Yet another silicon-optics breakthrough is afoot, as a startup claims it's produced a 10-Gbit/s modulator out of silicon using mass-production techniques.

Luxtera Inc. is announcing today that it's built that modulator using complementary metal-oxide semiconductor (CMOS) processes. That means the device can be built using the manufacturing lines that already churn out millions of microprocessors and other chips, a high-volume factor that keeps the price of manufacturing down.

The 32-employee startup did this through three years of work with Freescale Semiconductor Inc. (NYSE: FSL), the former semiconductor arm of Motorola Inc. (NYSE: MOT). Luxtera's parts were built on the same assembly line that builds PowerPC chips for Apple Computer Inc. (Nasdaq: AAPL), says Cary Gunn, Luxtera's chief technology officer.

Luxtera's is the latest in a string of silicon-photonics projects bearing fruit after years of work. Researchers at the University of California at Los Angeles (UCLA) unveiled their silicon-laser work late last year. And Intel Corp. (Nasdaq: INTC) has produced a modulator and a laser in silicon (see UCLA Claims First Silicon Laser, Intel's Modest Modulator , and Intel Claims Laser Breakthrough).

Luxtera emerged from research at the California Institute of Technology (Caltech) in 2001. In two rounds of funding, it's picked up $22 million from the likes of August Capital, New Enterprise Associates (NEA), and Sevin Rosen Funds. Luxtera also got funding from the Advanced Technology Program (ATP) at the U.S. Department of Commerce in 2002 (see Hot Components Research Rewarded).

Unlike Intel, Luxtera is avoiding the difficult task of building a laser out of silicon. Even so, the modulator wasn't exactly child's play. "The key thing is to make very small, very well controlled features," CTO Gunn says.

It's a problem of scale. A CMOS semiconductor can have line widths of 0.13 micron, but by comparison, a fiber-optic cable carries light in a core that measures 10 microns across. Like water aimed at a bottle opening too small, the light "spills around the edges," causing unacceptable loss, Gunn says. Luxtera gets around this using a proprietary means of coupling the fiber and modulator, but details aren't being disclosed.

So, what good does it do to have a CMOS modulator without a silicon laser? The idea is to bring costs down. Silicon can produce smaller and cheaper devices than materials such as lithium niobate (LiNbO₃).

Luxtera also has plans on integrating other functions. Its modulator is a Mach-Zender device that works using waveguides, and that means the modulator chip can be augmented with other waveguide-based elements, such as filters and multiplexers. Based on lab work so far, the company believes it can "produce robust DWDM [devices] in silicon," says Alex Dickinson, Luxtera's CEO.

Intel, by contrast, is trying to get the whole optical shebang into silicon. "We're going to integrate all the stuff into silicon and bring down the whole cost of the transceiver," says Victor Krutul, Intel's senior manager of silicon photonics technologies.

The Local-Area Level
All this might be nice for the telecom market someday, but Luxtera and Intel are targeting enterprise networks instead, where there's more money to be made. "It just seemed to be a better financial opportunity," Dickinson says. "We think the value is in taking these technologies, which have typically been used at the long-range, and bringing them more to the local-area level."

In addition to making optics smaller and cheaper, Luxtera promises to bring higher performance to the LAN. Typical enterprise optics use 850nm wavelengths, which come in cheaper lasers. Luxtera is concentrating on the 1550nm case using single-mode fiber.

Similarly, Intel's interest lies in driving the use of optics as interconnects between systems or even between chips -- markets close to Intel's heart and more in line with the company's high-volume production. Make those into optical markets, and the volumes would dwarf telecom, Krutul says.

"In the long-haul or metro telecom markets, we're talking about units in the tens of millions," Krutul says. By contrast, "Intel makes over 50 million CPUs a year, and on top of that, you add chipsets" and other related devices, he says.

Luxera is claiming to have the upper hand on Intel so far. The Intel modulator was introduced at 1-Gbit/s speeds, and Luxtera officials point out that the device is too large and power-hungry to be a practical product.

Intel counters by noting that its results so far haven't been product releases. They've been research results published in the journal Nature. Intel does expect to turn the results into products eventually, and a spokesman says the team is working on getting the size and power consumption down accordingly.

As for the speed, Intel has gotten its silicon modulator to the 10-Gbit/s level and plans to publish those results next month, Krutel says.

Luxtera's first product samples -- an unspecified type of "integrated" device, Dickinson says -- are planned for the second quarter of 2006. Intel isn't disclosing its product roadmap.

— Craig Matsumoto, Senior Editor₃, Light Reading