Laboratory of Photonics and Microwave Engineering, School of
Information and Communications Technology, Royal Institute of
Technology (KTH), SE-164 40 Kista, Sweden
Hewlett-Packard Laboratories, Palo Alto, California 94304, USA
Joint Research Center of Photonics of the Royal Institute of Technology
(KTH) and Zhejiang University, Hangzhou, China
Photonics communications: From global reach to photonic interconnect
networks on multicore architecture chips: The role of low power nanophotonics in data centers
Lars Thylen
1
Acknowledgements
• Lech Wosinski, Petter Holmström, Fei Lou, Alex Bratkovski, Min Yan, Min Qiu....
• Funding
– ADOPT
– Vinnova
– HP
– VR
– ......
3
Outline
• Optical interconnects, overview and rationale
• Integrated photonics: – A 30+ year long story – ”Moore’s law” for integrated photonics
• Integrated low power (nano)photonics fabrics for interconnect
– Basics – Plasmonics with metals & Loss & Gain – New negative e materials? – Near field coupled QDs (FRET) for integrated nanophotonics
– Emerging materials & technologies
• Electro optic polymers • Chalcogenides
• Concluding remarks
3 3
From Rod Tucker
New scenario
• Data processing architectures have followed Moore’s Law for decades, but..
– the performance today is no longer limited by transistor speed (which improves with scaling) but rather by time delays and power dissipation in data transfer between components.
• A case for Optical interconnects (which certainly have been around for a long time…)
6
Optical interconnects • Rapidly increasing application complexity and limited
computing power budgets =>more lightweight cores replacing fewer bulky cores in emerging processor chips.
• Increase in core counts has puts pressure on communication fabrics for supporting more streams of higher bandwidth data transfers.
• => chip power and performance are now dominated not by processor cores but by the need to transport data between processors and to memory.
• => critical that power, bandwidth, and latency of communication scale to meet the needs of processing chips in the near future.
• One solution: Low power integrated nanophotonics for network fabric
8
From Rod Tucker
10
From Rod Tucker
12 Small, power,cheap
?
• The cold climate of Luleå helps cooling the Facebook servers • With all three server halls, the electrical energy consumption will be around 1 TWh/year ( appr .1 GW) • Swedish industry electricity consumption: 55 TWh/year) So: Power input to data centers, >> MW , signal output << MW, difference is basically heat dissipation
Somewhere close to the polar circle in northern Sweden..
IEEE Aug.
2002Spectrum
14
Photonics ICs ... and relation to electronics ICs
16
Vision when the term Integrated optics was coined.
16
S.E. Miller, "Integrated Optics: An Introduction”, Bell Syst Tech J, Vol 48, 2059-2069, Sept 1969.
From Rod Tucker
18
State - of- the Art Photonic IC (U of Eindhoven)
Optical Cross-connect: 100’ish components
State-of the Art Electronic IC (Intel Website)
Pentium 4: 42 M Transistors
Photonics is far behind electronics in maturity
=> Excellent research and business opportunities
New concepts, new materials are required.
After ~30 years of development ....
18
A Moore´s law for integration density in terms of equivalent number of elements per square micron of integrated photonics devices: Growing faster than the IC Moore´s law
L Thylen et al, J. Zhejiang Univ SCIENCE 2006 7(12) p.1961-1964 http://www.zju.edu.cn/jzus/
19
Applications of integrated (nano)photonics
• ICT: from global reach to photonic interconnect networks on multicore architecture chips and perhaps further
• Sensors for nearly everything
– Power grid monitoring
– Photonic sensing of single-cell biomolecule
– .............
• ...
Low power dissipation integrated nanophotonics
• Interesting per se • Nano (small volumes)& efficient materials=>low power * • Optical materials have a central role in enabling
nanophotonic low power switches in two ways: – By enabling an efficient conversion of applied stimulus
(electric field, heat…) into a change of the complex linear or nonlinear refractive index.
– By making possible high confinement of light fields to small volumes
– already treated in T. Tamir, Ed., “Integrated Optics”, in Topics in Applied Physics, Springer, 1975.
