Transcript
Silicon Photonics Opportunity, Applicatoins & Recent Results
Mario Paniccia, Director Photonics Technology Lab Intel Corporation
Agenda Opportunity
for Silicon Photonics Copper vs optical Recent advances Intels SP Research Recent results – Intel’s Silicon Laser Summary
ELECTRONICS: Moore’s Law Scaling MIPS
Pentium ® 4 Processor Pentium ® III Processor Pentium ® II Processor
10000
1000
$/MIPS 100
10
Pentium ® Pro Processor
100
10
1
Pentium ® Processor Intel486TM DX CPU Intel386TM DX Microprocessor Microprocessor
0.1 MIPS $/MIPS
1
0.01 1985
1987
1989
1991
1993
1995
1997
1999
2001
Integration & increased functionality Volume economics – faster, better, cheaper
The Opportunity of Silicon Photonics
Take advantage of enormous ($ billions) CMOS infrastructure, process learning, and capacity – Available tools: litho requirements typically >90nm – Draft continued investment going forward
Potential to integrate multiple optical devices
Micromachini ng could provide smart packaging Micromachining
Potential to converge computing & communic communications ations Industry standard silicon manufacturing processes could enable optical. enableintegration, integration,bring bring“volume economics”totooptical.
To benefit from existing infrastructure optical wafers must run alongside product.. i.e CMOS fabrication compatible..
Today's High Speed Interconnects Primarily Primarily Optical Copper
Chip to Chip 1 – 50 cm
Metro & Long Haul
Billions
0.1 – 80 km 0.1 –
Board to Board 50 – 100 cm 50 –
Rack to Rack
V o l u m Millions e s
1 to 100 m
Thousands Decreasing Distances
Need to drive volume economics to drive optical closer to chip
Copper Approaching Limits Simulation of 20” channel transmitter w/ equalization 0
] B d l [ e n o n i n t a a h u C n e t t A
18G
-10
Red Zone = Eye Closes
-20 -30
12G
-40 -50
Low Loss Ro4350 Standard FR4
0
10
20
30
40
Data Rate [Gb/s]
Copper scaling more challenging. Headroom getting squeezed. Howard Heck
Electrical to Optical Enterprise Distance: 0.1-10km
10G
Silicon Photonics?
>= 40G
OPTICAL
Rack-Rack Distance: 1-100m
3.125G
10G
40G Optical Tech
Board-Board Distance: 50-100cm
3.125G
5-6G
Chip-Chip
ELECTRICAL
Distance: 1-50cm
3.125G
5-6G
2005
10G Copper Tech
20G
T r a an s i t ti i o on Z o n e
10G
15-20G
2010+
Transition driven by cost
The Photonic Dilemma Fiber
has much more bandwidth than copper
However,
it is much more expensive…..
Photonics: The technology of emission, transmission, control and detection of light (photons) aka fiber - optics & opto- electronics electronics Today: Most photonic devices made with exotic materials, expensive processing, complex packaging Silicon Photonics Vision: Research effort to develop photonic devices devices using using silicon as base material material and do this using standard, high volume silicon manufacturing techniques in existing fabs Benefit: Bring volume economics to optical communications communications
Agenda Opportunity
for Silicon Photonics Copper vs optical Recent advances Intels SP Research Recent results – Intel’s Silicon Laser** Summary
Silicon Pro’s and Cons + Transparent in 1.3-1.6 µm region + CMOS fabrication compatibility + Low cost + High-index contrast – small footprint − −
No electro-optic effect
− −
No detection in 1.3-1.6 µm region
− −
High index contrast – coupling
− −
