daniel_manuscript

A simple method to obtain the electroluminescence spectrum
of opto-electronic devices and engineering light emitting
metal-oxide-silicon devices
Daniel Oler, Hasan Goktas, Volker J. Sorger, and Ergun Simsek
Department of Electrical and Computer Engineering,
The George Washington University
Washington, DC 20052
August 7, 2016

Contents
Introduction 4
1 Experiment Method 5
1.1 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Proposed Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Testing new structures 18
2.1 Diﬀerent Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Tapered Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Thinner Oxides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Smaller Footprint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Nanoparticle Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.2 Simulation Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Conclusion 28
Bibliography 29
1

List of Tables
1.1 Lorentz Drude Parameters used in FDTD solver. . . . . . . . . . . . . . . . 9
2.1 Nanoparticle array simulation variables . . . . . . . . . . . . . . . . . . . . 27
2

List of Figures
1.1 (a) An optical image of the fabricated MIS-LED, (b) a zoomed-in version
while working; spectrum of the emitted light for bias voltages changing from
1 to 6 V (c) experiment and (d) simulation results. . . . . . . . . . . . . . . 5
1.2 (a) Two- and (b) three-dimensional views of the simulated structure; (c)
simulation result for the spectrum of the emitted light. . . . . . . . . . . . . 6
1.3 (a) Gold’s density of states [15] (b) Spectrum obtained with convolution of
FDTD solutions with Au’s density of states.) . . . . . . . . . . . . . . . . . 7
1.4 Project tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Design settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Electromagnetic properties for Si . . . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Creating the Si Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.8 SPICE circuit design for DC voltage across the gate and ground layers . . . 12
1.9 Array settings for point observers . . . . . . . . . . . . . . . . . . . . . . . . 12
1.10 Final structure with arbitrary colors . . . . . . . . . . . . . . . . . . . . . . 13
1.11 MATLAB code to plot the Wavenology results . . . . . . . . . . . . . . . . 14
1.12 MATLAB code to convolve density of states and Wavenology results . . . . 16
1.13 (a) Power spectrum directly from Wavenology results. (b) Power spectrum
from convolution with gold density of states. . . . . . . . . . . . . . . . . . 17
2.1 (a) Classical and (b) quantum power spectrums of gold, silver, aluminum,
and copper metals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 (a) Plain structure with no taper. (b) Tapered structure with height de-
scending from 20 nm to 2 nm. (c) Power spectrum of the two structures. . . 20
2.3 Power spectrum of SiO2 with varying layer thicknesses . . . . . . . . . . . . 21
2.4 Power spectrum of 1500 nm x 1500 nm and 1000 nm x 1000 nm footprints. 22
2.5 Longitudinal, Transverse1, and Transverse2 modes of a nanoparticle cylinder
array with a radius of 45nm, height of 55nm, with spacing of 140nm. . . . . 24
2.6 Two longitudinal mode outputs of a nanoparticle cylinder array with the
same radius of 45nm, same height of 60nm, and diﬀerent spacing of 91nm
and 131nm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3

