Electronics, photonics vie for optical-net switching

Electronics, photonics vie for optical-net switching
SAN DIEGO — One revelation from the SPIE's 45th International Symposium on Optical Science, Engineering and Instrumentation last week is the fact that silicon-based electronics has irreversibly moved ahead of photonic computing, at least in two crucial areas of information technology: logic processing and mass data storage. However, a third pillar, interconnects, is rapidly yielding to advances in photonics technology.

In a keynote address, Peter Delfyett, an optical researcher from the University of Central Florida, made a case for teamwork between the two technologies. "Photons can coexist easily since they have no charge, but that makes them impractical in logic applications where switching functions are needed," Delfyett said.

Optical networking is moving ahead rapidly because of the ability to multiplex photonic signals onto a single fiber. For example, commercial systems have already reached the multi-terahertz level, thanks to dense wavelength-division multiplexing (DWDM), which uses all optical components. Each multiplexed channel, however, is in the gigahertz range, an area easily reached with advanced electronic devices using silicon germanium or gallium arsenide compound semiconductors.

Delfyett believes that the two frontiers — optical multiplexing and high-speed electronic circuits — will continue to advance in a balanced way so that crucial logic functions can be performed with electronic circuits.

While some companies are pursuing an all-optical approach to network switching, Delfyett is convinced that such schemes will run into inherent size limitations imposed by the wavelength of light. Since photons have no charge, any strong interactions required for switching depend on interference techniques based on their wave nature.

"The components in any optical circuit cannot be shrunk significantly below the wavelength of visible light, which puts a lower limit on optical-circuit density," he said. In contrast, silicon-based electronics is moving rapidly into dimensions defined by deep-ultraviolet and X-ray optical sources, effectively escaping competition from all-optical information-processing systems, a least in the crucial factor of circuit density.

It is unlikely that photonic systems will ever be able to operate at such short wavelengths due to the difficulty of building shortwave lasers. In addition, as the wavelength of photons shrinks, it becomes more difficult to control them since they tend to penetrate most materials.

Essentially the same factor has entered the race between optical- and magnetic-disk data storage technologies. "At one point it looked like optical storage was going to move decisively ahead of magnetic-disk systems, but the basic problem with photons is that they are just too fat," said David Thompson, a magnetic-disk expert at IBM Corp.'s Almaden Research Center (San Jose, Calif.).

Optical storage has turned out to be commercially viable, since it is ideal for removable-disk products such as compact disks and DVD systems. However, magnetic-storage technologists have relentlessly pushed denser magnetic media and more sensitive read/write sensors using the giant-magnetoresistance effect. In contrast, optical-disk systems are limited to a basic data storage cell defined by how small a laser beam can be focused. In effect, that means a limitation imposed by the wavelength of visible light, just as in all-optical switching.

"Some researchers are proposing near-field optical techniques to get around the wavelength barrier, but that will require a read/write head that can fly just a few nanometers off of the surface," Thompson said. As for holographic storage, he said, the problem is finding a suitable medium. So far holographic media have have been plagued by short lifetimes, prohibitive expense or toxicity.

But another group at IBM's Almaden lab has been making progress in optical holographic storage. One advantage of that technology is the ability to read and write data in parallel. Typically a "page" of data — about 1 Mbyte of information — can be read out in one operation.

For data communications, keynoter Delfyett was confident that electronics and optics will form a team to take optical networking from its current terahertz level to a next-generation range of greater than 50-THz operation. Getting there will take devices with higher speeds and more flexible operations — in short, devices based on new physical principles.

One critical area is signal amplification. Currently, DWDM systems have benefited from the ease of amplification offered by the erbium-doped fiber amplifier. Essentially, this is a length of nonlinear optical fiber energized with an external light source. Any type of optical signal entering one end of the fiber is pumped with an internal laser response, emerging in an amplified form at the other end. The operation is simple to insert into a fiber system and has inherently low noise characteristics.

This type of device tends to be bulky since it requires a loop of fiber and the additional light source. Delfyett proposed replacing it with semiconductor optical amplifiers, which can switch at higher speeds, can be packaged more compactly and are able to be integrated on a printed-circuit board. In addition, they offer several other advantages, such as the ability to be electronically gated and to produce high gain, and they have the capability for wavelength conversion.

Tunable laser sources
Another area that badly needs development, in Delfyett's view, is tunable laser sources. Existing technology offers tunability over about 10 wavelengths with microsecond switching. These devices are expensive, running anywhere from $2,000 to $5,000. By 2003, Delfyett expects the same technology to support up to 100 wavelengths at the same switching speed and carry a price tag of around $1,000. By 2008, lasers operating at nanosecond switching speeds tunable to 1,000 wavelengths should be available for less than $500, he said.

To get to the required cost and performance levels of next-generation optical networks, several key system parameters will have to be improved. One is insertion loss, which occurs whenever optical components are connected to one another. Also, optical systems will have to be transparent to a wide range of wavelengths, not the few that now are preferred due to the current state of the art with optical components and light sources.

A related problem is preferred polarization. Many components are sensitive to the polarization of the light they are processing. That can lead to bottlenecks in design. The objective should be "polarization diversity," Delfyett said. As with electronics, systems need to scale down in size and power consumption while achieving higher reliability.

Switching fabrics
One type of component that will need a lot of improvement is the switching fabric. Optical-networking companies are now building mostly micromechanical-based optical switches. For example, Lucent Technologies (Murray Hill, N.J.) has devised a switch based on micromirrors. Agilent Technologies Inc. (Palo Alto, Calif.) has produced an integrated optical 32-channel switch based on waveguide technology.

The switches themselves are a novel bubble mechanism — a fluid at the intersection is heated, producing a reflecting bubble that redirects a given channel. Corning Inc. (Corning, N.Y.) is pursuing an optical switch fabric based on liquid-crystal light modulation. These first-generation optical switches redirect high-speed optical channels, but they switch slowly, due to their dependence on electronically controlled physical processes.

One thrust is to build all-optical ICs that can decode the routing information contained in Internet data packets and send them off in new directions based on their IP address. Many industrial and academic research labs are racing toward that goal with a wide variety of materials systems. Nonlinear optical materials such as lithium niobate, polymer thermo-optic systems, compound semiconductors such as indium phosphide, and photonic bandgap materials are some of the technologies that could support integrated photonics. However, due to cost and scaling issues, Delfyett was skeptical of this approach.

Another possibility is to use an emerging hybrid field: terahertz electronic/photonic systems. "For example, it is possible to place specially designed photo-detectors at the end of a terahertz fiber to generate a terahertz electromagnetic field," Delfyett said. "There is one problem, however: It is only a local field." Such high-frequency electromatic fields, like light, need waveguides to propagate efficiently.

Delfyett and his colleagues at the University of Central Florida are pursuing a combined WDM/time-division multiplexing approach to terahertz optical networking based on a new type of laser source they have developed.