Lacks efficient light emission
Silicon will not win with passive devices.. Must produce active devices that add functionality
Silicon Photonics Breakthroughs Are Accelerating
Raman Net Pulsed Gain 9/6: Intel 9/20: Cornell 9/29: UCLA 9/29: CUHK
SRS UCLA
Si LEDs STM, Trento
Integrated APD+TIA UT
Low Loss Strip
Inverted Taper NTT, Cornel
MIT
2001
2002
Raman
Conversion UCLA
30GHz SiGe Photodetector IBM
Modeled GHz PIN Modulator
GHz MOS Modulator Intel
Surrey, Naples
PBG WG <25dB/cm
PBG WG <7dB/cm
IBM
IBM, FESTA, NTT
2003
2004
Progress In Recent Years Is Accelerating still not there…
CW Raman lasing Feb 05
Agenda Opportunity
for Silicon Photonics Copper vs optical Recent advances Intel’s SP Research Recent results – Intel’s Silicon Laser** Summary
Intel’s Silicon Photonics Research 1. Develop photonic building blocks in silicon
1) Light Source
2) Guide Light
3) Modulation
Waveguides devices
First Continuous Silicon Laser
4) PhotoPhoto-detection
5) Low Cost Assembly Passive Align
6) Intelligence CMOS
1GHz (Nature ‘ 04) 04) 4 Gb/s ( ‘ ‘05) 05)
(Nature 2/17/05)
Mirror
SiGe Photodetectors
First Prove that silicon is viable material for photonics
Packaging Approximate Optical Product Cost Breakdown
Packaging 1/3 Device 1/3 Testing 1/3 In addition to device costs, packaging and testing costs must drop with to enable high volume photonics
Micromachining for Packaging Use standard pick and place technologies along with litho defined silicon micro -machining
U-Grooves
Tapers
Mirror
V-Grooves
Laser Attach
45° Mirrors
Facet Preparation
Intel’s Silicon Photonics Research 1. Develop photonic building blocks in silicon 2. Integrate increasing functionality directly onto silicon Integrated in Silicon Photodetectors
DEMUX Taper
Receiver Chip
Passive Align
Driver Chip
Lasers
MUX
Intel’s Silicon Photonics Research 1. Develop photonic building blocks in silicon 2. Integrate increasing functionality directly onto silicon 3. Long term explore monolithic integration ECL Modulator Filter
Multiple Channels
Drivers
CMOS Circuitry
TIA
TIA
Photodetector
Passive Alignment
SILICON LASER What we announced on Feb 17th
The First Laser Developed by Ted Maiman, published in Nature, August 6, 1960. this ruby laser used a flash lamp as an optical pump Fully Reflective Mirror
Flash Lamp
Partially Reflective Mirror
LASER BEAM RUBY CRYSTAL ROD
Raman:
(Historical Note)
Raman Effect or Raman Scattering: A phenomenon observed in the the scattering of light a it passes through a transparent medium; the light undergoes a change in frequency and random alteration alteration in phase due to a change in rotational or vibrational energy of the scattering molecules.
• Disc Discover overed ed a mate material rial effe effect ct that that is name named d after after him him •Nature published his paper on the effect on March 31, 1928 •He received the Nobel prize in 1930 for his discovery
• The first first lase laserr using using the the Raman Raman effect effect was built built in 1962 1962 • Today Raman based amplifiers are used throughout telecom • Most long distance phone calls will go through a Raman amplifier
Typical Raman Amplifier
The Raman Effect Materials
Raman gain coefficient (10 -8 m/MW)
Silicon Indium Antimonide (III-V) Quartz Lithium Niobate (used for modulators) Diamond Glass Fiber (Raman lasers/amps)
0
Kilometers of fiber
...