2.7 Example of increasing radius of nanoparticle array with high inter-particle
spacing showing red shifted bandwidth enhancement. . . . . . . . . . . . . . 26
2.8 (a) Power spectrum of plain simulation without nanoparticles, and arrays
with nanoparticles. (b) Power spectrums from (a) divided by the plain power
spectrum. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Introduction
The electroluminescence capability of metal-insulator-metal (MIM) structures has been
known and studied for almost a half century [1–4]. They indeed are very simple structures
to fabricate and they can shine when small amounts of DC voltages are applied. With
the growth of the CMOS industry, metal-insulator-semiconductor (MIS) structures have
become popular due to their compatibility with silicon [5–10]. In thin MIS structures, the
gate tunneling characteristics can be interpreted in terms of four voltage regions: tunneling
into surface states, indirect conduction band, direct conduction band, and oxide conduction
band. At low fields, when the voltage drop across the oxide, Vox, is less than the silicon
band gap, inelastic tunneling occurs into surface states. When Vox exceeds silicon band gap,
inelastic tunneling continues into the indirect conduction band of silicon. At higher fields,
elastic tunneling into the direct conduction band occurs. With further increase, electron
transport occurs directly to the oxide conduction band. At high fields the observed current
is primarily due to electron tunneling from cathode to the oxide conduction band, which
is typically modeled by the Fowler-Nordheim equation. The Fowler-Nordheim tunneling
current density, J, is given as a function of applied voltage V by
J =
q2V 2
8πhd2Φ
e−α/V
(1)
where A = 4 2qm∗m0Φ3d/ , Φ metal/insulator barrier height, m∗ is the tunneling effec-
tive mass of the charge barrier in the barrier material, d is the tunnel barrier width, m0 is
the free electron rest mass, q is the magnitude of electronic charge. Hence the ln(J/V 2) ver-
sus 1/V plot is a straight line having a slope of α, and is a characteristic of Fowler-Nordheim
tunnel current. We initially studied transfer characteristics of several MIM (Au/SiO2/Au)
and MIS (Au/SiO2/Si) structures using SILVACO Atlas device simulator [11] and observed
typical Fowler-Nordheim tunnel currents for voltages changing between 0.2 V and 6 V using
α ≈ 1 V/nm, which is a realistic number for our devices that will be explained below.
The Fowler-Nordheim equation helps us to determine I −V characteristic of our device
but it does not say anything about the electromagnetic fields and waves created by these
currents. However, if we avoid the two ends of the bias voltage (i.e. voltage is neither too
small nor too large), then we can predict the spectral behavior of MIM and MIS tunnel
junctions. In other words, the third region -where the current is created with the elastic
tunneling into the direct conduction band- can be simulated by solving Maxwell’s and
circuit (SPICE) equations simultaneously.
4

Chapter 1
Experiment Method
1.1 Experiment
Figure 1.1: (a) An optical image of the fabricated MIS-LED, (b) a zoomed-in version
while working; spectrum of the emitted light for bias voltages changing from 1 to 6 V (c)
experiment and (d) simulation results.
We fabricated a MIS structure and measured the power and spectrum of the light being
emitted. Fig. 1.1 (a) and (b) show the optical images of a MIS-LED, which is fabricated
on a (100) p-Silicon wafer with a resistivity of 50 Ω/cm. After cleaning the wafer with
piranha solution, 300 nm thick oxide is etched all the way down to silicon wafer by using
a mask with a square hole pattern with a side length of 2 µm. A ∼2 nm thin native oxide
is grown, and then a 20 nm thick Au ﬁlm is sputtered on top. A 100 nm thick Au pad is
created with a liftoﬀ to be used as a gate. Figure 1.1 (c) shows the spectrum of the emitted
light for bias voltage of 3 V.
5

1.2 Simulation
Figure 1.2: (a) Two- and (b) three-dimensional views of the simulated structure; (c) sim-
ulation result for the spectrum of the emitted light.
We use Wavenology, a commercial 3D full-wave field-circuit co-simulation software package
[12] for the simulation of MIS-LEDs. Figure 1.2 (a) and (b) shows the simulated structure
from different perspectives. We assume a 20 nm thick gold (Au) film placed on top of a
SiO2 covered Si substrate. The silicon substrate and SiO2 film are assumed to be 3 µm and
2 nm thick, respectively. The dispersive permittivity of Si and SiO2 are taken from [13],
whereas complex permittivity of the Au is taken from [14]. We calculated the intensity
of the light emitted from this structure over a 0.5 µm × 1 µm region 200 nm above the
Au film at a distance 1.5 µm far from the plate where voltage is applied. In order to
prevent any plasmonic effect, which might come from the plate, the plate itself is assumed
to be a perfect electrical conductor (PEC). Another very thin PEC film placed under the
Si substrate and a DC voltage is applied between these two PEC spots. To guarantee
high numerical accuracy, a 20 points-per-wavelength sampling density is utilized over the
whole simulation domain. In order to take doping level into account, the conductivity of
Si is set to σ = q(µnn + µpp), where µn and µp are the mobility of electrons and holes,
respectively, where n and p are their doping levels. The blue curve in Fig. Figure 1.2 (c)
shows the numerically obtained spectrum of the emitted light, which is very different than
the experiment result.
By convolving the simulation results obtained from Wavenology with the density of
states model of the metal used, the quantum power spectrum can be plotted. The results
shown in Figure 1.2 (c) is convolved with the density of states model of gold in Figure 1.3
(a) to see similar agreement with the experimental result from Figure 1.1. The resulting
6

