased and the transition at the critical

phase shift broadens. When the reflectivities and the finesse of the

cavity are large, then the transitions are sharp. For these

conditions, the transmitted image from a diffuse source through the

Fabry-Perot will appear as a series of sharp concentric rings, as shown

in Figure 2b.

For a nonlinear Fabry-Perot, however, the Airy function must be

slightly modified to account for the nonlinear index by

A(e)NL = 1 (2.21)
1 + F sin2( t I eff 6)

where i is a constant describing the nonlinear refraction,10 and Ieff

is the effective mean intensity within the cavity. The total Fabry-

Perot fractional transmission can be written asI0

T 1 R ) ( I -A) A(e) (2.22)
(1 -R ( 1 A ) 2


where the intensity absorption per pass A is given by

A = 1 ead (2.23)

and a is the absorption coefficient. A second equation can be written

which is parametric in Ieff for the Fabry-Perot transmission:I0

T = a d ( 1 R ) ( 1 A ) eff (2.24)
A (1 -R( 1 A ) I

The condition for optical bistability can be determined by

simultaneously solving equations 2.22 and 2.24. A graphical solution

of these equations which shows the criterion for optical bistability is

shown in Figure 3.

The critical intensity Ic for the onset of bistability is given by

the intensity Io which gives more than one intersection with the line

and curve. This is the intensity at which the saturation of the

feedback occurs. Once this saturation is achieved, it is found that if

the input intensity is reduced, the output intensity does not drop

until a finite decrease in the input has occurred. In other words, a

hysteresis effect is observed. The switching between the two intensity

levels can be considered as a change in logic state. The nonlinear

Fabry-Perot interferometer therefore can be used as an optical logic

a) oc



8c 8 >8
1 eff 8

Figure 3 Graphical solution to the Airy function showing the
critical phase shift required for the onset of


gate. The great interest for logic gate applications is that the

switching between the two levels can occur on a subnanosecond time

period in some materials.

The saturation or critical intensity Ic is found to be inversely

proportional to the nonlinear refractive index and is given by

I = 1 (2.25)


3 n
2 = (2.26)
X a

and p is a figure of merit value for the cavity, relating the

reflectivities of the two mirrors and the attenuation of light as

passes through the cavity. Equation 2.25 indicates that for fixed

cavity conditions, a small saturation intensity corresponds to a large

nonlinear refractive index. The total index of refraction is given by

nt = n1 + An (2.27)

An = n21c.




The value of nt is what actually determines the phase shift but as

indicated by equation 1.1, the figure of merit for nonlinear optical

applications also includes the switching times and the absorption



Bulk Properties of Cadmium sulfide

Nearly all of the early work on cadmium sulfide was carried out on

single crystal platelets which were made by a chemical vapor phase

growth process. The natural crystal structure of cadmium sulfide is

hexagonal wurtzite, although single crystals of the cubic zincblende

structure have been fabricated. Within either of the two crystal

structures it is possible to have regions which are made up of the

alternate crystal structure. The transition from a hexagonal to a

cubic lattice or visa versa can occur through a well known twinning

mechanism,12 in which the twinned region is bound by stacking faults.

The twinned regions can be manifested during deformation of the crystal

or under particular growth conditions, although it is difficult to

differentiate these two sources when crystals are grown from the vapor

phase. In either case, the two crystal structures do not have a center

of symmetry or inversion, which leads to the unique properties of

noncentrosymmetric crystals such as piezoelectricity, pyroelectricity,

and third order optical susceptibility.

Another important physical property of CdS is the stoichiometry of

the crystal. Very little work has been done on the defect chemistry of

CdS; however, the work done on other II-VI semiconductors such as ZnS

and CdTe indicates that the range of nonstoichiometry at room

temperature is very small, e.g. 0.01 to 0.1 %.13 Early work by Collins


which involved sulfur atmosphere heat treatments and electron

bombardments, showed that sulfur vacancies were the predominant native

defect and that they acted as the recombination center responsible for

the green edge emission associated with CdS luminescence.14 Other

studies which investigated impurity doping effects are described in a

following section on photoluminescence.

A detailed knowledge of the band gap structure of CdS has come

from the extensive study of exciton states. Cadmium sulfide is found

to be a direct gap semiconductor with a band gap equal to 2.59 eV at 0

K. The wurtzite lattice of the material is described by a p-like

valance band consisting of two gamma-7 states and one gamma-9 state,

and a s-like conduction band made up of one gamma-9 state. A diagram

of the band extrema is shown in Figure 4. The three states in the

valance band are also known as the A, B, and C free exciton states.

These intrinsic exciton states are modeled as Wannier excitons; i.e.

the electron and hole behave like a hydrogen atom. The orbital

movements of the electron and hole are determined by their effective

masses within the band extreme. As shown in Figure 5, the solution for

the wave function of this model results in a series of discrete

parabolic bands below Eg which merge into a continuum at higher

energies.15 Because of the unique band structure of this material

there are a large number of possible exciton states. Any of the

exciton energy levels (i.e. n=1,2,3, etc.) can be associated with the

three primary states (A,B and C) in the valance band. In addition, any

one of these free excitons can be associated with an impurity center,






Figure 4 Band gap structure for wurtzite crystals near k=0.14






Figure 5 Energy diagram for Wannier excitons as a function of
exciton momentum K, showing "hydrogenic" states which merge
into a continuum at energies greater than E .15

forming a bound exciton complex, which will have a lower energy than

the corresponding free exciton.

Of all the II-VI semiconductor materials, the exciton states in

bulk CdS have been studied the most and are perhaps the best

understood. Reflection, absorption, and luminescence studies dating

back to the mid-fifties have investigated the exciton states in this

material. Excitation of the states can be accomplished by either

electron bombardment or by photon absorption. When the emission is due

to the latter process it is known as photoluminescence and the results

reported for CdS are detailed below.

Photoluminescence of CdS

When CdS is excited by photons of energy greater than the band gap

the characteristic luminescence which results form the decay of excited

states is shown to consist of two primary emission bands. The first,

known as the "green-edge emission" is due to the edge emission of

various states in the band gap, i.e. shallow donor-acceptor

recombinations. Studies of the edge emission of CdS were made by

Kroger as early as 1940.16 At a slightly higher energy, a band known

as the "blue-edge emission" occurs and is due to emission from free and

bound exciton complexes.17 A typical low resolution spectrum

displaying these two bands is shown Figure 6, and a high resolution

spectrum of a portion of the blue band is shown in Figure 7. The

intensity of the bands is dependent upon the polarization of the

incident light with respect to the c-axis of the crystal.









0.20 -

0.10 -

0 _

/ -. E11C -
05 4
00 5200 5300 5400 5500


Figure 6 Low resolution photoluminescence spectrum of CdS single
crystal showing "green-edge" emission due to band edge
recombinations and "blue-edge" emission due to excitons.
Figure shows emissions are dependent upon the polarization
of the excitation source.18




Figure 7 High resolution photoluminescence spectrum of CdS single
crystal at 4.2 K, showing bound exciton peaks for
excitation perpendicular to the c-axis. Peaks labeled P
are observed with parallel excitation.19


The exciton states were first extensively studied and

characterized by Thomas and Hopfield.19 They have shown that within

the higher energy band a number of sharp luminescence lines occur which

correspond to transitions of both free excitons (A, B, and C) and

excitons bound to neutral donors or acceptors. A designation of I1 was

given to excitons which are bound to neutral acceptors and 12 was given

to the excitons bound to neutral donors. These two peaks are labeled

in Figure 7. The distinctions between the various transitions were

made by using the Zeeman effect. When a strong magnetic field is

imposed on the sample, many of the luminescent lines split due to the

spin moments of the ground and excited states. From the group theory

of bound complexes,19 they were able to assign the transitions to the

different defect states.

A later study by Henry, Faulkner, and Nassau showed for the first

time donor-acceptor pair lines in the photoluminescence spectra of

CdS.20 These pair lines are narrowly spaced transitions which were

observed in the green-edge emission band and correspond to closely

spaced donor-acceptor pair-recombination bands. Again the confirmation

of these lines was made by Zeeman experiments. The significance of the

study is that for the first time direct spectroscopic evidence for the

existence of these states was made. This is important point because in

the study by Thomas and Hopfield it was assumed that these states must

exist based on the Zeeman experiments, but they had no direct evidence.

