Moore hits hurdlesMoore hits hurdles

Santa Clara, CA-As the dot-com bubble bursts, investors' soaring expectations for the "new economy" are coming back to Earth. Wary of the promises of software and Internet companies, they are concentrating more on hardware, storage, and processing.

This is good news for chip makers, but it comes at an awkward time. Since 1965, the microprocessor industry has lived by Gordon Moore's famous dictum that the number of transistors on computer chips will double every 18 to 24 months. And indeed, there were 2,300 transistors on Intel's 4004 in 1971, 7.5 million on the Pentium II in 1997, and 42 million on today's Pentium IV.

But even the most optimistic engineers agree this pace cannot continue forever. The question is exactly when the industry will hit the wall. Now recent announcements from engineers at IBM, Intel, and HP may buy a little more time for Moore.

In June, Intel announced they could make a transistor just 20 nm (20 billionths of a meter, or 0.02 microns) in size, calling it "the world's fastest silicon transistor." This advance could enable production of a billion-transistor processor by 2007, says Rob Willoner, an Intel market analyst. That's like fitting 24 Pentium IVs on a single chip. And it's not just small, but it's fast. At 20 nm, transistors can switch faster than 1 picosec (trillionth of a second), so such a chip would likely operate at 20 GHz, compared to the Pentium IV's 1.5 and 1.7 GHz models available today.

Willoner cites the advance as Intel's entrance into nanotechnology, which is generally defined as building structures smaller than 30 nm. And the technique could "push Moore's Law through the end of the decade."

Another major advance came with IBM's April announcement that it had built an array of transistors out of carbon nanotubes, cylinders of carbon atoms as small as 10 atoms across. They expect to see the full benefits of this technology after silicon-based chips cannot be made any smaller, a barrier they predict is 10 to 20 years away.

IBM perfected a process for mass-producing nanotube transistors, instead of painstakingly building them in the lab. The tubes form in bundles of metallic and semiconducting types, but only the latter can be used as transistors. IBM engineers discovered how to use a burst of electricity to destroy the metallic nanotubes.

Finally, researchers at Hewlett-Packard and UCLA won a patent in July for making chips from individual molecules instead of semiconductors. They plan to build a 10 Kbyte chip by 2005.

Still, there are physical conditions that may limit the progression of Moore's Law.

The current challenge of chip design is not so much lack of processing power as lack of efficiency. Portable electronics like PDAs and cell phones must still carry batteries as large as the appliances themselves. Another challenge is the perennial problem of electronics overheating when they get too small. Reducing voltage helps, as it shrinks power demand and boosts reliability, but it also slows down transistors, Willoner says. So with every new chip design, Intel must find a balance point.

Another physical hurdle for Moore is that the gates in Intel's 20 nm transistors are just three atomic layers thick. If they get any thinner, electric current will leak between them, making them useless.

For more information about chips from Intel: Enter 533

For more information about chips from IBM: Enter 534