A Fast Switch
Work on a high-speed optical switch could provide one component for future optical communications networks.
Jade Wang's office is nothing fancy, just a cubicle in one corner of a lab within Lincoln Laboratory's Optical Communications Technology Group. But it's here that Wang, who just this year earned her PhD, is collaborating with electrical engineering professor Leslie Kolodziejski from MIT’s Research Laboratory of Electronics on finding ways to speed up optical communications. Wang's particular contribution to this complex problem is a method that may help make optical switches practical.
As more and more information streams along the Internet, it strains the capacity of the fiber-optic network. One way to get more data through the system, of course, is to increase the bit rate—and indeed, the latest communications hardware can handle 40 gigabits per second, up from the current standard of 10 Gb/s.
But there's a problem. After coursing speedily through the optical fibers as pulses of infrared light, data transmission comes to a screeching halt when signals get to network nodes, where they have to be rerouted into a different fiber. The routers and switches at these nodes read the data packets' information to figure out their destination and send them along the right paths. To get that information, these devices need to perform logical operations on the data stream, and they must do it electronically—optical computing devices are not yet practical. Converting the optical data to an electronic signal, examining and rerouting it, and then converting it back into light takes time and requires additional equipment, slowing networks and driving up costs. "There's a bottleneck there," Wang says.
The color-coded map shows constructive and destructive interference as a
function of bias current on the two semiconductor optical amplifiers (SOA 1
and SOA 2). Constructive interference between the two interfering
light beams, shown in deep red, results in high optical transmission (a
digital 1), while destructive interference, shown in blue, produces
low optical transmission (a digital 0).
The optical switch that Wang is developing is based on interferometry: the incoming optical signal is split in two, and each component travels along a separate channel before recombining. If the signal in one of the channels is processed, it can speed up or slow down. Depending on how much the signal was changed, when it meets the signal from the other channel, it can interfere constructively, making the combined signal stronger, or destructively, making the combined signal weaker. If this merged signal is stronger, that counts as a digital "one;" if it is weaker, it’s a "zero."
The signal-altering devices in these switches are semiconductor optical amplifiers (SOAs)—lengths of a semiconductor material, usually indium-gallium-arsenide-phosphide, that work much like a laser. Pumping an electrical current through the SOA produces charge carriers (negative electrons and positive holes). An optical pulse traveling through the electrically energized device causes the electrons and holes to recombine, emitting photons with the same characteristics as the original pulse, thereby amplifying it.
The SOA is made of nonlinear material—that is, the optical signal passing through it alters the material’s index of refraction. So if a second pulse follows right on the heels of the first, it is slowed down or speeded up by an amount proportional to the change in index. Therefore, Wang says, "I can control how much index change I get by controlling how much power I send in."
To create optical logic, the researchers build an interferometer with an SOA in each channel. A control pulse passes through one, altering its index of refraction. A signal pulse follows, passing through both channels and then recombining at the other end. The power of the pulses controls how much out of phase with one another the signal pulses are.
With the pulses adding up to ones and zeroes, the routers can now read the headers of data packets and decide where to send them without having to go through the electronic conversion step. Previous researchers have shown this design can perform switching at 80 Gb/s.
An SOA/interferometer optical switch was first demonstrated about a decade ago, but it was built out of discrete components on a circuit board—much too large for integration into optical networking equipment. Wang and company are working on integrating the same design into a much smaller package, making it easier to use and less expensive to produce. "We've gone from a 12-by-18-inch board to an inch-long device that performs optical logic," Wang says. The next step is to put multiple logic gates on a single chip. One issue the researchers are trying to tackle is that the optical pulses lose power as they pass through the device. The amplification effect of the SOAs can compensate for that, but using too many SOAs adds complexity and introduces noise that can interfere with the signal. "There may be a way to reduce that loss by being clever about fabricating the chip," Wang says.
For instance, light moves around the device through a passive waveguide; the SOAs, by contrast, are active waveguides, altering the signal. Chip makers can fabricate the passive and active waveguides out of similar materials, and determine which is which by changing the mix of materials slightly. Wang uses Kolodziejski's design, which moves the light from the passive to the active waveguides by slowly tapering the end of the waveguide; the change in shape forces the light to move from one waveguide to the other.
Wang is also focusing on creating a streamlined method for checking the devices’ performance. Because variations in fabrication produce devices that are slightly different from each other, each logic circuit has to be tested to determine what combination of current and signal timing gives the best match for constructive interference—the strongest "on" signal—and the best match for destructive interference—the strongest "off" signal. "The better we can turn things off compared to on, the better the switch operates," Wang explains. Wang has devised a way to measure the output power of the device at each current setting. Her technique produces a color-coded "bias map" of the device that shows the optimal current setting, or bias. Her method takes about 10 minutes, as opposed to the hours needed for earlier testing techniques. Better testing and better fabrication should eventually yield a reduction in costs that will help make these switches practical, she says. Wang's work, under the direction of her former thesis advisors Scott Hamilton (currently associate leader of the Optical Communications Technology Group) and Erich Ippen (a professor of both physics and electrical engineering at MIT) is part of the Defense Advanced Research Projects Agency's (DARPA) Data in the Optical Domain program.
According to DARPA, this program has, for the first time, shown a path toward a network that eliminates electrical-to-optical-to-electrical conversion at each router—a transformation that could potentially speed up transmission tenfold. Efforts like Wang’s will help to "enable new optical networks that can meet the growing Department of Defense need for bandwidth while also minimizing latency," says DARPA spokesperson Jan Walker.
Wang doesn't want to project how long it might be before optical switches actually make their way into communications systems. For one thing, researchers still have to figure out how to get multi-gate logic, which would allow data to pass through the gates sequentially, on a single chip. With such a serial approach, it's especially important to reduce signal loss in each gate. "Things with optical logic have been limited to a few gates, so the functionality's not really there yet," she says. And complex optical computation will require some sort of optical memory, which does not yet exist. But she's hopeful that one day an all-optical network will eliminate the electronic bottlenecks and data will whizz around the Internet at speeds much greater than are now possible.