Indeed, several companies have demonstrated outstanding optical performance with fabrics achieving very low optical losses. Although not everyone is aware that the technology is mature and commercially available, early deployments of new high-speed Internet and data networks in Asia and the United States already use it to implement an intelligent and dynamic optical network layer.

Moreover, while electro-optical switches are often placed at odds with all-optical switches, the two technologies are in fact complementary, with all-optical switches working at the network core and electro-optical circuit switches and routers aggregating sub-wavelength services at the edge of the network.

The 3-D-MEMS switching provides virtually unlimited switching capacity; no other optical switching technology is on the horizon. With the integration of ASIC drivers on one chip and of optical functions like variable attenuation, power monitoring or wavelength division multiplexing, improvements and cost reduction will continue.

Like tunable lasers, MEMS-based optical switches are a key building block for developing the next-generation network communication infrastructure, bringing us closer to realizing the goal of transparent and reconfigurable optical networks. After the successful deployment of optical amplifiers and wavelength division multiplexing in the 1990s for point-to-point transmission, optical switching represents a natural evolution toward all-optical networking.

The main technology driver is the need to sustain a continual improvement in increased capacity and cost reduction per transmitted and switched bit of information. Intelligent optical switching systems also pave the way to high-capacity and dynamic services, such as bandwidth-on-demand.

The concept of 3-D-MEMS switching is simple enough. A beam of light from an input fiber bounces between two planes of mirrors that couple the beam to an output fiber (Fig. 1.). Each mirror can rotate around its two axes, interconnecting any input fiber to any output fiber. The fibers are collimated with an array of lenses to control the size of the beam propagating in free space and to minimize optical losses.

In an optical switch configuration with arrays of MEMS mirrors and collimated fibers, each mirror moves freely around two axes to switch light in three dimensions.

An example of a 256 x 256 packaged optical switch, the DiamondWave 256 x 256 photonic switch from Calient Networks, shows the two half-switches consisting of a collimated fiber array aligned to a mirror array collapse in the form of a handshake (Fig. 2). The insertion loss of this core switch is around 1.5 decibels typical and 3 dB maximum. The underlying technology is more complex, requiring many critical technologies, not only in MEMS fabrication and optical design but also in packaging, electrical and optical interconnections, mirror control, assembly and testing.

For instance, because the production of 256 x 256 switches requires hundreds of thousands of insertion-loss measurements per day, an adequate testing infrastructure is a manufacturing necessity. In addition, the achievement of low optical losses and high reliability demands an optimization between the optical design MEMS fabrication and packaging. For example, the mirror size, pitch and angle of deflection must be judiciously chosen to minimize optical losses and sensitivity to mechanical vibration, while allowing the beams to move freely between the two collimated fiber arrays. And good antireflection coatings are key to switch the entire spectrum of single-mode-fiber (1,260 nm to 1,625 nm) with little wavelength or polarization dependency.

There are many design decisions to make while developing a 3-D-MEMS switch. MEMS mirrors etched out of a single-crystal silicon wafer provide higher aspect ratio and structural integrity over polysilicon surface fabrication, more easily making the mirror flat and therefore reducing optical losses. Electrostatic actuation limits power consumption compared with electromagnetic actuation.

The choice of the electrode configuration is also critical. Parallel plate MEMS mirrors suffer from a snap-down problem: If the mirror rotates too far, the actuation force grows exponentially and the mirror sticks to the electrode. This limits the angle of operation of the mirror to single digits, which inturn limits the scalability of the switch fabric. Also, reliability issues arise when the switch is subject to external vibration. Comb-drive actuation avoids these problems, providing much higher actuation forces with larger surface and shorter distance to the electrodes. As a result, double-digit deflection angles are easily achieved.

3-D-MEMS mirrors have shown outstanding reliability, which can be explained by the simplicity of the switch and the low number of elements. In accelerated aging tests, more than 2 billion switching cycles are routinely performed with no failure. The MEMS switches have been subject to office and shock tests up to 500 g's with no effect.

MEMS mirrors can sustain such shocks because shock sensitivity is proportional to mass. The smaller the switch and the optical path length, the better the performance, cost and reliability. The small size yields many mirror and lens array dies per wafer, thus reducing cost. Short paths optimize optical performance and limit vibration sensitivity.

The 3-D-MEMS switches offer many promising applications, but their deployment has not been as widespread initially as was anticipated a few years ago. The reasons lie in the end of the recent technology bubble, delaying the need for high-capacity optical switches. But they also are being deployed more slowly because all-optical switches affect the rest of the network, and they have to be designed carefully to allow a smooth integration.

The DiamondWave 256 x 256 packaged photonic switch from Calient Networks connects any of 256 input fibers to any of the 256 output fibers.

For those reasons, Calient Networks chose to vertically integrate the steps for developing switching systems, from MEMS fabrication to networking software, and opted to work closely with telecom carriers and service providers.

At least as important as understanding the core technology is appreciating the value that the 3-D-MEMS technology can deliver to a product that meets an application's specifications.There are two types of applications for common optical telecommunications: standalone fiber switching and core-network wavelength switching. The former already has several niche applications in lab and automated test equipment. But it is the latter that ultimately promises the highest value proposition for providing a converged and intelligent optical-network backbone.

An important part of that is the ability to dynamically set light connections throughout the network. This involves software integration into a common control plane with Internet Protocol routers and Sonet/SDH circuit switches. Much effort has been expended by such groups as the International Telecommunications Union and the Internet Engineering Task Force to standardize automated switched optical-network architectures and such open-network protocols as generalized multiprotocol label switching.

The ability to reduce the network overbuilds with more efficient optical protection and restoration mechanisms is also attractive, but initial deployments must fit with the mode of operation of current networks that support 60-ms service restoration. Mirror control mechanisms must perform fast and smooth switching on the order of 10 ms.

Also, input and output power monitors must be integrated to detect failures. More advanced performance-monitoring capabilities, like optical signal-to-noise ratio and bit error rate, can be shared economically using the nonblocking property of the switch matrix.

Finally, switching systems must exhibit high availability to support large-scale deployment, having capacities of hundreds of terabits per second.

Recent earthquakes have provided occasions to demonstrate the solidity of the switches. In one instance, bit-error-rate testers were connected to an active switch during a nearby earthquake that reached 5.2 on the Richter scale. The testers measured no errors.

Olivier Jerphagnon is a product manager at Calient Networks Inc. (Goleta, Calif.).