SIMULATIONS AND SILICON WAFER COMPATIBILITY OF A
VOLTAGE-CONTROLLED OPTICAL SWITCH USING ITO/NbOx
By
Kevin Burghardt

A THESIS
Submitted to
Michigan State University
in partial fulfillments
for the degree of
Electrical Engineering - Master of Science
2017

ABSTRACT
SIMULATIONS AND SILICON WAFER COMPATIBILITY OF A
VOLTAGE-CONTROLLED OPTICAL SWITCH USING ITO/NbOx
By
Kevin Burghardt
The story of optics and processing has always been on of silicon devices making strides faster and
cheaper than optics. The idea of creating optical switches has been generally relegated to academic
exercises or niche markets. This research takes a view of optical processing that is complimentary to
silicon. Silicon wafers produce extremely dense, high quality devices but producing truly 3D integrated
circuits has been a challenge. It would be advantageous to not need to bond wafers to create a 3D
active structure. An argument for an optical switch that has a simple structure and uses industry
established fabrication methods is given. The proposed switch uses the material indium tin oxide
nanoparticles in niobum oxide glass (ITO/NbOx) as the active layer. The transmittance through this
material is proportional to the electric field applied to it meaning the structure of a capacitor could be
used to control it. It uses a metal for one plate of the capacitor and the ITO/NbOx as the other plate
with the light running through ITO/NbOx plate. Each of the plates are separated from one another and
surrounded by a dielectric material. Simulations show that silicon dioxide (SiO2) can be used
effectively to turn the ITO/NbOx into a light guide with a transmittance controllable using an applied
voltage and that the proposed structure can be created using industry established wafer fabrication
processes.

TABLE OF CONTENTS

LIST OF TABLES..........................................................................................................iv
LIST OF FIGURES.........................................................................................................v
INTRODUCTION............................................................................................................1
CHAPTER 1................................................................................................................3
LITERATURE REVIEW.....................................................................................................3
Optical Processing.............................................................................................3
3-Dimmensional Integrated Circuits.........................................................................6
Optical Switching..............................................................................................9
Optical Properties of Materials.............................................................................20
Sub-wavelength Light Guides................................................................................33
CHAPTER 2...............................................................................................................44
PROPOSED RESEARCH...................................................................................................44
CHAPTER 3...............................................................................................................47
TASKS.....................................................................................................................47
ITO/NbOx Material Evaluation for use as the Core.......................................................47
Baseline Simulation of a Square ITO/NbOx Light Guide.................................................52
Cladding Material Evaluation................................................................................59
Manufacturability .............................................................................................67
.

CHAPTER 4...............................................................................................................72
FUTURE WORK...........................................................................................................72
CHAPTER 5...............................................................................................................73
SUMMARY.................................................................................................................73
BIBLIOGRAPHY...........................................................................................................74

iii

LIST OF TABLES

Table 1. Transmittance................................................................................................48
Table 2. ITO/NbOx Refractive Index..................................................................................49
Table 3. Cross-sectional area comparison...........................................................................51
Table 4. Transmittance versus length...............................................................................51
Table 5. Baseline simulation set-up..................................................................................54
Table 6. Field reach from the edge of the light guide in y-axis for ITO34.......................................55
Table 7. Field reach from the edge of the light guide in x-axis for ITO34.......................................55
Table 8. Field reach from the edge of the light guide in y-axis for ITO69.......................................58
Table 9. Field reach from the edge of the light guide in x-axis for ITO69.......................................58
Table 10. Cladding material refractive indices.....................................................................60
Table 11. Field reach from the edge of the light guide in y-axis for ITO34/SiO2................................61
Table 12. Field reach from the edge of the light guide in x-axis for ITO34/SiO2...............................61
Table 13. Field reach from the edge of the light guide in y-axis for ITO34/Teflon-AF 1601..................62
Table 14. Field reach from the edge of the light guide in x-axis for ITO34/Teflon-AF 1601..................62
Table 15. Field reach from the edge of the light guide in y-axis for ITO34/nanoporous silica................63
Table 16. Field reach from the edge of the light guide in x-axis for ITO34/nanoporous silica................63
Table 17. Field reach from the edge of the light guide in y-axis for ITO69/SiO2................................64
Table 18. Field reach from the edge of the light guide in x-axis for ITO69/SiO2...............................64
Table 19. Field reach from the edge of the light guide in y-axis for ITO69/Teflon-AF 1601..................65
Table 20. Field reach from the edge of the light guide in x-axis for ITO69/Teflon-AF 1601..................65
Table 21. Field reach from the edge of the light guide in y-axis for ITO69/nanoporous silica................66
Table 22. Field reach from the edge of the light guide in x-axis for ITO69/nanoporous silica................66

iv

LIST OF FIGURES

Figure 1. Lenslet EnLight256 single board............................................................................5
Figure 2. Image of a FINFET.............................................................................................6
Figure 3. 3D IC Wafer Stack.............................................................................................7
Figure 4. 3D IC Cross-section TEM Image..............................................................................7
Figure 5. Liquid Crystal Optical Switch Example...................................................................12
Figure 6. SEM Image of MEMS Torsion Mirror........................................................................13
Figure 7. SEM Images of a VO2 Plasmonic Heater Switch..........................................................15
Figure 8. Magnetic field distribution of a self-collimated beam propagating through bend (a) and splitter
(b).........................................................................................................................16
Figure 9. Self-collimated AB switch..................................................................................17
Figure 10. Optical switch (a) and modulator (b) using gold nanowires on top of an SOI evanescent Xcoupler...................................................................................................................20
Figure 11. Schematic for plasmon oscillation for a sphere.......................................................22
Figure 12. Au-silica concentric shell particle and SEM images showing the silica particle (A), the Au
coating on the silica (B), the second layer of silica (C), and the second Au coating on the finished
concentric shell particle (D)...........................................................................................23
Figure 13. SEM images of Au nanoparticle pairs (diameter 150nm) spaced center to center distances of
450nm, 300nm and 150nm (left to right)............................................................................23
Figure 14. Schematic and SEM images of a BaTiO3-Au-SiO2 concentric shell nanoparticle.................24
Figure 15. VO2 transistor showing self-heating that is proportional to channel width in the VO2............26
Figure 16. Transmittance and reflectance of VO2 film and nanoparticles above and below the transition
temperature.............................................................................................................27
Figure 17. Au-coated VO2 nanoparticles.............................................................................28
Figure 18. Surface polariton plasmon propagating at the surface of the graphene in the region of LiNbO3
that is down polarized (a) and being absorbed in the depolarized region (b).................................29
Figure 19. Transmittance and refractive index of ITO that is prepared with RF sputtering on substrates
of varying temperature................................................................................................31
Figure 20. Optical extinction as a function of wavelength for differing toping levels of tin and the
concentric shell model of the resulting ITO nanoparticle.........................................................32
Figure 21. ITO/NbOx (34% ITO nanoparticles) transmittance at different voltages...........................33

v

Figure 22. SEM image of a sub-wavelength grating.................................................................34
Figure 23. Simulations of a Si-Ge subwavelength grating elbow and Y-splitter...............................36
Figure 24. Metal-dielectric-metal side coupled wave guide......................................................36
Figure 25. Aluminum cavity wave guide showing the electric fields for two different spacings of L = 30nm
(top) and L = 40nm (bottom) while keeping the ridge to upper plate gap constant at 10nm and a
wavelength of 1300nm.................................................................................................38
Figure 26. Wave guide with layered dielectrics....................................................................39
Figure 27. Observation method for measuring the light confinement of the TiO2 light guide...............40
Figure 28. Measurements of the TiO2 light guide with the incident beam entering the light guide (a) and
with it hitting other areas of the substrate (b) and (c) to confirm the observed light was in fact traveling
through the light guide and not some other effect................................................................41
Figure 29. High magnification stitched image showing the measured light confinement in the light
guide.....................................................................................................................42
Figure 30. SEM image of a TiO2 Y-splitter (top) and imaging of the measured light confinement
(bottom).................................................................................................................42
Figure 31. Comparison of proposed optical switch on Si wafer 3D IC versus a traditional wafer-on-wafer
structure.................................................................................................................44
Figure 32. Concept drawing of a capacitor based optical absorption switch...................................45
Figure 33. Cross-section and Nanoparticles.........................................................................50
Figure 34. Axis definitions for Matlab implementation of the extended Marcatili approach for rectangular
light guides..............................................................................................................53
Figure 35. ITO34 at 475nm (x and y axes are in Âľm)...............................................................54
Figure 36. ITO34 at 475nm Ex Zoomed...............................................................................54
Figure 37. ITO69 at 1.55Âľm (x and y axes are in Âľm)...............................................................56
Figure 38. ITO69 at 1.55Âľm Ex Zoomed..............................................................................56
Figure 39. ITO69 at 1.55Âľm Ez Zoomed..............................................................................56
Figure 40. ITO69 at 1.55Âľm (x and y axes are in Âľm). Width is 1Âľm and height is 500nm.....................57
Figure 41. ITO69 at 1.55Âľm Ex Zoomed (1Âľm x 500nm)............................................................58
Figure 42. ITO69 at 1.55Âľm Ez Zoomed (1Âľm x 500nm)............................................................58
Figure 43. ITO34/SiO2 at 475nm (x and y axes are in Âľm)..........................................................61
Figure 44. ITO34/Teflon-AF 1601 at 475nm (x and y axes are in Âľm)............................................62
Figure 45. ITO34/nanoporous silica at 475nm (x and y axes are in Âľm)..........................................63

vi

Figure 46. ITO69/SiO2 at 1.55Âľm (x and y axes are in Âľm).........................................................64
Figure 47. ITO69/Teflon-AF 1601 at 1.55Âľm (x and y axes are in Âľm)...........................................65
Figure 48. ITO69/nanoporous silica at 1.55Âľm (x and y axes are in Âľm).........................................66
Figure 49. SEM images comparing acid-catalyzed (a) to base-catalyzed (b) nanoporous silica..............69
Figure 50. Full stack of ITO/NbOx, SiO2, and Al creating the basis for the capacitors/switches.............70
Figure 51. Photo resist to create the actual individual switches.................................................70
Figure 52. After etching to produce the individual switches......................................................71
Figure 53. After final SiO2 deposition and any necessary planarization.........................................71

vii

INTRODUCTION
The fields of electronics and optics have had an increasing level of combined interested over the years.
Starting with things like LEDs, photo sensors, and fiber optics there has been more and more interest in
looking at the two as one. From an industry perspective, the focus has been finding an all optical
switching solution for fiber optic networks. Currently these systems generally are required to go
optical-electrical-optical in order to perform things like AB switching operations [1]. To this end, the
commercial focus for Silicon-based chips is geared specifically towards optical switching. While there
have been attempts at optical processors nothing has been made commercially available [2]. The main
reason for this is Silicon processes are well understood and the cost of creating devices with this
material is low enough that it is very difficult for devices based off of other materials to compete.
Additionally, device density can be pushed very small with the theoretical limit of Silicon-based
transistors down in the 10nm to 20nm range for gate width [3]. One drawback to Silicon-based MOSFET
technology is the RC constant associated with the gate capacitance and the necessity of moving that
charge from one part of a circuit to another. There is a physical limit to both how fast the fields can
be created in the channel region and to how fast the electrons can be moved onto and off of the gate
effectively limiting switching speed. While using photons in place of electrons will not result in moving
the particles at the absolute speed limit of light because they will be traveling through a medium, it
may be possible to push processing speeds faster. Moving directly to photon only based technology
would be a massive undertaking; however with recent advancements in materials, developing a hybrid
electron-photon device may be possible. The exact interaction between the electrons and photons
would be dependent on the material. Vanadium dioxide, for example, requires heating and cooling to
transition optical states so electrons would need to be able to manage localized temperatures. This
could be done through something as simple as ohmic heating or as complicated as an LED that converts
limited amounts of heat to light [4]. Another example, indium tin oxide (ITO) in niobium oxide glass
(NbOx), could potentially be controlled using the accumulation of charge similar to a traditional
MOSFET [5]. Each different method has pros and cons associated with its specific electron-photon
interaction. This approach represents an intermediary step between electron-based logic and photon-

1

based logic. The ultimate goal is system and fabrication process that will be simple enough to allow
for a relatively fast research to fabrication cycle allowing for greater flexibility in experimentation.

For logic or processing applications an electron-controlled photon switch would be needed. Optical
switching is once again an area where the focus on research has been on communications. The goal of
that research is the ability to keep all signals optical in a network. The existing ideas for optical
switching consists of AB or array switches. These types of switches require that light changes
directions often through changing the index of refraction of a material that the light passes through.
The advantage to this method is there usually aren’t any moving parts. Conversely, something like a
micro-mirror requires mechanical movement [1]. Even the original optical processor relies on a liquid
crystal cell to effect the phase of the light resulting in either constructive or destructive interference
[2]. Finding a way to create a light switch that uses existing fabrication techniques could be very
advantageous in that it could allow for the creation of 3D structures or multiple layers of active
structures that are simpler to make and design. The problem with the more traditional methods of
switching is the reliance on changing the direction that light travels. Once feature sizes get small
enough, things like edge effects and diffraction effectively start canceling out this change in direction.
The direction of propagation may be changing but the light is spreading out enough that any sensor
would still be illuminated. For this reason an absorptive switching scheme is being considered. In this
scheme, light is either blocked or allowed to pass. This could be thought of as a fiber optic cable in
which the core material could be turned “on” or “off” allowing light to pass or be absorbed
respectively. Using this method should allow structures to get smaller than the switches that rely on
light redirection. The trade-off is that diffraction and the edge effects become very large compared
to the light guide’s cross-sectional dimensions become smaller than the light’s wavelength. The
challenge then becomes managing these effects such that adjacent light guides are not affected by
cross-talk.

2

CHAPTER 1 LITERATURE REVIEW
Optical Processing
The subject of optics and processing has periodically made its way into articles as Silicon-based
semiconductors have met their periodic roadblocks as the industry has been scaling the technology
down. Prior to that, optics had a fairly rich period of research when the relative superiority of
semiconductors was less certain. This period of research and its subsequent evolution is detailed
extensively in an Advances in Optical Technologies article called “Optical Computing: A 60 Year
Adventure” [6]. The article discusses how the focus of optics and information processing has changed
from the goal of all optical to viewing it as a compliment to Silicon-based semiconductors. One of the
early goals was to develop methods of taking advantage of the parallel processing potential of optics.
This is a goal which has never really gone away. Even today there is a company in United Kingdom
called Optalysys which is researching using Fourier transforms for parallel processing with analog light
signals [7]. While parallel processing using light seems to be getting looked at again, advancements
have been made using Silicon-based CMOS designs in quantum computing which also claim parallel
processing capabilities [8, 9]. This is indicative of the historical relationship between optics and
Silicon. As it seems the field of optics starts making some progress, Silicon makes a comparable
advancement and is able to do so at either a smaller scale or cheaper cost. In this case, potentially
developing parallel processing utilizing existing fabrication methods would be a huge advantage over
anything in the optics field. Where optics has made a major impact on information processing is in
digital communications [6]. Fiber optics has largely enabled the revolution in connectivity and
information sharing that has defined beginning of the 21st century. Fiber optics provide a low loss high
data density method for transmitting information over long distances. Demands for faster speeds have
persisted highlighting the challenges to the interface between optics and electronics. It is here where
much of the research in optics related information processing has focused for the last twenty or so
years.

Optical processing can take place in an analog form often using some sort of planar parallel processing
or in digital form which can be thought of as more analogous to serial processing. In traditional digital

3

processing there is true serial processing like communications and traditional multibit processing like in
microprocessors and microcontrollers. While microprocessors and microcontrollers process multiple
bits during each clock cycle a planar parallel system can process multiple microprocessors with each
clock cycle representing another order of parallelism. For example a 64-bit system can process one 64
bit operation representing 264 possible unique values each clock cycle. A 64 x 64 bit planar system
could process 64 different 64-bit operations per clock cycle or all bits could be looked at together as
potentially representing 24096 possible unique values. This illustrates why the subject of parallel optical
processing never really fades away completely. Unless Si-based quantum computing becomes viable
there will always be a window for optical processing to do something traditional processors can’t. As
noted above, quantum computing may actually be on the horizon with the demonstration of a device
with the first reprogrammable qubits or quantum bits [10]. The work horse of optical based parallel
systems is the spatial light modulator (SLM). This devices allows for control of light in two dimensions
(either amplitude, phase, or both). The most common example give is an overhead projector where an
LCD is used to perform the modulation which in this example is the projected image. LCD pixels,
usually containing three complementary sub-pixels, either block light or allow it through recreating or
generating an image. The configuration of these pixels is updated as a function of time. LCDs are by
no means the only way to do this. They are appealing because they can translate a digital number into
a 2D image. This concept can also be used to perform logical operations. Since light is a wave, Fourier
transforms are often employed to do the processing where digital signals are converted to light,
processed, then converted back into digital signals or performed directly on light signals. An example
of this is a Fourier transform based system that encodes/decodes lower bit-rate tributaries into
extremely high 10.8 and 20.6 Tbit s-1 data streams [11].

