October 9, 1999
Chip Progress May Soon Be Hitting
Barrier
By JOHN MARKOFF
AN FRANCISCO -- For more than three decades, it has been an
unshakable principle of the computer industry: every 18 months,
the number of transistors that will fit on a silicon chip doubles.
The phenomenon, known as Moore's Law for
the semiconductor pioneer who first observed it,
has been the basic force underlying the
computer revolution and the rise of the Internet.
As transistors have been scaled ever smaller,
computing performance has risen exponentially
while the cost of that power has been driven
down. And it has been assumed in the industry
that the rate of progress would hold for at least
another 10 to 15 years.
But now a researcher at Intel, the world's
leading chip company, has reported glimpsing a
potentially insurmountable barrier to the
advance of Moore's Law much closer at hand,
perhaps early in the coming decade.
In an article in the journal Science, the Intel
scientist, Paul A. Packan, says it is not clear
whether the most common type of silicon
transistor can be scaled down beyond the
generation of chips that will begin to appear next
year, because semiconductor engineers have not
found ways around basic physical limits.
"These fundamental issues have not previously limited the scaling of
transistors," he wrote in the Sept. 24 issue. "There are currently no
known solutions to these problems," he added, calling it "the most difficult
challenge the semiconductor industry has ever faced."
Dennis Allison, a Silicon Valley physicist and computer designer, said:
"The fact that this warning comes from Intel's process group is really
significant. This says that they see actual limits."
The report by the Intel scientist will be echoed by researchers from the
University of Glasgow in a paper to be presented in December at a
conference in Washington.
Without further advances in the miniaturization of silicon-based
transistors, hopes for continued progress would have to be based on
technologies that are promising but unproved: new materials, new
transistor designs and advances like molecular computing, in which single
molecules act as digital on-off switches.
To be sure, such dire warnings have been made periodically in the past --
an article in Scientific American in 1987 said Moore's Law was unlikely
to be maintained through the 1990's -- and each time semiconductor
designers have shown remarkable ingenuity to surmount seemingly
impossible barriers.
Indeed, Moore's Law -- first stated in 1965 by Gordon Moore, an Intel
co-founder -- proved to be understated; Moore had to revise his initial
prediction of 24 months for each doubling of chip capacity. And while it
is not an actual physical law, his observation has taken on an almost
mystical quality as the clearest expression of the power of human science
and engineering and many industry executives have come to see it as a
self-fulfilling prophecy.
In the last decade the advances described by Moore's Law have had an
accelerating impact on the personal computer industry, driving the cost of
desktop machines down from $3,000 to as low as $500 while increasing
their power.
The inventors of the original semiconductor design technology are for the
most part still bullish about extending that progress, whatever the
immediate hurdles.
"Historically the economic incentives to find new methods for device
improvement have regularly overcome the predicted scaling limits," said
John Moussouris, a physicist and semiconductor designer. "The physical
challenges may be getting harder, but the people and financial resources
to surmount them are also growing each year."
But for the first time the global semiconductor industry is grappling with
transistors so small that the placement of individual atoms will soon
become crucial.
For example, in the current generation of semiconductors, the wires that
interconnect transistors are etched as fine as 0.18 micron -- one
five-hundredth the width of a human hair -- and the individual insulating
layers that are inside a transistor may be only four or five atoms thick.
Semiconductor factories in Japan plan to begin mass production of chips
based on widths of 0.13 micron early next year, and such chips should
be in widespread use within two years. But beyond that generation, the
industry's leading researchers acknowledge there remain far more
questions than answers.
The next step would be widths of
0.10 micron, a milestone that in the
Moore's Law progression would be
expected three to five years from
now. But at that scale, Packan writes,
transistors will be composed of fewer
than 100 atoms, and statistical
variations in this Lilliputian world are
beyond the ability of semiconductor
engineers to control.
Packan said he had written the Science article to challenge the industry
and academia to focus on areas where breakthroughs are needed. "For
the last 30 years we've been engineering the device, and now what's
required is fundamental science," he said in a telephone interview today.
Intel executives cautioned against reading too much gloom into their
technical papers, saying that while they did not yet have precise
engineering solutions for breaking the 0.10 micron barrier, they were
confident that answers would be found.
They suggested that part of the reason for Intel's recent pessimism might
have more to do with the need for corporate secrecy than the arrival of
fundamental technical limits.
"We face serious challenges," said Mark Bohr, an Intel technology
development director and the co-author of an internal Intel technical
paper that enumerates the company's unsolved problems. "We all have
ideas to address some of these problems and admittedly they are iffy and
not fully developed, and you don't want to tip your cards too soon."
And Carver Mead, a physicist and a pioneer in semiconductor design,
says he still adheres to what has been the conventional industry wisdom,
suggesting that Moore's Law will continue to account for the pace of
silicon technology advances until at least 2014. "There are still some open
issues," he said. "and so the Chicken Little sky-is-falling articles are a
recurring theme."
But James Heath, a chemistry professor at the University of California at
Los Angeles who is a co-inventor of the carbon 60 molecule known as
the Buckyball, said the industry might be overly optimistic because it had
such a vast investment in today's silicon technology.
With researchers at Hewlett-Packard, Heath has developed a prototype
memory cell the size of a single molecule that operates on different
principles from today's semiconductors.
"I think their optimism for being able to continue until 2014 is not very
realistic," he said. "When you get to very, very small sizes, you are limited
by relying on only a handful of electrons to describe the difference
between on and off."
Executives at IBM, which along with Intel and Motorola is one of the
nation's dominant chip makers, acknowledged that it might be accurate to
warn of an impending limit to the shrinking of today's dominant chips,
known as C.M.O.'s, or complimentary metal oxide semiconductors. But
they said they believed they had found an alternative approach, known as
silicon-on-insulator, that held great promise at dimensions of 0.10 micron
and smaller.
"This paper is quite consistent with work we've published," said Randall
Isaac, vice president for systems technology and science at IBM's
Watson Laboratory in Yorktown Heights, N.Y. "But when a given
technology saturates, it is usually replaced by a new one."
fred |
