MIT professor says 'strained' silicon holds the promise of faster microchips. Article from Boston Globe 4/11/02.
STRESSED FOR SUCCESS
MIT PROFESSOR SAYS `STRAINED' SILICON HOLDS
THE PROMISE OF FASTER MICROCHIPS
Author(s): Hiawatha Bray, Globe Staff Date: April 11, 2002 Page: C1
Section: Business
Some people work better under stress. Same thing for computer chips,
according to an MIT materials science professor. And he's set up a business
that aims to replace today's "relaxed" microchips with a new generation
made with "strained" silicon.
"It's the perfect solution for how to get more performance out of silicon,"
says Gene Fitzgerald, 39, cofounder of AmberWave Systems Corp. of
Salem, N.H. AmberWave's 35 employees have designed a new way to
make the silicon wafers that form the basis of most microchips. The
company, founded in 1998, has attracted $47 million in venture capital from
Adams Capital Management, Arch Venture Partners, and Dow Chemical
Co., among others. AmberWave's strained silicon technology has attracted
the attention of chip designers since the mid-1980s. It's being developed
independently by IBM Corp., which shares Fitzgerald's view that strained
silicon offers a relatively easy way to make more powerful chips.
Unlike other techniques for boosting chip speeds, strained silicon holds out
the promise of a big performance increase without the need for a major
overhaul of the chip manufacturing process. Strained silicon chips are made
using the same processes as today's relaxed chips. The higher performance
comes from the properties of the silicon itself.
At the heart of the problem is the relentless demand for faster chips that use
less electricity and run cooler. Everybody knows about Moore's Law - the
principle that says chip makers can double the number of transistors on their
chips, and hence their speed, every 18 months. But there are other things you
can do to improve chip performance. And some of these tricks involve
changing the chemical composition of the chip's foundation, or substrate.
Most chips use a silicon substrate. They are etched, dozens at a time, onto
wafers that are usually either eight or 12 inches in diameter. But faster chips
can be made by using other materials, such as an alloy called gallium
arsenide. Chips made of this material can move electrons much faster than
silicon. Gallium arsenide chips are widely used in digital switches and routers,
where sheer speed is at a premium.
But gallium arsenide is much more expensive than silicon. Worse yet, making
chips of it requires an entirely different process than that used for silicon. A
typical chip plant or "fab" costs at least $1 billion, much of that tied up in
highly specialized machines used to make silicon chips. Switching to gallium
arsenide would mean replacing nearly all of this gear. The resulting chips
would be far too expensive to compete with slower silicon in home
computers and many other tasks.
"You're never going to get off the silicon platform if you want to have
low-cost chips," Fitzgerald says. So he needed a way to get silicon chips to
work faster, without the need to replace the fab's chip-making hardware. He
claims that strained silicon is the answer.
The strained silicon is produced by depositing a mixed layer of silicon and
germanium onto a silicon base. The atoms of the silicon-germanium alloy are
larger than the silicon alone. Next, a layer of pure silicon is deposited. The
silicon atoms conform to the shape of the silicon-germanium, which causes
the atoms to be pulled out of their normal shape, stretched over the
underlying molecules. It's become a piece of strained silicon. This simple
change in the silicon's shape causes electrons to flow through it much more
rapidly than a normal piece of silicon.
"It allows you to achieve much faster device performance," says Steve
Byars, managing director of IQE PLC in Cardiff, Wales. Byars's company is
a leading maker of silicon wafers, and one of the first to begin production of
wafers based on the AmberWave process.
Byars said that he expects strained silicon chips to offer performance that
compares well to gallium arsenide.
"You'd be expecting a 40 to 50 percent speed improvement," Byars says, and
"round about a 25 to 30 percent power reduction benefit."
But unlike gallium arsenide, which requires a complete redesign of the chip
plant, chip firms could switch to strained silicon wafers with only minor
adjustments to the manufacturing process.
"You don't need to go out and buy a whole new wafer fab," says Byars.
The downside is that strained silicon wafers cost a lot. Byars says a standard
eight-inch wafer sells for about $120, and he expects strained silicon to cost
two to three times more. The chip plant in Dresden, Germany, where
Advanced Micro Devices Inc. makes its Athlon processors can use up to
20,000 wafers a month. At that rate, strained silicon would cost the company
an extra $58 million a year in materials. AMD would either have to eat the
extra cost, or pass it on to customers, which would undercut AMD's
low-price challenge to rival Intel Corp.
The cost of strained silicon is its biggest drawback, says chip analyst Nathan
Brookwood of Insight 64 in Saratoga, Calif.
"People have demonstrated that it works," says Brookwood. But the question
for potential customers, he says, is "does it improve performance more than it
adds cost?" If not, he warns, strained silicon will only catch on in markets
"where you absolutely have to have the performance and you couldn't get it
any other way."
IBM can wink at the higher cost of strained silicon, because the company
isn't planning to offer the technology to the mass market. IBM is a major
chip maker in its own right, producing custom processors for its own line of
mainframe computers, as well as the PowerPC chips used in Apple
Computer Inc.'s elite Macintosh line. Jeff Welser, a senior manager in IBM's
strained silicon project, says the company wants the new wafer technology
for its own internal use only.
"We have no plans to try to sell our silicon wafers to anybody," Welser said.
AmberWave doesn't make strained silicon wafers; just the technology to
create them. It has to persuade the world's wafer makers to adopt the new
technology - and that means convincing mass market chip producers like
Intel Corp. and AMD. These firms are desperate to shave their production
costs, and they won't embrace strained silicon unless they're convinced that
its higher performance justifies its premium price.
For his part, Fitzgerald has no doubts that strained silicon will be cheap
enough and fast enough to become the standard raw material in much of the
chip-making world. And with a portfolio of patents on the underlying
know-how, he thinks every company that wants to use the technology will be
paying AmberWave for the privilege.
"When we become profitable," Fitzgerald said, "we'll be very, very
profitable."
Hiawatha Bray can be reached at bray@globe.com.
SIDEBAR: |