Henry, Nassua, and Shiever studied the impurity doping of CdS and

showed that Na and Li are the only shallow acceptors that can act as

substitutional impurities.21 These shallow acceptors give rise to the


I, bound exciton that was described by Hopfield and Thomas. Usually

two II lines are observed, and by varying the doping level these

authors proved the lines to only be due to Na and Li. High purity

crystals grown in clean reactor tubes were found to only exhibit the

I(Li) line. When Na was added a considerable broadening of the I,

line occurred, but as successive runs were made in the same tube, these

authors showed the Na line could be resolved. Doping with K, Rb, or

Cs, only resulted in a sharp I(Li) line, and P was found to give a

complex shallow acceptor. The identity of the shallow donor level

responsible for the 12 bound exciton could not be determined; however,

the authors showed by donor-acceptor pair line splitting that the donor

was not a native double donor such as a cadmium vacancy, or a sulfur

interstitial. The authors reasoned that the donor may be Na or Li

interstitials because of the small size of these atoms, and when

crystals are heavily doped, they become highly compensated.

Unfortunately they were unable to prove the identity of the donor.

Later work by Henry and Nassau involved measuring the

spectroscopic lifetimes of the two bound exciton complexes.22 They

were basing their work on another study by Thomas and Hopfield which

showed the measured oscillator strength of the bound excitons to be

very large, which meant that the radiative decay of the weakly bound

exciton would be very fast. Thomas and Hopfield determined an

oscillator strength of 9 +/-2 corresponding to a radiative lifetime of

0.4 +/-0.1 nsec.23 This oscillator strength is about 104 greater than

the strength of a free exciton.II Henry and Nassau with their

experimental setup were able to measure a radiative lifetime for the 12


exciton to be 0.5+/-0.1 nsec. The very fast decay time of this state

is what makes CdS so attractive for nonlinear optical application.

Resonant Raman Scattering in CdS

Further studies of exciton levels in CdS have involved the

measurement of multiple phonon scattering from exciton states. Leite,

Scott, and Damen found that the scattering of longitudinal optical (LO)

phonons was enhanced when the scattering-phonon frequency coincided

with that of excitons.24 They were able to show up to nine orders of

resonant Raman scattering occurring at frequencies shifted less than 1%

from multiples of the 305 cm-I line (the first LO line). These authors

were not able to determine the identity of the exciton state (i.e. free

or bound) which was acting as the intermediate state for the resonant

scattering in this study, butin a latter paper by Leite, Scott, and

Damen free excitons were proven to be the primary intermediate state.25

They also showed that bound excitons participated as intermediates by

observing phonon sideband features on the photoluminescence of the

bound excitons. The I1 exciton was found to be a stronger resonant

state compared to the 12 bound exciton.

In a more recent paper by Mashshenko,61 the temperature dependence

of the LO and 2LO lines associated with the A exciton were

investigated. At 77 K the A-LO phonon predominates the emission, but

as the temperature is increased to 110 K, this line decreases and the

A-2LO line increases. This increase in the 2LO line indicates that an

increase in the probability of a two-phonon process occurs at high


Nonlinear Susceptibility of CdS

Even with the obvious advantages of using CdS for nonlinear

optical applications, very few groups have investigated this material.

The primary amount of work in this area on CdS has been carried out by

Dagenais.3,4,11 As would be expected from the large oscillator

strength of the bound exciton, a very large nonlinear refractive index

results when this level is saturated, and the decay time is very short.

By use of a Fabry-Perot arrangement and a narrow bandpass tunable dye

laser, Dagenais reported a cw saturation intensity for this level of

only 58 W/cm2, which corresponds to a nonlinear index of refraction of

1 X 10-4O cm2/W.3 In a later publication, Dagenais and Sharfin1 report

that by using a high finesse Fabry-Perot, a saturation intensity of

only 26 W/cm2 is required, which for the experimental setup corresponds

to a nonlinear refractive index of 2 X 10-2 cm2/W. They also reported

a switch up and switch down time of one and two nanoseconds

respectively. These are the largest values of a nonlinear refractive

index ever to be reported.

Other work on CdS has been done by Bohnert, Kalt, and

Klingshirn.26 They did not, however, study nonlinearity by exciton

saturation, but rather by the formation of an electron-hole plasma.

High intensity laser pulses of energies just above the band gap energy

were used to study this effect. By measuring the temporal line shape

of the transmitted laser pulse they were able to determine the

renormalization of the band gap due to the formation of an electron-

hole plasma. Because this is a much smaller effect, intensities of 120

kW/cm2 were required to produce a change in the transmitted pulse.


CdS Thin Films

Vacuum Deposition

A great number of studies have investigated the thermal

evaporation of CdS for the deposition of thin films. Although all thin

films for this study were deposited by RF-magnetron sputtering, some of

the results of these investigations are relevant to the present work.

Many of the studies of thermal evaporation were undertaken to research

the electrical properties of CdS thin films, although some optical

properties have been studied. The problem is that most of these

studies present results which both show differences from single crystal

results and also differ from each other.27 The reasons for the

discrepancies are related to the difficulties in evaporating CdS, which

lead to problems with maintaining stoichiometry and crystal structure

in the deposited films. The difficulty in evaporating CdS, as well as

other chalcogenide compounds, is that complete dissociation of the

compound occurs during evaporation.28 If too high a temperature is

used, then the compound will dissociate incongruently and because Cd

has a higher volatility than S, non-stoichiometric thin films result.

Source temperatures between 650 and 700 must be accurately controlled

to avoid excess cadmium.28,'29 Due to differences in the sticking

coefficients of Cd and S, a non-stoichiometric thin film will also

result if the an improper substrate temperature is used. Cook and

Christy27 have found that if fused quartz substrates are used at

temperatures greater than 200 C, then non-uniform films result. In

comparison, Wohlgemuth et al.29 have found that if the substrate is


cooled to LN2 temperatures, then an amorphous film results which is Cd


The optical properties of CdS thin films are central to this

study; however, the reported results for vacuum vapor deposited thin

films are found to vary greatly. Most notably is the variation in the

reported optical band gap and absorption coefficient, and the

calculated index of refraction. Some of these differences can be

realized by examining Figure 8 and Figure 9 which show the refractive

index and absorption coefficient obtained by several authors.29

Referring to Figure 8 the results for single crystal CdS obtained by

Cardona and Harbeke30 are given by curve a. Curve b is for

polycrystalline film deposited onto fused silica at 180 C by Khawaja

and Tomlin. 31 Curve c corresponds to a polycrystalline film also

deposited on silica at 180 C by Wohlgemuth et al.29 and curve d is for

an amorphous film deposited at LN2 temperatures by Wohlgemuth et al.

The same authors correspond to the same curve letters of the reported

absorption coefficients shown in Figure 9. As shown in Figure 8 the

results by Khawaja and Tomlin show the closest resemblance to single

crystal results, while the Wohlgemuth et al. results show a marked

difference. In contrast, as shown in Figure 9, the absorption

coefficient of polycrystalline films deposited by Wohlgemuth et al.

model single crystal results best. The highest quality films in these

studies were deposited at high temperatures, although a more recent

study by Cook and Christy27 reports similar results for room

temperature depositions. It is obvious that subtle differences in the

deposition technique result in a wide variation of thin film

3.0 3.0

2.8 -- 2.8

2.6 2.6
n/ n
2.4 2.4

2.2 -2.2

2.0 2.0

--I I I I I -- I 1 1
2000 4000 6000 8000 10000

Figure 8 Variation of index of refraction reported by several




103- I I I
1.0 2.0 3.0 4.0

hfi w (eV)

Figure 9 Variation of the absorption coefficient reported by several


properties. This is probably due to differences in thin film

microstructures, which in the above studies were only determined by X-

ray diffraction. No direct measurements (i.e. electron microscopy)

were used to examine thin films.

Although some of the optical properties of evaporated thin films

have been investigated, albeit there are differences in the results,

apparently very few studies have been made of the photoluminescence of

thin films produced by this technique. Christmann et al.32 describe

evaporation deposited epitaxial CdS films which displayed both green

and blue edge luminescence. This is one of the first studies to relate

both the morphology and composition of thin films to the observed

photoluminescence. Thin films were deposited on cleaved surfaces of

SrF2 at temperatures from 210 to 310 C. Variation of the

supersaturation by controlling either the source temperature or the

substrate temperature resulted in three basic morphologies. At low

supersaturations, smooth CdS thin films were produced which only

displayed broad band green photoluminescence. By decreasing the

substrate temperature or increasing the source temperature, films were

produced which displayed a structure with many hexagonal flat tops.