Optical processing systems can be thought of as fitting into two broad categories with respect to
hardware: all-optical and hybrid optical-electrical. There are a few instances where all optical
operation is a real benefit such as communications where operations to be done on fiber signals [11,
12, 13, 14]. In this field, switching between electrical and optical is a relatively slow operation and is
very inefficient. The few times this conversion must be made the better. To transmit information or

4

very long fiber lengths, optical repeaters are needed to boost signal strength, reform the signals, and
retime the signal. These represent the three classifications of optical repeaters. An R repeater is one
that only does reamplification, a 2R repeater is one that does reamplification and pulse reshaping, and
a 3R repeater does all three including retiming [15]. While hybrid systems are not desirably in the
middle of communication chains, they are necessary since most technology people directly interact
with is electronic. In the early 2000’s a small start-up company named Lenslet announced an optical
processor making similar claims to Optalysys; however this endeavor was focused on a computer-like
processor as opposed to the large data/server type focus of Optalysys. Although they demonstrated a
working prototype in 2003, by 2006 the company had announced it was closing [16]. They produced a
single white paper describing a hybrid system called EnLight256 with the promise of a single chip
solution that never came [17].

Figure 1. Lenslet EnLight256 single board [17]
Hybrid systems employ some form of electronically controlled manipulation of optical properties.
Refractive index manipulation is common, but phase manipulation is also used [1, 2]. Traditionally

5

these methodologies have struggled to compete with Si-based with transistors on the basis of size. As
Si-based transistors have driven from microscale to nanoscale feature sizes with the FINFET (fin field
effect transistor) pushing those sizes to the potential physical limits of FET (field effect transistor)
technology, moving structures into three dimensions has become an increasingly import subject [18].

Figure 2. Image of a FINFET [19].

3-Dimmensional Integrated Circuits
Three-dimensional systems were proposed as far back as 1960 with a paper by Jim Early where he
discussed issues of temperature and device element density among others [20]. While current papers
discuss cost justification and fabrication methods/issues, the subject of temperature is almost always
discussed. Thermal budgets are nothing new to IC design. While two-dimensional systems can utilize
simple heatsinks effectively, three-dimensional systems run into significant issues that require more
advanced techniques such as designing the heat sink into the structure in the form of thermal vias
which in turn reduces the maximum theoretical device density [21, 22]. There are significant
challenges to three-dimensional structures reaching their promised potential, but there are
applications where it just makes sense. One such application is micro electrical mechanical systems
(MEMS) where the MEMS devices is fabricated using a Silicon-on-glass process and the electrical system

6

is most likely fabricated on a traditional CMOS process or something similar. Housing both die in the
same package in a stacked-die configuration makes the most sense [23]. Similar to MEMS, is the idea of
mixing wafers from different processes that are then connected together using through-silicon vias
(TSV) and wafer to wafer bonding [23]. In addition to this, is the desire to utilize multiple active layers
to develop truly three-dimensional wafer-fabricated structures. This would allow for separating
circuitry within the device in advantageous ways like locating memory overtop of a processor core or
placing all NMOS transistors physically below all their PMOS counterparts [24].

Figure 3. 3D IC Wafer Stack [23]

Figure 4. 3D IC Cross-section TEM Image [24]

There are two ways circuitry can be expanded in three dimensions: 3D packaging and 3D integrated
circuits. With 3D packaging multiple wafers created using traditional fabrication methods and then
connected using methods like wire bonding or flip chips to connect them. Integrated a MEMS die and a
traditionally processed integrated circuit is an example of this method. There are three methods of
accomplishing this which are described in various ways throughout literature: system in package (SIP),

7

3D wafer level package (WLP) and package on package (POP). SIP stacks separate dice connecting
them with either wire bonding or using bonding points like solder bumps. This is a useful method for
building single chip MEMS devices. 3D WLP is very similar to SIC in that TSVs are prepared at the wafer
level but the actual connections between dice are made at the package level [25]. POP uses finished
die that are connected together in stacks by using an interposer that is integrated into the chips.
Connections are made using ball grid array (BGA) connections during PCB processing such that one chip
has the solder points available on the top of the device where a different BGA device would get
installed [26]. In all of these cases, there is an advantage to testing where individual systems can be
tested prior to mating die or wafers together increasing the chances of a finished device that functions
properly. Using a fully realized 3D fabrication method complicates testing as a defect on any of the
active layers can cause the entire system to fail [27].

The other way of moving into three-dimensions is to make a truly three-dimensional integrated circuit.
There are two ways of doing this: 3D stacked integrated circuit (3D SIC) and Monolithic 3D integrated
circuits. As previously mentioned, 3D WLP and 3D SIC are somewhat similar since both utilize TSVs, but
3D SIC uses TSVs to directly connect multiple wafers using things like wafer thinning and Cu-Cu
contacts. In this scheme, wafers are connected to a dummy wafer for mechanical support, then
thinned down as thin as 100Âľm then connected to another the main wafer stack using wafer alignment
techniques [23, 24]. While both wafers are tested prior to being mated there is a risk that a nonfunctional die location on the wafer stack is mated with a functional die location on the thinned wafer
or vice versa thus reducing overall yield [25]. Although this risk exists, being able to mate so many
dice at one time does have its advantages over having to mate them one at a time. Monolithic 3D ICs
are created using a sort of bottom-up type process [24]. This is where active layers are processed
directly onto a wafer using a technique such as Silicon on Insulator (SOI). This methodology solves
many of the challenges posed by TSVs like wafer alignment and the fact that Cu-Cu contacts present
field risks in large-scale manufacturing environments as the switch from Au to Cu bond wires has
demonstrated; however it offers its own challenges such as processing temperature. Doping processes
can be upwards of 1000°C, but solid-phase epitaxy can add a doped layer of Si around 600°C. This

8

process is done by laying down a layer of doping atoms and amorphous Si then heating everything up to
crystalize the Si [28]. Creating Monolithic 3D ICs using simpler, more widely established fabrication
methods would be advantageous. For example Indium-Tin-Oxide nanoparticles in Niobium Oxide
(ITO/NbOx) is a thin-film material that can be deposited using spin-coating then annealed at 400°C
[29]. This material however would have to be used with an optical switching scheme rather than an
electrical one using traditional transistors. The three pieces needed for such a scheme would be a
metal layer, a dielectric layer, and the ITO/NbOx layer which could all be deposited using industry
established methods then etched; the details of which will be discussed later. The tradeoff here is
that in order to use simpler fabrication methods, a change from traditional transistors to optical
switching would have to happen.

Optical Switching
For reasons previously mentioned, optical switching has generally been tied to telecommunications and
data networks as light has always had distinct advantages over wired electrical signals over great
distances. Using fiber optic cables as allowed systems to get around Ohm’s Law and improve
efficiency. Data network usage has drastically increased placing demands on capacity and power that
has called for more efficient switches and fewer optical/electrical transitions [30, 31]. This focus on
long range transmission and serial data interfaces has affected the constraints placed on problems
being researched. In general, the goal is to do all-optical switching of traditional fiber-optic signals.
This results in a number of constraints that do not translate well to integrated circuits. One such
constraint is the core of a fiber must be larger than the wavelength of the light that travels through it,
usually by about five to ten times. This is done to minimize the effects of the evanescent field and
diffraction since they remain small compared to the size of the fiber and thus can be neglected. Since
integrated circuits have feature sizes from the low micro-scale to the nano-scale that would mean the
wavelength of light needed would range from the visible range (1Âľm/1.24eV) to just into the x-ray
range (1nm/1.24keV). While a 1Âľm wavelength might seem feasible, this would be very old technology
with 10Âľm feature sizes (assuming a 10x relationship between wavelength and feature size). A more
common feature size at the present time is 100nm corresponding to wavelengths of 10nm and an

9

energy of 124eV which is in the ultraviolet range. Even a feature size like this is relatively old
technology with processes now down to 14nm [3]. This is related to another common design constraint
where beam theory is used for the light. This is seen in almost every switching scheme where the
effects of evanescence and diffraction are not dealt with in favor of demonstrating proof of concept
most often by demonstrating the material properties using thin-films. These constraints make perfect
sense when considering their intended use in component-level designs where the entire device is
dedicated to one or two switching operations. These are the designs that are practical for large data
networks. Though any reliable miniaturization or improvement in efficiency or switching speed would
be welcomed by data centers, a large leap in such things are not necessary for data centers to realize
impactful improvements. While component-level processes and physics do not necessarily translate to
wafer-level processes it is well worth considering this large body of work.

An optical switch is a device that controls the flow of photons by a number of effects, with each effect
defining the switch’s type. Most of these effects rely on changing the direction of the propagation of a
beam of light, often by changing the refractive index of a material like in bubble switches. MEMS
devices can also directly change the direction of a light beam using micro mirrors. Another approach is
to find some what of blocking or allowing light to pass such as dual polarization like in LCDs or using
destructive interference like in thermo-optic or Mach-Zehnder interferometers [1].

Optical switching by manipulating refractive index is a very common approach. This is because it is
relatively easy to do, is extremely closely related to traditional fiber optic theory, and provides an
acceptable level of accuracy in large data networks where switching speeds are at a far greater
premium than physical device size. This category often ends up being very similar with absorptive and
interference switching as the physical mechanisms end up being shared between the two. The main
difference between the two categories lies in what is done with the beam of light. In refractive index
switching the beam of light’s trajectory is changed such that it leaves the device at different location
relative to its original state. This could be thought of as an AB switch or switching between an A and a
B output. Conversely, absorption/interference switching blocks light or allows it to pass. Examples of

10

refractive index switching are liquid crystal switches, bubble switches, and electro-holographic
switches [1]. All of these examples use various physical conditions to change refractive index by
applying a voltage to a material. Liquid crystals are perhaps the most familiar as these have been used
on very large scales in screens for many years now. In screens, three liquid crystals are placed very
close to one another in a grouping commonly called a pixel. Each of the three switches, called subpixels, manipulate a specific band of light with each of the three bands making up a complimentary set
of colors. The most common set is red, green, and blue. In this configuration, the job of the liquid
crystal is to the polarization of the light which is then put through passive polarizer. If the light
passing through the sub-pixel is polarized in one orientation, then it is blocked. If it polarized in the
opposite orientation, then it passes out of the screen. The three sub-pixels combine to create a signal
pixel with a color that can be controlled with voltage and the many pixels combine to create an image.
In this case, the liquid crystals are controlling polarization where as in the communications the main
goal is to move a beam of light from one place to another. Using a liquid crystal to perform a direction
AB switching operation is simply a matter of having two output ports: one port located in line with the
expected trajectory of the light traveling through the liquid crystal in the passive state and one port
located in line with the expected trajectory of the light traveling through the liquid crystal in the
active state. The active state is achieved by applying a voltage to the liquid crystal. In the case of
liquid crystal in polymer, a simple structure is employed where as a traditional liquid crystal requires
the use of polarizers to achieve the same effect [1].

11

Figure 5. Liquid Crystal Optical Switch Example [1].

An example of a polymer-stabilized liquid crystal switch is a cholesteric liquid crystal positioned
between two Si plates. The unique aspect of cholesteric devices is their potential to continue
functioning when power is removed which has obvious advantages in mobile technology. This specific
example considers infrared wavelengths which of course are of interest in communication networks
[32]. It should be noted that this particular case doesn’t make any claims about power usage as its
focus is on the refractive index and polarization. As previously mentioned, liquid crystals generally
have heavy polarization dependencies which combined with passive polarizers makes them a good fit
for displays. What makes this liquid crystal unique is that it is able to effect the refractive index in a
way that is independent of polarization and is also isotropic. This is demonstrated using a Fabry-Perot

12

interferometer where the polymer-stabilized liquid crystal is positioned between two reflective plates
which in this case are Si plates. Experimentation with a gap between the plates of 20Âľm with respect
wavelengths showed modifications to the refractive index changing from 1.58 to 1.49 at 18V with no
appreciable change to polarization although voltages lower than 18V did show some continued
polarization dependence [32].

MEMS switching consists of using electrostatic control of micro mirrors to redirect light from one port
to another. They are generally considered to be highly efficient from an optical loss perspective while
also producing very little crosstalk but those strengths are tempered by relatively slow switching
speeds and high voltages needed to control the micro mirrors. These voltages often need to be on the
order of 10’s of volts to nearly 100V to manipulate them, although newer designs have been shown to
actuate as low as 0.5V [33, 34].

Figure 6. SEM Image of MEMS Torsion Mirror [33].

13

Traditionally, MEMS optical switches relatively slow switching speed best suits them for connections
sending large bursts of data over them. Once the switch is configured it is relative cheap from a power
perspective thus creating a tradeoff compromising transition speed for lower power. MEMS switches
can be used in combination with switches based on other technology which transition faster to create
systems that can leverage the strengths of MEMS switching while also using the versatility of the other
technologies to meet both the bus load and plasticity demands of modern networks [30]. The micro
mirror can be applied in a number of ways. Torsion mirrors pivot a micro mirror in one or two
dimensions where the light comes in at some angle from above the die surface and is angled to another
location in the device. Changing the angle of the mirror thus redirects the beam a light to a different
port. Another approach is the direct light to a port using a fixed mirror then having another mirror
that can be moved into the path of the light directing it to a different port [35]. Since most designs
rely on beam theory, smaller scale structures are problematic. Even designs that use diffraction to
their advantage are still technically using beam theory for much of the system such as is the case with
a switch combining a micro mirror and a Fourier lens [36].

Aside from redirecting light, it can be blocked. This can be accomplished in two ways: absorption and
destructive interference. Absorption is kind of the brute force method to switching. Instead of
worrying about exactly where the beam is headed or what other effects may be present, it’s all sent
through or none is. This is analogous to digital logic. It differs from digital logic most critically in the
area of power. MOSFETs work by using fields to allow charge flow or prevent it. The main power draw
in these devices is the actually switching which consists of the charging and discharging of the gate
capacitors. In absorptive switching the device must absorb and dissipate all the energy entering it
when in the blocking state which means heat. Heat management is a very important issue when
discussing any IC especially 3D ICs. Absorption and destructive interference is an area where research
has focused on in recent years. Nanowires and nanoscale objects are of specific interest as researchers
keep pushing dimensions smaller with MOSFET gates in the 20nm range. In nano-structures, light
interacts with materials by a process called plasmonics. This is concept that helps describe how things
like metals interact with light since the large number of mobile electrons in metals makes them behave

14

similarly to plasmas in electromagnetic fields. Plasmonics does a good job describing how light with a
wavelength greater than the dimensions of a nano-structure interacts with that structure and will be
discussed in greater detail in a later section about material properties. An example of an optical
switching using plasmonics is a low voltage broad band hybrid plasmonic-vanadium dioxide switch [37].

Figure 7. SEM Images of a VO2 Plasmonic Heater Switch [37].

This is a sub-wavelength plasmonic switch relying on VO2’s phase changing properties. The switch uses
grating-couplers made from the VO2 light guide overtop of an ohmic heater built out of silver to create
the switch. The grating-couplers exist to get the light into and out of the switch with the active region
of the switch being the VO2 light guide. A voltage is applied to the silver electrodes produces localized
heat under the VO2 light guide causing the phase shift resulting in an increase in extinction ratio: direct
absorption in VO2. It serves as an important proof of concept with the dimensions of the switch being
roughly 80Âľm by 60Âľm. It should be noted that most of this area is taken up by structures necessary to
properly test the concept of the device like the electrodes which are significantly larger then would
necessary in a fully integrated device. The width of the VO2 light guide is 8Âľm, the thickness is 280nm,
and it has a length varying from 5Âľm to 15Âľm and the wavelength of the light being 1.55Âľm. While this
is not small enough for dense integrated circuits, it does demonstrate the principles well. It should

15

also be noted that wave-guide dimensions smaller than 10x the wavelength of the light traveling
through them can be considered sub-wavelength since below this point all the issues associated with
sub-wavelength light guides begin to come into play. This VO2 switch has a light guide only 5x larger in
width and a little over 5x smaller in thickness than the light’s wavelength. Due to silver being used for
the heating element a low activation voltage of 0.4V was reported [37]. Based on the dimensions given
in the paper, an activation voltage of 0.4V and an assumed localized temperature around 75C
corresponds to something on the order of 40ÂľW.

