Businessweek Archives

Inside Intel


Cover Story

INSIDE INTEL

Some 170 employees at Intel Corp.'s plant in Chandler, Ariz., wait expectantly as their shirt-sleeved leader, Andrew S. Grove, throws the meeting open to the floor. It's the latest in a long-running series of forums at Intel facilities around the world where the chief executive of America's No. 1 chipmaker asks employees to download what they're thinking. On this hot morning in April, that's plenty.

For 90 minutes, engineers, assembly workers, and managers alike pepper Grove with tough questions: What's Intel doing to thwart rivals from stealing more business? How will Intel compete with the blazing speed of Digital Equipment Corp.'s new Alpha chip? Is Intel facing the type of problems that plague IBM and DEC? Grove offers no platitudes. Intel, he says, is "under attack."

Hold it, you say. Why are these people so worried? Intel is the fastest-growing and by far the most profitable of the world's top makers of integrated circuits, the tiny slices of silicon that run everything electronic. It's the king of desktop computing, the company whose microprocessors are the brains of some 100 million IBM-style personal computers, five times what its nearest competitor can claim. Intel's grip on the PC market has sent its sales soaring nearly fourfold since 1986, to $4.8 billion last year, and doubled its profits over that period, to $819 million.

FAST BREAK. Despite such strengths, however, Intel's main business is facing its biggest challenge in a decade. Led by Advanced Micro Devices Inc. (AMD), cloners of Intel chips are denting its bottom line. In May, the No. 3 U.S. chipmaker, Texas Instruments Inc., announced two souped-up microprocessors. MIPS Computer Systems Inc. and Sun Microsystems Inc. have new designs that outrace Intel's--and will even run the same huge library of programs. Even IBM, Intel's No. 1 customer, has joined with Apple Computer Inc. and Motorola Inc., maker of Apple's microprocessors, to build a new desktop brain. More disturbing, PC sales have slowed, cutting demand for chips. All this has raised doubts about how long Intel can count on operating margins of 18% to 25%, its range since 1987. Investors are worried: Intel's stock has tumbled from 68 3/4 last Feb. 19 to 47. Could Intel's desktop dominance finally be coming to an end?

Not without a fight, it won't. Grove is striking back on multiple fronts, hauling cloners to court, slashing prices as never before, and for the first time, advertising his chips on TV. He has revamped Intel's strategic vision to focus more on its core microprocessor business. He is personally nurturing "extremely close contacts with the hardware and software worlds," a key to customer satisfaction. Most important, he has launched a crusade to whisk products to market in a flash. "Ultimately," says Grove, "speed is the only weapon we have."

Intel is no longer content to introduce one or two new-generation chips annually and a whole new microprocessor family every three or four years. In 1992, it will market nearly 30 new variants of its cutting-edge 486 chip. This summer, several months ahead of schedule, it will unveil its next-generation chip, a veritable one-chip mainframe that may or may not be named the 586. For most of the '90s, Intel expects to give birth to new chip families every two years, a blistering pace that Grove thinks will keep cloners in a perpetual catch-up mode.

ALL-OUT FIGHT. This speedup will affect everyone who uses computers. Already, the ever-increasing power of Intel microprocessors has propelled the spread of office automation, computer networks, and "friendly" software any novice can fathom. From banking to fast food, nearly every industry relies on Intel chips in a cash register or a PC. Each new generation has spawned advances in products, services, and productivity. So, as Intel gets faster chips to market sooner, worldwide competition will grow more hectic and intense.

Staying ahead is more than a matter of pride for Grove, 55, who fled the Soviet invasion of Hungary in 1956 and has run Intel's day-to-day operations since the mid-1970s. Microprocessors provide about 55% of Intel's sales and the bulk of its profits, which Grove vows to "fight with everything we've got" to protect. Those earnings enable Intel to outspend, outdevelop, and outmanufacture the competition. They also support its spadework in new arenas, such as supercomputers, interactive digital video, and flash memory, a semiconductor alternative to storing data on magnetic disks (page 90). If its rivals keep gaining, Intel could eventually lose ground all around.

This is no idle threat. Cyrix Corp. and Chips & Technologies Inc. have re-created--and improved--Intel's 386 without, they say, violating copyrights or patents. AMD has at least temporarily won the right in court to make 386 clones under a licensing deal that Intel canceled in 1985. In the past 12 months, AMD has won 40% of a market that since 1985 has given Intel $2 billion in profits and a $2.3 billion cash hoard. The 486 may suffer next. Intel has been cutting its prices faster than for any new chip in its history. And in mid-May, it chopped 50% more from one model after Cyrix announced a chip with some similar features. Although the average price of a 486 is still four times that of a 386, analysts say Intel's profits may grow less than 5% this year, to about $850 million.