These films showed green edge emissions and very low intensity blue

edge emissions. Finally, at very high supersaturations, films which

displayed a morphology with many hexagonal pyramids were found to show

very intense blue edge, or bound exciton photoluminescence.

Through cathodoluminescence and microprobe studies it was

determined that the hexagonal pyramids were cadmium rich, and they did

not contribute to the luminescence. Only the areas adjacent to the

hexagonal pyramids were found to show bound exciton luminescence. The

pyramids were thought to have nucleated from Cd droplets, and as they

grew, the adjacent areas were depleted of cadmium. The required donor

states for 12 bound excitons were therefore provided by cadmium

vacancies. Other films were found to be uniformly slightly cadmium

rich, and it was thought that cadmium interstitials were responsible

for the broad band photoluminescence.

Humenberger et al.33 also reported thin films which displayed

bound exciton luminescence. These films, however, were made by a hot-

wall epitaxial technique, which is similar to the vapor phase technique

used to make bulk single crystal platelets. Films were deposited on

BaF2 substrates. The surface morphology was found to be smooth with a

low density of hexagonal flat tops. No compositional data were given

for thin films; however, acceptor states were provided by indium doping

thin films during the deposition process. Free carrier concentrations

as a result of the indium doping were measured to be on the order of

1017 to 1018 cm-3.33

The above two studies are the first to show by photoluminescence

the presence of exciton levels in thin films. Studies of several other

thin film deposition techniques such as chemical bath,34 or spray

pyrolisis35 have reported photoluminescence, but none have shown

exciton emissions. To see these transitions, it is apparent that very

high quality thin films are required, and it is obvious that standard

vacuum techniques or other techniques such as those described here are

not capable of producing thin films of the necessary degree of quality.

A deposition technique which permits greater control over the


deposition process is RF sputtering, and several investigations of the

deposition of CdS by this technique are presented below.

RF Sputtering

Only recently has the deposition of CdS thin films by RF

sputtering been investigated to any great extent. One of the first

investigations was reported by Lagnado and Lichtensteiger.36 They

described some of the properties of CdS thin films produced by RF-diode

sputtering. A very strong preferred orientation of the CdS

crystallites was found to occur with the c-axis parallel to the

substrate plane. Electrical resistivity measurements at different

temperatures revealed two activation energies for conduction,

interpreted to show the activation of an unspecified trap below the

conduction band, although no data on the chemical analysis for

impurities was presented.

Recently the technique of RF-diode sputtering has received

considerable attention for producing CdS thin films for solar cell

heterojunctions.37 The large photoconductivity of CdS makes it an

attractive material for both solar cell and sensitive photodetector

applications. A number of investigations on the RF diode-sputtering of

CdS for these applications have been reported by Martil et al.38-41

These workers have investigated the dependence of the physical,

electrical, and optical properties on both deposition parameters and

post deposition heat treatments.

The earliest publication by Martil et al.38 outlines the

dependence of deposition rate, thin film grain size and resistivity as

a function of sputtering power, pressure, substrate temperature and

substrate bias. Thin films were deposited onto fused silica

substrates. The first effect they describe is the dependence of the

deposition rate and resistivity on the pressure of gas used. A maximum

in the deposition rate and a minimum in resistivity occurred for a

sputtering pressure of 5 pm (5 X 10-3 torr). For higher pressures they

found the rate actually decreased, and the resistivity increased. The

decrease in deposition rate at higher pressures was explained by a

backscattering mechanism which increases the probability that more

sputtered atoms will return to the target at higher pressures. The

minimum in resistivity at 5 pm pressure was claimed to be a result of

the maximum deposition rate that occurred at this pressure. Although

no direct proof was given, the authors claim the high deposition rate

decreased the number or concentration of impurities that were trapped

in the film.

The second effect described in this early paper by Martil et al.38

was the dependence of thin film grain size on the substrate temperature

and substrate bias. The grain size was determined by SEM observations.

As would be expected, the grain size of films increased from 300 A to

3500 A as the substrate temperature was increased from 90 C to 300 C.

The deposition rate accordingly decreased as the temperature was

increased. When the substrate bias was increased to above -100 volts,

the resulting thin films were found to be amorphous. The temperature

effects can be explained by considering that with increasing

temperature 1) the critical size for a nucleus increases, 2) the

surface diffusion coefficient increases and 3) the sticking coefficient


for a material decreases.28 The appearance of an amorphous structure

with high substrate bias was explained by the authors as a result of

structural damage that occurred due to ion bombardment (which was

increased by the high negative bias).

The second report published by Martil et al.39 further explored

the effects of substrate temperature and bias on the electrical

properties of CdS thin films. They found that the resistivity

increased from 10 to 108 9 cm as the temperature was increased from 60

C to 250 C, which they claimed was due to a change in stoichiometry

(i.e. a loss of cadmium with increasing temperatures), although no

chemical analysis was presented to back up this claim. A minimum in

resistivity was also found with a substrate bias of -50 volts, or with

a floating substrate (which developed a self bias of -28 volts). This

effect was explained by the ion bombardment that results from these

bias voltages preferentially resputting oxygen and other impurities

from the growing film. Oxygen has been shown to be a acceptor-like

trapping center, located 0.9 eV below the conduction band.42 Again no

chemical analysis was reported to prove that a decrease in oxygen

occurs. Also the variation in activation energy for conduction that

occurs with temperature as seen by Lagnado and Lichtensteiger36 was not

explored by these authors.

The third publication by Martil et al.40 reports the influence of

the above sputtering parameters on the optical properties of thin

films. The structural changes which occur under different conditions

were also further explored. Increasing the substrate temperature

resulted in similar structural changes to those reported in the earlier


study; the grain size varied between 500 A and 3000-4000 A as the

substrate temperature was increased from 60 to 300 C. This report,

however described the sputtering pressure dependence of the

crytstallinity. The sputtering pressure determines how much structural

damage occurs due to ion bombardment. At low pressures, bombardment is

enhanced by a large self-bias that develops on the substrate, and an

amorphous structure results as previously described. At higher

pressures, the crystallinity was also found to decrease, which was

explained by the authors as due to a porous structure that develops

from trapped gases.

Optical properties were found to be a function of both substrate

bias and sputtering pressure. A maximum in the optical band gap (2.36

eV) was found to occur for a floating substrate bias, whereas the

minimum in the band gap (2.30 eV) occurred for a -110 volt bias. The

band gap was found to increase with an increase in sputtering pressure

and became nearly constant for pressures above 10 pm. The pressure

dependence of the refractive index however contradicts the pressure

dependence of the band gap. The index of refraction was found to be a

maximum with 5 pm pressure, and it decreased with increasing pressure.

This was explained in terms of the porous structure which occurred at

higher pressures; however, if the refractive index is reduced at higher

sputtering pressures due to a more porous, less crystalline structure,

then the band gap energy should also decrease. An increase in the

number of crystalline defects should cause tailing of the band gap,

which is seen as a decrease in the gap energy.29


The latest investigation to be reported by Martil et al.41

describes the effects of heat treatments on the electrical and optical

properties of sputtered films. Heat treatments were carried out under

H2 and N2 atmospheres at temperatures ranging from 100 C to 550 C and

for times ranging from 20 minutes to 5 hours. The primary effect on

the electrical properties was a two order of magnitude reduction in the

resistivity to 4 X 10-2 Q cm for heat treatments at 200 C. Carrier

mobilities accordingly increased to 50 70 cm2/Vs for this treatment.

Higher temperatures were found to increase the resistivity and decrease

the mobility. The decrease in resistivity was explained by claiming

that oxygen desorption from the grain boundaries occurs at 200 C. As

previously described, oxygen is a trapping center which in this case

decreased the carrier concentration and simultaneously increased the

scattering which reduces the mobility.41 No explanation was given for

the increase in resistivity with increasing temperature, although it

could be due to a loss of cadmium.

The optical absorption edge was shown to become sharper and occur

at a higher energy for heat treatments at 200 C. Temperatures higher

than this were not reported to significantly alter the absorption edge;

however, treatments above 550 C produced an overall decrease in the

transmission. This was explained by a dissociation or reevaporation

process that occurred as a consequence of the higher temperatures. The

optical band gap was found to increase from 2.36 eV to 2.39 eV for

treatments at 200 C, which was again explained in terms of oxygen

desorption from the grain boundaries, although no chemical analysis was

reported to prove this hypothesis.