* L Thylen, et al, "Nanophotonics for Low-Power Switches", in "Optical Fiber Telecommunications VI", I. P. Kaminow, T. Li, and A. E. Willner, eds., Elsevier Science and Technology Books,Oxford, U.K (2013)
21
2
33
3
00
44.1
rnV
E m
EO
ee
E switch energy, VEO volume of electrooptic medium, m losses per m, optical field confinement to EO medium, r33 electrooptic coefficient, e RF epsilon of EO medium L Thylen, et al, "Nanophotonics for Low-Power Switches", in "Optical Fiber Telecommunications VI", I. P. Kaminow, T. Li, and A. E. Willner, eds., Elsevier Science and Technology Books,Oxford, U.K (2013)
RFeff
eff
ErnN
LkN
33
3
0
Mach Zehnder modulator
Nanophotonics low power devices:
Tentative requirements for switch etc elements
• Size – Longitudinal < or <<1 mm
– Transverse << 104 nm2 (much below diffraction limit)
• Dissipation (Joule) < 1 fJ/bit
• Insertion loss < or << 1 dB/device
• Speed >> 10 GHz
• Transmitted signal power: 100-1000 photons/time slot
• Reasonable temperature properties
• Low cost
• ........
23
Question: Why not electronic interconnect for short distances??
• It is well known that photonics is the technology of choice for broad band long distance communications. But photonics is now also regarded as necessary for the much shorter connections in high performance data centers, ranging from cabinet to cabinet interconnect to eventually intra chip interconnect, i e connection between cores on a chip. An intermediate step is interconnects between chips on a board.
• But why would photonics be better than metal electronic leads for high bandwidth interconnects to connect a chip of order of magnitude 10 cm2 area to other chips on a board (which can be rather large)? The issue is to get very large chip edge bandwidth ( i e total bandwidth from/to chip edge) and large (say >>10) pin count per chip edge.
• See slides 25 and 26 in the presentation, and look up the reference given there.
D. A. B. Miller and H. M. Ozaktas: “Limit to the Bit-Rate Capacity of Electrical Interconnects from the Aspect Ratio of the System Architecture”, JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING 41, 42–52 (1997) ARTICLE NO. PC961285 NOTE: Does not assume advanced modulation formats, which can help to some extent but at the cost of complexity
20l
ABB
A is lead crossectional area, L lead length
Nanophotonics???
But light wavelength is micrometer?
Is 1000 nm nano?
Total field width in microns vs core width in microns for a slab waveguide at a wavelength of 1.55 microns and for various core and cladding refractive indices: from top to bottom ncore = 1.5, ncladding=1.4, as in a glass waveguide, ncore = 3.4, ncladding=3.1 representative for IIIV waveguides and ncore = 3.5, ncladding=1, representing silicon in air.
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
1
2
3
4
Total field width vs core width um , lambda 1,55 um, n1,n2 1.5,1.4 , 3.4,3.1 , 3.5,1
Kramers Krönig
Research related to nanophotonics
• Photonic xtals • Si photonics • III V on Si • EO polymers
But ”real” nanophotonics requries more: • (Hybrid) plasmonics • QDs in various materials and applications
– Near field coupled nanoparticles
• ?? Visions ? Ubiquitously useful technology
And..Plasmonics has been a field of choice for implementing nanophotonics
30 Source: Min Yan , KTH
ISI statistics Topic=(plasmonics OR surface plasmon)
Introduction and background
Photonics
Electronics
Opera
ting S
peed
Critical dimension 1 mm
GHz
10 nm
THz
MHz
100 mm
Plasmonics
Plasmonics: the next chip-scale technology
Fig. Operating speeds vs critical dimensions.
(a)PhC fiber, image courtesy MAX Plank
(b) Silicon photonics, image courtesy IBM
Core Diam.~ 6 µm
Width~550 nm
From Mark L. Brongersma,
Standford
Size mismatch: orders of magnitude
Photonic technology State of the art
Optical interconnects: chip to chip and intra-chip
45 nm
32 nm
22 nm
14 nm
10 nm
2007
2009
2011
2013
2015
Intel Technology Roadmap
Tri-gate, 22 nm litho, 2011
High k metal gate, 45 nm litho, 2007
Electronic Technology State of the art
Image courtesy Intel
Gate width ~ 80 nm
31
Surface plasmon polariton
32 32
Surface plasmon polariton
33
Characteristics: • TM wave • Field enhancement at the interface • Group velocity -> 0 and field confinement -> at resonance
Color: Hy arrows: Ex and Ez
(ω=0.5ωp, λp=1.5μm)
34
Sergey I. Bozhevolnyi et al, Nature Photonics 440, p. 508. (2006)
Deep metallic V-groove waveguide L. Liu, et. al., Opt. Express 13, 6645 (2005)
