Chip startup Aèluros Inc. is out to prove it's got what it takes to build physical layer (PHY) chips using standard CMOS silicon technology at speeds up to 10 Gbit/s serial (see Aeluros Emerges From Stealth).

Today, vendors usually turn to more exotic materials like silicon germanium or gallium arsenide to reach serial speeds of 10 Gbit/s or above. Electrons move faster in these materials, so it's easier to build high-speed circuitry out of them. But the performance comes at a price -- literally.

Using standard CMOS (complementary metal oxide semiconductor) process technology would be more desirable, because it results in circuits with lower power consumption, as well as being less expensive. In the past few years, this has started to look possible. The speed of CMOS technology has improved with each new generation of CMOS process technology, based on smaller transistors.

"The next frontier for CMOS is 10-Gbit/s serial links," says David Gamba, Aèluros's VP of marketing. The industry is in broad agreement on this. The question is: How soon?

Chip heavyweight Broadcom Corp. (Nasdaq: BRCM) is one of the few vendors delivering high-speed CMOS today. Yet, Aèluros is claiming it can deliver a power reduction of 30 percent and significantly improved jitter performance, compared to solutions from Broadcom.

Power is crucial, says Gamba. Some PHY devices in production consume nearly 3 Watts of power. Considering that emerging optical module standards like XPAK and X2 have a maximum power budget of only 4.5 W for the entire module, this clearly isn't good enough.

Of course, Aèluros is probably comparing its nascent technology with older products from Broadcom. But Broadcom, which employs around 3000 people, said it couldn't find anyone to give us details about its products.

Building 10-gig CMOS chips isn't as easy as some would make out, adds Gamba. A lot of companies have tried and failed to get the performance they need, even after several years of development. Vendors have quit work on 10-gig CMOS chips in recent months, suggesting that it's not an easy technology target. Startup Transpectrum closed down a few weeks ago (see Transpectrum Flunks Out ).

According to Gamba, and others in the industry, even the early chips from Broadcom suffered from too much jitter, or electrical timing noise -- problems which it may have corrected in later versions.

So what makes Aèluros think it can succeed, and even surpass the likes of Broadcom? Gamba says it's all down to the skills of the engineering team: "What surprises people is the fact that as speeds go up, the skill set required on the team changes. Now you need RF engineers, packaging experts, and so on. The folks that put together the right team have the best chance of success."

Aèluros's founders have worked on 10- and 40-Gbit/s chip development at Agere Systems (NYSE: AGR/A); high-speed device modeling at Atheros Communications; signal integrity at Rambus Inc. (Nasdaq: RMBS); and high-speed packaging at Sirenza Microdevices Inc.

Rather than rushing to bang out a product real fast, Aèluros has instead started with a couple of test chips, to ensure that device modeling gives the results it expects. One of the major challenges of designing high-speed CMOS, Gamba contends, is that the models supplied by the foundry aren't accurate enough at high speeds. "Now we have the models, it will speed up our time to market considerably."

Aèluros's second test chip, called Bobcat, integrated four 10-Gbit/s transcievers, to prove that it was possible to integrate multiple 10-gig circuits on the same chip. Aèluros hasn't disclosed what its first product will be, but does say it is likely to be released early next year.

Founded in June 2001, the startup has received $10 million in funding from New Enterprise Associates (NEA) and WorldView Technology Partners.

— Pauline Rigby, Senior Editor, Light Reading
www.lightreading.com