Silicon Chameleon Transmeta Challenges Intel

Its chips are fast, cheap--and upgradeable

David R. Ditzel knew he had pulled off a public-relations coup when no fewer than three CNN news crews, each with its own satellite-TV van, showed up last Jan. 19 for the unveiling of his company's super-secret chip. The chief executive of startup Transmeta Corp. had waited nearly five years for that moment. But even he was stunned by the turnout. So many newshounds came sniffing that Transmeta executives could spend only 15 minutes with each reporter. Starting at noon that day and continuing for the next 72 hours, the company's Web site was one of the 10 most visited places on the Internet, and Ditzel received hundreds of phone calls and e-mails. "It took two days for it all to sink in," he recalls.

Now, the hoopla has started up again. Some top names in the electronics industry have kicked in $88 million to help Transmeta begin producing its speedy Crusoe processors. Sony, Gateway, Compaq, and America Online have all bought a stake in Transmeta. And now it's just started shipping in volume. By late June, the first Crusoe-based notebook PCs, handheld computers, and information appliances should begin showing up from companies such as IBM, Compaq Computer, former chipmaker S3 Corp., and Taiwanese giant Quanta Computer. These gizmos will run on batteries far longer than any predecessors. They could be smaller and lighter, too. "Transmeta's chips let us consider whole new market segments," says Andrew L. Wolfe, chief technology officer for S3, which plans this fall to unveil a 2-lb., tablet-shaped computer that can surf the Web wirelessly.

Why all the excitement? Until now, designers of mobile PCs and portable gadgets faced a frustrating conundrum: There are a half-dozen low-power processors on the market, from suppliers such as Hitachi Ltd. and Mips Technologies Inc., but these chips can't handle Windows software. And while Intel Corp.'s chips are adept at Windows, they gulp so much power that batteries usually last only a few hours. "People are tired of their batteries running out, tired of shoulder burn from heavy laptops," Ditzel says.

ON THE FLY. For its alternative, Transmeta relies on some digital trickery. Strictly speaking, its design is not compatible with Intel's Pentiums, which process instructions 32 digital bits at a time. Transmeta's circuits gobble instructions in 128-bit chunks, a technique called very long instruction word (VLIW). To pull that off, the chip's processing engine is wrapped in a silicon-and-software envelope that translates Pentium instructions, on the fly, into VLIW code. The upshot: A power-stingy 700-megahertz Crusoe can run Windows programs as fast as a 500-MHz Pentium III, the current champ in laptops. Yet Crusoe chips are smaller and have fewer transistors, so they cost as much as 50% less.

What's more, unlike any other microprocessor, Transmeta chips can be upgraded with software to remain on the cutting edge for years at a stretch. No longer will road warriors have to keep buying new laptops to get the latest gee-whiz features. Instead, they'll simply download VLIW software to acquire new capabilities.

Not everyone considers Transmeta a shoo-in, however. Among the skeptics is the closely read Microprocessor Report, which published a lukewarm review, wondering if Transmeta was setting its prices a little too high. And Joseph J. Byrne, an analyst with market researcher Dataquest Inc., says Intel has the resources to paint Transmeta into a corner while it develops something comparable. Indeed, when it comes to challenging Intel, history is not encouraging. A half-dozen have tried. Only longtime nemesis Advanced Micro Devices Inc. (AMD) has made a lasting dent and now enjoys a 17% share of the market for Intel-type chips.

Don't pester Ditzel with such details. The 43-year-old entrepreneur, who once worked for Bell Labs and headed processor development at Sun Microsystems Inc. for eight years, has raised some $188 million from venture capitalists and Microsoft Corp. co-founder Paul Allen, and the new investor-customers. Buyers of the chips, which are manufactured by IBM, are playing their own game of say-little, to whet expectations about upcoming products. But analysts figure there are some pretty safe bets (table), such as portable systems from Compaq and IBM.

Besides lots of money and big-name customers, Ditzel has assembled a team of high-profile engineers and execs. One is James N. Chapman, who spent 11 years at Intel and then flogged sales of Intel-compatible chips at Cyrix Corp. Another example: Linus Torvalds, the legendary author of the popular Linux operating system. With such talent, Ditzel declares, "we have no qualms about competing in the market."

The opportunity is enormous. Transmeta's high-end chip, dubbed the Crusoe TM5400, is priced from $120 to $300, depending on configuration, and is aimed at laptops, the fastest-growing segment of the PC business. The simpler TM3200, costing $65 to $90, is meant for information appliances, a still-nascent market that could explode over the next few years. All told, the potential for Crusoe chips could top 160 million units by 2003. Grabbing even a 10% share would mean billions in revenues. And some observers figure Transmeta could do far better. "This is the first truly innovative development in processors in the last 12 years," says Drew Peck, an analyst at Boston brokerage SG Cowen & Co. Even more enthusiastic is Robert A. Enderle, a chip watcher at Giga Information Group Inc. "Transmeta has the potential to overthrow the king," he says.

Analysts are especially intrigued by the clever software layer that translates Intel instructions into VLIW. This linguist could easily handle other types of software code as well. So one computer could run not only Windows but also Macintosh programs and Sony PlayStation games. In effect, the Crusoe is a silicon chameleon. This "software microprocessor," Ditzel stresses, gives Transmeta significant competitive advantages. It can take years to revise the complex silicon circuits on an Intel or AMD chip. But Crusoe designs can be upgraded with simple keystrokes. Case in point: A customer in Japan found a bug in an early Crusoe version. Transmeta engineers quickly wrote a fix, then zapped the new code to the customer electronically. "We essentially sent them a new microprocessor over the Internet in a week," Ditzel says. "It blew them away."

So far, rivals are hardly quaking, but they're definitely taking notice. AMD Senior Vice-President Robert R. Herb praises Transmeta's technology but cautions that surviving in the cutthroat market for Intel-compatible chips will be tough. And Transmeta can't get by with just a snappy processor, he adds: It also needs to develop the usual companion chips and development tools for customers. James R. Edwards echoes that point. He's a marketing manager at National Semiconductor Corp., which fled PC processors last year but continues to make a special-purpose chip for info appliances. Called Geode, this chip has most of the required silicon companions built in. Still, the lack of Transmeta support chips didn't stop S3 from chucking Geode in favor of Crusoe.

Intel won't comment on Transmeta. But at its Apr. 27 meeting with Wall Street analysts, Intel pledged to deliver a Pentium this summer that matches Crusoe's power consumption. That, sniffs Chapman, is an empty promise to keep customers from jumping ship. Could be. On the other hand, if Transmeta does begin stealing laptop customers en masse, Intel will certainly unleash a ferocious counterattack. For Transmeta, too much success could precipitate a backlash. To survive, Ditzel will need all the lavish publicity, capital, and talent--and cunning--he can muster.

    Before it's here, it's on the Bloomberg Terminal.