Intel's chips face another challenge, too. Ebbing demand for personal computers has slowed innovation in advanced PCs. This has left a gap at the top--and most profitable--end of the desktop market that Sun, Hewlett-Packard Co., and other makers of powerful workstations are working to fill. Thanks to microprocessors based on a technology known as RISC, or reduced instruction-set computing (page 94), workstations have dazzling graphics and more oomph--handy for doing complex tasks and moving data faster over networks. And some are as cheap as high-endPCs. So the workstation makers are now making inroads among such PC buyers as stock traders, banks, and airlines.

Grove's plan for meeting this challenge is to stimulate PC advances. For one thing, Intel is trying to make such gear as its networkingcircuit boards and graphics chips easier and cheaper to put in PCs, he notes, because new technology "won't help unless the marketplace accepts it." That would also line Intel's pockets twice, since these features often need the extra processing power found only in the newest chips.

Leveraging Intel's ancillary products also dovetails with Grove's grand vision of making the company the center of computing--everything from palmtops to supercomputers. The 586, with its 3 million transistors, will begin to tie all these strands together. It will process 100 million instructions per second, more than double the fastest 486 and equal to RISC chips from Sun and MIPS Computer.

DOUBLING UP. The mounting RISC-chip threat was the main thing that persuaded Grove to rush chips to market quicker, but the cloners forced his hand, too. Intel had always waited until one generation was ready for production before starting work on the next (chart). That changed in mid-1990. Amid talk of looming competition, Grove "doubled up" on design. He launched work on the 686 generation, due around the end of 1993, two years before the 586's planned christening. By chance, Intel had just canceled a joint computer project with Germany's Siemens. That freed dozens of engineers at the company's Hillsboro (Ore.) site who had been working on 686-type features, including sufficient circuitry to allow redundancy. That's a must for fail-safe computers that can't suffer crashes, such as those handling bank transactions. Now, members of the 586 team, their work largely done, are shifting to the 786 generation, which may be unveiled in 1995. By then, what is now the 686 team should have beenat work for two years on the family that will be the 886.

Developing two microprocessor generations at once is a monumental endeavor. The 486, with 1.2 million transistors, took more than $250 million and 450 work-years to develop. The goal for the 786 is far more formidable: Its 20 million transistors will process some 250 million instructions per second (MIPS), more than 10 times what the first 486 could do. Somewhere around 2000, maybe for the 886, Intel envisions a quantum leap to 100 million transistors, divided into as many as four independent chips for so-called parallel processing. The overall chip will execute 2 billion instructions a second--that's supercomputer territory now, but in 10 years it could be a desktop province.

Building on unknown foundations, as Intel is doing with the 686, will strain the research budget, even for a company accustomed, as Grove once quipped, to betting millions on science fiction. Its engineers will feel the heat, too.

Each new generation starts as the dream of a handful of designers who have just finished a previous chip family. In early 1989, for instance, engineers wrapping up the 486 spent weekends brainstorming the 586. They were trying to balance two ingredients: the number of transistors manufacturing could put on a chip three years hence vs. the performance Intel would need by then to stay ahead of its rivals.

By mid-1990, the 586 project was growing fast, toward the current 200 engineers in Santa Clara, Calif. The "real estate" specialists were laying out blocks of silicon for various features. As a section was finished, other engineers translated their diagrams into thousands of transistors and wires. By late last year, the work went late into the night. There were frequent round-the-clock scrambles to change the design--say, if manufacturing tripped over a last-minute problem. Just weeks ago, the team loaded its labors onto a computer tape that generated the masks, or stencils, that print the circuitry on silicon wafers--the first of which are just seeing the light of day. Now, the P6 team that Intel formed in Oregon in mid-1990 is tapping the 586 engineers' accumulated knowledge to nail down precisely what the 686 will be building on.

Such efforts depend on a critical resource: Intel's proprietary design tools. As the complexity of microprocessors climbed into the hundreds of thousands of transistors, making manual design unthinkable, Intel invested heavily in design-automation software. "People have probably been wondering what we've been doing with those 386 profits," says David L. House, a senior vice-president. "They've gone into `enablers' that make it possible to design successive chips, each with two or three times as many transistors, at no increase in development time."

In fact, the product cycle may actually shrink, thanks to a breakthrough from Quickturn Systems Inc., a startup in Mountain View, Calif. Intel Vice-President Albert Y.C. Yu says the Quickturn system should trim at least six months off the usual four-year cycle from conception to volume production.

