Continually needing to add computing power to its
microprocessors, Santa Clara behemoth Intel (INTC)
this year announced it was venturing beyond its traditional method of cramming
more and more transistors into a flat pieces of silicon in favor of a different
approach -- building chips in three dimensions.
With the 3-D devices slated to show up in personal computers this spring,
Intel CEO Paul Ottlina recently exulted that the technology would pay dividends "for
generations to come."
Yet even as he said that, his company and others were actively exploring
other ways to keep up with the increasing data demands of the myriad consumer
gadgets hitting the markets these days. And if the remarkable prescience of
Moore's Law is any guide, they stand a good chance of being successful -- at
least for a few more years.
 |
| Gordan Moore |
Initially postulated in 1965 by Intel co-founder and then-CEO Gordon Moore,
the law forecast that the number of tiny transistors squeezed onto chips would
roughly double annually, a prediction he later changed to every two years. And
despite critics who have repeatedly voiced doubts about the plausibility of
shrinking components any more, the industry has proved over and over that the
law has plenty of life left in it.
"All rumors of its death have been
completely exaggerated," said Liam Madden, a vice president at San
Jose-based chipmaker Xilinx, which recently claims to have made "a
bold step into the era of 3-D" with one of its own chip designs. While
steadily reducing chip components has posed monumental challenges, he
added, "amazingly, we always seem to be able to pull it off in the
end."
In 2005, the transistors in one
of Intel's most advanced microprocessors were 90 nanometers wide, more than
1,000 times thinner than the width of a human hair. Today, Intel is selling
chips that are 32 nanometers wide and will have 22-nanometer versions
incorporating its 3-D technology on the market soon.
Among the various chip-making
technologies being explored, 3-D has generated particular industry
interest.In traditional flat-chip
transistors, an insulated gate turns electrical signals on and off to
produce the "ones" and "zeros" that constitute digital
information. But the transistors tend to leak current, wasting energyWith Intel's so-called Tri-Gate
chip, the transistor is built with tiny upraised fins, which guide the
current along a three-dimensional channel. That helps the gate take maximum
advantage of the flowing current while limiting leaks, the company
contends. In addition, the design makes it possible to pack more
transistors closely together, enabling Intel to put 2.9 billion of them on
a chip about the size of a dime.
IBM and 3M also are intrigued
with the concept and in September announced they are working jointly to
make "silicon towers," with up to 100 chips. Using a 3M adhesive
that diverts heat from the chips so their circuity isn't damaged, the
semiconductor mini-skyscraper "would create a computer chip 1,000
times faster than today's fastest microprocessor," enabling more
powerful smartphones, tablets, PCs and gaming devices, the companies said.
Because the copper connections
used in chips today can become overloaded, degrading electronic signals,
another way some companies hope to advance Moore's Law is by shuttling
information around with beams of light.
In July, Intel said it had
developed a prototype fingernail-size "silicon photonic link,"
comprised of mini lasers, which could send 100 hours of digital music from
one device to another in a second. Although the Santa Clara company said a
commercial version wouldn't be available for several years, it eventually
expects to make a chip with the ability to transmit the Library of
Congress's entire printed collection in less than two minutes.
Infinera of Sunnyvale already
sells equipment containing photonic chips it developed to cable companies,
Internet content providers and others who transmit data across fiber-optic
networks. Referencing Moore's Law in its most recent annual report, the company
said it expects to double the data-carrying capacity of its photonic chips
every three years.
Such improvements are essential
given the enormous increases in information being passed around by
Infinera's customers, according to Gaylord Hart, marketing director for the
company's cable business. "It's being driven largely by increased
telecommunications and Internet access, and certainly video is the largest
bandwidth driver," he said.
Other companies exploring
photonics include Hewlett-Packard
(HPQ). But
the Palo Alto tech giant also is working on something called
"memristors" with South Korea's Hynix Semiconductor.
Because memristors function on
the principal that electrical resistance increases when current flows
through a device one way and decreases when it flows the opposite
direction, HP believes memristors could be turned into tiny electrical switches,
reducing the number of transistors traditionally used on chips, while
boosting the chip's processing power.
While other materials eventually
may be determined to be superior, HP already has successfully made
memristor switches from titanium dioxide on a layer of platinum and placed
it within a tiny grid of wires, said Stan Williams, an HP senior fellow and
director of the company's nanoelectronics research group.
By running current through a
part of the titanium dioxide where oxygen atoms were removed -- slightly
altering its resistance -- HP has caused the memristor to open and close
like a switch, a technical success that Williams said eventually might
enable memristor-populated grids to be programmed to store and process
information.
Noting that it may be possible
to replace a dozen transistors in certain types of circuits with a single
memristor no bigger than about 3 nanometers wide -- roughly nine atoms --
he said, "we can easily see factors-of-10 improvements in certain
types of chips" using such a switch.
But Williams cautioned that it
would be physically impossible to build anything smaller than an atom, even
with memristors. And while no one knows when the limit will be reached for
shrinking chips under Moore's Law, "we are very close," he said.
"There is not that much room to go."
CREATING MOORE'S LAW
Nearly a half-century ago, Intel co-founder Gordon Moore foresaw a future filled with remarkable devices made possible by cramming microchips with an increasing array of transistors, tiny components that amplify and switch electronic signals.
In a 1965 Electronics Magazine article, he first postulated that the number of chip transistors would roughly double every year, though in 1975 he revised that to every two years. So far, his prediction has largely proved correct. Intel's first microprocessor in 1971 contained 2,300 transistors. That number had risen to 275,000 by 1985, to 42 million by 2000, to 592 million by 2004 and to nearly 3 billion today.
CREATING MOORE'S LAW
Nearly a half-century ago, Intel co-founder Gordon Moore foresaw a future filled with remarkable devices made possible by cramming microchips with an increasing array of transistors, tiny components that amplify and switch electronic signals.
In a 1965 Electronics Magazine article, he first postulated that the number of chip transistors would roughly double every year, though in 1975 he revised that to every two years. So far, his prediction has largely proved correct. Intel's first microprocessor in 1971 contained 2,300 transistors. That number had risen to 275,000 by 1985, to 42 million by 2000, to 592 million by 2004 and to nearly 3 billion today.
|
|
No comments:
Post a Comment