2012 Peking University Summer School on Si Photonics ...spm.pku.edu.cn/summerschool file/lecture/Peking SC_WadaS.pdf · 2012 Peking University Summer School on ... You should always

Kazumi WadaUniversity of Tokyo

2012 Peking University Summer School on Si Photonics Technology and Applications

June 29, 2012

[email protected]

Three Myths in Ge-Si Photonics-锗硅有源器件单片集成的进展

Lecture Notes

A Lecturer has designated this material as copyright-protected.

•Material protected by copyright has restrictions on usage and reproduction, which may be subject to Fair Use exceptions for nonprofit, educational purposes under limited circumstances.

•You should acknowledge that this material may be protected by copyright law and its usage may be restricted.

2

3

About me

1975

光通信 IT

1998 2004

エレクトロニクス, E

フォトニクス, P

E+P

2010

Electronics

Electronics

E + P

Electronics

Optical Fiber Comm. IT

You should always ask questions, the bigger the better. If you ask big questions, you get big answers.

Francis Crick8 June 1916-28 July 2004

Outline

• Introduction

• High performance computing and communication

• Problems and Solutions

• Three myths in Ge research in Si photonics

• Epitaxy, detection limit, and lasing

• Summary

5

World Economy

6

7

Important is to understand where the bottlenecks are

END USERS

CONTENT & APPLICS

APPLI-ANCES

SERVICE PROVIDERS

NETWORKOWNERS

EQUIPMENTMAKERS

COMP-ONENTS

MATERIALS &PROCESS EQUIP

SemiconductorsMetals InsulatorsMagnetic mat.PolymerSteppersEtchersMEMSInsertionetc

LasersAmplifiersTransceiverFiltersProcessorsMemoriesFibersASICSMEMESDSPsMicroprocessorsDisplaysetc

RoutersSwitchesHubsBase StationsSatellitesServersSoftwareOSComputersetc

WirelessBackboneMetroAccessSubstationsSatellitesBroadcast SpectrumCommunication Spectrumetc

Long distanceLocal phoneCellularISPBroadcastHot Spots Cable TVSatellite TVetc

MusicMoviesEmailVoIPPOTSShoppingBusinessetc

ComputersPhonesMedia PlayersCamerasPDAsWeaponsetc

BusinessConsumerGovernmentMilitaryEducationMedicaletc

Bruce and Fine @Sloan “02

IT Value Chain

8

Moore’s Law

4004400480808080

80868086

80088008

PentiumPentium®® Processor Processor

486486™™ DX Processor DX Processor386386™™ Processor Processor

286286

PentiumPentium®® II Processor II Processor

PentiumPentium®® III Processor III Processor PentiumPentium®® 4 4ProcessorProcessor

ItaniumItanium®® 2 Processor 2 Processor

1,000

10,000

100,000

1,000,000

10,000,000

100,000,000

1,000,000,000

1970 1980 1990 2000 2010

Microprocessor transistor countMicroprocessor transistor count

Moore’s Law

9Quarks & Co

Minimum Structure Sizes in Memory Devices

> 2020 ??

10

Where we are?

Photons meet Si-LSIs Data processing by electrons and transmission by photons.

Electronics and Photonics Convergence

Large (‘70s) Disktop (‘80s) Note (‘90s)

Cell (‘00s)Keyholder (‘10s)Quark

History of Computer

11

Development of Supercomputers

• 1000x in 10 years, ~16 yrs from top to note.

6-8 yrs 8-10 yrs

500th

Top

Notebook

From ECOC 2008 tutorial by U. Vlasov

http://www.top500.org/

12

• However, everything has limitation in human history.

13

CPU Performance Evolution

1970 1980 1990 2000 2010

105

104

103

102

10

1

0.1

Clo

cksp

eed

(MH

z)

Year

DA. Muller, Nature Mat. 4, 645, 2005

14

Saturation

Speed of Aircrafts

15

1900 1925 1950 1975 2000

Year

CommercialMilitary

1,000

100

10

Spe

ed (

m.p

.h.)

D.A. Muller

How to Stop “Saturation”

• Many core architecture and parallelism.