Centimeters of silicon
5000
10000
15000
20000
The Raman effect is 10,000 times stronger in silicon than in glass fiber This allows for significant gain in centimeters instead of kilometers
Fabrication of low-loss silicon waveguides is challenging
Raman Gain in Silicon Silicon Waveguide Pump in
Pump out
Probe in
Pump/probe experiment
Probe out
Raman Gain and WG loss vs. Input Pump Power 2.5 2.0
) B d ( 1.5 n i a g n 1.0 a m a R0.5 0.0 200
400
600
Input pump power(mW)
800
-0.2
3
-0.4
s s o L 2.5 G W / ) 2 n B i a d ( G 1.5 n a m 1 a R 0.5
(b)
0
3.5
1000
-0.6 ) B d ( -0.8 s s o L -1 n i a -1.2 G
Raman Gain WG Loss Loss w/o Pump Gain-Loss
-1.4
0
-1.6 0
200 400 Pump Power (mW)
600
CW Gain Saturation due to TPA induced FCA
Two Photon Absorption In silicon, one infrared photon doesn't have the energy to free an electron
e e
e
e e
Free
e
Electron
e
e
e SILICON WAVEGUIDE
But, occasionally, two photons can knock an electron out of orbit. Free electrons absorb individual photons and cancel Raman gain
Overcoming TPA induced FCA −
V
+
laser beam p-type silicon
n i a G n a m a R
electrons
Gain needed to make a laser oxide
intrinsic silicon
Gain limit due to Two Photon Absorption problem
Pump power
n-type silicon
SiO2 passivation Al contact
p-region
Lifetime=16 ns Lifetime=6.8 ns Lifetime=3.2 ns Lifetime=1 ns
500
Si rib wav wavegu eguide ide
H
W
Al contact
h
Buried oxide Si su subs bstr trat ate e
n-region
) 400 W m ( r 300 e w o p t 200 u p t u O100
25 V 5V short open
0 0
200
400
600
800 1000 1200
Input power (mW)
PIN Cross-section TPA coeff ~ 0.5 cm/GW cm/GW,, 0.39 dB/cm dB/cm,, FCA cross sect 1.45e-17 1.45e-17 cm^2 cm^2 @ 1550 nm. nm. The lifetime is used as a fitting parameter
CW gain vs. reverse bias voltage WG= ~1.5um by 1.5um
NET GAIN
NO NET GAIN
Pump =1550 nm
Signal =1686 nm
With gain can build Laser: Silicon Waveguide Cavity 16 mm Rf
V bias
n-region
Rb
m m 2
SOI rib waveguide
Pump beam Laser output Dichroic coating
24%/71%
p-region
Broad-band reflective coating
90%
Experimental setup
Pump power monitor Polarization controller
Pump at 1,550 nm
Lensed fibre
Silicon waveguide
De-multiplexer
90/10 Tap coupler LP filter
0
Optical spectrum analyzer
-10 -20 -30 -40 -50 -60 -70 -80 1684
1685
1686
1687
1688
1689
1690
1691
1692
90/10 Laser output power meter
Tap coupler
Laser output at Dichroic 1,686 nm coating
High reflection coating
Experimental Set up Test chip with 8 laser WG’s
Laser chip
Typical Lasing Criteria •Threshold behavior: rapid growth in output power when gain > loss •Spec •S pectr tral al lilinew newid idth th na narro rrowi wing ng:: Coherent light emission
Threshold, Efficiency, and PIN effect 10.0 9.0
25V bia b ias s
8.0
5V bia b ias s
) W 7.0 m ( t 6.0 u p 5.0 t u o r 4.0 e s 3.0 a L
25V slope 5V slope
2.0 1.0 0.0 0
200
400
600
800
Coupled pump pu mp power (mW) (mW)
Laser turns on at threshold, when gain per pass in cavity becomes greater than the loss.
Spontaneous emission vs. laser spectrum 2.50 2.50
) 2.00 ) . . 2.00 u u . . a a ( ( r r 1.50 e e1.50 w w o o p p l l 1.00 a a r r 1.00 t t c c e e p p S S0.50 0.50 0.00 0.00 11666688.5 .5
Lasing Lasingsignal signal Spontaneous Spontaneous emmission emmission
Magnified 10^ 5x
11666699
11666699.5 .5
11667700
11667700.5 .5
Wavelength Wavelength (nm) (nm)
When lasing, the spectrum becomes much more narrow and much higher in power.
