Science & Technology
THE UPSTART CHIP DESIGNER THAT'S CHIPPING AWAY AT INTEL
On a frigid night last January came the moment of truth for Cyrix Corp. Late on Friday the 24th, the startup received the first preproduction model of its new brain for IBM-compatible personal computers--a microprocessor it hoped would break Intel Corp.'s lock on the high end of that business. At 2:20 a.m., as most of the 17 engineers who had labored more than a year on the chip looked on, the verdict came in. A PC using the Cyrix chip had no trouble with Microsoft Corp.'s Windows, a complex program that should have crashed if the chip had serious flaws. The group erupted with cheers. "To do a microprocessor in a short time and have it functional on the first pass--you can't ask for anything more exciting," says Senior Engineer Mark W. Bluhm.
Except to do it again--and eventually beat Intel to market. Cyrix intends to do just that. Even before it finished last winter's "workalike" version of Intel's 486 microprocessor, Cyrix was starting on the next generation, a 586 that will compete with Intel's P5. Intel should be first out with that one early next year. But for the following model, the 686, the Richardson (Tex.) upstart is betting it can take the lead. "If we do 18-month design cycles, and they do 3- to 4-year design cycles, we are going to get there before them," says Kevin C. McDonough, Cyrix' engineering vice-president.
Those are major ifs, but should they occur, Cyrix would become a classic in entrepreneurial annals, the microprocessor market could see a flurry of new competitors, and Intel's mystique, if not its growth, could suffer. "We've broken through the psychology that says it costs $250 million to do these designs, that you've got to have hundreds of engineers, and that it takes four years," says Cyrix President Gerald D. Rogers. Cyrix had little choice. Only four years old, with profits last year of $13 million on revenues of $58 million, it had to do its 486 on the cheap--for under $10 million, or 4% of Intel's investment.
It will be a challenge to keep doing it that way. Intel's head microprocessor executive, Vice-President Paul S. Otellini, argues that "it's always easier to imitate something than to create the original." Designing a 686 from scratch, he contends, "is going to take architectural techniques that will be difficult for anyone to implement." Moreover, he and other industry executives doubt Cyrix' claim that its 486 virtually matches Intel's design and performance at a fraction of the cost. They even question whether it's really a 486.
Things are confusing because the new chip is a hybrid: Cyrix scrimped on on-chip memory, so it slows down when it runs large programs. The chip also has the same "pin-outs," or pattern of tiny plugs on the bottom, as Intel's earlier and slower 386--though that can be an advantage. Owners of 386-based PCs who want more power can simply replace the microprocessor--a $120-per-machine steroid injection. That's what the chief technical official at a large bank plans to do. In his tests, the Cyrix design did most banking jobs 70% faster than a 386.
MATH WHIZ. Just who is this brash challenger, anyway? Privately held Cyrix was founded in 1988 by two former Texas Instruments Inc. engineers. Rogers was TI's microprocessor chief when he left. Thomas B. Brightman had jumped ship in 1982 to develop chips for Commodore International Ltd. and then Atari Corp., climbing to Atari's head of business planning. Once Rogers decided to go after Intel, Brightman was his choice for co-founder. "We knew we could build a microprocessor," says Rogers." But we didn't have the money to go do it. So our plan was to do a co-processor to generate the cash to bootstrap ourselves into the microprocessor business."
A co-processor is a chip tailored to zip through heavy-duty math jobs that tax the main processor. The math co-processor for 386 chips is called a 387, and that's what Cyrix first set out to build. Its goal was not a clone in the usual sense but an original design that beat Intel on performance. Cyrix' version runs faster than Intel's--up to 20 times faster at some tasks--and was developed in just 14 months on a penny-pinching $4 million. Since Cyrix introduced its 387 in late 1989, it has cornered 10% of a $427 million market, according to market researcher In-Stat Inc. That's enough to earn the No. 2 ranking. Cyrix' spurting sales also helped push Intel to cut its 387 prices by 70% last year.
As its math chip took off, Cyrix set its sights on the 486, a business that In-Stat estimates will reach $1.8 million this year. It followed up last spring's entry with another 486 hybrid this summer, and in the fourth quarter, Cyrix plans to unveil a "real" 486, with 486 pin-outs.
