10 February 2010—A laser made of germanium may open the door to optical interconnects on computer chips that have multiple processing cores, say researchers at MIT, who recently demonstrated such a laser.
The group, led by Jurgen Michel, a research scientist at MIT’s Microphotonics Center, achieved low-level lasing at room temperature with a laser built from layers of germanium grown on a silicon wafer and powered by pulses from a separate laser. The paper is set to be published soon in Optics Letters.
Michel envisions using such a laser to make an optical bus, which would carry data around a multicore microprocessor. On a chip with 64 cores, such a system could have 40 times the power efficiency of traditional wire connections, he estimates. And in the next few years, chips with hundreds of cores are expected to make the concept even more useful. Another approach to such a system involves bonding lasers that are made of compound semiconductors and built separately onto silicon chips, but that could prove too costly and error prone for mass production. Germanium is already used in silicon manufacturing processes, and growing a laser directly on a chip should be vastly more efficient, Michel says.
Germanium, like silicon, is an elemental semiconductor that doesn’t easily emit light, because it has an indirect band gap. So when electrons within the germanium are excited by an outside energy source, such as a laser, and then drop back to a lower-energy state, they emit the excess energy as heat rather than light. To get the material to produce photons instead, the researchers had to alter the germanium to achieve a direct band gap.
“Luckily with germanium, the difference between the direct and the indirect band gap in terms of energy is not that great,” Michel says.
The researchers used two tricks to make up that difference. First they doped the germanium with phosphorus. The electrons in the phosphorus occupy the lower-energy state so that it’s no longer available to excited electrons, which instead drop to the direct band-gap energy. ”We filled the indirect band gap so much that the electrons would spill into the direct band gap,” Michel says.
The team doped the germanium with 1019 atoms of phosphorus per cubic centimeter. Michel says they need to increase that to at least 5 x 1019 to get about 10 times as much light out of the laser and thus make it practical. At the high temperatures used to make the laser, phosphorus tends to leak out faster than the researchers can put it in, so such an increase will require changes to their doping methods.
Increasing the doping will also produce a wider range of laser wavelengths. In the experiment, the team saw light from 1590 to 1610 nanometers, but with more doping that could become 1500 to 1620 nm. Silicon is transparent at those wavelengths, so the same material used to build circuits could also build waveguides to route the light around a chip.
The team also reduced the difference between the two energy states by putting mechanical strain on the germanium. They built the device by depositing a silicon oxide on a silicon chip, with a 1.6-micrometer-wide trench in the oxide; then they grew the germanium in the trench at about 600 °C. They annealed the device at 850 °C. Because silicon and germanium shrink at different rates as they cool from the annealing temperature, the process stretched the germanium so that its atoms were farther apart.
“The key issue for applications is whether an electrically pumped laser can be made,” says Philippe Fauchet, a professor at the University of Rochester’s Institute of Optics, who calls the work “really important.” Using electricity instead of another laser to power the device is necessary to make the laser practical for on-chip use.
It will also need to emit light in a continuous wave instead of in pulses, as it does now. Michel says there appears to be a clear path to reaching those goals, though he expects it to take at least a couple of years.
About the Author
Neil Savage writes about optoelectronics and other technology from Lowell, Mass. In January 2010 he reported on a new way to get silicon to emit light by using plasmonic waveguides.