16

• The enhancement is limited by a fraction must have serially performed.

Amdahl’s Law

17€

Overall speedup =1

F +1− FN

,

where F - fraction of a program to serially perform,           N - number of cores

G.M. Amdahl, In AFIPS Conference Proceedings vol. 30 AFIPS Press, Reston, Va., 1967, pp. 483.

5%

10%

20%

50%

F=0%

18

10-2

100

102

104

106

108

1010

1012

1014

1880 1900 1920 1940 1960 1980 2000 2020 2040

Re

lativ

e In

form

atio

n C

ap

ac

ity (

bit/

s)

Year

Telephone lines first constructed

Carrier Telephony first used 12 voicechannels on one wire pair

Early coaxial cable links

Advanced coaxial and

microwave systems

CommunicationSatellites

Single channel (ETDM)

Multi-channel(WDM)

OPTICAL FIBER SYSTEMS

Photons to the rescueLC. Kimerling, ECS Interface, p.28, Summer, 2000.

Short Summary

• HPC gets faster exponentially. • Top to notebook seems ~16 years.

• Optics is “MUST HAVE” on Si to suppress “saturation”. • On-chip DWDM should be the clue.

• The key is Light emitter (EMT), modulator(MOD), detector(DET), multiplexor/demultiplexor(MUX/DEMUX) on a chip.

19

20

History of Device Development

ProcessSi-LSIs

MaterialsMagnetic disk

PrincipleOptical disk

21

Innovation by Materials

1956 IBM Ramac 305 vs. 2000 IBM Microdrive5 MB 1 GB50 x 24“ dia. disks 1 x 1”“ diskweighs “a ton“ < 1 oz.$50,000 $500

22

Innovation by Processing

INTEL

23

Si Crystal

• Si• Dislocation-free• 400mm

• made by Super Silicon research Lab (2000).

24

Innovation by Principle

Auto Piano　　　Optical Disk

Innovation in Si Photonics

25

Si Electronics Si Photonics

Materials Si CMOSSi CMOS + Bonding

Process Si CMOS Si CMOS

Principle ElectronicsElectronics + Photonics

Electronics + Photonics

• Current tartget

• Signal processing by electronics (transistors)

• Signal transmission by photonics

• Future

• Both by photonics

26

27

Optical Fiber Comm.

10-2

100

102

104

106

108

1010

1012

1014

1880 1900 1920 1940 1960 1980 2000 2020 2040

Relat

ive In

form

ation

Cap

acity

(bit/s

)

Year

Telephone lines first constructed

Carrier Telephony first used 12 voicechannels on one wire pair

Early coaxial cable links

Advanced coaxial and

microwave systems

CommunicationSatellites

Single channel (ETDM)

Multi-channel(WDM)

OPTICAL FIBER SYSTEMS

Rel

ativ

e In

form

atio

n C

apac

ity (b

it/s)

Light is f

ast and has colors

Optical Fiber and Laser

28

29

砂浜

海

海岸線

君

友

遅い

速い

高屈折率(遅い)

低屈折率

Optical Fiber Fab. (VAD)

30

31

Optical Fiber and III-V Lasers

C Band:1535-1565nm

L Band:1565-1620nm

32

Photon Energy (eV) → 2.76 1.55 1.1eV 0.41Wavelength (µm) → 0.45 0.8 1.12µm 3.0

Si Bandgap

• Weak EO coeficient• transparent 1.3-1.6 μm• Cost-effective • Weak light emission• CMOS Compatibility • Weak detection <1.1 μm

Communications

Si as a Photonic Material

33

World’s First E + P Convergence

R=800um

850nm

Core: (1.5µm)2

PECVD SiOxNyCladding:SiO2 Δ n=0.05Collaboration with Intel 2000

34

E + P Convergence

• Short distance by electrons

• Long distance by photons

optical clocking

local electricaldistributionH-Tree

35

Si Waveguide

• Core:Si

• 0.2 x 0.5 µm2.