COSTLY NIGHTMARE. Called hardware emulation, Quickturn's technology can build a model chip using programmable chips that pretend to be transistors. Quickturn intended the system for simple chips but joined with Intel to apply it to the 586. The result is a room-size collection of circuit boards that reflects the performance of the design more faithfully than software simulations--and runs engineering tests up to 30,000 times faster. Thus able to perform many more tests, the engineers caught numerous bugs before production began. That's important. On the 386, one bad transistor sneaked through--until customers started seeing math errors. It took months to fix the goof and replace faulty chips in the field.

The emulator had one more benefit: blunting the spread of RISC. At a technology forum for PC companies and software developers last November, Yu dialed it up and ran a Lotus 1-2-3 spreadsheet from a terminal. The crowd was astonished that a model was already working. Six months later, Compaq Computer Corp. scrubbed its plans for a RISC-based PC. "Intel is back on the high-performance track," says Doug Johns, a senior vice-president at Compaq. The Houston company's decision may also doom the Advanced Computing Environment. This consortium, spearheaded by Compaq, Microsoft, and MIPS, was formed to establish an alternative, RISC-based PC standard.

The other key to shortening the "time-to-money" cycle, as Intel calls it, is concurrent engineering--getting customers and manufacturing involved early in the design. This was a sea change within Intel, which has been infamous for its high-handed treatment of customers. Before the 586, Intel didn't always bother to ask what they wanted--or listen closely when they volunteered. "They used to say, 'Here's what we've got--how many do you want to buy?' " recalls an executive at one PC maker.

CONFRONTATIONS. Intel's arrogance might have been understandable during the preclone days. It was founded in 1968 by Gordon E. Moore, now chairman, and the late Robert N. Noyce, who in 1957 had invented one form of the integrated circuit. The company went on to create the DRAM, or dynamic random-access memory, now the most common chip in every computer, plus other widely used chips. In 1971, it invented the microprocessor, which ultimately won IBM's approval and became the PC standard. Little wonder Intel felt special.

Grove added a confrontational edge when he took over the reins. With an engineer's logic, he believes that pulling no punches is simply efficient. For instance, any employee can demand an "AR"--action required--of any executive on any issue. Grove encourages people to iron out disputes by speaking their minds, and they do. It can be unpleasant, but problems get solved faster.

Grove has mellowed since his 1987 elevation to CEO: The company no longer keeps lists of those who miss the 8 a.m. check-in. And starting with the 586, he insisted that Intel put on a new face, especially in dealings with customers. "Historically, we thought of ourselves as a chipmaker," says Grove. That insular attitude began to change in the late 1980s. Today, he says, there's a "strong allocation of resources" devoted to customer support.

The makeover program's Mr. Inside is Craig R. Barrett, a former materials-science professor at Stanford University. After transforming Intel's once-weak manufacturing arm into a fortress of respectability, he was promoted in January, 1990, to executive vice-president--and heir apparent to Grove. Barrett took charge of daily operations, includingrevamping the product-development process. He's an appropriate choice to head the speed crusade--an avid athlete who relishes marathon bicycle trips.

Both the internal and external efforts were crucial for the 586, an unusually pivotal product. It is aimed at large computers as well as PCs, so Intel engineers had to envision how it would fit in those various systems. The designers first visited every major customer, plus major software houses such as Borland International, Lotus Development, and Microsoft, to ask what those companies wanted. The result was a list of 147 specific features--many ranked differently from what Intel expected. Doing this at the front end, says Vinod K. Dham, general manager of the 586 group, more than paid for itself by avoiding most revisions in later stages.

Dham is most pleased about the mind-set change. Pre-586, Intel engineers got their kicks from the intellectual joy of devising elegant or efficient solutions to circuit-design problems, regardless of their value to customers. Now, says Dham, engineers think like customers. The most stinging epithet in meetings is: "C'mon, stop being a chiphead."

This attitude is mirrored in marketing, too. Intel surprised many customers late last year by furnishing software simulations of the 586 so they could start engineering their new computers. In the days of arrogance, customers were in the dark until Intel was set to start etching silicon--about now for the 586. The early notice should trim at least six months from the time it takes computer makers to debut 586-based systems, says Yu. For Intel, that should bring in more revenue sooner, cushioning the blow of a chip generation's shorter life cycle.

In yet another departure, the 586 project included help for software companies. Intel engineers wrote so-called software compilers to complement their chip. Compilers are the critical tools that convert programs written in software languages into the special machine language a microprocessor understands. The lack of compilers is one reason that software has lagged at least a generation behind Intel's silicon. Today, six years after the 386's birth, only a smattering of programs takes full advantage of its power.

`RESPONSIVE.' Better service seems to be working. An executive at NCR Corp., pointing to the new collaboration, calls Intel his "microprocessor division." Adds Carl Stork, director of business development for Microsoft's next-generation operating system, or basic software: "They're a much more responsive company."