P. Holmström , et. al., Appl. Phys. Lett. 97(7), 073110 (2010).
Nano-particle chain waveguide
Slot waveguide
Hybrid plasmonic waveguide
Dielectric-loaded plasmonic waveguide
X. Zhang, Nature Photonics 2, 496 - 500 (2008)
PHYSICAL REVIEW B 75, 245405 2007
Plasmonic waveguides: beyond the diffraction limit of light
35
Near field mediated Ag nanoparticle waveguide
eh=1
H
p
rese
21a
pn
em()
eh
d
Longitudinal polarization wave propagation
light line
d=60-80 nm
a=25 nm
35
Wavelength= 350 nm corresponds to wplasma/(31/2)
Model: W.H Weber and G. W. Ford, PRB 70, 125429 (2004) phonon
36
Coupling length vs. nanoparticle array waveguide separation
c= 75 nm
90 nm
100 nm
110 nm
120 nm
130 nm
6.5
0
d
cllc
l0=170 nm
0=371 nm
Nanoparticle array directional coupler: Simulation of lossless silver spheres, diameter 50 nm, separation 75 nm, = 371 nm,
coupling length appr 500nm • Coupling length vs. waveguide separation described by power law
• Exponential law using conventional dielectric waveguides
lc c
Petter Holmstrom, Jun Yuan, Min Qiu, Lars Thylen, and Alexander M. Bratkovsky, "Theoreticalstudy of nanophotonic directional couplers comprising near field-coupled metal nanoparticles", Opt. Express (2011)
36
Theoretical performance of metal nanoarray directional couplers, examples
• Coupling lengths =500 nm for 90 nm array separation
• Staggered configuration: 2.4 nm bandwidth FWHM for a 3
mm long coupler
• A 5 mm long device would require 4×10-3 host index change, reachable with LiNbO3. for high extinction ratio switching (For a conventional coupler: at 1 mm and a 1 mm long device, a phase change requires an effective refractive index change of 0.5)
• Stored RF electric switch energy is on the order of fJ
• BUT: Possible ways of utilizing these characteristics yet to be explored due to very high light propagation losses
37
tW
WQ
/
Q value, basic definition
is angular frequency, W energy
• Plasmonics: At the resonance ( where there is maximum field • confinement) we have a quasistatic condition, • with the H field -> 0 and • At or close to resonance the Q value depends only on the plasmonic medium and is a useful parameter.
metal
metal ddQ
e
e
2
/
*Feng Wang and Y. Ron Shen, “General Properties of Local Plasmons in Metal Nanostructures”, PRL 97, 206806 (2006)
*
Plasmonic materials are LOSSY
• The Q value of a plasmonic material per se is a good gauge for ICT usefulness (e ´´ << magnitude of e ´)
• Subwavelength confinement=> large fraction of energy resides in plasmonics medium=> high losses
• Q value of Ag at RT and 1.5 mm: appr 40, determined by electron scattering rates (order 10s of fs at RT)
• Desirable Qplasmonic material for ICT applications order 1000s • Propagation losses close to resonance in planar dielectric/metal
waveguide order dB/mm or higher ( since group velocity -> 0)
metal
metal ddQ
e
e
2
/
39
Loss happens...
How about gain ??
Note: There are numerous applications of plasmonics where optical losses are not so important , e g sensors and SERS (Surface enhanced Raman scattering)
41
E0
|Ex|/E0
Compensate loss with gain?? Electric fields in InP/ZnS/Ag nanoparticle Ag with loss, QD with
gain
E0
|Ex|/E0
=1.8 eV
=1.833 eV
=2.5
(a)
(b)
2
InP/ZnS
Ag
1 3
2
1
2
3
QD Ag host
(a) (b)
• Frölich resonance strongly broadened by Ag loss => Field enhancement in the QD suppressed, |Ex|/E0~10 @ F
• Field enhancement increases near the QD resonance 0: |Ex|/E0~60
0QD res.
