Intel's Development of 386 Computer Chip Took $100 Million

Intel's Development of 386 Computer Chip Took $100 Million and Four Years of Difficult Work

By Brenton R. Schlender, Staff Reporter
The Wall Street Journal

August 29, 1986

Santa Clara, Calif. -- One Friday night last August, two young engineers drove 40 miles from Intel Corp.'s laboratories here to its plant in Livermore to pick up prototypes of a device that may well change the entire computer industry.

"I'll never forget it," says Pat Gelsinger, who went along on the drive with another computer-chip designer, Joseph "Chip" Krauskopf. "We had the top down on Chip's convertible, playing an old Rolling Stones tape full blast all the way there. But coming back to Santa Clara, we put the top up, turned the stereo off and were almost afraid to breathe because these were the company jewels we were carrying."

Those "jewels" were the first experimental samples of one of the most powerful microprocessor chips ever developed -- the Intel 80386, or 386 for short. The product of $100 million and four years of effort, the 386 crams more than 275,000 transistors and a million electronic components onto a quarter-inch square of smelted sand.

Though vital to Intel's future, the 386's significance is much broader. As the blueblooded scion of the microprocessor at the heart of International Business Machines Corp.'s PC personal computers, the 386 is the core around which IBM and some 150 other computer makers are designing the next generation of desktop computers.

Those new machines, which will begin hitting the market next month, will cost $8,000 to $20,000 each. But they will enable people to run on their own computers elaborate business simulations and engineering and scientific programs that heretofore required big mainframe computers costing hundreds of thousands of dollars. What's more, they will be able to run any of the thousands of popular programs written for IBM PCs and do it faster.

"I'm convinced the 386 represents the biggest shift in the industry since the introduction of the IBM PC itself," says William H. Gates III, the chairman of Microsoft Corp., the company writing many of the computer languages and software systems that the chip will use. Adds Michael Murphy, the publisher of the California Technology Stock Letter: "The 386-based machines will finally give desktop computers all the power that most people will ever, ever need."

However, the 386 is making its debut amid sluggish demand for higher-performance computers. Early sales of the chip could prove disappointing because fully equipped 386-based machines initially will be priced too high for the mass market. Moreover, the software needed to fully utilize 386-based computers probably won't be available for two years.

Nonetheless, the 386's birth demonstrates how a watershed high-tech product is created as much by old-fashioned business, marketing and management methods as by pure science. Indeed, Intel's youthful team of engineers and salesmen struggled to balance customers' wishes with their own technical and profitability goals. Yet the 386 people produced, ahead of schedule, a chip that worked on the very first try despite its unrivaled complexity.

Intel started planning the 386 in March 1982. John Crawford, then 29 years old, launched the project. His job was to come up with an "architectural sketch" outlining the basic functions of a faster, more powerful chip that could run programs written for existing IBM PC-type computers.

Intel had learned the importance of compatibility the hard way. In early 1981, the company unveiled a fancy microprocessor called the 432 that also would put mainframe power in a chip. Engineers bragged that the 432 was so advanced because it was "unfettered by compatibility" with earlier microprocessors. But Intel's customers wanted to keep using earlier products, and the technological marvel was a business flop.

In another break with past practices, Intel also decided to include customers in its early planning for the 386. For the next six months, Mr. Crawford and a marketing specialist crisscrossed the country canvassing customers.

"Designing a successful microprocessor used to be a little like throwing a dart at the wall and then drawing a circle around it and calling it a bulls-eye," says George Alexy, the 37-year-old marketing chief for the microprocessor division. "But in the very beginning of the 386 product cycle, we were able to get a pretty good idea from our customers what the market wanted."

In October 1982, Mr. Crawford finished a 180-page document combining his ideas with those of the customers. Called a software reference manual, it described how programmers could use the 386. Because the manual could tip off competitors, it was guarded like a state secret. Only selected customers were shown copies.

During the next 18 months, the design team grew to about 30 members (and eventually 40) as it honed Mr. Crawford's basic chip "architecture" into a full-blown, 600-page "microarchitecture" that described exactly the chip's eight functional components.

Meanwhile, Mr. Alexy and his marketing team were trying to persuade customers not to buy the Motorola Inc. 68000, a competing chip introduced in 1983 that drives the Apple Computer Inc. Macintosh computer as well as computer-aided design machines made by several other companies. The 386's big advantage over the Motorola chip, which is nearly as powerful, is its ability to run without modification thousands of programs written for existing IBM PC-compatible machines -- something Intel marketing people think made the 386 worth the wait.

Apparently their efforts worked; several customers put off new-product development to await the 386.

