Chip Advances Surge Ahead
Computer progress follows Moore's Law, but the limits may be in sight on a distant horizon
PITTSBURGH — IF there's one thing you can count on in the computer industry, it's that next year's product will be smaller, faster, and cheaper.
This rapid development is so well established it even has a name: Moore's Law. Every year and a half, says Gordon Moore, chairman of the Intel Corporation, the computer industry doubles the number of transistors it can put on a chip. It took about three years to move from producing 4-megabit chips (4 million bits) to today's high-end 16-megabit chips. If Moore's Law holds true, the 64-megabit and 256-megabit chips now in labs will follow the same curve.
Moore's Law has tremendous implications for computer development. If the industry can pack four times as many transistors on a single chip, it can produce smaller, faster machines. Until now, prices have followed a similar curve. Over time, 1-megabit (1 million bits) chips have become less expensive than four 256-kilobit chips (256,000 bits each); now 4-megabit chips are starting to undercut 1-megabit chips.
But the economic benefits of Moore's Law are starting to look less favorable.
In March, Mr. Moore told the American Physical Society that the industry was running into economic limits. It is becoming increasingly expensive to build the factories that make advanced chips. So expensive, in fact, that it will be difficult for any one company to build the next generation of semiconductors. He called it "a major change in the industry's dynamics."
A big reason is equipment costs. A factory that can turn out 5,000 chips a week requires $1.5 billion worth of equipment. Add another $500,000 to construct a building with the necessary advanced clean rooms. "Not too many companies can afford $2-billion factories," Moore says.
Other chip manufacturers agree. "What Gordon's looking at is [that] his wafers are getting more complicated and expensive, and that curve he likes to draw ... is getting dinged up," says T. J. Rodgers, president and chief executive officer of Cypress Semiconductor Corporation in San Jose, Calif.
"Economically, we're not sure any of this makes sense beyond even 16 megabits," adds Robert Dennard, an International Business Machines fellow at the Watson Research Center in Yorktown Heights, N.Y. "You have to make near-perfect manufacturing facilities, elaborate clean-room procedures.... Even though you get a denser memory cell, the cost of doing it may not be such that you get a cheaper memory cell."
But technologically, the limits are still some years off.
"I think the discussion Gordon Moore has initiated will start to really take hold in the year 2000," Mr. Rodgers adds. "I don't see for the next decade any slowdown."
The changes wrought by Moore's Law are revolutionary.
"The average person wears more computing power on their wrist today than all computing power combined before 1956," says Paul Saffo, research fellow at the Institute for the Future in Menlo Park, Calif.
That may be an exaggeration, Dr. Dennard says. But it's clear that more computing power can fit in a pocket than existed worldwide in the early 1950s.
Dennard is the inventor of the 1-transistor memory cell. When he was designing the 1-transistor chip in the late 1960s, he envisioned a device where each bit took up about 5,000 square microns. (A bit is a single unit of binary information; a micron is one millionth of a meter.)
By the time the first 4-kilobit memory chips came out, they contained one bit every 1,000 square microns. By contrast, today's advanced 16-megabit chips contain one bit every four square microns. That makes today's technology some 250 times denser than the devices of the early 1970s.
As computers have gotten smaller and faster, they have also gotten cheaper. The 10 million transistors in today's desktop computer cost substantially less than 10 million staples. Many scientists wonder whether the price declines will continue. But perhaps that's not so bad, some technologists say, since the world is still trying to catch up with the advances already made.
"You are building something that is more powerful and costs less," says Gordon Bell, former research director at Digital Equipment Corporation in Maynard, Mass. "You essentially eliminate a whole chunk of the economy."
In the short run, technological advances can indeed create economic havoc. In the past year, the United States economy grew by 2.6 percent, while more than 500,000 clerical and technical positions disappeared.
"People have worried about this problem of technological unemployment for a long time," says Frank Lichtenberg, professor of finance and economics at Columbia Business School in New York City. "That tends to be more of a short-term adjustment problem." In the long run, technological advancements mean not only more jobs but better jobs as well, economists say.
"People who use computers at work ... earn about 15 percent higher income than they otherwise would," says Alan Krueger, professor of economics at Princeton University in Princeton, N.J.
Even if semiconductor manufacturers conquer the economic challenge and find ways to make tomorrow's chips cheaper than today's, they still face physical limits that threaten to make Moore's Law obsolete.
For example, connecting ever-more-minute elements could become physically impossible. Chip designers will have to reduce the voltage of the devices (so the electric field across the device doesn't get too big), but there are limits to how low they can go. Today's manufacturers are already moving their systems from 5-volt to 3.3-volt technology. Dennard says the designers can get down to somewhere around 1 or 1.5 volts before the voltage is so low that the transistors won't turn on and off cleanly.
More density also means more waste heat. (See related story, Page 12.) And declining size means that the insulators between the electronic gates and the underlying channels of transmission will get so small that they will reach the limits of quantum mechanics, Dennard says.
What will happen when the industry runs into a brick wall? It will slow down, these experts say, and that might be all right.
"Moore's Law may be hitting the edge, but it doesn't matter, because microprocessors aren't the driving force" behind the direction advances take anymore, Mr. Saffo says. He says laser-driven communications - everything from fiber-optic phones to multimedia - will allow computers to handle data, voice, and video in more interesting and complicated ways. The real achievement of the computer revolution has already been achieved, he adds. Today's transistors are cheap enough to create disposable computers.
"The computer technology that's available to the world at the prices that they're available for is already pretty extraordinary," Dennard says. "Computers are pretty cheap today and you have 64 times more power waiting in the wings."
Moore's Law will still pack its punch - at least for a few more years.