ll-optical rings form an express lane for the rest of the network. They shunt traffic from one major city to another, ultimately handing it off to a conventional electrical network. And after more than a year of vague promises and hype, these rings are starting to get built.

Broadwing Communications (Austin, Texas) this month completed its three-ring all-optical network spanning the United States. And last September, Global Crossing Ltd. (Hamilton, Bermuda) began running traffic on a trial line using the WaveStar LambdaRouter, the all-optical switch that Lucent Technologies Inc. (Murray Hill, N.J.) began shipping in July.

But optical networking has some growing up to do. The new equipment required for all-optical networking is expensive and most of the technologies behind it are still undergoing carrier tests. And because the shift to optics actually takes away some crucial network capabilities, particularly in the area of management, the transition raises concerns that this whizz-bang technology might not be useful in every part of the network.

"What I find fascinating in a young industry is that a lot of things are done because you can do them, not because you need them," said Tim Cahall, chief executive officer of Trellis Photonics, who termed some of the optical experiments going on in the industry "science projects."

For years, telephone companies have been using optical signals, translating the zeroes and ones of a particular message into infrared light pulses carried by fiber-optic cables. At the receiving end, the message is translated from light into electrons for reading by traditional electronic circuits.

The problem is that along the way, the light signal needs to be regenerated, and that process required electronics to read the signal. Hence, intermediate stops had to be installed where the light (optics) would be converted to electricity, then back into light-an "OEO," in industry parlance.

Moreover, an OEO stop was needed any time a piece of the signal had to be split off, a situation that became particularly relevant with the advent of wave-division multiplexing-allowing different wavelengths of light to carry different messages on one cable.

The dream of all-optical networking is to eliminate the OEO steps, creating so-called "OO" switches that transfer light without translating to electrons in between. Or, at the very least, to create a way to carry light signals farther without having to regenerate them electronically.

Broadwing and others also see all-optical technology as a way to handle data at a more granular level. For example, conventional systems allow only 20 to 30 percent of the traffic to be dropped at a particular point, compared with 100 percent for the optical equipment from Corvis Corp. that's used in the Broadwing network, said Dale Richardson, director of transmission-systems engineering at Broadwing. "As we kept building rings, as we kept adding colors to the system, it became extremely complex," he said. We were stranding bandwidth."

Provisioning in the network core is different from doing it at the access level, because much larger chunks of traffic are involved, Richardson said. It's easy for an inexperienced operator to allot too much bandwidth for a particular stream, creating the "stranded bandwidth" problem faced by most service providers.

Mirror mirror
The most common method of all-optical switching involves microelectromechanical systems, in which an array of tiny mirrors reflects light from one port to another. Lucent was among the first to announce a MEMS-based optical-switching technology, and others quickly followed.

The 256 x 256 array seems to be the size preferred by switch makers, with 1,000 x 1,000 the next step, said Philippe Marchand, project manager at Optical Micro Machines Inc. (San Diego), which has been shipping MEMS switches in sizes up to 16 x 16. "Then when you talk to [companies such as] MCI, they say they really want a 4,000 x 4,000," he said.

Optical Micro Machines sees a need for letting switch makers build to intermediate sizes as desired. "There's a clear path for things starting at 256 [x 256], maybe even 128. But the key is, can you offer some level of modularity?" Marchand said. That's because carriers are accustomed to paying only for the capacity they use, adding more capacity by putting extra line cards into existing systems. The all-optical switch as yet doesn't offer that option-a 1,000 x 1,000 switch will cost the same regardless of whether all the capacity is used.

Optical Micro Machines expects to ship prototypes of the 256-input switch fabric this year. Marchand was reluctant to give a time frame for the 1,000 x 1,000 switch, noting that it's unclear when such a system will be needed. Carriers are interested in hearing about such systems but not in buying them right away, he said.

Many early MEMS-based systems incorporate mirrors that pop up when they need to reflect light. One mirror is needed for every possible combination of input and output ports, so that an N x N switch uses N2 mirrors.

To get to larger sizes of switches-1,000 x 1,000 being a common milestone-it's necessary to move to so-called 3-D mirrors. These can swivel to access multiple ports, so that only 2N of them are required for an N x N switch. Each input port gets a pair of mirrors; one deflects light toward the proper output line, while the second "centers" the beam to hit the output line dead-on, maximizing the strength of the light captured.

