Computer researchers have fashioned
infinitesimally tiny electronic
switches using conventional
chipmaking equipment, demonstrating
that the semiconductor industry
will be able to continue shrinking
its basic building blocks at
a torrid pace at least until the end
of this decade.
At a technical conference
being held this weekend in Kyoto, Japan,
a scientist for the Intel Corporation
reported that the company had
successfully made a handful
of silicon transistors no more than 70
to 80 atoms wide and 3 atoms
thick. They are capable of switching
on and off 1.5 trillion times
a second, making them the world's
fastest silicon transistor.
Although the Lilliputian
switches do not represent the smallest
experimental transistors yet
invented, they are being hailed by the
industry as a watershed because
they were made using standard
commercial techniques and the
same materials used in today's
microprocessors and memory
chips.
The achievement indicates
that researchers have once again found a
way around technical barriers
that might have slowed or stopped the
phenomenal four-decade march
that has led to the speed and power of
today's computers.
"This is a very good piece
of technical work," said Hon-Sum Philip
Wong, an I.B.M. research scientist
who presented a similar paper in
Japan this weekend focusing
on related technical problems in
building transistors three
generations in the future.
Although semiconductor
engineers can now predict with great
accuracy the amount of computing
power that the transistor advance
will bring with it by 2007,
computer scientists cautioned that
speculating about labor-saving
applications or dazzling
technological consequences
can be more risky.
While the semiconductor
manufacturing process underlying the
modern microchip continues
to improve steadily, specific new
applications are frequently
unpredictable. Some of the most
prominent advances of recent
years ó for example the electronic
wristwatch, the personal computer
and the Internet ó all largely
came as surprises to experts.
The research will make
make possible computer processor chips with
as many as one billion transistors
and 20 gigahertz speeds. That is
more than 23 times the number
of transistors used in Intel's
current state-of-the-art Pentium
4 microprocessor, which has 42
million transistors and is
capable of executing 1.7 billion
instructions a second.
It will also make possible
a generation of fingernail-sized memory
chips that can each store four
billion bits of data ó more than 333
copies of "Moby Dick."
Industry researchers said
that shrinking the transistor size was a
technical tour de force.
"It proves that Intel
is a technical heavyweight," said G. Dan
Hutcheson, president of VLSI
Research, a semiconductor industry
research firm based in San
Jose, Calif. "What Intel does so well is
take a technology that is not
meant to be extendable and pushing it
past its limits, making computing
incredibly cost-effective."
Moreover, the new chips
will consume less than one volt of
electricity ó perhaps
less than half of today's Intel
microprocessors ó making
it likely that within half a decade the
world's fastest processors
will be portable and perhaps even hand-
held.
Significantly, in an era
when vast farms of Internet and World
Wide Web servers are being
blamed for a share of the nation's
electrical power crunch, the
Intel scientists reported that the
power consumption of each new
generation of chips is shrinking more
quickly than the computer power
is increasing.
Indeed, the dwindling
power requirement of the Intel transistors
represents one of the fortunate
paradoxes of silicon-based
computing: these transistors
are able to switch on and off more
than three times faster than
today's microprocessors while
consuming only slightly more
than half as much electricity.
The Intel technical paper,
which was presented this weekend at the
Silicon Nano Electronics Workshop
in Japan, comes as good news for
an industry that in the last
two years has been experiencing
self-doubt about the longevity
of its guiding principle, known as
Moore's Law.
First noted by the Intel
pioneer and cofounder Gordon E. Moore in
1965, the rule has held that
the number of transistors that can be
etched on a single chip of
silicon doubles on average every 18
months.
But engineers and physicists
note there is nothing technically or
physically inevitable in the
continued shrinking of the fundamental
size of electronic circuitry.
Indeed, as long ago as 1993, Dr.
Moore himself publicly voiced
his doubts about the permanence of
the progression that received
his name. In a speech given in
Silicon Valley that year, he
suggested that a process known as 0.25
micron might be the point at
which the industry ran into
fundamental physical limits.
A micron is a millionth of a meter.
Nevertheless, chip makers
blew past the 0.25-micron size two
generations of chips ago ó
roughly four years ó and now use exotic
optical techniques to manufacture
the most advanced processors and
memory chips, which are based
on transistors with 0.13-micron
technology.
Now Intel scientists are
saying that they can see their way at
least three more generations
into the future, to transistors with a
0.045-micron technology ó
less than one-fifth the size limit that
Dr. Moore speculated about.
"Everyone who predicts
the end of Moore's Law has always been
wrong," said Gerald Marcyk,
Intel's director of components research
technology manufacturing group
in Hillsboro, Ore.
The new results are a
dramatic contrast to a pessimistic article
written by another Intel researcher
two years ago in the journal
Science.
The Intel scientist, Paul
A. Packan, wrote that it was not clear
whether the most common type
of silicon transistor could be scaled
down beyond the 0.13-micron
generation of chips that began to
appear last year, because semiconductor
engineers had not found
ways around basic physical
limits.
Now that those potential
physical limits have been surmounted, the
researchers acknowledged they
are in a realm where the switches are
so small that arcane physical
effects such as the quantum behavior
of electrons become a factor,
making chip-building as much an art
as a science.
The Intel scientists performed
some "heroic efforts" to make a
handful of transistors work,
Dr. Marcyk said, adding that the
specific methods used were
Intel manufacturing trade secrets.
One of the principal challenges
in creating these kinds of
switches is that their smallest
dimensions are much smaller than
the wavelengths of light generated
by the machines that etch
patterns on silicon wafers.
In conventional lithography,
a mask is used with transparent areas
and opaque areas, like a photographic
negative. However, the light
passing through the transparent
areas bends slightly, leaving a
slightly blurry image. As circuits
shrink below the width of a
lightwave, this becomes a profound
problem.
However, engineers have
found a way to turn that manufacturing
defect into a remarkably precise
tool through a process called
"phase shifting," which makes
use of interfering patterns of light
to create ultrafine circuit
lines.
The semiconductor industry's
technical road map has generally
forecast that integrated circuits
will stay on their doubling path
until 2014. But doubters have
continued to emerge, suggesting
either that there will be insurmountable
technical obstacles or
that at some point the cost
of each new generation of semiconductor
factory will soar astronomically.
The Intel researchers
said they had been faced with another
challenge besides building
ever tinier chips. Dr. Marcyk said that
each time he presented his
results to Andrew S. Grove, Intel's
chairman and co-founder, Dr.
Grove insisted, "I want you to show me
the limit."
"I have to tell him, `Andy,
I haven't found the limit yet,' " he
said.
Copyright 2001 The New York
Times Company