dcsimg
RealTime IT News

All-Star Lineup Heads Chip Startup

Let's see, there's Intel's Pentium and XScale, IBM's Power, Sun's Sparc line, FreeScale, ARM, and the list of microprocessors goes on. Why on earth do we need another on?

P.A. Semi plans to answer that question emphatically on October 28, when it unveils the fruits of its two years of stealth design. The Silicon Valley startup is expected to preview its first chips at the Fall Processor Forum conference sponsored by In-Stat.

The company isn't saying much until then, except for a few tantalizing details and the background of some of its impressive engineering talent. In an email to internetnews.com, P.A. Semi said it plans to "face the 800-pound gorillas head-on -- including Intel, AMD and Freescale -- by building a high-performance processor at unprecedented low-power." At the P.A. Semi Web site, the company said it will have products for high-end consumer portables, server blades and network infrastructure.

Just a few years back, in 2000, another highly-regarded Silicon Valley chip startup burst on the scene with similar grand ambitions. And for a while it looked like that company, Transmeta, might just pull it off. Transmeta had world-class engineers, big investors and a software emulation scheme that let its low-power Crusoe chip offer Intel compatibility at a fraction of the price.

But Transmeta suffered a lengthy delay getting Crusoe to market and finally got but a few notebook computer and blade manufacturers to sign on. Meanwhile, the company's bold marketing tactics stirred Intel to action; the chip giant responded with a competitive low-power push of its own that continues to this day. Transmeta sold Crusoe earlier this year.

"Transmeta totally discredited Intel and did everything it could to tick them off, and they succeeded. That was a mistake," Kevin Krewell, an analyst with Processor Forum host In-Stat, told internetnews.com. "P.A. Semi is not going after the same market that Transmeta was, like ultra-portable, Intel-compatible PCs. They will have a different instruction set and will be looking at markets like low-power, high-performance servers."

Krewell also said P.A. Semi's presentation will be a preview of products not due to ship till next year.

"Intel, AMD and IBM all have a low-power and multi-core focus, so, at some level, [P.A. Semi] is betting they can outdo those companies by a lot," Gordon Haff, analyst with Illuminata, told internetnews.com. "It really depends on their target market, but even a 20 or 30 percent improvement isn't going to be enough to get manufacturers to switch to another chip."

Krewell said P.A. Semi has an "all-star team of designers, and its technology looks good." A non-disclosure agreement forbids him from getting any more specific.

Heading P.A. Semi is president and CEO Dan Dobberpuhl, creator of the highly regarded Alpha, a pioneering, innovative 64-bit chip that was rated the fastest processor in the world at the time of its release in 1992; and StrongARM for DEC, now owned by Intel, which has evolved it to the XScale processor for cell phones and mobile computing devices. P.A. Semi also has engineers who helped lead the design of the original AMD Opteron, Intel Itanium and Sun UltraSparc processors.

Before P.A. Semi, Dobberpuhl headed SiByte, an embedded processor company that designed high-performance networking chips using MIPS Technologies' design architecture.

"There is quite an impressive group of people involved with this company, but it does take more than that to be successful," noted Haff. He added that, as Transmeta discovered, the market for low-power blades has not been nearly as strong as analysts expected, and P.A. will have to offer a breakthrough to make serious headway.

"Power in the data center is definitely a concern," said Haff. "And the idea that servers have low utilization, so let's get max density with a blade solution, sounds great on paper, but it hasn't been widely accepted by IT departments."