Insiders at Advanced Micro Devices Inc. have known since last summer that Intel Corp.'s monopoly in 386 chips, the world's best-selling microprocessor, was over. The break came on Aug. 10. That's when AMD's hush-hush design team, dubbed the Cave Dwellers because they labored in a windowless laboratory nicknamed The Cave, rushed out to revel under the Texas sun. Their competing chip, which was formally unveiled in late March as the AM386, had passed its initial tests with flying colors.
The news was flashed to W. J. Sanders III, chief executive of the Sunnyvale (Calif.) chipmaker. He tracked the celebrating engineers to the Salt Lick, a barbecue joint in the Texas hill country town of Dripping Springs, 25 miles from AMD's Austin plant. Summoned to the phone, team leader A. Benjamin Oliver got Sanders' tongue-in-cheek congratulations: "What the hell are you doing out drinking beer at 2 p.m. on a Friday?"
Champagne would have been more fitting: The story of the AM386 is one of beating extremely tough odds to clone Intel's hottest product. Even among AMD's engineers, "only Ben thought it was feasible," says Geoffrey R. Tate, a former AMD executive vice-president who now heads a semiconductor startup. Indeed, pulling it off took two years. But now, Sanders is betting that his 386 clone, which is faster and consumes less power than Intel's best design, will rake in revenues of some $100 million in its first full year. That should go a long way toward returning the company to profitability after four losses in the past six years, including a $53.6 million deficit in 1990 on sales of $ 1.1 billion.
INTO THE BREACH. The tale of the new chip dates to 1982, when AMD won a license as a "second source" of Intel microprocessors. But the smaller company ultimately ran past Intel: It has 52% of total sales for the 286 chip Intel developed, to Intel's 33%. So in 1987, as the microprocessor king readied its new 386, it had second thoughts. Intel decided not to license the 386 or any subsequent design for the "brains" of IBM-type personal computers. AMD sued to force Intel to reverse that stance, a legal contest that continues today. Without a 386 entry, AMD knew it was destined for trouble: A new generation of microprocessors appears every three years or so, and the 286 would be slipping by the early 1990s.
To hedge against losing the court battle, AMD looked for a fallback. Tate opted for a clone, and he drafted Oliver, who was then 36, to create it.
Oliver had joined AMD in 1985 after nine years of designing chips for Texas Instruments Inc. Having seen TI's calculators evolve into shirt-pocket portables, he decided to tailor his PC chip for the then-nascent portable computer market. By February, 1989, he had assembled a dozen engineers who were willing to reinvent the 386. That was crucial, notes Oliver, because many chip designers have egos too big to be satisfied by copycat work. AMD clamped a lid on the project, worrying that if Intel found out, there would be no chance ever to win a 386 license. Thus was born The Cave.
AMD did have one edge: A 1976 cross-licensing deal that gives it access to Intel's patents. This meant the team could risk a "reverse-engineering" approach--taking the Intel design apart bit by bit. The half-dozen other companies trying to build a 386 clone--including Chips & Technologies, Cyrix, Meridian Semiconductor, and NexGen Microsystems--worry about inadvertently violating Intel's patents. So, they're trying to create chips from scratch, borrowing only publicly divulged specifications.
CHIP SHOTS. To start, the Cave Dwellers "shaved" an Intel chip, whittling off successive layers just a few atoms thick. Each slice would reveal new details of the 386's intricate circuitry, which consists of tiny wires linking 275,000 transistors stacked in several tiers. As each layer came off, the team snapped photos of every square millimeter through a 1,000-power microscope. During several months in 1989, a local supermarket took in tens of thousands of dollars for developing these photos.
By studying mosaics of photos, the engineers plotted the location of all 275,000 transistors and traced the gossamer web of connections among them. Then, two teams independently recreated the circuitry on computer-aided engineering systems. During this tedious process, a computer continually compared the two versions, catching hundreds of errors. When a simulation of the new chip's final design was tested, several more glitches turned up--identical mistakes made by both teams.
Meanwhile, other engineers were hunting for ways to improve on Intel's original design. They moved transistors that exchange signals frequently closer to each other. They also spotted shortcuts Intel had taken to save transistors; by adding 6,000 and using different technology, their design draws less power and "remembers" even when the power is turned off. AMD's design also runs faster, processing signals 40 million times a second (40 megahertz), vs. Intel's fastest "heartbeat" of 33 Mhz.
UP AND RUNNING. Last July 30, AMD began building a few prototype chips. Just 11 days later, Oliver plugged one into a PC in place of an Intel 386, crossed his fingers, turned on the computer--and started running programs. Although it was three months ahead of schedule, the chip performed without a hitch.
Can AMD capture a profitable chunk of Intel's business? Intel Vice-President Paul S. Otellini isn't ready to concede that the chips are 100% compatible, despite months of testing by AMD, potential customers, and independent labs. He notes that months after Intel's own 386 was first shipped, a flaw was found. Beyond that, Intel has already introduced its own power-saving version of the 386 for portables, though it is slower than AMD's. And on Apr. 22, Intel plans to unveil a low-cost version of its latest and most powerful 486 chip, according to Michael Slater, editor of Microprocessor Report. A cheap 486 could touch off a price war in the 386 market, he adds, and pare profits to the bone.
Even so, Sanders is counting on the brisk laptop market to lap up the AM386. He predicts that the chip will dominate the portable PC market -- and that within a year, all top 20 PC makers will be using it. That remains to be seen. Executives at Japan's leading laptop makers express reservations. And Kazuhiko Nishi, president of Ascii, Japan's largest PC software company, declares that "no Japanese company will go with the AMD chip." Still, he concedes, Korean and Taiwanese companies probably will. And in the U. S., two computer companies plan to ship desktops with AMD's chip this summer.
In the meantime, AMD faces more legal challenges. It claims that its cross-license agreement with Intel gives it the right to use the software, called microcode, that Intel builds into its chips. Microcode is what lets a microprocessor "understand" programs such as spreadsheets. Intel has sued, charging that AMD can use this microcode only internally, not resell it in chips beyond the 286. That case should go to trial this summer. If AMD loses, Oliver hints, he may reverse-engineer the microcode, too: "We'll do whatever it takes to stay in this business," he adds.
Of course, the lifespan of the 386, like the 286, won't be very long. That product will be running out of steam by 1993. Can AMD really expect to clone the 486, a chip with 1.2 million transistors? "Why not?" asks Oliver, confident as ever. His Cave Dwellers--now tripled in number, to about 40--are back in their lair. And they're busy taping together lots of new pictures.