|Table of Contents|
Table of Contents
Chapter 1. Introduction
Chapter 2. Conventions and organization
Chapter 3. Background and literature review
Chapter 4. Materials, methods, and equipment
Chapter 5. Results: Pro0cessing parameters of spark-processed silicon
Chapter 6. Results: Microscopy and chemical spectroscopies of spark-processed silicon
Chapter 7. Results: Other spark-processed elements
Chapter 8. Results: Other luminescent spectroscopies
Chapter 9. Results: Miscellaneous
Chapter 10. Further discussion
Chapter 11. Summary and conclusions
Chapter 12. Future work
Appendix A. Electroluminescence and photoconductivity
Appendix B. Calibration of the PL system
Appendix C. Computer programs
PRODUCTION AND NATURE OF HIGHLY LUMINESCENT SPARK-PROCESSED
POROUS OXIDES OF SILICON AND OTHER ELEMENTS
MICHAEL EMERY STORA
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
Michael Emery Stora
Dedicated to my parents, Michelle Monique and Emery Lester Stora
First and foremost, I would like to express my appreciation and heartfelt
thanks to Dr. Hummel who has served as my research advisor, graduate advisor,
supervisor, supervisory committee chairman, and mentor. His enthusiasm, support,
interest, and humor have been invaluable to me. I also wish to thank Dr. Ludwig for
his friendship, creativity, and technical contributions.
I thank Drs. Holloway and DeHoff for both serving on my supervisory
committee and for their influence as professors. It was a pleasure to learn from both
of them as their classes were among the most interesting and productive I have taken.
I graciously thank Drs. Sigmund and Srivastava for their service on my
I am grateful to my current and former coworkers including Nigel Shepherd,
William "Grif" Wise, Jeliazko Polihronov, David Burton, Damien "Cross" Reardon,
Jon Hack, Dr. Sung-Sik Chang. and Dr. David Malone for both their technical
contributions and valued friendship. I also wish to thank Dr. Philip Rack. Dr.
Jonathan Gorrell, Jay Lewis, and Loren Reith, for their friendship and constant
willingness to help.
Dr. Holloway's secretary, Ludie, deserves special recognition for being the de
facto "secretary" for all of us graduate students.
I am grateful to the staff and graduate students of the University of
Rochester's Center for Photo-Induced Charge Transfer, particularly Steve Atherton
and Chris Collison, for the equipment-time and training they so graciously provided,
Dr. Robert Walko at Sandia National Laboratories and Dr. Sey-Shing Sun at Planar
Systems for their help with constructing electroluminescent devices, and Dr. Regina
Miiller-Mach at Agilent Technology for help with photoluminescence excitation
measurements. I also wish to thank Drs. Gerd MUiller and B. C. Blasse for their
interest and helpful advice and Dr. Herbert Ruefer at Wacker-Chemitronic for
supplying the large quantities of silicon wafers used in this study.
My thanks go to all my friends in the UF Kodenkan Jiu-Jitsu Club and UF
Judo Club who are too numerous to list. I owe a great deal to Professors Alex
Limbaugh, Donald Cox, and Bill Beach, and Senseis Margaret Limbaugh, Kevin
Book (deceased), and John Nelson whose lessons of hard work, discipline, and
dedication were as important as any academic lesson.
Thanks to my college roommates and long-time friends Dr. Dmitry Golovko
and "soon-to-be Dr." David Harrison for the friendship and encouragement they have
given me over the years.
Last, but certainly not least, I wish to thank my parents, to whom this study is
dedicated, and my fiancee Heather Rose for the undying love and support they have
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ............................ ................. iv
A BSTRA CT .............................. ......................... viii
1 INTRODUCTION ............................................... 1
2 CONVENTIONS AND ORGANIZATION .......................... 5
U nits of M measure .................... ............................ 5
Definitions and Rules ........................................... 10
Organization of Experimental Results .............................. 13
3 BACKGROUND AND LITERATURE REVIEW.....................14
Review of Spark-Processed Silicon Literature ........................ 14
Other Literature on Spark-Processing and Sparks in General............ 57
Review of Quantum Size-Effect Literature ......................... 61
Review of Defects in Silica and Silicon Oxynitrides ................... 71
Review of an Important Transient Defect in Silica .................... 75
General Review of Luminescence Processes ........................ 93
4 MATERIALS, METHODS, AND EQUIPMENT .................... 107
5 RESULTS: PROCESSING PARAMETERS OF SPARK-PROCESSED
SILICO N .......................................... ......... 119
Electrical, Temporal, and Physical Processing Parameters .............. 119
Chemical Processing Parameters ................................. 180
6 RESULTS: MICROSCOPY AND CHEMICAL SPECTROSCOPIES
OF SPARK-PROCESSED SILICON ............................ 209
Scanning Electron Microscopy (SEM) ............................. 209
Fourier Transform Infrared Spectroscopy (FTIR).................... 215
Secondary-Ion Mass Spectroscopy (SIMS)........................ 220
7 RESULTS: OTHER SPARK-PROCESSED ELEMENTS..............235
Preliminary Exploration ........................................ 235
Additional Work with Selected Elements .......................... 258
Chemical Processing Parameters ................................. 265
8 RESULTS: OTHER LUMINESCENT SPECTROSCOPIES ........... 300
Variable Excitation Power ...................................... 300
High-Temperature Photoluminescent Spectroscopy ...................306
Photoluminescent Excitation Spectroscopy (PLE).................... 326
Time-Resolved Photoluminescent Spectroscopy (TRPL).............. 336
9 RESULTS: MISCELLANEOUS ................................ 367
Useful Results ................................................ 367
Interesting Results ............................................. 371
Uninteresting Results .......................................... 372
10 FURTHER DISCUSSION ...................................... 374
Final Word on Quantum Dots .................................... 374
Integrating Results from Multiple Chapters ........................ 376
11 SUMMARY AND CONCLUSIONS ............................. 389
12 FUTURE WORK ............................................. 393
A ELECTROLUMINESCENCE AND PHOTOCONDUCTIVITY.........395
B CALIBRATION OF THE PL SYSTEM .......................... 402
C COMPUTER PROGRAMS ..................................... 416
REFERENCES ..................................................... 429
BIOGRAPHICAL SKETCH .......................................... 438
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
PRODUCTION AND NATURE OF HIGHLY LUMINESCENT SPARK-
PROCESSED POROUS OXIDES OF SILICON AND OTHER ELEMENTS
Michael Emery Stora
Chairman: Professor Rolf E. Hummel
Major Department: Materials Science and Engineering
Spark-processed silicon (sp-Si) is a novel nano-porous material with unique
optical and magnetic properties. Sp-Si exhibits two broad, intense, room-temperature
photoluminescence (PL) bands, which are located near 370 nm and 530 nm when
excited by 325 nm UV light. These are known as the UV/blue band and the green
band, respectively. In addition, a number of other elements have been show to
exhibit significant PL when spark-processed.
Several PL spectroscopic methods were used to characterize sp-Si and other
spark-processed materials. These included continuous-wave PL (CWPL), time-
resolved PL (TRPL), PL excitation spectroscopy (PLE), and temperature-dependent
PL measurements. In addition scanning electron microscopy (SEM), Fourier
transform infrared spectroscopy (FTIR), and secondary-ion mass spectroscopy
(SIMS) were utilized.
The influences of the chemical, physical, electrical, and temporal processing
parameters on the PL of sp-Si and other spark-processed elements were extensively
explored and characterized. Both PL bands of sp-Si as well as similar bands in sp-
Ge, sp-Al, and sp-Ta were found to be strongly dependent on the presence of nitrogen
in the processing ambient. The PL of sp-Si and several other spark-processed
materials was found to exhibit extremely fast sub-10 ps lifetimes. Longer-lived
components attributed to trapping and evidence of competition between the two PL
bands of sp-Si indicate charge transfer between separate absorption and emission
A model for the PL mechanisms of sp-Si is proposed. This model explains
the observed properties of both PL bands of sp-Si and may be extended to other
spark-processed materials. Absorption of UV light is attributed to a stable version of
a normally metastable defect found in silica, and many other materials, known as the
self-trapped exciton (STE). The rapid melting and solidification during spark-
processing is believed to "quench-in" this defect. PL emission is the result of charge
transfer of carriers from STEs to nitrogen-related deep defect states within the silica
band-gap. The UV/blue band is attributed to defects containing nitrogen and the
green band is attributed to more complicated defects containing two or more nitrogen
atoms, similar in nature to the N-N nearest-neighbor center in GaP.
Silicon has dominated the semiconductor industry for more than half and
century. While other materials are often used for high power switching and
rectification, the role of silicon in the microelectronics industry is of paramount
importance. The vast majority of integrated circuit technology is based on silicon
[Pea88]. Silicon is economical (costing about a tenth as much as gallium arsenide, its
nearest competitor) and easy to manufacture. The technology to dope silicon with a
number of donor and acceptor elements is mature and well behaved. The most
significant advantage of silicon is that it has a stable, adherent, and easily grown
oxide that is epitaxial and self-passivating (especially for the (100) crystal
orientation) [May90]. This oxide may be selectively patterned and etched by
hydroflouric (HF) acid, which crystalline silicon resists. Finally, silicon is non-toxic
and silicon production and processing has a minimal impact on the environment.
This is not true of gallium arsenide. These technical advantages, combined with the
maturity of, and investment in, silicon-based technology make it likely that silicon
will remain the dominant semiconductor material for the foreseeable future.
Despite its advantages, silicon has major limitations. It is both a low band-
gap and indirect band-gap material. It is limited to very inefficient optical emission
in the near-infrared (IR) region of the electromagnetic spectrum. It is poorly suited to
both electro-optical communication and display technology. For these reasons,
gallium arsenide and other III-V (or II-VI) semiconductors and their alloys are
typically used in applications where light emission is necessary, while other non-
semiconductor technologies dominate the display market. Photonics (the integration
of photo-optics and electronics) promises to help overcome basic speed and
bandwidth limitations of electronic communications, not only over long distances, as
common today, but on the inter-chip and intra-chip levels as well. This will only
become economically and technically feasible when a phosphor materials technology
is developed that is completely compatible with silicon-based microelectronics. For
these reasons, the development of silicon-based phosphors has generated much
excitement for a number of years.
In order to be successful, such a material must have a higher effective band-
gap and a much higher radiative efficiency than silicon. The second issue can be
addressed by doping silicon with isoelectronic centers or activators, but this actually
decreases the energy of emission. For a time in the nineties, it appeared that porous
silicon (PS), offered the solution to both issues. PS was discovered by Uhlir lUhl56]
and its efficient room temperature photoluminescence (PL) was reported by Canham
[Can901. While the mechanisms for the luminescence of PS are still controversial,
practical photonic devices have never materialized and the excitement has largely
A new material, known as spark-processed silicon (sp-Si) was discovered by
Hummel and Chang in 1992 IHum92|. It is a highly luminescent porous oxide
produced by an electrical discharge between an anode and a silicon cathode in air or a
mixture of nitrogen and oxygen. While sp-Si and PS have some superficial
similarities, they exhibit different emission wavelengths, emission time constants,
stability under annealing and ultraviolet (UV) exposure, microstructure. and other
important properties [Lud96a, Cha94]. Sp-Si and PS are very different materials with
unique properties which likely result from unique mechanisms.
Despite its efficient PL, useful electroluminescence (EL) from sp-Si has
proven elusive. While strong PL is no guarantee of useful EL (it is not even a
prerequisite), a more detailed understanding of the nature and mechanisms of the PL
of sp-Si may allow the design of more intelligent approaches to generate EL in sp-Si
or a future material derived from sp-Si.
The author has undertaken this study to better understand the nature of sp-Si
and its luminescent properties. This endeavor consists of two main thrusts:
correlating the properties of sp-Si to the conditions under which it was produced and
attempting to glean an understanding of the mechanisms responsible for the
luminescence of sp-Si. The former involves the development of new precision
production techniques to accurately quantify and verify the previously reported
properties of sp-Si as well as to conduct new experiments by taking advantage of the
new degrees of freedom offered by these techniques. The latter involves attempting
to identify the species and mechanisms responsible for the PL of sp-Si by conducting
various experiments to find evidence that may suggest, support, or exclude proposed
In an attempt to further understand the properties of sp-Si. the author
conducted spark-processing on a number of other elements besides silicon. It was
discovered that strong, stable, room temperature PL is not unique to silicon. In fact,
every elemental semiconductor and semi-metal (with the exception of selenium,
which is too flammable to spark-process), as well as a number of metals, exhibit PL
after spark-processing. In addition, the PL of several of these spark-processed
materials was found to be dependent on the presence of nitrogen in the processing
ambient, indicating that a common mechanism may be involved. Thus. spark-
processing of materials other than silicon evolved from a limited exploration to
become an integral part of this study.
The two goals of this endeavor have been met. The processing parameters of
sp-Si and their influence on its properties have been extensively explored. Not only
are the absorption and emission mechanisms of sp-Si well characterized, but the
author is also able to present a comprehensive theory for the luminescence of several
CONVENTIONS AND ORGANIZATION
Units of Measure
Physical Units of Measure
The author uses units of measurement pragmatically. Preference is given to
the International System of Units (SI), or "Metric System," when there is no obvious
alternative. SI-derived units such as the cgs (centimeter-grams-seconds) System are
used when these are more appropriate to the scale of the measurement than SI units.
As all the experiments reported in this study were performed in the United States, the
American version of the Imperial, or "English" system of units cannot be ignored. In
most cases, equipment and supplies were manufactured to US specifications. It is
difficult for the reader to have such measurements arbitrarily converted to SI units
only for the sake of conformity, as some authors do. Because SI and SI-derived units
are the standard for scientific literature, whenever an Imperial unit is used, a Metric
equivalent will follow in parentheses. If this is a scientifically important
measurement, significant digits will be stated and preserved in the conversion. If this
is a nominal measurement, than the metric conversion will be nominal as well. An
example of this is the size of semiconductor wafers, which are usually manufactured
in even-inch diameters. For example: "the 4" (10 cm) silicon wafer was . While
the wafer may very well have been manufactured to very high tolerances (4.00", for
example), 4" is a standard size and terminology used in industry. The metric
conversion is equally nominal (10 cm, not 10.16 cm).
