This document summarizes Daniel B. Riley's M.S. defense on modeling and optimization of an electrically-pumped silicon laser. It motivates the need for integrated silicon photonics to overcome bandwidth bottlenecks in microelectronics. It describes a Multi-University Research Initiative project to develop a silicon laser using alternating layers of erbium-doped oxide and silicon nanocrystals. Electromagnetic simulations and theoretical modeling are presented to optimize the layer structure for high optical confinement factors. Gain-loss analysis using finite-difference time-domain and transfer matrix methods aims to determine threshold conditions for net optical gain. Future work is proposed to better understand polarization behavior and energy transfer processes.
1. Toward an Electrically-PumpedToward an Electrically-Pumped
Silicon Laser: Optimization andSilicon Laser: Optimization and
ModelingModeling
Daniel B. Riley
M.S. Defense
Department of Electrical and Computer Engineering
University of Rochester
2. 2
Acknowledgements
Dr. Philippe Fauchet
Fauchet Research Group
Yijing Fu & Jidong Zhang
Vicki Heberling
MURI Silicon Laser
Participating Institutions
Funding Sources
M.S. Thesis Examination Committee
Dr. Thomas Hsiang – Electrical and Computer Engineering
Dr. Miguel Alonso – Institute of Optics
4. 4
Limit of Microelectronics
Dimensional shrink of microprocessors Moore’s Law
Barriers materials, power dissipation, parasitic capacitance,
bandwidth bottleneck, lag of front side bus
Performance limited as 10Mb/s/km threshold is approached
Available: http://www.intel.com/technology/mooreslaw
5. 5
Communication Links
Transition from electrical links to optical links within the next
10 years – more solutions at all levels
Chip to chip and intra chip stand to benefit most as strain on
processors increases – also most challenging
L. Pavesi and D.J. Lockwood. “Silicon Photonics” in Topics in Applied Physics”. 94. 1 – 90.
6. 6
Benefit of Photonic Systems
L. Pavesi and D.J. Lockwood. “Silicon Photonics” in Topics in Applied Physics”. 94. 1 – 90.
8. 8
Silicon: Good and Bad
Silicon is cheap
Easily integrated with existing
CMOS processes
Poor light emitter
Free carrier absorption at infrared
wavelengths (1.55μm)
L. Pavesi and D.J. Lockwood. “Silicon Photonics” in Topics in Applied Physics”. 94. 1 – 90.
10. 10
Project Task
Extrinsic gain laser
Horizontal slot waveguide structure w/ alternating nanolayers of Er-
doped oxide and nc-Si for optical cavity of Si laser system
Si nc
Er 3+
SiO2:Er (low index)
nc-Si (high index)
11. 11
Electrical Injection – Dipole Energy Coupling
1 Walters, R., Bourianoff, G., Atwater, H., Nature 04. 143, Feb. 2005.
Energy Transfer
Si-nc
Er3+
Exciton
Recombination
Resonant
Absorption
12. 12
MURI
Optical gain engineering – minimize losses, maximize gain
in cavity by increasing optical mode confinement factor (CF)
Confinement factor
Net gain
Threshold current density for injection
Power consumption and dissipation
Device efficiency
13. 13
Outline
Motivation
MURI Silicon Laser Project
Theoretical Background
Simulation and Results
Conclusions and Future Outlook
14. 14
Theory – Important Concepts
t∂
∂−
=×∇
B
E
J
D
H +
∂
∂
=×∇
t
0=⋅∇ B
0=⋅∇ D
02
=
∂
∂
−∇
t
E
E
2
µε
Maxwell’s Equations
Wave Equation
Vector Algebra
µε
1
=v
000 ε
ε
εµ
µε
==n
HB µ=
ED ε=
Stokes’ Theorem
Divergence Theorem
0)(ˆ
0)(ˆ
0)(ˆ
0)(ˆ
12
12
12
12
=−⋅
=−⋅
=−×
=−×
DD
BB
HH
EE
s
s
s
s
Boundary conditions
at a dielectric interface
Continuous components
“D-B normal, E-H tangential”
D1 D2
B1 B2
E1
E2
H1 H2
15. 15
Planar Waveguide
Maxwell’s Equations
Wave Equation
Boundary conditions for EM
fields
Snell’s Law TIR
Waveguides
– n2 > n1 and n2 > n3
– Evanescent decay
n1
n2
n3
16. 16
Slot Confined Waveguide
Recall boundary conditions at
a dielectric interface
D1,normal = D2,normal
D1 = n1
2
E1
D2 = n2
2
E2
n2 > n1
E1 > E2 by factor of n2
2
/ n1
2
n1
n2
n1
n2
n1
Libson, M, et.al. “Guiding Light in Void nanostructure”. Optics Letters. 29. 1209, Jun 2004.