A type of destructive interference switching based on the self-collimation effect in photonic crystals
has been proposed on a theoretical basis since at least the early 2000’s [38, 39]. The self-collimation
effect is that beams of light propagating through perfect photonic crystals effectively ignore many of
the unwanted small feature size effects that cause the beam to disperse.

Figure 8. Magnetic field distribution of a self-collimated beam propagating through bend (a) and
splitter (b) [38]

Feature sizes claimed in these papers are on the order of lattice constants of the photonic crystals.
For one such example, the photonic crystal is silicon rods in air [39]. This design consists of a square
layer of silicon with air holes arranged in a repeating lattice structure. Along one diagonal of the
structure there is a single line of air holes with a smaller radius than the rest of the structure. This

16

creates a 3dB splitter where a beam of light entering on side of the structure is outputted both in the
transmitted direction but also one perpendicular to the original trajectory in the reflected direction.
The transmitted beam sees no phase shift but the beam that is reflected at a 90° angle and is phase
shifted 90°. The inputs are perpendicular to one another as are the outputs such that there is always a
transmitted direction and a perpendicular direction for each input. One input’s transmitted direction
is the same as the other input’s perpendicular reflected direction. If a beam of light of the same
wavelength enters both inputs with one of the beams phase shifted 180° then one output sees
constructive interference and the other output sees destructive interference. Exactly which output
sees what, is a function of the incoming beam’s phase shift. The proposed design is a 12√2a by 12√2a
square of silicon with air having a radius to lattice constant of 0.35 [39].

Figure 9. Self-collimated AB switch [39].

The proposed design and subsequent simulations do not consider any finished dimensions but rather
keep things in normalized terms. The dimensions of photonic crystals are generally referenced to the
desired wavelength. For example, a common wavelength in fiber optic networks is 1550nm making the
lattice constant about 525.5nm [40]. This would make the proposed structure about 8,918nm by
8,918nm which is quite large when compared against current FET technologies. Extrapolating this
relationship out to the limits of visible light results in structures with side lengths of 4,027nm (Îť =
700nm, red) and 2,301nm (Îť = 400nm, blue). Even into the near ultraviolet range the side lengths only

17

get down to 575.3nm (Îť = 100nm, ultraviolet) which is a full order of magnitude larger than current
FET process feature sizes. Despite this possible short coming for integrated circuits, this design
provides an interesting possible solid state switching solution for communication networks which is
likely its intended application.

The switches discussed thus far focus on devices that switch an optical beam on or off. There is also
the need to eventually convert that light back into electricity if used in an integrated circuit. While
this is a deep field of study by itself and falls somewhat out of the scope of this paper, it is worth
mentioning a couple of them. ZnO nanowires were shown to have an optically induced, reversible
resistance change making them a potential option for photon/electron conversion [41]. The nanowires
were analyzed with diameters ranging from 50nm to 300nm. They were exposed to light with
wavelengths below 380nm and were found to have resistance changes of 4 to 6 orders of magnitude
with a high level of sensitivity. It was also confirmed the ZnO nanowires were not responsive to
wavelengths above 370nm which is a material dependent cut-off for the effect [41]. InP nanowires
were shown to exhibit anisotropy with respect to polarization [42]. These were analyzed in air such
that the refractive index difference between the air and the material was larger than it would be in a
normal semiconducting environment. In addition to this sensitivity to orientation, the nanowires were
observed to change their conductance by two to three orders of magnitude [42]. These findings
suggest the materials could be an interesting option for detection applications since any switching
operations would require changing the orientation of a beam of light or the nanowires themselves.
Positioned correctly with respect to the phase of light guide, nanowires can also be used to absorb
light. An example of gold nanowires placed on top of SOI waveguides shows how the phase of the light
effective can control whether it is absorbed in the nanowires or not [43]. Unlike some of the other
nanowire examples given, this is a direct plasmonic effect. The gold nanowires where simulated with
the width and thickness held constant at 40nm and 25nm respectively. The antennas were laid on top
of and perpendicular to the waveguide such that varying the length of nanowire effectively changed
the cross-sectional area the nanowire could cover. The simulations showed that the optimal condition
was three equal sections of gold nanowire 105nm long, spaced 40nm apart for a 400nm wide and 200nm

18

thick Si waveguide and a wavelength of 1550nm. It is important to note that the effectiveness of the
nanowires is wavelength dependent. The interactions between nanowires and light can be thought in a
similar fashion to electromagnetic waves interacting with antennas whose lengths are shorter than the
fundamental wavelength of the incoming wave. Two waves are fed into the waveguide with the same
frequency and phase matching such that a standing wave is produced with the nanowires positioned at
one of the power minimums thus minimal absorption by the nano wires takes place. By shifting the
phase of this standing wave, the location of the power maximum shifts to the same location as the
nanowires allowing for maximum absorption. An example of combining this basic structure into a
switch is given using an evanescent X-coupler. This is a device that routes two waveguides very close
together such that the evanescent field from one guide bleeds into the adjacent waveguide
transferring some power between the two. Each light guide has three rows of three nanowires
positioned in same place with respect to its adjacent lightguide’s nanowires. Waves enter both inputs
with the transmission or absorption depending on the phase of the resulting standing wave. In the
event of absorption, the standing wave’s power maximum is shifted to the location of the nanowires
providing maximum absorption and minimum power transferred through to the two outputs.
Conversely when the power minimum is shifted to the location of the nanowires, minimum absorption
occurs allowing for the maximum power to be transferred to the two outputs. This effectively creates
a phase dependent switch which functions similarly to a Mach-Zehnder interferometer. The key
difference between the two is a Mach-Zehnder interferometer functions more like an AB switch rather
than an on/off switch [43]. This example is still large by comparison with the full structure being at
least in the low micrometer range. It is interesting from the perspective of relying on more readily
available manufacturing processes like SOI and metal deposition. Reading the various properties of
different nanowires illustrates the material dependencies there are in the subject of lightguides and
optical switching.

19

Figure 10. Optical switch (a) and modulator (b) using gold nanowires on top of an SOI evanescent Xcoupler [43].

Optical Properties of Materials
The optical properties of materials essentially is a discussion of how electromagnetic radiation
interacts with materials. Generally when talking about optics, the discussion is limited to the
wavelengths around the spectrum that is visible to the human eye. With this distinction in mind, the
properties important to consider are absorbance, birefringence, reflectance, refractive index, and
transmittance. Absorbance, reflectance, and transmittance effective describe what happens to the
energy at a material interface and in the material. Reflectance describes how much energy is
reflected back from the material interface, and transmittance describes how much energy moves
through the material interface. Absorbance describes how much energy is lost to the material as the
radiation moves through it usually in the form of photons transferring energy to electrons. Refractive
index is the relationship between the radiation’s phase velocity in the material and in a vacuum. The
classical consideration of this property is to describe how a light ray is bent moving through a material
interface. Birefringence is a property where the refractive index of a material is a function of
polarization. Often a polarization independent switch is desired which allows for a more generic
switch that can easily be placed into any situation and still function, but this is situationally
dependent. For switching applications material properties that are not static are desired. An example

20

of a liquid crystal that has a polarization-independent change in refractive index was given in the prior
section [32]. Other switches rely on changing the absorbance of a material. For effective integration
into electrical systems, this changed would need to be controlled electrically whether directly like in
the case of electrochromic materials or indirectly like with VO2.

Plasmonics is an effective way of describing how light interacts with electrons in materials like metal.
In metal, there are a large number of free electrons such that can interact with an electric field. This
is similar to a plasma where all the electrons of a material have been given enough energy to be free
electrons. In the presence of an electric field, these free electrons are attracted or repelled by the
field and thus move. When this electric field is alternating like is the case with light, these electrons
are alternately attracted then repelled as the polarity of the field changes. This interaction absorbs
some of the energy from the light; however this attenuation is wavelength dependent. Eventually, the
electrons reach a maximum speed as they move back and forth and increasing the frequency no longer
has any effect. This higher frequency light can pass through the plasma with no effect due to the
electron attraction/repulsion interaction. Since in metal there is a similarly large number of free
electrons, there is a very similar interaction between light and the electrons. Like in a plasma, at
some point the electrons reach a maximum speed and can no longer interact with higher frequency
light. The frequency at which this happens is proportional to the color seen reflected back from a
metal. This is only the most obvious and observable effect described by plasmonics. What makes
plasmonics of interest in optics is the electron attraction/repulsion effect holds true for objects much
smaller than the wavelength of the light: for example how infrared light (λ≈1550µm) interacts with a
gold nanoparticle (d≈20nm). Considering the simplest case where a spherical particle is exposed to
light, the electromagnetic field pushes and pulls on the electron cloud in the material due to the
Coulomb force; the motion of which is called a dipole plasmon resonance of the particle [44].

21

Figure 11. Schematic for plasmon oscillation for a sphere [44].

When half the electron cloud moves parallel to the electromagnetic field and half moves
perpendicular, the plasmon excitation is called a quadrupole mode. Relating higher mode plasmon
resonance can be very complicated like when it relates to d-orbital electrons, but this can be related
to the dielectric constant of the material. In the case where the particle is much smaller than the
wavelength of the light, the electromagnetic field can be considered constant. This is known as quasistatic approximation. For more complex particle shapes, numerical methods are needed to describe
their plasmon resonance. The Modified Long Wavelength Approximation for spheroids is based on the
quasi-static approximation which can effectively show the red shifting as the spheroid becomes oblate.
Beyond that, an effective way to describe plasmon resonance is the Discrete Dipole Approximation
(DDA) which looks at structures as made up of individual discrete dipoles which have solutions to
Maxwell’s equations, and solve for the final structure numerically. It should be noted that other
numerical methods exist such as multiple mutipole method and the finite difference time domain
method each with their own advantages and disadvantages [44]. Looking at complex structures as
composites of simpler geometries has been used to look at things like triangles, cubes, cylinders,
spheroids and concentric spherical shells [44, 45, 46].

22

Figure 12. Au-silica concentric shell particle and SEM images showing the silica particle (A), the Au
coating on the silica (B), the second layer of silica (C), and the second Au coating on the finished
concentric shell particle (D) [45].

In addition to the shape of a particle having a major influence, the proximity of a particle to other
particles also has an impact. The closer two particles get to one another, the more they act like a
single larger particle. In the case of two spherical particles, they will act increasingly as a single
spheroidal particle red shifting if aligned in the parallel direction and blue shifting if aligned in the
perpendicular direction [47]. In the parallel direction, the two spherical dipoles reinforce one another
since the positive end of one is near the negative end of the other resulting in an additional attraction
force in addition to the applied electromagnetic field. In the perpendicular case, the ends of the
dipoles end up near their like-charges resulting in a force that effectively works against the applied
field [47].

Figure 13. SEM images of Au nanoparticle pairs (diameter 150nm) spaced center to center distances of
450nm, 300nm and 150nm (left to right) [47].

23

Theoretical study of the plasmon resonance behavior of nanoparticles shows a very strong dependence
on size and shape which has interesting implications, but it also yield very practical problems for
experimentation. If one wants to gather measurements on spherical nanoparticles, how closely the
particles are to perfect spheres will have a major impact on the results. Often making large quantities
of particles results in a relatively large variation of sizes and shapes which both theoretical work and
experimentation have shown to significantly change the results [48]. While complex shapes can be
challenging, they also have been shown to have interesting results. Such is the case with a spherical
concentric shell made from BaTiO3-Au-SiO2. The core is made from BaTiO3 with a very thin shell of Au
which is then surrounded in SiO2. This structure was shown to produce a difference-frequency
generation effect which is a stepping stone in developing an optical parametric amplifier [49].

Figure 14. Schematic and SEM images of a BaTiO3-Au-SiO2 concentric shell nanoparticle [49].

The difference-frequency generator discussed here effectively combines an input signal field with a
very strong pump field producing an amplified field with the wavelength that is the difference between
the signal and pump fields. Another interesting practical application is using surface plasmons to probe
semiconductors in a touchless manner [50]. The surface plasmon effect has primarily been studied in
metals, but technically anything with an excess of charge carriers has this effect contributing to its

24

behavior in the presence of an electromagnetic field. Charge carriers in metals are on the order of
1023cm-1 while Si-based semiconductors can be on the order of 1016 to 1021cm-1 depending on doping.
Since current doping processes are inaccurate when compared to the needs of controlling nanoparticle
properties, significant challenges exist; however being able to probe using surface plasmons provides a
new way to determine localized properties of materials and these effects become more dominant as
structures get smaller.

A material that garners attention due to its relatively unique properties is Vanadium Dioxide. This
material is able to undergo a phase transition at relatively low temperatures. This phase transition
happens between 66°C and 68°C. Specific to the material’s electrical properties, there is a
semiconducting to metal transition where properties like resistance and opacity in the infrared range
can change by up to four orders of magnitude. Below the transition temperature, VO2 is transparent to
infrared wavelengths and above in the metallic state becoming reflective [51]. It should be noted that
sources alternate between calling the room temperature phase of VO2 between “insulating” and
“semiconducting”. The physical effect being described is a sort of low conductance. Likely papers
discussing the material properties in terms of resistance are more likely to call the low conductance
state “insulating” whereas papers looking at other properties might be more likely call it semiconducting. “Semiconducting” is probably a more accurate description since the effect is reversible
and doesn’t cause any obvious physical damage which makes for a good distinction between
“insulating” and “semiconducting”. While the semiconducting and metallic phase transition is a
temperature based effect, it has been shown be possible to locally affect material properties in a way
that is analogous to Si-based transistors or diodes with a current conducting channel whose width varies
proportional to the current flowing through it. This demonstration places two Ti/Au electrodes on a
layer of VO2 with a 15Âľm gap between them [52]. The increase in current flowing through the channel
has a self-heating effect where more current means more ohmic-heating and thus more VO2
transitioning into the metallic phase with the maximum width of the channel being the width of the
electrodes at the gap.

25

Figure 15. VO2 transistor showing self-heating that is proportional to channel width in the VO2 [52].

In the metallic phase, the transmittance decreases. Comparing VO2 thin films to VO2 nanoparticles
shows a significant difference in this property. The thin film shows superior transmittance changes but
this comes at cost. The overall transmittance is lower and the reflectance is higher in the
semiconducting state [51]. Additionally the greatest transmittance difference between the
semiconducting and metallic states happens at long wavelengths, 2000Âľm to 2500Âľm. The
nanoparticles show a region near 1000Âľm to 1500Âľm that shows a comparable transmittance
difference. Also, the overall reflectance of the nanoparticles is quite low. Smaller wavelengths mean
smaller feature sizes and less reflectance means less light is reflected at interfaces which helps with
managing noise when devices are jammed very close together.

26

Figure 16. Transmittance and reflectance of VO2 film and nanoparticles above and below the transition
temperature [51].

Experimentation by adding gold nanoparticles to VO2 showed a change in resistivity of four orders of
magnitude and a small decrease in transition temperature [53]. The sample was prepared by
alternating a number of laser pulses between high purity Vanadium and Au targets in the presence of a
Sapphire substrate until the total thickness of the VO2/Au was 200nm. All the other parameters where
in line with pure VO2 but the transitions temperature was about 3.3K less when heating up and about
3.5K less while cooling down. This difference was attributed to the presence of Au nanoparticles
acting as electron sources. This idea was taken to the nanoscale where Au nanoparticle coated VO2
nanoparticles where shown to have ultrafast switching properties. Picosecond laser pulses where used
to generate hot electrons that are injected across the Au/VO2 interface inducing sub-picosecond phase
transitions of the VO2 [54]. Samples of nanoscale VO2 coated with Au nanomesh were hit with 60ps
laser pulses with the samples near the 68C transition point. This produced a nearly step-like response

27

as measured by monitoring the response at 3100nm which was chosen to maximize the difference
between the metallic and semiconductor states.

Figure 17. Au-coated VO2 nanoparticles [54].

Graphene is a form of carbon that is effectively two-dimensional. It is a single atom thick lattice of six
carbon atoms linked in a hexagonal configuration. Graphene has many interesting properties like its
high strength to weight ratio, electron mobility, and thermal conductivity. Graphene by itself is a
more like a metal than a semiconductor with a 0 eV band gap. Introducing oxygen produces a non-zero
bandgap. It has been demonstrated that graphene oxide can have its bandgap, and thus its optical
properties, tuned by using IR radiation, varying the exposure time and power to shift the absorption
and photoluminescence [55]. In addition to graphene oxide, Graphene ribbons have shown optical
properties like photoluminescence as well. These properties are highly dependent on the size and
shape of the graphene [56]. Graphene has also shown charge carrier mobility dependence on substrate
material as well. SiO2 is often used, but the roughness which can cause scattering or charge traps.
Hexagonal boron nitride has been shown to improve mobility by an order of magnitude. Atomically flat
substrates produce a very low mobility while layered dichalcogenides, semiconductors made from
transitional metals and elements from the oxygen family, appear to have the potential for good
substrate material [57]. This brings up the potential need for substrate material that is less common
and most likely more expensive than Si. A surface plasmon polariton (coupled photon and electron)
modulator has been proposed using graphene on a lithium niobate substrate. Depending on the

28

polarization of the lithium niobate, a surface plasmon polariton in the graphene will be absorbed or
pass through [58]. When the lithium niobate is down polarized, the surface plasmon polariton energy is
highly consentrated near the surface of the graphene and when the lithium niobate is depolarized, the
energy flows into the graphene where it is absorbed. Control over the polarization of the lithium
niobate results in the electrostatic doping of the graphene with the example given in literature being
p-doped graphene in the down polarized state [58].

Figure 18. Surface polariton plasmon propagating at the surface of the graphene in the region of LiNbO3
that is down polarized (a) and being absorbed in the depolarized region (b) [58].

Combining graphene with other materials like cadmium sulfide (CdS) has been shown experimentally to
enhance the nonlinear optical properties of organic glass like polymethyl methacylate [59]. Testing
compared samples with varying degrees of graphene/CdS with respect to transmittance and absorption.
Interestingly, the results showed that simply increasing the graphen/CdS didn’t necessarily increase the
absorbance of the material. Various concentrations and combinations were tested such as the glass by

29

itself, graphene in glass, CdS in glass, and the various concentrations of graphene/CdS. There was one
specific concentration of graphene/CdS that appeared to outperform the others (7.8mg dispersed in
the glass). These results are attributed to three things: the structure of graphene causes the material
to have a large number of delocalized electrons which enhance nonlinear optical effects, the CdS
nanoparticles on the graphene allow for a large number of light active sites, and the charge transfer
between the graphene and CdS suppresses charge recombination [59].

Indium tin oxide (ITO) has been used in displays for many years because of its ability to conduct current
while allowing light to pass. This is used as a thin film on glass that functions to connect things like
LCD pixels with their driver chip pins. ITO is an n-type degenerate semiconductor with an optical
bandgap of around 3.75eV. This bandgap makes it mostly transparent in the visible light range and
opaque in the infrared and near infrared range [60, 61]. The conductivity of the ITO comes from the
high carrier concentrations which result from oxygen vacancies and tin dopants [61]. The form that
ITO comes in can have an impact on its properties. Experimentation with RF sputtering on quartz
substrates show significant metal grain structure variation with respect to substrate temperature. This
ranges from amorphous ITO to polycrystalline which was verified both in SEM imaging and X-ray
diffraction measurements. In addition to grain structure changes, the temperature also resulted in an
increase in the carrier concentration that tracked increasing substrate temperature and a shifting and
changing of the refractive index curve [61].

Figure 19. Transmittance and refractive index of ITO that is prepared with RF sputtering on substrates
of varying temperature [61].

30

Thin films made from ITO nanoparticles experience similar results during annealing. Experimentation
varying annealing temperature and time showed similar results to the RF sputtering on the heated
substrates. This testing showed a similar change in conductance near the 250C mark, but then went on
to show the conductance begins increasing again. In this study, the decrease in conductance near the
250C mark is attributed to a decrease in carrier concentration due to oxygen diffusion into the ITO
[62]. It attributes the increase in conductance at very high annealing temperatures (up to 1000C) to
sintering of the ITO nanoparticles which was shown visually through SEM imaging. Similar results were
reported by an earlier study which showed the same sintering at 800C along with the same changes in
conductance and transmittance [63]. It was also shown that annealing time also can result in a
decrease of conductance by testing various times while annealing at 550C. This was carried out in ever
increasing time increments until eventually annealing for 4080 minutes, showing the conductance
effectively maxes out at the same value as the initial peak near the 25 minute mark. In addition to
this, both annealing temperature and time was shown to also increase the range of wavelengths with
good transmittance. This was attributed to the decrease in carrier concentration which was reported
to happen faster than the decrease in measured conductance [62]. ITO nanoparticles have also been
shown to have dependencies on Sn doping levels. This produces a shift in the absorption that is
associated with charge carrier concentrations which is consistent with ITO bulk and thin-film results
[64]. As dopant levels increased, it was noted that the doping became less uniform and began showing
significant sample to sample variation. This shifting was explained as a Sn building up on the outsides
of the nanoparticles resulting in a structure that is closer to a concentric spherical shell described in
several papers on plasmonics [45, 46]. The resulting of this Sn build-up was a semiconductor-like
behavior with the center of the ITO particle having a different Sn content than the outside.
Conductivity experiments supported the Drude model descriptions of particles having large potential
barrier from particle to particle consistent a depletion region near the surfaces of the particles [64].

31

Figure 20. Optical extinction as a function of wavelength for differing toping levels of tin and the
concentric shell model of the resulting ITO nanoparticle [64].

The conductivity experiments showed much higher carrier concentrations in samples with significant
levels of Sn surface segregation. The study found good agreement between the Drude model and the
experimental results suggesting their theory of nanoparticle Sn surface segregation is the primary cause
for their observations. ITO is very good at absorbing wavelengths in the infrared range. Combining ITO
nanoparticles with niobate glass produces a material that has an extended range of frequencies it can
absorb. This material produces an amorphous thin-film that is able to absorb light in both the infrared
and visible range and do so selectively as a function of voltage [5]. This material is able to transition
between three states: transparent when 4V is applied, blocking in the infrared range when between
1.5V and 4V is applied, and blocking in both the infrared and visible ranges when less than 1.5V is
applied. These three points are obviously generalized, but the study’s data shows various
concentration levels of ITO nanoparticles compared against baselines of pure niobate glass and ITO
nanoparticle thin film. There is a little bit of a tradeoff in the concentration levels where the
absorbance of the ITO decreases but the absorbance of the niobate glass increases as more the ITO
concentration levels reach the 30% to 50% range [5]. The combination of the two results in a material
that is slightly less effective at absorbing infrared wavelengths but significant more effective at

32

absorbing visible wavelengths. From a light guide and device feature size perspective this is good since
the higher the wavelength the smaller the feature sizes can be.

Figure 21. ITO/NbOx (34% ITO nanoparticles) transmittance at different voltages [5].

Sub-wavelength Light Guides
A light guide is a structure that confines light along a path, for example a fiber optic cable. The
principle at play is something called total internal reflection. When light hits an interface between
two materials a portion of the light is reflected back and a portion of the light is refracted. The angle
of this refraction is given by the relationship between the two material’s index of refraction called
Snell’s law. When this angle becomes extreme enough the light that is bent is essentially bent back
into the original material, called total internal reflection. This view of a light guide is one that
considers light as a ray. When the angle of the light ray hitting the interface is such that total internal
reflection happens, it is said that all of the light stays in the original material. The inner material of a
light guide is called the core and the outer material is called the cladding. By choosing the materials
correctly, most of the light traveling down the light guide stays in the core. In reality, the light
doesn’t stop exactly at the interface but rather extends out into the adjacent material in a field that
exponentially decays called an evanescent field. Usually the dimensions of the light guide is chosen
such that this portion can be ignored. Another aspect of light guides that is ignored in ray theory is
diffraction. When a field passes through a smaller and smaller opening or when the dimensions of a
material transition become small compared to the wavelength, the field will disperse out more and

33

more. For example a fiber with a core diameter of 8Âľm and a cladding diameter of 125Âľm operating at
wavelength of 1250nm has a core that is a little over 6 times larger and cladding that is 100 times
larger than the wavelength of the light traveling through it. This mitigates the effects of diffraction
and ensures plenty of space between adjacent cores thus negating the effects of the evanescent field
and ensuring most of the power stays in the core. An additional mitigating factor for fiber optic cables
is the distance between the inputs and outputs. Having these relatively far from each other means
that any light that diffracts probably won’t make it to a sensor or another interface. Because of these
size constraints, fiber optic switching centers in communication networks can be quite large. To move
such things to more compact systems, sub-wavelength light guides need to be considered. In such
systems, the evanescent field must be considered since it doesn’t necessarily get smaller when feature
sizes are reduced so it becomes very large with respect to the diameter of the light guide. In the
example of a silica core fiber at subwavelength diameters, it can be shown the power is no longer
mostly confined to the core as it is in a normal scaled fiber. A significant amount of power exists as
the evanescent field outside the core [65]. This example also illustrates how the evanescent field’s
effects appear to scale up with respect to the core as the core scales down. For 633nm light in 400nm
and 200nm diameter cores the power of the evanescent field dissipates to negligible levels near a
radius of 500nm in both examples though it gets there much quicker comparatively [65]. In addition to
subwavelength fibers, waveguides can be created using periodic structures. The simplest physical
structure is a periodic waveguide consisting of rectangles spaced evenly apart [66].

Figure 22. SEM image of a sub-wavelength grating [66].

34

This type of structure is a diffraction grating. The theory behind it is quantum mechanics. When large
numbers of photons are sent through a structure, then classical physics generally describe what
happens (whether ray theory or wave theory). Quantum mechanics looks at it as a set of probabilities
and these probabilities have a spatial dependence. Macro structures exist effectively as the summation
of all the different spatially dependent probabilities where the probabilities of the odd outcomes
cancel each other out resulting in the classical result being the only one that is observable. Removing
the waveguide material at specific locations relative to the wavelength of the wave results in either
reinforcing the probabilities of classical results or removing some aspect of them [67]. In the case of
the rectangular subwavelength grating (SWG) the exact dimensions can allow the structure to operating
in several different regimes: diffraction, reflection, and subwavelength. The diffraction regime
scatters an incoming beam in different orders, the reflection regime is able to reflect an incoming
beam backwards, and the subwavelength regime allows for the suppression of diffraction effects [66].
For light guides like SWGs, metal is the preferred material since the surface plasmon resonance is well
studied and the properties of which can be controlled through particle/structure dimensions. As was
previously stated, metals are not the only materials to exhibit surface plasmon resonance. Technically
anything with free charge carriers can be used like semiconductors [50]. In addition to this, it was
been theoretically demonstrated that dielectric materials with sufficiently high permittivity can also be
used as subwavelength light guides. It should be noted that this permittivity is very high, needing to be
greater than 14. The study does note that Silicon-Germanium in the proper molar ratio has a
permittivity that is high enough in the real part while the imaginary part is low enough to significantly
reduce losses [68]. The main point of using a high permittivity material is to reduce the losses from
0.2dB per 100nm (0.6dB per 100nm actual) to a theoretical 0.01dB per 100nm. Nanoscale light guides
were simulated using spherical particles with radii of 50nm and spaced 150nm apart showing good
confinement of the light and even the ability to create Y-splitters and angled elbow paths. Where
these light guides changed direction or split, there were higher losses predicted in the simulations of 2.9dB for the Y-splitter and -5dB for the angled elbow [68].

35

Figure 23. Simulations of a Si-Ge subwavelength grating elbow and Y-splitter [68].

Similar in physical structure to the SWG waveguide is a metal-dielectric-metal (MDM) waveguide with
side coupled resonators [69].

Figure 24. Metal-dielectric-metal side coupled wave guide [69].

In this regime, there is a very narrow path made up of dielectric material with metal on either side.
Periodically, there are cavities on either side of the path of the same width as the main dielectric
path. The lengths of these cavities are not the same on both sides of the path, but they are located at
the same points along the path’s length. The asymmetry of the cavities leads to the light guide taking
on behavior similar to electromagnetically induced transparency. The difference between the two
side’s cavity lengths directly correlates to the bandwidth of the transparency window: as the
difference between the cavity lengths increases, the width of the transparency window increase.
Simulations then show an inverse relationship between the cavity length difference and the group
velocity of the light traveling through the structure which is expected behavior for electromagnetically

36

induces transparency [69]. The study also shows the correlation between the periodicity of the cavity
spacing to center frequency of the transparency window. As the distance between the cavities
decreases, the transparency window shifts higher in frequency but also shows a greater bandwidth.
This is attributed to coupling between adjacent cavities. The closer the cavities are to one another,
the greater the coupling between them is [69]. A different cavity based MDM structure is numerically
demonstrated by placing two Ag plates very close together (10nm to 40nm) and using low ridges to
form the light guide itself and a very small diameter hole acting as the input [70]. This structure relies
on surface plasmon polaritons for the actual transport of energy and was shown to be able to confine
light to a cross-sectional area of 10nm x 20nm over a distance of 10Âľm. This structure consisted of a
single hole in the bottom Ag plate with a diameter of about 20nm (that dimension isn’t specifically
noted in the study) with a 20nm tall, 10nm wide ridge protruding from either side out 5Âľm in either
direction. The distance between the top of the ridge and the upper Ag plate was fixed at 10nm.
Simulations where run on light with wavelength of 1300nm with the ridge height set to 20nm and 30nm
which corresponds to a distance between the Ag plates of 30nm and 40nm respectively. This showed
surface plasmon polariton wavelengths of 740nm and 769nm respectively but most importantly, showed
that 40nm Ag plate separation results in extremely good confinement of light. This demonstrated some
dependence on the distance between the Ag plates. The closer the plates were places to one another,
there is a shift to larger wave vectors for the same wavelength. This distance is independent of the
height of the ridges and can serve as an additional parameter that can be tweaked in design. By red
shifting the surface plasmon polariton 29nm they achieved a roughly 39% increase in signal isolation
between adjacent light guides (estimated from the graphical simulation results) [70].

37

Figure 25. Aluminum cavity wave guide showing the electric fields for two different spacings of L =
30nm (top) and L = 40nm (bottom) while keeping the ridge to upper plate gap constant at 10nm and a
wavelength of 1300nm [70].

Similar to a MDM waveguide is a multi-layered plasmon-dielectric hybrid system. This is a relatively
simple structure of multiple dielectric layers whose permittivities and thicknesses are chosen such that
the high order frequency based dispersion is effectively eliminated [71]. The concept is to start with a
simple metal-dielectric structure. Then, between the dielectric and the metal another layer of a
different dielectric is placed with a higher permittivity. Additional layers of dielectric can be placed at
the bottom of the stack but this time with permittivities incrementally smaller than the layer on top of
it (and decreasing thicknesses).

38

Figure 26. Wave guide with layered dielectrics [71].

On the side with metal, the surface plasmon polariton is confined to the metal-dielectric interface but
on the dielectric side, it is able to extend relatively far into the dielectric. For small wave vectors, the
surface plasmon polariton extends very far into the dielectric effectively seeing only the outer most
material binding the surface plasmon polariton to the interaction between the metal and that layer of
the dielectric. For very large wave vectors, the surface plasmon polariton to the next layer and so on
incrementally down to the physically closest dielectric layer. For small wave vectors, the behavior is
mostly unchanged, but for higher order wave vectors the inner layers begin to dominate the response.
If the permittivities and thicknesses are chosen correctly, a structure can be theoretically created that
allows for slow and subwavelength propagation of short pulses with no phase or amplitude distortion
[71]. A somewhat similar, though geometrically more complex and metamaterial dependent, approach
to managing light at the subwavelength scale is given in a set of simulations described as
subwavelength ray optics. By carefully managing dispersion throughout the path of the light, the study
claims to be able to guide light at diameters of Îť/10 with spacing of Îť/5 along curved paths using
metamaterials [72]. This is done by doing a conformal transform of three-dimensional space to get an
equivalent two-dimensional space. Metamaterials are then alternated with their physical design being
defined in the transformed two-dimensional space. A simulation was performed using alternating
curved layers of Ag and GaN with curves defined in their equivalent two-dimensional space though it
should be noted that no actual experiments were performed and fabrication methods would be
extraordinarily difficult. However difficult the fabrication might be, the paper does show through

39

simulation an interesting reduction of both diffraction and evanescent fields [72]. On the other hand,
a different study showed good experimental data demonstrating good light confinement of a subwavelength light guide. A set of light guides made from TiO2 on glass were imaged using photon
scanning tunneling microscopy (PSTM) to gather the spatial distribution of electric field intensity in
subwavelength light guides [73]. The glass used was BK7 (alkali-lime silicates with 10% potassium
oxide) with a refractive index of about 1.515. TiO2 was chosen because its dielectric function of 5.3 is
higher than the glass which helps offset problems with coupling the incident light as the light guide
cross-section gets smaller. Using standard electron beam lithography and reactive ion etching, 200nm
wide, 150nm high, 40Âľm long light guides made form TiO2 were formed atop the BK7 glass substrate.
Spacing was chosen to be 20Âľm between adjacent light guides to make sure there was no cross-talk
between them. The incident light was provided by a He-Ne laser with a wavelength of about 633nm.
PSTM measurements were obtained using a fiber tipped with 7nm of Cr to shield the tip from scattered
light. Electric field intensities were obtain with PSTM at a constant height yielding a spatial
distribution of the field intensities. Using this method allows for accurate mapping of the light
confinement in the guide since power measurements require both electric and magnetic field
measurements to be obtained simultaneously which the Heisenberg uncertainty principle forbids. The
probe was placed about 350nm above the top of the light guides (500nm above the substrate).