Units of pressure deserve some special consideration. Most American
literature uses pounds per square inch (psi) for pressure above atmosphere and torr
(mtorr, Rtorr) for vacuum. The torr is a SI-derived unit defined as the pressure
necessary to support a 1 mm column of mercury against gravity. However, torr is not
a very common unit in international literature. It is also not particularly practical
(there are 760 torr in an atmosphere (atm)). The SI unit of pressure is the Pascal (Pa),
but this is not particularly practical for vacuum, and not very common in the
literature. This study will use a cgs unit, the bar (mbar. [tbar) for vacuum as well as
pressure above atmosphere, which is very common in international literature. The
[ibar is a dyne cm-2. It is quite convenient and intuitive, as 1 bar is very close to 1
atmosphere (0.9869 atm--within about 1%) and 1 mbar is very close to 3/4 of a torr
(0.7453 torr--within less than 1%). Atmosphere, or atm, in this study always refers to
the SI standard atmosphere and not the technical atmosphere. The technical
atmosphere is defined as one kilogram-force (the weight of one kilogram mass on
earth) per square centimeter. It is essentially equal to the pressure exerted per 10
meter depth of fresh water. While it is a very convenient engineering unit, and
deserves much more acceptance in the opinion of the author, it is relatively unknown
and will not be used in order to avoid misunderstandings.
Unless otherwise stated, absolute vacuum serves as the common reference
point for all pressure measurements in this study, both vacuum and above
atmosphere. For example, 1.5 atmospheres means 0.5 atmospheres above
Optical Units of Measure
Units of optical intensity are quite simple but are the cause of much confusion.
The author uses the following rules.
If no units are given for an "intensity", the number represents uncorrected
instrumental units (typically counts for a photomultiplier or charge-plate detector).
Quantum units represent the number of light quanta (photons) emitted at a
particular wavelength per unit time. These differ from instrumental counts in that
they take into account a wavelength-dependent correction for the sensitivity of the
Radiometric units are units of optical power (watts, for example) at a
particular wavelength. These are related to quantum units by the wavelength
dependence of the energy of a photon (the amount of energy carried by a photon is
inversely proportional to its wavelength). This is rigorously defined as the radiant
flux (intensity integrated over a sphere), but the author will use the term intensity as it
is so often used in the literature. Many sources indicate that the use of watts for
radiant flux is to be deprecated, but this has not stopped their almost universal use.
The alternative is the clumsy ergs per second. There is often a reality gap between
the people who propose standards and those who must use them.
Photopic units are obtained from radiometric units by taking into account the
wavelength-dependent sensitivity of the vision of a "typical" (and gender-less) human
being. The term photometric is more commonly encountered in the literature, but
could refer to either the photopic vision of the retinal cones (which dominates at light
levels above I millilambert (mL)), or the scotopic vision of the retinal rods (which
dominates at low light levels and is more sensitive in the blue). The author prefers
the term photopic to prevent this ambiguity. These units indicate the apparent
brightness of a light source at a given wavelength. Photopic units are derived by
multiplying radiometric units by a correction curve known as "The International
Standard Eye", or more formally, The Relative Luminosity Function of the
Commission Internationale de L'Eclairage (CIE) [Ber76, Sze811. As the sensitivity of
the eye varies with viewing angle, this curve was defined for light entering the eye in
a 2 acceptance angle (other curves exist for larger angles). This curve peaks at 555
nm, which is close to the peak wavelength of sunlight, and resembles a slightly
asymmetrical Lorentzian curve with a greater half-width in the long-wavelength
direction, and a slight shoulder near 450 nm. It has finite values from 380 nm to 770
nm and is defined as zero outside of this range. The proper term for these units is
luminous flux, which is measured in lumens (1m). It is neither an Imperial nor an SI
unit, but was defined as the apparent brightness of 1/680"h of a watt of light at 555 nm.
Luminous flux density can be measured in footcandles (fc), lux (Ix), or phot (ph) in
the Imperial. SI, and cgs systems, respectively. Flux can also be measured per unit
solid angle and per unit solid angle per unit area, with and without factors of 7t, giving
rise to candelas (cd), stilbs (sb), nits (nt), footlamberts (fL), lamberts (L), and
apostilbs (asb). There is no equivalent set of radiometric units for these as people just
use "watts per..." If all this gets confusing, the reader should just keep in mind that
these units were contrived from psychometric measurements made on small groups of
volunteers in the early years of the 20' Century. There is really no underlying
physical meaning behind these units for the reader to understand.
It is common to see statements in the literature referring to intensity (power or
its photopic equivalent) at a particular wavelength. The author has essentially done
the same thing in the proceeding paragraphs. This is rather sloppy, because, except in
the case of a theoretically monochromatic light source, power is actually proportional
to the area under the curve of the intensity vs. photon energy plot. This is usually
referred to as the "integrated intensity" in the literature. In addition, most authors use
the area under the curve of the intensity vs. wavelength plot, which is not really the
same thing. Such imprecisions are common to every field of scientific study and are
generally not problematic if the reader is cognizant of them.
When conducting photoluminescence (PL) measurements, it is generally not
possible or even desirable to know exactly how the measured quantities correspond to
true quantum or radiometric units as long as it is known that they are related by an
unknown linear proportionality constant. This is often due to uncertainty over the
spatial distribution of emitted light as well as arbitrary sensitivity adjustments made
to the instruments (such as the adjustment of optical slits, detector alignment, and
geometry). In such cases, the units of measure are referred to as arbitrary units. If
intensity is measured at different wavelengths, it makes sense to refer to arbitrary
quantum units or arbitrary radiometric units, etc. However, if intensity is measured at
only one wavelength while some other parameter is varied, there is no difference
between these wavelength-dependent arbitrary units (one arbitrary constant is as good
as another). In such cases, only the term "arbitrary units" will be used.
As a final note on optical measurements, it should be pointed out that the peak
location of an optical band may be reported in nanometers (nm) or electron-volts
(eV). These are inversely proportional to each other with a proportionality constant
of 1240 nmeV.
Definitions and Rules
Stages of Apparatus Development
In the experimental chapters of this study, the reader will encounter the terms
Stage I, Stage II, and Stage III. These refer to the degree of sophistication of the
equipment and instrumentation as it evolved during the course of the experimental
work and will be defined in detail in Chapter 4. In summary: Stage I involved spark-
processing with a Tesla coil, DC power supply, or pulsed high voltage supply from a
video monitor. Stage II involved the use of a sophisticated controllable spark-current
supply that was relatively insensitive to changes in experimental parameters. In
addition, the spark current could be crudely measured. Stage III involved the ability
to precisely measure and control the spark current to completely separate it from
other processing variables. In addition, Stage III involved the implementation of
means to prevent nonlinearities in the PL system resulting from chromatic aberrations
and scattering/diffraction at small slit openings. This finally allowed good calibration
curves to be made for the PL system which can be applied to Stage III measurements
with great confidence and to Stage I/II measurements with a fair degree of
Rules for Figures
When multiple PL spectra or peak intensities are presented in one figure, it is
to be assumed that these were measured at identical conditions or that the author has
corrected for varying conditions so that the intensities may be compared with
meaning. When PL spectra are in different figures, it should be assumed that their
intensities are not directly comparable unless otherwise stated. The term
"normalized", when applied to a PL spectrum or other data, means that the largest
value of the dependant variable was set to 1 and that all other data points were scaled
accordingly. The term "self normalized", when applied to multiple spectra or other
data sets, means that each one was individually normalized to 1 with different scaling
factors. In the case of PL spectra, this allows peak shape, width, and location to be
compared independent of peak intensity.
Except in the background and literature review chapter (Chapter 3), when the
intensities, peak intensities, or peak wavelengths of more than one PL band of a
sample or samples are shown in a figure, the shortest peak-wavelength data are
denoted by squares and the longer peak-wavelength data are denoted by circles.
When applicable, the shortest peak-wavelength data points are connected by a solid
line and the longer peak-wavelength data points are connected by a dashed line.
When only one dependant variable is plotted, squares and solid lines are used. In the
case of the UV/blue and green bands (to be defined) of spark-processed silicon (sp-
Si), the UV/blue is represented by squares and solid lines and the green is represented
by circles and dashed lines.
Multiple PL spectra are often presented in figures by two-dimensional (2-D)
projections of three-dimensional (3-D) parameter spaces. In general, the X-axis
represents the processing variable in question, the Y-axis represents wavelength, and
the vertical Z-axis represents intensity. The orientation of the Z-axis is never changed
(increases vertically). However, in order that the reader obtain a clear view of the
data, the azimuthal angle and rotation about the Z-axis vary from figure to figure. In
many cases, this results in the unfortunate but necessary reversal of the direction of
increasing numerical values of one or both of the X and Y-axes. This is better than
having one set of data obscure another and is not a problem if the reader looks closely
at the figure. The projections are made with no perspective (infinite horizon, or
"cubic perspective") and the "back planes" are drawn to assist the eye of the reader.
These 3-D figures may represent two or more samples or measurements per
approximate value of a processing parameter. In these cases, the author is forced to
resort to the use of color so that the reader can clearly differentiate between them.
Highly contrasting solid colors are used. For the case of PL spectra, the spectrum
with the highest peak intensity (of the shortest wavelength band if more than one) is
assigned the color blue followed by red, purple, and green in that order. Even for the
microfilm-reader, who will see shades of gray, this arrangement is superior to the use
of black and white.
In addition, color is used in any graph where two or more dependent variable
data sets overlap in a figure or are otherwise hard to follow with the eye. In the case
of the PL peak intensities of the UV/blue and green bands of sp-Si plotted as a
function of a processing variable, the colors used are blue and green, respectively.
There are three possible standards for the scale and range of a particular
parameter's axis in a figure. The first is a fixed range for all figures, the second is to
scale the axis to the distribution of the data in each figure, and the third is to adjust
the range and scale of the axis to best show the trend of the data that the author
wishes to demonstrate (a smaller range to amplify a small trend, or a larger range to
hide scatter when there is no trend). The author feels that the third approach is
misleading. In the case of the location of the peak intensity of PL bands, the first
approach is impractical, leaving only the second method. Accordingly, in such
figures, the Y-axis is scaled to the distribution of the data.
Organization of Experimental Results
Due to the large number of experimental parameters varied in this study and
the sheer quantity of data to be analyzed, there are several experimental chapters,
each of which may contain its own introduction/motivations, experimental, results,
and discussion sections. It would be unwieldy to attempt this with one chapter for
each where the discussion would be too far removed from most of the corresponding
results (particularly figures). The experimental chapters are arranged in a logical
order and the discussion sections will be cumulative (each may refer to previous
chapters). This is important because the results build on each other and, in many
cases, experiments were designed based on previous results. Following these
experimental chapters is a further discussion chapter (Chapter 10) covering all the
previous chapters as well as the summary and conclusions chapter (Chapter 11) and a
future work chapter (Chapter 12).
Because this is a scientific study with the primary goal of understanding the
nature of the luminescence of spark-processed materials, the author chose to use
separation of variables (scientific method) rather than the Taguchi method. Instead of
a uniform-density parameter space, the author endeavored to find a set of optimal or
near-optimal values to serve as the origin of the parameter space and to fill it with
data points along one axis (parameter) at a time. If it is ever desirable to produce
spark-processed materials for commercial uses, the empty areas of the parameter
space, which represent simultaneous large deviations from the standard values of
multiple variables, may have to be explored.
BACKGROUND AND LITERATURE REVIEW
Review of Spark-Processed Silicon Literature
Hummel and Chang first observed photoluminescence from spark-processed
silicon (sp-Si) in 1992 [Hum92J. The vast majority of literature on sp-Si and spark-
processing in general has been published by Hummel and associated researchers and
students, a group that includes the author. Work by other groups has concentrated on
areas not directly related to this dissertation. These include work on the
electroluminescence of sp-Si IYua95], doping of sp-Si with hydrocarbons [Rut94],
and doping of sp-Si with rare earth elements (Stj971.
Due to the relative novelty of the field and the fact that most of the work on
sp-Si has been done locally, the traditional concept of a "literature review" chapter is
of limited applicability. This section will cover the majority of the work done by
others while reserving most of that done by the author for later chapters. However,
both were conducted concurrently and neither category was done in an intellectual
vacuum. While most of the content to be presented herein is unpublished, work by
others has strongly influenced the author and the author has played a part in the
research to be presented in this section. The following sections will cover work by
other research groups on sp-Si, work in closely related fields, and other work
applicable to the understanding of this presentation.
The majority of work published by Hummel, Ludwig, et al., has been
periodically assembled in large review articles I Hum01. Lud96a. Lud971 and this
background draws heavily on these three publications. The literature will be
presented in the as published state. Much of this information is confirmed by the
author's research, but in some cases it must be modified or even contradicted. Some
of these instances will be alluded to in this background section. Such comments and
criticisms will be indented, single-spaced, and in a bold font to clearly differentiate
them from the review.
Introduction to Spark-Processed Silicon
Spark-processing substantially modifies the properties of materials. Spark-
processed silicon has been shown to exhibit intense room-temperature
photoluminescence (PL), electroluminescence (EL), and cathodoluminescence (CL)
in the visible part of the electromagnetic spectrum as well as in the near UV and near
IR. Sp-Si also exhibits some novel magnetic properties.
When excited with 325 nm light from a He-Cd laser, sp-Si exhibits either or
both of two intense PL peaks. These have been located at approximately 385 nm and
525 nm in the published literature and are known as the UV/blue peak and the green
peak, respectively. Several processing parameters influence the absolute and relative
intensities of these two bands and will be discussed.
Measurements by the author combined with detailed
calibrations of the PL system place these peaks at about 372 nm
and 531 nm when plotted in radiometric (power) units and 373 nm
and 542 nm when plotted in quantum units.
Much has been published about the major differences between the production,
properties, luminescence, and mechanisms of sp-Si and another material known as
porous silicon (PS), which is produced by a wet chemical or electrochemical etch
JCan90J. Electropolishing of Si predates the discovery of luminescence by decades
[Uhl56, Tur58]. There is little in common between these materials, and PS will not
be discussed to any significant degree in this study. Ludwig has dealt with these
distinctions in great detail [Lud96al. It should be pointed out however, that when
excited by 325 nm light, sp-Si exhibits similar photoluminescent intensities (at
significantly different wavelengths) to PS samples purported to have quantum
efficiencies of approximately 5%. The conclusion has been drawn that sp-Si has
similar quantum efficiencies [Hum95a, Hum95b].