x
z
17. 17
Transverse E-field of fundamental TM mode
Libson, M. “Guiding Light in Void nanostructure”. Optics Express. 12. 2004.
18. 18
Outline
Motivation – Photonics Overview
MURI Silicon Laser Project
Theoretical Background
Simulation and Results
Summary, Conclusions and Future Outlook
19. 19
Device Structure
Single layer gain medium
Si nc and Er grouped
together in one layer
Alternating layers of nc-Si and Er
Higher mode confinement
Easier electrical injection into Si nc
Controls dipole interaction length
between Si nc and Er during energy
transfer
n type device layer
p type device layer
SiO2 BOX
~3nm
~2nm
Si nc
Er 3+
SiO2 (low index)
nc-Si (high index)
20. 20
Device Geometry
Cap Layer (SiO2)
Substrate
(SiO2)
T_SiO2
T_Si
T_C
Multilayer Region
50nm
1μm
y
x
z
nc-Si
SiO2:Er
SiO2:Er
nc-Si
Slab Height
~370 – 400nm
21. 21
Tools and Methods
3D device analysis is difficult
RSoft Photonics CAD – numerical simulation
FullWave – based on FDTD
Matlab – data analysis
Clarification of RSoft axis conventions - see previous page
Transverse field oriented along x axis
TM modes are of primary interest
Transfer Matrix Method (TMM)
Algorithm by Yijing Fu
Comparison to FDTD
22. 22
Effective Index Method
Think of multilayer region as single layer with effective index neff
Effective index depends on both the thickness and index of each layer
Effective index like a weighted average
Effective index method determines index of refraction “seen” by
propagating mode
neff
23. 23
CF vs. Ratio for 2D Slot Waveguide
Based on equations for 2D slot structure
Ratio of thickness between SiO2:Er layers and nc-Si layers is critical
Saturation behavior of CF
Confinement Factor vs Ratio of Slot Width to
Waveguide Width
55
56
57
58
59
60
61
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Ratio of Slot Width to Waveguide Width
ConfinementFactor(%)
24. 24
Setup & Limitations
Computing power
Virtual memory limitation – much longer times, crashes
Resolutions (x,y,z) >5nm per grid point
Limitation on values of thickness, height, width, length
Whole number values only
Evenly divisible by resolution value for that dimension
Restriction on possible devices for simulation
Only a few structures chosen
Scripting not possible
Structures chosen based on above limitations, cutoff values
for single mode operation, and 2D model for optimum
thickness ratios
28. 28
Table of Structures & Results
Effective ratio accounts for extra layer of Si
Range of optimum ratios covered
Total height < 400nm for single mode operation
N Ratio
(SiO2:Si)
Effective
Ratio
T_Si
(nm)
T_SiO2
(nm)
Total Height
(nm)
CF (%)
6 2.00 1.71 20 40 380 42.57
7 1.50 1.31 20 30 370 48.56
7 1.75 1.53 20 35 405 52.03
8 0.80 0.71 25 20 385 61.56
8 1.25 1.11 20 25 380 58.46
9 1.00 0.90 20 20 380 60.41
29. 29
FDTD and TMM Results
Graphs of CF versus effective thickness ratio betw. SiO2:Er and nc-Si
Left: FDTD and TMM simulation results for both TM and TE modes
Right: TMM results for both TM and TE modes
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
Thickness ratio between SiO
2
and Si layer
Confinementfactor
TMM and FDTD result for multilayer thickness of 0.38 µm
TE mode confinement
TM mode confinement
FDTD data points for TM
FDTD data points for TE
0.5 1 1.5 2 2.5
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
Thickness ratio between Si and SiO
2
layer
Confinementfactor
TMM result for multilayer thickness of 0.52 µm
TE mode confinement
TM mode confinement
30. 30
Gain/Loss Analysis to Model Actual Device
TM modes lower loss than TE
More of mode w/in gain layers
Less in lossy Si
Gain/loss coefficient ratio
Lower limit such that net gain is achieved
Choose 3 optimum structures from previous graphs
Redo simulations with loss and gain mechanisms
Vary gain/loss coefficient ratio
Loss coefficient set – gain coefficient increased until net gain = 0
Value of gain/loss coefficient when net gain = 0 is sought
Determine value for net gain from maximum power at time
