Material systems based on silica on silicon, gallium arsenide, lithium niobate, and indium phosphide are contenders for the role of "optical silicon."
For roughly 15 years optical engineers have harbored a kind of microelectronics envy, longing for a day when very large-scale-integration technology could be applied to optical components as it has with electronic components. Our analysis suggests that it will be a long time before optical engineers have something like an optical microprocessor to work with.
But our research also indicates that a major trend toward integration exists that could lead to a 12% decrease in cost of equipment within five years. This type of integration—crude and low level by electronics standards—is taking various forms. One key difference between optical integration and microelectronics is that, while electronic engineers have found a philosophers' stone in the form of silicon, nothing with the same ability to embody multiple complex optical functionalities is yet available for optical integration, and may never be.
While many different roads to optical integration are being explored, and a wide variety of integrated optical products are already being sampled, silica on silicon is the nearest to a mainstream technology for photonic integrated circuits. Companies taking this approach include NTT Electronics (NEL), Hitachi Cable, JDS Uniphase, Corning, and Lightwave Microsystems, together with several startups. We are forecasting that the market for integrated components based on silica on silicon will grow from $1.5 million in 2001 to $282 million in 2005 (see table).
These forecasts do not include discrete arrayed waveguide gratings (AWGs), which are usually the first products to come out of the silica-on-silicon process. AWGs can be used for creating multiplexers and demultiplexers for high-end DWDM systems and they are already being shipped in production quantities by NEL, Hitachi, and JDS Uniphase. We believe that shipments of standalone AWGs will total $193 million by 2005 and that they will prove very successful for multiplexing once the number of channels goes much above 16 channels. This type of product is not an integrated component in any interesting sense. However, such products and the precursors of integrated optical components—especially combined AWGs and variable optical attenuators (VOAs)—are already being produced in sampling quantities.
Silica on silicon is the optical-integration mainstream today, but it is a mainstream with some obvious limitations. Because the functionality of silica on silicon is based on the thermo-optic properties of silica, it is relatively slow—millisecond switch times are expected. More important, it requires complex thermal controls to isolate the region being heated and, because silica needs a fair amount of heat to get it to do something useful, lots of power is needed to drive a silica-on-silicon optical circuit. Today, a 40-channel VOA would require up to 14 W to drive it. Widespread use of such circuitry could tax the power availability in a central office and would certainly lead to thermal management problems.
It gets worse as new levels of functionality—such as light sources, switches, and dispersion compensators—are added to relatively simple integrated products such as the AWG/VOA. Even higher power requirements will be needed. Other issues include relatively high losses at the point where the fiber joins the integrated optical circuit and when the circuit is bent to any degree. The former is a limitation of silica on silicon for obvious reasons; the latter is a limitation because it restricts the number of circuits that can be built on a given wafer. This is an area where Kymata (recently acquired by Alcatel) claims to have made significant improvements using a process licensed from IBM.
THE POWER OF POLYMERS
Optical integration is at a very early stage of development and it is possible that some clever circuit designer will find ways around many of the limitations of silica on silicon. An interesting variation is the sol-gel process, which uses a deposited colloidal silica and results in a faster process and higher yields than either chemical-vapor or flame-hydrolysis deposition. In North America, this process is most closely identified with Lumenon, but some European startups are also taking this approach.
One potential solution to the power problems associated with silica on silicon is to use certain polymers to selectively add a cladding level to a silica-on-silicon waveguide. This technique has the advantage that the polymers used are more thermo-optically sensitive, so less power is required to control them. However, the stability of the polymer upon exposure to humidity and temperature is questionable and polymer cladding for silica-on-silicon lightwave circuits has yet to be commercialized.
Similar problems have plagued integrated optical circuits built completely of polymer. Several attempts at commercialization of such products have taken place, although many of the companies that have tried this road to optical integration have found it wanting and have moved on to other processes—
Zenastra and Lightwave Microsystems are cases in point. The truth is that polymers are about the most frustrating of the materials being explored for optical integration.
Manufacturers continue to work in this area and some are beginning to claim that problems associated with polymer are solved. Corning, Gemfire, Optical Crosslinks, and Telephotonics, and even Lightwave Microsystems and Zenastra have not abandoned this material entirely. In our view the best hope for polymer-based integrated optical circuitry will be for products that are not mission critical, where cost is critical, or where replacements can be easily made. Applications such as fiber to the curb or certain equipment backplane applications spring to mind here. We think polymers will have to prove themselves in certain niches before they will be widely accepted.
I began by mentioning that there is, as yet, nothing quite like an optical silicon—that is, a material from which optical integrated circuitry can be built, which doesn't cost much, and in which many different kinds of functionalities can be embodied. Bookham Technology, it should be noted, is building optical circuitry using silicon and we believe that silicon may have been too readily dismissed by many manufacturers. The performance of the current generation of silicon optical circuitry is only slightly worse that the silica-on-silicon mainstream and it offers a variety of advantages including relatively fast response times and low manufacturing costs. But no one is really claiming that silicon can do for "microphotonics" what it did for microelectronics.
Another possible contender is gallium arsenide (GaAs), which, like silicon, is well-understood. It is already being used by Marconi and Nortel Networks in a limited way for modulators, but we have not seen any real signs of it emerging as a major material for optical integration. Lithium niobate (LiNO3) has similar properties to GaAs and is also used to build modulators. Lithium niobate was once touted as the wonder material for optical integration, but its star has fallen because available LiNO3 wafers are limited to 4 in. and because manufacturing circuits from LiNO3 is relatively complex.
Lithium niobate is also losing ground in its one successful niche—modulators—to indium phosphide (InP), which we believe, for the time being, provides the nearest thing to an optical silicon. It is certainly the most versatile of any of the platforms and is the only platform that promises anything like large-scale integration of active components (lasers and detectors), passive components, electronics, and even microelectromechanical systems (MEMS). In our survey of the industry we were surprised at how high the expectations were for InP.
Indium phosphide is already fairly widely used for integrated active components. However, manufacturing techniques for making integrated circuits out of this material are still quite primitive, only relatively small circuits can be made in commercial quantities, and InP is still very expensive. In addition, available InP wafers are 3 in.—silicon is moving to 12 in.—and yields from InP are still quite low. Indium phosphide can also be quite lossy, can exhibit optical nonlinearities, and can be quite temperature sensitive. Nevertheless, companies such as Agere Systems, Alcatel, Corning, Genoa, Marconi, Nanovation (currently in Chapter 11 bankruptcy), and ThreeFive Photonics are already seeking to tap into its potential.
Lawrence Gasman is president of Communications Industry Researchers, PO Box 5387, Charlottesville 22905. He can be reached at email@example.com.