Figure 27. Observation method for measuring the light confinement of the TiO2 light guide [73].

40

Figure 28. Measurements of the TiO2 light guide with the incident beam entering the light guide (a) and
with it hitting other areas of the substrate (b) and (c) to confirm the observed light was in fact
traveling through the light guide and not some other effect [73].

Observing the entire guide including the incident light source resulted in an a scale for the amplitude
that made the guided light difficult to distinguish from the incident light; however even in this scaling
both the light at the exit of the guide and the guided light were still visible. The incident light was
then tested at to other locations relative to the entrance of the light guide showing a lack of coupling
which would be expected. The successful capturing and guiding of light with a properly oriented
incident light sources was reproduced on over 30 other light guides on the test substrate. To get
better resolution on the light confinement in the light guide, five PSTM images covering a 5Âľm by 5Âľm
area at a probe height of 10nm to 50nm were stitched together. This imaging showed good light
confinement to a width of roughly 300nm (given the TiO2 widths were 200nm).

41

Figure 29. High magnification stitched image showing the measured light confinement in the light guide
[73].
A comparison of the measured and numerically predicted electric field intensities was done showing
nearly a nearly identical normalized intensity profile with respect to the light guide’s width. The very
small differences noted are attributed to TiO2 roughness leading to scattering. The numerical result
was obtained using Green’s dyadic method to compute the electromagnetic near-field distribution in a
plane parallel to the sample surface, though the sample length in the simulation was reduced to 10Âľm
to keep calculation times reasonable [73].

Figure 30. SEM image of a TiO2 Y-splitter (top) and imaging of the measured light confinement (bottom)
[73].

42

Literature Review Summary
There is promising research in the area of plasmonics making subwavelength light guides appear like a
plausible way of creating integrated photonic circuits. History suggests the idea of a fully integrated
photonic circuit that can compete with Si-based integrated circuits is not a feasible one. It would be
advantageous to find an application that would make the two complimentary. While light-based
switching has obvious practical applications in the telecommunications industry, it is possible that such
similar switching code meet a need in the field of integrated circuits. We are exceedingly good at
making 2D structures in Si wafers, but moving effectively into the 3rd dimension has proved difficult.
Finding a method to add another type of switch on-top of existing Si-based wafer technologies would
provide a possible alternative avenue to achieving some 3D integrated circuit designs without the need
for SOI or bonding wafers together. This thesis examines a potential technology.

43

CHAPTER 2 PROPOSED RESEARCH
This idea is to create a switchable photonic trace that can be created using existing wafer processes
that can be built up on top of Si-based wafers. The switchable photonic trace uses a material that
whose transmittance can be changed by applying an electric field. Since generally such a material is
also conductive and an electric field is needed to produce the switching effect, a simple parallel plate
capacitor makes a convenient structure to base the switch off of. One plate is proposed to be the
switchable optical material with a dielectric surrounding it on all sides with a lower refractive index to
satisfy total internal reflection. Then a layer of metal is proposed to run parallel to the switchable
material. Charging the capacitor would apply an electric field causing a transmittance change and
discharging the capacitor would remove the electric field producing the opposite transmittance effect.

Figure 31. Comparison of proposed optical switch on Si wafer 3D IC versus a traditional wafer-on-wafer
structure.

44

Figure 32. Concept drawing of a capacitor based optical absorption switch.

To fully realize such a device, a number of problems will need to be solved. Since the proposed switch
is an optical effect that is controlled by an electrical effect and because the switch is proposed to
function in a complimentary fashion to Si-based wafers, optical/electrical conversion will be something
that needs to be addressed. This takes the form of two problems: Light generation/distribution and
light sensing. Additionally device properties will need to be addressed such as switch core/cladding
materials, switch length, and minimum distance between adjacent switches. Finally the proposed
structure and materials will need to be compatible with traditional Si-based wafer technology. The
scope of this thesis is to evaluate a specific material as the switchable material: ITO/NbOx [5]. It will
be evaluated in conjunction with three potential cladding materials (SiO2, Teflon AF 1601, and
nanoporous silica) at two wavelengths (475nm and 1.55Âľm). The other plate of the proposed
capacitive structure is suggested to be Aluminum because of it prevalent use as integrated circuit
metal lines. This will address two of the three main problems that need to be solved to make this
technology viable: device structure and compatibility with Si processes. The problem of light
generation, distribution, and sensing is not within the scope of this thesis; however it should be noted
that the wavelengths chosen for simulations were chosen such that LEDs can produce them. 1.55Âľm

45

was chosen because of its usage and existing technology related to fiber optic communications and
475nm was chosen because it is near the wavelength of blue LEDs. Additionally there is research
demonstrating diffraction gratings being used to get light into subwavelength light guides and this
effect works at right angles. Shining light on a diffraction grating from above makes the light move out
in-line with the grating path.

46

CHAPTER 3 TASKS
ITO/NbOx Material Evaluation for use as the Core
Significance: Determine the necessary parameters of a light guide consisting of ITO/NbOx as the core
material for 475nm and 1.55Âľm wavelengths to run simulations
Method: Literature Review and Simulation
Results:
ITO/NbOx is a material that is switchable in both infrared and visible wavelengths. This material was
reported in as a thin-film in a Nature Letter for selectively switching infrared and visible sunlight. It is
made up of ITO nanoparticles ranging between 5nm and 15nm that are processed with hexaniobates.
The ITO portion of the material is responsible for the switch-ability in the infrared range and the NbOx
is responsible for the switch-ability in the visible range. By bonding the ITO nanoparticles to the NbOx
the researchers describe the ITO and NbOx sort of merging together producing a material with different
properties than either of the original constituents. By trading off some of the ITO’s infrared light
switching performance, the visible light switching performance increased significantly [5]. Because
this is not a traditional material with well-defined material properties, it was necessary to extrapolate
the necessary properties from what has been reported thus far. The first task was determining the
percentage ITO and NbOx. The ITO gets bonded to hexaniobates and then to alter the exact properties
the researchers added this bonded ITO/NbOx to a solution of decaniobates. By altering this percentage
they were able to get differing transmittance switching profiles. Examining these profiles at the two
wavelengths of 475nm and 1.55Âľm it was determined that for 475nm an ITO concentration of 34% and
for 1.55Âľm an ITO concentration of 69% provide the best switching performance. The basis of which
was the greatest contrast between the minimum and maximum transmittance combined with a
reasonable maximum transmittance as to avoid unnecessary losses. These transmittance profiles were
measured through a thin-film 150nm thick. For the sake of simplicity, a switch that is absorbing light
will be said to be in an ON state and a switch that is allowing light to pass will be said to be an OFF
state. From the transmittance profiles the following was determined:

47

Wavelength
475nm
1.55Âľm

ITO Concentration
ON (150nm)
OFF (150nm)
34%
43%
90%
69%
35%
85%
Table 1. Transmittance [5].

ON (1nm)
99.44%
99.30%

OFF(1nm)
99.93%
99.89%

These numbers pulled from graphs are the transmittance through 150nm of material. To determine the
transmittance per nanometer the transmittance through 300nm of material was calculated
representing two units of 150nm in series.
(0.43)(0.43) = (0.43)2 = 0.185
This answer was then used to solve for the transmittance per nanometer where each unit is 1nm, then
0.185 represents the transmittance of 300 units in series.
𝑋 300 = 0.185
𝑋 = (0.185)

1⁄
300

𝑋 = 0.9944
For the sake of clarity, from here out the ITO/NbOx mix consisting of 34% ITO nanoparticles will be
referred to as ITO34 and an ITO/NbOx mix consisting of 69% ITO will be referred to as ITO69.

Next it was necessary to determine the refractive index of each material at the desired wavelengths.
Unfortunately there isn’t any information available for this material regarding refractive index since
this isn’t a particularly important parameter for switching in a thin-film form, but is critical to a light
guide. To determine this, the refractive indices for ITO nanoparticles and niobate glass were looked up
in literature then were combined to estimate the refractive index of ITO/NbOx. Refractive index
describes how light travels through a medium referenced to how it would travel through a vacuum.
Effectively this is the speed of light through a medium. Since ITO/NbOx consists of part ITO
nanoparticles and part niobate glass, it could be reasoned that the light will travel some distance
through ITO nanoparticles and some distance through niobate glass for a given length through the
material. The amount of each would be proportional to the percentage of the material made of up of
each material. It can be reasonably estimated that light traveling through ITO34 would travel through
ITO nanoparticles for 34% of the material and niobate glass through the rest. While the refractive

48

index of bulk ITO is a well-studied parameter it is significantly different than ITO nanoparticles. For
example, at 475nm bulk ITO has a refractive index of about 2 while simulations shows that ITO
nanoparticles having something closer to 1.5 [74, 75]. For the NbOx, the bulk properties were used
since ITO/NbOx effectively uses NbOx as the bulk material with ITO nanoparticles suspended in it which
has a refractive index around the 2.2 to 2.4 range [76]. The differences between bulk ITO and ITO
nanoparticles show that smaller structures of a given material may have lower refractive indices than
their bulk counterparts. This effect is not considered for the NbOx but it is likely there is some
reduction of the refractive index for the hexaniobates that are bonded to the ITO nanoparticles. This
would mean some portion of the NbOx will be both a structure on the nano scale and be somewhat
altered due to the presence of the ITO it is bonded to. To estimate the overall refractive index of a
given concentration of ITO/NbOx the following formula was used where ITO% is the fraction ITO
nanoparticles in the material. From this equation refractive indices for ITO34 at 475nm and ITO69 at
1.55Âľm where estimated where ITO34 was the optimal ITO concentration chosen for switching at
475nm and ITO69 was the optimal ITO concentration chosen for switching at 1.55Âľm.
(𝐼𝑇𝑂%)𝑛𝐼𝑇𝑂𝑛𝑎𝑛𝑜 + (1 − 𝐼𝑇𝑂%)𝑛𝑁𝑏𝑂𝑥 = 𝑛𝐼𝑇𝑂%

Wavelength
475nm
1.55Âľm

ITO Concentration
34%
69%
Table 2. ITO/NbOx Refractive Index [75, 76].

Refractive Index
2.1181
1.4521

The proposed dimensions of each switch depends on the transmittance of a given ITO concentration at
a given wavelength. Since the transmittance profile actually depends on the concentration of ITO
nanoparticles, it was decided to examine the effects of non-uniformity on the ITO concentration. At
very small dimensions having one too many or one too few nanoparticles would have a significant
impact the transmittance properties of that particular switch. The following analysis looks at the
cross-sectional areas showing the error in ITO concentration as a function of area to get an idea of the
relationship.

49

Figure 33. Cross-section and Nanoparticles.
The number of nanoparticles needed to create the same ratio of materials as ITO% given a nanoparticle
radius r and wave guide height H and width W is given by the following equation. Assuming a maximum
diameter of 15nm the following table shows the expected window for concentration error assuming
whole numbers of nanoparticles. For example in a 100nm by 100nm light guide 4.81 nanoparticles are
needed to create the proper 34% ITO concentration. Having only 4 results in a nearly 17% error and
having 5 results in a 4% error. It should be noted this effect might be avoidable due to etching, but the
effects of partial nano-particles in a very small structure is not known at this time.
𝐼𝑇𝑂% (

𝐻𝐿
) = 𝑁𝑛𝑝
𝜋𝑟 2

50

ITO%

H (nm)

W (nm)

r (nm)

n

Error (+%)

Error (-%)

Error Range

34

100

100

15

4.810016

16.84%

-3.95%

20.79%

34

100

200

15

9.620032

6.45%

-3.95%

10.39%

34

100

300

15

14.43005

2.98%

-3.95%

6.93%

34

100

400

15

19.24006

1.25%

-3.95%

5.20%

34

100

500

15

24.05008

0.21%

-3.95%

4.16%

34

200

200

15

19.24006

1.25%

-3.95%

5.20%

34

200

300

15

28.8601

2.98%

-0.48%

3.46%

34

200

400

15

38.48013

1.25%

-1.35%

2.60%

34

200

500

15

48.10016

0.21%

-1.87%

2.08%

34

200

600

15

57.72019

1.25%

-0.48%

1.73%

69

100

100

15

9.761503

7.80%

-2.44%

10.24%

69

100

200

15

19.52301

2.68%

-2.44%

5.12%

69

100

300

15

29.28451

0.97%

-2.44%

3.41%

69

100

400

15

39.04601

0.12%

-2.44%

2.56%

69

100

500

15

48.80752

1.65%

-0.39%

2.05%

69

200

200

15

39.04601

0.12%

-2.44%

2.56%

69

200

300

15

58.56902

0.97%

-0.74%

1.71%

69

200

400

15

78.09203

0.12%

-1.16%

1.28%

69
69

200
200

500
15
97.61503
0.63%
600
15
117.138
0.12%
Table 3. Cross-sectional area comparison

-0.39%
-0.74%

1.02%
0.85%

The actual length of each switch was determined using the per-nanometer ON/OFF transmittances. A
balancing act is required here to provide the most difference between ON/OFF states while still
maintaining reasonable transmittance in the OFF state.

ITO%

Length (nm)

ON (T)

OFF (T)

ON/OFF Diff

34

100

0.570311

0.932371

0.362060209

34

200

0.325254

0.869316

0.544061258

34

300

0.185496

0.810525

0.625028582

34

400

0.10579

0.75571

0.649919245

69

100

0.495364

0.89578

0.400415434

69

200

0.245386

0.802422

0.557035674

69

300

0.121555

0.718793

0.597237683

69

400
0.060214 0.64388 0.583666203
Table 4. Transmittance versus length.

51

A trade-off in the dimensions was accepted for this thesis. A wider light guide provides for less
susceptibility to ITO concentration non-uniformities as demonstrated above. Additionally, the length
of each switch is likely best at shorter distances because of better transmittance in the OFF state and
less overall absorption. The height and width of 200nm and the length of 400nm was chosen as a
compromise between performance and shape. Since the method of simulation run below assumes a
light guide which is very long, actually simulating a switch that was longer than it was wide made for a
more meaningful result. If changing the shape of the guide that drastically results in effects not
captured in the simulation, then hopefully the chosen shape mitigates those effects.

Baseline Simulation of a Square ITO/NbOx Light Guide
Significance: Get a baseline for how much light needs to be dealt with between adjacent light guides.
Will be used to help evaluate how densely populated the device can be while avoiding cross talk.
Method: Simulation
Results:
These are the baseline simulations of ITO34 and ITO69 fully surrounded by air. The purpose of these
simulations was to gain insight into the scales that will need to be considered at each ITO
concentration level and wavelength and any potential issues that would prevent the material from
functioning as a light guide. To run simulations for this thesis, the Matlab implementation of the
extended Marcatili approach for rectangular light guides was used [77]. This method was chosen in
part because of its ability to more accurately model rectangular light guides with high refractive indexcontrasts. This allows the same tool to be used to model very small scale structures which generally
only function with high refractive index-contrasts. For reference, fiber optic cables using silica cores
have refractive indices around 1.5 and are considered low contrast compared to silicon having an index
around 3 which is considered high contrast. ITO34 falls somewhere between with an index of around 2
and ITO 69 is about 1.46 which is worse than a normal fiber optic cable (at 1.55Âľm). By using the
extended Marcatili method it can be reasonably assumed the plots of ITO34 should be fairly accurate
since it is lower than silicon. This method is also used to model small rectangular silicon light guides
meaning again modeling the ITO34 and ITO69 light guides a small dimensions should provide reasonably

52

accurate results. Both light guides were simulated at the same dimensions of 200nm by 200nm. This
simulation method uses five regions: the core defined as region 1 and four surrounding regions 2-5
defining the surrounding materials. The regions are only defined for each of the directions primary
axes x and y (i.e. no results are calculated for the diagonals since enough information is gathered from
the primary axes making the diagonals either redundant or unnecessary). The z-axis is defined to be
the direction of propagation of the light. To simplify the code this simulation just changes the
orientation of the light guide to solve for the y-axis information. This is done so that only code solving
for the x-axis has to be written. Figure (a) shows the orientation of the light guide, (b) be shows the
profile, (c) and (d) show how the light guide is rotated during simulation to cut done on the code they
needed to write, and finally (b) illustrates the lack of solving for the diagonals.