Frölich resonance
F
41 41
P. Holmström, L. Thylen and A. Bratkovsky, APL 97, 073110, 2010
SPASER?? Surface plasmon amplification by stimulated emission of radiation
(b)
42
For plasmonics in ICT the fact seems to be…
• ..that using plasmonics it is possible with currently available materials and understanding to significantly improve the single device footprint, power dissipation and functionality of PICs as compared to Si and III-V
• ..but not to cascade to any significant number of components, especially if device interconnect is done by plasmonics waveguides, all this even if amplification is used*
*L Thylen, P Holmstrom, A Bratkovsky, JJ Li, S Y Wang, "Limits on integration as determined by power dissipation and signal-to-noise ratio in loss-compensated photonic integrated circuits based on metal/quantum-dot materials", IEEE J. Quantum Electron. 46, pp. 518-524, (2010)
42
Nonresonant “hybrid” structures
Jianwei Wang, Xiaowei Guan, Yingran He, Yaocheng Shi, Zhechao Wang, Sailing He, Petter Holmström,
Lech Wosinski, Lars Thylen and Daoxin Dai, ”Sub-μm2 power splitters by using silicon hybrid plasmonic
waveguides”, OPTICS EXPRESS, Vol. 19, No. 2 (2011 ), p 838
43
44
Fig. Intrinsic quality factor as function of radius, Hs, HSiO2.
Hybrid plasmonic microring/disk resonators
Theoretical intrinsic quality factor
1/Qint= 1/Qrad+1/Qabs
Fig. Sketch of hybrid plasmonic microdisk.
Fig. Field distributions at a 4th order resonance.
Fei Lou et al, FMI KTH
Modulator:Nonresonant Mach-Zehnder vs ring resonator
• Requirement for switching in Mach-Zehnder interferometers
• The resonator version is
where Q is the Q value of the resonator. Then:
• Conclusion?
If the potential improvement in device performance with (much) lower loss
negative epsilon materials is so huge...
Why so little effort to research lower loss materials?
Negative epsilon with no loss, according to Mackay et al PRL 99, 189701 (2007)
2 4 6 8 10Angular frequency
40
20
20
40
60
80
Re and Im part of permittivity
Negative epsilon with zero loss from Mackay
e’ and e’’ vs normalized radian frequency •Point of operation is normalized radian frequency “5”, which could correspond to e g 300 THz •Peak e’’ at normalized radian frequency 1 (60 THz)
47
Die
lect
ric
typ
e
RF
me
tal t
ype
Op
to m
eta
l typ
e
?
47
e′
e″
Equivalent circuit from electrical network synthesis Eilert Berglind, Petter Holmstrom, and Lars Thylen, "On the Possibilities to Create a Negative Permittivity Metamaterial with Zero Imaginary Part of the Permittivity at a Specific Frequency:Electrical Network Theory Approach", IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. 48, NO. 4, APRIL 2012
ba
8
222 )(
e
a=0.9
b=5.5
=5
Electrical network synthesis (Re(Y()) Y(s)):
4
4
3
3
2
210
3
3
2
21)(sbsbsbsbb
sasasajsY
)()(00
e
e jj
E
Pj
E
JY P
C0 L1 L3
R
L2
C2
C0=78.6
L0=0.0215
L1=-0.000687
L2=0.000817
C2=49.0
L3=0.00433
R=0.0165
L0
51
22
2, CL
res
Definition of a ”polarization admittance”:
JP
e0E
Re(Y)
Im(Y)
Re(Y)
Re(Y) Re(Y)
Im(Y)
Circuits with Light at Nanoscales: Optical Nanocircuits
Inspired by Metamaterials
Nader Engheta SEPTEMBER 2007 VOL 317 SCIENCE
Other materials for nanophotonics
50
Electrooptic (EO) polymers: finally a disruptive technology?
• Getting RF field mediated refractive index changes much higher than in LiNbO3
51
Larry Dalton et al, U Washington, 2007
RFErnn 33
3)2/1(
r33 (Electrooptic coefficient) for LiNbO3=30 pm/V
52
EOP
SiO2
EX field
=1.55 mm
y [m
m]
250/100/250 nm
200 nm n-Si n+ Si 70 nm
=1.7 cm-1 Lactive=0.1 dB
n+ Si 70 nm
Si-polymer Mach-Zehnder modulator – simulation of one arm in a push-pull configuration
film
r33=500 pm/V Vbias=2 V Switch energy= 16 fJ Lactive=/(4neff)=80 mm (push-pull) f3dB=1/(2RCEOP)=190 GHz (CEOP=4.1 fF)
x [mm]
n-Si
Polymer, n=1.7
1.5 mm 1.5 mm Au Au
53
Calculations by Petter Holmström
Chalcogenides
54
55
Chalcogenides: Material phase change • Material containing one or more chalcogen elements • (e.g. sulphur, selenium or tellurium) as a substantial base constituent. •Electronically, optically or thermally controlled refractive index changes between the amorphous and crystalline state •Example of phase change mediated index change: > .5 •But: Switching times: ms range, ps in nanostructured materials? • Large losses in crystalline state
Optoelectronics Research Centre (ORC) University of Southampton 55
..so
.. short of any break throughs in
lower loss plasmonics materials, any
further options for low power
nanophotonics?
