But two new factors on the optics scene today may allow photonic designers finally to get in on the VLSI game. One is a broad research effort into "nanophotonics," where nanometer-scale optical devices are able to operate below the wavelength of light. The other, which will be an essential enabler of such devices, is the creation of planar-silicon optical-bandgap materials.

Eli Yablonovitch, who invented photonic-bandgap materials in the late 1980s while a laser researcher at Bell Communications Research, has a back-of-the-envelope computation that seems persuasive: "With the advance of photolithography, critical dimensions in silicon chips have now reached the point where they are less than lambda/4n = 100 nanometers, where lambda is the optical telecom wavelength 1.55 microns, and 'n' is the refractive index of silicon," he said. "This is the same as for electronic gate lengths, so the two technologies now match right up, and will scale together in the future."

Indeed, recent research in fabricating photonic-bandgap structures using 0.018-micron lithography has shown that virtually all optical components used in optical networks can be fabricated alongside silicon circuits.

Yablonovitch, who now heads the optoelectronics group at the University of California, Los Angeles' department of electrical engineering, founded a company, Luxtera Inc., two years ago to build optoelectronic chips as replacements for the racks of hybrid optoelectronic boxes that now switch and route optical networks. Luxtera (Pasadena, Calif.) has managed to raise $7 million in venture capital funding in a down market, which may say something about the practicality of Yablonovitch's approach.

Luxtera plans to build a few "nanophotonic integrated circuits" that will perform optical-switching functions now requiring large racks of components. The company's stated goal is to build these chips with standard semiconductor processing equipment.

The chips will be optoelectronic rather than all-optical. That runs up against conventional wisdom in the optical-networking world, where the drive has been to remove electronics entirely from optical switching. Dense wavelength-division multiplexing has been partially successful in doing this. It has been fairly easy to build optical multiplexers-demultiplexers to add or remove individual wavelengths from a fiber. Data is routed by wavelength rather than content, so these optical components do not have to read header addresses.

But these "transparent" networks can achieve only a coarse-grained level of routing. Sooner or later, the data content of signals needs to be decoded, and only electronics has been able to handle the task up to now. Yablonovitch is conservative on this point, asking why execute heroic projects to build optical transistors when electronics can already do that quite well. He doesn't believe that an integrated photonic-electronic circuit is going to be too slow, and with further scaling of the electronics, speed will be even less of a problem.

"There are a lot of problems with all-photonic circuits," he said. "For example, logic is very difficult to implement, memory is very difficult to implement. I don't see that anyone has ever gotten around those issues, and I don't think it is very likely that they will get around them."

Axel Scherer heads the Nanofabrication Group at Caltech and has been working closely with Yablonovitch. "I share Eli's conservative attitude. Electronics does many things well, so do the computation in electronics," Scherer said. "Optics has other advantages, such as high-bandwidth communication."

A critical new capability not available before is the ability of photonic-bandgap devices to shrink to the size of current optical devices. The tendency of light to scatter has been a fundamental barrier to scaling photonic devices. One breakthrough of the 1980s, the vertical-cavity surface-emitting laser, was made possible by the discovery of a "native oxide" for gallium arsenide: aluminum oxide. That allowed the light to be efficiently contained inside the laser cavity. And Yablonovitch first came up with the photonic-crystal concept in an attempt to prevent light-scattering lasers.

Scherer makes the point that the current shift toward integrated photonics has a lot to do with the ability of today's lithography systems to build two-dimensional patterns with a precision once achieved only with molecular-beam epitaxy crystal-growth processes. By patterning photonic-bandgap structures in two dimensions, a far more interesting and complex device geometry becomes available to photonic designers.

Scherer and his group have wasted no time in putting that capability to use. At the recent Optical Fiber Conference in Atlanta, they reported on the design and fabrication of lasers, modulators, add/drop filters, polarizers and detectors. The lasers have to be built in compound semiconductors, but the remaining devices can be built on a submicron scale on silicon. For most photonic functions, lasers are not required, Scherer pointed out. But for cascading chips, lasers become essential for creating optical gain.

As with electronic devices, the performance of photonic-bandgap (PBG)-defined devices improves as they scale down. Scherer reported a record quality factor of 20,000 in a tiny microcavity laser measuring only a third of a wavelength on a side. The ability of PBG structures to confine optical fields is also leading to even smaller-scale devices based on near-field techniques-the exciting field of nanophotonics.

MIT's 'photonic micropolis' illustrates many different 3-D photonic crystal devices tied by red 'roads' of one-dimensional periodic crystals.

Another scalable optical-confinement technique plays a role in these devices. "You get a maximum quality factor in cavities around half a wavelength on a side," Scherer said. "If you are willing to sacrifice some of that quality, you can go lower using surface-enhanced plasmon confinement on metal surfaces."

Not all photonic researchers share the conservative view of Yablonovitch and Scherer. One believer in photonic logic is Michal Lipson, who directs Cornell University's nanophotonics group. "We use photonic bandgap for modulating light on silicon," she said. "This will allow the equivalent to the transistor: an all-optical switch."

Lipson is working with an induced nonlinear effect in PBG cavities. Injecting free carriers into a silicon cavity has a strong effect on the index of refraction, she said, producing nonlinear optical effects that allow photons to strongly interact. The optical transistors can be defined using electron-beam methods, but that could switch to lithography methods to produce a manufacturable optical logic, said Lipson.

---

The roads not taken: '60s to '80s

The 1960s saw attempts to build digital computers using conventional optical components. The systems were promising, but bulky and power-hungry.

In the '70s, nonlinear optical materials appeared, enabling photons to interact strongly and produce efficient logic operations. But integration did not proceed smoothly, and optical systems have a difficult time with memory.

Compound semiconductor systems and an efficient means for building arrays of laser diodes emerged in the '80s. That led to new attempts to build digital logic processors merging compound semiconductor devices with silicon circuits via strained-layer lattice growth, liftoff schemes and techniques for gluing optical components to silicon chips.

An important spin-off was the development of SiGe technology, essentially a transference of compound semiconductor techniques into silicon. Had mainstream electronics switched to GaAs substrates, as some in the early '80s believed would happen, the industry might have a fully integrated optoelectronic technology. But silicon VLSI proved an unstoppable force.