Skeptics say that to reduce development time, Intel is limiting the power of its chips a bit. And Intel concedes that a two-year interval between generations can't be maintained forever, mainly because manufacturing advances don't come that fast. But for this decade, Intel sees an opening. In fact, Patrick P. Gelsinger, who heads chip design in Hillsboro, says Intel could reach its turn-of-the-century, 100 million-transistor chip a year or two early.

To pull it off, Intel is spending $300 million on a new Hillsboro facility, slated to open this summer, just to develop the chipmaking processes for the next three chip generations. After that, given the pace of chip technology, the facility will be obsolete. If Intel can discover techniques for packing more transistors onto smaller chips before its competitors do, it will own a defense against cloners as well. The same methods could shrink existing chips, which would cut costs and prices.

Reaching such goals means sparing no expense, however. Intel plans to raise its research-and-development spending 29% this year, to $800 million--tops of any chipmaker in the world. That excludes a $1.2 billion capital budget, up 25% from last year and also No. 1 worldwide. And then there are its technological contributions to joint ventures, such as one with Japan's NMB Semiconductor Co., which is making Intel's flash-memory chips. As mind-boggling as it sounds, plowing 30% of revenues into R&D and equipment isn't outlandish in semiconductors: Reinvesting 20% or more of sales is required just to keep up.

Whether even $2 billion will fend off competitors hinges on how PC makers and buyers respond to the faster and gentler Intel. Intel's past disdain for customers--particularly its sole-source strategy that fostered shortages of the 386 several years ago--has left a bitter aftertaste. "I would pay more to get an equal-performance microprocessor from anyone else," declares an irritated buyer for one PC maker.

Unless, of course, his customers want machines with Intel chips. Intel scored a coup with a 1988 ad campaign promoting its low-cost, SX version of the 386. That previously obscure product took off to become the company's all-time best-seller. So, in peddling the 486, Intel chartered 100 salespeople to call on corporations and stress the benefits of "Intel Inside," its new advertising slogan. Grove says nearly every visit resulted in a big order for PCs from a computer maker. Still, it's uncertain how long loyalty will last if low-cost clones are available. "It takes more than advertising to convince most customers," says Genelle Trader, marketing vice-president at Everex Systems Inc., a PC maker in Fremont, Calif.

MORE RISK. Just in case, Intel is staying lean. It has fewer employees now than in 1984. Instead of hiring for new projects, it transfers workers from low-return products. Last year, 7% of the work force moved to new jobs and per-employee sales hit a record $197,000. Offices at Intel's headquarters reflect Intel's spartan values. Even Moore and Grove occupy open cubicles--though a bit larger than most.

Today, Intel is holding its breath as the 586 moves into "first silicon." These initial production runs are always a period of great uncertainty. For the 486, the ramp-up to volume output took nine months because of a glitch in the chip's math-processing circuits, which had been beefed up for heavy-duty number-crunching chores. The 586 presents a bigger departure--and thus a greater risk. It is Intel's first microprocessor made with a process called BiCMOS, a hybrid of two chipmaking technologies. Bipolar transistors switch on and off extremely fast, generating so much heat they can melt without refrigeration. By using bipolar only for a few transistors that need extra dash, BiCMOS chips keep cool enough that refrigeration is not necessary.

Once the 586 is proven, look for Intel to focus on steering PC makers toward more innovation. For example, it has designed a new machine called the Panther, a PC with workstation power, that customers can license. But some PC makers are wary. Suppose the campaign to seed the PC business with new technology doesn't pay off in quick acceptance of the 586--might Intel market its own computers to recoup its R&D costs? It already rakes in $400 million building PCs for companies such as DEC and Unisys Corp., which resell the computers under their own names. Grove insists Intel will never compete directly with buyers of its chips. Says Morgan Stanley & Co. analyst Andrew J. Kessler: "They're skating barefoot on a sword."

So far, Intel hasn't slipped. If anything, most analysts say, its efforts to speed development and cultivate customers have put it on firmer footing. Chances are that even if Intel can't reclaim its near-monopoly in microprocessors, it will remain king of the desktop for years to come. Even Sun Microsystems and NeXT Computer Inc. are resigned to that: They have both promised to adapt their basic operating software to run on Intel chips. That's the sort of news that gets Grove even more fired up. "The more people nip at our heels, the faster we go," he declares. "One by one, these mosquitoes will be swept away by the wind." Meantime, Intel will be spending a bundle on insect repellent.Robert D. Hof in Santa Clara, Calif.


Later, Baby
LIMITED-TIME OFFER SUBSCRIBE NOW

(enter your email)
(enter up to 5 email addresses, separated by commas)

Max 250 characters

 
blog comments powered by Disqus