Figure 34. Axis definitions for Matlab implementation of the extended Marcatili approach for
rectangular light guides [77].

Making the length and width the same further simplifies the final product. The baseline simulations
were run as follows:

53

Wavelength
475nm
1.55Âľm

Width
200nm
200nm

Height
200nm
200nm

Core
n1(core) Substrate n2(sub)
ITO34
2.1181
Air
1
ITO69
1.4521
Air
1
Table 5. Baseline simulation set-up.

Figure 35. ITO34 at 475nm (x and y axes are in Âľm)

Figure 36. ITO34 at 475nm Ex Zoomed

54

Dielectric
Air
Air

n3-5(delectric)
1
1

ITO34 at 475nm shows very good initial results. This shows very good light confinement. The last data
point that shows up in the plot occurs at 117.5nm form the edge of the guide for the y-axis (width) and
167.5nm in the x-axis (height). The Ex component of the field is the strongest (which is in part to how
the simulation runs). The Ey and Ez components actually extend out farther than the Ex component in
the x-axis; however they are 14% and 56% of the intensity of the Ex component making their actual
impact significantly less the Ex component. This means the Ex component dominates the distance out
from the edge of the light guide the light reaches. This is supported by comparing the absolute limit of
each field component to the point at which it is reduced to 10% of the absolute maximum field
intensity.

Field (y-axis)
Limit
10%
Ex
117.5nm
67.5nm
Ey
57.5nm
n/a
Ez
127.5nm
52.5nm
Table 6. Field reach from the edge of the light guide in y-axis for ITO34

Field (x-axis)
Limit
10%
Ex
167.5nm
102.5nm
Ey
172.5nm
27.5nm
Ez
177.5nm
87.5nm
Table 7. Field reach from the edge of the light guide in x-axis for ITO34

The x-axis, which represents height, is of the least concern since the proposed structure will have
regions of interconnects between each active region, although it is worth noting. The y-axis, which
represents the distance between two adjacent switches, is the dimension of primary concern. Placing
switches at least 255nm apart will effectively mean that not even the edges of the fields will overlap.
Adding to this dimension a guard band would be the safest way to do an initial build of the proposed
technology ensuring a minimum amount of cross talk which would be advantageous for any initial
testing. The actual switch density would not only be a function of the fields outside of the switches
but also the length of the switch (affects the ON/OFF transmittance difference) and optical/electrical
converter sensitivity that would need to be at the end of the switch. For the purposes of this thesis,
the 10% point was chosen because 10% cross-talk likely would be tolerable. It should be noted that it

55

may be possible to tolerate more cross talk meaning switch density could be further compressed.
Based on this 10% point, the minimum distance between adjacent switches would be 67.5nm for ITO34
surrounded by air. This of course isn’t a structure that can be built as a 3D integrated circuit. A
dielectric that will not perform as good as air will be needed.

Figure 37. ITO69 at 1.55Âľm (x and y axes are in Âľm)

Figure 38. ITO69 at 1.55Âľm Ex Zoomed

Figure 39. ITO69 at 1.55Âľm Ez Zoomed

56

The simulation of ITO69 at the nanoscale shows a failure of the switch to support much of anything
inside the switch core. The Ex component which is supposed to be the primary component effectively
doesn’t travel at all inside the switch but rather in the cladding. Changing the dimensions of the light
guide from 200nm by 200nm to 1Âľm by 1Âľm results in a normal looking light guide response; however a
1Âľm thickness is not practical. Experimenting shows that in order to support the Ex component inside
the switch the thickness needs to be about 500nm. Even at this point the performance is not very good
as the intensity inside the switch is 24% less than it is outside. Additionally the intensity of the Ez
component of the field is actually comparable to the Ex component. While this is possible, it is not
really practical for 3D integrated circuit applications.

Figure 40. ITO69 at 1.55Âľm (x and y axes are in Âľm). Width is 1Âľm and height is 500nm

57

Figure 41. ITO69 at 1.55Âľm Ex Zoomed (1Âľm x 500nm) Figure 42. ITO69 at 1.55Âľm Ez Zoomed (1Âľm x 500nm)

Field (y-axis)
Limit
10%
Ex
643nm
343nm
Ey
703nm
n/a
Ez
843nm
303nm
Table 8. Field reach from the edge of the light guide in y-axis for ITO69
Field (x-axis)
Limit
10%
Ex
1567nm
1048nm
Ey
1467nm
223nm
Ez
1528nm
698nm
Table 9. Field reach from the edge of the light guide in x-axis for ITO69

Despite its practical short comings, there can be reasonably good switch density compared to the
switch dimensions at 343nm between each switch compared to a 1Âľm switch width. Looking at the yaxis however shows another problem. There would need to be a minimum of 1Âľm between layers of
ITO/NbOx which again while not impossible, puts a significant limitation on design of a 3D integrated
circuit. None of this takes into consideration that much of this light would be absorbed adding
additional heat to the device which is known to cause issues in 3D integrated circuits.

58

The baseline simulation of ITO34 at 475nm looks very promising showing good light confinement, the
ability to quite densely pack switches, and likely little layer to layer concerns. The baseline simulation
of ITO69 at 1.55Âľm shows at the 200nm switch scale it fails to confine light. The dimensions of the
switch can be scaled up to produce a potentially functional switch, but its performance is not very
good. There are two things working against ITO69. The wavelength this material works well at
switching, is relatively long which requires a very high refractive index to compensate. Unfortunately
the refractive index of ITO69 at 1.55Âľm is low further exasperating the confinement problem. This
isn’t an insurmountable problem if a material can be found with the ability to switch in the infrared
range can be found with high refractive index. Experimenting with simulations show that at a
refractive index of 5, the confinement is good but there still isn’t a good ratio of light in and out of the
switch. At a refractive index of 6 this ratio starts looking good. ITO/NbOx doesn’t look like a good fit
for building an infrared switch at the 200nm scale, but it does look like it will work for a 200nm scale
blue light switch.

Cladding Material Evaluation
Significance: Compare various materials for use in the cladding focusing on refractive index and
availability in semiconducting manufacturing.
Method: Literature Review and Simulation
Results:
To create a switch that uses a capacitor as its fundamental structure, dielectric materials must be
chosen. These materials were chosen based on their existing association with Si-based fabrication then
their refractive indices were compared to the indices of ITO34 and ITO69 to see which would be
compatible for both materials. One obvious choice is SiO2 which is the most prevalent dielectric
material in the semiconductor industry. In addition to this nanoporous silica, hydrogensilsesquioxanes
(HSQ, Teflon-AF (PTFE), silicon oxyflouride (FSG), hafnium silicate (HfSiO4), zirconium silicate (ZrSiO4),
and barium titanate (BaTiO3) where looked at [78, 79]. This list was narrowed down based on
searching for the approximate refractive index. Materials with refractive indices lower than SiO2 were
looked for. It should be noted that most references to refractive index for these materials are

59

reported as a constant and no wavelength is given for the stated index resulting in some sources
reporting seemingly conflicting numbers. Looking at various materials in reference [74] shows that
most materials have a significant dependence on wavelength with the general shape of the index
showing a spike at short wavelengths and then a somewhat exponentially decreasing curve towards
longer wavelengths. Knowing that total internal reflection must be satisfied helped reduce the list of
materials to those with refractive indices lower than ITO69 at 1.55Âľm. This narrowed things down to
three: SiO2, Teflon-AF 1601, and nanoporous silica.
Dielectric Material
n(475nm)
n(1.55Âľm)
SiO2
1.4638
1.444
Teflon-AF 1601
1.31
1.295
Nanoporous Silica
1.22
1.20
Table 10. Cladding material refractive indices [80, 81, 82].

The simulations in this section use a 200nm by 200nm structure for ITO34 and a 0.5Âľm by 1Âľm structure
for ITO69. While the reason for this is apparent from the baseline simulations run on ITO69, it should
also be noted that the simulation will not actually run for ITO69 at the 200nm scale because the
refractive indices are too close together. The simulation just errors out. At the large scale, the
simulations will run. Also the bottom substrate material is always chosen to be SiO2 for ITO34 to
evaluate its compatibility to be just put on top of a wafer using SiO2 for all of its electrical layers. This
was not done on ITO69 because that material even at 0.5Âľm has issues with how far down the electric
field penetrates. With the extremely week index contrast between ITO69 and SiO2 (1.4521 and 1.444
respectively), this will only get worse. This material would absolutely need a barrier layer between it
and the rest of the wafer.

60

Simulations of ITO34 on a SiO2 substrate and a dielectric of SiO2, Teflon-AF 1601, and finally
nanoporous silica:

Figure 43. ITO34/SiO2 at 475nm (x and y axes are in Âľm)

Field (y-axis)
Limit
10%
Ex
163nm
93nm
Ey
138nm
n/a
Ez
168nm
58nm
Table 11. Field reach from the edge of the light guide in y-axis for ITO34/SiO2
Field (x-axis)
Limit
10%
Ex
193nm
123nm
Ey
93nm
n/a
Ez
218nm
98nm
Table 12. Field reach from the edge of the light guide in x-axis for ITO34/SiO2

61

Figure 44. ITO34/Teflon-AF 1601 at 475nm (x and y axes are in Âľm)

Field (y-axis)
Limit
10%
Ex
163nm
93nm
Ey
128nm
n/a
Ez
168nm
63nm
Table 13. Field reach from the edge of the light guide in y-axis for ITO34/Teflon-AF 1601
Field (x-axis)
Limit
10%
Ex
178nm
113nm
Ey
198nm
8nm
Ez
203nm
93nm
Table 14. Field reach from the edge of the light guide in x-axis for ITO34/Teflon-AF 1601

62

Figure 45. ITO34/nanoporous silica at 475nm (x and y axes are in Âľm)

Field (y-axis)
Limit
10%
Ex
127nm
93nm
Ey
127nm
n/a
Ez
182nm
68nm
Table 15. Field reach from the edge of the light guide in y-axis for ITO34/nanoporous silica
Field (x-axis)
Limit
10%
Ex
163nm
107nm
Ey
183nm
13nm
Ez
192nm
93nm
Table 16. Field reach from the edge of the light guide in x-axis for ITO34/nanoporous silica

Simulations of ITO69 totally surrounded by dielectric of SiO2, Teflon-AF 1601, and nanoporous silica.
Note SiO2 is not used as a substrate for ITO69 because of its poor performance.

63

Figure 46. ITO69/SiO2 at 1.55Âľm (x and y axes are in Âľm)

Field (y-axis)
Limit
10%
Ex
> 8000nm
> 8000nm
Ey
> 8000nm
n/a
Ez
> 8000nm
n/a
Table 17. Field reach from the edge of the light guide in y-axis for ITO69/SiO2
Field (x-axis)
Limit
10%
Ex
> 8000nm
> 8000nm
Ey
> 8000nm
n/a
Ez
> 8000nm
n/a
Table 18. Field reach from the edge of the light guide in x-axis for ITO69/SiO2

64

Figure 47. ITO69/Teflon-AF 1601 at 1.55Âľm (x and y axes are in Âľm)

Field (y-axis)
Limit
10%
Ex
1500nm
900nm
Ey
1500nm
n/a
Ez
1500nm
300nm
Table 19. Field reach from the edge of the light guide in y-axis for ITO69/Teflon-AF 1601
Field (x-axis)
Limit
10%
Ex
2800nm
1900nm
Ey
2750nm
n/a
Ez
2750nm
700nm
Table 20. Field reach from the edge of the light guide in x-axis for ITO69/Teflon-AF 1601

65

Figure 48. ITO69/nanoporous silica at 1.55Âľm (x and y axes are in Âľm)

Field (y-axis)
Limit
10%
Ex
1000nm
650nm
Ey
1150nm
n/a
Ez
1100nm
350nm
Table 21. Field reach from the edge of the light guide in y-axis for ITO69/nanoporous silica
Field (x-axis)
Limit
10%
Ex
2100nm
1200nm
Ey
1750nm
n/a
Ez
1750nm
750nm
Table 22. Field reach from the edge of the light guide in x-axis for ITO69/nanoporous silica

ITO34 shows very good results with good light confinement for all three materials. Since the
performance of this material with SiO2 is good it allows the material to be placed directly on wafers
using SiO2 as their dielectric even if a different dielectric is desired for the switches. Adding the
dielectric does result in some loss of switch density moving from a minimum of 67.5nm between
switches to a minimum of 93nm between switches for all three dielectric materials. ITO69 performed
extremely poorly with SiO2. This means the most common dielectric in semiconductor manufacturing
can’t be used with this material. The other two dielectric materials simulated didn’t perform too

66

poorly. The minimum distance between switches was 900nm and 650nm for Teflon-AF 1601 and
nanoporous silica respectively. Considering the wavelength of 1.55µm, that wasn’t too bad. The issue
with ITO69 again was the performance in the x-axis (height of the switch). The resulting minimum
distance between switches layered on top of one another was 1.9Âľm and 1.2Âľm. While increasing the
height of the switch allows for better light confinement allowing for some amount of optimization,
micrometer scales for thicknesses on wafers is not very practical. Simulations show that ITO34 with all
three dielectric materials tested should function well, but ITO69 will only potentially function with
Teflon-AF 1601 and nanoporous, but there are questions about how viable even those materials could
be.

Manufacturability
Significance: Compare the materials from the core and cladding material evaluation for
manufacturability.
Method: Literature review
Results:
The manufacturability of the materials chosen is critical to this thesis. The whole point was to find a
process for optical switching that functioned on small enough scales to be added as a 3D integrated
circuit on top of an existing Si-based wafer process. Four materials were examined here, ITO/NbOx,
SiO2, Teflon-AF1601, and nanoporous silica.

ITO/NbOx is a material that is synthesized first in solution where ITO nanoparticles between 5nm and
15nm in diameter are linked with hexaniobates forming the material in ITO/NbOx that is primarily
responsible for its effects. This solution is then mixed with decaniobates to the desired ITO
concentration where ITO34 is 34% ITO/hexaniobate and 66% decaniobate. The final product of this
preparation is an aqueous colloidal solution that can then be deposited with spin coating. The
substrate needs to be treated to make it hydrophilic, which in the case of the original study was
treating the surface with ultraviolet-ozone. After spin coating, the resulting film needs to be annealed
at 400°C for 20 minutes using a rapid thermal annealing furnace at 10°Cs-1 (annealing in air is fine) [5].

67

The simple spin coating deposition is one of the main reasons this material was chosen for this
application making it something that can be applied to a wafer easily. SiO2 is a staple of the
semiconductor industry. This can be deposited by chemical vapor deposition using SiH4 and O2 in
temperatures ranging from 300°C to 500°C at atmospheric pressure [83]. There is a very significant
amount of literature written and industry experience regarding this material. This material effective
creates the definition of what other materials must do at a minimum. Teflon-AF 1601 is the DuPontbranded name for polytetraflouethylene or PTFE. This material claims to be unique in that it has the
lowest refractive index of any polymer [81]. There are three versions of this material where the
amount of dioxole monomer is varied (AF 1300, AF1601, AF2400). Deposition was performed using spin
coating where the Teflon-AF was diluted in FC-40 surfactant. The refractive indices for all three
versions are relatively close to one another differing by only about 0.015. Since the data showed that
thinner films were possible at slower spinning speeds for Telfon-AF 1601 given the same spinning time,
it was chosen for simulations. A speed of 2000rpm for 60s resulted in about a 55nm film thickness
whereas Teflon-AF 2400 required 5000rpm for 60s to achieve a 60nm film thickness. The bake
temperature for the film was 140°C for 60 seconds [81]. It should be noted that the glass transition
temperature for Teflon-AF 2400 is given as 240°C which is well below the 400°C required for annealing
ITO/NbOx. This calls into question how well this material could handle the high temperatures required
for this annealing process. Nanoporous silica is a process quite similar to ITO/NbOx. The chemistry
here seeks to create silica nanoparticles suspended in solution using a base process (using TEOS,
ethanol, and water). The research showed that although it is possible to produce the material using an
acid and two step acid/base process, it was only the sample prepared using the base process that
showed the extraordinarily low refractive index [82]. SEM imaging of films made with acid-catalyzed
and base-catalyzed nanoporous silica show the base-catalyzed version to have smaller structures
showing a dependence of optical performance on the size of the pores. The deposition method for
nanoporous silica was dip-coating. The film thickness was controlled by controlling the draw speed of
the wafer out of the solution. The silica solution was treated with an ultrasonator for 30 minutes prior
to dip deposition to separate unstable aggregate. After coating, the film was placed in an ammonia

68

atmosphere for 24 hours and then baked at 400°C for 2 hours [82]. This process also places it within
the same range of temperatures as the ITO/NbOx and SiO2.

