A Law of Continuing Returns

If Moore’s Law -- the enduring axiom about increases in computing power -- held true in other industries, a New York-to-Paris flight that in 1978 cost $900 and took about seven hours would today cost about a penny and take less than a second.

The laws of physics make such lightning-fast travel all but impossible. But those laws have also permitted remarkable advances in technology, at a pace first predicted 40 years ago Tuesday by a young engineer at Fairchild Semiconductor.

In the decades since Electronics magazine published Gordon Moore’s article “Cramming More Components Onto Integrated Circuits,” Moore’s Law has evolved from an obscure theoretical postulation to the very foundation of Silicon Valley’s entrepreneurial ethos.

And although Moore’s Law has proved strikingly accurate for nearly half a century, the engineers who made silicon one of the most important ingredients in modern life are asking themselves: “What comes next?”

Simply put, Moore’s Law posits that the number of components on integrated circuits such as silicon computer processors will double every couple of years, while the cost per component declines at a commensurate rate.

It explains how computing gets cheaper even as it gets ever faster. By making transistors tinier and tinier, researchers have cut the distances that electricity must travel, boosting chip speeds.

“I wanted to get across the idea that integrated circuits were going to be a way to build things, and that things were going to get cheaper as they got more complex,” Moore, now 76, who co-founded chip giant Intel Corp. in 1968, said last month in an interview.

The squeezing of progressively more components onto silicon chips has enabled technology to pervade virtually every aspect of daily life, affecting the way people communicate, work, learn and relax.

“Moore’s Law is indelibly linked to the history of our industry and the economic benefits that it has provided over the years,” chip expert Dan Hutcheson of semiconductor consultancy VLSI Research wrote in a recent article. “The integrated circuit developed rapidly, leading to Moore’s observation that became known as a law -- and in turn, launched the information revolution.”

Moore’s maxim has weathered industry downturns, skepticism by experts and breathtaking technological advances. But it endures. Every 10 years or so, Moore himself predicts that his law has another 10 years or so before the technology on which it is based begins to hit its limits and the pace of advance slows.

Today many researchers peg the date as 15 years out. Already, though, they are stretching their brains -- and the boundaries of physics and chemistry -- to figure out what comes next and supplants the transistor-on- silicon computer chips built as the drivers of the Digital Age.

“Whether it’s carbon nanotubes, latches or something optical or DNA, there’s something out there, but it is really too early to call any winners,” said Fred Weber, chief technology officer of chip maker Advanced Micro Devices Inc. of Sunnyvale, Calif.

For ordinary computer users, PCs have long been plenty fast. They are unlikely to notice much difference between a chip that operates at 4.0 gigahertz instead of 2.0 GHz.

But for university computers doing complex mathematical research, or government systems crunching census data or simulating nuclear explosions, the need for speed and power grows ever greater. Researchers and companies also strive to make things ever smaller, hoping that someday the equivalent of a PC can be worn on the wrist, or that tiny robots can perform surgery inside blood vessels, and even pills can be equipped with circuitry to help with medical treatment.

The competition to achieve such performance is intense, as universities and companies around the world seek to reach markets for such inventions.

Understanding Moore’s Law, and its potential limits, first requires a basic understanding of how integrated circuits, or chips, work. The building blocks of a chip are circuits that turn on and off to produce electrical pulses that act as digital instructions.

Hundreds of millions of tiny transistors do most of the work in modern chips, switching current on and off millions of time a second. The more transistors there are, the faster and more powerful the chip.

The transistors are built onto ultra-thin layers of silicon wafers. Silicon, which is one of the most abundant elements on Earth, doesn’t conduct electricity. But when mixed with other elements, it can conduct a weak current -- hence the name semiconductor.

In his now-legendary article, buried back on Page 114 of the April 19, 1965, issue of Electronics, Moore figured that the number of transistors would about double every year. Few noticed at the time, but as the years passed, Moore’s optimistic observations proved surprisingly astute -- and the article became iconic for Silicon Valley’s optimism.

