May 2000

The End of Moore's Law?

Continued from page 1

By Charles C. Mann

Transistors, diodes, capacitors and other now-familiar electronic devices had just been invented, but already some far-sighted people, many of them in the Pentagon, were thinking about lashing together these individual components into more complex circuits. Texas Instruments was trying to hook up with the Army's Micro-Module program, in which individual components were built on small wafers and stacked like so many poker chips. Kilby thought this approach was ludicrous-a kludge, in engineer's slang. By the time a module was large enough to do something interesting, the stack of wafers would be ridiculously big and cumbersome.

On July 24-exactly a month after Bell Labs celebrated the 10th anniversary of the public unveiling of the transistor-inspiration paid a visit to Kilby in the empty factory. Instead of wiring together components in modules, he wrote in his lab notebook, engineers should scatter "resistors, capacitors and transistors & diodes on a single slice of silicon." Classic inventors' stories usually include a chapter about how management ignores the inventor's brilliant new idea. At Texas Instruments, Kilby's boss immediately asked him to build a prototype. By September, Kilby had assembled one. It was simple and crude, but it worked. The company filed for a patent on its revolutionary "Solid Circuit" in February 1959.

Two weeks before the filing, a similar idea occurred to Robert Noyce, an engineer at Fairchild Semiconductor, one of the first startup tech firms in Silicon Valley. (Noyce and Moore would later leave Fairchild to found Intel.) Whereas Kilby had linked the components of his integrated circuit by gold wires and solder, Noyce realized the connections could be painted on the silicon with a kind of stencil-a photomicrolithograph, to be precise. Noyce's bosses, like Kilby's, were enthusiastic. And in July Fairchild, too, filed for a patent.

Litigation inevitably ensued. It lasted 10 years and ended with the companies fighting to a draw. But while the lawyers argued, both companies raced to create ever-more-sophisticated integrated circuits-"chips," as they came to be called. The first chip appeared on the market in 1961, to less than universal acclaim; engineers, accustomed to designing their own circuits, initially regarded these prefabricated gizmos as annoyances. But the companies kept going. By 1964 some chips had as many as 32 transistors; when Moore wrote his article in 1965, a chip in his R&D lab had twice as many.

One component (1959), 32 (1964), 64 (1965)-Moore put these numbers on a graph and connected the dots with a line. "The complexity [of cheap integrated circuits] has increased at a rate of roughly a factor of two per year," he wrote. Then he got out a ruler and extended the line into the future. It sailed off the top of his graph and into the stratosphere. "Over the longer term...," Moore argued, "there is no reason to believe [the rate of increase] will not remain constant for at least 10 years." In other words, the companies that were then laboring to create microchips with 64 components would in a decade be manufacturing microchips with over 65,000 components-a jump of more than three orders of magnitude.

Moore's Law was not, of course, a law of nature. It was more like an engineer's rule of thumb, capturing the pattern Moore had discerned in the early data on microchip production. But law or no, by 1975 engineers were designing and manufacturing chips a thousand times more complex than had been possible just 10 years before-just as Moore had predicted. That year, Moore revisited his prediction at the annual International Electron Devices Meeting of the Institute of Electrical and Electronics Engineers, the professional association of electrical engineers. Acknowledging the increasing difficulty of the chip-making process, Moore slightly revised his "law." From that point on, he said, the number of devices on a chip would double every two years. This prediction proved correct, too. Today, some people split the difference and say that microchip complexity will double every 18 months; other people loosely apply the term "Moore's Law" to any rapidly improving aspect of computing, such as memory storage or bandwidth.