Semiconductor Doped Filter Glasses

Sharp cutoff filter glasses are silica based glasses which contain

a fine dispersion of semiconductor microcrystallites. The variation of

the sharp absorption edge position is achieved by either varying the

composition and or heat treatment schedule. Great interest has been

generated recently because it is thought by many authors that the

crystal size developed in these materials by thermal treatments is on

the size order for quantum confinement effects to occur. For glasses

containing a mixture of CdS and CdSe many authors have reported a blue

shift in the absorption band edge. In addition, a blue shift of the

high energy exciton photoluminescence peak has been interpreted in

terms of quantum confinement.

Warnock and Awashalom43,'44 in two publications on mixed crystal

glass (those containing CdS.27/Se.73) report that the glasses which

displayed smallest size distribution (average size = 94 A) exhibited

the largest shift in the exciton photoluminescence, and this

photoluminescence was shown to decay on a time scale of only 18 psec,

which is nearly two orders of magnitude faster than the lifetime

displayed by excitons in bulk material.44 The shifts in peak position

and fast lifetime were interpreted in terms of a quantum confinement


A very recent paper by Borrelli et al.45 has shown that the blue

shift in the photoluminescence displayed by this particular mixed

crystal glass is not due to confinement effects, but rather to

compositional effects. A careful study of the crystal structure by X-

ray diffraction revealed that a change in the stoichiometry occurs


during the heat treatments which develop the microcrystallites. It was

postulated by these authors that more of the selenium remains in glass,

while sulfur is more easily incorporated in the crystallites.45

Therefore, at the lower temperatures which produce the smaller

crystallites, the crystallites end up containing more sulfur, so the

optical properties of glasses containing these crystallites are blue

shifted toward the optical properties of pure CdS. True quantum

confinement effects were, however, shown to occur in a series of

experimental glasses which contained either CdS or CdSe, but not both.


Vacuum Deposition System

Thin films for this study were made with a specially designed

vacuum system. A schematic of the deposition chamber is shown in

Figure 10, and a picture of the complete system is shown in Figure 11.

The deposition chamber consists of a rectangularly shaped stainless

steel box which measures 12" X 18" X 18". This chamber was custom

built by MDC Vacuum Corp. (Haywood, CA.). Numerous ports and

feedthroughs on the chamber permit a wide variation of deposition

configurations. The vacuum pumping system utilizes a 330 1/sec

turbomolecular pump (Balzers, Hudson, NH), a molecular sieve trap (MDC

Vacuum Corp.), and a 300 1/min two-stage mechanical pump (Sargent

Welch, Skokie, IL). With the molecular sieve trap activated, pressures

in the high 10-5 torr range were possible using the mechanical pump

only. Pumpdown from atmosphere pressure to 1X10-6 torr could be

accomplished in 2 hours with use of the turbopump. Vacuum gauging was

performed with a Leybold Heraues (East Syracuse, NY) model CM 330

combined Penning discharge and thermocouple gauge controller.

Pressures for sputter deposition were controlled by using a

micrometer adjustable throttle valve (Sputtered Films Inc. Santa

Barbara, CA.) and a mass flow meter/controller (Matheson Gas Products

Norcross, GA). High purity argon gas (99.9995%) was used as the






Figure 10 Schematic diagram of sputter deposition chamber. Computer
figure shown to indicate computer control of system.

Photograph of complete deposition system showing
sputtering chamber and instrumentation.

Figure 11


sputtering gas. Initially the thermocouple vacuum gauge was calibrated

with a capacitance manometer (MKS Instruments Inc. Burlington, MA) for

different settings of the throttle valve and mass flow controller.

Based on the previous work on CdS sputtered films, a pressure of five

microns (5X10-3 torr) was chosen and the system was calibrated for this

pressure. To obtain this pressure an argon flow rate of 20 cc/m was


Film thickness during deposition was monitored with a Leybold-

Heraeus IC 6000 deposition controller. This instrument utilizes a

quartz crystal oscillator microbalance for determination of film

thickness. All parameters of calibration and deposition control are

software programmable with this unit, either via the front panel

controls, or through an RS-232 communications link to an external

computer. The IC 6000 permits monitoring of up to six different films,

and by using a sample and hold program, two different films can be

monitored and controlled simultaneously. Close loop control of

deposition rates was achieved by connecting the IC 6000 thickness

monitor to the RF power supplies.

For semi-automatic control and constant monitoring of the

deposition process a Zenith Z-158 personal computer with 512K RAM and a

20 megabyte hard disk was interfaced to the system. A Dascon 1 I/O

board (Metrabyte Corp.) which provides four channels of analog and 12

bits of digital input/output was installed in the computer. With this

interface board and with RS-232 communications to the IC 6000

deposition controller, it was possible to monitor nearly every aspect

of the deposition process. The RS-232 communications permitted direct


programming control over deposition parameters and constant monitoring

of deposition rate, thickness, power output and crystal oscillator

status. The analog portion of the I/O board was connected to the

pressure gauge controller and to the RF power supplies. Connection to

the power supplies permitted recording of the DC bias which develops on

the target as a result of the sputtering process. All of these process

parameters were stored on disk during a deposition run. This made it

possible to go back and review the deposition process, if a film was

later found to have anomalous properties.

A unique feature of the deposition system has to do with the

various substrate holders which could be used. For low temperature

depositions, a liquid nitrogen (LN2) cooled substrate holder was

employed. This holder is made up of two double concentric tubes in

which the center tube acts as a reservoir for LN2 and the outer tube,

which is open at the bottom to the vacuum system, acts as an insulating

vacuum jacket. An aluminum plate with a slightly recessed area to hold

the substrate is attached to the end of the inner reservoir tube with a

copper screw. To reduce the amount of contamination that would occur

on the cold substrate, an additional LN2 cooled coil was positioned

around the substrate holder assembly. This coil is referred to as the

cryoshield in Figure 10. For room temperature depositions, the same

holder was used, however, without filling the LN2 reservoir.

High temperature depositions were accomplished by replacing the

LN2 feedthrough assembly with a resistively heated substrate holder.

This holder was made by drilling holes lengthwise in a thin aluminum

block and inserting nichrome wound ceramic heaters. A Chromel-Alumel


thermocouple was embedded in the core of the holder, and control of the

temperature was made by controlling the input power to the holder with

a variable autotransformer. Substrates could be heated to 300 C with

this arrangement with a temperature control of +/- 5 C.

A single planar magnetron sputter gun supplied by US Guns

(Campbell,CA) was used for sputter depositing the CdS thin films. The

gun accepts two inch diameter targets, which can range from 1/16th to a

quarter of an inch in thickness. An Eratron RF power supply

(Campbell,CA) operating at 13.57 MHz with a power capability of 600

watts was used to supply RF power to the gun, through an auto load

match tuning network. The matching network is require to balance the

impedance of the gun and plasma to the 50 ohm output from the power

supply. If the impedance is not matched, then the RF power is

reflected, and the only thing that is accomplished is heating of the

heat sinks in the power supply. The utility of an auto load network is

that when conditions of the plasma change, the network will

automatically match the impedance, thereby eliminating any reflected RF


High purity cadmium sulfide sputtering targets were obtained from

CVD Industries (Woburn, MA). These targets are made by a chemical

vapor deposition process so they are supplied with a bulk density of

nearly 98% of theoretical density. High bulk density reduces the

amount of outgassing during sputtering. All impurities were less than

1 ppm and the stoichiometry of the target material was slightly cadmium

rich (50.46%). Nearly all thin films for this study were made from CVD

targets; however, a different type of sputtering target was obtained

from EM Chemicals Inc. (Hawthorne, NY) for testing. This target was

made by a hot pressing technique, which results in a lower bulk density

and a higher impurity content. Impurity information was not given by

the company for these targets; however, thin films that were sputtered

from these targets were found to contain much higher levels of iron

than the levels found in films sputtered from CVD targets.

Targets that were used for COSAD were made from a piece of glass

obtained from Schott Glass Industries (Duryea, PA). The glass was a

special type of borosilicate crown glass (#8329) which is similar in

composition to a BK-7 glass, but it is processed in such a way so that

all the volatile impurities in the glass are removed. This glass is

sold by Schott as a special type of over coating glass used for thermal

evaporation. Targets were made by first sectioning the as supplied

cylinder of glass into 3/16 inch discs from which two inch blanks were

cut. Both sides of a blank were polished to 600 grit before being used

as a sputtering target. After it was found that a high power level was

required to sputter this glass, a 1/16th inch copper plate was bonded

to one blank to improve thermal contact to the sputter gun. The

bonding was accomplished with an indium tin solder.