56
Near field coupled QD chains: alternative to plasmonics to get low loss
nanophotonics?
InAs
GaAs
InAs
GaAs
d=10 nm
GaAs
InAs
~6 nm
polymer or ligand molecules
Dipole-dipole interaction energy:
2
, 5
0
3( )( )1
4
nl nl nl
n l
h nl
RV
R
e e
μ μ μ R μ R
, 1,..., ,n l N , , ,x y z
Model adapted from Y. Kubota and K. Nobusada, “Exciton–polariton transmission in quantum dot waveguides and a new transmission path due to thermal relaxation,” J. Chem. Phys. 134, 044108 (2011). Cooperation with the Ohtsu group, Univ of Tokyo
57
• Initial condition: QD 1 excited with one exciton • No inhomogeneous broadening, no phonon interaction, no dephasing • QD spacing d=10 nm
Pulse propagation on QD chain
Transverse Longitudinal
Longitudinal Transverse
58
Gaussian pulse propagation on QD chain
Transverse Longitudinal
FWHM=10 ps
vg=4.5×103 m/s
FWHM=5 ps
vg=9.0×103 m/s
• Initial condition: Gaussian pulse including phase relations of QDs
• No inhomogeneous broadening, no phonon interaction, no dephasing
• QD spacing d=10 nm
Transverse Longitudinal
59
A low power nanophotonics modulator • Modulator structure comprising InAs/GaAs core-shell QDs in organic host material and electrode arrangement for controlling exciton-polariton propagation along QD array. • Electrode separation of the top contact and conducting substrate 20 nm. • COMSOL calculations of electric fields
QD chain modulator: predicted
performance • Switching by detuning QDs
• Example: Dissipated switch energy Esw =2×10-19 J=.2 aJ=1 eV
• Speed example: T=77 K, time slot 10 ps, 100 photons=>1 mW (100 parallel strands required, with 100x increase in switch energy= 20 aJ)
• Comparison to electronics: – 22 nm (Intel state of the art) feature size gives sub-ps gate delay
time and ~20 aJ dissipated switch energy (CMOS very fast per se, but slowed down in circuits by interconnects)
• Thus: QD array devices comparable in footprint and switch energy, but not in speed for single arrays
Petter Holmström and Lars Thylén, “Electro-optic switch based on near-field-coupled quantum dots”, submitted to Optics Express
61
Issues
• In and out coupling efficiencies (nanoantenna, nanofountain..)
• Phonon interaction
• Propagation directionality and temperature properties (other mechanisms than coherent transport are possible)
• (Size and position dispersion of QDs (DNA positioning possible))
• Nanoelectrodes
• …
K(京)-Computer and its High Performance Architecture
System Spec:
8 cores / node
~ 83,000 nodes (~ 664,000 cores)
Has already achieved full spec (664,000 cores) parallelization.
One of the world’s fastest machines
1016=
10PFlops
TDDFT-Maxwell Coupled Equation
(RIKEN, Kobe, Japan)
Unified first-principles calculations of near-field excitation dynamics in nanostructures
Katsuyuki Nobusada Institute for Molecular Science
Photonics for everything??
Electronics for RAM type memory and digital logic operations
Photonics for Data communication (transmission,
routing...)
My conclusion: Never compete with electronics for the functions listed above
Fermions vs bosons
64
All optical vs electrooptical technology for low power and footprint integrated photonics
switches ?