spectrum in Figure 1.3 (b) shows similar peaks at 700 nm and 575 nm surrounded by decay.
Figure 1.3: (a) Gold’s density of states [15] (b) Spectrum obtained with convolution of
FDTD solutions with Au’s density of states.)
1.3 Proposed Method
Using the Wavenology software we can simulate structures to compare their power spec-
trums to determine the optimal properties to increase light emission. To demonstrate how
the software was used in this report, a Silicon, Silicon Dioxide, and Gold layered structure
will be constructed with the settings and properties documented below.
Design Settings First, the Wavenology project settings were set in the project tree
under the Design label by double clicking ’Unit’ shown in Figure 1.4. The settings used
for the 6 tabs are shown in Figure 1.5.
Materials Before the structure is created the materials are added for Si, SiO2, and
Au. Materials are added by right clicking Materials on the project tree and selecting
’Add Material’ or ’Import Material from Library.’ Au can be imported from the library
with the title ’Au (dispersive)’ whereas Si and SiO2 can be added manually by selecting
’Add Material’ and changing the Electromagnetic properties appropriately. The dispersive
properties of Au are already handled by the library, but they can be edited by selecting the
material’s Electromagnetic ’Edit dispersion’ button where the parameters should match
those in Table 1.1. The electromagnetic properties for Si are shown in Figure 1.6. The
Elastodynamic properties can be left unchanged. The last material needed for this structure
can be imported with the title ’PEC’ (Perfect Electrical Conductor). The PEC will be used
to apply the DC voltage across the structure shown in Figure 1.2 (a).
7

Physical Structure Due to the structure’s simplicity of flat layers, only the box creation
tool is needed to design the structure. Right click on ’Solids’ under the project tree and
select ’Create Box’ to name the solid and set the upper and lower corners of the box. For
the Si layer, we designed the footprint to be 1000x1000 nm with a height of 500 nm, shown
in Figure 1.7. The SiO2 layer was placed on top of the Si layer with lower and upper
corners (0,0,500) and (1000,1000,502) respectively, The Au layer was then placed on top
of the SiO2 layer with corners (0,0,502) and (1000,1000,522). The ground layer using the
PEC material was placed under the Si layer with corners (0,0,-5) and (1000,1000,0). The
upper gate layer was placed over the Au layer with corners (0,0,522), and (1000,50,527.6).
SPICE Circuit To apply a DC voltage across the gate and ground layers, a SPICE
circuit was created by expanding the ’Circuits’ subtree in the project tree and right clicking
on ’Spice Circuits’ to select ’Create a new circuit by text.’ Then select the gate and ground
layers on the structure, the ’Create Lumped Circuit’ dialog should open. Edit the properties
to match the settings shown in Figure 1.8.
Observers In order to obtain results from the simulation, 64 point observers are placed
230 nm above the structure in an 8 × 8 grid. To do this, right click on ’Observers’ under
the project tree and select ’Create Observer’ to bring up the observer editor. Set the
observer’s name appropriately and set the location XYZ to (150, 150, 750) and set the
captured components to only Ex, Ey, and Ez. Select the ’Array Creation’ button and fill
in the dialog to match Figure 1.9 and select ’OK.’
Running The Simulation With the structure complete, the simulation is ready to
begin. The simulation duration depends on how complex the structure is and how many
points per wavelength are defined in the design settings ’Mesh’ tab (Figure 1.5). To slightly
reduce the simulation duration, more threads can be assigned in ’Simulation->Multiple
Threads.’ Then the simulations can be started by selecting ’Simulation->Start Simulation.’
Processing the results After the simulation is complete, the results will be stored in
the ’res’ folder. MATLAB is then used to plot the results along with the convolved results.
The code used to plot the direct simulation results is shown in 1.11. The plot of this code
produces the power spectrum shown in 1.13 (a). To convolve the density of states of gold
with the results, the MATLAB code shown in 1.12 is used. This gives the final power
spectrum shown in 1.13 (b).
8