For producing COSAD films, the deposition chamber was reconfigured

to accept a second RF planar magnetron sputter gun, for sputtering the

glass target. This gun was used with a second 600 watt Eratron supply

and the two power supplies were driven by a common oscillator to phase

match the two plasmas. In addition, a set of baffling plates were

mounted inside the chamber as shown in Figure 12. The center baffle

plate was used to isolate the two plasmas. A pair of three inch


"ii l .. . .... .. ... ...



iAlt MU




Figure 12 Schematic diagram of deposition chamber configured for

A Tr I IhI nrnm .?.'.'.- .%-z .%-.- -.-#


diameter apertures centered over the two sputter guns were cut in the

top plate to provide access to the two plasma sources. A dual

substrate holder was rotated in the plane above the two apertures, so

that the substrate was alternatingly exposed to one source and then the

other. The dual holder provided the capability of producing two films

during a single deposition run. Initially a DC motor with a reducing

gear drive was used to rotate the substrates with a linear speed. A

more sophisticated direct drive stepping motor was later added to the

system to permit variable exposure to the two plasma sources. The

stepping motor was interfaced with the Dascon I/O board in the Z-158

computer via an external digital control circuit.

One disadvantage of the COSAD configuration was that substrates

could not be heated or cooled and the temperature of the substrate

could not be monitored during a deposition. The utility of this

particular system configuration, however, is that single component

films could still be produced, simply by positioning the substrate

holders over the respective apertures, and exposing the substrate to

deposit a film.

RF Magnetron Sputtering

As described in the Bibliographic Review section, the only sputter

deposition of CdS was carried out by RF-diode sputtering. In this

configuration two parallel plates are used. The target is attached to

the cathode plate, and substrate is attached to the anode plate, which

can either be at ground or floating potential. A plasma is generated

between the two plates by applying RF power to the cathode plate.


Electrons in the plasma can be accelerated by the positive half of the

RF field, which results in both the ionization of argon atoms, and the

bombardment of the target. This bombardment results in the development

of a negative potential on the target surface, which is called the DC

bias. The argon ions in the plasma are too massive to be accelerated by

the RF field; however, they are accelerated towards the target by the

DC bias, which results in sputtering of the target material.

The configuration of diode sputtering results in exposing the

substrate and growing films to the full energy of the plasma. The

growing film is continuously bombarded by a number of energetic

particles, both charged and neutral. This bombardment can either be

enhanced or reduced by applying a potential to the substrate during

deposition. Substrate biasing can only reduce the amount of

bombardment to a certain extent, essentially because the substrate is

totally emersed in the plasma during deposition. Considerable heating

of the substrate and film occurs as a result of the bombardment, even

at moderate plasma powers.

RF planar magnetron sputtering reduces the amount of bombardment

by incorporating a strong permanent magnet behind the cathode assembly.

The configuration of a US sputter gun is shown schematically in Figure

13. Curved magnetic field lines force the electrons in the plasma into

circular orbits, which both enhances the plasma immediately above the

target, and reduces the amount of electron bombardment on the growing

film. The target is a circular disk that is secured to the magnet

housing, which comprises the cathode. The anode is comprised of a

circular cap (called the ground shield in the Figure) which fits over


& MA N T--__





Figure 13 Cross-sectional view of RF magnetron sputter gun made by US
Guns Inc.46


the gun assembly. Most of the plasma is constrained to the area

immediately above the target, so a large reduction in substrate

bombardment and substrate heating is realized. The substrate and

growing film, however, can still be bombarded by energetic neutral

particles, which has been shown to be a major cause of substrate

heating.47 Still the heating is reduced and as will be pointed out in

the Results section (Chapter 5), a considerable difference in the as

deposited film properties occurs with RF planar magnetron sputtering,

compared to RF diode sputtering.

Thin Film Deposition

Prior to deposition of thin films, all substrates were subjected

to a three phase cleaning procedure. The first phase of the process

was an ultrasonic bath in DI water for 10 minutes. After this

treatment the substrates were removed from the water and rinsed with

isopropyl alcohol (IPA). A second ultrasonic cleaning was then carried

out for 10 minutes in IPA. The final phase of the cleaning procedure

was a 15 minute treatment in a vapor degreaser, which employed IPA as

the solvent. Substrates were slowly removed from the vapor, which

allowed the condensed vapor droplets to evaporate. To check for

cleanliness substrates were examined by edge illumination against a

black background. This permitted observation of any contaminates on

the surface of the substrate in addition to any small flaws in the


To deposit thin films, substrates were mounted on a substrate

holder and positioned in the deposition system immediately after

cleaning. The deposition system was pumped down to a pressure of < 1.0

X 10-6 torr before backfilling with the sputtering gas. The IC-6000

was usually programmed for a fast ramp up to the desired power level

and then a rate controlled presputter was initiated before the shutter

was opened. If a sputtering target was newly installed, a presputter

burn-in of 30 minutes was carried out; otherwise the target was

presputtered for 5 minutes before exposing the substrate to the plasma.

The IC-6000 automatically controlled the deposition rate and thickness.

Deposition rate was varied between 1 to 5 A/sec and films for optical

analysis were typically made 1 pm thick, although the film thickness

for other analysis was varied.

Substrate materials

Several different types of substrates were used to support thin

films, depending on how the films were to be analyzed. Standard

substrates for optical analysis were made of optical quality fused

silica, typically one inch square and 1 mm thick. These substrates

were found to have very few surface flaws and were essentially

transparent over the optical region in which films were analyzed. The

low index of refraction (n=1.46) of these substrates was also necessary

for measuring the refractive index of the film and for planar waveguide

measurements. Some films were also deposited on silica for TEM

analysis. Self-supporting TEM specimens were prepared by floating

these films off the silica with a 1% HF solution.

Other films for optical analysis were deposited onto single

crystal sapphire substrates. This was done to determine if the high


thermal conductivity of the sapphire would aid in reducing heating by

the laser beam during low temperature spectroscopic measurements.

These substrates were of very high quality with no detectable flaws and

a very smooth surface finish. Some COSAD films were deposited onto

sapphire substrates to aid in X-ray chemical analysis, since the

sapphire had no X-ray lines which overlapped the lines in these films.

For other TEM analysis, films were deposited on slabs of single

crystal NaCI and on TEM grids which held carbon support films. The TEM

grid-carbon film combination was used for quick analysis of deposited

films. Once the grid-carbon film assembly was prepared, thin films

could be deposited directly onto the carbon film, and no other sample

preparation was necessary. Films of thickness from 500 to 3000 A were

typically deposited for TEM analysis. The carbon support film was made

by thermally evaporating approximately 200 A of amorphous carbon onto

sheets of single crystal mica. After the carbon film was floated off

the mica, 300 mesh TEM grids were carefully placed on the floating

film. The film with grids was then picked up on a thin cardboard card

and allowed to dry.

Single crystal NaCl was used as a substrate because it was found

that after heat treatment or as a result of a high temperature

deposition, thin films of CdS could not be floated off silica

substrates. Normally the 1% solution of HF would readily float films

off silica, but after any type of heat treatment above 200 C not even

a high concentration of HF was successful for floating the films.

Apparently some type of chemical bond forms between CdS and the silica

even at relatively low temperatures. Single crystal NaCI substrates


were also required for TEM analysis of COSAD films, since the HF

solutions normally used to lift CdS off silica slides would strongly

attack these films.

Substrate temperatures

Several different substrate temperatures were used to investigate

the effect of deposition temperature on thin film grain size. With

the LN2 feedthrough assembly that was described above, substrates could

be cooled to LN2 temperatures in about 20 minutes. For some

depositions a low temperature thermal contact paste was use to mount

the substrate in the holder, although most frequently, standard hold

down clips were used. An Iron-Constantan thermocouple was initially

used to monitor the temperature of the substrate holder during a

deposition. To use a digital meter to measure the output from this

thermocouple during a plasma deposition, an RC circuit had to be used

to decouple the RF signal that was imposed on the DC thermocouple

signal. The RC circuit is essentially a low band pass filter. A

schematic diagram of the circuit is shown in Figure 14. If this filter

is not used, a very large RF signal can be induced in the lead wires,

which will destroy most D/A converters used in digital meters.

For depositing films at room temperature, the LN2 feedthrough

assembly (without LN2 in the reservoir) was used to hold substrates.

