Optical Interconnects

Speeding Up Computing

Matt Webb

PICTURE HERE

In the near future All logic

operations solved using with optics

Have to start somewhere

Picture of something to signify amazing computing speeds

Why do we care? Problems with scaling of electrical

interconnects (EI) Ex: Telecommunications already

moved away from electrical lines Performance between chips is

already affected by EI Near future will be a problem on

chips

Solutions Several possibilities

New architectures• Minimize interconnections

New design approaches• Emphasize interconnection layout

New medium for interconnection• Optics

Background Research on optical interconnections

has been going on for >20 years With optoelectronic digital

computing Development of SEED’s and VCSEL’s Practical to implement

SEED/VCSEL picture?

Aspect ratio limitations

Bandwidth limitations

Delay limitations

Timing for optical signals Virtually independent of

temperature Virtually no degradation of signal on

the scale of meters Slower propagation, but very reliable

and predictable Could eliminate high power clock

circuits

Benefits of OI Optoelectronic devices can be used

as impedance transformers No inductance on an optical line Do not generate or detect radio-

frequency signals or interference Long or short does not matter

More benefits Larger synchronous zones, even on

multiple chips Allows “fire-hose” architectures Lower power dissipation after

“break-even length”• 100μm – 10’s of cm

Benefits cont. Voltage isolation Increasingly important due to

smaller power supplies Larger density for long distance on-

chip and off-chip interconnects No need for hierarchy of

interconnects

Scaling of OI transmitter/receiver Only viable if

technology for the design of TX/RX can keep up with future generations of silicon technology

Typical model Laser diode driver Transimpedance Amplifier

Latency vs. Length 3Gb/s, 0.25μm CMOS

Channels vs. Length

Power consumption vs. Length

Main Challenges for OI Young and expensive Systems that could take advantage

of optics will most likely have different architectures than today's

Problems and benefits are misperceived by those not involved in recent research

Receivers Power dissipation Small capacitance Larger noise immunity “Receiverless”

Transmitters Quantum-well modulators VCSEL’s LED’s Si based optoelectronic devices

Technology Absence of low-cost and practical How to integrate

III-V devices Compatibility Hybrid integration

Solder-bonding

Misperceptions Wavelength is too large

Not true for longer interconnect lengths Conversion of optics to electronics is

inefficient Power, area, and time Current generation of technology

Conclusions Use of OI could solve many of the

problems faced in today’s electrical systems

Much work remains to make the technolgy feasible

When it’s all said and done OI’s are the future