Figure 1.4: Project tree
Lorentz Drude Parameters
Drude Lorentz
Metal
Damping
Coeﬃcient
(rad/s)
Plasma angular
freq. (rad/s)
Relative
at Zero
Freq.
Damping
coeﬃcient
(rad/s)
Resonant
angular
freq. (rad/s)
Au 8.05212×1013 1.19599×1016 12.3629 3.66143×1014 6.30496×1014
2.18364 5.24147×1014 1.26099×1015
1.65677 1.32176×1015 4.51071×1015
3.64549 3.78905×1015 6.53893×1015
3.01483 3.36366×1015 2.02366×1016
Ag 7.29248×1013 1.25831×1016 8.9247 5.90387×1015 1.23972×1015
1.50133 6.86709×1014 6.80784×1015
1.01333 9.87524×1013 1.24352×1016
1.82655 1.39165×1015 1.37995×1016
2.11334 3.67511×1015 3.08259×1016
Al 7.14056×1013 1.64588×1016 1941.97 5.05916×1014 2.46121×1014
5.70651 4.74011×1014 2.34575×1015
12.3955 2.05253×1015 2.74684×1015
1.55813 5.13816×1015 5.27642×1015
Cu 4.5578×1013 1.24766×1016 85.4891 5.74283×1014 4.42107×1014
2.39504 1.60435×1015 4.49247×1015
4.01886 4.88141×1015 8.05212×1015
1.59868 6.54045×1015 1.69854×1016
Table 1.1: Lorentz Drude Parameters used in FDTD solver.
9

Figure 1.5: Design settings
10

Figure 1.6: Electromagnetic properties for Si
Figure 1.7: Creating the Si Layer
11

Figure 1.8: SPICE circuit design for DC voltage across the gate and ground layers
Figure 1.9: Array settings for point observers
12

Figure 1.10: Final structure with arbitrary colors
13

clear; clc;
%%%%%%%% Defaults %%%%%%%%%%%%%%%%%%%%%%%%%%
set(0,'defaultlinelinewidth',2)
set(0,'DefaultAxesFontSize',18)
set(0,'DefaultTextFontSize',18)
%%%%%%%% Constants %%%%%%%%%%%%%%%%%%%%%%%%%%
prefix = 'Si_SiO2_Au'; %Name of the wavenology .WNT file
Exr = load([prefix '_resobservers' prefix '_rev_ex_freq_real.txt']);
Exi = load([prefix '_resobservers' prefix '_rev_ex_freq_imag.txt']);
Eyr = load([prefix '_resobservers' prefix '_rev_ey_freq_real.txt']);
Eyi = load([prefix '_resobservers' prefix '_rev_ey_freq_imag.txt']);
Ezr = load([prefix '_resobservers' prefix '_rev_ez_freq_real.txt']);
Ezi = load([prefix '_resobservers' prefix '_rev_ez_freq_imag.txt']);
n = Eyr(1);
m = Eyr(5);
lambda= 3e8./linspace(Eyr(2), Eyr(3),m);
lambda=lambda*1e9;
Ex = reshape(Exr(6:end,1),m,n)+1i*reshape(Exi(6:end,1),m,n);
Ey = reshape(Eyr(6:end,1),m,n)+1i*reshape(Eyi(6:end,1),m,n);
Ez = reshape(Ezr(6:end,1),m,n)+1i*reshape(Ezi(6:end,1),m,n);
Ptotal = abs(Ex).^2+abs(Ey).^2+abs(Ez).^2;
figure(120);
plot(lambda,sum(Ptotal.'))
ylabel('P_{C} (a.u.)')
xlim([400,900]);
xlabel('lambda (nm)');
Figure 1.11: MATLAB code to plot the Wavenology results
14

