Chameleon cuts staff, plans to improve processor

Chameleon cuts staff, plans to improve processor
By Patrick Mannion , EE Times
January 9, 2002 (9:26 a.m. EST)
URL: http://www.eetimes.com/story/OEG20020108S0079

MANHASSET, N.Y. — Chameleon Systems Inc. has laid off 50 percent of its work force and restructured its operations in an effort to address questions about the validity of the technology and business model for its Reconfigurable Communications Processor (RCP), which it called "a complete re-think of the signal processor."

Chameleon's problems reside in part with its tools, said Will Strauss, president of Forward Concepts, a consultancy in Tempe, Ariz. "It wasn't as though you could take your present software running on conventional DSPs and port it [to RCP]. You had to rewrite a lot of your code to utilize the reconfigurable capabilities. That can be a killer."

As a fabless IC company, Chameleon had also attempted to break into high-volume OEM houses while wrestling with the issue of reliable delivery. "There's no way a top-tier cellular company will buy from anyone except those who own the fabs," said Strauss. "ZSP went under for that ver y reason, despite having one of the best-performing products out there. Nokia wanted it, but wouldn't entrust volumes they needed to a small company." ZSP was subsequently bought by LSI Logic Corp.

Jeremey Donovan, chief analyst with Gartner Dataquest, said the company's difficulties related to technology and implementation. "Their trouble would be more with a fundamental technology limitation, like a performance problem, or a market-timing issue," he said.

Charlie Cump, a Chameleon founder and its vice president of sales and marketing, said, "We decided to take advantage in the downturn in the economy and the softness in the 3G marketplace to look at our product and improve its performance."

The company's flagship product, the CS2112, has been under evaluation by telecom companies for 18 months, and has created feedback that told Chameleon it needed to move to a more scalable architecture, Cump said. "We're going to increase the price performance by 40 to 80 times," he said. "We've also learned a lot about the operation of the reconfigurable fabric and will optimize it tremendously. We're working on the marketing requirements document right now."

The sole announced customer for the RCP is Lockheed Martin, which is currently prototyping a military wireless communications application.

Chameleon is also retargeting its processor to areas outside of communications. "We've had interest from other market segments, such as digital video and imaging, which tends to have its own set of requirements," Cump said. "We want to roll those into this as well so we can serve a broader market."

To meet customer demands in both telecom and video markets, Chameleon will migrate the CS2112 RCP from the current 0.25-micron process yielding 100-MHz processors to 0.18/0.15-micron technology to produce 300-, 400- or 500-MHz versions. "It's not a clean-sheet redesign, as we're keeping the basic architecture, basic cell blocks," Cump said. "We're focusing more on the pipelining and the registration between the c omputing blocks to support a higher clock rate." Pricing targets have not yet been decided. The 100-MHz RCP costs $100.

On the tools front, Cump said Chameleon will "architect a complete new software suite by adding higher-level design tools to allow the conventional DSP engineer to bring their algorithms and IP into our toolset at a high level and have that compiled down to our reconfigurable structure." To date, designers have had to operate at a lower level, similar to an FPGA. "This isn't comfortable for DSP engineers, so we're taking it up another level or two to let them get their designs done quickly," he said. Two other points of entry, starting at the assembly level, will also be provided.

Cump said he doesn't see the fabless model as a barrier. "We're using TSMC [Taiwan Semiconductor Manufacturing Co.], like a lot of companies doing off-the-shelf work," he said.

Chameleon is following a littered path. "Before Chameleon, Xilinx and Altera had large investments in reconfigurable processi ng and those projects were terminated," said Donovan of Dataquest. "Then the next generation of on-the-fly reconfigurable processor companies came along in the 1998 time frame," including Chameleon, Quicksilver, TriSend and others. Even if they fail, the concept of reconfigurable processing will "pop up again," he said.