• Elastic, essentially lossless parametric processes – Pockels
– Kerr
• Carrier generation (plasma effect) mediated index changes – PN junctions
– Optical absorption
L Thylen, “A comparison of optically and electronically controlled optical switches”, Applied Physics A, invited, DOI 10.1007/s00339-013-7914-x, (2013)
65
Pc, pulsetrain
nc
Ppulsetrain
ns
nc
ns
WDM WDM
PRF,c
PCW Ps
Capacitor
c
cSW
Pn
LPW
cRF
cRFSW
Pn
LPW
,
,
L
L
All optical
Electrooptical
Kerr & Pockels effect
66
nc
nc
Pc
Pc
Pcw or Ppulsetrain
Pcw or Ppulsetrain
ns
ns
Ps
Pcw
ns
Pc
ns
Ps
(a)
(b)
Electrical RF signal Optical power: CW, pulse train or signal
OO(a) and EO (b) switching circuit diagrams, based on the control information being electronic (Pc) . Abbreviations: OO: optically controlled optical switch, EO: electronically controlled optical switch, c: control, s: signal , RF: radio frequency, n: optical frequency, P: optical or RF power , CW: continuous wave, pulse train: Periodic sequence of short pulses (return to zero, RZ format)
Single switches All information in the electronic domain
67
The role (rule?) of economics in electroncis & photonics
Moore’s famed law is somehow an economics one...
– Power input to data centers, >> MW , signal output << MW, difference is basically heat dissipation
– But: Factor of >10-100 improvement possible in electronics device power before hitting any thermodynamic limits
– Large fraction of power dissipated in electronics interconnects, photonics is envisaged to mitigate this
– but still development in interrelated issues of photonics footprint and power dissipation required
Better to burn power using existing technology than to develop new?
68
MIT Local mesh to global switch
Example of architecture of chips in data centers
69
− Point-to-point plus passive mesh
− DRAM-centric architecture
− 20W power envelope (network)
− 10x improvement over electrical
• Each ring in main fig => 16 double rings modulating or filtering 16 different wavelengths
• Each optical power wavegudie=> 16 waveguides
Device
and (mm)
V L
(V mm)
V
(V)
L
(mm)
IL
dB
(dB/mm)
Confine-
ment
Capaci-
tance (fF)
(fJ/bit)
Comments
P Layered
metal/chalcogenide
waveguide 1.55
0.66 0.33 2 7
(3.5)
0.01 μm2 .01
(0.003)
Chalcogenide thickness
4nm,
index change 0.1
P Array of Ag
nanoparticles in
EOP matrix
0.680
3 15 0.2 2.4
(12)
Appr 0.01
μm2
(Very approxi-
mate)
0.01
(2)
200 nm electrode
separation. Very rough
approx, probably much
better. Trading lower
voltage for length impeded
by loss
P Slotline
Si/EOP/Si
1.55
160 2 80 0.1
(0.001)
Appr .3x.7
μm2
4
(16)
Doped Si serves as
electrodes. 100 nm EOP
A III V
Electroabsor-ption
QCSE
1.55
400 2
(Vpp)
200
active
500
total
3-5 4 μm2 200 Travelling wave type EAM,
50 Ohm transmission line
>100 GHz bw
Experiment
-Min Yan, Lars Thylen, and Min Qiu, Opt. Express 19, 3818, (2011).
-Petter Holmstrom, et al, Opt. Express (2011)
-P Holmström, L Thylen, unpublished
-M. Chacinski, U. Westergren, B. Stoltz, L. Thylén,“Monolithically Integrated DFB-EA for 100 Gb/s Ethernet”, IEEE Electron Device Letters,
vol29, (2008)
Comparison modulators ( A: amplitude, P: phase), Electrooptic polymer (EOP): 500pm/V or Chalcogenide with index
change = 0.1; V L for phase shift or > 10 dB extinction ratio. V and L as examples. THEORY except where otherwise stated
70
With existing plasmonic materials unpublished
Challenges in photonics • Integrating III V, Si, EO polymers, metals and other
materials in “electronics like” generic foundries: – III V for light generation, modulation and detection +… – Si(GeSn) as a waveguide and device platform: (tunable)
WDM, filtering, modulation, switching, detection… +.. – EO polymers for all high speed high performance optical
phase change functions?
• Lower loss plasmonic materials! SPASERs • Chalcogenides? Graphene?... • Footprint and power dissipation approaching
electronics: – Förster (FRET) near field coupled QD arrays? – …
• Monolithic integration with electronics! 71
A Moore´s law for integration density in terms of equivalent number of elements per square micron of integrated photonics devices: Growing faster than the IC Moore´s law
L Thylen et al, J. Zhejiang Univ SCIENCE 2006 7(12) p.1961-1964 http://www.zju.edu.cn/jzus/
72
?
Next quantum
Mike Tan, HP labs