monitor with no loss or gain
31. 31
S&R – Gain/Loss Analysis: FDTD vs TMM
Net gain vs. gain/loss coefficient ratio for both TM and TE modes
Left: modal gain using TMM
Right: propagation gain using FDTD
Note agreement between the two methods
TM net gain intercept ~ 0.25 for both as expected
TE net gain intercept ~ 2.50 for both – not expected (should be ~1.00)
0 0.5 1 1.5 2 2.5 3
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Modal gain for TE and TM polarization by TMM
gain/loss coefficient ratio
modeloss/gain(1/cm)
TE mode gain
TM mode gain
0 0.5 1 1.5 2 2.5 3
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Gain/Loss coefficient ratio
ModalGaincoefficientA.U.
Propagation Gain from FDTD simulation
TM propagation gain
Reference
TE propagation gain
Reference
33. 33
Summary & Conclusions
Silicon photonics
2D slot waveguide as a model
3D waveguide for cavity of Si based laser
Gain/loss analysis
Lower threshold current densities realized for less power
consumption and more efficient devices
34. 34
Future Considerations
Further simulation in more powerful computational
environment for improved accuracy
Propagation lengths > 10 μm – closer to actual device lengths
Higher resolution values
More diverse structures with varying geometrical dimensions
Better understanding of TE mode behavior
Explanation beside coupling effects?
With respect to device – better understanding of Si nc Er3+
energy transfer process
35. 35
Thank you for ListeningThank you for Listening
Please ask questions if you have them.
Editor's Notes
Motivation primarily involve a discussion of photonics especially with respect to silicon
MURI larger body of work of which this work was a small part
Performance requirements will increase and electronic circuits will not be able to meet them as they cross the 10Mb/s/km threshold
More distributed architectures will be needed to meet these demands
This indicates optical carrier utilization
To better understand it, let’s look at the current situation for all levels of the interconnect hierarchy
Increased performance requirements for interconnects
Speed, bandwidth, low loss, low latency
For speeds &gt;1Gb/s and transmission distances &gt;100m
Exclusively optical solutions – benefits outweigh cost
Long-haul telecommunications fibers
Intermediate distances (1-100m)
Available solutions
Chip-to-chip level None
Cost/technical issues dictate electrical solutions
Largest potential benefits at this level
Need low cost-high volume manufacturing capabilities
Why bother? Optical solutions, in particular integrated chips offer many benefits (especially when silicon is used)
General advantages of photonics over electronics
Higher bandwidth multiplexing and information carrying capacity
Lower loss transmission
Less electromagnetic noise/interference
Smaller size, weight, power consumption
Lower cost assuming high volume application
Performance scaling through parallelism
Photonic Integrated Circuits
New Platform electronic/photonic integration on Si wafer
Eliminates need for active alignment of off-chip light sources
Si light sources are key
Leverage existing infrastructure for low cost-high volume
Required Functionality for material to be used
Benefits
Low cost – Microelectronics industry based on Si
Potential for integration using CMOS compatible processes
Problems
Indirect bandgap inefficient light emitter
Large free carrier absorption at infrared (1.55μm) wavelengths
Solutions / Alternatives
Raman laser – UCLA, Intel – Raman amplification; nonlinear process
Hybrid laser – UCSB, Intel – specialized bonding technique
Hybrid laser - III-V direct gap SCs (InP) – lattice mismatch with Si - defects
Extrinsic material as light emitter (erbium) – MURI Si laser
Deliverables:
First CMOS compatible Er doped laser diode at 1.55micron
First CMOS compatible Er doped waveguide amplifier optical micro amplifier for 1.55micron
Slot confinement structure
Horizontal slot for easier fabrication (no high precision etching for sidewall)