These depositions were very close to room temperature, since it was

found that even with the highest deposition rates, the holder would

only heat up to 35 C. When substrate temperatures greater than this

were required, the LN2 assembly was replaced with the resistively

heated holder. Temperatures as high as 300 C could be obtained in

r O.1F------------
'' .OI^F


'. 0.01 pF
O.... ..1

Schematic diagram of RC decoupling circuit used protect
digital equipment from RF transients.48

Figure 14

about 10 minutes with this holder, although most depositions were made

at 200 C. The unique feature of this holder is that it permitted

outgassing of the substrates prior to deposition. This was done by

heating to 200 C for 30 minutes in high vacuum.

Deposition rates

A variation in deposition rates was also studied to determine its

effect on thin film properties. For pure films of CdS, rates from 1

A/sec to 5 A/sec were used. These rates were monitored and controlled

by the IC 6000 deposition controller. Figure 15 shows a graph of the

sputtering rate versus power density for both CdS and BK-7 targets. As

shown, the glass target used for making COSAD films is much more

difficult to sputter, as an input of 275 watts (13.5 W/cm2) only

resulted in a deposition rate of 2 A/sec. Power levels greater than

300 watts could only be used with the BK-7 glass target which had a

copper backing plate bounded to the back of the target. Even at 500

watts, however, a deposition rate of only 4 A/sec could be obtained.

At this power level, considerable heating of the substrate was found to

occur, even when the substrate was rotated during COSAD runs.

Co-Sputter Alternating Deposition (COSAD)

By using the dual sputter gun configuration described above, a

very unique type of thin film could be produced. The acronym describes

the process by which these films were made; by rotating the substrate

above the two sputter guns, an alternating layer of glass and the CdS

was deposited, which was repeated continuously. Initially the two guns

1 I

0 -T-

Figure 15

2 4 6 8 10 12 14

Graph of sputtering rate versus input power density for CdS
and BK-7 targets.


were set up to sputter deposit at a static rate of 3 A/sec. This is

the maximum rate that could be used with the glass target without

causing excessive heating. The circle over which the substrate was

rotated was eight inches in diameter, so with the three inch diameter

apertures used with the guns, the duty cycle was rather low, only 0.12

for each gun. With a rotation speed of 20 RPM, the resulting total

dynamic deposition rate is less than 1 A/sec. One possible way to

increase this rate would have been to reduce the rotation speed. This,

however, would also mean that the substrate would be spending more time

over areas which were not depositing film and were actually depositing


The use of the computer controlled stepping motor to rotate the

substrate holder alleviated the problem with the linear drive system.

With this setup is was possible to increase the amount of time the

substrate spent over a given source. At the same time, it reduced the

amount of impurities introduced into the film because the stepping

motor could be driven full speed between the two sources, which was

about 120 RPM. By simply altering the computer program, this system

permitted easy variation of deposited structure.

Post Deposition Treatments

To improve many of the physical properties of thin films a post

deposition heat treatment had to be instituted. Generally, heat

treatments on CdS thin films were made in a fused silica muffle tube

furnace under a flowing argon atmosphere. The argon used for heat

treatments was of the same grade as the gas used for sputtering. For


annealing COSAD films high purity compressed air was used with the same

furnace setup. Flow rates of 100 cc/m were typically used.

Temperature control was accomplished with a single setpoint temperature

controller (Love Control Co.) which provided temperature control to +/-

1 C. Two thermocouples were used, one mounted outside the tube for

control and one positioned inside the tube adjacent to the sample.

Heating rates of up to 35 C/min could be realized, and cooling rates

of 100 C/min could be obtained by sliding the muffle tube out of the

furnace hot zone. Temperatures for heat treating thin films ranged

from 200 C to 700 C, with times ranging from 1 minute up to 5 hours.

To reduce the amount of grain growth that would occur as a result

of high temperature heat treatments, some films were heated by a

technique known as rapid thermal annealing. This process utilizes high

power quartz lamps to rapidly heat the sample. Because of the low

thermal mass that is achieved with this technique, heating rates in

excess of 500 C/sec are possible. The sample is sandwiched between

two thin sheets of graphite which act as black body absorbers and help

to transfer the heat generated by the lamps to the sample. In this

particular system lamps with a total power of 5000 watts were used.

Heat treatments up to 650 C with times ranging from 10 to 60 seconds

were done under a flowing nitrogen atmosphere.

Thin Film Characterization


Determination of thin film microstructure was achieved using

several analytical techniques. Direct examination of the


microstructure was made by electron microscopy. The various TEM

specimens that were described in a previous section were examined in a

JEOL 200 CX analytical microscope (Japanese Electron Optics Laboratory,

Boston, MA.). This instrument was used in both transmission (TEM) and

scanning transmission (STEM) modes to directly observe and measure the

ultra-fine structure of thin films. Selected area diffraction (SAD)

mode was used to conduct electron diffraction experiments for

determination of crystal structure and orientation. In the SAD mode it

was possible to obtain diffraction information from areas as small as 5

um. The line to line resolution observed in the TEM mode was 8 A.

Another technique used for determination of crystal orientation was

dark field imaging. By tilting the primary beam in the microscope, the

contrast that is observed in a dark field image is only due to those

crystals or portions of crystals which are oriented for a specific

Bragg reflection. Dark field imaging makes it possible to observe

preferred orientation in a thin film or any portion of a crystal that

contains defects (such as microtwins).49 On the JEOL 200CX changing

from bright field to dark field was accomplished with a single switch.

To study the topographical structure of thicker films a JEOL 35CF

scanning electron microscope was used. Films which were made for

optical analysis were frequently examined with this instrument. The

advantage of this technique for CdS films was that no sample

preparation was required except for sample mounting. The conductivity

of these films was high enough not to warrant coating with a conductive

film, although for COSAD films a thin carbon film had to be deposited

before these films could be examined. With this instrument the


presence of gross defects in the film such as pinholes or cracks could

be observed, as well as the fine grain structure of certain films. The

major disadvantage with this microscope, however, was a limiting

resolution of about 300 A which made it difficult to observe the grain

size of as-deposited thin films.

X-ray diffraction was another technique used to measure the

crystal structure and orientation of thin films. An APD 3720 computer-

controlled diffractometer (Phillips Co.) was used to carry out

diffraction experiments. Typically, diffraction was measured from two-

theta values ranging from 15 to 900 with a angular step of 0.05. Use

of the computer to store X-ray diffraction spectra on disk greatly

facilitated the determination of the change in crystal structure with

heat treatments. Changes in peak positions and intensities could be

readily determined, and by making high resolution scans (0.02/step)

over certain peaks it was possible to calculate the grain size and film

strain using certain utility programs available in the computer system.

Again, no sample preparation was necessary for this technique as

diffraction from 1 Pm films supported on silica substrates was readily


Chemical Analysis

Initially, Energy Dispersive Spectroscopy (EDS) was used to

measure the stoichiometry of thin films during both SEM and STEM

observations; however, it was found to give results that were not

quantitative. Particularly in the SEM, it was found that the results

were very dependent on sample geometry with respect to the detector. A


technique know as X-ray secondary fluorescence was used instead to

measure the stoichiometry of nearly every thin film that was made. The

particular instrument used for these measurements utilized a silver X-

ray tube to produce the primary X-ray beam. Tube voltages of 50 kV

were used with a current of 30 mA. One of four different secondary

targets could be selected to generate the X-rays that were used to

analyze the sample. The advantage of this technique is that the

background radiation from the X-ray tube (bremsstrahlung) is totally

removed and only the very narrow X-ray line corresponding to the

secondary target reaches the sample which results in a great reduction

of the background counts. The particular geometry of this system

limits the radiation to only one polarization which further reduces the

background count.50 Detection levels for most transition metals was

0.1 ppm. Beam size incident on the sample is 1 cm in diameter and

penetration depth is only a few microns which makes this technique

ideally suited for measuring thin films supported by substrates. The

X-rays emitted from the sample are analyzed by EDS with this system;

however, due to the above arrangements, a much more quantitative

determination could be made when the results were compared to a

standard reference.

