By fielding its own tunable lasers, Intel joins a small but growing club of companies betting that carriers will prefer to buy laser modules that can transmit many wavelengths of light across a wide spectrum rather than the single-wavelength laser widely used today.

While the economic downturn has decimated demand for lasers of any kind, Intel executives insisted they were jumping into the arena for a sustained period and expect to see profits despite the recession.

The tunable technology comes after several years of stockpiling optical expertise through acquisition, as Intel sought to strengthen its comms business. The company said it is now working with 15 leading telecom companies in optical communications and calls the tunable laser technology further proof that it can no longer be considered a neophyte. "When we visit customers, we don't get questions about what we're doing in optical," said Mike Ricci, co-vice president and general manager of the optical products group at Intel. "It's more like, 'Let's talk about road maps.' "

By creating a device that acts more like a standard, single-wavelength laser, officials said, Intel has overcome some nagging problems associated with today's tunables. Intel's product is an external cavity laser with a high output power of 20 milliwatts, enough to meet most system specifications today. "An external cavity is the best way to get high output power, so you can do most of the link without amplification," said Gary Wiseman, director of marketing in Intel's optical products division.

With no moving parts, the external cavity meets power, package specs.

Having sufficient power has proved a problem with early tunable lasers. "The more you build in tuning, the more you compromise semiconductor properties," said consultant Jay Liebowitz, president of Liebowitz Strategies (Newton, Mass.). "There's more loss inside the laser so you would need to drive it harder, which causes more heating. And the more it heats up the more inefficient it gets."

The device has no moving parts, the company said, which improves reliability and underscores a key difference from tunable lasers that use microelectromechanical systems (MEMS) to select their wavelength actuators.

In Intel's product, the main components include mirrors, an integrated thermal tuner, a gain chip, an isolator and a number of flexures that precisely align the laser to the fiber before it is mounted in the package. The components are said to take up little space and can be packaged in a compact "butterfly" can.

Working samples of the tunable laser modules are being tested now at Intel's lab in Newark, Calif. Intel said it will disclose further details when it demonstrates the technology at the Optical Fiber Communication Conference and Exposition in March.

Intel expects to ship the first modules using the new laser technology by the end of 2003 or early 2004, a spokeswoman at the company said.

Single spare, multiple wavelengths

Carriers have several reasons to embrace tunable lasers, experts say. The most immediate advantage: Carriers can purchase just one type of laser instead of many versions with different wavelengths, helping to reduce inventory costs. Tunable lasers can also replace many spare single-wavelength lasers that serve as network-node backups, a practice known as "sparing."

Beyond that, tunable lasers could be used to build more flexible networks where traffic can be redirected by switching the colors of light. "You can start to talk about networks where you can change wavelengths on the fly," Liebowitz said.

Among Intel's competitors is four-year-old Agility Communications Inc. (Santa Barbara, Calif.), which has fielded its own laser that integrates most of the components in one indium-phosphide-based semiconductor device — including the mirrors and tuner — and packages it in a butterfly can. Like Intel's laser, Agility's device has a wide tuning range and can meet 20-milliwatt power requirements, said Arlon Martin, vice president of sales and marketing at Agility.

Agility's lasers can switch channels in 10 milliseconds, well within the 50 milliseconds needed to meet the Sonet restoration specification. Intel has not disclosed the switching speed of its tunable laser, but Martin expects the chip giant to be "a very good competitor."

Manufacturing leverage

Intel plans to separate itself from its competitors with manufacturing expertise. In semiconductor technology, the company soon expects to leverage its 90-nanometer process, which will incorporate high-speed silicon germanium for many physical-layer functions. Standard CMOS can be used for other components.

The goal is to build components needed to bridge electrical and optical signals of a laser similar to the way Intel makes digital logic. That means finding a way for optical components to keep up with Moore's Law.

"We have the capability to do in 90 nanometers that which can't be done with processes today," said Gordon Hunter, co-vice president and general manager of Intel's optical products group.

But Intel will continue to buy InP-based laser chips from outside companies because they are plentiful. "The cost of a laser chip today is less expensive than a laser driver," Wiseman said.

To assemble the devices into the laser can, Intel will leverage an automated manufacturing technology that it acquired in buying Light Logic (Newark) in 2001. Starting with a metallized ceramic substrate, an automated pick-and-place machine populates the substrate with the transmitter, receiver and other components.

To align the fiber to the laser, Intel uses proprietary flexures on which the components are adjusted before a technician welds them into place. This process takes about one-tenth the time a technician normally spends aligning a laser by peering into a microscope and making adjustments. Manufacturing yields are greater than 90 percent, said Joe Young, director of operations for Intel's optical platforms division.

Intel executives insist they're in the optical business for the long haul and say they expect to turn a profit before the telecom recession comes to an end.

Still, Intel has yet to prove there's enough demand for lasers, tunable or not, to push down manufacturing costs. Since the downturn, demand has fallen from millions of units a year to a couple of hundred thousand. Tunable lasers will make up just 3 to 5 percent of the market this year, though demand is growing, said Agility's Martin.

"We've seen other attempts at automated manufacturing. Unfortunately, the volumes didn't materialize," said industry consultant Liebowitz.

"The question is: How do you match the right manufacturing technique at a given volume and when will the volume be large enough to justify increased automation and ways to achieve greater economies of scale?" he asked.