Professional Documents
Culture Documents
John Bowers
Director, Institute of Energy Efficiency
University of California, Santa Barbara
bowers@ece.ucsb.edu
http://optoelectronics.ece.ucsb.edu/ 1
Acknowledgements
Collaborators
UCSB: Dan Blumenthal, Larry Coldren, Steve DenBaars, Nadir Dagli,
Jared Bauters, Jock Bovington, Hui-Wen Chen, Hsu Chang, Daoxin Dai, Martijn
Heck, Sid Jain, Geza Kurzveil, Phil Mages, Jon Peters, Jason Tien, Zhi Wang
Intel : Richard Jones, Yimin Kang, Mario Paniccia, Brian Koch, Hyundai Park, Matt
Sysak
Aurrion: Alex Fang, Greg Fish, Eric Hall
Hewlett Packard: Di Liang, Marco Fiorentino, Ray Beausoleil
Financial Support
DARPA MTO (Rodgers, Shah), Intel, Hewlett Packard, Rockwell Collins
2
IEE Electronics/Photonics Group
Ultra-Narrow Δν and
Tunable InP/Si Lasers
& Laser Arrays
Capability
THz-Bandwidth Chirped
Lidar & mmW Sources 256 QAM
1Tb/s Integrated Coherent PICs
Tx/Rx Capacity Epitaxial InP on Si
PICs
st
1 Gen Hybrid InP/Si
Laser Technology 100Gb/s All-Optical
Coherent Regeneration
100Gb/s Integrated
Tx/Rx Capacity 350 GHz fmax HBT
OPLL ASICs
st
1 Gen Optical
Phase-Locked Loops
QPSK Coherent
PICs
Collaborations
Stanford
Clean Slate
Institute-
McKeown
6
The Problem: Power
File sharing
8
Top 10 Global Web Companies
Human Genomics
7 EB/yr, 200% CAGR
Ave. Files on HD Retail Customer DB Clinical Image DB HD video forecast Physics (LHC)
54GB 600 TB ~1PB 12 EB/yr 300 EB/yr
Aerospace Point of Sale Security IP Services Test & Transportation Energy Factory IP Media
Surveillance Measurement & Utilities Automation Phones
Estimating the Exaflood, Discovery Institute, 1/08; Amassing Digital Fortunes, a Digital Storage Study, CEA, 3/08 Courtesy: Rattner(2010)
State-of-the Art Electronic
Terabit IP Router
• Problem: Bandwidth demands scaling faster than both silicon and
cooling technologies
Maximum configuration for CRS-1 92 Tbps (80 racks)
~1 Megawatt!!!
486
0 386
10
1.5μ 1.0μ 0.8μ 0.6μ 0.35μ 0.25μ 0.18μ 0.13μ
T ec hnolog y
node
The solution to the heat problem is multiple cores
with a Terabit optical bus
Opportunities
for
3D
and
L ow
P ower
Multi‐C ore
B ipolar C MO S
10
Module
Heat
F lux
S quadrons
T ‐R ex
Mckinley
8 F ujits u
V P 2000
IB M
G P
IB M
3090S
6 NT T
IB M
R Y 5
16
The Solution: Optical Interconnects
•
3D layer stacking will be
prevalent in the 22nm
Enough Pitfalls?
timeframe
al I/O
• Intra-chip optics can take
advantage of this
Optic
BUT: Silicon is reciprocal. How to make an isolator?
affic
technology
ptical tr
• Photonics layer (with
BUT: SiO2 is thermally
supporting electrical resistive. So, power dissipation of
active circuits)
devices moreis a problem, particularly for rings and DWDM
easily
o
signa ical
On-chip
integrated with high
ip opt
performance logic and
ls
BUT: Silicon is centrosymmetric (not
memory layers electro-optic)!
Photonic
Off-ch Plane
Memory Plane
So, • how to can
Layers integrate modulators?
be separately Logic Plane
optimized for performance
and yield
BUT: Silicon has an indirect gap and is a Kash,
poor absorber
“Photonics (not
in Supercomputing:
Lasers 50
years
~50
Silicon years
Lasers Si Manufacturing
OPTICAL
ANYWHERE,
INCREDIBLE
Very high High volume, POTENTIAL
bandwidth low cost
Long distances
InP Highly Silicon
Photonic integrated Photonic
Immunity Integrated
to Six Generations
Integrated
electrical noise
circuits Scalability circuits
Year
of
Production 1995 1998 2001 2004 2007 2010 2013 2016
DRAM
1/2
Pitch
(nm) 270 190 130 90 65 43 32 22
Wafer
Size
(mm) 150 200 200 200 300 Courtesy:
300 300
Rattner 450
(Intel)
Why Silicon Photonics?