Theoretical models to explain above band-gap luminescence in modified
semiconductors with nanoscale features traditionally fall into three main categories:
(1) a classical quantum confinement model [Can90, Bru94, Cal94], (2) a surface-
modified quantum model [Koc93, Cam86], and (3) models based on molecular
species and localized defects (many of which will be discussed in a later section).
The first category of theories predicts a widening of the optical band-gap of
semiconductors due to physical confinement of carriers in quantum wells, wires, or
dots (one, two, or three-dimensional confinement). These models usually use a
classical particle in a box approach IOha90O or an effective mass approximation
approach IBru86, Kay881. In addition, materials with an indirect optical band-gap
(like silicon) can assume a pseudo-direct behavior due to Heisenberg-related
smearing of the momentum of confined carriers ISan921.
The second category also involves broadening of the optical gap of
nanocrystalline semiconductors, but rather than a direct recombination mechanism
across this gap, the enlarged gap hosts optically active tail states associated with the
surface interface of the quantum dot and the surrounding material. The high surface
to volume ratio of nanoscale particles makes such a mechanism plausible.
The third category assumes either the presence of luminescing molecular
species present in the matrix (often complexes of Si, 0, and H known as siloxenes)
[Fuc93, Tam94] or luminescence from highly localized defects in the oxide itself
Production of Spark-Processed Silicon
The method of producing sp-Si samples in the majority of published work has
evolved from the use of a Tesla coil or DC high voltage supply to the use of the
pulsed high voltage output of the flyback coil from a cathode ray tube (CRT). The
spark intensity varies from one CRT to another. The open circuit voltage is
approximately 15 kV. The repetition rate of the high voltage pulses is equal to the
horizontal scan rate of the CRT. This is 16 kHz for a television (525 columns X 30
frames per second) and is about 19 kHz for a VGA monitor. The pulses are applied
between an anode (often a sharp tungsten tip) and the silicon to be processed, which
forms the cathode, see Figure 3-1. They have typically been separated by a gap on
the order of 1 mm. The pulse width created when driving a spark gap with a VGA
monitor has been quoted as being 10 ns or 20 ns |Lud96a, HumOl I. Based on the
assumption that the power supply charges the capacitance of the spark gap (estimated
to be 30 pF) to the full potential of 15 kV after which it discharges across the gap in
10 ns. Ludwig calculated a value of 45 amps for the instantaneous current of each
This analysis is inherently flawed and will be addressed in
Spark-processing is done in air or a mixture of pure gases. As it requires a
discharge, it cannot be done in a vacuum. In air or mixtures of nitrogen and oxygen it
Figure 3-1: Schematic representation of the preparation of spark-processed materials.
High-voltage DC pulses cause a discharge between an anode tip and the substrate
(cathode). Adapted from Lud96b with permission.
results in the build-up of a circular whitish-gray porous oxide layer on the surface of
the silicon. When the power source is a monitor, the growth rate is about 3-5
im/min. and the central region is often surrounded by a light brown ring that is
substantially thinner than the central region (this is not always true of samples
produced by the author, but depends on the processing conditions).
Very little surface modification occurs when the sample is the anode and the
tip the cathode. It is believed that gas molecules are positively ionized and
accelerated toward the cathode transferring kinetic energy to the substrate surface.
When a high-voltage DC potential is applied between two tungsten electrodes in air,
the cathode glows red hot and the anode remains relatively cool. The cathode is
consumed at a much greater rate.
This rapid transfer of kinetic energy melts and/or evaporates the silicon and
ejects a plume of material, which reacts with the ambient gases and recondenses on
the substrate, see Figure 3-2. The deposited material is constantly bombarded by
more ions and more material deposits on top of it. The result is a nanoporous matrix
composed largely of silicon-rich silica with a quantity of embedded silicon
The luminescence has been reported to vary with a number of processing
parameters. The area and thickness of the porous region increases with processing
time. A linear relationship between PL intensity and processing time has been
reported. The resulting PL appears to be unaffected by the doping type, doping level,
or crystalline orientation of the wafer during processing [Cha941.
/ vaporized Si
Si rich vapor
Figure 3-2: Schematic representation of the formation of silicon nanoparticles within a
silicon-rich silica matrix during spark-processing. Adapted from Lud96b with
Morphology of Spark-Processed Silicon
Figure 3-3a shows a plan view scanning electron microscopy (SEM)
micrograph of sp-Si processed with a Tesla coil power source. Figures 3-3c and 3-4
show cross-sections of the substrate and porous region. The porous region consists of
deposited clusters surrounding deep "worm-holes". These holes appear to be
channels through which the spark reaches the more conductive substrate. In this
example, they are about 15 ptm in diameter and 200 [tm in length. Ludwig believed
that these channels represent the volume eroded by a single pulse and has calculated
an evaporation rate of 8 g/s and a current density of 107 A/cm2 [Lud97].
The author believes that these channels result from the
gradual deposition of material around the discharge sites. As the
thickness builds up, the spark is drawn to the least resistive paths
to the substrate and repeatedly strikes areas with less coverage.
This causes growth near these areas, limiting their size. As a
result, long narrow channels form through the porous region. The
author neither believes that the pore volume represents the
amount of material removed by a single discharge, nor that the
pore diameter indicates the diameter of the plasma channel, which
is likely to be highly forked at the substrate end.
Figure 3-3b shows a high resolution transmission electron microscopy (TEM)
micrograph of a sp-Si sample produced using a Tesla coil. Silicon particles with
diameters in the nanometer range can be seen (four are outlined in ink). The
crystalline orientation appears to be random and the particles are surrounded by an
amorphous matrix. Also in the figure are electron diffraction patterns of the silicon-
rich region in the upper right of the figure and of the amorphous region in the lower
left. These confirm that the silicon is crystalline but randomly oriented and the
Figure 3-3: Micrographs of sp-Si:
(a) SEM plan view. Adapted from Hum93 with permission.
(b) TEM. Some nanocrystals have been outlined. Satellite pictures show electron
diffraction patterns taken at locations indicated on the micrograph. CR =
crystalline and AM = amorphous. Adapted from Hum93 with permission.
(c) SEM cross-section at a fresh fracture. Adapted from Hum93 with permission.
Figure 3-3- continued
Figure 3-4: Cross-sectional SEM micrograph of sp-Si taken at a fresh fracture.
Adapted from Lud96a with permission.
matrix is amorphous. The X-ray energy dispersion spectrum of the amorphous region
indicates that it is largely silica (SiO2) [Cha941.
The presence of silicon nanoparticles has been confirmed by the observation
of their expected effect on the Raman spectra (Stokes shift) of sp-Si vs. bulk silicon
[Cha941. Campbell and Fauchet describe how the Raman peak shift and line width
depend on particle size and shape (Cam86]. This is caused by phonon confinement.
In addition, relaxation of momentum conservation of optical phonons in nanoparticles
may cause splitting, resulting in a double Raman peak ITsu92, Wu_95]. Figure 3-5a
shows the Raman peak for bulk silicon at 520.9cm'. Figure 3-5b shows the
corresponding peak of a green luminescing sample that is shifted toward lower
energies and broadened. Figure 3-5c shows the shifted, broadened, and split peak for
a UV/blue luminescing sample. In this case, both single peak and double peak best-
fit Gaussian deconvolutions are superimposed on the data. Both samples were
produced by monitor sparking. The degree of shift and broadening is consistent with
spherical particles of 3-5 nm in diameter ICam86, Sui921.
While nanoparticles are present, there seems to be no correlation between the
size of the nanoparticles and the color or intensity of the PL. A more detailed study
found no significant difference between the Raman shift and broadening between
UV/blue and green sp-Si samples (Rup97]. Furthermore, the less luminescent edge or
"halo" regions of monitor-sparked samples appear to have a smaller particle size than
the center of the porous region, which is highly luminescent. The sizes of the
particles detected in this study, regardless of region, ranged from 8 to 15 nm. These
are much too large to be the cause of either PL band. Finally, the observed Raman
peak shifts and broadenings were not consistent with one another. This is very strong
evidence that quantum size effects are not responsible for the PL of sp-Si.
1000 1 1 1-1-1
'^ 600- \
".': .. .* sy .^
1000 ------------------------ "--
200 (.). -s
460 480 500 520 540
Raman Shift [1/cm]
Figure 3-5: Room temperature Raman shifts (Xee = 488 nm, 400 W cm 2):
(a) crystalline Si 520.9 cm peak. 4.0 cm 'FWHM
(b) green luminescing sp-Si 513.8 cm 'peak, 6.5 cm FWHM
(c) UV/blue luminescing sp-Si 510.9 cm 'peak. 24.0 cm-' FWHM
Single Lorentzian fit shown for all samples (solid line), double Lorentzian fit shown for
(c) (dotted line). Adapted from Hum95a with permission.
Composition of Spark-Processed Silicon
X-ray photoelectron spectroscopy (XPS) has been performed on monitor
sparked sp-Si samples. The core binding energies of the Si 2p electrons have been
found to be 103.4 eV [Kur97]. This is very close to that of silicon in the +4 valence
state as measured in amorphous SiO2 (103.3 eV). This implies that the majority
component of the sp-Si surface is near-stoichiometric silica.
Small-spot XPS has been conducted on the cross-section of a freshly fractured
sample that was processed in air, and reveals that the composition varies with depth
[Lud96b]. The resulting depth profile is shown in Figure 3-6. The concentration of
Si02 is essentially 100% at the surface and falls monotonically as the depth increases
until the substrate is reached. The spark-processed layer in question was 250 Im in
depth. The concentration of crystalline silicon increases monotonically from 0% at
the surface to 100% at the interface with the substrate. The depth profile of two
minor constituents differed. The first represents silicon whose signal is charge
shifted. This must be due to isolated islands or clusters of silicon not connected to the
substrate and surrounded by an insulating matrix. These first appear at a depth of
about 50 [tm, and constitute about 13 at% (atomic percent) of the sample from about
100 tm to 200 Rm. The charge shifted signal finally disappears at the interface with
the substrate. Finally, nitrogen in concentrations as high as 6 at% is found at similar
depths. This signal is also exhausted at the interface with the substrate. Within the
detection limit of the instrument (approximately 1 at%), the surface appears to be
pure SiO,. Depth resolved X-ray emission spectroscopy results are consistent with
only silica to a depth of at least 200 nm (10 keV beam) ILud96bI.
50 100 150 200
Figure 3-6: Composition of sp-Si as a function of depth as derived from small-spot
XPS measurements. Adapted from Lud96b with permission.
Transmission and diffuse reflection Fourier transform infrared spectroscopy
(FTIR) spectra were taken for a monitor-sparked sample in the "fresh" state (Figure 3-
7a), after annealing at 600 C in air for 20 h (Figure 3-7b), after an additional 24 h
anneal at 800 C (Figure 3-7c), and for another sample sparked in a stream of flowing
nitrogen (Figure 3-7d). The first three spectra show strong vibrational modes
associated with Si and 0. These are labeled on the figure. Structures in the 3600
cm-1 to 3750 cm' range assigned to vibrational frequencies of OH groups including
the aSi-O-H group (silanol) are present and are shaded in the figure. It is clearly seen
that the strength of these modes decreases with annealing. OH groups are known to
desorb at temperatures above 450 C. A number of hydrogen related vibrational
modes that are seen in porous silicon produced by a wet etching process are not seen
in sp-Si. These include the scissors and wagging modes of the Si-H bond at 910 cm'
and 630-670 cm ', respectively, as well as a Si-O-H mode at 870 cm-'. A strong N2
related vibrational mode is seen in all four spectra indicating that nitrogen is
incorporated in sp-Si processed in either air or nitrogen.
Room temperature photoluminescence
Sp-Si prepared in air using a monitor as a power supply exhibits two primary
photoluminescence (PL) bands under 325 nm excitation. A peak maximum near 385
nm is known as the UV/blue band and a peak near 525 nm is known as the green
band. These are shown in Figure 3-8a and 3-8b. A tail extending into the red can be
seen in samples exhibiting only the UV/blue peak. This peak can be resolved into a
discrete peak near 650 nm when sp-Si is excited by the 488 nm or 514.5 nm line of an
^ .. z (C) 6 I. ,: .-"
I. -.. ::.I-.
i (a) : (a)
i I / i/ I,
4000 3500 3000 1500 1000 500
Figure 3-7: FTIR spectra of:
(a) freshly sparked sp-Si produced in air
(b) same sample annealed for 20 h at 800 C in air
(c) same sample after an additional 24 h anneal at 800 C in air
(d) sp-Si produced in UHP nitrogen (99.999% pure)
Adapted from Hum95a with permission.
300 mW/cm2 2-
300 400 500 600 700
Figure 3-8: Room-temperature PL spectra of: (325 nm 0.3 W cm2 excitation)
(a) UV/blue luminescing sp-Si
(b) green luminescing sp-Si
(c) red PL of PS shown for comparison
Adapted from Lud97 with permission.
Argon ion laser as well as when PL is measured at high temperatures. Hummel has
assigned this luminescence to a specific radiative transition rather than to an extended
distribution of states [HumO1I. The peak intensities of the UV/blue and green peaks
are similar. When excited by about 5 W of 488 nm laser light, the red peak has
similar intensities to the other peaks when excited by about 20 mW of 325 nm laser
light. This indicates a room temperature quantum efficiency ratio of about 1:250. It
is possible that the tail also exists in green samples, but it cannot be individually
The relative proportions of the UV/blue and green peaks can be changed by
varying several processing parameters. A more intense spark produces more green
and less blue. A large heat-sink under the sample produces more blue. Blowing a
stream of air on the sample during processing almost guarantees a blue emitting
sample. Heating the substrate during processing enhances the green at the expense of
the blue. It is likely that temperature is the common factor in all these observations.
While a sample can exhibit a superposition of the UV/blue and green peaks, there is
never a peak in between or a shift in the peak locations beyond that expected by the
mathematical addition of the two curves.