The standard used, in this case, to determine the stoichiometry of

thin films was a piece of target material. The stoichiometry of the

target was verified by a wet chemical technique and electron probe

microanalysis (EPMA) (JEOL 730 Superprobe). For solution analysis a

portion of the target was dissolved in nitric acid and then this

solution was diluted a given amount. The amount of Cd and S in


solution was determined by using an induction coupled plasma (ICP)

spectrometer (Allied Analytical Systems, Waltham, MA). By comparing

these amounts to those in standard solutions the stoichiometry was

determined. For EPMA the stoichiometry of the target material was

determined by use of standards and a ZAF or Opz program.51 The ratio

of the integrated peak intensities for Cd and S in the target was then

determined by X-ray fluorescence. The ratio obtained from thin films

was compared directly to this ratio to quantitatively determine the


Film thickness could be readily determined with this technique by

measuring the intensity of the silicon K-alpha peak that was emitted by

the substrate through the film. The intensity of this line is

exponentially attenuated by the thickness of material that it passes

through and can be related by the following equation:

I = 0Io exp-at (2.1)

where Io is the intensity from a bare substrate and t is the film

thickness. From equation 2.1 a linear relationship between a function

of the intensity, F(I), and t can be obtained by55

F(I) = log10 I (2.2)

The intensity from three different films of known thickness (verified

by profilometry) were measured and by plotting F(I) versus t, a master

curve was generated. The thickness of an unknown film could then be

found by solving the above equation for F(I) using the intensity from

the unknown. The x-coordinate on the curve for the unknown F(I) would

give the thickness. Film thickness of up to 3 im could accurately be

determined with this technique. Changes in density or loss of material

as a result of heat treatment could also be monitored with this


Several other analytical techniques were used to analyze the

composition of thin films. In each case the results for thin films

were compared to the results obtained from target material. The

different techniques were used to both acquire information, which the

specific technique is most suited for, and to cross-check the

composition of thin films. Auger electron spectroscopy (AES Perkin-

Elmer, Norwalk, CT) gave the least quantitative determination of the

absolute composition, but the technique was very quantitative for

determining differences between samples. X-ray photoelectron

spectroscopy (XPS Kratos,UK) gave semi-quantitative results and also

showed how the surface of thin films were altered by heat treatment, a

result which was not detected with any other technique.

Optical Measurements

By far the most extensive measurements made on thin films were

optical measurements. The availability of several spectrometers,

monochromaters, lasers, and optical hardware greatly facilitated these

measurements. The UV-VIS spectrophotometer used for many measurements

was a Perkin-Elmer Lambda 9 (Norwalk, CT). Photoluminescence was

measured with both a Princeton Applied Research Optical Multichannel

Analyzer (Princeton, NJ), and an Instruments SA Raman spectrometer


(Metuchen, NJ). A variety of lasers were used for different

experiments,, although most photoluminescence experiments were made with

a Spectra-Physics model 2025 Argon Ion laser (Mountain View, CA). Many

measurements were made at both room temperature and at temperatures

down to 9 K. Low temperatures were obtained with an Air Products

(Allentown, PA) closed-cycle helium cold finger and digital temperature

controller. Experiments carried out with these instruments will be

extensively detailed in the following sections.

Index of Refraction

A very unique technique for measuring the dispersion or index of

refraction as a function of wavelength was used on thin films.

Normally the index of refraction of a material is related to the

percent of transmitted light by the well known equation:

T = (1 R)2 exp (-at) (4.1)
1 R' exp (-2at)

where the reflection, R, is given by (n l)2/(n +1)2. The situation

is much more complicated, however, for a thin film supported on a

substrate with a different refractive index. The reflection that

occurs at the film/substrate interface must be dealt with as well as

the reflections that occur at the other interfaces. This leads to a

more complicated expression:

T = (1 RI) (1 R2) (1 R3) exp (-at) (4.2)
(1 R2R3) { 1 [ RIR2 + RIR3 (1 R2) ] exp (-2at) }


where RI, R2, and R3 are the reflectivities of the air-film, film-

substrate, and substrate-air interfaces, respectively.16 The values of

the RI, R2, and R3 are given by the same equation stated above with the

refractive indices of the three materials (air, film, and substrate)

substituted for n, respectively. Even if the value of the absorption

coefficient is known, the transmission must be measured to 0.5% to get

a 1% accuracy in the refractive index,52 which is experimentally very

difficult to do unless a specially designed spectrometer is used.

When the refractive index of the film is significantly different

from that of the substrate (as the case with CdS on silica) a very

strong wavelength dependent interference effect occurs. The effect can

be explained by examining Figure 16. When the incident light beam

passes through the thin film and encounters the interface it will be

reflected due to the differences in refractive indices. Depending on

the thickness and the refractive index of the film (the product of

these two values determines the optical path length) the reflected ray

will either constructively or destructively combine with the incident

beam. This is the same process that occurs in a Fabry-Perot

interferometer described previously. However, instead of changing the

physical distance of the cavity to change the optical path length,

resonance is achieved by scanning the wavelength of incident light. In

terms of the total transmission that is measured, a series of

interference fringes is observed, which vary with wavelength. These

fringes are known as fringes of equal chromatic order (FECO), and the

wavelengths at which maxima and minima in transmission occur is given

by the same equation used for resonance conditions in a Fabry-Perot

/ ------


Tf T8S TfTR8R;

Figure 16 Interference effects at thin film, substrate, and air

T,.T; RS

m X = 2 n d (4.3)

where m is the order number of the fringe, d is the thickness, and X is

the wavelength. To take advantage of this effect to determine the

index of refraction two assumptions must be made. The first is that

the order number is not a discrete number but that it can vary

continuously. Why this is important will be described shortly. The

second assumption is that the dispersion of the refractive index can be

described by the following equation

n2 = A ____ + B(44)

where the square root of B gives the infinite wavelength refractive

index and A and x are constants.

To make use of this effect, a FECO spectrum is obtained at normal

incidence. A typical spectra from a 2 pm CdS film is shown in Figure

17a. Curves are drawn tangent to the maximum and minimum peaks and the

values of the wavelength of the tangent points are determined. The

sample is then rotated in the beam by 30, resulting in an increase in

the optical path length by a factor of 1/(cos 30). The assumption

that the order number is a continuous mathematical variable means that

the normal incidence spectrum can be shifted to a shorter wavelength by

increasing each order number by an amount 6m.34 This increase produces

a shift in the FECO spectra shown in Figure 17b. Again, tangent curves

are drawn and the tangent points to the interference spectrum are

determined. The difference in the pairs of tangency points are then

0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9


1 1.2 1.4 1.6 1.E


Fringes of equal chromatic order (FECO) spectrum for 2.0 um
thick film of CdS on a silica substrate, a) Normal
incidence spectrum showing many orders; b) Expanded
abscissa scale showing shift in FECO spectrum for 30

Figure 17


used in an iterative calculation to determine the values of A, B, x and


The problems associated with this technique involve precise

placement of the sample at the focal point of the beam and accurate

rotation of the sample about the axis which passes through the focal

point. These problems were addressed by using a micrometer adjustable

x-positioner and a micro-rotation stage, which are shown placed in the

Lambda 9 spectrophotometer in Figure 18. Normal incidence to the film

was determined by rotation of the stage a few degrees on either side of

the initial zero and observing the shift in position of a given

interference fringe. The true zero point was taken as the position

which produced the midpoint between the two shifts. The focal point of

the beam was then determined by observing the peak value of a given

fringe as the sample holder was moved with the positioner.

The most difficult part of this technique (except for writing the

iteration program for the analysis) was the manual determination of the

value of the tangency points from the FECO spectra. The wavelength of

the tangent point must be known to a precision of 0.1 nm to obtain an

accuracy of three decimal places in the refractive index. Wavelengths

can be measured with the spectrometer to this precision, but this

precision could not be obtained by manual measurement of the tangent

point from the printout of the spectra.

The problem was alleviated by interfacing the Lambda 9 to an IBM

AT computer. Use of the computer greatly facilitated all aspects of

using the spectrometer as spectra could be stored on disk and

I.. i

Photograph of micro-positioner stage used for acquiring
FECO spectrum, shown mounted in Lambda 9 spectrophotometer.

Figure 18

manipulated later. The BASIC program used to remotely program, run,

and accumulate data from the Lambda 9 is listed in Appendix A.

For the determination of the refractive index a particular

software package available with the AT was found to be very useful.

This software called is Asyst (McMillain Software Co.) and it is a very

powerful mathematical analysis program. To determine the refractive

index, the numbers representing the FECO spectra were imported into

Asyst and stored in a two dimensional array. Using a very short

program, the array was plotted, and the equations for the curves

tangent to the maximum and minimum were fit. Next, arrays were

generated to represent each of the tangent curves. Asyst could then be

used to very easily determine the points where the three arrays

intersected, with an accuracy of 0.1%. The intersection points were

stored in a fourth array. This process was repeated for a 30

incidence spectra. The very long iterative BASIC program that was

originally used to calculate the dispersion equation could be shortened

considerably with utility functions available in Asyst. In fact, by

manipulating the numbers as arrays, the calculation was much simpler

and much faster.