clear; clc;
%%%%%%%% Defaults %%%%%%%%%%%%%%%%%%%%%%%%%%
set(0,'defaultlinelinewidth',2)
set(0,'DefaultAxesFontSize',18)
set(0,'DefaultTextFontSize',18)
%%%%%%%% Constants %%%%%%%%%%%%%%%%%%%%%%%%%%
load('../gold_dos'); %Density of states of gold data
dosy0 = dosy;
numSamples=5000;
yi = zeros(1,numSamples);
x2=linspace(−8,8,numSamples*2−1);
Eii = linspace(−4,4,numSamples);
dosy = interp1(dosx,dosy0,Eii,'linear');
dosy(isnan(dosy))=0;
prefix = 'Si_SiO2_Au'; %Name of the wavenology .WNT file
Exr = load([prefix '_resobservers' prefix '_rev_ex_freq_real.txt']);
Exi = load([prefix '_resobservers' prefix '_rev_ex_freq_imag.txt']);
Eyr = load([prefix '_resobservers' prefix '_rev_ey_freq_real.txt']);
Eyi = load([prefix '_resobservers' prefix '_rev_ey_freq_imag.txt']);
Ezr = load([prefix '_resobservers' prefix '_rev_ez_freq_real.txt']);
Ezi = load([prefix '_resobservers' prefix '_rev_ez_freq_imag.txt']);
n = Eyr(1);
m = Eyr(5);
lambda= 3e8./linspace(Eyr(2), Eyr(3),m);
Eiw = 1240e−9./lambda; %keyboard
y=zeros(1,length(lambda));
c=zeros(1,numSamples*2−1);
Ex = reshape(Exr(6:end,1),m,n)+1i*reshape(Exi(6:end,1),m,n);
Ey = reshape(Eyr(6:end,1),m,n)+1i*reshape(Eyi(6:end,1),m,n);
Ez = reshape(Ezr(6:end,1),m,n)+1i*reshape(Ezi(6:end,1),m,n);
%Continued on next page...
15

%Continued...
Ptotal = 1*abs(Ex).^2+1*abs(Ey).^2+1*abs(Ez).^2;
sumtotal=sum(Ptotal.');
y(1,:)=sumtotal(1:length(lambda));
yi(1,:) = interp1(Eiw(1:length(lambda)),y(1,1:length(lambda)),Eii,'linear');
yi(1,isnan(yi(1,:)))=0;
c(1,:)=conv(yi(1,:),dosy);
lambda=lambda(:)*1e9;
figure(745); clf;
subplot(311);
plot(dosx,dosy0);
ylabel('Au DoS (a.u.)'); xlabel('E (eV)')
title(prefix, 'Interpreter', 'none')
subplot(312);
plot(lambda,y(1,:))
ylabel('P_{C} (a.u.)')
xlabel('lambda (nm)');
xlim([400 900])
subplot(313);
plot(1240./x2,c(1,:))
ylabel('P_Q (a.u.)'); xlabel('lambda nm')
xlim([400 900])
Figure 1.12: MATLAB code to convolve density of states and Wavenology results
16

Figure 1.13: (a) Power spectrum directly from Wavenology results. (b) Power spectrum
from convolution with gold density of states.
17