The PL peak of an anodically etched porous silicon sample purported to have
5% external quantum efficiency is shown in Figure 3-8c. This was measured under
identical conditions as the sp-Si peaks. The intensity of PS is comparable to both
UV/blue and green sp-Si leading to the published assertion that the quantum
efficiency of sp-Si is in the neighborhood of 5%.
The methods reported in this dissertation allow the creation
of UV/blue sp-Si several times brighter than samples of this
vintage. For the green PL, they can be over an order of magnitude
brighter. However, when the author attempted to make a direct
measurement of the quantum efficiency of a typical green sample,
the result was only about 1/3%. The brightest green samples are
about 5 times more intense than a typical green sample, which
indicates a quantum efficiency of approximately 1 -2%. This is
still quite high, and due to the methodology of the measurements,
is probably a conservative or "worst case" value. This issue is
addressed in Chapter 9.
Both major PL peaks vary linearly in intensity with excitation powers varying
from 0.01 mWcm2 to 10 mWcm2. The spectral shape is also preserved.
These PL bands of sp-Si cannot be accounted for by currently known PL
bands in stoichiometric and non-stoichiometric silica and other glasses. In Figure 3-
9, the primary peaks of sp-Si are represented by curves (a) and (b). Also shown on
this graph are a number of silica-related peaks. These samples were measured under
identical conditions and include: (c) X-ray damaged borosilicate glass exposed to 40
keV X-rays, (d) 99.999% pure silica powder, (e) Silicon-rich silica created by Si ion-
implantation (25 keV, 2 X 101 cm2), and finally, (f) industrial grade fused silica.
Within the high bandgap of silica (9 eV for crystalline silica, about 8 eV for dense
amorphous thermal silica) there are reported luminescent bands at 185 nm, 264 nm,
288 nm, 400 nm, 563 nm, and 650 nm |Fri85, Pio9O, Sta871. These have been
assigned to defect states as they have been found to increase in PL intensity under X-
ray, gamma, or neutron irradiation of silica. These bands have been successfully
excited by below-gap photon energies. 165 nm, 240 nm. 258 nm, and 620 nm
absorption bands have been found in silica. Photon energies over 5.0 eV (248 nm)
are reportedly necessary to activate luminescence at 264 nm, and 288 nm. These are
associated with triplet-to-singlet and singlet-to-singlet transitions of electrons in Si-Si
bonds across an oxygen vacancy |Toh89b, Jon76]. The 460 nm luminescence is seen
in Si-rich silica containing a low concentration of OH groups. The 650 nm emission
350 (a) 325 nm
3- ( 300 K
/ \ (b)
300 / \
l \ /--
I \ / \
IL 10 0
100 /2 ) \
50 / / \ \
100 / \- \ V -
O i ^ i < i i ^ i
350 400 450 500 550 600 650 700
Figure 3-9: PL spectra of various samples measured under identical conditions:
(a) UV/blue luminescing sp-Si
(b) green luminescing sp-Si
(c) X-ray damaged borosilicate glass
(d) SiO2 powder (99.999% pure)
(e) Si ion-implanted fused SiO2
(f) fused SiO,
Numbers indicate multiplication factors. Adapted from Hum95a with permission.
band has been attributed to a non-bridging oxygen hole center (NBOHC) [Sta87,
Gri91] or an oxygen interstitial [Lud95a, Sku84, Pro94, Shi94]. It is associated with
the 258 nm and 620 nm absorption and is unique among defects in silica due to its
small relaxation (most others have Stokes shifts of 2 eV or more). This band is
rapidly degraded under UV excitation [Hum95b].
It appears that the existing literature on silica cannot explain the main PL
peaks of sp-Si. Also, He-Cd laser radiation (325 nm) is well below the peak
absorption energy associated with all of the known emission bands except the one at
650 nm. Finally, under 325 nm excitation, the PL of monitor produced sp-Si is at least
three or four orders of magnitude more intense than that from these various defects in
silica and glasses.
Thermal dependence of the PL of sp-Si
The PL spectra of monitor-sparked sp-Si samples were measured at a variety
of temperatures. Samples with UV/blue or green luminescence lose intensity above
room temperature and gain intensity below room temperature, see Figure 3-10. The
peak location and shape of the UV/blue luminescence is unchanged, while the green
peak is blue-shifted and broadened by increasing temperature. As in the case of
lower energy excitation, the red tail of UV/blue sp-Si can be resolved as a separate
peak at high temperature. This band appears to be less susceptible to thermal
quenching than the main part of the spectrum. The optical band-gaps of Si. Si02, and
silicon quantum dots all decrease (red shift) with increasing temperature |Blu74,
Thu751. This is not observed in either primary peak. The blue shift of the green peak
with increasing temperatures resembles the behavior of some silicon oxynitrides and
400 500 600 400 500 600
Figure 3-10: Temperature dependence of the PL spectra of:
(a) UV/blue luminescing sp-Si
(b) green luminescing sp-Si
Measurement temperatures are indicated adjacent to curves. Adapted from Lud96b with
non-stoichiometric silicon nitride excited by below band-gap energies [Aus60,
The integrated PL intensity is plotted as a function of inverse temperature in
Figure 3-1 la. At temperatures as low as 25 K, the PL intensity of sp-Si is 10 to 20
times more intense than at room temperature. Above room temperature, the PL is
thermally quenched with activation energies of 156 meV for the UV/blue peak and
241 meV for the green peak. These values indicate a high degree of localization. The
scale of this localization can be estimated by assuming that these activation energies
correspond to excitonic binding energies. Using a simple electrostatic potential
model (AE = (47tEn0) 'e2d1) and an average of the two values, a carrier separation
distance, d, can be estimated to be 0.6 nm in Si (c=12) and 1.9 nm in SiO2 (E=3.9).
The lattice constant of Si is 0.543 nm by comparison. There are a number of ways to
trap and localize defects besides classical excitons, but this exercise demonstrates a
very high degree of localization, nevertheless.
Below room temperature, the activation energies vary with temperature. Such
effects are generally attributed to the thermal emptying of exponential band-tails or
other localized states that are inhomogeneously broadened. Such a phenomenon has
been seen in band-tail to band-tail luminescence in a-Si:H ICol801. In such cases, the
intensity varies with temperature according to the equation I(T) = 1 / (1 + ceTr'),
where c is a constant. T, is the characteristic temperature, which is a measure of the
disorder in the system. It is related to the degree of inhomogeneous broadening and is
proportional to the "width" of the density of band-tail states. If Io / I(T) 1 is plotted
logarithmically vs. temperature then a straight line should result according to the
theory. To can be determined from the slope of the line. As seen in Figure 3-11 b, the
thermal data for sp-Si fits this model very closely. In its simplest form, this model
200 100 50 33
0 10 20 30 40
1000 /Temperature [1/K]
T. 81 K 101
r ^? \ 100 6
^^ blue 10
T =96 K
I I I I
0 100 200 300 400 500
Figure 3-11: Temperature dependence of sp-Si PL:
(a) Temperature dependence of the integrated PL intensity of UV/blue and green
luminescing sp-Si. Activation energies of thermal quenching are shown.
(b) Plot of II/I(T)| 1 vs. temperature for the same data. Characteristic
temperatures are shown (see text).
Adapted from Aug96 with permission.
also predicts a redshift with increasing temperature as the band-tail states closest to
the mobility edges are thermalized first. At higher temperatures, the remaining
occupied band-tail states lie further in the gap and have less potential energy.
However, real systems can be more complex. An inhomogeneously broadened state
might not lie near a mobility edge and might depopulate by a different mechanism. In
the case of SiNx, a red shift is seen at small values of X, a blue shift at high X, and no
shift at all at some intermediate X value. Thus, neither the lack of a shift in the case
of the UV/blue peak, nor the blue shift of the green peak can be taken as a significant
contradiction of the theory. The excellent fit of the integrated intensities indicates
that either exponential band-tails or exponentially broadened interband states are
involved in both species of luminescence. Measured values of the characteristic
temperature of the PL of sp-Si have varied from 70 K to 110 K. This indicates a
higher degree of disorder and band broadening than in PS where To is in the range of
40 K to 70 K [Ros93, Mus93]. It should also be noted that in PS, PL intensity is
actually decreased at very low temperatures and has a maximum between 100 K and
200 K [Zhe92, Per92]. Thus, luminescence in PS is thermally activated. In sp-Si,
thermal quenching predominates at all temperatures measured.
Figure 3-1 la is actually highly misleading. By putting straight lines and the
activation energies on this Arrhenius-plot, Augustin strongly implies that these
energies were derived from a slope in this I(T) vs. I/T plot. Correspondence with
Ludwig verified that these energies are in fact derived from the slope of an 10 / I(T) -
I vs. 1/T plot (not shown). This treatment is a consequence of the following theory
for the change in PL intensity during thermal quenching: I(T) = 1 [(Tr') / (T'- + Tn')],
where xTr and T, are the time constants of the radiative process and non-radiative
processes. In processes that are not thermally activated, T, usually is relatively
insensitive to temperature and t, is strongly temperature dependent. Some
manipulation of the previous equation yields: (Tn' / T,-') = I, / I(T) 1. The ratio on
the left can be approximated by an exponential giving: exp(-Ea/KbT) I / I(T) 1,
where Ea is the activation energy for the non-radiative process. This model is much
simpler than the treatment of Collins and is often not sufficient for amorphous or
disordered materials [Col801. However, at sufficiently high temperatures, this
thermally activated quenching usually predominates and Ea approaches a constant
Stability of sp-Si
Sp-Si is highly stable against degradation by UV light exposure, thermal
annealing, and HF acid etching. Figure 3-12 shows how the PL intensity of sp-Si
compares to that of PS with continued 325 nm UV excitation. Curves (a) and (b)
represent typical sp-Si samples. Curve (c) represents a sp-Si sample annealed in 900
C air for three hours. It is even more stable than freshly produced sp-Si. Curve (d)
represents a sp-Si sample produced in UHP nitrogen. Curves (e) and (f) represent
two typical PS samples. All curves are self-normalized to an intensity of I at zero
time. The difference between sp-Si and PS is stunning, especially after the
logarithmic intensity axis is taken into account.
Figure 3-13a shows how the PL intensity of sp-Si is influenced by annealing
in ultrapure nitrogen. For each data point, the same sample was introduced to the
furnace at the indicated temperature and held for 30 minutes. It was then removed,
allowed to cool to room temperature, and the PL spectrum was measured. Each
successive data point from low to high temperature indicates several accumulative
annealings at the indicated temperatures. The green peak is relatively stable until 600
\ ~(a) \
V \ (b)
: ~~~(f) ^ ^ ^ ^
325 nm, 300 mW/cm2 por-Si
0 5 10 15 20
Figure 3-12: Degradation of PL as a function of time of exposure to 325 nm 0.3 W cm2
(a) UV/blue luminescing sp-Si
(b) green luminescing sp-Si
(c) UV/blue luminescing sp-Si annealed in air for 3 h at 900 C
(d) sp-Si produced in UHP nitrogen (99.999% pure)
(e) and (f) porous Si samples for comparison
Adapted from Hum 95b with permission.
400 800 1200 0
Annealing Temperature rC]
400 800 1200
Annealing Temperature [C]
Figure 3-13: Annealing characteristics of sp-Si (consecutive anneals of the same
sample in UHP nitrogen (99.999% pure) for 30 min. at each temperature):
(a) PL peak intensities of UV/blue and green PL bands
(b) PL peak positions of UV/blue and green PL bands
All data was taken at room temperature with 325 nm excitation. Adapted from Hum95a
-* *eep* ~e ** *-
C and begins to increase with annealing, tripling in intensity by 1,100 C. The
UV/blue peak remains relatively stable until 900 C when it begins to decrease
somewhat, loosing less than half its intensity by 1,100 C. The apparent peak
locations of both bands are independent of annealing temperatures as seen in Figure
3-13b (410 nm and 520 nm are instrumental locations of the peaks before correction).
The apparent increase in PL around 400 C has been attributed to quenching of non-
radiative centers from "naked" silanol groups. This is seen in the 460 nm defect
luminescence in silica, see Figure 3-14. The silica sample was boiled in pure
deionized water for 5 hours to saturate dangling silicon bonds with -OH groups.
From room temperature to about 110 C, the surface of silica is believed to be
covered with chemisorbed -OH groups which are in turn covered by physisorbed
H20. This combination does not provide good surface passivation. The H20 desorbs
between 200-250 C leaving an -OH terminated surface that provides excellent
passivation. Between about 450-600 C, the -OH groups desorb, and the passivation
is lost. Thus, similar peaks in the annealing curves are seen between 400 C and 450
C for the UV/blue and green peaks of sp-Si as well as the much less intense defect-
related luminescence in silica. The double line in the figure indicates the temperature
above which there should be no chemisorbed -OH groups. It is believed that silanol
groups enhance the luminescence of sp-Si within a narrow temperature range, but are
not directly responsible for either luminescent band. It should be noted that no
difference was seen for sp-Si after boiling in water vs. a standard air-sparked sp-Si
sample, indicating that dangling bonds are already passivated in sp-Si prepared in
damp Florida air.
The PL from sp-Si has been found to be remarkably stable against HF etching.
HF etches SiO, quickly, indicating that the excess silicon and nitrogen in the bulk
0 200 400 600
Annealing Temperature [oC]
Figure 3-14: PL peak intensities of 460 nm defect luminescence in silica after
consecutive anneals of the same sample in UHP nitrogen (99.999% pure) for 30 min. at
each temperature. Adapted from Hum95c with permission.
slow etching substantially [Cha94]. Because of this, it has been concluded that the
PL of sp-Si is a bulk, and not surface, phenomenon.
PL lifetimes of sp-Si
Nanosecond PL decay was originally reported for the PL band of sp-Si. More
recent measurements by the author have revealed much faster picosecond
luminescence using equipment with a much faster time resolution. This will be
discussed in detail in Chapter 8.
However, the original published conclusions about the speed of these
processes in sp-Si are still true (perhaps more so) and will be presented in this review.
The fast response suggests that recombination is based on geminate carriers
recombining at highly localized luminescent centers.
