February 10, 2003 2:27 PM PST
Moore's Law to roll on for another decade
"Another decade is probably straightforward," Moore said, speaking at the International Solid-States Circuits Conference. "There is certainly no end to creativity." The conference, a gathering of top semiconductor researchers organized by an IEEE group, takes place in San Francisco this week.
Moore's Law--which states that the number of transistors on a given chip can be doubled every two years--has been the guiding principle of progress in electronics and computing since Moore first formulated the famous dictum in 1965. And, for the same amount of time, people have predicted it would hit a wall.
So far it hasn't, meaning chips and computers have become simultaneously more powerful and less expensive. The number of transistors produced annually is now roughly equal to the number of letters and/or characters printed annually--and they cost about the same to produce, Moore noted. The amount of transistors produced each year outnumbers the worldwide ant population by 10 to 100 times.In some markets, "you can get 50 million transistors for a buck these days," Moore said. In the late 1950s, some chips had 200 transistors; by 2005, Intel will produce chips with 1 billion transistors. Semiconductor industry revenue has grown 800-fold since the late '50s.
Still, future progress will be difficult and likely to some degree slower, Moore noted. Chips now require vast amounts of electricity, a growing portion of which is dissipated through leakage. Designers are going to have to add technologies such as strained silicon to their chips and to redesign transistors to control energy consumption.
"I don't want a kilowatt in my laptop," Moore said.
Intel, like other companies, is using the ISSCC conference to promote ideas for reducing power consumption. One proposed solution is to create subsections in chips, dedicated to specific tasks such as handling network protocols or multimedia calculations, said Shekhar Borkar, an Intel research fellow.
"General-purpose hardware is generally not power-efficient," Borkar said.
Progress in lithography, the science of "drawing" circuits on chips, will also have to be made. Extreme Ultraviolet (EUV) lithography uses light with much smaller wavelengths and will start to come onto the market in 2007. Perfecting lithographic techniques, though, takes time. Air, Moore noted, can absorb EUV light and cause errors.
The tension between the optimists and the pessimists concerning Moore's Law largely seems to revolve around whether an individual sides with engineers or with the laws of physics. Physics dictates that transistors can only get so small. (Shrinking transistors is the principal mechanism for doubling the population on a piece of silicon.) Engineers, though, have typically found workarounds.
"I remember thinking 1 micron (a milestone the industry blew past in 1986) was as far as we could go" because of the wavelength of visible light used at the time, Moore said. Engineers then switched to ultraviolet light.
Progress was then supposed to stop at 0.25 microns, but it didn't. Now, the lines drawn on chips are smaller than the wavelengths of light used to draw them, which Moore likened to "breaking the laws of physics."
Someday progress will stop. "No physical quantity can continue to change exponentially forever," Moore noted. Nonetheless, none of the current looming problems represents a wall, he added.
Moore also managed to dispel some of the misconceptions about his dictum. For one thing, he didn?t name it. California Institute of Technology professor Carver Mead did after seeing Moore's original paper on the potential growth of the transistor.
Moore also affirmed he never said transistor count would double every 18 months, as is commonly said. Initially, he said transistors on a chip would double every year. He then recalibrated it to every two years in 1975. David House, an Intel executive at the time, noted that the changes would cause computer performance to double every 18 months.
House actually came close. Computer power is doubling around every 20 months. Nonetheless, "House said 18 months, not me," Moore said.
The observation has also proved much more resilient than Moore thought. "I never expected it to be precise. It turned out to be much more precise than I ever imagined," he said.
Not all his observations, though, have weathered the test of time. At the same time in the early '60s, Moore also admits he predicted that wafers today, the round silicon sheets that chips get popped out of, would measure 57 inches in diameter. Twelve-inch wafers are just coming into the market.
Moore admits that he failed to predict that engineers could continue to pile layers of circuits on top of each other.
1 commentJoin the conversation! Add your comment