Cyrix is about the last company you'd expect to make a run at Intel. Advanced Micro Devices Inc. (AMD), a 386 clonemaker several times Cyrix' size, is still working on its 486. And Intel's resources should give it a commanding edge. It has a vast storehouse of knowhow from earlier chip designs, and it's spending $1 billion to acquire more. Still, the former general manager of a chipmaker that built 286 microprocessors under license from Intel says it also shares a characteristic common to big companies: "The bureaucracy is incredible."
`FABLESS.' Intel's alleged weaknesses aside, Cyrix credits two strategies for its success. First is a rapid design approach that exploits the most advanced tools in computer-aided design and engineering and stresses a "get it right the first time" attitude. Rogers was convinced that if Cyrix gave the best tools to talented engineers and purged the design process of hang-ups and errors, it could develop microprocessors in a fraction of the time the task had typically taken. The other strategy is to stay "fabless"--that is, avoid spending tens of millions on a chip-fabrication plant, which Cyrix couldn't afford anyway. Instead, it farms out production of its designs, currently to SGS-Thomson Microelectronics Inc. in Carrollton, Tex., and to TI.
This meant that the profits from Cyrix' 387 could fund its 486. Development of that chip, code-named M.5, kicked off in September, 1990. The first job was to come up with the specifications that the chip would have to meet to be compatible with Intel--without regard to how Intel had achieved similar results. McDonough says it's not necessary to copy Intel's approach because cloning its designs "isn't really the measure of compatibility anymore. The real measure now is, can this chip run all the software that the user is going to run?"
To guarantee this compatibility, Cyrix spent six months debugging the design--the biggest single slice of the development cycle. Its engineers ran countless tests with their computer-simulated model, then checked the results against an actual Intel chip. After numerous revisions, especially to the microcode--the instructions built into a microprocessor that tell it how to execute programs--Cyrix got down to mapping the actual circuitry last October. Just before yearend, the finished design went to SGS-Thomson. As expected, the first chips had minor bugs. But they were soon fixed, and volume production began in April. James N. Chapman, a former Intel marketing director who now heads Cyrix' marketing, was impressed. The Cyrix chip "was ready to go much faster than any microprocessor that I've been involved with," he says.
Cyrix had lined up SGS-Thomson partly to head off an expected Intel suit, since Thomson owned a license to key Intel patents, acquired when it bought Mostek Corp. in 1985. It didn't work. Intel sued even before it inspected Cyrix' chip. Intel's patents are so comprehensive, says General Counsel F. Thomas Dunlap Jr., that "if a chip runs all the 386 software, it has to infringe these patents." So far, the courts have sided with Cyrix, ruling that because SGS-Thomson has licenses from Intel, there is no infringement. Intel is appealing.
`VALUE.' Meanwhile, several PC makers have recognized Cyrix as a new source of microprocessors. Tandy, Zeos International, and CompuAdd are using the Cyrix 486 in some of their computers. "The price-performance value was such that we couldn't ignore it," says Bryan Kerr, a vice-president at Tandon Corp. The competition also keeps Intel in line. Since spring, it has lowered its 486 prices by as much as 54%, to only a little more than Cyrix' in some cases.
Armed with its fast-design capability, Cyrix plans a new product every quarter, "weaving in and out of the Intel product road map" and hitting market niches, says Chapman. The company figures that with marketing support from its producers, it can grab 20% to 30% of the market for Intel chips. Analyst Thomas A. Thornhill at Montgomery Securities thinks that's high but says Cyrix could get 15% "largely by displacing AMD." AMD, which has snatched 30% of the 386 market, is having a tougher time with the 486 because of a recent courtroom loss that could keep it from using Intel's microcode.
Intel's pockets are deep enough to keep the pressure on Cyrix, in the marketplace and in court. Because Cyrix depends so heavily on so few chips, "Intel could still cripple them" by cutting prices, says Michael Slater, a microprocessor consultant in Sebastopol, Calif. Rogers concedes that the threat exists. Still, he vows "to overcome that by becoming a broad supplier of microprocessors. We're going to take control away from Intel." A cocky call from a fighter who just stepped into the ring.Stephanie Anderson Forest in Dallas, with Richard Brandt in San Francisco