Wavelength tuning (comparison)
0
pump
-10
0 -10
1548 nm
-60
) -20 B d ( r -30 e w o p -40 l a r t c -50 a p S
-70
-70
-80
-80
) -20 B d ( r -30 e w o p -40 l a r t c -50 a p S
1680
1550 nm 1552 nm 1554 nm 1556 nm 1558 nm
1548 nm 1550 nm 1552 nm 1554 nm 1556 nm 1558 nm
-60
1685
1690
1695
Laser wavelength (nm)
Silicon Raman laser
1700
1542
1547
1552
1557
Laser wavelength (nm)
Commercial ECDL
1562
Potential Applications
Communications Applications PUMP LASER
Si Raman Amplifier
passively aligned waveguide coupler
weak data beam 101110
amplified data beam
101110
Si Multi-Channel Transmitter
silicon waveguide (cm’s)
laser cavity
P
modulators
passively aligned
MOD
MUX
PUMP LASER
MOD
MOD
N splitter
Si Raman Modulator integrated mirrors
MOD
Optical Fiber
Covering the Gaps • Different wavelengths require different types of lasers • Mid-Infrared very difficult for compact semiconductors • Raman Lasers could enable lasers at these wavelengths • Applications in sensing, analysis, medicine, and others Compact Semi. Lasers
2.1µm Ho:YAG laser
PUMP LASER
>2 µ m
2.9µm Er:YAG laser cascaded mirrors
Could enable lasers for a variety of applications
Summary Long term true convergence opportunities are with silicon B/W will continue drive conversion of optical into interconnects Tremendous progress from research community Need
to continue pushing & improving performance Research breakthrough with CW silicon laser Integration is next set of challenges In order to benefit Technologies must be CMOS fabrication compatible to benefit from HVM & infrastructure Silicon will not win with individual devices, but with integrate d modules that bring increased total functionality & intelligence at a lower cost
BACKUP
Benefits of Integration Photonic
Integration:
Reduction
in interfaces – lower loss
Reduction
in size
Simpler assembly, testing, packaging Cost
Optoelectronic Integration: Reduce
parasitics, improved high-freq performance
Further
size, testing, packaging reductions
? Cost Integration is only useful if integrated device has benefit (functionality,, cost, performance) over discrete devices (functionality
CMOS Integration Challenges – Film topology – Coupling to fiber – Contaminating the fab – Yield metrology – Thermal budgets – Heat dissipation – Complexity / yield
Optoelectronic Integration
To benefit from existing infrastructure optical wafers must run alongside product, introducing additional pragmatic challenges
Surface Topology: Litho vs DOF • Depth of focus (DOF) shrinks as litho improves • Many optical devices are much taller than transistors
For 0.18µm and better, topology exceeds DOF New planarization techniques required for advanced litho DOF vs. Litho Technology ( µm)
0.25 0.5 µm
0.18 0.35µm
0.09
Transistor on 90nm
0.9µm Rib
0.2µm 0.2µ 0.3µm Strip 0.1µm gate
8µm Taper
Fiber Coupling Taper from (W x H): 10 x 8 µm to 2.5 x 2.3 µm Assume zero roughness roughness Tip=0.5 Tip=1.0 Tip=2.0 10
) B d ( s s o l r 1 e p a T
• Coupling from standard fiber
to Si waveguides requires special structures (tapers, gratings, etc). 2dB
1dB
• For wedge tapers, etch angle as well
as the tip lithography impact loss.
0.1 80
82
84
86
88
Sidewall angle (degrees)
90
• Sidewall roughness is also a factor
Source: Intel
Getting light from fibers into silicon waveguides will require couplers. For certain structures litho and etch parameters must be carefully controlled.
Yield Metrology • CMOS fabs monitor thousands of parameters across wafer in line • Tight control – e.g. CMOS gate width held to 10’s of angstroms • Significant per - -wafer w afer cost savings from screening out yield early
• In-line wafer level optical probing is very immature • Most optical device testing is performed after wafer dicing
To truly gain from HVM processing, automated & non-destructive techniques for probing optical devices at the wafer level must be developed
Opto-Electronic Integration (cont) Thermal:
Simulated multimulti-core thermal map
For optoelectronic integration , optical devices must tolerate heat generated by CMOS circuits.
IO Pads Core Core
Other Logic
Process compatibili compatibility: ty: @ 10Gb/s CMOS IC’s need 90nm technology Silicon Photonic devices may only need ~.25um
Temp °C
Cache
Core Core IO Pads
Yield: Typical industry IC yields are high, but the process windows are extremely tight. Tweaks to enable opto-electronic opto-electronic integration may effect effect IC yield yield
Trade off off of yield and and process compexi compexity ty will determi determine ne if opto-electrical integration valuable
80-85 75-80 70-75 65-70 60-65
Animation Click in box while in slide show mode to start
Extending and Expanding Moore’s Law Sensors
Mechanical
Discrete
SSI
E X P Biological A EXTENDING D Fluidics I N G
LSI VLSI
Wireless
Optical
Two Photon Absorption in Silicon Conduction band
Silicon band gap 1.1 eV
Pump λ=1.55µm
Valence band
Two photons can simultaneously hit an atom Combined energy enough to kick free an electron