• Cladding: SiO2 (or air)

Fabricated at MIT Lincoln Labs

Si

36

World’s Smallest Ring

1 2 3 4

8 μm

In Li,m 2000Thru port

1x4 WDM in Silicon

Ge-based Devices

Devices Materials

Waveguide Si, SiON, SiOx

Modulator Si(ER), Ge(EA)

Photodetector Ge, Si

Light emitter III-V on Si (bonding), Ge

MUX/DEMUX Si, SiON, SIOx

Fiber coupler Si, SiON, SiOx

Ge as an active photonic material b/c ofhigh quality and strain.

Intermission

38

Three Myths in Ge Research

1.Epitaxy needs lattice matching.

2.Ge only works below 1550 nm to detect.

3.Laser needs direct transition type semiconductors.

39

First Myth

1.Epitaxy needs lattice matching.

• High quality epitaxial layers are only available when the lattice constant matches with that of its substrate.

40

41

Lattice Matching in Epitaxy

Lattice constant (A)

Lattice Mismatch: Growth

Ge flat epilayers A two-step (low-high temp.) growth process.

300˚C

550˚C

Islandingsdue to long migration to get energy minima.

42

Lattice Mismatch: Defects

• Dislocations: • 4%-mismatch generates

1 misfit dislocation every 25 Si atoms.

Ge

Si

SiO

2

SiO

2

Dislo

catio

n

10μmGe SiO2

1 cycle

10 cycles

Dislocation-free Ge on SiLuan

43

44

Lattice Matching in Epitaxy

Lattice constant (A)

GaAs on Ge on Si

45

Si

Ge

GaAs

• This opens up new fields; • Integrated light emitters, cost-

effective GaAs solarcells, and more.

• XTEM• Dislocation-free GaAs on Ge

• Off-cut (100) Si substrate used.

Second Myth

• Ge works below 1550 nm to detect. • The typical communication wavelength

range is 1530-1620 nm (C+L band) because of Er doped fiber amp.

46

Absorption Spectra

47

• Direct bandgap of bulk Ge is 0.8 eV (1550nm).

• Absorption is weak when light is beyond 1550 nm in wavelength.

Extended Absorption of Ge Epi

• Red shift of Ge absorption

• Beneficial for detection of C+L band.

48

101

102

103

104

1300 1400 1500 1600 1700 1800

Wavelength (nm)

Ab

sorp

tion

Co

eff

icie

nt

(cm

-1)

The presentBulk

L-ba

nd

C-b

and

30 meV

49

Thermal expansion coefficientGe > Si

Ishikawa et al. (MIT), APL 82, 2044 (2003); JAP 93, 13501 (2005).

Strain by Thermal “Shrinkage” Mismatch

0.2-0.3% strain-tensile

50

Direct Bandgap under Strain

0.2%30 meV

• Ge direct gap shrinks under biaxial tensile strain.

• At 0.2% tensile, that shrinks by 30 meV. • 0.80 eV (1550 nm) to

0.77 eV (1610 nm)

• Thus, Ge on Si covers ~C+L band.

Third Myth

• Laser needs direct transition type semiconductors. • Indirect semiconductors never lase.

51

52

Ge: indirect-gap Semiconductor

0.80 eV0.66 eV

(M. L. Cohen)

• Direct gap is 0.80 eV while Indirect gap 0.66eV.

• Optical transition: needs phonons to conserve k.

• Thus, it never lases.

Three Strategies for Ge Lasing

• The Γ valley to populate electrons via, 1.heavily n doping

2.highly strain-tensile Ge

3.high carrier injection

• No scattering of electrons to the L valley.

• Quasi-direct gap.

53

Monolithic; n+Ge Laser

54

Jifeng Liu, J. Michel, et al., Opt. Lett. 35, 5 (2010).

Ge lasing reported mid September 2009. !

• n+Ge Lasing• Optical pumping

• Rib structure• Multimode

Short Summary

Three myths and the breakthroughs1.Epitaxy needs lattice-matching.

• Dislocation-free Ge on Si using sheer stress.

2.Ge cannot detect wavelength beyond 1550 nm that InGaAs can. • Strain-tensile Ge to detect 1610nm.

3. Indirect semiconductors never lase. • Quasi-direct Ge via n-type and strain-tensile to

suppress electron scattering to the L valley.

