Why Transmeta Failed

Posted 06/27/2005 at 7:42am | by

0

Comments

Print

By Tom R. Halfhill
Another one bites the dust. Well, almost. Transmeta isn’t completely out of business, but it has been forced to downsize and radically change course in order to slow the arterial bleeding still threatening to kill the wounded company. What went wrong?

Almost everything went wrong. The worst effect of Transmeta’s downfall, however, isn’t the millions of dollars bled from stockholders or the retreat of Transmeta’s processors from the market. (Transmeta’s last-ditch strategy is to gradually stop selling chips in favor of licensing its technology and engineering services to other companies.) No, the worst fallout is the impact on the investment community. It will be years—if ever—before another startup gets the massive funding needed to challenge Intel’s supremacy in PC processors. For the foreseeable future, only AMD will offer any significant competition against the mighty empire. (VIA still makes PC processors, but it’s a bit player.)

When Transmeta was founded in 1995, it began with big ambitions and a big disadvantage. Intel is a huge company with vast financial resources, world-class engineering, and its own chip-fabrication plants. By contrast, Transmeta is a small, fabless semiconductor company that subcontracts its chip manufacturing to independent foundries. After years of design effort, Transmeta realized it couldn’t beat the performance of Intel’s desktop processors.

Part of the problem was Transmeta’s radical new approach to microprocessor design. By themselves, Transmeta processors can’t run x86 software. They rely on emulation—the company prefers to call it “code-morphing software”—to achieve x86 compatibility. Despite using the best emulation technology available, Transmeta couldn’t match Intel’s desktop performance. But by moving some complexity from the chips into the emulation software and inventing its LongRun voltage/frequency-scaling technology, Transmeta slashed power consumption. So the company decided to focus on notebook and embedded processors.

Five years ago, when Transmeta introduced its first Crusoe chips, company officials objected to my opinion that emulation overhead would cripple Crusoe. “Overhead” was obsolete, they insisted—code morphing wasn’t a handicap. Now we know better.

When Intel, AMD, VIA, and other companies imitated LongRun, Crusoe was in stormy seas. When Intel did as I predicted and created its own low-power design (Pentium M/Centrino), Crusoe was shipwrecked. Transmeta was overwhelmed by Intel’s superior resources. Although I salute Transmeta for having the gumption to try, and for creating some innovative technology, I marvel at the investors who bet so much money on such a long shot.