At this week's Intel Developer Forum, senior photonics researcher Andrew Alduino gave an overview of his lab's work in that area. It's not exactly the most popular topic from Intel's research division -- Alduino's talk drew only six attendees, including two Intel speakers who'd arrived early from the next session.
That's to be expected, though. Intel's microprocessors are the stars at IDF, which is geared towards server designers and engineers. Intel's photonics work is geeky in a different way, and much of it won't be found in products this year or, in some cases, even this decade. "We're a longer-term research lab," Alduino said.
When it comes to optics, Intel has made perhaps too much fuss over things like silicon modulators (see Intel's Modest Modulator ). Passive alignment seems more pragmatic, because it involves extremely precise manufacturing techniques, an area where Intel excels.
Optical modules are expensive to build because the process is difficult to automate. The elements have to be aligned so that enough of the laser's light gets into the fiber. This is done through "active alignment," where the laser is turned on and its output gets monitored while an operator manually tweaks the alignment. The procedure produces fabulous results, but it's time-consuming and expensive.
With passive alignment, the process could become more like snapping Legos together -- no laser monitoring required. But that means the assembly technician has to trust the parts can line up perfectly, so that none of the active-alignment tweaking is necessary.
That's where semiconductor manufacturing can come in handy, because it involves the creation of microscopic structures etched out of a block of silicon. The right shapes can help make passive alignment work, as Alduino illustrated with a few textbook examples, all of which Intel has produced in the lab:
A common approach is to etch grooves into the silicon, to hold the fiber in place. These can be U-shaped -- rectangular channels resembling the Death Star trench in Episode IV -- or V-shaped.
- Laser attach
Lasers aren't made of silicon yet, so if you're building silicon-based integrated optics, the laser has to be built separately and glued atop the silicon substrate. By placing gold bumps on the substrate and the laser, the laser's positioning can be made predictable enough to avoid having to do active alignment.
An ultra-smooth 45-degree mirror can make a horizontal beam reflect "up" or "down." Useful in adapting certain lasers to certain module designs.
- Waveguide trenches
With properly shaped waveguides, one can change the shape of the optical beam, increasing its chances of hitting the target within the fiber's diameter. "We expand the beam enough so we aren't aligning a needle to a needle," Alduino said. "It's a garbage can to a garbage can."
But new optical manufacturing techniques have been slow to commercialize, now that easy funding has fled the photonics industry. Newport Corp. (Nasdaq: NEWP), for example, tried applying its chip-building expertise to photonics but had to scale back those plans as the market dried up (see Newport Still Believes in Fiber).
Those companies continuing the work admit the industry is still taking baby steps. For example, Arrayed has developed a way to stack semiconductor wafers precisely, allowing it to build vertically oriented modules en masse. Passive alignment is achieved using silicon tunnels -- chips with tiny holes -- that hold the fiber in place (see Arrayed Fiberoptics Intros Tech Platform). Nice trick, but the company is starting slow, with stacks of just two elements: a fiber and a lens, to make collimator arrays.
"The question is similar to the beginning of the IC industry. You had people asking how long it would take to create a television on an IC. You don't start there. You start with something simpler," says Ben Jian, Arrayed founder. On the plus side, Arrayed's collimator arrays are commercially available and will be in Thorlabs Inc.'s catalog next month, Jian says.
Assuming passive alignment finds its way into volume production, the next step would be to further automate the process, because it still requires a human touch. Alduino said Intel has a "clear vision" of how that automated tool should operate, but he added that Intel wouldn't be the one to build and sell such a machine. — Craig Matsumoto, Senior Editor, Light Reading