Figure 49. SEM images comparing acid-catalyzed (a) to base-catalyzed (b) nanoporous silica [82]

ITO/NbOx, SiO2, and nanoporous silica all show good compatibility with standard Si-based wafer
fabrication. Teflon-AF 1601 raises questions about whether it could actually survive the temperatures
required for the ITO/NbOx and SiO2 (if used again) and whether it would retain all of its properties
afterward. Routinely taking the material above the glass transition temperature would also mean the
structural integrity of other things like the metal lines might be compromised and potentially put
additional stress on the ITO/NbOx during processing. Given there are alternatives with fewer
questions, it is likely SiO2 or nanoporous silica should be considered for the dielectric material.

The following is a process recommendation for the structure of the switch. Note this process doesn’t
take into consideration light generation or optical/electrical conversion. The dielectric material
chosen is SiO2, the metal lines and vias chosen are the Aluminum and Tungsten respectively, and the
active layer is ITO/NbOx (ITO34). For the sake of this example, vias are ignored to focus on building up
the actual proposed switch. Additionally, the bottom electrode here is the metal layer and the top
electrode is the ITO/NbOx but it should be noted that there is no reason this must be true. There is
nothing preventing the bottom electrode from being the ITO/NbOx and the top electrode being the
metal layer. The intention is for this process to be built up on top of a Si wafer. First a layer of SiO2 is
necessary to provide isolation from the lower metal interconnect layers for the Si wafer (note that vias

69

would obviously be needed in this layer to ensure electrical connections to the Al bottom electrodes.
Next the Al metal layer needs to be deposited over the whole wafer (there are established processes
for chemical vapor and sputtering deposition for Si wafer processes). After the Al another layer of SiO2
must be deposited after which the ITO/NbOx layer must be put down. Then a photo resist needs to be
put down to create the pattering for the switches. The next step is to etch all three layers of
ITO/NbOx, SiO2, and Al to create small capacitors that will function as switches due to the ITO/NbOx
electrode (through which the light will travel). Another layer of SiO2 needs to be added to fully
surround the switches and provide electrical isolation for the conductive layers. The final step needs
to be a planarization since the switches are relatively tall and the final SiO2 layer will likely be quite
uneven at the top. At this point another etch process would need to occur to add the necessary vias to
create the electrical connections to the ITO/NbOx top electrodes.

Figure 50. Full stack of ITO/NbOx, SiO2, and Al creating the basis for the capacitors/switches

Figure 51. Photo resist to create the actual individual switches

70

Figure 52. After etching to produce the individual switches

Figure 53. After final SiO2 deposition and any necessary planarization

The final product will be small capacitors that, in the case of ITO/NbOx, will allow light to pass through
the top electrode when charged and will block light when discharged.

71

CHAPTER 4 FUTURE WORK
This thesis only addresses the idea of an optical switch controlled by an electric field for a specific
material: ITO/NbOx. It seeks to evaluate its possible use as an active device that could be fabricated
on top of a Si wafer. This is not the only material that could potentially have applications as a switch.
There also are a number of problems that fall outside the scope of this initial work. Most notably is the
generation and distribution of light and converting that light back to electrical charge that can be
deposited onto the switches to turn then on and off. Nor does this thesis address how charge could be
bleed off the electrodes in order to switch the other way. ITO/NbOx was chosen for this work because
of its simple deposition method and annealing temperature making it compatible with existing wafer
processes. This is a somewhat unique material that has very interesting properties but hasn’t been
exhaustively researched. The refractive index needs to be experimentally determined not just
estimated based on the individual ITO and NbOx properties. A very important piece of information that
needs to be experimentally determined is the isotropicity of the material. Research into thin films
examine how light interacts with them in a direction that is actually perpendicular to how it is
intended to travel through this switch. The concern here is related to how the light is propagating with
respect to the applied electric field that is intended to produce the transmittance change. In the case
of this switch the electric field will also be perpendicular to the direction the light is propagating. This
is something that needs to be investigated since this is not really a scenario for which there has needed
to be much research. While it is reasonable to assume that if the material is isotropic this will not be a
concern, it would be wise to consider the light propagation direction and applied electric field
direction scenario simply because of a lack of research.

72

CHAPTER 5 SUMMARY
The goal of this research was to investigate the feasibility of creating an optical switch that would be
compatible with Si wafers. The literature review provided a promising material in ITO/NbOx that
showed itself to have potential for use in the short wavelength visible range, specifically 475nm.
Unfortunately this material did not show good results for the infrared range due to the poor refractive
index at longer wavelengths. A material with a significantly higher refractive index would be necessary
to provide good light confinement. The 475nm simulations showed good results with SiO2 showing that
switches with sizes similar to late 1990s Si transistor node sizes. A 100nm process would be needed to
produce switches placed 100nm apart and 200nm wide and could use industry established processes for
metallization, SiO2 deposition and ITO/NbOx deposition (spin coating/annealing). While there is
nothing stopping the simulations from going to smaller sizes, it is likely that much below 200nm widths
would result in problems getting consistent performance from switch to switch due to the use of ITO
nanoparticles in ITO/NbOx. This research provides the ground work developing a voltage controlled
optical switch that could be placed onto Si wafers to build a 3D integrated circuit without the need of
additional wafers or through-Si vias.

73

BIBLIOGRAPHY

74

BIBLIOGRAPHY

[1]

Roland Lenz. “Introduction to all optical switching technologies.” Available:
http://imedea.uibcsic.es/~salvador/coms_optiques/addicional/tutorials/intro_allopticalswitching.pdf, Jan. 30,
2003 [Accessed Aug. 12, 2015].

[2]

I. Glaser. “Lenslet array processors.” Applied Optics, Vol 21, Issue 7, pp. 1271-1280, 1982
doi: 10.1364/AO.21.001271

[3]

Usman Pirzada. "Intel ISSCC: 14nm all figured out, 10nm is on track, Moores Law still alive and
kicking.” Internet: http://wccftech.com/intel-isscc-14nm/. WCCF Tech. July 2, 2015.

[4]

Parthiban Santhanam, Dodd Joseph Gray, Jr., and Rajeev J. Ram. “Thermoelectrically Pumped
Light-Emitting Diodes Operating above Unity Efficiency.” Physical Review Letters, Vol 108,
Issue 6, 097403. Feb 27, 2012.
http://journals.aps.org/prl/abstract/10.1103/PhysRevLett.108.097403

[5]

Anna Llordés, Guillermo Garcia, Jaume Gazquez, and Delia J. Milliron. “Tunable near-infrared
and visible-light transmittance in nanocrystal-in-glass composites.” Nature, Vol 500, pp 323326, Aug 15, 2013.
doi: 10.1038/nature12398

[6]

Pierre Ambs. “Optical Computing: A 60-Year Adventure.” Advances in Optical Technologies, Vol
2010, Article ID 372652, Feb 19, 2010
doi: 10.1155/372652

[7]

Optalysys company website [Online]. Available: http://optalysys.com/technology/ [Accessed
April 11, 2017].
M. Fernando Gonzalez-Zalba, Sergey N. Shevchenko, Sylvain Barraud, J. Robert Johansson,
Andrew J. Ferguson, Franco Nori, and Andreas C. Betz. “Gate-Sensing Coherent Charge
Oscillations in a Silicon Field-Effect Transistor.” Nano Letters, Vol 2016, pp 1614-1619, Feb 11,
2016.
doi: 10.1021/acs.nanolett.5b04356

[8]

[9]

M. Veldhorst, C. H. Yang, J. C. C. Hwang, W. Huang, J. P. Dehollain, J. T. Muhonen, S.
Simmons, A. Laucht, F. E. Hudson, K. M. Itoh, A. Morello, A. S. Dzurak. “A two-qubit logic gate
in silicon.” Nature, Vol 526, pp 410-414, October 15, 2015.
dio: 10.1038/nature15263

[10]

S. Debnath, N. M. Linke, C. Figgatt, K. A. Landsman, K. Wright, and C. Monroe. “Demonstration
of a small programmable quantum computer with atomic qubits.” Nature, Vol 536, pp 63-66,
Aug 4, 2016.
dio: 10.1038/nature18648

[11]

D. Hillerkuss, R. Schmogrow, T. Schellinger, M. Jordan, M. Winter, G. Huber, T. Vallaitis, R.
Bonk, P. Kleinow, F. Frey, M. Roeger, S. Koenig, A. Ludwig, A. Marculescu, J. Li, M. Hoh, M.
Dreschmann, J. Meyer, S. Ben Ezra, N. Narkiss, B. Nebendahl, F. Parmigiani, P. Petropoulos, B.
Resan, A. Oehler, K. Weingarten, T. Ellermeyer, J. Lutz, M. Moeller, M. Huebner, J. Becker, C.
Koos, W. Freude and J. Leuthold. “26 Tbit s-1 line-rate super-channel transmission
utilizing all-optical fast Fourier transform processing.” Nature Photonics, Vol 5, pp 364-371,
May 22, 2011.
dio: 10.1038/NPHOTON.2011.74, Alternate

75

[12]

S. A. Hamilton, Member, IEEE and B. S. Robinson, Student Member, IEEE. “40-Gb/s All-Optical
Packet Synchronization and Address Comparison for OTDM Networks.” IEEE Photonics
Technology Letters, Vol. 14, No. 2, pp 209-211, Feb 2002.
dio: 10.1109/68.980524

[13]

A.J. Poustie, K.J. Blow 1, A.E. Kelly, and R.J. Manning. “All-optical full adder with bitdifferential delay.” Optical Communications, Vol 168, pp 89-93, Sept 1, 1999.
dio: 10.1016/S0030-4018(99)00348-X

[14]

Francisco M. Raymo and Massimiliano Tomasulo. “Optical Processing with Photochromic
Switches.” Chemistry European Journal, 12, pp 3186-3193, Jan 17, 2006.
dio: 10.1002/chem.200501178

[15]

Ramaswami, Rajiv, Kumar Sivarajan, and Galen Sasaki. Optical networks: a practical
perspective. Morgan Kaufmann, 2009.

[16]

Batya Feldman. “Optical processing co Lenslet closes” [Online]. Available:
http://www.globes.co.il/en/article-1000067464. March 2, 2006 [Accessed April 10, 2017].

[17]

“EnLight256® 8000 Giga MAC / sec fixed point DSP.” Available:
http://besho.narod.ru/reviews/newage/EnLight256.pdf. 2003 [Accessed April 10, 2017]

[18]

D. Hisamoto, Wen-Chin Lee, J. Kedzierski, E. Anderson, H. Takeuchi, K. Asano, Tsu-Jae King, J.
Bokor, and Chenming Hu. “A Folded-channel MOSFET for Deep-sub-tenth Micron Era.” IEDM
Tech. Dig 1998, pp 1032-1034, Dec, 1998.
dio: 10.1109/IEDM.1998.746531

[19]

Arabinda Das and Alexandre Dorofeev. “Intel's 22-nm process gives MOSFET switch a facelift”
[Online]. Sept 6,2012 [Accessed April 10, 2017].

[20]

J. M. Early. “Speed, power and component density in multielement high-speed logic systems.”
1960 International Solid-State Circuits Conference, pp 78-79, March 1960.
dio: 10.1109/ISSCC.1960.1157266

[21]

Paul D. Franzon, W. Rhett Davis, Michael B. Steer, Steve Lipa, Eun Chu Oh, Thor Thorolfsson,
Samson Melamed, Sonali Luniya, Tad Doxsee, Stephen Berkeley, Ben Shani, and Kurt
Obermiller. “Design and CAD for 3D Integrated Circuits.” Proceedings of the 45th annual Design
Automation Conference, pp 668-673, June 8, 2008.
dio: 10.1145/1391469.1391642

[22]

Cristinel Ababei, Yan Feng, Brent Goplen, Hushrav Mogal, Tianpei Zhang, Kia Bazargan, and
Sachin S. Sapatnekar. “Placement and Routing in 3D Integrated Circuits.” IEEE Design & Test of
Computers, pp 520-531, Volume: 22, Issue: 6, Nov.-Dec. 2005.
dio: 10.1109/MDT.2005.150

[23]

Rafael Reif, Andy Fan, Kuan-Neng Chen, and Shamik Das. “Fabrication Technologies for ThreeDimensional Integrated Circuits.” Quality Electronic Design, 2002. Proceedings. International
Symposium on, Aug 7, 2002.
dio: 10.1109/ISQED.2002.996687

[24]

A. W. Topol, D. C. La Tulipe, Jr., L. Shi, D. J. Frank, K. Bernstein, S. E. Steen, A. Kumar, G. U.
Singco, A. M. Young, K. W. Guarini, and M. Ieong. “Three-dimensional
integrated circuits.” IBM Journal of Research and Development, Vol 50, No. 4.55, pp 491-506,
July, 2006.
dio: 10.1147/rd.504.0491

76

[25]

“What is 3D Integration?” [Online]. Available: http://www.3dincites.com/3d-incitesknowledge-portal/what-is-3d-integration/. [Accessed Oct 2, 2016].

[26]

Steven M. Pope, and Ruben C, Zeta (Maxim Integrated Products, Inc.). “Package-on-Package
Secure Module Having Anti-Temper Mesh in the Substrate of the Upper Package.” U.S. Patent 7
923 830 B2, April 12, 2011.
Alt: 7 923 830 B2

[27]

Robert S. Patti. “Three-Dimensional Integrated Circuits and the Future of System-on-Chip
Designs.” Proceedings of the IEEE, Vol 94, Issue 6, pp 1214-1224, June, 2006.
dio: 10.1109/JPROC.2006.873612

[28]

Rachel Courtland. “The Rise of the Monolithic 3-D Chip.” IEEE Spectrum, [Online]. Available:
http://spectrum.ieee.org/semiconductors/design/the-rise-of-the-monolithic-3d-chip/. Feb
24th, 2014 [Accessed Oct 4, 2016].

[29]

Anna Llordés, Guillermo Garcia, Jaume Gazquez, and Delia J. Milliron. “Tunable near-infrared
and visible-light transmittance in nanocrystal-in-glass composites.” Nature, Vol 500, pp 323326, Aug 15, 2013.
dio: 10.1038/nature12398

[30]

Christoforos Kachris, Ioannis Tomkos. “A Survey on Optical Interconnects for Data Centers.”
IEEE Communications Surveys and Tutorials, Vol 14, No. 4, pp 1021-1036, Jan 11th, 2012.
dio: 10.1109/SURV.2011.122111.00069

[31]

Xiaohua Ma, Geng-Sheng (G. S.) Kuo. “Optical switching technology comparison: optical MEMS
vs. other technologies.” IEEE Communications Magazine, Vol 41, Issue 11, Nov, 2003.
dio: 10.1109/MCOM.2003.1244924
Mitsunori Saito, Aki Maruyama, and Junki Fujiwara. “Polarization-independent refractive-index
change of a cholesteric liquid crystal.” Optical Materials Express, Vol 5, Issue 7, pp 1588-1597,
June 12, 2015.
dio: 10.1364/OME.5.001588

[32]

[33]

Shi-Sheng Lee, Long-Sun Huang, Chang-Jin Kim, and Ming C. Wu. “Free-Space Fiber-Optic
Switches Based on MEMS Vertical Torsion Mirrors.” Journal of Lightwave Technology, Vol 17,
No. 1, pp 7-13, Jan, 1999.
dio: 10.1109/50.737414

[34]

Hsin-ta Hsieh, Chen-wei Chiu, Tom Tsao, Fukang Jiang, and Guo-Dung John Su. “Low-ActuationVoltage MEMS for 2-D Optical Switches.” Journal of Lightwave Technology, Vol 24, No. 11, pp
4372-4379, Nov, 2006.
dio: 10.1109/JLT.2006.883674

[35]

Yao-Joe Yang, Bo-Ting Liao and Wen-Cheng Kuo. “A novel 2 × 2 MEMS optical switch using the
split cross-bar design.” Journal of Micromechanics and Microengineering, Vol 17, No. 5, April
3rd, 2007.
dio: 10.1088/0960-1317

[36]

Pierre-Alexandre Blanche, Daniel Carothers, John Wissinger, and Nasser Peyghambarian.
“Digital micromirror device as a diffractive reconfigurable optical switch for
telecommunication.” Special Section On Emerging MOEMS Technology And Applications, Vol 13,
Issue 1, Dec 2nd, 2013.
dio: 10.1117/1.JMM.13.1.011104

77

[37]

Arash Joushaghani, Brett A. Kruger, Suzanne Paradis, David Alain, J. Stewart Aitchison, and
Joyce K. S. Poon. “Low-voltage broadband hybrid plasmonic-vanadium dioxide switches.”
Applied Physic Letters, Volume 102, Issue 6, Feb, 2013.
dio: 10.1063/1.4790834

[38]

Xiaofang Yu, and Shanhui Fan, “Bends and splitters for self-collimated beams in photonic
crystals.” Applied Physic Letters, Volume 83, Issue 16, Oct, 2003.
dio: 10.1063/1.1621736

[39]

Yuanliang Zhang, Yao Zhang, and Baojun Li. “Optical switches and logic gates based on selfcollimated beams in two-dimensional photonic crystals.” Optics Express, Vol. 15, Issue 15, pp
9287-9292, July 13, 2007.
dio: 10.1364/OE.15.009287

[40]

Photonic Crystals, Theory, Applications and Fabrication, Page 11.

