Intel Eyes Silicon Photonics As Future Speedway For Computing

The problem with today's computers is that, as fast as they are, they aren't fast enough. More precisely, the thin copper wiring used to link state-of-the-art processors can't support bandwidths greater than 15 to 20 gigabits per second. That's where optoelectronic devices come in, and that's why Intel is working on silicon photonics -- aka chips with built-in lasers. They can deliver communications speed of 40 Gbps and more.

Alexander Wolfe, Contributor

October 17, 2007

4 Min Read

The problem with today's computers is that, as fast as they are, they aren't fast enough. More precisely, the thin copper wiring used to link state-of-the-art processors can't support bandwidths greater than 15 to 20 gigabits per second. That's where optoelectronic devices come in, and that's why Intel is working on silicon photonics -- aka chips with built-in lasers. They can deliver communications speed of 40 Gbps and more.There are several threads in this complicated story, so let's take them one at a time. First, let's define what we're talking about. The devices Intel has demonstrated -- and for which it received a bunch of patents on Oct. 16 -- combine both a working laser and conventional silicon technology on the same chip. That's why this stuff is such a big deal. Laser fabrication isn't hard anymore, nor is making silicon. But putting the two together, or manufacturing a working, integrated optical-silicon interface, is a very significant breakthrough indeed.

Think of it as the chip world's version of bridging the blood-brain barrier. And engineers have had almost as hard a time as biologists. It's taken years of messing around in the lab with just the right combination of arcane materials, to come up with photonics devices which can be reliably manufactured at a reasonable cost.

In July, Intel demonstrated an important breakthrough in the form of the world's first working, 40G silicon-laser modulator. In September, it showed off the photo detector for the other end of the connection. (You need both of these to get a working link.) Mario Paniccia, an Intel Fellow and the director of the company's Photonics Technology Lab, led the development effort. Read about it on the Reseach@Intel blog, here (detector) and here (modulator).

Intel has built working laser-on-silicon modulators and demodulators, which make possible communications links delivering speeds of 40 gigabits per seconds. (Click picture to enlarge, and to see other opto slides.)

OK, back to our quick course in why this stuff is important. Intel's modulator/demodulator combo may have more immediate applicability to telecom networks than to computers. That's fine. Either way, photonics are ultimately going to replace copper as the preferred interconnection medium in electronics, just as surely as electricity (via either batteries or fuel cells) will soon edge out gasoline in our cars.

When applied to computers, understand that silicon-laser devices aren't going to handle messaging within microprocessors. Rather, they'll enable the hundreds of processors within large-scale teraflops computing systems to talk to all their sister CPUs without unacceptable latency or delay.

The need for this stuff isn't as far off as you might think. For one thing, there already are hundreds of teraflops-class systems extant in academic and research settings. These are the high-performance computing set-ups, which used to be called supercomputers.

More immediately, with quad-core processors here today, we're looking at a world where hundred-core servers are soon going to be commonplace. Such designs could benefit from light-fueled interprocessor communications, because copper is just too darn slow once you have too many adjacent parts vying for access to the same bus.

So you know this post isn't just a rehash of the stuff Intel demonstrated in September, I got onto the subject because I learned that Intel received three relevant patents on Oct. 16. The first is patent number 7,283,703, for a "Movable lens beam steerer." This is described as optoelectronic module, which has the added tweak of being able to position the lens to steer the light beam.

Schematic from Intel patent number 7,283,703, for a optoelectronics device with a steerable lens. (Click picture to enlarge, and to see other opto slides.)

The second patent is number 7,283,699, for an "optical package." It's described in the abstract as a technique for combining a chip and waveguide in the same package. The third item is patent 7,283,689, entitled "Optical waveguide having high dielectric constant contrast between cladding and core."

Thus, along with showing working prototypes, Intel has now corralled a significant collection of intellectual-property rights.

Perhaps optoelectronics will provide the path toward diversifying Intel's business away from its too-heavy reliance on microprocessors. If so, it would be a huge feather is CEO Paul Otellini's cap, and that's an understatement.

About the Author(s)

Alexander Wolfe

Contributor

Alexander Wolfe is a former editor for InformationWeek.

Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.

You May Also Like


More Insights