ONE SMALL STEPThe next big advance in chip
design arrives one year early
With grand fanfare, the electronics giant Texas Instruments announced in May that it had perfected a process that can produce silicon microchips of far greater detail and complexity than any currently available. Newspapers widely marveled at the innovation; many pointed out that TI is the first to produce chips with features as small as 0.18 micron (millionths of a meter) wide. Some predicted that the microchips would launch a generation of wonderfully smart and compact contraptions.
Such reports were wrong on two counts, but correct on the third. TI was not first. Although that company has prototypes on hand and hopes to have a factory constructed by next year, IBM began shipping small quantities of equally detailed integrated circuits in May. And both TI's and IBM's processes create tiny transistors that are 0.25, not 0.18, micron in width. (The much misunderstood 0.18-micron measurement refers not to feature size but to the distance current must travel to switch a single transistor.) This long-expected advance is the logical next step beyond the 0.35-micron features that make up the Pentium Pro and PowerPC chips now on the market, but it was not anticipated to occur until 1997.
Hyperbole aside, the new semiconductors may indeed have a dramatic impact on computers over the next five years or so, for several reasons. First is power consumption. TI claims its devices run on as little as one volt-about one third the voltage required by Intel's Pentium. Such low-power chips could significantly extend battery life in portable gadgets. The second benefit is sheer size.
Whereas a Pentium Pro today spreads about 3.3 million transistors across four layers, the new processes draw smaller switches onto six layers, with even more layers to come in the near future. All told, TI estimates it can turn out chips with up to 125 million switches, a number that was expected to be beyond reach until about 2005, based on historical trends. Whether it can do so without also charging 10 times more for its chips than Intel does for the Pentium Pro remains to be seen.
Greater breadth and depth lead directly to the final advantage: speed. Smaller transistors always switch faster, but the real boost will come from combining into a single chip functions that previously required several processors. "A lot of speed is lost when you have to move signals between chips," observes G. Dan Hutcheson, an industry consultant at VLSI Research. In particular, he suggests, chip designers will want to combine memory with logic circuits in ways they never could before.
For all its promise, the advance also represents a threat to the computer industry. "As we push below 0.25 micron, the software tools available to design integrated circuits are not going to be able to keep up with the added complexity," Hutcheson warns. If manufacturers have to add dozens of engineers to produce each new design, chips will not remain cheap-and fast-evolving-for long. To head off what it calls a "productivity gap," the industry consortium SEMATECH awarded a multimillion-dollar contract to Synopsys in May for an advanced design system that can handle circuits of 0.25 micron and the next two or three smaller increments. Beyond that lie limits that will force chipmakers to look for great technological leaps rather than small, safe steps. [See "Technology and Economics in the Semiconductor Industry," by G. Dan Hutcheson and Jerry D. Hutcheson; Scientific American, January.] -W. Wayt Gibbs in San Francisco