[41]

Hannes Kind, Haoquan Yan, Benjamin Messer, Matthew Law, and Peidong Yang. “Nanowire
Ultraviolet Photodetectors and Optical Switches.” Advanced Materials, Vol. 14, Issue 2, pp 158160, Jan 16, 2002.
dio: 10.1002/1521-4095(20020116)14:23.0.CO;2-W

[42]

Jianfang Wang, Mark S. Gudiksen, Xiangfeng Duan, Yi Cui, and Charles M. Lieber. “Highly
Polarized Photoluminescence and Photodetection from Single Indium Phosphide Nanowires.”
Science, Vol. 293, Issue 5534, pp 1455-1457, Aug 24, 2001.
dio: 10.1126/science.1062340

[43]

Roman Bruck, and Otto L. Muskens. “Plasmonic nanoantennas as integrated coherent perfect
absorbers on SOI waveguides for modulators and all-optical switches.” Optics Express, Vol. 21,
Issue 23, pp 27652-27661, Nov 4, 2013.
dio: 10.1364/OE.21.027652

[44]

K. Lance Kelly, Eduardo Coronado, Lin Lin Zhao , and George C. Schatz. “The Optical Properties
of Metal Nanoparticles: The Influence of Size, Shape, and Dielectric Environment.” Journal of
Physical Chemistry B, Vol. 107, Issue 3, pp 668–677, Dec 21, 2002.
dio: 10.1021/jp026731y

[45]

E. Prodan, C. Radloff, N. J. Halas, P. Nordlander. “A Hybridization Model for the Plasmon
Response of Complex Nanostructures.” Science, Vol. 302, Issue 5644, pp. 419-422, Oct 17,
2003.
dio: 10.1126/science.1089171

[46]

Iván O. Sosa, Cecila Noguez, and Rubén G. Barrera. “Optical Properties of Metal Nanoparticles
with Arbitrary Shapes.” Journal of Physical Chemistry B, Vol. 107, Issue 26, pp 6269–6275, June
7, 2003.
dio: 10.1021/jp0274076

[47]

W. Rechberger, A. Hohenau, A. Leitner, J.R. Krenn, B. Lamprecht, and F.R. Aussenegg.
“Optical properties of two interacting gold nanoparticles.” Optics Communications, Vol. 220,
pp 137-141, March 17, 2003.
dio: 10.1016/S0030-4018(03)01357-9

[48]

Emilie Ringe, Richard P Van Duyne, and Laurence D Marks. “Correlated structure-optical
properties studies of plasmonic nanoparticles.” Journal of Physics: Conference Series, Vol. 522,
conference 1, 2014.
dio: 10.1088/1742-6596/522/1/012006

78

[49]

Yu Zhang, Alejandro Manjavacas, Nathaniel J. Hogan, Linan Zhou, Ciceron Ayala-Orozco,
Liangliang Dong, Jared K. Day, Peter Nordlander, and Naomi J. Halas. “Toward Surface
Plasmon-Enhanced Optical Parametric Amplification (SPOPA) with Engineered Nanoparticles: A
Nanoscale Tunable Infrared Source.” Nano Letters, Vol. 16, Issue 5, pp 3373-3378, April 18,
2016.
dio: 10.1021/acs.nanolett.6b01095

[50]

Jacob A. Faucheaux, Alexandria L. D. Stanton, and Prashant K. Jain. “Plasmon Resonances of
Semiconductor Nanocrystals: Physical Principles and New Opportunities.” Journal of Physical
Chemistry Letters, Vol. 5, Issue 6, pp 976-985, Feb 27, 2014.
dio: 10.1021/jz500037k

[51]

G A Niklasson1, S-Y Li and C G Granqvist. “Thermochromic vanadium oxide thin films:
Electronic and optical properties.” Journal of Physics: Conference Series, Vol. 559, conference
1, 2014.
dio: 10.1088/1742-6596/559/1/012001

[52]

Shuyan Zhang, Mikhail A. Kats, Yanjie Cui, You Zhou, Yu Yao, Shriram Ramanathan, and
Federico Capasso. “Current-modulated optical properties of vanadium dioxide thin films.”
Applied Physics Letters, Vol. 105, Issue 21, Nov, 2014.
dio: 10.1063/1.4902924

[53]

J.-C. Orlianges, J. Leroy, A. Crunteanu, R. Mayet, P. Carles, and C. Champeaux. “Electrical and
optical properties of vanadium dioxide containing gold nanoparticles deposited by pulsed laser
deposition.” Applied Physics Letters, Vol. 101, Issue 13, Sept, 2012.
dio: 10.1063/1.4754708

[54]

Kannatassen Appavoo, Bin Wang, Nathaniel F. Brady, Minah Seo, Joyeeta Nag, Rohit P.
Prasankumar, David J. Hilton, Sokrates T. Pantelides, and Richard F. HaglundJr. “Ultrafast
Phase Transition via Catastrophic Phonon Collapse Driven by Plasmonic Hot-Electron Injection.”
Nano Letters, Vol. 14, Issue 3, pp 1127-1133, Jan 31, 2014.
dio: 10.1021/nl4044828

[55]

R Maiti, A Midya, C Narayana, and S K Ray. “Tunable optical properties of graphene oxide by
tailoring the oxygen functionalities using infrared irradiation.” Nanotechnology, Vol. 25, Issue
49, Nov 21, 2014.
dio: 10.1088/0957-4484/25/49/495704

[56]

Alexander I. Chernov, Pavel V. Fedotov, Alexandr V. Talyzin, Inma Suarez Lopez, Ilya V.
Anoshkin, Albert G. Nasibulin, Esko I. Kauppinen, and Elena D. Obraztsova. “Optical Properties
of Graphene Nanoribbons Encapsulated in Single-Walled Carbon Nanotubes.” American
Chemical Society Nano, Vol. 7, Issue 7, pp 6346-6353, June 24, 2013.
dio: 10.1021/nn4024152

[57]

A. V. Kretinin, Y. Cao, J. S. Tu, G. L. Yu, R. Jalil, K. S. Novoselov, S. J. Haigh, A. Gholinia, A.
Mishchenko, M. Lozada, T. Georgiou, C. R. Woods, F. Withers, P. Blake, G. Eda, A. Wirsig, C.
Hucho, K. Watanabe, T. Taniguchi, A. K. Geim, and R. V. Gorbachev. “Electronic Properties of
Graphene Encapsulated with Different Two-Dimensional Atomic Crystals.” Nano Letters, Vol.
14, Issue 6, pp 3270-3276, May 20, 2014.
dio: 10.1021/nl5006542

[58]

Hao Wang, Hua Zhao, Guangwei Hu, Siren Li, Hang Su, and Jingwen Zhang. “Graphene Based
Surface Plasmon Polariton Modulator Controlled by Ferroelectric Domains in Lithium Niobate.”
Scientific Reports, Vol. 5, Article Num. 18258, Dec 14, 2015.
dio: 10.1038/srep18258

79

[59]

Qiuyun Ouyang, Hailong Yu, Zheng Xu, Yue Zhang, Chunyan Li, Lihong Qi, and Yujin Chen.
“Synthesis and enhanced nonlinear optical properties of graphene/CdS organic glass.” Applied
Physics Letters, Vol. 102, Issue 3, Jan, 2013.
dio: 10.1063/1.4789367

[60]

Andrea Ambrosini, Angel Duarte, Kenneth R. Poeppelmeier, Melissa Lane, Carl R. Kannewurf,
Thomas O. Mason. “Electrical, Optical, and Structural Properties of Tin-Doped In2O3+M2O3
Solid Solutions (M 5Y, Sc).”, Journal of Solid State Chemistry, Vol. 153, Issue 1, pp 41-47, Aug,
2000.
dio: 10.1006/jssc.2000.8737

[61]

Sok Won, Kim; Manil, Kang; Inkoo, Kim; Minwoo, Chu; Ji-Wook, Ryu. “Optical Properties of
Sputtered Indium-tin-oxide Thin Films.” Journal of the Korean Physical Society, Vol. 59, Issue
51, pp 3280-3283, April 30, 2011.
dio: 10.3938/jkps.59.3280

[62]

Michael Gross, Albrecht Winnacker, and Peter J. Wellmann. “Electrical, optical and
morphological properties of nanoparticle indium–tin–oxide layers.” Thin Solid Films, Vol. 515,
Issue 24, pp 8567-8572, Oct 15, 2007.
dio: 10.1016/j.tsf.2007.03.136

[63]

J Ederth, P Heszler, A Hultåker, G.A Niklasson, and C.G Granqvist. “Indium tin oxide films
made from nanoparticles: models for the optical and electrical properties.” Thin Solid Films,
Vol 445, Issue 2, pp 199-206, Dec 15, 2003.
dio: 10.1016/S0040-6090(03)01164-7

[64]

Sebastien D. Lounis, Evan L. Runnerstrom, Amy Bergerud, Dennis Nordlund, and Delia J.
Milliron. “Influence of Dopant Distribution on the Plasmonic Properties of Indium Tin Oxide
Nanocrystals.” Journal of the American Chemical Society, Vol. 136, Issue 19, pp 7110-7116,
April 30, 2014.
dio: 10.1021/ja502541z

[65]

Limin Tong, Jingyi Lou, and Eric Mazur. “Single-mode guiding properties of subwavelengthdiameter silica and silicon wire waveguides.” Optics Express, Vol. 12, Issue 6, pp 1025-1035,
March 9, 2004.
dio: 10.1364/OPEX.12.001025

[66]

Robert Halir, Przemek J. Bock, Pavel Cheben, Alejandro Ortega-MoĂąux, Carlos Alonso-Ramos,
Jens H. Schmid, Jean Lapointe, Dan-Xia Xu, J. Gonzalo Wangüemert-Pérez, Íñigo MolinaFernández, and Siegfried Janz. “Waveguide sub-wavelength structures: a review of principles
and applications.” Laser and Photonics Reviews, Vol. 9, Issue 1, pp 25-49, Sept 30, 2014.
dio: 10.1002/lpor.201400083

[67]

Richard Feynman. “QED: Fits of Reflection and Transmission -- Quantum Behaviour”, The Sir
Douglas Robb Lectures, University of Auckland, 1979.

[68]

Junjie Du, Shiyang Liu, Zhifang Lin, Jian Zi, and S. T. Chui. “Dielectric-based extremely-lowloss subwavelength-light transport at the nanoscale: An alternative to surface-plasmonmediated waveguiding.” Physical Review A, Vol. 83, Issue 3, March 14, 2011.
dio: 10.1103/PhysRevA.83.035803

[69]

Yin Huang, Changjun Min, and Georgios Veronis. “Subwavelength slow-light waveguides based
on a plasmonic analogue of electromagnetically induced transparency.” Applied Physics
Letters, Vol. 99, Issue 14, Oct 7, 2011.
dio: 10.1063/1.3647951

80

[70]

Jennifer M. Reed, Haining Wang, Yingnan Guo, and Shengli Zou. “Subwavelength light
confinement and propagation: A numerical study of a two-layer silver film with perforated
holes.” Applied Physics Letters, Vol. 103, Issue 16, Oct 17, 2013.
dio: 10.1063/1.4826441

[71]

Aristeidis Karalis, J. D. Joannopoulos, and Marin Soljačić. “Plasmonic-Dielectric Systems for
High-Order Dispersionless Slow or Stopped Subwavelength Light.” Physical Review Letters, Vol.
103, Issue 4, July 24, 2009.
dio: 10.1103/PhysRevLett.103.043906

[72]

Seunghoon Han, Yi Xiong, Dentcho Genov, Zhaowei Liu, Guy Bartal, and Xiang Zhang. “Ray
Optics at a Deep-Subwavelength Scale: A Transformation Optics Approach.” Nano Letters, Vol.
8, Issue 12, pp 4243-4247, Oct 29, 2008.
dio: 10.1021/nl801942

[73]

R. Quidant, J.-C. Weeber, A. Dereux, D. Peyrade, Y. Chen, and C. Girard. “Near-field
observation of evanescent light wave coupling in subwavelength optical waveguides.”
Europhysics Letters, Vol. 57, Issue 2, pp 191-197, 2002.
dio: 10.1209/epl/i2002-00560-3

[74]

“Refractive Index of ITO, Indium Tin Oxide, InSnO” [Online]. Available:
http://www.filmetrics.com/refractive-index-database/ITO/Indium-Tin-%20Oxide-InSnO.
[Accessed April 8, 2017].
M. Baum, I. Alexeev, M. Latzel, S. H. Christiansen, and M. Schmidt. “Determination of the
effective refractive index of nanoparticulate ITO layers.” Optics Express, Vol 21, Issue 19, pp
22754, Sept 23, 2013.
dio: 10.1364/OE.21.022754

[75]

[76]

“Optical constants of Nb2O5 (Niobium pentoxide)” [Online]. Available:
https://refractiveindex.info/?shelf=main&book=Nb2O5&page=Lemarchand. [Accessed April 8,
2017].

[77]

Wouter J. Westerveld, Suzanne M. Leinders, Koen W. A. van Dongen, H. Paul Urbach, and
Mirvais Yousefi. “Extension of Marcatili's Analytical Approach for Rectangular Silicon Optical
Waveguides.” Journal of Lightwave Technology, Vol. 30, Issue 14, pp 2388-2401, July 15, 2012.
dio: 10.1109/JLT2012.2199464

[78]

“Dielectric Materials” [Online]. Available: https://www.gelest.com/applications/dielectricmaterials/. [Accessed April 8, 2017].

[79]

“Dielectric materials for semiconductor chips” [Online]. Available:
http://www.eeherald.com/section/design-guide/dielectric_in_semiconductors.html. Accessed
April 8, 2017].

[80]

“Refractive Index of SiO2, Fused Silica, Silica, Silicon Dioxide, Thermal Oxide, ThermalOxide”
[Online]. Available: http://www.filmetrics.com/refractive-index-database/SiO2/Fused-SilicaSilicon-Dioxide-Thermal-Oxide-ThermalOxide. [Accessed April 8, 2017].

[81]

Min K. Yang, Roger H. French, and Edward W. Tokarsky. “Optical properties of Teflon® AF
amorphous fluoropolymers.” Journal of Micro/Nanolithography, MEMS, and MOEMS, Vol 7,
Issue 3, Aug 13, 2008.
dio: 10.1117/1.2965541

81

[82]

Yiqun XIAO, Jun SHENy, Zhiyong XIE, Bin ZHOU and Guangming WU. “Microstructure Control of
Nanoporous Silica Thin Film Prepared by Sol-gel Process.” Journal of Materials Science and
Technology, Vol 23, Issue 4, pp 504-508, Aug 27, 2007.
dio: 10.3321/j.issn:1005-0302.2007.04.014

[83]

John Foggaito. “Chemical Vapor Deposition of Silicon Dioxide Films.” Handbook of Thin Film
Deposition, Chapter 3, pp 120, 2001.
Available online: https://booksite.elsevier.com/9781437778731/
https://booksite.elsevier.com/9781437778731/past_edition_chapters/Chemical_Vapor.pdf

82