Chapter 2
Testing new structures
In this chapter we make use of the method described in chapter 1 to simulate different
structures and see how it affects the power spectrum. Each section below demonstrates
how the geometry, material, or permittivity of a different structure compares to a control
structure much like in Figure 1.10.
2.1 Different Metals
Design To see how different metals performed against each other, gold, silver, aluminum,
and copper were simulated. The density of states model for each metal was used from [15].
Results From Figure 2.1 (b) it is clear that gold produces the most light of the 4 metals,
with copper in second place. Though Figure 2.1 (a) shows silver has the greatest classical
power output, the density of states model shown in [15] illustrates a narrow and smaller rel-
ative probability. Aluminum has a similar problem with low relative probability compared
to gold and thus has a much smaller quantum power output.
18

Figure 2.1: (a) Classical and (b) quantum power spectrums of gold, silver, aluminum, and
copper metals.
2.2 Tapered Metals
Design A structure with a tapered metal is one with the topmost layer tapered instead
of ﬂat, as illustrated in Figure 2.2 (b). We created two tapered metal structures, one with
heights 20 nm to 2 nm, and another with heights 100 nm to 20 nm.
Results What is seen in Figure 2.2 (c) is that the larger tapered structure with heights
100 nm to 20 nm is unfavorable as it produces much less power than the plain structure.
Though the case of heights 20 nm to 2 nm taper produces slightly more power at its
maximum peak, along with an enhancement of two local maximum peaks. The maximum
peak on the tapered structure increased 10% in power from the plain structure, and the
local maximum peaks increased in power by 50%. These increases show that tapered metal
structures help produce more broadband light, and thinner metal will produce more light.
19

Figure 2.2: (a) Plain structure with no taper. (b) Tapered structure with height descending
from 20 nm to 2 nm. (c) Power spectrum of the two structures.
2.3 Thinner Oxides
Design The structure created in Chapter 1 which had an SiO2 thickness of 2 nm was
used as a base for this design. The structure was then adjusted to increase the thickness
to 5 nm.
Results From Figure 2.3, the results show that a structure with 2 nm SiO2 is more
eﬃcient than one with a thickness of 5 nm. The maximum peaks of the spectrums show
that the 2 nm version is 130% greater than the thicker layer around the 600 nm wavelength.
Between wavelengths 500 nm and 700 nm, the 2 nm thick layer spectrum is twice as large
as the thicker layer. The 2 local maximum peaks surrounding the 2 nm SiO2 peak seem to
be reduced in the 5 nm SiO2 spectrum.
20

Figure 2.3: Power spectrum of SiO2 with varying layer thicknesses
2.4 Smaller Footprint
Design Using gold as the metal, two similar structures were created with diﬀerent foot-
print sizes. Areas of 1500 nm × 1500 nm and 1000 nm × 1000 nm were chosen.
Results What can be seen from Figure 2.4 is that the smaller footprint produces more
power. This is because there is less ohmic loss when less gold is used, allowing more power
to be produced.
In the future we will plot power vs. footprint with more data to understand where the
optimal footprint size is.
21

Figure 2.4: Power spectrum of 1500 nm x 1500 nm and 1000 nm x 1000 nm footprints.
2.5 Nanoparticle Arrays
2.5.1 Model
The theoretical model used for the design of nanoparticles was found in [16]. The article
introduces a theoretical model for the analysis of multi-layered medium field enhancement
through the use of nanoparticle arrays. By using layered medium Green’s functions and
treating the nanoparticles as oscillating point dipoles the model is able to determine the
three surface plasmon resonance modes (longitudinal, transverse 1, and transverse 2). The
theoretical model shows strong agreement with experimental results and can enhance field
strength to increase surface plasmon propagation lengths up to nearly 1 µm.
Application Using the theoretical model, users can determine how a nanoparticle array
with certain specifications (radius, height, spacing) can enhance an existing field. For
example, Figure 2.5 shows the output of the model for a nanoparticle array of cylinders
with a radius of 45nm, height of 55nm, and spacing of 140nm.
These plots are inversely normalized to show clearly in red where the enhancement will
be. In Figure 2.5’s case, the longitudinal enhancement will be between 650-690nm. This
was determined by noting that the x-axis shows the real part of the angular frequency
whereas the y-axis shows the imaginary part which indicates the loss. By noting that
22