Avoids potential high index contrast sidewall scattering losses
Ring resonator – high Q factor required to compensate for low emission cross section in Er
Why Er? – Extensive library of knowledge, Forster dipole-dipole energy coupling to Si nc,
emission peak at ~1550nm communications wavelength
Electroluminescence mechanism in Si nc floating gate transistor
Excitons excited in nc-Si by MOS field effect injection exhibiting Fowler-Nordheim tunneling
High internal radiative QE for excitons in Si nc (~60%)1
Si nc excitation and energy transfer to Er both faster than radiative emission rate for Er easy population inversion
My role in MURI project
Loss mechanisms:
Scattering
Shown to be negligible for nm-scale nanocrystals
Lower limit ~0.4 dB/cm
Radiative turning loss
due to resonator structure
&lt;0.1 dB/cm for diameters &gt; 100 μm
Free Carrier Absorption (FCA)
inherent problem with Si – density of states in conduction band
result of using tunnel/band injection currents – accumulation layer
losses ~ 1.1 dB/cm
primary optical gain engineering parameter for slot waveguide cavity
Maxwell Eqns. – in a source-free medium
Wave Eqn. – describes propagation of EM waves in a source-free medium
Maxwell’s Equations
Wave Equation
Boundary conditions for EM fields at dielectric interface
Snell’s Law Total Internal Reflection (TIR) Waveguides
Planar Slab Waveguide
Propagating modes
Slot Confinement WG
2 waveguides side by side
Enhanced E-field
Redistribution of photons
Shows that high confinement of light can be achieved in nm-thin layers when WGs are close together (slot &lt;&lt; decay length for mode)
n1 = 1.44; n2 = 3.46; Enhancement by factor of ~6 at low index – high index interfaces
Confinement factor - % of mode in a specific region of WG
Power in specific region divided by total power
Slot waveguide structure
High confinement within nanometer-thin, low index slot regions
Due to BC for D and high index contrast
Multiple layer waveguide for higher optical confinement
alternating SiO2:Er and nc-Si layers – repeating periods of nc-Si and SiO2:Er - Notice there is always one more layer of nc-Si
Optical cavity for Si laser
Si nanocrystals form from amorphous Si by appropriately tuning thermal budget
Er and nc-Si in separate layers to help ensure proper Er – nc-Si interaction length in energy transfer process
Also helps improve effective emission cross section of Er
Horizontal slot for easier fabrication (no high precision etching for sidewall)
Avoids potential high index contrast sidewall scattering losses
Cap layer confines mode in region of multi layer region directly underneath it
– determines mode width
Width and Slab height restricted by single mode cutoff values ~400nm for slab height
Slab height determined by thickness of Si, thickness ratio and total number of repeating
nc-Si, SiO2:Er periods
Hard to apply theoretical analysis to 3D WGs - Need for numerical simulation software
RSoft – FullWave, etc
Based on Finite-Difference Time Domain numerical methods for electromagnetic waves
TM vs TE difference for RSoft – explain using structure
Initially a good way to think about structure
Index of refraction for nc-Si layers ~ 3.46
Index of refraction for SiO2:Er ~ 1.44
Expect effective indices ~ 2.45 if width of nc-Si layers and SiO2:Er layer thicknesses equal
2D Slot Confinement as a model for 3D
Good indication of performance for more realistic 3D model
Effective index method
Determines index of refraction “seen” by propagating mode
Ratio of thickness between nc-Si layers and SiO2:Er layers is critical in determining how well guided and well confined the mode is
Graph shows saturation behavior – optimum range of thickness ratios for 2D slot model
As ratio increases, more area for enhanced E field but also mode is not as well guided since effective index is decreased – Si claddings steal more of the mode
Minimum layer thickness with reasonable results = 15nm (3 grid points)