The ps decay of sp-Si is much faster than red-luminescing PS [Ook921 and
defects in a-SiO, [Toh89cj, which are both in the 1 10 .us range. It is also three
orders of magnitude faster than the 10 ns decay observed in blue-luminescing highly
oxidized PS IKoc93, Kov94]. The 1.9, 2.2, 2.7, and 3.1 eV transitions in silica have
reported time constants on the order of 10 us, 100 ns, 10 ms, and 100 1us, respectively
[Sta87, Nis92, Sku78, Sku84, Sku92]. All of these mechanisms are excluded as
likely candidates to explain the UV/blue and green PL of sp-Si.
Effects of processing parameters on luminescence
The intensity and color of the PL emission from sp-Si depend on several
experimental parameters including the pressure of the gas ambient, the total
processing time, and the substrate temperature. Two of the most important
parameters are the spark current and the chemical composition of the gas ambient.
Figure 3-15 shows how the UV/blue and green PL intensity change with gas
20 40 60 80
Oxygen Content [Vol %]
Figure 3-15: Peak intensities of UV/blue and green PL bands as a function of the
composition of the gas ambient during processing. X-axis shows the percentage of UHP
oxygen. The balance is UHP nitrogen. Adapted from Hum98 with permission.
composition between pure N2 and pure 02. Both gases must be present during
processing for either of these dominant bands to exist in a sample [Hum98]. Samples
sparked in pure 0O exhibit only an orange/red peak at 1.9 eV (650 nm) which is 4-6
orders of magnitude smaller in intensity than the PL from a typical sp-Si sample.
This peak has been attributed to nonbridging oxygen hole center (NBOHC) defects in
silica. The samples represented by this figure were spark-processed with a monitor as
a power source, which is essentially a constant-voltage supply. Oxygen is more
readily ionized than nitrogen. It is clear to the casual observer that as the nitrogen
concentration is increased the spark intensity rapidly decreases. Thus, without the
luxury of a constant current source, this experiment cannot differentiate between two
of the most important parameters (gas composition and current). With such a
constant-voltage supply, the intensity of both peaks is maximized at N2:02 ratios near
1:1. While nitrogen has been shown to passivate non-radiative centers in Si [Luc95],
it appears to play a more direct role in the luminescence of sp-Si.
The PL intensity of sp-Si increases with processing time, see Figure 3-16
[Aug96]. This can be explained by the continuous production of luminescing centers.
This increase is not accompanied by any change in the peak location. If the
luminescence were due to quantum dots of continuously decreasing size, one would
expect both a blue shift and nonlinear intensity increase. According to one model
[Hyb941 quantum dot optical transition probability (and thus, the efficiency) varies
with diameter by 1/d6.
As previously mentioned, heating the substrate is known to increase the
intensity of the green peak at the expense of the UV/blue peak. Figure 3-17 shows
the intensity of the green peak as a function of substrate temperature. Without a
constant-current supply, increasing substrate temperature can increase the current as
Figure 3-16: Intensity of the UV/blue PL peak of sp-Si as a function of processing
duration. Adapted from Aug96 with permission.
I II I I I
I I I
Substrate Temperature [C]
Figure 3-17: Intensity of the green PL peak of sp-Si as a function of the substrate
temperature during processing. Adapted from Aug96 with permission.
well because the intrinsic carrier concentration of the silicon increases. However,
because of the large relative increase of the green peak compared to the UV/blue
peak, there is definitely evidence of a thermal component to this effect.
If the spark-processing is done in flowing air, the intensity of the UJV/blue
peak is observed to increase with increasing air velocity. Figure 3-18 shows the
increase in PL intensity as a function of the air pressure applied to a nozzle blowing
on the sample during spark-processing. Such "spray cooling" is known to be a highly
efficient method of cooling a small area, and this effect is probably thermal in nature.
It is not due to additional oxygen from the flowing air. X-ray emission measurements
indicate that sp-Si processed in flowing air is less oxidized than that processed in
stagnant air, a somewhat counterintuitive observation [Kur97I.
As briefly mentioned, variations in the substrate including the type and
concentration of dopants (n-type and p-type) do not effect the peak locations of the
PL bands in sp-Si. However, a substantial difference in current can result.
Tungsten has proven to have excellent properties for use as an anode material
for spark processing, and it is often used for this reason. This led to the argument that
the UV/blue luminescence of sp-Si originates from tungsten contamination IVep971.
The luminescence of sp-Si sparked using a variety of anode materials exhibited
identical peak locations and similar PL intensities [Hum98I. The anode materials
included W. Ti, Mo, V, Ni, and, most importantly, Si. Some electrodes oxidized more
easily than others, and some exhibited an adherent insulating oxide which "choked
off" the spark. Differences in current can explain the observed differences in
intensity. These experiments conclusively rule out the role of contamination from the
anode as a mechanism for the UV/blue and green peaks of sp-Si.
10 20 30 40 50
Air Pressure [Ibs/sq.in.]
Figure 3-18: Peak intensity of the UV/blue band of sp-Si as a function of the air
pressure used for spray cooling of the area being processed. Adapted from Aug96 with
Cathodoluminescence of sp-Si
Sp-Si is known to exhibit two cathodoluminesence (CL) peaks near 480 nm
and 650 nm [Lud95al. This luminescence is commonly observed in silica and glasses
and is easily explained. It appears to have no relation to the PL of sp-Si. It is beyond
the scope of this study to go into the CL in detail, but it should be pointed out that the
penetration depth of electrons (10 KeV was used for the CL work) is only a few [tm,
while silica is nearly transparent to 325 nm UV light. CL is therefore believed to be a
surface (or near-surface) phenomenon, while PL is a bulk phenomenon.
Electroluminescence of sp-Si
Some work has been done by the author on electroluminescence of sp-Si and
will be covered in Appendix A. In brief, thin film AC EL devices were found to be
very inefficient and produce a faint greenish luminescence. No meaningful spectra
could be taken of such a dim source. Roughly half (apparently the younger half) of
people asked to observe it could see it in a darkroom. Thick film (high frequency)
AC devices based on sp-Si in an oil dielectric worked significantly better (intense
enough to measure) but were short lived. The EL spectrum in this case is very similar
to the CL spectrum. It should also be noted that a much more intense and stable
broad red/IR luminescence has recently been observed by Nigel Shepherd. a member
of the same research group as the author, in a DC device based on metal contacts over
an extremely thin spark-processed layer (sparked for only seconds). Previously,
Yuan and Haneman observed visible PL in a similar system [Yua95]. However, as
the author was not involved in this work and the EL peaks appear totally unrelated to
the PL, these results will not be discussed here.
Doping sp-Si with rare-earth elements
Sp-Si has been doped with highly luminescent Ce, Eu, and Tb salts [HumOl].
This is also beyond the scope of this study except to note that the PL of these narrow
line emitting elements becomes very broad when they are dispersed in a sp-Si matrix.
The emission of Eu in sp-Si:Eu has a full-width at half-maximum of 200 nm. This
demonstrates that the disorder in sp-Si is great enough to inhomogeneously broaden a
very specific state into a wide PL band. Thus, a distribution of states is not needed to
explain the broad PL bands in undoped sp-Si.
Spark-processing of materials other than silicon
Spark-processing leads to visible PL in a large variety of materials including
Ge, Sb, Bi, Sn, Te, As, and GaAs ILud94, Lud95b]. With the exception of GaAs, this
was the work of the author and will be covered in the experimental section.
However, certain aspects of sp-Ge and sp-Zn have also been studied by others,
particularly Chang [ChaOOa, ChaOOb I.
Three deconvoluted PL peaks are reported in air-sparked sp-Ge near 420 nm
(blue), 520 nm (green), and 620 nm IChaOOa/Figure 3-19]. The author consistently
observes the first two peaks in sp-Ge but never the third. Chang sparked primarily in
air, while the majority of the author's work was in mixtures of UHP N2 and UHP 02,
which may explain the difference. The overall PL intensity increases at low
temperatures (as low as 21 K) and the increase is greater for the blue peak than the
green peak. Furthermore, the blue band of sp-Ge also maintains its peak location at
all temperatures (as does the yellow 620 nm band). This thermal behavior is very
similar to that of sp-Si.
**b // l
II I ,
^~~~~ #- /^^
400 450 500 550 600 650 700
Figure 3-19: PL spectrum of sp-Ge (solid line) and Gaussian deconvolution (dashed
lines) of same into three distinct peaks. Adapted from ChaOOb with permission.
Several spark-processed metals exhibit visible PL and will be discussed in this
study. With the exception of recent conference proceedings [StoOO], only sp-Zn has
been covered in the literature [ChaOOb]. Unlike sp-Si and other sp-materials
including some metals, the PL of sp-Zn can be readily explained. A UV/blue 380 nm
PL peak is attributed to a direct band-edge transition of ZnO [Sze81], while a broad
green 540 nm PL peak is attributed to any of many well-documented non-
stoichiometric defects in ZnO [Van96, Pro95, Xu_96, Byl78, Ort97. Liu92].
Laser-processing of silicon
Pulsed laser ablation of silicon in an atmosphere containing nitrogen and
oxygen (Ip-Si) results in PL similar to, but substantially weaker than, the UV/blue
peak that results from spark-processing [HumOO]. Lp-Si exhibits no green peak
analogous to that of sp-Si.
Magnetic Properties of Sp-Si
Sp-Si exhibits unique magnetic properties that can best be described as being
strongly paramagnetic with a narrow, but clearly discernable. ferromagnetic
hysteresis loop IHac971. It is likely that spark-processing produces such a high
concentration of paramagnetic centers that their unpaired electrons are close enough
to couple, giving rise to a ferromagnetic exchange energy. Electron Paramagnetic
Resonance (EPR) studies indicate the presence of approximately 3.8 x 1020 unpaired
spins/cm3 associated with the E' center in silica. The E' center is one of the major
sources of unpaired spins in silica. The remanent magnetization (a measure of the
ferromagnetic behavior), the total magnetization (a measure of the paramagnetic
behavior) and the strength of the EPR signal (a measure of unpaired spin density)
correlate to one another and all decrease rapidly with annealing above 600 C. This
magnetic behavior appears to be independent of the PL mechanism, which is not
strongly effected by annealing in this temperature range.
The PL intensity of the UV/blue peak increases slightly when the PL is
measured in a magnetic field as high as 2 x 104 Oe. When the field is reduced, a
permanent reduction in intensity of around 6% is observed. This trend continues
when the field is reversed, see Figure 3-20 IHumOOl. This phenomenon is not easily
explained, but it is interesting to note that it does not occur to the green peak of sp-Si.
This is further evidence that the mechanisms for the two peaks are different.
No peak splitting is observed in the PL bands of sp-Si at high magnetic fields,
but it is unlikely that this effect could be observed even if present given the broad
nature of these bands.
Finally, it should be noted that once a sample is exposed to a magnetic field
over 104 Oe, it no longer displays a ferromagnetic hysteresis loop.
Other Literature on Spark-Processing and Sparks in General
Three papers on spark-processing and spark erosion not published by
researchers at the University of Florida are of interest and are discussed in this
section. St. John performed spark processing on silicon covered by erbium nitrate
salts IStj97]. The erbium was incorporated into the sp-Si structure and found to be
coordinated to oxygen and exhibit broadened PL. These results are consistent with
those discussed in the previous section.
Hsu reported the formation of titanium-rich titanium carbide nanoparticles
produced by spark erosion of titanium electrodes in an organic solvent [Hsu95I. The
reaction between Ti and C is thought to occur in the vapor phase. A bimodal size
....... I I 1 .1 1 1 1 1 1 1 ....... a.. -I T ...... M ......... Il...... 111 l11
0.0 0.5 1.0 1.5 2.0 x104
Magnetic field strength (Oe)
Figure 3-20: Evolution of the peak intensity of the UV/blue PL band of sp-Si with
exposure to a varying magnetic field. Adapted from HumOO with permission.
-2.0 -1.5 -1.0 -0.5
..t........., ......... .........l.... 1.....t.........t......... ......... ..... ..........
" - ... .. .... .... .... I
distribution of particles was found. 5-50 nm particles were attributed to the rapid
condensation of TiC and Ti vapor while larger 5-20 rtm particles were attributed to
solidification of remelted TiC:Ti.
Zhang reported on a novel method of performing spark erosion by electrical
discharge machining (EDM) with a constant voltage supply [Zha97]. An ultrasonic
repetition rate of sparks was achieved by vibrating the cutting electrode at ultrasonic
frequency. While this paper belongs to a very different field than spark-processing, a
thorough explanation of the processes that occur in a spark discharge is discussed in
If a potential below the breakdown voltage is applied between two electrodes
in air, a tiny current will flow. The conductivity is provided by gas molecules that are
ionized by various forms of ionizing radiation including UV light and cosmic rays.
This current is stable if the potential is unchanged and is sometimes referred to as the
If the potential is increased further, a halo-discharge occurs. Here the current
is large enough to cause some additional ionization. However, even at this stage,
there is not enough current to create a specific ionization channel and the halo
expands in a diffuse manner.
If the potential is increased further still, the halo-discharge becomes a self-
continuous discharge. The production rate of charged particles overcomes the de-
ionization rate and the number of ionized particles increases dramatically with time.
This rapid increase in current is known as the spark discharge phase. The entire spark
discharge phase lasts about 107 to 105 seconds. During this short period, a large
amount of energy is released resulting in highly localized heating. This creates a
tremendous overpressure. Typical temperatures and pressures are on the order of
10,000 K and hundreds of atmospheres. A shock wave travels away from the spark
column through the surrounding gas.
When the number of ionized particles again reaches equilibrium, the final
stage is reached. This is known as the arc discharge phase and is stable as long as the
potential is maintained. In this phase, many ionizations and recombinations occur in
equilibrium and the current and total number of ions are constant. The arc can last a
long time and results in significant melting and/or evaporation of the cathode surface
(the heavy ions striking the cathode carry much more momentum than the electrons
striking the anode). When the potential is reduced below the point necessary to
maintain an arc, the hot column of gas rapidly cools resulting in a sudden reduction of
pressure to near vacuum. This, in turn, collapses forming another shock wave.