the red area rests in between 2.6 × 1015 rad/s and 2.8 × 1015 rad/s, a simple calculation
converting angular frequency to wavelength shows the wavelength enhancement.
In another example, to build a nanoparticle array to enhance a bandwidth, the model
can be used to tweak the variables of radius, height, and inter-particle spacing to find a
combination that matches the wanted enhancement. In practice, one can see that adjusting
the inter-particle spacing will adjust the range of the bandwidth as shown in Figure 2.6.
It is clear that to have a narrower bandwidth enhancement the inter-particle spacing must
be as great as possible while still following the model’s rule of the inter-particle spacing
must be no greater than three times the radius.
If in search of a single wavelength enhancement, the number of variables to change has
reduced from three to two if the inter-particle spacing is set to a length slightly smaller than
three times the radius. From here it is easiest to leave the height constant and tweak the
radius of the nanoparticle cylinder varying linearly across an estimated range. For example,
if 600 nm is the enhanced target, a range of a radius between 30 and 50 nm would be a
good set to test as shown in Figure 2.7. It is clear that as the radius of the nanoparticles
increase, the enhanced wavelength is redshifted. This can used to easily target the wanted
enhanced field specifications.
23

Figure 2.5: Longitudinal, Transverse1, and Transverse2 modes of a nanoparticle cylinder
array with a radius of 45nm, height of 55nm, with spacing of 140nm.
24

Figure 2.6: Two longitudinal mode outputs of a nanoparticle cylinder array with the same
radius of 45nm, same height of 60nm, and diﬀerent spacing of 91nm and 131nm.
25

(a) Longitudinal and Transverse1 modes of nanoparticle array with 30nm radius, 55nm height,
89nm spacing
(b) Longitudinal and Transverse1 modes of nanoparticle array with 38nm radius, 55nm height,
113nm spacing
(c) Longitudinal and Transverse1 modes of nanoparticle array with 46nm radius, 55nm height,
137nm spacing
Figure 2.7: Example of increasing radius of nanoparticle array with high inter-particle
spacing showing red shifted bandwidth enhancement.
26

2.5.2 Simulation Testing
Design Using the theoretical model from above, gold nanoparticle arrays were made in
Wave-nology by using the cone creation tool along with the array copy tool. Simulations
were made for 5 cases, the first case had no nanoparticles, the other 4 had nanoparticle
arrays with different combinations of variables shown in Table 2.1. The table also shows
the targeted wavelength to be enhanced.
Array
Number (nm)
Radius (nm) Height (nm)
Particle
Spacing (nm)
Targeted
Wavelength (nm)
1 37 50 110 600
2 37 55 110 600
3 46 55 140 600-615
4 52 50 155 850
Table 2.1: Nanoparticle array simulation variables
Results The results shown in Figure 2.8 show that the expected enhancement did not
occur. Instead of increasing the illumination strength, the nanoparticles absorb the energy
specifically between 400 nm and 600 nm and lose the energy as heat. A 10% increase was
observed for wavelengths greater than 700 nm but the overall spectrum is still weaker.
Figure 2.8: (a) Power spectrum of plain simulation without nanoparticles, and arrays with
nanoparticles. (b) Power spectrums from (a) divided by the plain power spectrum.
27

Conclusion
The method introduced in Chapter 1 works well to find the power spectrum of nanoscale
light emitting devices. However, engineering light emitting devices to be more efficient is
a difficult task. From Chapter 2 we have learned that thinner oxides and smaller foot-
prints help increase illumination in the structures, whereas nanoparticle arrays reduce the
illumination. Out of gold, aluminum, silver, and copper, gold is the best metal to use for
metal-oxide-silicon devices. To create a more broadband power spectrum, the metal could
be tapered while keeping the tallest height within skin depth of the metal.
To keep investigating efficiency, we will try simulations with new variables including:
temperature, quantum dots, etching, and thin films of transition metal dichalcogenides.
We will also go into more depth with the topics already presented such as footprint size to
find the optimal dimensions for maximum power output.
28

