Light Strategy

Intel pushes to make optical nets more efficient

Intel is springing into optical networking with chips that can correct transmission errors on the fly and quadruple the reach of fiber across a metro area.

The worlds biggest chipmaker has been quietly building capacity to enter the optical realm with eight acquisitions in 18 months, including last weeks announcement that it will pay $748 million to buy Xircom and its technology to link mobile devices to optical networks.

A key acquisition was the $1.25 billion Intel paid last March for Copenhagen, Denmark-based Giga, a maker of processors that can handle traffic at 10 gigabits per second to 40 Gbps, and the linchpin of Intels plan to quicken and extend the reach of optic fiber across a metro area.

Intels specialized network processors hold the promise of dramatically lowering the cost to a carrier jumping into a new metro market and to existing carriers looking to build up their networks, said Tony Stelliga, general manager of business planning in Intels telecom components division.

Analysts said the competition is formidable, yet Intels leap into the market in the next few months just might change everything about how networks are built in the future.

Intel said it will be the leader in bringing Forward Error Correction (FEC), a decades-old algorithm technology, to the metro space, extending the reach of lower-cost fiber and lowering equipment costs by as much as 90 percent.

FEC allows data to get scrambled a little bit — but not enough that the errors cant be corrected by smart decoding and smart algebra. The decoding algorithm, attached to Intels IFX30001 FEC transceiver sitting near the optical switch, wraps around bytes of data. Like a code book, it can figure out whats out of place in the ones-and-zeroes pattern and make corrections so the information does not have to be retransmitted.

Being able to speed traffic to the point that data actually collides a bit is the only way to get the most robust transmission on an optical network, Stelliga said. The fewer ports and platforms needed to descramble and reamplify the optical system, the more money can be saved.

Intel believes FEC will be the key piece in its chip sets ability to extend the reach of the lower-cost fiber for metro areas from about 12 miles to 48 miles. With Intels smart chips, a carrier wouldnt have to purchase as much expensive equipment from systems vendors, Stelliga said.

"We can extend the reach of equipment beyond the normal footprint by allowing the signal to degrade more and then correcting it. It can bypass expensive SONET [Synchronous Optical Network] rings designed to carry voice," he said.

Quiet Determination

"Im impressed with how Intel has been defining and implementing its communications semiconductor strategy," said Jay Patel, an analyst at The Yankee Group. "Theyre not making a lot of noise, doing it very quietly and shrewdly."

"A lot of [equipment companies] see Intel as a company that can change the competitive landscape," said Sean Lavey, research analyst at International Data Corp. "Things would dramatically change if Intel or one of its competitors could inspire stronger adoption for the network processor instead of the big boxes from Cisco [Systems], Nortel [Networks] and Alcatel. It would really lower the barrier for others to develop a far less expensive system around an off-the-shelf, programmable network processor."

Currently, the key equipment on optical networks is routers and switches made by dozens of companies, including Alcatel, Avici Systems, Cisco, Lucent Technologies, Nortel and Sycamore Networks. If the chip companies can make network processors sufficiently sophisticated, they could replace some of the functions of the traditional routers and switches, would cost less and could be offered off the rack. That way, start-ups without the hardware engineering expertise or deep-pocket capitalization of the giants could more easily compete with them in metro areas, Lavey said.

The metro area, where most of the office parks housing the corporations that play on the Internet reside, is where the action is. But the largest metro areas are so big that optical signals typically must be reamplified when going from, say, NBC in Manhattan to CNN in New Jersey. Each time that equipment has to be housed in another colocation center, costs escalate.

Intels eight acquisitions give the company a complete package to vie with the likes of Broadcom and PMC-Sierra as a supplier of chips for the switches and platforms that power optical networks, analysts said.

The market for systems for optical networks is expected to grow from $31 billion in 1999 to $90 billion in 2003, according to analysts at RHK.

Intel is laying plans to move into optical networks as demand for personal computers lags. The chipmaker cited a decline in PC sales last week as it forecast a 15 percent decline in revenue for the first quarter, compared with the $8.7 billion in sales the company had in the fourth quarter of 2000.

Most of the large network processor companies are working on FEC for high-speed traffic, but are awaiting the decisions of a standards body, said Armond Hairapetian, senior director of optical transport at Broadcom. When metro areas ramp up to 40-Gbps speeds, "you absolutely have to have it," Hairapetian said.

Intels acquisitions have been successful so far in giving it the expertise to jump into the optical market, Patel said. "I dont think theyve had any blowups yet. Theyve had a good strategy," he said.