Intel, for instance, last week offered a $10,000 bounty for a mint copy of the magazine -- prompting a wave of thefts from libraries around the country.

The Santa Clara, Calif.-based company delivered its first processor in 1971; the chip, called the 4004, had 2,300 transistors. By 1978, the 8086 processor had 29,000 transistors. The number hit 1.2 million in 1989 with the Intel 486 processor. Last year’s Pentium 4 processor for PCs had 410 million, and the Itanium 2 processor for server computers, to be released this year, has a mind-spinning 1.7 billion transistors.

Engineers and chip designers recognize that as the number of transistors surpasses a billion on a single chip the size of a postage stamp, they are becoming so minuscule -- approaching the dimensions of molecules -- that it will be physically impossible to get transistors any smaller.

Chip engineers are confident they can build transistors, which today are as small as 90 nanometers across, or 90 billionths of a meter, as small as 22 nanometers.

Then transistor manufacturing gets tricky. Scientists say they can theoretically produce the devices at 10 nanometers or even 5 nanometers, but they will have to overcome formidable issues of overheating and power leakage. At less than 5 nanometers, those problems become insurmountable, and the techniques used to build transistors -- building up layers and dissolving away portions to produce a three-dimensional structure -- are no longer viable.

That’s why researchers at Intel, IBM Corp. and other chip makers need to come up with new technologies to take semiconductors beyond the transistor-on-silicon model -- known as CMOS, or complementary metal-oxide semiconductor, technology -- and, they hope, keep Moore’s Law alive.

At Intel -- by far the world’s largest maker of computer processors, with more than 80% of the world market -- scientists are evaluating many paths, but the one holding the most promise concerns what are called carbon nanotubes.

Scientists grow such nanotubes by causing a spark between two electrodes made of carbon, creating a cloud of carbon atoms that attach to themselves. The resulting lattice-like tube is something like a roll of chicken wire but far stronger and tinier -- 100,000 nanotubes could fit across the diameter of a human hair.

The nanotubes, only 1 nanometer wide, could function as on-off switches, replacing the much larger transistors.

Carbon nanotubes right now grow in a jumbled bundle, making it difficult for researchers to extract usable tubes. “But we try to make some order of it,” said Paolo Gargini, Intel’s director of technology strategy. The nanotubes need not necessarily be lined up in perfect parallel rows. As long as they touch what are called the source and the drain, they can conduct electricity. Intel is partnering with Stanford University for research and producing nanotubes in an Intel lab in Oregon.

“It will still take five to 10 years before they’re manufacturable,” Gargini said.

Intel also is looking at compounds made from elements other than silicon. Compounds such as gallium arsenide or indium antimonide can replace silicon and transistors, and can sometimes create faster switches while consuming less power. But they also present manufacturing difficulties that have not yet been overcome.

Research powerhouse IBM also is putting its faith in carbon nanotubes, which are unruly when produced and difficult to separate and align but have extraordinary properties in conducting electricity with low power consumption.

“The question is, can we get them to the purity we want, and place them where we want?” said Gian-Luca Bona, an IBM fellow and manager of the science and technology research division at IBM’s Almaden Research Center in San Jose, where the first carbon nanotubes were produced.

One way IBM is looking to bring nanotubes under control is with biological agents such as DNA, the building blocks of genes and chromosomes. Researchers at IBM and other institutions have been able to wrap the double helix structure of DNA neatly around carbon nanotubes in a candy-cane-like pattern, giving them rigidity for easier handling.

Enzymes could be used to cut the DNA, typically from simple bacterial organisms, into set lengths, much as biotechnology researchers use enzymes to snip and recombine DNA.

Bona calls it “bio-inspired engineering.”

Intel, IBM and other chip makers are keeping an eye on other potential technologies, including optical computing that manipulates photons, weightless packets of electromagnetic energy; harnessing the spin magnetism of electrons; and “quantum dots,” where electrons or photons are confined and controlled within minute three- dimensional spaces to create tiny transistors.

For his part, Moore said he was skeptical of the ability of quantum dots to replace the transistor.