If a series of spark and arc discharges occur repeatedly, a large amount of
melted and vaporized material can be produced. The free material is drawn into the
gap by the vacuum associated with the end of the arc discharge and blown outward by
the shock wave associated with the spark discharge. At the same time, ions are being
produced and the cathode is being melted/evaporated (mostly in the arc phase). This
repetition results in a continuous mixing and remixing of molecules from the gas
phase and the cathode in an environment with an abundance of available thermal and
electrical energy to drive chemical reactions.
Understanding the nature of sparks and how they develop and progress in time
provides the necessary background needed to interpret the microstructure of sp-Si and
the effects of changing the spark-processing parameters on the PL of sp-Si. These
processing parameter variations will be presented in detail in Chapter 5 and newer
SEM micrographs of sp-Si will be presented in Chapter 6.
Review of Quantum Size-Effect Literature
While a very strong case against the involvement of quantum confinement or
"quantum dots" in the PL of sp-Si has been made in the literature, the topic deserves a
closer second look. if only for the reason that nanoscale silicon structures are present
in the material.
Radiative Lifetimes in Quantum Systems
The radiative lifetimes of sp-Si will be discussed in Chapter 8. It is difficult to
compare the lifetime of sp-Si to published lifetimes of quantum structures, because
there is substantial disagreement among the published lifetimes for both experimental
and theoretical treatments of quantum systems.
Beginning with the experimental papers: Bimberg observed lifetimes of
"dozens of picoseconds" in GaAs/AIGaAs and InAs/GaAs quantum dots |Bim97]. It
should be pointed out that these materials already exhibit efficient direct PL in the
bulk without quantum effects and the PL was only in the range of 1.1 to 1.3 eV even
for dot diameters as small as 0.5 to 2.5 nm.
Linnros prepared Si nanoparticles in a Si02 matrix by ion-implantation of Si
into silica (Lin971. The dosage was varied by changing the thickness of an
amorphous silicon (a-Si) implantation mask above the silica layer. Phase segregation
of the excess Si was achieved with a 1 hour 1100 C anneal under N,. The resulting
PL wavelength varied with the implantation dose (longer wavelength for more Si) and
was believed to correlate with particle size. In all cases, the PL decay took the form
of a stretched exponential similar to that reported for PS. The primary exponential
time constants decreased exponentially with increasing PL energy from 20 to 50 as
for 1.45 eV PL to 5 to 15 is for 1.9 eV PL. The extended tails were attributed to
interactions between adjacent quantum dots. The PL was believed to be excitonic
even at room temperature.
Okamoto observed low temperature PL in 2D Si/SiO2 single quantum wells at
low temperature (2 K) [Oka97]. A small size-dependent PL peak was seen
superimposed on a much larger size-independent peak which was located at 1.6 eV
and had a decay time of 1.1 ms. The size-dependent portion had the following
locations and decay times: 1.8 eV and 0.3 ms for a 0.6 nm well, 1.7 eV and 0.6 ms for
a 1 nm well, and 1.5 eV (no decay time reported) for a 1.3 nm well. Stokes shifts for
resonant excitation were from 0.1 eV to just under 0.4 eV, which is consistent with
theoretically expected exciton binding energies in quantum dots. The majority of the
luminescence was attributed to defects at the Si/SiO, interface while the small size-
dependent portion was attributed to a quantum size effect.
Min et al. prepared Ge quantum dots in SiO2 by ion-implantation of a super-
saturated concentration of Ge IMin96]. No correlation was seen between nanocrystal
size and the peak energy and lifetimes of the PL. In addition, the PL shows only
weak temperature dependence, which is inconsistent with the published theory for Ge
quantum dots. The PL was identical to that seen in silica damaged from Xe
implantation with no quantum dots present. The luminescence was attributed to
radiative defect centers in the SiO, matrix.
In a theoretical paper, Takagahara and Takeda used first-principle calculations
to describe the development of pseudo-direct behavior in quantum dots of indirect-
gap materials such as Si and Ge [Tak921. The interesting claim is made that the
resulting band folding in a nanocrystal is similar to the local action of an isoelectronic
center such as the nitrogen impurity in GaP:N. The calculated lifetimes were found
to vary from the ns range for 1 nm dots to the ms range for 3 nm dots of either Si or
Ge. Effective gaps were calculated to be 1.5 eV for 3 nm particles for both Si and
Ge. To achieve 3.0 eV gaps, sizes as small as 1.3 nm in Si and 1.7 nm in Ge were
The large disagreement in time constants for processes attributed to quantum
confinement makes it difficult to argue that a process is not due to such effects based
on a measured time constant. However, there seems to be no experimental evidence
of time constants in the 1-10 ps range for known, well-characterized, quantum
structures. This is important because, as will be shown in Chapter 8, the primary time
constants of sp-Si are in this sub-10 ps range.
Particle Size vs. Energy Gap
Several of the papers in the preceding section also discuss the correlation
between PL peak wavelengths and particle size or the lack of such agreement. It is
evident, even in theoretical models, that very small particles would be needed to
account for the UV/blue peak in sp-Si. Takagahara's calculation that a 1.3 nm Si
quantum dot would have an absorption band-gap of 3.0 eV is in close agreement with
the simpler effective mass approximation model. If one considers the high Stokes
shift observed in quantum systems (hundreds of meV) and the fact that the UV/blue
emission of sp-Si (with 325 nm excitation) is at 3.3 eV (not 3.0 eV), then a gap on the
order of 4 eV would be necessary for a quantum dot model of this PL band. This
would correspond to a particle size significantly smaller than 1 nm. In addition, there
seems to be a wavelength vs. particle size discrepancy between theoretical papers and
experimental reports in the literature. The wavelengths often do not correlate well to
particle size, or when they do, they are often longer for a given particle dimension
than predicted by theory.
It is interesting that even in a very well characterized system, such as that
studied by Okamoto, where some luminescence that correlated with size was found, it
appeared to contribute only slightly to the total luminescence intensity. Thus,
assigning the PL bands of sp-Si to non-size effect related systems is not necessarily a
challenge to the validity of quantum theories. Such phenomena may simply be
masked by more prominent ones.
Papers by Gotza and Wu and their coworkers provide more examples of well-
defined quantum structures that fail to produce observable luminescence. Gotza et al.
prepared arrays of free-standing isolated 2-3 nm diameter silicon wires using reactive
ion etching of silicon-on-insulator wafers followed by a self-limiting oxidation
[Got981. The buried insulator layer not only isolated the wires electrically from the
substrate, but served as an effective etch stop. These wires were then passivated by a
layer of a-SiN,:H followed by an anneal in a reducing atmosphere (forming gas).
Under 325 nm excitation, a broad luminescence could be seen with a peak
wavelength which depended on the value of x in the SiN,:H capping layer. In
addition, PL bands were observed at 400 nm, 480nm, 560 nm, and 650 nm. All of
these emission bands were attributed to various defect states in SiN, and SiO2. Wu et
al. observed PL spectra from nanocrystalline Si:H that has no correlation with particle
size, even for mean particle sizes as low as 2.2 nm IWu_981. The PL was attributed
to the radiative recombination of carriers at interface defects in the nanocrystalline
Excitonic Nature of Quantum Dots
Several of the papers already discussed report or predict that various
nanoparticle systems show excitonic behavior, even at room temperature. According
to Yip, quantum systems should exhibit strong Stark-effect red-shifting and intensity
quenching due to an applied electric field:
In response to an electric field, the electron and hole
eigenstates follow the contours of the linear energy gradient in
opposing directions so that a large red-shift necessarily induces a
strong polarization of the eigenstates with the corresponding reduction
in the oscillator strength of the absorption resonance. [Yip98, 8021
The same should follow for the emission resonance as well, which differs only
by the Stokes shift corresponding to the exciton binding energy. While Yip worked
with InAsP/InP and InAsP/InGaP quantum wells, this quenching may be expected to
be even more prominent in indirect-gap materials that rely on relaxation of the
momentum selection rules for luminescence. By reintroducing a spatial dependence
to the eigenstates that increases the expectation value of the hole/electron separation
distance, the external field decreases the localization of the system. In addition to the
decrease in resonance that occurs in direct-gap quantum systems, there should be an
additional decrease in the oscillator strength from the loss of Heisenberg-related
There is a strange disagreement about the quantum confined Stark-effect in
the literature. In bulk (non-quantum) semiconductors which show excitonic behavior
(often at low temperatures), the Stark-effect results in a blue-shift because the binding
energy of the exciton subtracts from the gap energy. When the exciton is destroyed
by the electric field, the emission energy increases by an amount equal to the binding
energy. Haung describes the same phenomenon in quantum confined systems
[Hau891. While this blue shift contradicts the red-shift described by Yip et al., Haung
confirms that highly confined quantum systems are excitonic at room temperature and
that moderate electric fields cause large decreases in the oscillator strength of such
systems. While the author finds the argument of Yip more compelling, it may not be
necessary to resolve the red/blue-shift dispute for purposes of this study.
If a distribution of quantum dot sizes were responsible for either broad PL
band in sp-Si, it may be impossible to observe any Stark red or blue-shift because its
magnitude may be insignificant compared to the total width of the band. However,
intensity quenching would apply to each nanoparticle in isolation and should be seen.
A total lack of any Stark-effect quenching in the UV/blue band sp-Si will be reported
in Chapter 9.
Limitations of the Various "Size-Effect" Models
An ideal theory would be capable of predicting the whole range of
semiconductor properties from a single atom to an infinite bulk crystal. This is
clearly not available. Different approaches are used for different scales or ranges.
Bloch functions do a good job of predicting the band structure of infinite crystals.
other models can account for surface relaxation in finite crystals, etc. The quantum
confinement approach, especially in its simplest form (the particle in a box) seeks to
extrapolate the behavior of small particles from their bulk properties. At the other
extreme end of the spectrum, molecular models seek to explain the behavior of
molecules and small clusters by building them up from individual atoms and atomic
bonds. More sophisticated "quantum" models blur the line as to exactly what
category they belong to. Of course, the natural system is not changing, only the
model. The ultimate test of any model is how well it works at predicting the behavior
of real systems. A number of papers, considered under the broad umbrella of
quantum confinement literature, blur these lines between the molecular approach and
the quantum size effect approach.
Khurgin reports that real Si/SiO2 quantum dot systems do not exhibit
luminescence that corresponds well to the ideal model where the blue shift is
inversely proportional to the square of the particle diameter [Khu96I. After
acknowledging that this lack of correlation has led to models based on oxidation and
surface states. Khurgin suggests that the discrepancy may be accounted for by
considering that a majority of the luminescence may come from a small portion of the
size distribution of particles. A model similar to the effective mass approximation,
but that takes into account all three lattice directions, demonstrates that the oscillator
strength (OS) of luminescence depends only on the smallest dimension of a
nanoparticle, not on the average dimension. Furthermore, the OS is proportional to
d6, where d is the smallest dimension. With such a nonlinear relationship, the mean
particle size measured by Raman or other means is less important than the size of the
smallest population of particles, which will contribute most of the luminescence.
While the author agrees with this in principle, the d6' relationship is questionable as
nowhere in the derivation did Khurgin account for the finite barrier of a Si/SiO,
quantum dot. A more rigorous treatment by Hybersten finds that the OS of a
quantum dot varies with d 6 only for the idealized case of an infinite barrier and with
as little as d3 for finite barriers IHyb94|. Finally, Khurgin et al. achieved a close fit
to empirical results by assuming a 2 nm size distribution about the mean. These
results fit data for systems exhibiting luminescence between 1.7 and 2.7 eV. Despite
the good fit between this modified theory and the observations, Khurgin has an
excellent grasp of the limitations of the quantum confinement model:
It is important to note that this statement is not contradictory to
the alternative model: since size of the NC's [nanocrystals] is of the
same order of magnitude as the size of the exciton in the silicon oxide
or siloxene, and/or the spatial extent of the surface state, the
probability of electron-hole recombination determined by the volume
in which they are confined, should be roughly the same for the exciton,
surface state, or quantum confinement. In this respect, making the
distinction between the quantum confinement, surface state, and defect
makes very little sense on the nanometer scale. IKhu96, 1243]
Quantum Dots as Large Molecules
Three papers that blur traditional lines are discussed in this subsection.
In the first paper, Delley uses a density functional approach to calculate the
(absorption) band-gaps of small silicon structures ranging from a single silicon atom
to a 3 nm (706 Si atom) nanoparticle as well as an extrapolation all the way up to bulk
silicon fDel931. The claim is made that dangling bonds would completely mask the
effective band gap, so all dangling bonds were considered to be hydrogen terminated.
After correcting for self-energy considerations using a Green's function dynamically
screened interaction approximation, the calculated band-gaps at both extreme ends of
the size scale are very close to the known values (within 0.1 eV for bulk Si and within
0.3 eV for pentasilane). Through a symmetry consideration, the model also predicts
the pseudo-direct gap behavior that occurs in small particles. This may be about as
close as one can get to an ideal model for Si particles of all scales, at least in these
respects. The model predicts that the band-gap varies with N"3 where N is the
number of Si atoms in the cluster (this is essentially d' for spherical clusters) while
the OS is nearly constant for clusters of 1-30 atoms and decreases exponentially with
the number of atoms above 30. A band-gap of over 6 eV is calculated for a single
silicon atom and a band-gap of 2.5 eV is calculated for a 3 nm (706 atom)
nanoparticle. The latter is claimed to correspond to the smallest observed particles at
the time of submission and to be in agreement with the observed absorption energy
In the second paper, Ebihara reports on the PL and EL of polydihexylsilane
(PDHS) [Ebi971. While no models are presented, if the preceding papers are
accurate, such a small molecule is the logical extension of the concept of a silicon
quantum dot to smaller sizes and higher energies. PL emission is reported to be at
3.35 eV (370 nm) at 77 K and 3.23 eV (384 nm) at 4.2 K. The quantum efficiency is
less than 0.1 % at 300 K and it increases to almost 0.7% at temperatures near 4.2 K.
The trend is fairly linear. At room temperature, this peak is close in energy to the
UV/blue peak of sp-Si but, as will be seen in Chapter 8, the thermal quenching of sp-
Si is nonlinear. Also, Ludwig observed no thermal shift in the UV/blue peak. but the
author did observe a blue shift with increasing temperature.
