Where Industry Meets Innovation

  • Contact Us
  • sign in Sign In
  • Sign in with certificate
mit campus

Resources

Search News

  • View All
  • ILP News
  • MIT Research News
  • MIT Sloan Management Review
  • Technology Review
  • Startup Exchange

ILP Institute Insider

April 10, 2012

Shining Light on a New Age in Computation

Thinking on a Different Scale

Eric Markowsky

Most people today are accustomed to thinking about information in terms of gigabytes, like the hard drive memory in our laptops or the data plans on our smartphones. For more sophisticated technology professionals working in fields that produce enormous data sets, such as nuclear science or biomedical research, terabytes might be the norm. But for Lionel Kimerling, MIT Thomas Lord Professor of Material Science and Engineering and the Director of the Microphotonics Center, questions about information capacity have reached a whole new scale.

Lionel Kimerling
Director, MIT Microphotonics Center
“We can handle petabits of information per second,” Professor Kimerling explains while discussing a current collaboration with Professor Anant Agarwal and MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL).

Together, researchers at CSAIL and the Microphotonics Center have been studying the potential for incorporating photonic communication with multicore computing, first developed at MIT by Agarwal in 2002. By using an optical broadcast to allow simultaneous communication between all the cores, the collaboration seeks to address some of the key challenges facing performance, programming, and power efficiency in multicore processors today. It’s called “All-to-all” computing, and it might be the key to unlocking the full potential of parallel processing.

“CSAIL developed a simulator that enables us to start plugging this in with real computer applications to see what the advantages are,” Kimerling says. “We’re learning a lot about where the energy goes in and how it’s being used, but right now we’re at the stage where we can’t find applications that require enough communication to meet the capacity that we’ve created.”

Petabits – that’s one order of magnitude beyond terrabits – per second. “No one ever thought about having that capacity,” Kimerling adds.

Photonics, Communication, and Computation: Bringing it All Together
The collaboration that produced “All-to-all” computing is only one of the most recent groundbreaking projects undertaken by the Microphotonics Center over the past two decades. Since arriving at MIT in the early nineties, Professor Kimerling has focused his research on the application of photonics to enhancing existing processes and solving new problems.

Kimerling started as a Bell Labs researcher during an era of revolution in communications technology. “I was able to participate in the development of integrated circuit technology on silicon, semiconductor lasers and fiber technology for the network,” Kimerling says. “The question was, if photonics did so well for the network, did it have anything to offer other functions where communication was important? About the time that I got to MIT it became apparent that in computing, communication was the limiting factor.”

In the standard computer model, the information from each computation needs to be transported from the central processing unit to memory, parked there, and then found and retrieved every time that data is required for another computation. Companies manufacturing chips were primarily concerned with scaling this process forward, but Professor Kimerling saw the potential to change the process.

He began searching for a solution that addressed energy, bandwidth density, and cost. “It was my feeling,” Kimerling says, “since I had been both involved in integrated circuit development and fiberoptic development that maybe the two of them could come together on a silicon photonic platform.”

So Kimerling and his group began developing photonic on silicon devices: waveguides, modulators, filters, high precision photodetectors, and, most recently, lasers. They have since begun looking at on chip integration, first making photonics and electronics work together to push down costs. “We call it electronic photonics synergy,” Kimerling says. “People are beginning to see how to do it properly, that’s creating the platform for silicon photonics.”

The CSAIL collaboration provided yet another avenue for exploring the application of photonics to computing. “It took us about two years to figure out how to talk the same language,” Kimerling remembers, “because they think at the architecture level and we’re thinking about electrons and atoms. But once we got together and understood each other, everything clicked, and it became very clear what the big issues are.”

While multi-core computing creates incredible potential for performing a much larger number of computations, it actually exacerbates the communication challenges in computing because the cores need to coordinate with each other, and the information that any one core might need could be parked in any of the other cores. This poses problems for energy use and performance, since abundant energy can be expended searching the cores, and it poses a special challenge for programming because the location of pertinent data needs to be written into the code.


View related video "Lionel Kimerling on The Communication Technology Roadmap and Sustainability"


Using the optical broadcast to communicate among the cores addressed all three problems at once. When one core finishes a computation, it sends data to all the cores. By assigning each core its own wavelength of light, it would be immediately apparent to the other cores where the information had come from. “The cores who want it can use it and the ones who don’t want it don’t have to take it,” Kimerling explains, “but each one knows what the others are doing and you can efficiently allocate computing resources between many cores, and it turns out that the communication cost isn’t that high when you do it with photonics.”

The collaboration has expanded to include circuit designers and network specialists, all with an eye on using photonic communication to enable a whole new kind of computing.

On a Long Road – With a Map
Even with all the breakthrough developments coming out of the Microphotonics Center, there’s still a long way to go toward making silicon photonics a standard in the marketplace. “We saw that early,” Kimerling says, “and so we created the Communication Technology Roadmap (CTR).”

The CTR is the product of a broad-based collaboration between the Microphotonics Center and its associated Industry Consortium. The Consortium hosts two meetings a year at MIT: the fall meeting, focused on new research in photonics and communication, and the spring meeting, focused on mapping the future of the industry.

“When we work with industry, we create more or less a place where people can have coffee and donuts and discuss what the real problems are in a way that they don’t have to worry about proprietary uses and so forth because we’re discussing them at a high level, looking far out, and that’s turned out to be very beneficial,” Kimerling says. “Our CTR and our microphotonics center are scalable, so the more companies that want to participate the better.”

Companies that have signed on with the Consortium are helping MIT continue to be the world’s hub for research and development in silicon microphotonics. “It was started here,” Kimerling says. “We’re still able to create most of the leading-edge developments. And I think we have the insights into where it’s going and what the next key technology issues are going to be.”