Integrating optics and electronics without MEMS

The remarkably rapid growth of the Internet and telecommunications markets has brought an infusion of money into optoelectronics research in the past year. Researchers are studying and developing devices using photons for tasks that have been completed by electrons in the past. Two of the myriad examples are work at MIT on developing light emitters in silicon, and at Agilent Technologies on developing and commercializing an optoelectronic switch that does not use MEMS.

An optoelectronic chip of the future would ideally include electronic circuits and optical devices (sources, waveguides, switches, and detectors) that could be grown at the same time via the same processes. Silica-on-silicon waveguides have been developed that provide on-chip waveguides, but sources can only be connected to silicon chips via hybrid manufacturing techniques that bond dissimilar materials (such as Si and GaAs) together. The difficulty lies in the materials properties of silicon: single-crystal silicon used for microelectronics barely emits any light. Ideally, light sources would be devices that could be grown on silicon so they could be fully integrated with electronic circuits on a single chip.

Fiber lasers and amplifiers exist, but these require meters of length, which is an even less reasonable requirement for an optoelectronic chip than the hybrid designs now used. The materials properties of fiber lasers, however, provide a springboard for research now being done by Michal Lipson together with Kevin Chen and Lionel Kimerling at MIT in Cambridge, MA.

Erbium-doped fiber amplifiers entered the telecommunications market remarkably quickly within the past decade, because they provided a way to increase the signal-to-noise ratio optically, rather than having to convert the optical signal to electrons for processing before re-converting it to a stronger optical signal. The long interaction lengths required for the fiber amplifier are not a major problem for telecommunications systems. The lengths are required because erbium has a small optical cross-section -- it neither absorbs nor emits light efficiently at room temperature.

Lipson is using microcavities in silicon to boost the efficiency of erbium. Instead of passing light through a long fiber once, the researchers are investigating whether a very efficient cavity could cause pump light to travel several meters as it bounced back and forth before escaping the cavity. To create such a device, they are embedding an erbium-doped gain material, either Er:Si or Er₂O₃, into cavities bounded by distributed Bragg reflectors (Figure 1).

Figure 1. Transmission electron micrograph of an active erbium-oxide layer between distributed Bragg reflectors. Image courtesy of Xiaoman Duan.

There are obvious long- and short-distance applications for this kind of work. While hybrid devices are available, the integration is a costly process -- if lasers and amplifiers could be made using the same sorts of processing, with the same equipment, that chip manufacturers use now, then this work could lead to cheaper optoelectronic devices. Telecommunications manufacturers would welcome such a device.

The group is collaborating with D.A.B. Miller at Stanford Univ. (who is working on hybrid devices) on optical systems for chip-to-chip communications as part of the Interconnect Focus Center, a multi-university collaboration based at Georgia Tech for developing faster chip interconnection technology.

If the work can provide practical emitters on silicon, these two applications could filter through the basic technology in computers and telecommunications equipment. "It could change the whole face of electronic chips today," Lipson said.

Figure 2. A commercial 32 X 32 optical switch based on inkjet and planar waveguide technologies. Image courtesy of Agilent Technologies.

Photonic switches are already changing the face of telecommunications, but most are based on MEMS technology, involving movable micromirrors. Last spring, Agilent Technologies (Palo Alto, CA) announced the development of an unusual optical switch without any moving parts. Agilent (formerly a subsidiary of HP) is commercializing 32- X 32- and dual 16- X 32-port switches based on the technology (Figure 2). The switches are compact; the 32 X 32 device is about the size of two dimes. While the product was announced in March, the device has not yet shipped in large quantities.

The switch combines inkjet technology and planar waveguides, and is a hybrid silicon and silica device. The silicon section includes thermal actuators -- i.e., small heaters. The silica section includes solid optical waveguides intersected, at the cross points, by trenches filled with index-matching fluid. When the device is flip-chip bonded, the heaters are located near the cross points of the waveguides. Inkjet technology uses small heaters near a liquid (in printers this is an ink, in the photonic switch it is the index-matching fluid) to create a bubble at that spot.

When not switching, the heaters are not turned on, and the beam (and the signal encoded onto it) passes unimpeded straight across the trench and back into the solid waveguide.

When the optical signal needs to be rerouted, a bubble is created, filling the trench near the waveguide. Julie Fouquet, the lead inventor and project manager for the switch at Agilent, said, "Total internal reflection occurs when light comes to a surface at a sufficiently oblique angle from a higher index medium to a lower index medium" such as the waveguide to the gas bubble inside the trench. The beam reflects off the surface of the bubble, and moves in a new direction into a different waveguide (Figure 3).

Figure 3. Waveguides in the switch are interrupted by trenches filled with index-matching fluid. Unswitched beams pass straight through the trenches and back into the waveguide. Switched beams bounce into the new waveguide by total internal reflection off a bubble that overfills the trench. Image courtesy of Agilent Technologies.

Switching time for the device, including the control software, is 10 ms. "This is very important for SONET systems," Fouquet said. "Telecom service providers want to be able to change a connection within 50 ms so they don't drop calls. They need to be able to do the hardware part in 10 ms or less."

Crosstalk in this technology is very low compared to traditional planar waveguide technology. The commercial 32- X 32-port switch has a specification of -50 dB crosstalk. "The switching operation is digital," Fouquet said, "either the signal reflects or it doesn't."

"All the complex alignments within the device occur during manufacturing using mask aligner-type instruments," Fouquet said. Standard photolithography and flip chip bonding have the tolerances to get the angles and positions correct.

"The switch," Fouquet said,"is conceptually simple. Switching is digital and there is no need for complex analog control. You don't need good aim to operate it," as opposed to some micromirror switch designs. Nor, once manufactured, will the optical alignment change.

The switch has no moving mechanical parts, which may translate to longer lifetimes and better reliability. Operating voltages are pretty standard: the device runs on 15 and 5 V DC.

Fouquet said, "You get add ports and drop ports for free -- the switch architecture is flexible and allows signals to be added and dropped. By contrast, she said, in a beam-steering MEMS switch additional mirror pairs must be added to accomplish these functions, resulting in a larger switch.

The device can switch several wavelength channels on a single fiber at once. Alternatively, a wavelength demultiplexer can separate the channels (from Los Angeles, for example) before the switch so the device switches individual wavelengths. One could then drop the signals in channels 5 and 19 at the local node (Phoenix, for example) while adding new signals in channels 5 and 19 before the signal is multiplexed for subsequent long-haul transmission (to Dallas, for example).

The technology is scalable. When the company asked customers how big switches should be built, it learned that one big switch is not necessarily best. "They handle the possibility of a switch failing by splitting the signal at the rack and sending the signals to both a regular and backup switch," Fouquet said. If the main switch fails, then the goal is to change over to the backup switch and repair the first switch as soon as possible. "If you have one humongous switch, it's really hard to get it out quickly and put in a new one with so many fiber connections. It's risky for an unskilled operator to do it in the middle of the night." The potential for confusing the fibers is alarmingly high.

With Fouquet's switch, a large switch is built up from smaller modules, each connected to fewer fiber arrays, which makes replacing a module less complicated than replacing a large switch. Therefore, the mean time to repair a switch should be shorter for a modular switch than a single big one.

The 32 X 32 and 16 X 32 switches can be built up to 512 X 512 size for a strictly nonblocking architecture. For wavelength-selective architectures, the switches can be built up to 1000s X 1000s.