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