Bibliography
[1] T. W. Hickmott, “Electron Emission, Electroluminescence, and Voltage-Controlled
Negative Resistance in Al-Al2O3-Au Diodes,” J. Appl. Phys., vol. 36, no. 6, pp. 1885–
1896, 1965.
[2] W. Pong, C. Inouye, F. Matsunaga and M. Moriwaki, “Electroluminescence in Al-
Al2O3-Au diodes,” J. Appl. Phys., vol. 46, pp. 2310–2313, 1975.
[3] J. Lambe and S.L. McCarthy, “Light Emission from Inelastic Electron Tunneling,”
Phys. Rev. Lett., vol. 37, no. 14, pp. 923–925, 1976.
[4] L.C. Davis, “Theory of surface-plasmon excitation in metal-insulator-metal tunnel
junctions”, Phys. Rev. B, vol. 16, pp. 2482–2490, 1977.
[5] Z. A. Weinberg, “On tunneling in metal-oxide-silicon structures,” J. of Appl. Phys.,
vol. 53, no. 7, pp. 5052–5056, 1982.
[6] K. R. Farmer, R. Saletti, and R. A. Buhrman, “Current ﬂuctuations and silicon oxide
wear-out in metal-oxide-semiconductor tunnel diodes,” Appl. Phys. Lett., vol. 52, pp.
1749–1751, 1988.
[7] R. Ramaswami, H. C. Lin, and R. Kuchimanchi, “Simulation of tunneling current
from gate edges of metal-oxide-silicon structures,” J. of Appl. Phys., vol. 69, vol. 9,
pp. 6679–6684, 1991.
[8] C. W. Liu, M.-J. Chen, I. C. Lin, M. H. Lee, and C.-F. Lin, “Temperature depen-
dence of the electron-hole-plasma electroluminescence from metal-oxide-silicon tun-
neling diodes,” Appl. Phys. Lett., vol. 77, no. 8, pp. 1111–1113, 2000.
[9] C. W. Liu, S. T. Chang, W. T. Liu, M.-J. Chen, and C.-F. Lin, “Hot carrier recombi-
nation model of visible electroluminescence from metal-oxide-silicon tunneling diodes,”
Appl. Phys. Lett., vol. 77, no. 4347, 2000.
[10] M.-J. Chen, J.-F. Chang, J.-L. Yen, C. S. Tsai, E.-Z. Liang, C.-F. Lin, and C. W. Liu,
“Electroluminescence and photoluminescence studies on carrier radiative and nonra-
29

diative recombinations in metal-oxide-silicon tunneling diodes,” J. Appl. Phys., vol.
93, 4253, 2003.
[11] ATLAS from Silvaco, Inc., Santa Clara, CA.
[12] Wavenology from Wave Computation Technologies, Inc., Durham, NC.
[13] E. D. Palik, Editor, Handbook of Optical Constants of Solids, Vol. 1, Academic Press
Publishers, 1985.
[14] A. D. Rakic, A. B. Djurisic, J. M. Elazar, and M. L. Majewski, “Optical properties
of metallic ﬁlms for vertical-cavity optoelectronic devices,” Appl. Opt., vol. 37, pp.
5271–5283, 1998.
[15] Ravishankar Sundararaman, Prineha Narang, Adam S. Jermyn, William A. Goddard
III, and Harry A. Atwater, “Theoretical predictions for hot-carrier generation from
surface plasmon decay.” Nat Commun, 5, Dec 2014. Article.
[16] E. Simsek, “Full analytical model for obtaining surface plasmon resonance modes of
metal nanoparticle structures embedded in layered media.” Optics Express, vol. 18,
No. 2
30

daniel_manuscript

More Related Content

Viewers also liked (8)

Similar to daniel_manuscript (20)

daniel_manuscript