Utilize advanced fabrication technologies for low cost, high volume integrated photonics.
Silicon Does Have Advantages…
• Cheaper substrates
• Larger substrates (>300 mm)
• Large fabrication infrastructure (32 nm 300 mm fabs)
• Improved process control-critical for large scale integration.
• Reduced two photon absorption (100x less)
• Lower loss waveguides: <0.3 dB/cm
• Higher thermal conductivity (κ) of
– Waveguides (Si has 30x higher κ than InGaAsP)
– EAMs or PD Absorbers: Ge has 16x higher κ than InGaAs
• Excellent APD characteristics
– K factor: 0.02 versus 0.6
– Gain bandwidth product: 850 GHz versus 160 GHz
22
Silicon light emission – How?
• Bulk silicon
• Low dimension Silicon
– Silicon nanocrystal (Pavesi, …)
– Periodic nanopatterned crystalline silicon (Jimmy Xu)
• Er dopants (Dal Negro,…)
• Avoid direct interband transition
– Raman laser (UCLA/Intel)
• Another material for gain (hybrid approach)
– Epitaxial
• Ge
• Quantum Dot
• Pillars
• InGaAsP (Mages)
– Bonding
• Dice level
• Wafer level (BCB or Molecular) 23
Bulk silicon LED
Sequential injection
of electron and hole
25
R. J. Walters, G. I. Bourianoff and H. A. Atwater, Nature Materials 4, 143 (2005)
Nanopatterned Crystalline Silicon
26
26
S. G. Cloutier, P. A. Kossyrev and J. Xu, Nature Materials 4, 887 (2005)
Emission mechanism
J. Xu, IEEE International Conference on Group IV Photonics, Ottawa, Canada (2006), FA1. 27
Silicon nanoclusters: waveguides
28
N. Daldosso and L. Pavesi, Laser & Photon. Rev., 2009
Silicon nanoclusters: LED
29
N. Daldosso and L. Pavesi, Laser & Photon. Rev., 2009
Rare earth doped light emitting MOS
31
Directional coupler 20
Pump A (0.22, 0.20)
B (0.21, 0.04)
15 Fit A
Laser
n-region V bias 10
23% slope efficiency
5
10
p-region Ring cavity
0 0
-10 0 50 100 150 200
Relative Power (dB)
region
Active
GaAs 350 Å ×5
GaAs 350 Å
GaAs 500 Å
Al0.4Ga0.6As:Si 1.0 µm
GaAs:Si 0.8 µm
GaAs 500 Å
QD buffer ×10
layer
In0.15Ga0.85As
50 Å
GaAs 500 Å
InAs
GaAs buffer 2.0 µm 180 mA
QDs
Si substrate
36
Die bond lasers one at a time (Luxtera)
Ceramic Package
Heterogeneous integration
Step 1: Bond InP-dies on SOI waveguide
InP III-V dies
Si
SOI-waveguide wafer
Step 2: Remove substrate
BCB or SiO2
SiO2
Si
8 lasers
Step 3: Process lasers at wafer scale
SiO2
Waveguide bundel
Si
38
Heterogeneous integration
Two alternatives for the die-to-wafer bonding process
• Adhesive layer bonding • Molecular bonding
– Planarization and bonding in single – InP on SOI-waveguides (UCSB,
step (IMEC-Ghent University) Intel, CEA-LETI, TRACIT) [2,3]
– Ultra-thin bonding layers (sub 200nm
demonstrated) [1]
InP-layer
Si-wire
[1] G. Roelkens et al., “Adhesive Bonding of InP/InGaAsP Dies to Processed Silicon-On-Insulator Wafers using DVS-bis-Benzocyclobutene”, J.
Electrochem. Soc., Volume 153, Issue 12, pp. G1015-G1019 (2006)
[2] D. Liang469. D. Liang, G. Roelkens, R. Baets, J. E. Bowers , "Hybrid Integrated Platforms for Silicon Photonics," Materials , 3 ( 3 ), 1782-1802 ,
March 12 , 2010 39
39
[3] M. Kostrzewa et al., 'InP dies transferred onto silicon substrate for optical interconnects application ', Sensors & Actuators A 125 (2006) 411-414
Hybrid Silicon Photonics
Silicon
A.W. Fang, et al. EEE Photonics Technology Letters , 18 ( 10 ), 1143-1145 , May 15 , 2006
Liang and Bowers, Nature Photonics, 4, 511, Aug. 2010. 40
Scaling of Bonded Wafers
150 mm (6”)
Di Liang
2 cm
These wafers have patterned optical waveguides on SOI with 2 micron GaInAsP
layer on top.
