MEMS are appearing in many forms - as large-scale optical cross-connects, RF switches, tunable capacitors, inductors, resonators, variable optical attenuators and filters.

Analog Devices, Lucent Technologies, OMM Inc and Onix Microsystems are all staking out territory in this new area with MEMS-oriented research projects.

"People were looking at MEMS to solve some of the most difficult problems in the telecommunications infrastructure," said Jim Doscher, director of optical MEMS business development at Analog Devices Inc. The constant demand for bandwidth has made scalability a prime factor in network design, and MEMS components often are able to deliver the required capacity.

"As you look into the future you have the potential [with MEMS] for all of the applications that people have always talked about and wanted to do but didn't have the bandwidth to do - videoconferencing, video-on-demand, faster Internet access," Doscher said.

Fibre optics has appeared as the solution to the bandwidth crunch, but current systems still rely on conversion to electronic form for switching, and scaling up these systems has proved difficult.

"A few years ago people ran the signals at sub-gigahertz speed," said Doscher. "Now we are at 2.5Gbits/second and trying to get to 10Gbits/s and then to 40Gbits/s. Every time you make one of these jumps, you have to replace all of the electronics in the cores. This is tremendously expensive, and the cores are complicated because they have to be tuned to a particular protocol" - Ethernet, ATM and voice.

"So there is this vision for keeping the signals in the optical domain," Doscher said. One way to do that is to use electromechanical components to control such optical effects as reflection and refraction of light. Not only does that eliminate the electronics from the data stream, it turns out to be transparent to the signal content itself. For example, a mirror knows nothing about the content of a light beam; it just reflects the light.

This type of control works best with high-capacity channels and is now being implemented in long-haul and metropolitan-area networks. However, with experience, companies will inevitably scale down the technology to smaller networks, Doscher said.

"The idea is to create a nationwide network of pipelines that is configurable on demand to meet the requirements of capacity," he said.

Companies like Analog Devices are building on existing experience to enter MEMS networking markets. "For example, we have been building automotive airbag accelerometers for about 10 years," Doscher said. "That [involves] taking a motion signal and converting through the MEMS device into an electrical signal. So it's mechanical and it's electrical, but now [in telecom] you're adding a third element" - optical.

Analog Devices has created a process that is tuned to optical MEMS. Called optical iMEMS (the "i" stands for integrated), the approach uses a bonded-wafer process to create mirrors out of single-crystal silicon with the desired mechanical properties. It is very flat and stable for the mirror device, the company reports.

ADI acquired another company last year to make sure it had a stable source of supply for the bonded wafers, Doscher said. "We plan to add some of the same features done with the accelerometers. It turns out that with these cross-connects, there is a great deal of value in having the electronics on board with the mirrors."

Analog Devices has demonstrated a micromirror-based device designed to make all-optical, 1,000-port cross-connect switch fabrics practical. The manufacturing process technology will make it possible for signal processing, actuation and multiplexing functions to be integrated onto a single chip, Doscher said.

Production experience needed

But while MEMS have tackled a wide variety of problems, not many R&D projects have produced commercial products. Pressure sensors, inkjet heads, accelerometers and display technologies are among the success stories. But only through volume commercial production will the thorny telecom issues of quality, reliability and performance be solved.

Proving the technology in commercial runs will be a key hurdle for optical MEMS, said Marlene Bourne, senior analyst for MEMS at Cahners In-Stat (Scottsdale, Arizona). "MEMS mirror arrays are relatively easy to design and demonstrate, but it's an entirely different matter to go into high-volume production," she said. According to Bourne, only three of the three dozen companies in this market are in production at all. Of that trio, OMM and Lucent Technologies are in low-volume production, and Onix Microsystems started production at the end of the first quarter.

A recent study by Venture Development Corp (Natick, Mass.) showed that many OEM users are skeptical of MEMS as a practical solution .

"The way the MEMS industry has evolved, everything has become application-specific. It is a very secretive , protective industry," said J. Eric Gulliksen, Venture Development's project manager. "A company goes into a MEMS fab and says, 'I need this kind of device,' [and the resulting development] becomes a proprietary design owned by them. That's why it is not mainstream."

Gulliksen believes certain elements of MEMS technology could be made mainstream and commoditised if there were organisations such as a standards bureau to do registration. "That would make it easier for companies to get into using MEMS without having to foot the bill with an expensive, proprietary design," he said. "It shifts the risk to the MEMS manufacturer, but it also gives them potential for a much wider market."

He noted that the MEMS industry could learn some lessons from the early years of the semiconductor industry. Semiconductors were assigned a Jedec number so that member suppliers could submit a version of a device that conformed to the performance and package configuration specification for that type. The Jedec convention introduced interchangeability and the possibility for multiple sourcing of a given IC.

A MEMs special-interest group is being formed now, according to Gulliksen.

"When it comes to second sourcing, Agere is big with multisource agreements," said Ray Nering, director of strategic marketing for optoelectronics at Agere Systems (Breinigsville, Pa.). "For functionality, we would like to work in some areas with a common footprint and protocol with these devices."

But Nering said it is possible for optical MEMS component makers to build their own markets. "Our focus is on 3-D MEMS, where we are interested in working with people who can scale to those kinds of dimensions," he said. "It may not be a 256 x 256 [port] device, because not everyone has that capability. But we are willing to talk about smaller devices."

Nering sees a push to bring switches closer to the edge of the network. In those cases, the switches tend to be a bit smaller - at 32 x 32 or 64 x 64 ports. "So, will it become as small as 16 x 16 or go up to a 128 x 128? We see people approaching the market in different ways." Agere is ramping up its 64 x 64 3-D optical switch (the 5200 series) to test the market.

"From our standpoint, even though we are using a different technology or someone else may be using a different technology, we think there is the ability to package the base technology in a way that makes it indistinguishable from another company's product," he said. That would make it possible for customers to have second-source options, even if there are no industry standards in place right now, said Nering.

"There may be an optical standard, determined by some type of industry standardisation at some time," he said, "but the footprint and any interface to it - basically the nuts and bolts to engineering a product - will be done by the players themselves."