Researchers at Intel and the University of California, Santa Barbara announced on Monday that they’ve succeeded in building a silicon-based laser that could be easily fabricated using the same manufacturing tools as those used to make microprocessors. They believe that the light source, dubbed a hybrid silicon laser, is the device that will finally allow engineers to integrate photonics inexpensively into computer chips.
The advantages of adding lasers to microprocessors are evident in the fiber optics industry: by encoding data in light, it’s possible to pipe information through fiber at a speed of gigabytes per second. The catch is that optical devices, such as lasers, modulators, and detectors, are relatively expensive and complicated to make; hence, the computer industry hasn’t been able to take advantage of this high-bandwidth technology.
Instead, today’s microprocessors rely on copper wires to route electrons between transistors. With billions of transistors in each processor, and multiple processors built into computers, copper creates a significant bottleneck.
The hybrid laser would let data zip between transistors and chips at unprecedented speeds–it might allow engineers to rethink computer architecture, says Mario Paniccia, director of Intel’s Photonics Technology Labs. “It could really change the way you look at computing,” he says. “We’ve found a way to integrate a light source into silicon in a volume manufacturing sort of way,” says Paniccia, “and the performance is good.”
By engineering a new type of laser that combines the light-emitting properties of a material called indium phosphide, and the light-routing properties of silicon, Paniccia and John Bowers, professor of electrical and computer engineering at UCSB, have overcome earlier challenges that kept silicon-based lasers from being feasible.
While silicon is not naturally a good light emitter, it does have the ability to confine and route light. This makes it an ideal material for the laser’s cavity, where photons bounce back and forth, building up enough intensity to eventually produce a laser beam.
Some researchers have tried to affix external light sources to silicon cavities. The problem with this approach, says Paniccia, is that it is prohibitively expensive and difficult to perfectly align an external light source with nanometer-scale silicon cavities in the manufacturing process.
To solve this problem, the researchers built their light source directly onto the cavity. They first etched laser cavities in silicon, using the same lithography process used to produce Intel’s microprocessors. Separately, they built an indium phosphide light emitter. Next, the silicon and the indium phosphide were bonded together in a unique process that uses a thin layer of “glass glue” only 25 atoms thick. The glue is needed, explains Paniccia, because the atoms of silicon and indium phosphide don’t naturally line up when directly bonded together, resulting in a nonfunctioning device.
Finally, metal contacts are connected to the indium phosphide, and an electric current applied. When the current is turned on, negative electrons and their positive counterparts combine within the indium phosphide, explains Bowers, a process that produces photons. These photons are then easily captured and concentrated in the silicon cavity, which emits the beam of light.
The research was funded in part by the Defense Advanced Research Projects Agency, and will appear in an upcoming issue of Optics Express. And it’s the most recent advance in a flurry of work out of Intel and its partner universities in the past few years. Last year, for instance, Paniccia’s team published three articles in the journal Nature, detailing significant advances in building silicon-based photonic devices (see “Intel’s Breakthrough,” July 2005).
Other groups are also trying to use well-established silicon processes to put light into computer chips, says Harry Atwater, professor of applied physics and material science at the California Institute of Technology in Pasadena. He and researchers at other institutions, including MIT, are looking to silicon quantum dots for their light-emitting properties, which may eventually be less expensive to manufacture than hybrid lasers using indium phosphide, he says. Indium phosphide, he adds, could still be too expensive for the semiconductor industry to adopt widely. “I’m enthusiastic about [Intel and UCSB’s research],” he says, “but [cost is] the thing that tempers my enthusiasm.”
He adds that when a company such as Intel finally adopts a process like this in its microprocessor manufacturing plants, “that’s the ultimate proof that it’s production worthy.” Until then, he thinks the researchers have “a long way to go” to demonstrate the cost effectiveness of the technology.
Indeed, Paniccia and Bowers suspect that it could take five years for their device to be incorporated in chip-making facilities. But they’re excited by early results, Paniccia says.
His team has already built a laser that emits infrared light with a wavelength of about 1,800 nanometers. The next steps will be to modify the design of the silicon cavity, he says, improve efficiency, and to tune the laser to emit light at many other wavelengths. Simply modifying the cavity to change the properties of the hybrid laser is one example of the benefits to using silicon, says Paniccia. Instead of fabricating a completely different light source, he says, they can just exploit the well-established silicon processes. “It puts all the complexity on the silicon.”