Design methodology development for VCSEL-based guided-wave optical interconnects

1
Design methodology developmentfor VCSEL-based
guided-waveoptical interconnects

• Michiel De Wilde Olivier Rits
• Ghent University, Belgium
• IMEC

2
Overview

• Optical interconnect
• rationale structure
• Optical interconnect design space
• exploration optimization
• Simulations to extract system-level properties

3
Optical interconnect rationale (1)
Moores Law
(source Intel)

• Denser inter-chip interconnect requirement
• Rising clock frequencies

4
Optical interconnect rationale (2)
ground plane
signal wire
A
wire length L
Inter-wire spacing area

• Wire capacitance resistance skin effect
• limit the electrical interconnect bandwidth B

(Miller-Ozaktas)
5
Optical interconnect rationale (3)

• Problematic interconnectbetween ICs and at the
  IC access level
• Industrial packet routers
• Some parallel and distributed processing systems
• Use of optics a solution on physical grounds
• No electromagnetic interference problems
• Almost distance frequency independent losses
• Optical I/O integration with CMOS
  solvesinterconnect problems at the IC access
  level

6
VCSEL-based parallel optical I/O
Connector
Fiber bundle
Package
PCB
VCSELs
Photodiodes
Solder balls
CMOS substrate (top side visible)
7
Fields involved
8
Complex design space

• Several approaches are possible
• Some continuously valued parameters too
• ?, operating currents, numerical aperture,
  physical dimensions
• Decisions in one field may affect other fields

Increase of numerical aperture of fiber
better coupling
less bend losses
worse coupling
(not to scale)
9
Design optimization

• Systematic way of making choices design
  methodology
• Designer states constraints
• Tool suggests good solutions meeting constraints
• Important system-level property categories
• Technological feasibility
• Performance (timing/power characteristics)
• Reliability
• Implementation cost

10
Design methodology development
11
Estimating timing/power/reliability

• Issues for direct estimation(e.g. from tabular
  data)
• Non-linear interactions between different
  fields(electrical, optical, thermal)
• Impact of noise and process variations
• Interconnect simulation
• Simulator
• Simulator models
• Stimuli
• Calculation of properties from simulation results

12
Simulator models
13
Simulator choice

• Device-level simulators
• OptiWave, RSoft, WinLase,
• Too detailed (some parameters are IP)
• Too slow for this purpose (finite-element
  methods)
• Circuit-level simulators
• SPICE, Verilog-AMS, VHDL-AMS,
• More concise parameter set possible
• Faster (only integration over time)

14
Example photodiode model
module pin_photodiode(in,anode,cathode)
input in inout anode, cathode power
in electrical anode, cathode parameter
real Cdep0, Cbo0, Rbas0, Resp0,
Id0 parameter real pole-1/(CdepRbas)
parameter real laplace_coeff_0CdepCbo
parameter real laplace_coeff_1CdepCboRbas
charge rc analog begin
I(cathode,anode) lt laplace_zp(RespPwr(in)Id,,
pole,0) Q(rc) lt laplace_np(V(cathode,a
node),laplace_coeff_0,laplace_coeff_1,pole,0)
end endmodule

• Terminals
• Model parameters
• Equations describing internal state and outputs

15
Driver/receiver model

• Normal analog electrical circuits
• IP protection no real circuit provided
• Alternative parameterised flowchart
• Validation measurementson non-hybridized CMOS

Receiver flowchart
Photocurrent input
Transimpedance preamplifier
Postamplifier
Equalizer
Decision circuit
Limiting amplifier
Digital output
16
VCSEL model

• Nonlinear differential equation system
• Jungo, et.al. VISTAS software package
• equations without spatial integration
• Difficult parameterization
• Validation measurements on non-hybridized VCSELs

17
Fiber-based optical path

• Abstraction of dispersion (short distance)
• Coupling coefficients for losses crosstalk

VCSEL-fiber crosstalk
Fiber-photodetector crosstalk
Fiber-photodetector coupling losses
VCSEL-fiber coupling losses
absorption
macrobend losses
Connectorlosses crosstalk
(not to scale)
18
Fiber bend losses

• Bend losses can be approximated using a
  combination of raytracing results (H. Lambrecht,
  et.al.)
• Validation through optical path measurements

90 bend X axis NA fiber Y axis bending
radius Blue color high losses
(source H. Lambrecht)
19
Simulation illustration
(exaggerated VCSEL model parameters)
(inverted output)
20
Simulation setup

• Interconnection signals
• Digital pseudorandom
• Design-specific
• Parameters
• Process corners
• Monte-Carlo generated
• inter-device correlation

• Noise
• asynchronous
• synchronous
• mesochronous
• Predominant noise substrate noiseat the
  receiver preamplifier
• amplify some few µAof photocurrent

21
Conclusion

• Illustrated optical interconnect
• rationale structure
• Discussed optical interconnect design space
• exploration optimization
• Simulations to extract system-level properties
• Explained approach
• Discussed models
• Future work

22
Acknowledgements

• IST Interconnect by Optics project partners
• Hannes Lambrecht (Ghent University, IMEC-INTEC)
• Fiber bend losses modelling
• Marc Jungo
• VISTAS VCSEL modelling project
• Fund for Scientific Research Flanders (Belgium)
  (F.W.O.)
• Research assistantship