"We perfected the striptease method of marketing," Mr. Alexy says. "Each visit, we titillated them with a few more details."

As the design work grew more detailed, "we were managing what amounted to eight highly technical projects in parallel, hoping they all would come together at the same time," says Gene Hill, the 39-year-old 386 project manager. Mr. Hill also was counting on breakthroughs in manufacturing technology. And another problem was that much of the equipment for designing microprocessors wasn't advanced enough for full simulation and testing of the 386.

Intel either could slow down the 386 work to give manufacturers time to design more powerful equipment or could make do with existing gear on portions of the chip, gambling that the whole chip would function when completed. With the Motorola 68000 already in production, Intel chose to press ahead.

By late 1984, most of the circuits had been designed, and the schematic drawings had to be translated into actual circuits that fit into the chip. To do that, specialists called mask designers use large computers to draw the millions of intricate lines representing the circuitry in each of the chip's 10 layers.

These drawings later are converted into minutely detailed photographic negatives -- masks -- that ultimately are used to inscribe each of the 10 layers of circuits on silicon wafers. (The masks are copyrighted as works of art to protect the chip design from being legally replicated by rivals.) Although the lines loom large on the computer screens, the metal connections themselves are measured in millionths of an inch on the masks and the silicon. A single misplaced connection can ruin the whole chip.

As the design work neared completion last summer, the pace of work picked up. Intel set a deadline by appointing people to plan a 386 product-introduction party last October -- even though the entire chip industry was sliding into a slump and Intel had scheduled mandatory unpaid vacations for most of its employees in late August. "It became clear our salaries depended on the 386," says Mr. Gelsinger, one of the two engineers on the drive to Livermore.

The 25-year-old Mr. Gelsinger was in charge of the last major task in the design phase, the tapeout -- the process in which the image of each layer of the chip is converted into digital data and recorded on magnetic tape. The tapeout tied up half of Intel's computing power and all of its world-wide communications links for most of the Fourth of July weekend last year. The job, which usually takes three months, was done in less than three weeks.

The day the tapes were finished, they were flown to a company in Japan that specializes in turning tapes into the photographic masks used in manufacturing the chips. (Intel chose the company "because they do better work and they do it faster," Mr. Crawford says.) The first masks were hand-delivered by courier from Japan about a week later, and within a day Intel started a round-the-clock, four-week production run (compared with the normal three months).

The wait for the first samples was excruciating. The Livermore plant started three batches, dubbed Thunder Buns, Hell Rider and Lucky Runner. A scoreboard in Santa Clara monitored the daily progress for those placing bets. After three weeks, Thunder Buns faltered because of a production mistake. Then, Hell Rider ran into a snag. Lucky Runner, which had been trailing, suddenly became the front-runner. On Aug. 9, 1985, the first of Lucky Runner's four-inch-diameter wafers, each containing 55 chips, came out of the fabricator.

That was the night Messrs. Gelsinger and Krauskopf hot-rodded to Livermore. When they returned, after midnight, many members of the design team were still milling about outside the wafer-test area. As Mr. Krauskopf triumphantly strode in, his arms outstretched to show off a plastic recipe box filled with silvery disks, he stumbled -- spilling them on the floor and, as the crowd gasped, trampled them.

"It was a great practical joke," Mr. Gelsinger says. He was right behind Mr. Krauskopf and carrying the real wafers.

Once the laughter died down, the work began. Before long, one of the engineers probing several chips on the very first wafer let out a whoop. The 386 worked.

The following Monday, the design team celebrated with a cake decorated with a stork proclaiming, "It's a Chip!" Intel engineers began perfecting what Andrew S. Grove, the company's president, calls a "Normandy invasion of support products." And within two weeks, Intel began shipping samples to customers to enable them to begin testing, writing software and designing machines. Once again, the customers played a big role.

"We would beat the hell out of their chips and write them very specific memos about what went wrong," says Eric D. Carlson, the general manager for 386-based product development at Convergent Technologies Inc., a microcomputer maker that has invested more than $10 million developing 386-based products. Most suggestions were "tweaks," small adjustments that didn't alter the chip's basic design.

Early last month, Intel began shipping production quantities of the 386 at a price of $299 apiece. This year, it hopes to ship 100,000 chips. Next year, shipments may hit two million, with revenue reaching hundreds of millions of dollars.

Meanwhile, Intel's customers are scrambling to build computers around the 386. But most speculation concerns what IBM -- Intel's biggest customer --will do. "Having worked for IBM for 12 years, I'm sure this is not an easy marketing decision for them," Convergent's Mr. Carlson says. Because the 386 is so powerful, any machine IBM makes will cannibalize sales of other IBM products. "It creates a bit of a problem for IBM," he says.