12/09/96 THE SILICON AGE? IT'S JUST DAWNING

View items related to this story

THE SILICON AGE? IT'S JUST DAWNING
Happy 25th, Mr. Chips. My, how you've grown
Few inventions have changed the world as profoundly--and rapidly--as the microprocessor. The first brain chip was born just 25 years ago. Intel Corp. announced its arrival in a November, 1971, trade-magazine ad that heralded ''a new era in integrated electronics.'' But even Intel didn't anticipate the scope of the revolution it was unleashing on business and society.
Today, the world's chip population has swollen to 350 billion, including 15 billion microprocessors. That's more than two silicon brains for every person on earth. In the U.S. and other industrial countries, most homes--those without personal computers, that is--are filled with a couple of dozen microprocessors. They're in TV sets, wristwatches, cameras, kitchen appliances, vacuum cleaners, and just about everything else that plugs into the wall. Cars typically have at least 10 microprocessors, and the new Mercedes-Benz S-Class models have 50.
The little silicon rascals have certainly made good on Intel's proclamation. But all that has happened so far is merely the prologue for what's coming. At the electronics industry's recent Comdex extravaganza in November, Intel CEO Andrew S. Grove asserted that ''we are only at the beginning of this revolution-in-progress.'' His view is shared by executives at IBM, Texas Instruments, Zilog, Hitachi, NEC, and other microprocessor producers. ''You ain't seen nothin' yet,'' says R. Gary Daniels, general manager of Motorola Inc.'s Microcontroller Technologies Group in Austin, Tex. ''It's an electronics explosion.''
Ever since Intel's first microprocessor, the 4004, these chips have grown increasingly powerful in periodic leaps and bounds (table, page 150). Grove predicts that this inexorable march will continue for at least 15 more years, perhaps 30. By 2011, he envisions microprocessors with a billion transistors that will chew through 100,000 MIPS, or millions of instructions per second. Today's fastest Pentium Pro chips boast 5.5 million transistors and speeds of 400 MIPS. So that future chip will be crammed with the power of 250 Pentium Pros.
OBVIOUS TARGET. To boggle the mind further, there will be more microprocessors produced with each passing year. Combine the gains in processing speed with the growth in the number of chips, and the result is astounding: Every two years or so, the world's computing power doubles. So between now and 1999, chipmakers will churn out as much number-crunching power as the sum of what now exists. And a year or two into the next century, that total will double. The implications of such compound growth are staggering. Federico Faggin, a member of the Intel team that invented the microprocessor and now CEO of Synaptics Inc., believes one result will be autonomously intelligent machines specializing in certain jobs. You'll just tell them what to do, and they'll figure out how to do it. ''That will be a new dawn,'' says Faggin.
For chipmakers itching for a piece of this action, Intel is the obvious target. It is the world's biggest and richest chip company, with a valuation that's approaching $100 billion, or close to the combined worth of the Big Three auto makers. And the bigger Intel becomes, the more competition it seems to attract. However, money isn't the only thing that will drive more chipmakers to dive into the microprocessor business.
Chips are becoming almost too much of a good thing. As transistor counts climb into the tens of millions per chip, as is now starting to happen, entire products will ''collapse'' onto one sliver of silicon. Except for PCs, workstations, and other sophisticated systems, a single chip will be able to hold all the circuitry now scattered among multiple chips from various suppliers. And all those suppliers will want to be the survivor that makes the ''system on a chip.'' So they'll need a microprocessor circuit.
Even dynamic random-access memory chips (DRAMs) are getting too fat for their britches. Computers may always be able to use more memory, but the next generation of DRAMs will sport more transistors than most products need just for storing data. So DRAM makers are developing hybrid chips with microprocessor modules to broaden their business. Market watcher Dataquest Inc. says at least a dozen chipmakers are working on so-called application-specific integrated circuits, or ASICs, that combine DRAM and microprocessor-type circuits.