Oxygen plasma enhanced bonding: 300 C, 30 minutes
• D. Liang, G. Roelkens, R. Baets, J. E. Bowers , "Hybrid Integrated Platforms for Silicon Photonics," Materials , 3 ( 3 ),
1782-1802 , March 12 , 2010
Photoluminescence study
Epi Grown by Oakley et al.
Lincoln Labs
PL improves after bonding
Before
bonding
5.4 V 68 nm
After
bonding
5.7 V 40 nm
Wavelength Intensity FWHM
105 C CW 1310 nm laser
Alex Fang
44
QWI DFB Array
Siddharth Jain
Wavelength (nm)
-30
Bandgap Bandgap A
-40 B
Power (dBm)
-50
-60
-70
-80
-90
1240 1260 1280 1300 1320 1340 1360
Wavelength (nm)
ISLC 2010, Kyoto , Japan
WA3 9.15 - 9.30 Integrated Broadband Hybrid Silicon DFB Laser Array using Quantum Well Intermixing
Hybrid Silicon Microring Laser
Di Liang
Confinement: MQW: 5.5%; silicon: 52%
III-V
Si
Silicon waveguide on
SOI wafer
8 AMPs 8 detectors
• Impact
– Electrically pumped amplifiers (unlike Raman or Erbium
amplifiers)
– Wider wavelength range than erbium amps: 1310 nm, S, C,
L band operation
• Issues:
– Minimize reflections at transitions for spectrally flat gain,
and high gain.
48
H. Park et al., PTL, 19(4), 230, February (2007).
Silicon optical modulators
(1) MZ silicon modulator
49
Index Change by Carrier Depletion
Si
Si = Plasma +….
>10x
III-V = Plasma + BF + Pockels + Kerr +….
Δv*device length(mm% )
20 35 Ω
40 Ω
50 Ω
10
-20
• Modulation is done by loaded “T” section
40 50 60 70 80 90 100
– Implant is used to isolate different T sections
Filling factor(% )
• Bragg frequency is around 10 THz
• Velocity match and increase impedance to 50 ohm
• 25 Gbit/s Bandwidth
25 Gb/s
• 40 Gbit/s Modulation
• 500 micron long
Linear (1->4)
Linear (2->4) 23 (35ps) 24 (25ps)
-8 Linear (2->3)
Linear (BtoB)
-9
-10
-11
-30 -29 -28 -27 -26 -25 -24 -23
Received Power (dBm)
• BER test
– Pattern: 231-1 NRZ PRBS at 40Gb/s
– Switch voltage : 2V
– Power penalty : <1.5 dB for all ports configuration
1.E+03
1.E+02
Energy
per
bit
(nJ)
1.E+01
1.E+00
1.E‐01
100,000x lower
1.E‐02
1.E‐03
1.E‐04
1.E‐05
1.E‐06
Inherent
Amplified
32
Tbit/s
Ethernet
Core
PON
ONU IPTV
PIC PIC PIC
Switch Router Server
system
5 circulations
Geza Kurczveil
Buffer
AWG
Digital
Tunable
Laser
Triplexers
Important for fiber to the home (FTTH) receivers
Integrated:
Laser
1310/1500 nm MUX MMI
MZI 1550/1490 MUX
Andy Chang
PDs
1 Peta-
FLOPS OI at ~1 pJ/bit
Computational Throughput
100
TFLOPS
100
GFLOPS
10 MW 1 MW 100 kW 10 kW 1 kW
Power Consumption *D. A. B. Miller, IEEE Proc., 2009
FP, ML, DFB, and DBR lasers Waveguide detector and amplifier
(Fang et al., IEEE PTL, 20, 2008 )
Ring
laser
(Park et al.,IEEE PTL,19 , 2007)
What’s Missing?
MLL
Ultra low threshold lasers (low power is key).
High power optical amplifiers AMP
detector
Short pulse mode locked lasers
DBR
Isolators
Polarizers
Polarization Rotators Microring resonator laser
DFB
(Liang et al.,FA5, postdealine, GFP2009)
Migration to 300 mm substrates: ELO
Hui-Wen Chen | PS 2010 | 60