3D MEMS-Based Optical Switch Handles 80-By-80 Fibers

Cost-effective transparent connectivity is possible thanks to an optical switch core module that routes light from any of 80 input fibers to any of 80 output fibers. Developed by Glimmerglass, the palm-sized subsystem is based on 3D microelectromechanical (MEMS) mirror arrays. Called Reflexion, it can switch signals within 10 ms with an insertion loss of 2 dB ±1 dB, specifications that are well within the Telcordia requirements for communications applications. The nonblocking switch supports single-mode fiber and operates from 1260 to 1630 nm.

The Reflexion subsystem switching core is packaged in a small volume (Fig. 1). According to Glimmerglass, the cost per port should be five to 10 times less than what has been possible for an all-optical switch. Also, Reflexion complements next-generation optical communications systems. It's currently used in test and measurement applications, 10-Gbit/s Ethernet, grid-computing, business continuity, and even high-definition video applications. All of these applications benefit from the company's new transparent connectivity offerings, which are compatible with any network topology, every protocol, and any data rate that delivers deterministic, very high-bandwidth connections at significantly lower costs.

Other companies have designed and tried to market high-density 3D-MEMS-based all-optical switches, and they all have come up against the same technical hurdles: acceptable performance and reliability at cost-effective per-port prices. Compounding their challenges is the economic slide blanketing the telecommunications market.

What makes Glimmerglass different from the other companies who have tried but have not been very successful? "We've delivered real commercial products instead of science experiments. We spend our money bringing solutions to our customers, not on marketing events and hype," says CEO Mark Housley. "That's why we're thriving in a tough market." The three-year-old company has been shipping products to paying customers for over a year and possesses a core of experienced principals in microphotonics technologies.

One of the most notable 3D-MEMS-based optical switch products that ran into market difficulties is the LambdaRouter from Lucent Technologies, a 256- by 256-port switch introduced about three years ago. Lucent discontinued its development about a year ago. Calient Networks produced a 256-by-256-port switch, the DiamondWave. While it's now being tested in the field overseas, it's unclear if there's been any actual shipping to any customers. Optical Micro Machines (OMM) worked feverishly on producing a similar product, but it has since gone out of business. Other smaller companies have either tried or are continuing their efforts to break into the all-optical high-density switch market.

For a relatively small number of ports (16 or less), many 2D, planar-oriented optical switching architectures have been developed. But in 2D designs, the number of required switches, such as micro-mirrors, squares as the number of fiber ports. In other words, N ports would require N² switches. This could be insurmountable with a large number of ports, particularly considering realistic device yields. In addition, 2D designs have significant path-dependent optical losses, particularly with a large number of switches. Yet 3D designs scale switching elements as 2N, accommodating hundreds, even thousands, of ports. All the while, they still maintain path-dependent optical losses of less than 1 dB. That's why Glimmerglass chose a true 3D MEMS design (Fig. 2).

The key to Glimmerglass' success is the tradeoff the company made between the design of the MEMS mirrors and the rest of the optical system, two considerations that generally aren't complementary, according to Paul Alioshin, vice president of engineering for Glimmerglass. "We made the MEMS design as simple as possible. We designed the MEMS elements to be as easy as possible to fabricate using realistic manufacturing tolerances," he says. "The challenge for us was to solve the mechanical and optical issues involved."

"The biggest problem in optical switching is that the tolerances involved are very tight," Alioshin continues. "And when you add in Telcordia specifications, the problem becomes even more difficult to solve."

The Reflexion's MEMS design is very simple. A micro-mirror array rests atop a single piece of bulk silicon on a ceramic substrate (Fig. 3). No bonding pads or other integrated electronics exist on the chip. All routing to the rest of the actuation electronics is done on the back end of the ceramic substrate. Additional electronics, located on a photodetector card, include −50- to 20-dBm logarithmic amplifiers, 3-kHz antialiasing, constant-delay two-pole Bessel filters, and 14-bit (6000 counts/dB) analog-to-digital converters with a conversion-phasing time of 5.2 µs. The DSP-based control system is fed serial low-voltage differential-signaling (LVDS) data.

Parallel-plate electrostatic actuation of the mirrors, with potentials of about 200 V, was chosen for its manufacturing simplicity, even though it provides less force than alternative actuation methods. ASIC high-voltage linear amplifiers provide sufficient electrostatic actuation. And unlike other designs that servo mirror amplifiers on a round-robin basis, Reflexion servos all amplifiers simultaneously over a high bandwidth of 35 to 50 Hz.

Glimmerglass' MEMS engineers developed new processes to dramatically improve the flatness and uniformity of the micro-mirrors in their arrays. Variations in the nominal mirror curvature were reduced by a factor of 6, to 0.5 m⁻¹, and intra-array uniformity was improved to 0.2 m⁻¹. The result is lower optical insertion loss, improved loss uniformity, and 20 times less thermal sensitivity.

Another key element in the Reflexion core's design is a sophisticated and patented (U.S. patent 6556285) DSP-based mirror control system to optimize light coupling from input to output fibers (Fig. 4). Furthermore, there's a patented (U.S. patent 6484114) calibration method employed for free-space-coupled fiber-optic transmission.

The company also applied for a patent for its method to actuate a two-axis MEMS device using just three electrode elements. An open-loop slewing algorithm and synchronous detection not only move the mirrors to factory-calibrated positions, they also detect errors from optimum light coupling and apply corrected voltages to the mirror electrodes. The open-loop mirror trajectory is optimized to reduce residual mirror ringing at the end of a slew. Each mirror's voltage trajectory is formed by convolving a step function with a set of impulsive shaping functions, each of which is intended to yield zero ringing after the slew for a specific mirror's mechanical resonant frequency.

The Reflexion 80-switch pack is available as either a small core module tailored for communications-systems manufacturers or as a larger evaluation system. The evaluation system includes the Reflexion module as well as optical-channel power monitors, a fiber patch panel, and an Ethernet control interface. It's all packed into a 4.5- by 9.5- by 7.0-in. rack-mountable case that consumes just 20 W. The company provides all of the necessary development tools and engineering services to help customers build solutions.

The price for the core module ranges from $200 to $700 per port.