"The thing to be careful of is that the mirrors aren't digital any more. They're analog. That adds another layer of complexity," Marchand said. "You need a very linear driver behind each mirror so you can be sure you are moving the mirror in a predictable way."

Lucent is shipping 3-D MEMS in its LambdaRouter all-optical switch, which was shown in a 256 x 256 configuration at the Optical Fibers Conference in March. A 1,034 x 1,034 version is due for release late this year or in early 2002, said Eric Spurrier, vice president of product marketing for Lucent's optical networking group. Optical Micro Machines likewise is developing 3-D MEMS for 256 x 256 and larger systems, and Texas Instruments Inc. (Dallas) this month unveiled its own 3-D micromirror efforts.

Beyond MEMS
MEMS switching is a technology still in development, and critics are quick to note shortcomings that the industry is still trying to overcome. Hinges are a problem, for example, because they are roughly the same size as the mirrors.

Some companies are addressing these problems with MEMS-like products. Blue Sky and Texas Instruments, for example, move micromirrors using electromagnetic fields, avoiding hysteresis.

Others have diverged from MEMS entirely. Agilent Technologies Inc. last year announced a method derived from bubble-jet printing processes, and Corning Inc. is among the companies developing switches based on liquid crystals.

Among the more recent entries in the field is Trellis Photonics (Columbia, Md.), which has devised a holography-based method using Bragg gratings inside a special type of crystal. The company said its approach eliminates the need for collimation, the technique of putting a lens at the end of each fiber that connects into the switch. Collimation is used with some MEMS switches to reduce the distance each beam needs to travel.

In place of a MEMS fabric, Trellis uses fiber Bragg gratings-light-reflecting grooves-etched in a proprietary material that keeps the gratings dormant unless a voltage is applied. Once activated, the grooves will deflect a particular wavelength of light.

The process takes 10 nanoseconds, CEO Cahall said.

"If you imagine a 2-D MEMS with the pop-up mirrors, superficially it looks the same," Cahall said. But Trellis can activate multiple crystals in a row, each reflecting a particular wavelength of light, whereas mirrors reflect all wavelengths.

The first row of crystals demultiplexes the incoming signal, splitting it into individual wavelengths. And the reflected signals exiting the fabric are remultiplexed, combined into one beam. Finally, by using crystals that lie between the "on" and "off" states, Trellis can perform optical attenuation, a smoothing out of the power levels of various wavelengths.

Real problems
"That inability to deal with the demux, remux and the power balancing in the network causes real problems. You can do this with other devices, but you start inducing loss," Cahall said.

Trellis' Intelligent Lambda Switch, made for carrying 240 wavelengths (six fibers bearing 40 wavelengths apiece) in a one-bay frame, is going into beta testing at the end of June, Cahall said. Trellis has plans to build systems as large as 1,920 wavelengths by the end of the year.

The data being passed through an all-optical switch isn't visible, a condition that the industry has termed "transparency." Because of that, finding the failure inside an all-optical network will be a cumbersome process driven by trial and error, Cahall said. "You've got six MEMS in the network, you put a good signal in, you get a bad signal out. What do you do?" he said. "This is the fatal flaw that keeps them from being deployed in the network."

Crystal-based switching methods such as Trellis' have a built-in mechanism for monitoring. "Due to a flaw in the Bragg grating, there's half a decibel that cannot be absorbed," Cahall said. That fraction of the light signal passes all the way through the fabric, since none of the crystals block it or reflect it.

Trellis takes advantage of this phenomenon by collecting the unused light into a bit-error-rate tester, which gathers light like a discard bin at the end of the switching matrix. The Trellis system then works backward to determine the strength of the original light beams and can quickly detect when a signal is weakening, indicating a trouble spot in the network.

Network management could be a critical enough concern to keep Trellis alive in a crowded market, Cahall believes. "At the end of the day, it's all that matters. Things that are not serviceable cannot be used," he said. "MEMS, with the moving parts-they've got good scientists, they've already shown that. But service providers are agnostic on technologies. They'll do it with cookie dough if they have to."