A Distinguished Engineer Works to Bring Photonics to Computing

Story by Al Riske. Photography by Howard Friedenberg.

3.June.08 - Ashok Krishnamoorthy finds himself on the cusp of a dream.

He's leading an ambitious research project that aims to bring light speed (rather than wire speed) to microprocessors.

How? Through a combination of materials science, quantum mechanics, silicon photonics, and an optical extension to Proximity Communications, Sun's wireless chip-to-chip I/O technology.

The promise is so great that the U.S. government's Defense Advanced Research Projects Agency has awarded Sun $44 million to fund the project over the next five and a half years.

In fact, Sun's proposal was the sole winner in a highly competitive process.

"We created a vision all the way from atoms to systems, and that appealed to DARPA," Krishnamoorthy says.

For Krishnamoorthy, a Distinguished Engineer in Sun's microelectronics group, the project represents the culmination of a career-long goal to bring more photonics into computing.

"From a hardware perspective, you have processors, memory, and interconnects. The interconnect side is where my work focuses. How do we make better interconnects?" Krishnamoorthy says.

"My training is in optics. That's my background. And I believe light is really the right technology for communicating information."

No one is going to argue with that when it comes to sending data over long distances. Fiber-optic cable rules.

"When the distance times the bandwidth exceeds a certain metric, then light becomes a better choice than electricity," Krishnamoorthy points out. "Then the question is what is that distance and what is that bandwidth?"

So far, the distance has been measured in meters and hundreds of meters.

"So what do we need to do to get it to millimeters?" he asks.

Krishnamoorthy's interest in light-speed communication dates back to his college days.

"I've spent a lot of time in the photonics world," he says. "Before joining Sun I lead a startup doing optical technologies and Sun was actually one of my potential customers. Before that I spent a decade at Bell Labs doing R&D in photonics. Before that, my Ph.D. was in photonics and computing. So it has really been my career goal to bring more photonics into computing."

Krishnamoorthy joined Sun in 2003 to work with Sun Fellow Jim Mitchell and others on an earlier DARPA-funded project dealing with high-productivity computing systems.

"Some of the silicon photonics technology development happened there," he says. "Based on this technology and our ideas about how photonics can help computing, DARPA gave us this one-year contract -- a small seed program -- to go explore this and come up with the grand vision. We were one of five groups to do that. I was quite surprised that we were the only team chosen in the subsequent competition, but obviously quite pleased about it."

The reason for using photons instead of electrons for communication sounds iron-clad.

"We realized that photonics is the best technology for communications simply because light particles -- photons -- don't interact with each other. In fact they hate each other. They can pass right through each other and not say a word," he says.

"On the other hand, electrons are fermions, particles that have charge and half-integer spin, which means their nature is to interact, to change each other -- and for communications that can be a bad thing. You don't want want these things to be talking along the way; they'll just interfere with each other. The information will get scrambled and you'll have a problem."

But it's not really that simple. Significant challenges remain.

Perhaps the biggest roadblock has always been silicon itself. Most microchips are made from a type of silicon known as CMOS (complimentary metal-oxide semiconductor), which has never been considered particularly good at optical communications.

"In fact, silicon was thought of as an optically dead material," Krishnamoorthy says.

As recent studies have shown, however, that's not exactly true -- and there may be ways to enhance its abilities.

"So this becomes a real interesting opportunity now," he says.

No one had been interested in using anything but silicon because no one wanted to replace their existing foundries, but now it appears they won't have to.

"The direction we want to go is to use optics for what it's good at, which is communications, and electrons for what they're good at, which is computing," Krishnamoorthy says. "So the genesis of the project goes back to the roots of what each technology is really good at -- and the discovery that materials like silicon and germanium can be good for both."

That simplifies manufacturing.

"If it's in silicon it's much easier to manufacture because you can do it at the same time as you do the circuits. But so far, what's been done in silicon is inefficient compared to electrical wires. So, although the potential of photonics is large, the implementation is poor. Why? With silicon what you gained was manufacturing but what you lost was a good electro-optical coefficient. In other words, a strong electro-optical effect," he says.

"That's really the challenge ahead: How do we improve that effect in silicon? How can we do that by doing work at the materials level, changing the quantum mechanics of these material, improving the way light gets confined in silicon, and creating low-power, small-area components that will convert light to electricity and back again?"

There's obviously a lot of risk involved.

"We're trying for the first time to take fundamental new discoveries in science and in photonics all the way through to the systems level," Krishnamoorthy says. "We've got to improve the efficiency of these electrical-to-optical transmitters and receivers by two orders of magnitude."

On the other hand, he says, "We believe that our ideas for doing that are the most complete, the most aggressive and ambitious, and the most far-reaching from a systems perspective."