In the third and final paper, Filonov presents a fascinating report on the
calculated electronic properties of Si,4 clusters with varying degrees of oxygen
involvement 1Fi981. The Si,4 cluster is highly symmetric and corresponds to a
silicon nanoparticle about 0.8 nm in diameter. Filonov starts with a fourteen atom
lattice "fragment" with a vertical direction corresponding to the bulk Si <111> axis
and atoms at their standard diamond lattice sites. There are 16 Si-Si bonds and 24
dangling bonds that are hydrogen terminated. This initial molecule is, thus. Si14H24.
The effect of adding oxygen to the structure is determined by building additional
prototype molecules one additional oxygen atom at a time. Two possibilities were
studied: replacing terminating hydrogens by -OH groups and inserting -0- bridges
between Si-Si bonds. The second scheme is limited to the addition of 16 oxygens.
one into each of the 16 Si-Si bonds, and the decision was made to limit the number of
oxygen atoms incorporated by both schemes to 16. To this end, 21 prototype
molecules were constructed: Si14H24, Si4H2,(OH), with x between 1 and 16, and
Si14H240, with x between 1 and 16 (the following values of x were skipped in both
cases: 2, 4,7, 9, 11, and 13). The latter structure essentially approaches a silica, not
silicon, quantum dot. For each x value, the lattice relaxation of the structure is
estimated using an MM2 routine and the lowest energy locations for the oxygen
atoms are selected. The candidate structures are then further optimized using
molecular orbital theory. The specific theory used was MO LCAO theory using the
Modified Neglect of Diatomic Overlap-Parametric Method 3 (MNDO-PM3) in the
restricted Hartree-Fock approximation. Only valence electrons were considered as
individual bodies. Core electrons were modeled as part of the nucleus. This method
allowed both the ground and excited states to be modeled, neatly giving absorption,
emission, and Stokes shift energies for every structure. The simplest Si 14H24 structure
has an absorption band-gap of about 4.4 eV and an emission energy of 3.7 eV, giving
fairly close agreement with more common quantum confinement models.
Hydrogen replacement by -OH groups at the surface resulted in little change
in the emission energy of the cluster but there was a continuous decrease in the
absorption energy with increasing -OH incorporation. This also resulted in a
decreasing Stokes shift with more -OH that was attributed to a decrease in physical
relaxation of the excited state due to mutual repulsion among the electronegative -OH
groups at the surface. The addition of bridging -0- in the bulk of the dot gave
significantly different results. The Stokes shift was essentially constant for values of
x between 0 and 10. which is attributed to the lack of influence of the oxygen on the
surface that undergoes relaxation. The bulk is already too tightly bonded to suffer
significant relocation of atoms. The conclusion is made that the Stokes shift in a
quantum dot is essentially a function of surface termination. As x approaches 10, the
absorption and emission energies decrease in a linear fashion. With x between 14 and
16, the energy levels are radically different (they suddenly increase about 2 eV) and
the Stokes shift is tiny. It is actually negative when x is 16, indicating that the ground
state is unstable. When x is 12. the energies and Stokes shift are intermediate
between the two forms. It is in this range that the molecule looses its silicon
character and takes on properties similar to silica. The overall ground state energies
of the oxygen-incorporated structures are lower than those of the -OH terminated
ones and are more likely to occur in nature.
The excitation and emission wavelengths and constant Stokes shift of the
bridging oxygen-incorporated structure are remarkably similar to the behavior of the
UV/blue peak of sp-Si when excited by various wavelengths. However, this is only
true over a limited range of excitation energies so it is probably a coincidence. This
will be addressed in detail in Chapter 10.
Review of Defects in Silica and Silicon Oxynitrides
Before concluding that known defects in silica and silicon oxynitrides are not
responsible for the luminescence of sp-Si, one should consider some broader
luminescence literature beyond that which was reviewed in publications by Hummel,
Ludwig, et al. This review will concern itself with the phenomenological
characteristics of such luminescence only, instead of attempting to review the theory
of the luminescence from defects that do not apply to sp-Si. Absorption bands at 165
nm, 240 nm, 258 nm, and 620 nm have already been discussed as well as emission
bands at 185 nm. 264 nm, 288 nm. 400 nm. 563 nm. and 650 nm. All of these have
been ruled out as candidates for the UV/blue (375 nm) or green (525 nm) emission
bands of sp-Si. Some additional emission bands, as well as additional evidence
regarding these already presented, are discussed.
Low-Energy Emission Bands
States with emission energies of 2 eV or less are often reported for silica.
Three common examples are 1.7 eV (729 nm), 1.9 eV (653 nm), and 2.0 eV (620
nm). Kenyon observed all three emission peaks in silicon-rich silica [Ken96]. The
1.7 eV peak was attributed to oxygen deficient centers (ODCs), while there was no
attribution for the 2.0 eV peak. Interestingly, the 1.9 eV center was attributed to
quantum structures because it is red-shifted and looses intensity with annealing.
However, this band may be the same as the previously discussed 650 nm band.
Tohmon also observed 1.9 eV luminescence and claimed that it is a well-known
oxygen vacancy related defect IToh89a]. Shimizu-Iwayama observed the 2.0 eV and
1.7 eV bands but not the 1.9 eV band in ion-implanted SiO2:Si [Shi941. The 2.0 eV
center was destroyed by annealing at 600 C and its intensity followed that of the
electron spin resonance (ESR) signal of the E' center (believed to be a Si dangling
bond at an oxygen vacancy) in SiO.. The 1.7 eV emission was attributed to carrier
recombination at defects on the surface of Si nanoparticles. Clearly, there is
disagreement among the authors regarding the source of sub 2.0 eV luminescence in
silica. However, as the primary peaks of sp-Si are above this energy range, they can
be ruled out nevertheless.
Involvement of Other Elements
Several authors have reported on emission related to impurities in silica or
luminescence from silicon oxynitrides. Augustine observed 2.2 eV emission (564
nm) under 2.54 eV (488 nm) excitation from SiONy:H plasma-enhanced chemical
vapor deposition (PECVD) films made with a N20 source [Aug951. This
luminescence increased 10 times after a 20 minute rapid thermal anneal (RTA) and
was attributed to bandtail-to-bandtail emission of the oxynitride. Stathis observed the
ESR signal of a photoinduced defect in silica that he attributed to nitrogen
involvement [Sta84]. Two models were proposed: -Si-O-N-O-Si- and
,Si-O-N-Si.. Finally, Poumellec reports 7 absorption bands and associated
emission bands in SiO,:Ge [Pou97]. Clearly, the incorporation of other elements into
silica can be significant. In sp-Si. likely elements for incorporation are nitrogen,
tungsten (already ruled out as contributing to luminescence), and carbon (from
organic contamination). The luminescence of sp-Si produced in mixtures of nitrogen
and carbon dioxide are reported in Chapter 5. Carbon is associated with a third peak
in sp-Si located near. but distinguishable from, the green peak. This peak is not seen
in samples not deliberately processed in carboniferous gases, even when a measurable
amount of carbon contamination is present. This leaves only nitrogen as a viable
candidate for the UV/blue and green emission peaks.
Higher-Energy Emission Bands
Gee and Imai both observed similar defects in silica irradiated by much
different means JGee79, Ima88]. In neutron irradiated silica. Gee observed the 4.3
eV (288 nm) emission band previously reported in this chapter and found it to be
associated with a 7.6 eV excitation peak attributed to an ODC. Imai observed this
same emission and excitation peak in silica exposed to ArF excimer laser radiation.
An excitation band for the same emission was also found at 5.0 eV. All three peaks
were reduced in intensity after annealing in oxygen. Tohmon placed these absorption
peaks at 7.6 eV and 5.2 eV and claimed three associated emission peaks at 4.3 eV, 2.7
eV (459 nm), and 1.9 eV (653 nm) [Toh89]. Imai attributed the 7.6 eV absorption to
a relaxed oxygen vacancy and the 5.0 absorption to an unrelaxed vacancy. In an
unrelaxed oxygen vacancy, the Si atoms are near the same positions they would
occupy without a missing oxygen. The unbound electrons on the Si atoms have a
pseudo-bonding interaction with each other over such a long distance (-Si *Si-).
The potential of this system is rather high when the separation is large. If the Si
atoms are displaced somewhat, the system can reach a lower potential energy
configuration known as the relaxed vacancy. In the literature, the unrelaxed vacancy
is sometimes represented by aSi/ \Sia to distinguish it from the relaxed vacancy:
*Si-Si-. As these states do not reproduce the observed excitation spectrum of sp-Si
(to be presented in Chapter 8), these defects can be ruled out as well. However, a
related defect will be discussed in the next section.
Several authors report absorption bands in silica near 5 eV. Tohmon observed
a 5.25 eV absorption associated with 3.1 eV (400 nm) emission that he attributed to
2-coordinated silicon with two excess electrons. IToh89] While this emission peak is
near that of the UV/blue band of sp-Si when excited by 325 nm (3.8 eV) light, it will
be shown that this is not the case when sp-Si is excited by energies closer to 5 eV.
Kohketsu also observed similar peaks at 5.17 eV and 5.06 eV [Koh89]. The former is
associated with emission at 4.3 eV (288 nm) and 3.1 eV (400 nm), and the latter is
associated with emission at 4.2 eV (295 nm) and 3.0 eV (413 nm). The decay of the
3.1 eV band had a time constant of 80 ats, which will be shown to be completely
incompatible with the UV/blue emission from sp-Si. Kohketsu attributed the former
to excess Si in SiO, and the latter to small Si clusters containing at least 4 Si-Si
bonds. Bagratashvili reported absorption bands at 248 nm (5.0 eV) and 242 nm (5.12
eV) that he associated with emission peaks at 280 nm (4.4 eV) and 455 nm (2.7 eV)
for the former, and 296 nm (4.2 eV) and 396 nm (3.1 eV) for the latter. The former
was attributed to simple ODCs while the latter was attributed to ODCs associated
with germanium impurities. The 455 nm emission was reported to increase in
intensity as the temperature was increased from 290 K to 514 K. This is in contrast to
the peaks of sp-Si, which are thermally quenched, not activated. Finally, Griscom
reported these absorption peaks at 5.0 eV and 5.15 eV lGri91]. Clearly, there is much
disagreement in the literature on these similar peaks. None of which appear to fit the
behavior sp-Si. Griscom also discusses a 3.8 eV absorption band that may be linked
to peroxy linkages or interstitial Cl2 molecules.
Review of an Important Transient Defect in Silica
The observed properties and theory of one particular defect should be
discussed in detail, because, as will be shown, its absorption curve matches the
excitation spectrum of the green PL peak of sp-Si. This transient defect found in both
crystalline and amorphous silica is believed to be the self-trapped exciton (STE).
An excellent review of the properties and theory of the STE has been written
by Itoh Iito891. While the band-gap of silica in either the amorphous or crystalline
form is a matter of some debate, SiO2 of both varieties appears to have a direct
transition at 10.4 eV. Even the question of how to define a band-gap in an amorphous
material is controversial. The author prefers to call the separation between the
mobility edges the "gap" and consider local tail states to be within the gap. Often
quoted numbers such as 8 eV for amorphous and 9 eV for crystalline silica are for
indirect transitions and may include a significant portion of the tail states in the
former case. If one measures the optical absorption at many temperatures and
rigorously defines the temperature-independent convergence of the Urbach tails to
define the indirect gap, the numbers become 9.1 eV for crystalline and 8.7 eV for
Electromagnetic radiation above 10.4 eV, as well as ionizing radiation and
electron bombardment, have long been known to introduce transient defects in both
amorphous and crystalline silica. In amorphous (but not crystalline) silica, this
transient defect is credited with being a precursor of many of the defects discussed in
the previous chapter. Under such excitation, a 2.8 eV emission is observed. Its decay
is exponential in crystalline silica and both non-exponential and blue-shifted with
time in amorphous silica (this is characteristic of inhomogeneously broadened donor-
acceptor pair emission). Figure 3-21 shows the transient optical absorption reported
by Itoh for this defect. A 5.2 eV absorption peak with a 4.2 eV satellite peak are
observed. It is remarkably similar for both amorphous and crystalline silica
indicating that this state likely arises from the reorganization of bonds in the SiO2
backbone common to silica in all its forms. This transient state has a triplet ODMR
(Optically Detected Magnetic Resonance) signal shift an order of magnitude stronger
than that observed for a recombining free electron and hole. The most important
evidence comes from the observation that this transient state has a large volume
change on the order of one SiO. molecular volume per defect.
I ""I I I
3.0 40 SO 6.0
PHOTON ENERGY (.V)
Figure 3-21: Transient optical absorption spectra of the defect believed to be the self-
trapped exciton in crystalline and amorphous silica. Adapted from Ito89.
The Model for the Self-Trapped Exciton
The popular model to explain these observed properties is that of a self-
trapped exciton in silica, usually abbreviated STE. This is actually a double exciton
consisting of an oxygen vacancy with two extra electrons and an adjacent peroxy
radical with two holes. Both the defect as a whole, as well as each site individually,
have paired spins and are diamagnetic. The structure of this defect is shown in Figure
3-22. Note that in a Si02 crystal, the peroxy group must invade the adjacent
interstitial site for room and this configuration makes the STE very similar to the
close Frenkel pair defect in crystalline silica (and it can be considered a donor-
acceptor pair in amorphous silica as well). Most authors simply call it the Frenkel
defect, but Itoh is careful in pointing out that it is an isomer of the idealized Frenkel
defect because the peroxy group occupies both the standard oxygen site and part of
the adjacent vacancy site, giving a smaller vacancy-interstitial separation than in the
classical Frenkel pair. The classical interstitially-bonded Frenkel defect for silica
would have the extra oxygen simply inserted into the silica linkage in a non-resonant
manner instead of as a closely bonded peroxy radical. Thus, it is more rigorous to
represent the STE as mSi- Si-O,2-Si. than as mSi -Si-O-O+-Si-. This
distinction is not trivial, for reasons that will be addressed shortly.
The large Stokes shift between 10.4 eV excitation and 2.8 eV luminescence is
attributed to the lattice relaxation necessary to break one Si-O bond and create the
subsequent peroxy radical.
In another publication, Itoh makes several additional important points
Iltoh88j. Evidence is building that the STE is common in insulating solids with
strong electron-phonon coupling. Since SiO, is a "lone-pair" semiconductor with the
2 = LONG BOND
Figure 3-22: Proposed structure of the self-trapped exciton in silica consisting of an
oxygen vacancy with two electrons and an adjacent peroxy radical with two holes.
