The chip startup is opening a third front in the ASIC/FPGA war, targeting the communications space

Craig Matsumoto, Editor-in-Chief, Light Reading

September 5, 2006

6 Min Read
How Cswitch Got Programmed

Strictly speaking, Doug Laird doesn't come from an FPGA or network processor background. Yet, he decided to move into both groups' territories with the recent launch of Cswitch Corp. , a startup building what he claims is the ideal ASIC/FPGA hybrid for telecom.

Cswitch, which debuted in July, is set to begin sampling chips in the fourth quarter. (See Cswitch Debuts Switch Array.) Regardless of whether the company succeeds or gets ground into powder, its story is a prime example of how Silicon Valley's startup machine can work, and how venture capitalists prime so-called serial entrepreneurs for their next big ideas.

Laird is a processor guy, best known as a founder and executive of Transmeta Corp. He started Cswitch in 2003 by following the money. He says it all started by asking a question: Which semiconductor businesses are raking in the cash?

Answer No. 1 was microprocessors -- but Laird wasn't going to take on Intel Corp. (Nasdaq: INTC) again. "Transmeta was a long, hard road," he says now.

Next on the list were analog chips, a difficult market where success has to be spread across dozens of obscure devices. It worked for Linear Technology Corp. and Maxim Integrated Products Inc. (Nasdaq: MXIM), but it's not a rich vein for startups. Strike Two.

That brought Laird to the FPGA (field programmable gate array) market, where he found light at last in the numbers behind Altera Corp. (Nasdaq: ALTR) and Xilinx Inc. (Nasdaq: XLNX). "They make as much money, if not more, per part than Intel. Why is that?" Laird says. "We started looking into this as an academic exercise."

Laird was getting paid to think like that. In 2002, he'd become an entrepreneur-in-residence (EIR) at the Silicon Valley office of Charles River Ventures . Different firms define "EIR" differently; in Charles River's case, an EIR gets to help the firm evaluate other startups' technologies while, on the side, formulating ideas of his or her own. The position goes to entrepreneurs while they're between startups.

"Doug helped me evaluate dozens and dozens of startups in the [chip] sector," says Bill Tai, a Charles River partner. "We came to the conclusion that this particular segment -- which is essentially programmable silicon in the communications space -- is both underserved and about to go through disruptive change."

Other Charles River EIR projects have included Avvenu Inc. and M2Z Networks Inc. , which were hatched by former 8x8 Inc. (Nasdaq: EGHT) president Keith Barraclough and Milo Medin, a founder of @Home, respectively. It's worth noting that M2Z made Light Reading's coveted Top Ten New Startups list. (See M2Z Networks.)

Charles River is hardly alone in this. Take the example of Stretch Inc., started in 2002 by Gary Banta, an EIR for WorldView Technology Partners . Mulling the future of processors, Banta teamed up Tensilica Inc. chief engineer Albert Wang to develop a programmable chip built on Tensilica's microprocessor core. (Loosely put, a "core" is a fragment of a chip; companies like Tensilica sell them as building blocks to be used inside ASICs or FPGAs.)

Laird, meanwhile, had settled on the FPGA idea but still had to find something to do. He'd worked at ASIC firm LSI Logic Corp. (NYSE: LSI), giving him some exposure to the ASIC/FPGA rivalry -- but he needed to investigate where FPGAs might provide an opening. Laird would spend the next nine months talking to OEMs, trying to find an angle for a startup.

What Laird found so interesting about FPGAs was the fact that all vendors cited communications as their top market. The sector accounts for nearly half the revenues at Altera or Xilinx. But communications vendors told Laird the chips run too slow. "It's a fundamental problem which requires a change of architecture to solve," says Charles River's Tai.

The simple answer would be to drop FPGAs and go straight to ASICs -- but ASICs were just getting too expensive. "You couldn't bet that a product you'd bring to market would run in sufficient volume to cover the cost," Tai says.

A device with the right qualities of each, it seemed, could be as popular as a prom queen. And that's been known for some time -- for years, FPGA vendors have been adding more cores to their chips, making them more ASIC-like, while ASIC vendors have been adding reconfigurability.

The ASIC/FPGA tug-of-war has become a popular area for startups. Ambric Inc. debuted its reconfigurable processor arrays at last month's Hot Chips conference. And stealth startup Tabula Inc. is said to be doing work on crossing FPGAs and ASICs.

"It's a lot of people having the same idea at sort of the same time -- which doesn't mean there are too many companies," says Linley Gwennap, principal analyst of The Linley Group . "An ASIC or FPGA for one group isn't necessarily going to compete with an ASIC or FPGA for another group."

After repeated visits to OEMs, the possibilities began to take shape for what would become Cswitch's architecture -- the idea being to focus on the kind of packet handling network processors tackle. "What they wanted was a much faster chip that could do this packet processing," says Ed McKernan, Cswitch's VP of business development and one of Laird's earliest recruits.

So, Laird launched Cswitch in 2003, eventually raising $41 million in two rounds -- the most recent being a $30 million round in April -- from ATA Ventures , Bay Partners , Charles River, GF Private Equity Group LLC, Harris & Harris Group Inc., Mitsubishi UFJ Capital Co. Ltd., Masters Capital, and Micron Ventures.

CSwitch's product, the Configurable Switch Array, is a telecom-specific FPGA. It contains an array of programmable logic but also columns of prefab circuitry designed for telecom applications -- content-addressable memory or specialized arithmetic units, for example. The prefab elements are nailed down in the chip, but some of them can be configured by the OEMs.

A lot of the chip's inner workings are devoted to moving information from one section to another quickly, and those elements draw heavily from the processor architectures of Laird's past. "This is stuff we've been doing in processors for many, many years," he says. "The difference here is that we had to make it configurable."

Neither Altera nor Xilinx would comment on Cswitch. But Laird notes that one of them, at least, sees the potential in Cswitch's idea. (The other one says it stinks; Laird isn't telling which is which.)

Cswitch compares itself mostly to FPGAs, but its chip also treads on the turf of high-end network processor vendors such as Bay Microsystems Inc. , EZchip Technologies Ltd. (Nasdaq: EZCH), and Xelerated Inc. . Admittedly, those startups' success has been limited, but they've also got a multiyear head start on Cswitch as far as getting into customer labs and gaining favor with OEMs.

In this milieu, Cswitch might find its limitations. Network processors offer flexibility in software, which can make them tough to program but can be a more versatile approach, in the long run, than Cswitch's reconfigurable hardware. "Ultimately, I don't think you can do as many things with the Cswitch architecture as you can with a network processor," says Linley Gwennap.

— Craig Matsumoto, Senior Editor, Light Reading

About the Author(s)

Craig Matsumoto

Editor-in-Chief, Light Reading

Yes, THAT Craig Matsumoto – who used to be at Light Reading from 2002 until 2013 and then went away and did other stuff and now HE'S BACK! As Editor-in-Chief. Go Craig!!

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like