7 January 2005--Scientists at Intel Corp., in Santa Clara, Calif., reported on 5 January that they have built an experimental laser out of silicon. In principle, such lasers could allow the integration of electronics and optics in standard-issue silicon chips, rather than in chips made of exotic semiconductors. If that happens, fiber-optic connections now seen only in long-haul telecommunications networks could finally come to the lowly PC.
Many research groups have quested after a silicon laser, only to be frustrated by the material's uncooperative electronic nature: in response to a current, it usually generates far more heat than light. One of the best results had been achieved by Salvo Coffa, the research director of soft computing, silicon optics, and post silicon technologies for STMicroelectronics NV, in Geneva. But while Coffa's method of injecting current into a specially engineered silicon diode has yielded an efficient light-emitting diode, it cannot yet support a continous laser--the device most needed for optoelectronic applications.
Even Intel's laser doesn't work continuously yet, and it cannot operate directly from electrical stimulation; instead it gets "pumped" by a separate, non-silicon laser. Nonetheless, Coffa hails the experimental laser as "a very important step on the road to silicon optoelectronics."
Intel exploited what is know as the Raman effect, in which light scatters in certain materials in such a way as to produce another, longer wavelength. In a typical Raman laser, light is fed into a kilometer-long spool of optical fiber to produce the longer wavelength. "Silicon has a 10 000 times stronger effect," says Mario Paniccia, director of Intel's silicon photonics group. "We could do the same thing in a centimeter-long device."
Apparently researchers at the University of California, Los Angeles, grasped the same idea, and reported their development of the first silicon laser last October. However, their device required that a silicon chip be inserted in an 8-meter ring of optical fiber, whereas the Intel group made its laser all in silicon. They did so by replacing the fiber with a waveguide, basically an S-shaped ridge built on a 15-mm-by-15-mm silicon chip. The idea was to feed light from a separate laser into the chip and have Raman laser light emerge.
But, of course, it was not that simple. The power of a silicon Raman laser will quickly hit a plateau, because pairs of photons will smash into each other and release electrons. "The electrons absorb and scatter light," says Paniccia. "It's a diminishing return. As you pump the thing harder, instead of getting more gain, you are losing more light through the absorption process."One solution is to chop the incoming laser light into pulses so brief that by the time the troublesome electrons form, the lasing is already over.
But a practical laser must operate continuously, not in picosecond-long pulses. So the Intel researchers jiggered the waveguide so as to let longer pulses lase. The ridge in the silicon that forms the waveguide sits between two chemically altered tracks of silicon, forming a type of diode with the ridge at its center. Voltage across the diode sweeps the unwanted electrons away and keeps the light flowing through the chip. This allows the input laser pulses to last longer--130 ns--boosting the silicon laser's output power.
The Intel laser represents a big improvement over the first silicon laser, says Philippe M. Fauchet, chairman of electrical and computer engineering at the University of Rochester, New York. Fauchet, who has done research into silicon lasers himself, says that Intel's use of a voltage to eliminate stray electrons has let them get close to the continuous wave operation needed to transmit a steady stream of bits.
Fauchet notes that Intel's laser can never become the Holy Grail of the all-on-one chip laser some researchers envision, because it will always require a non-silicon external laser to drive it. Many other research groups, including his own and that of STMicroelectronics' are working on lasers that shine on electricity, not borrowed light. But Fauchet agrees with Paniccia that Intel's laser will still have many applications. "It's not electrical," he says. "But that doesn't mean it's not practical."
The silicon laser is just one part of Intel's recent drive toward remaking the optoelectronics world in its own silicon image. It is developing photodetectors, modulators, and other optical components in silicon in the hope that one day it will be able to put them together with existing chip-making processes and infrastructure.
Silicon may never beat the performance of optoelectronics made with expensive, exotic materials, such as gallium arsenide, but if silicon can do a decent job at a much lower cost, it may make entirely new applications possible. Not only may computers be optically linked into networks, but components might be optically linked within a single microchip. Before Intel introduces any of its silicon photonic devices, Paniccia says, it plans to get all of them to operate at 10 gigabits per second, the speed predicted for most long-haul optical telecommunications systems a year or two from now.