Adapted from Ito89.
valence band strongly associated with electrons at oxygen sites and the conduction
band strongly associated with electrons at silicon sites, such coupling is unavoidable
in silica. Strong experimental evidence of the STE exists for Y205 and A1203 as well.
It would not be unreasonable to expect the STE to exist in all oxides.
Upon closer examination, Itoh found that the STE can decay through a
completely intrinsic process and give the previously reported 2.8 eV luminescence as
well as decay by charge transfer to other nearby defects through an extrinsic process
resulting in 2.5 eV luminescence. The 5.2 eV transient absorption peak and the 4.2
eV satellite peak are not from separate absorption centers but are a superposition of
two transitions of the same center. In an optical bleaching experiment, the fractional
change in the optical density across the entire transient absorption spectrum of the
STE was found to be completely independent of probing energies between 3.0 and
Due to the degree of polarization of the 2.8 eV intrinsic luminescence of the
STE, Itoh was able to calculate that the STE is created by the breaking of one Si-O
long bond followed by the oxygen entering a peroxy arrangement with the only other
oxygen which is long-bonded to that same silicon atom. This is important because
Robertson has challenged the model of the STE on the basis that 10.4 eV is not
sufficient to break "three Si-O bonds" [Rob85]. It seems that Robertson is adhering
to the "ball and stick" model of the peroxy ion being two oxygen atoms with a normal
bond between them. It appears much more reasonable to the author that one Si-O
long bond is broken and the free oxygen atom resonantly rearranges its electrons with
the neighboring oxygen without ever breaking the bonds that the second oxygen atom
shares with its silicon neighbors.
It is believed that the 5.2 eV transition results from one of the excess electrons
at the vacancy site being promoted from a bonding to an anti-bonding orbital while
the 4.2 eV transition results from one of the same electrons being physically
transported to a higher orbital of the peroxy radical site, see Figure 3-23. The STE is
metastable, which is probably due to the tetrahedral pseudo-bonding that occurs
across the vacancy (the ground state of the STE has the two excess electrons in
bonding orbitals of the vacancy and two holes in the anti-bonding orbitals of the
peroxy radical). The STE decays with a time constant roughly on the order of
milliseconds if left alone and nearly instantaneously if excited by either transition.
The whole structure becomes unstable if one of the excess electrons cannot "see" the
other across the vacancy or if the peroxy radical has even one electron in an anti-
bonding orbital. Consequently, a single-exciton STE is never seen because there is
no energetic advantage to its formation. The STE relaxes nearly all of the 10.4 eV
used in its formation and is only about 2.8 eV in potential above the ground state.
There is a significant metastable energy barrier between the STE and the ground state
that can be overcome by energies in the range of 3.0 to 4.5 eV. However, the driving
force to return to the ground state is not that large and this may explain why the STE
readily transfers the offending electron to another defect in the extrinsic process,
probably by propagating it along the silica backbone and possibly even accepting a
lower energy replacement in the same manner. To occur with any likelihood, this
charge transfer process would have to be very fast since the optical bleaching itself is
known to be very fast.
The lifetime of the STE is about I ms |Hay84| in crystalline silica and about
10 4ts [Tani881 in amorphous silica. The decay of the former is in the form of a
stretched exponential or power law while the latter is exponential. Non-exponential
5.2 eV absorption AV
4.2 eV absorption
2.8 eV luminescence
Figure 3-23: Ground state of the self-trapped exciton (STE) in silica with two electrons
in bonding orbitals of the Si *Si pseudo-bond and two holes in the anti-bonding orbitals
of the peroxy radical along with proposed optical transitions:
2.8 eV emission from the intrinsic decay process of the STE to normal silica
5.2 eV absorption promoting a vacancy electron to an anti-bonding orbital
4.2 eV absorption transferring a vacancy electron to a higher orbital of the peroxy
Adapted from Ito88.
decay is usually observed in donor-acceptor pair luminescence when there is a
distribution of pair separation distances. This also explains the observed blue shift
with time seen in the PL of the a-SiO2 STE. While the favorable bonding
arrangement of the doubly charged STE (filled bonding orbitals in a pseudo-bond
across the vacancy in a configuration some authors refer to as a "relaxed vacancy"
and filled bonding orbitals at the peroxide with empty anti-bonding orbitals at both
sites) makes this state nearly stable, the driving force for the destruction of the STE
certainly comes from the very large strain it creates in the material. While one would
expect this strain to be more severe in the crystalline material, the faster time constant
of the amorphous case may be explained by the lower diffusivity of the oxygen in
crystalline silica. Once the oxygen occupies an interstitial site, it must displace other
lattice atoms to return to the vacancy site.
There are important points in three more papers, which involved Itoh, but with
either Tanimura or Tanaka as primary authors. These are discussed in chronological
Tanimura et al. studied the formation of the STE when silica is exposed to
electron bombardment and found the formation process was nonlinear with electron
dose |Tani83 I. This supports the idea that the singly ionized STE is unstable and that
a two-step ionization process is necessary for the formation of the metastable STE.
The current must be high enough that there is a significant likelihood of the same
"mer" on the silica backbone being doubly ionized within the transient lifetime of the
unstable singly ionized STE. In addition, the volume change of the STE was
carefully measured and found to be at least one molecular volume and essentially the
same for both crystalline and amorphous silica.
Tanaka et al. verified the volume change reported by Tanimura et al. and also
discussed the likelihood of a STE explanation for transient absorption bands observed
in As2Se3 [Tan851.
Finally, Tanimura et al. performed extensive experimental work on electron
irradiated amorphous silica JTani88]. The main transient absorption peak was found
to be 5.3 eV for amorphous silica rather than the 5.2 eV reported for crystalline silica,
while the satellite band remained at 4.2 eV in both materials, see Figure 3-24. The
transient absorption of high purity silica as well as silica with high concentrations of -
OH and Cl is identical indicating that this defect involves the backbone structure of
SiO2. A Gaussian deconvolution of the spectrum indicates the two absorption peaks
are 5.30 eV with a full-width half-maximum (FWHM) of 0.78 eV and 4.20 eV with a
Polished silica plates were placed on a 78 K cold finger and irradiated with a 2
MeV 20 ns electron pulse from a Febetron 707 accelerator creating electron-hole
pairs in concentrations up to 2 x 1018 cm-3. The time evolution of the PL intensity at
450 nm and the optical absorption at 250 nm were monitored following the electron
pulse. Finally, a 308 nm laser pulse from a XeCI excimer laser was used to bleach
the transient STE center. Figure 3-25 shows the resulting PL and transmission curves
vs. time. t, represents the electron pulse and t, represents the laser pulse. The width
of the laser pulse and the data sampling rate were not stated. The creation of centers
at te and the destruction of centers at t, are too fast for the instrument to clearly
resolve. The destruction or bleaching of the center verifies that its excited state is
unstable. It should be noted that these plates were only 1 cm thick, yet the material
experienced a 70% reduction in the transmission of 250 nm UV light. The absorption
So ) a-SiO2
(&A ) a-SiO2:OH
0 I a-SiO2: CI
3 5 6
Photon energy (eV)
Figure 3-24: Transient optical absorption spectra of the self-trapped exciton in
amorphous silica showing no difference among high-purity silica. -OH-rich silica, and
Cl-rich silica. Peaks at:
5.30 eV with a FWHM of 0.78 eV
4.20 eV with a FWHM of 1.16eV
Adapted from Tani88.
Figure 3-25: Decay of the self-trapped exciton (STE) in amorphous silica:
(a) PL intensity of decay emission (monitored at 450 nm)
(b) optical transmission of sample (monitored at 250 nm)
te denotes the 2 MeV electron pulse that creates the STEs while t, denotes the 308 nm
laser pulse that bleaches the STEs. Adapted from Tani88.
A B E B B
I m m It E
at the 5.3 eV (234 nm) peak would be about 20% stronger than it is at 250 nm.
Clearly, the STE is a strong absorber of UV light.
Additional Evidence of the Nature of the Self-Trapped Exciton
A number of other sources offer some additional clues to the nature of the
STE. Hayes performed OMDR measurements on crystalline quartz [Hay84]. In
quartz, as well as amorphous silica, each oxygen atom is bound to two silicon atoms
by one long bond (1.612 A) and one short-bond (1.607 A). Each silicon atom has two
oxygen atoms connected by long bonds and two by short bonds. Hayes observed that
long bonds are weaker and preferably broken. This is consistent with the structure
presented by Itoh. A physical explanation of the difference in lifetimes for the STE in
amorphous and crystalline silica has already been presented. However, Hayes
attributes the 1 ms lifetime of the STE in its unexcited state in c-SiO, to his
observation that the transition from the STE to the standard silica bonding
configuration is spin-forbidden. If this explanation for the time constant is true (as
opposed to the limited diffusivity argument previously proposed), one must make the
argument that the lOOx faster time constant in the amorphous case comes from some
relaxation of the spin selection rules that is not seen in the crystalline case.
Silin discusses the transient defect in silica which he also attributes to the STE
and describes how it can give rise to more traditional defects like the various E'
centers, NBOHCs, and neutral peroxy bridges in amorphous silica ISi1801. Careful
reading of this paper indicates that Silin is actually discussing the singly-ionized
version of the STE because of a specific reference to "an electron in an anti-bonding
orbital" which makes the exciton unstable. This may be the critical difference
between the mechanism that leads to defect formation and that which leads to the
metastable doubly-ionized STE.
The Physics of Self-Trapping in Polar Solids
Some explanation of what exactly "self-trapping" means and why it occurs
should be made. Mott and Davis are a good source for this but can get quite involved
for the casual reader IMot77]. In general, self-trapping occurs when an electron,
hole, or exciton interacts with the lattice in such a way that it becomes distorted and
the combined system has a lower energy than the free carrier. It is called self-
trapping because the defect did not exist prior to being induced by the carrier; rather,
the carrier created the defect to minimize the total energy of the system. Such a state
is essentially a molecular polaron. The hindered-rotor problem and the Jahn-Teller
effect from solid-state physics are analogues of self-trapping. Self-trapping occurs
only when the lattice distortion is large enough to create a global minimum in the
energy vs. coordination coordinate plot, see Figure 3-26 for an example. A local
minimum is not sufficient for self-trapping at finite temperature as such metastable
states are very shallow and free carriers can find their equilibrium position within a
few lattice vibrations. Such thermalization is a femtosecond process (Bla94j. Strong
interactions between the lattice and carriers occur in silica because it is a "lone-pair"
insulator with a valence band made up of oxygen 2p-orbitals. This, combined with a
reasonably large band-gap, guarantees that nearly every outer-shell electron in the
solid is at an oxygen site. This strong polar nature of the lattice provides the needed
carrier-phonon coupling. Self-trapping has been reported in many oxides including
MgO, CaO. SrO, ZnO, BeO, and Ai20,.
w^ W Uidi ___
0 ^<^ ^~~
(c) W (ii)
Figure 3-26: Theoretical self-trapping mechanism for a hole:
(i) denotes the elastic energy, (ii) denotes the electronic energy, and (iii) denotes the total
(a) no polaron is formed
(b) a polaron is formed with trapping energy W
(c) the case of trapping by a pre-existing defect.
Adapted from Mot77.
The Polar Nature of Oxides and "Lone-Pair" Insulators
The domination of the valence band by oxygen in an oxide is explained in
simple terms by Fowler [Fow86J. Taking SiO, as an example, each silicon atom
contributes 4 valence electrons while each oxygen atom has 6 electrons in its outer
shell. There are also two oxygen atoms for every silicon atom. This means that there
are 12 valence electrons contributed by oxygen for every 4 contributed by silicon.
Even if the bonding were covalent, the oxygen would contribute 75% of the total
valence electrons. However, the Si-O bonding is highly ionic with the electrons
favoring the oxygen. For this reason the valence band is almost totally dominated by
electrons at oxygen sites.
In this same paper. Fowler also discusses the nature of the bonding across the
oxygen vacancy in silica and supports the idea of a pseudo-bond bridging the
Even with the structural constraints imposed by the crystalline
environment the silicons will approach to =2.5 A 12.35 A for
amorphous apart.... The suggestion that the 0 vacancy is better
thought of as a Si-Si bond seems to be valid. IFow86, 1741
In addition. Fowler discusses the STE in some detail. The paper is supportive
of the position of Itoh, Tanimura. and Tanaka but does mention that some controversy
exists on the subject. For example, based on the same evidence for the relaxation of
the oxygen vacancy, Fowler points out that one would expect the oxygen vacancy
associated with the STE to lead to a contraction and not an expansion as observed.
The discussion closes with:
Although no calculations have yet been reported which bear on
the stability of this self-trapped exciton, it is reasonable to assume that
the system has found the lowest energy atomic arrangement consistent
with its being an excited electronic state; in a configuration-coordinate
picture it is at the minimum of the excited-state curve. Apparently
there are no low-energy crossovers to the ground-state configuration
coordinate since radiative de-excitation occurs. [Fow86, 179]
Summary of the Properties of the STE
Mott has stated unambiguously that the coordination coordinate is a stress-
related concept |Mot75]. Taking this into consideration, we can conclude that the
STE is chemically and electronically stable, but is not favored elastically. This elastic
constraint is relatively small in magnitude, as the chemical and electronic stability of
the STE relaxes all but 2.8 eV of the 10.4 eV potential of the classical exciton.
The singly-ionized STE and the electronically excited STE are unstable. The
excited STE decays either by returning to its own ground state by a fast charge
transfer process or by the destruction of the STE itself (driven by the elastic strain).
This charge transfer process has been observed and must be at least as fast as the
destruction process to be observed at all.
The STE is formed spontaneously from hole-electron pairs in silica and can be
formed by any process that provides the requisite 10.4 eV of energy needed to create
the pair. The lifetime of the STE in amorphous silica is on the order of 10 ps.
A Thought Experiment Involving the STE
Rather than leave the reader in the dark for several chapters as to why
the STE is important, the author believes that a small amount of argument
foreshadowing the eventual conclusions of this study is appropriate at this
point in the discussion. Consider the following thought experiment: What