Layers at least 20nm thick – easiest value to use; not realistic but point is validation of theory
Propagation length – 10 micron is sufficient
Whole number values only
Evenly divisible by resolution value for that dimension
Good results: 25nm layer thickness and resolution of 5nm
Bad results: 25nm layer thickness and resolution of 4nm
Only a few structures chosen
Scripting not possible
Comparison to TMM
Propagation length = 10 micron
Time monitor at ~9.4 micron
Pictures of contour graph of power in z direction at the time monitor,
X-cut cross section of the power across the structure, and 3D picture of the power distribution
Note high confinement in low index SiO2:Er layers and evanescent tail
Temptation to think that more low index material would increase overall CF indefinitely – NOT TRUE
Notice high power confinement in low index layers and large evanescent tail in substrate
Mislabeled – x and y labels switched
Evanescent tail is more prominent here
Effective ratio – total height of SiO2:Er layers divided by total height of nc-Si layers
Cannot accurately simulate structures with exactly a height of 380nm and also have usable values for other dimensions
380nm is the best possible height for the TMM simulations for comparison to FDTD
CF calculated based on output data file of z component of power at time monitor
380nm is the best possible height for the TMM simulations for comparison to FDTD
CF calculated based on output data file of z component of power at time monitor
by importing data into Matlab
CF vs Ratio graphs for TE/TM modes
Discussion of Results – optimum ratios, achievable CF
Notice similarity to graph for 2D slot confinement model
Left: Slab height = 380nm (ranges slightly for FDTD)
CF max &gt; 55% for thickness ratio ~0.80
Optimum range for ratios ~ 0.80 – 1.10
Right: Slab height = 520nm
CF max ~ 75% for thickness ratio ~1.00
Optimum range ~ 0.80 – 1.10
Saturation behavior due to increased SiO2 causing mode to be less well confined in multilayer region – some of the mode seeps into the cladding layers
Possible sources of error include lower limit on resolution, accuracy of FDTD method, and mainly propagation length since 10 microns is not nearly large enough to allow a true eigenmode to form. Limitations on computational power restricted this parameter however.
TM modes lower loss than TE
More of mode w/in gain layers
Less in lossy Si
Lower gain/loss coeff. ratio necessary for net overall gain in cavity
Gain/Loss coeff. ratio is relationship between gain seen in SiO2:Er and loss seen in nc-Si by the optical mode
Lower limit such that net gain still achieved within cavity
Uneven distribution of light among layers coefficient ratio not = 1
Graphs of Overall Modal gain vs. Gain/Loss coeff. Ratio
TM net gain intercept expected to be 0.25 because ~ 4x as much light is in SiO2 layers as is in Si layers
TE net gain intercept unexpected – possibly due to mismatch between gain profile and mode profile for TE?
TE mode essentially a smooth Gaussian but gain profile is step-like.
Photons in outermost low index layers possibly couple out of device into cladding and substrate layers as a result causing net gain intercept value for the gain/loss coefficient to be significantly higher than expected value of 1.
Major expansion – cost effective platform for smart partitioning of electronic and photonic functionality
Extend processing power of integrated circuits and performance of communication networks
Device combines excellent electrical properties of silicon with favorable optical properties of Erbium in order to side step the less than favorable optical properties of silicon and the inability to electrically pump Erbium. It utilizes a novel energy transfer process between nc-Si and Er3+ ions to accomplish a light emitting system in silicon. This energy transfer process is still relatively mysterious and much more work must be done in the way of understanding it.