Even though the above process produced an accurate result, a

second technique, ellipsometry, was used to verify these results.

Ellipsometry is based on the change in polarization of a light beam

when it is reflected from a surface. Generally a plane polarized light

beam is made incident on the surface of a material at an angle of 45.

The resulting reflected beam is elliptically polarized and from the


analysis of this polarization the optical constants of the material can

be determined.

The problems associated with this technique mostly originated from

the actual instrument. The computer-controlled Gaertner ellipsometer

(Chicago, IL) is the industry standard; however, the unit available for

use in this study was a manual model which greatly reduces the capacity

for quantitative analysis. The problem does not so much have to do

with manually setting the polarizer and analyzer, but rather in reading

the analog scale to determine the node points. This was a special

problem on this unit because the meter was nonlinear; the highest

sensitivity was between a reading of 2 an 3 on a scale of 1 to 5. In

addition, the gain for the photocell had to be manually adjusted as a

node point was found. The gain was also nonlinear and there was a flat

spot at one end of the range. These two problems contributed to

difficulties in reproducing the results; however, the greatest problem

with the analysis had to do with the computer program used to calculate

the values of n and t from psi and delta. Supplied by Gaertner, this

program did not actually calculate the values of psi and delta, but

instead it looked up the values in a table, based on the values of the

two pairs of angles that were inputed. From the table values of psi

and delta the program then calculated the refractive index and

thickness by an iterative technique. One difficulty with the program

occurred if the particular combination of angles inputed were outside

values in the table. The program would hang up and the only solution

was to boot the program out. This actually happened a surprisingly

large number of times, even when similar samples were analyzed. To


alleviate the problems with this program, a graphical method was

developed to calculate the refractive index and thickness from the set

of angles measured with the ellipsometer. A brief description of this

method is given in Appendix B.

UV-VIS Absorption

Measurement of the fundamental optical absorption edge in

semiconductors permits direct determination of the nature of the band

gap. Optical absorption properties of thin films were measured over a

variety of temperatures with the Lambda 9 spectrophotometer operating

in one of several modes. In the transmission mode a linear, zero to

100% scale was used, which was useful measuring FECO spectra. A more

useful mode for measuring the band edge, however, was to plot the

transmitted light in terms of absorbance. Because the absorbance scale

is the natural logarithm of reciprocal transmission, features of the

band gap absorption above the band edge could be studied. In the

transmission mode unless the scale was expanded these features would

appear indistinct, because transmission above the band gap was

typically less than 1% with the thickness of films usually studied. In

the absorbance mode this 1% transmission would correspond to a value of

4.6 on a possible absorbance scale of 6.0. Another advantage of using

the absorbance scale is that the spectra obtained qualitatively

reflected the shape of the true absorption curve. As explained earlier

in this chapter, the transmission spectrum that is measured of a thin

film on a substrate is the result of a complicated combination of the

reflections which occur at the various interfaces. The transmission,


however, is exponentially proportional to the absorption coefficient,

which is why absorbance spectra approximate absorption spectra. The

two terms will therefore be used interchangeably when describing


The band gap energy was taken as the point where the slope of the

absorption curve was a maximum.54 One unique feature of the Lambda 9

was that it allowed anywhere from the first to fourth derivative of a

spectra to be taken. The maximum of the first derivative of the

absorption edge was taken as the band gap energy. A calculation to

verify these results will be described in Chapter V. Basically, it

evolves solving equation 4.2 for a using the absorbance spectra, and

then plotting a2 verses energy. A direct band gap material will give a

straight line for this plot, and the extrapolation of the line to a2 =

0 gives the optical band gap of the material.54

The temperature dependence of the band gap energy was measured by

cooling the sample with the cold finger. The sample was mounted in a

ring sample holder which would then be attached to the end of the cold

finger. A radiation shield was then positioned around the cold finger

and a vacuum insulation collar with fused quartz windows was then slid

over the entire assembly. A vacuum of less than 1 micron pressure was

then obtained in a few minutes by pumping the system with a Balzer's 70

1/sec turbomolecular pump. Once this vacuum was obtained in the

collar, cooling of the sample to 9 K could be accomplished in less

than 1 hour, although most measurements were made after 2 hours of

cooling. Any intermediate temperature between 9 K and 250 K could

achieved with the digital temperature controller and heater assembly.


The cold finger assembly was mounted on a height-adjustable rack system

so that once a sample was cooled to low temperatures the cold finger

assembly could be moved so that measurements on different equipment

such as the OMA and Raman spectrometers could be accomplished.

All aspects of the measurements with the Lambda 9 were greatly

enhanced by interface to the AT computer. Spectra were stored as ASCII

files which later could be imported into a LOTUS worksheet (Lotus

Development Co., Cambridge, MA) for manipulation and plotting. The

BASIC program listed in Appendix A would accumulate spectra in the form

of a long column of numbers, each number representing the absorption or

transmission at a given data point. The data interval was based on the

slit width used for the analysis. The slit defines the bandpass that

determines the resolution to which features can be resolved. Usually a

data interval of one-third the slit width was used for the analysis.

Slit widths from 4 nm down to 0.5 nm were used depending on the

required resolution. As the slit width is decreased, the signal to

noise ratio is also decreased. This means that the amount of time that

is spent at a given data point must be increased. Computer control of

the spectrophotometer was indispensable for making high resolution runs

where very slow scan speeds were necessary.

Photoluminescence and Raman

Further investigations of the band structure of semiconductors can

be made by measurement of the emission bands or peaks produced by

photoluminescence and resonant Raman scattering. Measurement of these

emissions for this study were made with the two spectrometers described

above. The majority of the measurements, however, were made with the

optical multichannel analyzer (OMA) and a 0.33 meter Instruments SA

monochromater. Most measurements were made at low temperatures, using

the cold finger to cool the sample.

The detection system of the OMA spectrometer consisted of a linear

silicon photodiode array which was 1024 elements wide with the diode

centers spaced every 25 um. Approximately 700 of the center diodes

were intensified to increase their light conversion efficiency. The

diode array was connected to a multichannel analyzer and the array was

scanned by the MCA to generate a spectra. Scan rates as fast as 16.6

msec/scan could be used, although to obtain a higher number of counts,

scan rates of 500 msec were typically used. The computer system

controlling the MCA was setup to accumulate a number of scans so that

signal averaging could be used to reduce noise levels. Also, the dark

current or background noise could be automatically subtracted from a

spectrum. The diode array could be cooled to 5 C with a Peltier

cooler to further reduce the background noise level. With the

monochromater and grating combination the MCA window was about 20 nm.

This is the width of the spectra which was displayed on the screen.

The grating in the monochrometer was rotated to change the wavelength


The third meter monochromater was used with a 2400 lines/mm

grating. The f-stop or light gathering capability of the monochromater

was calculated with


focal length

f-stop = ____________ (4.5)

grating size

where grating size is the linear dimension of the grating. With a 55

mm grating, the f-stop of the monochromater is f6. The reciprocal

linear dispersion (RLD) of a monochromater is a measure of its ability

to disperse light, and can be calculated by

RLD = X dL (4.6)
f de

where d8/dL is the angular dispersion (in rad/mm) and is approximately

equal to the number of grating lines per mm. The RLD for this

monochrometer was calculated to be 1.25 nm/mm. The bandpass or

resolution of a monochromater is usually calculated by multiplying this

number by the width of the slit used. With a linear diode array

detection system, however, the resolution is determined by the spacial

resolution of the array. Assuming that three diodes can resolve a

peak, which correspond to 75 pm, the resolution of this system was 75

Pm X 1.25 nm/mm or 0.9 nm.

A schematic of the experimental setup is shown in Figure 19 and

photographs of the system are shown in Figure 20a and Figure 20b. An

f2 lens was used to focus emitted light into the spectrometer and a 100

mm focal length lens was used to focus the laser beam on the sample.

Typically a focused spot size of 50 um was used, so with an incident

power of 1 mW, the focused power density corresponds to 400 W/cm2. As

Fused silica window

He cryogenic cooler
sample chamber

Laser line filter

Ar laser

Probe-forming lens


Collection lens Slit

To vacuum pump

To optical multichannel analyzer

Schematic diagram of experimental setup used for measuring

Figure 19

Photograph of OMA setup which was illustrated in Figure 19.

Figure 20

,. + I : 0k L+"++
i+.+ .. ... ,'40