Foundry deals shaking processor-IP model at its core

Foundry deals shaking processor-IP model at its core
By Crista Souza, EBN
March 6, 2000 (11:08 a.m. EST)
URL: http://www.eetimes.com/story/OEG20000306S0016

Demand for portable communications is driving growth throughout the semiconductor industry, but the opportunities opening up on the embedded-processor front are causing a fragmentation that threatens to shake up the ranks of established suppliers.

The 32-bit embedded-RISC market, clearly dominated by ARM Ltd. and MIPS Technologies Inc., is under siege by an army of small players firing a volley of alternative cores at the hundreds of fabless system-on-a-chip designers that have gone largely ignored by the leaders.

Start-ups such as ARC Cores Ltd., Tensilica Inc., and Lexra Inc., as well as Internet-based freeware movements, are offering customers a wide assortment of competing architectures that promise to make an end run around the market's more venerable cores.

Hoping to bolster their dominance of the high-end embedded space, ARM and MIPS are relaxing their narrow licensing strategies and are separately mass marketing their intellectual pr operty through pure-play foundries.

In the first of what for ARM will be a series of foundry partnerships, the company last week authorized UMC Group-Taiwan's second-largest semiconductor manufacturer-to produce chips based on the ARM7TDMI core.

Previously, ARM worked only with integrated-device manufacturers under a licensing program that allowed it to maintain stringent quality standards. As the popularity of its core increased, however, the din from fabless chip suppliers clamoring for ARM's IP grew too loud to ignore.

"The demand got so large, we were compelled to respond to these requests," said Reynette Au, vice president of worldwide marketing at ARM, Cambridge, England. "With the huge success of the foundry model and the trend toward system-on-a-chip integration, a larger number of companies have the ability to create very advanced designs, which is where the ARM core does well."

ARM's is among the cores most demanded by UMC's foundry customers, according to Jim Ballingall, vice presi dent of worldwide marketing at UMC (USA) in San Jose.

The company's strategic shift comes at a time when the processor-core market is more competitive than ever, and foundries-the new repositories for commercial IP-are the levelers between the manufacturing haves and have-nots.As a result, ARM and MIPS have come to accept independent mass producers as a viable distribution channel. MIPS last year aligned with Singapore's Chartered Semiconductor Manufacturing Pte. Ltd. to boost its market penetration. The Mountain View, Calif., company is currently testing 0.25-micron MIPS32 4Kc cores "hardened" to Chartered's fab process.

UMC is initially licensed to manufacture the ARM7TDMI. Customer designs using the 0.25-micron core are already in production, and the foundry is accepting orders for 0.18-micron designs for production by the first half of 2000, according to Ballingall.

ARM will grant single-use licenses to use UMC as a foundry for specific designs, as opposed to the reuse licenses it traditional ly has granted to customers that manufactured their own chips, according to Au.

"By optimizing implementations of the ARM7TDMI to UMC's process, we're creating access to a core that is tested, proven, and can basically be dropped into a design with assurance that it will perform to the specifications of the design," she said.

Together, ARM and rival MIPS lay claim to 54% of the $6 billion licensable 32-bit RISC segment. However, swelling demand for powerful system chips designed on ever-shorter cycles has given rise to a host of alternative architectures that threaten to undermine the companies' market positions.

The new architectures are compelling, observers said. Their silicon-independent format gives fabless customers the freedom to shop around for the best foundry prices. Furthermore, these synthesizable cores can deliver ARM-like performance for less cost and risk, suppliers claim.

London-based ARC Cores Ltd., for example, is touting a 32-bit configurable processor with DSP extensions t hat is designed to weaken ARM's hold on the cellular-communications market by aiming squarely at 3G wireless designs. Its claim to fame is that its code is user-customizable, giving designers more control, said vice president of marketing Jim Turley.

"The difference is, the designer owns the changes-he doesn't have to go to the semiconductor vendor and say, 'Do the work for us,'" Turley said in a recent interview. "With ARM, the vendor brings the differentiation. With ARC, the core itself is the differentiator."

The list of choices continues to grow. Last week, a fringe group calling itself OpenCores surfaced on the Internet, offering free downloads of a 32-bit RISC clone. While traditional processor vendors are skeptical of what OpenCores can deliver, interest in such community-source programs is significant.

Sun Microsystems Inc.'s Microelectronics Division has had 1,200 downloads of its picoJava core in RTL format, and 1,500 downloads of the embedded microSparc IIep, according to Fadi Azhari, group marketing manager for Sparc and community-source licensing at the Mountain View-based company. How many will result in royalty-bearing designs is uncertain.

Still, ARM remains the clear front-runner in the 32-bit embedded-RISC market, and according to analysts, has an overwhelming lead in design wins for future SOCs. Based on 1999 figures, about 70% of cellular handsets are ARM-based, and a good portion of MP3 players have an ARM CPU, said Tony Massimini, an analyst at Semico Research Corp., Phoenix.

"It's not the only alternative, but it's got a very strong foothold-and not just in one part of the market," Massimini said. "If it runs on a battery, there's a good chance it will have an ARM in it."

With a slew of portable consumer-targeted communication devices on the horizon, there's tremendous opportunity for a lot of companies, not just ARM and MIPS, he said.

"But keep in mind, there's a huge gap between ARM and MIPS and everyone else," he added. "It will take quite a bit for the alte rnative architectures to catch up and build that kind of momentum in terms of development tools, application software, test, support chips. ... Infrastructure is everything."

Additional reporting by Mark Hachman