A View from Emerging Technology from the arXiv
1-Femtojoule Optical Interconnect Paves the Way for Low-Power, High Performance Chips
The wires that ferry electronic data across microchips are too power hungry for future microchips. But photons can now do the job instead thanks to a new generation of ultralow power optical interconnects.
Moore’s law is the observation that the number of transistors on integrated circuits doubles every two years or so. When Gordon Moore spelt out his eponymous law in 1965, the numbers involved were measured in thousands. Today, microchips contain billions of transistors, with the smallest having dimensions just a few nanometres wide.
All that makes for fantastically fast processing at huge data rates. But there’s another increasingly important factor in chip design: power consumption. Not only must the transistors themselves operate at low power but the power consumed in transmitting the data from one pat of a chip to another must also be low.
And therein lies a serious problem. An interconnecting wire on a chip just 1 millimetre long consumes about 100 femtoJoules for every bit it carries. That may sound small–a femtoJoule is 10^-15 Joules. But with data rates now hitting petabits per second (1 petabit = 10^15 bits), a large chip will eat about 100 Watts. And that’s just for the interconnecting wires. The power that transistors use up is on top of this.
The bottom line is that conventional interconnecting wires are at least ten times more power hungry than the next generation of chips can handle.
So the designers of future chips are turning from electrons to photons to do this job. The idea is to convert the electronic signals from transistors into photons and beam them around the chip at ultralow power.
That requires lasers to create the photons, modulators to encode the photons with data and detectors to receive the photons. But here’s the thing: all this has to be done with a power budget measured in just a few femtoJoules, something that hasn’t been possible.
Until now. Today Michael Watts and pals at the Massachusetts Institute of Technology in Cambridge say they’ve designed and built the first photonic modulator that operates at this ultralow power level. “We propose, demonstrate, and characterize the first modulator to achieve simultaneous high-speed (25 Gigabits per second), low voltage (0.5 peak-to-peak Voltage) and efficient 1 femtoJoule per bit error-free operation,” they say.
The new device is a hollow silicon cylinder that acts as a cavity for trapping light waves. It modulates this light thanks to a phenomenon known as the electro-optic effect in which the refractive index of silicon can be changed by modifying the voltage across it.
The modulator solves a number of problems that electronics engineers have been wrestling with. First, it is entirely compatible with the CMOS (complementary metal oxide semiconductor) process used for manufacturing chips and so can be made inside any existing fabrication plant. Previous attempts to make devices of this kind relied on indium which is not compatible with CMOS.
Next, the device is unaffected by the kind of temperature changes that occur in a chip. That’s been a problem for previous modulators of this type since their critical dimensions, and therefore the control they have over light, have always changed with temperature.
Watt and co fix this by exploiting the fact that the electro-optic effect is also temperature dependent. In the new design, this cancels out any changes in dimension ensuring that a temperature change has no overall effect.
And finally, they’ve got all this working with a measly 1 femtoJoule power budget.
Impressive stuff. Watt and co are justifiably pleased with the result. “The results represent a new paradigm in modulator development,” they say.
At the very least, it should make possible a new generation of powerful chips operating at lower power than ever before.
Ref: arxiv.org/abs/1312.2683: A One FemtoJoule Athermal Silicon Modulator