Intel spices up silicon for ultrafast future
Rupert Goodwins in San Francisco
ZDNet UK
March 10, 2006, 15:15 GMT
Materials like gallium, arsenic, indium and antimony may help Intel to achieve speeds of 250Thz or more with its future chips
At the Intel Developer Forum in San Francisco on Thursday, Intel's director of technology strategy Paulo Gargini unveiled plans to reach transistor speeds of 250THz in ten years time.
By adding a wide range of other semiconductor materials to silicon wafers, Gargini said, designers will be able to dial in the atomic properties of the materials they work with.
"For the past thirty years, we've been applying scaling to silicon" said Gargini. "We've now got layers only four or five atoms thick, and we can't scale past three, two, one, zero. To increase performance, we need to look at electron mobility. Strained silicon can double that, but then we'll reach the limits."
Electron mobility is a measure of the maximum speed at which electrons can move through a semiconductor, with higher mobility materials making faster devices. Strained silicon improves this by moving the atoms farther apart.
Gargini said that while there are six or seven semiconductors with higher mobility than silicon, comprising elements such as gallium, arsenic, indium and antimony, it isn't just a matter of switching over to them. "You can't make wafers from them," he said, "but you can deposit these materials on a silicon wafer. You can make a custom semiconductor, because elements are mixable, that changes from area to area on the chip"
The resulting transistors will be a lot faster, Gargini said, and will work at a lower voltage than current devices. "We think silicon transistors will have a top speed of 25THz," he said, "and we should be able to get between ten times and a hundred times better than that. If we get ten, that's reasonable. And by a quick rule of thumb calculation, they should work down to around 200mV."
Gargini said that the new technology would enable some long-predicted changes. "We will be able to integrate photonics a lot more efficiency, and finally get die-to-die optical interconnects, as well as on-chip lasers and sensors."
Although lab results were promising, he said products were still at least ten years off. "There seems to be a standard delay of between ten and fifteen years from an idea being demonstrably worthwhile and it appearing. Strained silicon was proposed in 1991, perfected in 2002 and in production in 2003