That includes the king of DRAMs, Samsung Semiconductors Inc. The Korean chipmaker in September unveiled an ASIC hybrid with 1 megabit of memory plus 60,000 logic gates. (A logic gate is the fundamental switch in a microprocessor circuit.) Because a logic gate consists of three or more transistors, Samsung's ASIC-DRAM is essentially an IBM PC/XT on a chip. Similar hybrids are offered by Toshiba, Texas Instruments, and IBM. Mitsubishi Electric Corp., which also has an ASIC-DRAM, is trading its DRAM savvy to Motorola for Motorola's latest microprocessor technology, called ColdFire--and has formed a Silicon Valley startup, Vsis Inc., to focus on system-chips. ''We expect to embed ColdFire into memory and create a system-on-a-chip,'' says Stephen D. Hester, executive vice-president at Vsis.
DRASTIC SHIFT. The Motorola-Mitsubishi collaboration points up a practical problem with system-chips: Even semiconductor giants will be hard-pressed to go it alone. When system-chips climb past 100 million transistors, they will be monstrously complex and costly to develop. No computer-aided design system that exists now could cope with so much detail. Already, chipmaking is the world's most expensive real estate speculation, says Intel Chairman Gordon E. Moore. Turning wafers into microprocessors costs $1 billion per acre of silicon.
Next year, IBM, Texas Instruments Inc., and others plan to shrink the width of transistors to an infinitesimal 0.25 microns, and by decade's end, 0.18 microns. By comparison, bacteria are about 8 microns wide and a human hair 10 times fatter than that. The equipment for etching silicon with millions of such minuscule devices steadily mounts in cost as the transistors shrivel. Today, a chip factory for printing 0.35-micron transistors costs $2.5 billion. The tab for a wafer-fabrication plant for superchips with 0.07-micron transistors could run $10 billion.
For the vast majority of chipmakers, the ante for staying in the game could soon be prohibitive. J.P. Dauvin, SGS-Thomson Microelectronics chief economist, reckons that a decade from now, only 9 or 10 chipmakers will be able to afford new factories. That's why 35 chipmakers, suppliers of semiconductor design systems, and users of chips recently founded the Virtual Socket Interface Alliance (VSIA). The goal: to develop standards that will let today's chips be converted into circuit modules that can be mixed and matched on future system- chips. If all goes as intended, designers will be able to combine, say, a microprocessor from Digital Equipment with a graphics accelerator from S3 and DRAM circuitry from NEC. And customers could put the finished design up for bids among participating chipmakers to get the lowest cost of production.
This would be a drastic shift from the current business model--each chipmaker using proprietary technology to lock in customers. Yet VSIA seems to be luring even the heavyweights--one notable exception being Intel. VSIA's membership has more than doubled, to 79 companies from Japan, Korea, Europe, and the U.S., since its start in September. Among its chipmakers: Cypress Semiconductor, Cirrus Logic, DSP Group, Fujitsu, LG Semicon, Matsushita, National Semiconductor, Philips, Samsung, SGS-Thomson, TI, and VLSI Technology.
Intel should be able to stay aloof as long as there are PCs. But the spread of ultrasmart system-chips into all types of consumer products could begin pinching demand for PCs--especially if the so-called Net Computer takes off. This scaled-down PC would be linked to the Internet and run software stored on big, remote computers. Thus, Net PCs wouldn't need Intel chips. Similarly, cellular phones with built-in mainframes might take over many tasks now handled by Intel-compatible PCs. And Sony Corp. and other Japanese consumer-electronics giants are salivating at the prospects of smart televisions and settop boxes.
Even if Intel comes through unscathed, its chips account for only 1.5% of all microprocessors. The likes of Motorola, Hitachi, and Zilog sell far more than Intel does, in unit terms, though many of their chips are much cheaper--some costing $5 or less, a fraction of what Intel gets for its PC chips (table, page 150). But just like Intel's silicon brains, all those other microprocessors will grow smarter as transistors shrink in size. Tomorrow, even your alarm clock may have enough smarts to download local traffic reports and adjust the time of its wake-up call to road conditions. So the number of PCs may be a poor barometer of how sweeping the microprocessor revolution will be. Says Motorola's Daniels: ''After the turn of the century, everything you touch will have a chip in it.'' And almost as many transistors as stars in the sky.
By Otis Port in New York, with Andy Reinhardt in San Francisco, Gary McWilliams in Houston, and Steven V. Brull in Tokyo

RELATED ITEMS

TABLE: March of the Microprocessor

WHOSE BRAINCHILD WAS THE BRAIN CHIP?

PHOTO: Bell Lab Researchers, 1948
Return to top of story