55

What we have learned

• There are always Myths in research and the ways to breakthrough (Not always).

• Shortcut is to ask questions.

• The bigger the better. You will get big answers (F. Click).

56

On-going Challenges

• WDM to further enhance signal transmission. • Increase the bit rate by 1000.

• In 1260 nm (O-band) to 1675 nm (U-band) : Δλ~400nm, 1000 signals available.

• On-chip Injection LD should be the major challenge.

• Wavelength locking of photonics on a chip (Fourth Myth)

57

carrier injection�

Carrier density from n-type doping (cm-3)�

Net

gai

n (c

m-1

)�

Toward Ge Injection LDs

• Heavily doping• n type carriers ~1020

cm-3

• Here, 0.25% tensile strain assumed.

• Challenge: such high n+ doping.

58

Net gain: Gain - Free carrier absorption

0.25 % tensile

Larger Net Gain (Ge) when:

5e18

1e192e19

in cm-3

5e19

Takinai, in this school

Toward Ge Injection LDs

59

• Large strain• Strain tensile ~1%,

• Here, n-type carriers can be low 1019 cm-3

• Challenge is how to strain Ge that high.

• No worry about wavelength.

The Fourth Myth (potential)

• LSIs and on-chip D(dense)WDM do not coexist under uncooled chip architecture (LSIs). • Thermal fluctuation on a chip malfunctions

DWDM because of wavelength fluctuation.

60

Hot chip and DWDM

61

From ECOC 2008 tutorial by U. Vlasov

Wanted:

Wavelength Locking

Computing: ENIAC (1946)

• Devices• 17468 vacuum tubes

• 70000 resistors

• 10000 capacitors

• metal interconnects

• Size: 30 tons

• Power consumption: 150kW

Thanks to Wikipedia

62

Principle of Wavelength Locking

• Temperature fluctuation resulting in:• EMT(emitter) grain spectrum shift: dG/dT

• MUX/DEMUX, EMT, MOD(Modulator) wavelength shift: dω/dT

• Locking by local strain ε; • dG/dT =-dG/dε (dEg/dT = - dEg/dε), • dλ/dT =- dλ/dε (dn/dT = - dn/dε)

63

• Si microbeam structure can be fabricated using SOI wafers.

• A few % strain at surface of the Si beam by a few μm pushing down.

Approach: Local Strain

64

Si

Stress-Bandgap relation

• Si bandgap shrinks under tensile and/or compressive stress.

• 1.3 μm under ~1 % strain-tensile.

65

Silicon

µ-Photoluminescence(PL)

• Excitation• 1 μm dia. Nd YAG at

beam edge (X).

• 2.6mW@ 457nm

• Detector• InGaAs@-100°C

• Detection limit ∼1750nm

• Stressing point• Bent 1, 2, 3.

66

Typical Si beam

Bent 3 2 1x

µ-PL

Evidence of Bandgap shift

• Red shifts in PL spectra under strain (Bent-1, -2, and -3).

• It reaches ~1.3 μm. • 1 % tensile strain in

the Si beam.

67

K. Yoshimoto, Opt Express 25,26492, 2010

Si beam

• However, no shifts in sharp (FP) peaks. • Refractive index does

not change with this level of strain.

• see Hirase in this school

Horie, in this school

Reconfigurable Emitter

68

O band U band

• One emitter emits various λ’s. • Reduce # of

emitters on a chip.

• GaAs on Si beams should cover from the whole optical communication wavelength range (O- to U-band).

Horie, in this school

Reconfigurable Emitter

69

GaAs on Ge on Si beam• One emitter emits various λ’s. • Reduce # of

emitters on a chip.

• GaAs on Si beams should cover from the whole optical communication wavelength range (O- to U-band).

Summary

• The three myths and the breakthroughs in early Si photonics reviewed

• Challenges of WDM implementation in Si electronics reviewed.• Wavelength locking via strained Si, Ge, and

GaAs.

• Demands on on-chip WDM: • Injection Ge laser: by n+ and strain+

• Reconfigurable(Tunable) emitter: GaAs on Ge on Si beam.

70