AMD tries to spin embedded gold

AMD tries to spin embedded gold
Austin, Texas — Fifteen months after plunking down $50 million for Alchemy Semiconductor and its highly regarded embedded-processor design team, Advanced Micro Devices Inc. is looking to spin gold from the melding of Alchemy's embedded expertise with AMD's little-known WLAN technology.

Alchemy is "the future," said Billy Edwards, general manager of AMD's Personal Connec-tivity Solutions Division. Analysts say that Alchemy is key to AMD's effort to diversify away from PC silicon, much as the Xscale effort has been for Intel Corp.

But AMD's plans have run up against obstacles that say a lot about today's embedded market: the importance of key operating software from Microsoft Corp. and the alleged inertia of system design engineers wedded to the ARM architecture.

Acquired in February of 2002, Alchemy is now the heart of the Personal Connectivity Solutions (PCS) division, which employs about 275 people, divided between Austin and and AMD's design center in Dresden, Germany. The division encompasses AMD's legacy embedded products in the X86 and Ethernet segments — products that Edwards said still bring in enough revenue to keep the division "breaking even."

He said the Alchemy initiative has the "complete commitment" of AMD CEO Hector Ruiz, who developed an affinity for the embedded market while running Motorola's Semiconductor Products Sector. Edwards worked for Ruiz at Motorola as a strategy planning manager.

Alchemy's MIPS-based cores "have an opportunity" to advance AMD's position in embedded because the dominant ARM architecture "is running out of innovation," said Jim Turley, an embedded processor analyst based in Monterey, Calif. "The ARM 11 is a big yawn. It is too much like the ARM 9 and 10. "

But for AMD to take advantage of that, Turley said, the Alchemy designers need to re-establish a distinct performance edge.

Before the current Alchemy team split off from Digital Equipment Corp., Turley said, it designed StrongARM cores with "godlike" reputations in terms of power consumption and speed. But the close working relationship between the process and product design engineers at Digital's Hudson, Mass., facility helped make such achievements possible, he said. They have proven more difficult to duplicate with a foundry production strategy.

Meanwhile, "there needs to be a buzz around the Alchemy products," Turley said. Given the proliferation of 32-bit processors, he said, designers are often driven "by advertising and market channel promotions" when choosing among them.

Edwards believes wireless crucial to customer success.

"AMD really needs to make Alchemy more of a recognized name among hardware designers," said Turley, a former Microprocessor Report editor who now runs Jim Turley Associates. "It has got to be seen as cool to use Alchemy. Instead, AMD has been very quiet about the whole thing. The reality is that MPUs are sold on marketing — oftentimes cutthroat marketing — and not technology by itself."

ARM is hugely popular, Turley noted, "and so people go with that. Engineers go with the herd, and a lot of ARM's success is due to that snowball effect."

"My biggest concern about Alchemy is that MIPS is not the default standard for handhelds," said Tom Starnes, embedded microprocessor analyst at Gartner-Dataquest. "[That segment] has been overtaken by ARM."

Seeking its voice

AMD has been quiet, perhaps, because it doesn't yet know what it wants to say.

Edwards, an energetic manager who clearly relishes the Alchemy challenge, is enthusiastic about the possibilities inherent in combining WLAN and processor cores. But he declined to be specific about how AMD will combine its 802.11 design expertise in Germany with that of the Alchemy team here.

"Some customers want a fully integrated product, some want the baseband function on the Alchemy die, others want it kept completely separate," Edwards said. "We are working the whole road map, including marrying them up" on systems-on-chip.

"AMD was late, relative to all of the [802.11] b stuff, and we're still defining our road map," he said. "But one thing we believe is that if you don't have a wireless connection to offer customers, you aren't going to succeed."

The Alchemy building in north Austin includes a demonstration room that shows off such devices as Web tablets or "thin clients," automotive navigation and vehicle entertainment systems, residential gateways and a new type of bar code reader. Marketing vice president Phil Pompa demonstrated a PDA prototype that he claimed offers video that is far superior to that of any system based on Intel's Xscale processor, arguing that the Alchemy processors have "a cleaner bus design" than does the Xscale.

But Alchemy faces a problem in the PDA sector: Both Microsoft's popular PocketPC operating system and the PalmOS support the ARM processor architecture, but not MIPS.

Indeed, the dominance of ARM cores in the mobile market has been a barrier to Alchemy since the company's founding in the mid-1990s. When Alchemy was getting started, Pompa noted, an ARM architectural license — one that would have allowed co-founders Rich Witek and Greg Hoeppner to apply their hardware design creativity to the ARM instruction set — was not available. (Intel got one when it purchased Digital's semiconductor operations.) So Alchemy took a MIPS license and designed to that architecture.

Witek and Hoeppner had earlier worked at Digital as members of the Strong-

ARM design team. When Intel acquired Digital's StrongARM assets, Witek and Hoeppner declined to join Intel and formed what was to become Alchemy. Thus, ironically, Intel's StrongARM, and in part its Xscale products, were based on technology from the Alchemy co-founders — who today compete against their own designs.

Pompa said AMD is talking to Microsoft often about extending PocketPC OS support to the MIPS architecture. "It may not be PocketPC 2002 but the follow-on that brings us in," he said. Meanwhile, "a number of applications are becoming Linux- and Java-based," opening the door to MIPS. And Microsoft is driving CE.Net with plug-in wireless access.

"You have to be careful when you talk about OS support, because that is a 'snapshot in time' kind of situation, and there may be a different story to tell nine months from now. The automation and residential-gateway markets are not locked in. Some markets may be closed off because of software, but there are a lot of open ones. The dominant OS is still being determined by the market."

But other issues loom. For one, said Dataquest's Starnes, AMD has yet to work out its strategy for DSP support. Intel has the Frio core, co-developed with Analog Devices Inc., and Texas Instruments Inc. is pushing the Omap coprocessor for its DSPs.

And even in the MIPS marketplace, AMD faces stiff competition. "AMD has got to do better than NEC, Toshiba and Integrated Device Technology on the level of integration and packaging. They have got to have the right feature set," Starnes said.

On the plus side, he said, "the embedded world is not as black-and-white as the PC space is"; it's a given that there's more than one way to do things.