“Can you connect a billion of them together?” he asked. But, he said, “I can see carbon nanotubes being incorporated.”

PC and printer maker Hewlett-Packard Co. has a healthy chip-producing business of its own.

“We manufacture four times as much silicon as Intel does,” said Stan Williams, HP Labs’ director of quantum science research. “Practically nobody knows that.”

Most of HP’s silicon goes into printer cartridges. But the Palo Alto-based company is developing its own technology, called “crossbar latches,” as a replacement for transistors. Crossbars are essentially a grid of platinum wires laid at right angles, with molecules at the junctures that can conduct electrical signals. Many such gates can be connected to perform computations.

This year HP will produce crossbars that are about 30 nanometers wide and spaced 100 nanometers apart, reduced from 40 nanometers wide and 125 nanometers apart last year -- an improvement that outpaces Moore’s Law, Williams notes.

Conceivably, crossbars could be made of nanotubes only 1 to 2 nanometers wide, shrinking the structure even further.

“We’re confident that the devices we’re working on will be the smallest electronic devices that can ever be made,” Williams said.

Although carbon nanotubes seem to be gaining some consensus support, “you really have to wait and let them settle for about a year, to see if there’s really a ‘there’ there,” said AMD’s Weber. “These trends are like the Internet: You’re overly excited about them initially, they all turn out to be harder than you thought, and some turn out be more important than you thought.”

Applied Materials Inc. of Santa Clara is the world’s largest maker of the equipment used to manufacture chips out of silicon wafers -- machines that can be as big as a school bus.

Mark Pinto, Applied Materials’ chief technology officer, sees the chips of the future as gradually adopting these esoteric technologies.

“The view of a lot of those things is that they’re not going to be replacements but could be useful elements to complement CMOS,” he said. “They may not be the solution but a piece of the solution.”

Humans have not hit on something that, simply put, is better than silicon.

“People talk about biology as a solution, and one appeal is the ability to replicate DNA,” Pinto said. “The problem is that it’s much less accurate than our silicon process. Silicon technology has a much higher level of reproducibility.”

Yoshio Nishi, a professor of electrical engineering at Stanford University and director of the Stanford Nanofabrication Facility, also sees a hybrid of technologies.

“My gut feeling tells me that the most likely [scenario] would be a combination of CMOS and either ... nanowire-based memory or carbon nanotubes,” said Nishi, a leading authority on silicon processes and new device structures. That, he added, is “assuming we will have some kind of engineering breakthrough which enables controlled growth of such materials.”

When he first came out with his prediction, Moore thought of it simply as a way to predict the near-term direction of his industry, one familiar to engineers but one consumers normally don’t see up close. But he says his theory can be applied to commercial commodities as well.

“Cellphones have been a phenomenal growth area, selling 600 million to 700 million cellphones a year ... one for every 10 people on Earth,” he said.

Moore’s Law “has moved on to apply to anything that changes exponentially, and I’m happy to take credit for all of it,” he quipped in his recent conversation with reporters.

And it could still evolve -- “maybe it will slow down, to a doubling every three years,” he said.

Meanwhile, chips have implicated themselves into every aspect of daily life, whether one is withdrawing money from the bank, checking in for an airline flight, starting the car to go pick up some bread or scanning that bread at the checkout counter. Chips are embedded in refrigerators and laundry machines, in jewelry that lights up and in talking dolls -- industry wags like to point out that the typical PC today has more computing power than the Apollo spacecraft that went to the moon.

Last year more transistors were produced, and at a lower cost, than grains of rice, according to the Semiconductor Industry Assn. Moore estimates that the number of transistors shipped in 2003 was 10 quintillion, or 10 to the 18th power -- about 100 times the number of ants estimated to be stalking the planet.

Some experts believe that even if Moore’s Law as it applies to transistors comes to an end, it will live on and be carried over to whatever technology is behind computing in the future. Moore, not surprisingly, is among them.

“No exponential lasts forever,” he said a few years ago. “But forever can be postponed.”