x
Optical components

Luxtera Launches Silicon Optics

Startup Luxtera Inc. has begun sampling its photonic chips, with the company now saying its future could be in consumer electronics as well as in data-center communications.

Luxtera has talked all along about applying its chip to the data center, where it would pitch the smaller size and lower cost of silicon-based optics. But the company now says it also wants a shot at becoming the optical interface between TVs and other equipment like DVD players.

"Years down the road, that's hopefully a high percentage" of revenues, says Marek Tlalka, Luxtera vice president of marketing. "Near term, it's going to be communications."

Founded in 2001, Luxtera began talking last year about its plans for photonics based on complementary metal-oxide semiconductor (CMOS) processes. In other words, its components would be built of "normal" semiconductor material and could be mass-produced on any old chip-building line. Photonics typically are built from more exotic materials such as indium phosphide (InP) or gallium arsenide (GaAS). (See Luxtera Chases Silicon Photonics.)

Now, the 50-employee company has started sampling a part that essentially combines two 10-Gbit/s transceivers on a chip. The CMOS die includes all the electronics needed in an XFP module and most of the optics, including the laser modules, photodetectors, and transimpedance amplifiers.

The lasers themselves aren't built in CMOS -- instead, Luxtera uses off-the-shelf, 1550nm, InP lasers and bonds them to the CMOS chip. Silicon-based lasers are still not commercially viable, although companies including Intel Corp. (Nasdaq: INTC) have been working on the concept. (See Intel Claims Laser Breakthrough.) What makes the hybrid model work is the low losses, about 1 dB, incurred between a laser and the CMOS chip.

The whole package, named the Aurora LUX21010, measures 20 mm by 12 mm, making it smaller than any 10-Gbit/s module on the market. It eats up only 1.5 W of power while allowing for distances up to 2 km.

This is just a calling card, intended as a sample for prospective customers to prove Luxtera isn't just smoking its own silicon. "We're sensitive to the fact that this is a new technology," says Luxtera CEO Alex Dickinson.

The "real" chips, due out in the fourth quarter, will have more ports -- Luxtera won't say how many.

As for those video applications, Luxtera is eyeing the High-Definition Multimedia Interface (HDMI) version 1.3, which would boost speeds to 10.2 Gbit/s from 1.5 Gbit/s today. More immediately, though, Luxtera is aiming for data-center markets, hoping to eventually get its chips situated inside XFP modules (where there would be "a lot of space left over," Tlalek notes) or posted directly onto network interface cards.

Among Luxtera's first targets is computer clustering, an area inhabited by InfiniBand and its notoriously thick bundles of copper cables. As InfiniBand moves to 10 Gbit/s, Luxtera hopes to convince OEMs to move to fiber, avoiding the hassle of an even thicker InfiniBand cable.

The InfiniBand crowd has some answers of its own, though. Some companies are already offering fiber interfaces for higher speeds. "Fiber would definitely make things a lot easier on the installation side," says Bill Lee, a product manager with InfiniBand chip firm Mellanox Technologies Ltd. (Nasdaq: MLNX) "Unfortunately, it's still prohibitively expensive." Still, companies are trying the idea; Zarlink Semiconductor Inc. (NYSE/Toronto: ZL), for example, announced a parallel optics product for speeds up to 20 Gbit/s. (See Zarlink Boosts InfiniBand.)

There's hope for InfiniBand cables, though, as companies such as W.L. Gore & Associates Inc. are working on slimming things down. "For shorter distances, you can go with a thinner cable, and there are a lot of vendors shipping 30 gauge cable, which is about as thick as Cat6," Lee says.

Luxtera has raised $46 million in two rounds, "a lot of it still in the bank," Dickinson claims. The company picked up $5 million in revenues last year from military and other custom work, such as a 2005 project for Sun Microsystems Inc. . (See Luxtera Lights Up Sun.) Luxtera expects to do $7 million this year.

— Craig Matsumoto, Senior Editor, Light Reading

redface 12/5/2012 | 3:43:44 AM
re: Luxtera Launches Silicon Optics "What makes the hybrid model work is the low losses, about 1 dB, incurred between a laser and the CMOS chip".

Are you talking about laser to fiber coupling efficiency? And it is only 1dB? That would be a record. Are you sure?
Pete Baldwin 12/5/2012 | 3:43:41 AM
re: Luxtera Launches Silicon Optics >>>Are you talking about laser to fiber coupling efficiency? And it is only 1dB? That would be a record. Are you sure?<<<

No, not laser-to-fiber. 1dB is the loss from the InP laser onto the CMOS chip. The light beams down onto the chip via these pyramid-shaped couplers, and the loss there is, apparently, around 1dB.

Dunno the loss from the CMOS chip onto the fiber.
deauxfaux 12/5/2012 | 3:43:39 AM
re: Luxtera Launches Silicon Optics Gunn's article from Photonics Spectra (March 2006, page 66) says 3dB...which does make sense at room temperature. Over temperature, lots of variation will occur. Losses in the modulator, coupling, waveguides, etc. will be on top of that. For short distances in a "cable" that probably doesn't matter.

Still looks like it needs a hermetic package.

But I really wonder why it would be less expensive as a 10G part. The modulator in the Luxtera design is essentially "free" but so is the modulator in a DML or an EML. Why do you need to couple the yields of the "optics" and the circuitry when wire bonds are so cheap?
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE