Network-chip makers may be in danger of extinction
By Dean Takahashi
Red Herring
August 17, 2001
This article is from the August 15, 2001, issue of Red Herring magazine.
Rupan Roy used to design graphics chips for Chips & Technologies. Now he's chief technology officer at Cognigine, a network-processor company in Fremont, California. He's part of a wave of chip designers who have joined the migration to communications chips.
Follow the engineers; that's one way to identify the coolest new chip technology. The best and the brightest engineers are placing their bets on network processors, programmable chips tweaked to process Internet data packets. "You join up when most of the companies are hidden from view, and you trust in people who have what it takes," Mr. Roy says.
In the '80s and '90s, many engineers started in microprocessors, but the startups faltered as Intel (Nasdaq: INTC) took over most of the market. Next, the hot thing was designing 3D-graphics chips. From 1995 to 2000, more than 50 new 3D-graphics-chip companies were born. In the end, market consolidation left three main vendors. One graphics-chip maker, PixelFusion, decided it was too late to release its design as a graphics chip, so it transformed the product into a network processor and reëmerged as ClearSpeed Technology. The lure? A chance to sell a chip for a few hundred dollars, compared with $10 to $30 for a graphics chip.
Analyst Linley Gwennap, president of The Linley Group, a market research firm, notes the shift: "I started covering microarchitectures for microprocessors, but all of the cool designs are now being developed for network processors."
HURRY UP AND WAIT
There are 26 companies -- from Agere Systems (NYSE: AGR.A) to ZettaCom -- working on full-scale network processors, and many others working on coprocessor chips that serve as companions to network processors, says Mr. Gwennap. The theory behind the trend is that network processors will replace custom silicon in switches and routers because the processors are both high performance and programmable. The latter feature allows chips to be tweaked to adapt to ever-changing communications protocols. Network processors can also be deployed more quickly than custom silicon, which can take a year or two to design. The market is expected to grow from $180.6 million in 2001 to $567 million in 2003, according to preliminary estimates from the research firm IDC.
The problem is that "everyone is coming up with these elegant architectures, but the deployment of the equipment that will use these chips is being delayed," says Richard Doherty, an analyst at The Envisioneering Group, a market research firm. Because companies like Cisco Systems (Nasdaq: CSCO), Lucent Technologies (NYSE: LU), and Nortel Networks (NYSE: NT) are struggling, they are postponing investments in infrastructure. The question is, How many network processors will be stillborn?
Intel and Applied Micro Circuits Corporation (Nasdaq: AMCC) have snared hundreds of customers, but they have focused on the low end of the market. Four others are shipping network processors: Agere (Lucent's former chip unit), IBM (NYSE: IBM), Motorola (NYSE: MOT), and Vitesse Semiconductor (Nasdaq: VTSS). Mr. Gwennap says about 350 design wins have been awarded to date. While Intel aimed for the low-end OC-12 and OC-48 networks, which run at speeds of up to 2.5 Gbps, newcomers are hitting a sweet spot in the OC-192 networks, running at 10 Gbps. A few of the most daring companies (and latest to market), like ClearSpeed, are developing chips running at OC-768 speeds, or 40 Gbps.
Bob Merritt, an analyst at Semico Research, likes the battle plan of Cognigine, which is targeting the OC-192 market with a network processor that combines the functions of several current chips. But Roland Van der Meer, a partner at the VC firm ComVentures, says the winner will also need good marketing and deep pockets to stay the course over the dry period of the next 18 to 24 months -- something only a big company is likely to provide.
IS THE GAME THE SAME?
The shakeout will probably start in 2002, analysts say. By that time, some startups will be running out of money, and those with chips that aren't ready will miss the next round of customer design wins. Will it play out the same as it did for the microprocessor and graphics-chip markets?
"The network-processor space will not collapse as far and as fast as the graphics industry did," says Peter Glaskowsky, an analyst at Cahners Micro Design Resources, "partly because there are more segments within networking where companies don't compete with each other. Some will get bought, and most will go out of business."
Theoretically, acquisitions could save the day, but large companies may have already bought their favorites, several for more than $1 billion: Conexant Systems (Nasdaq: CNXT) bought Maker Communications for $1.15 billion, Broadcom (Nasdaq: BRCM) bought SiByte for $2 billion, and AMCC bought MMC Networks for $4.5 billion. Even Intel acquired network-processor businesses in deals with Digital Equipment and Level One Communications. These investments have yet to pay off. "One thing for sure is that the prices this time around in the acquisitions will be a lot lower," says Mr. Gwennap.
It's unclear if the big companies will belly up to the bar a second time to rescue more startups. So the engineers who flocked to network-processor upstarts may soon find themselves looking again for the next big thing.
Write to dean.takahashi@redherring.com. |
