Integrated circuits have revolutionized electronics and have begun to make good on their promise to do the same for photonics, but challenges remain. Since the beginning of the most recent boom in the optical fiber industry, component vendors have introduced many technologies, among them planar optical integration, in which optical circuits are constructed out of waveguides fabricated on silicon wafers. The manufacturing process for these optical circuits uses the tools and techniques developed by the semiconductor industry, which was itself transformed 30 years ago by the integrated circuit chip. However, while planar optical integration has made substantial progress, both in technology and in market acceptance, the wavelength of a telecommunications photon is 10,000 times bigger than an electron, providing a fundamental limit to the amount of integration that can be achieved. That said, the state of the art is not at all close to these fundamental limits, and a lot of headroom is left to exploit the technology. Key measures of its developmental status are performance and level of integration. Equipment vendors demand components that push the boundaries of achievable performance. There is no place for subspecification parts in current or next-generation systems. But planar lightwave circuits must also deliver lower cost and size through increased integration. The most commercially advanced optical integration technology — silica-on-silicon — has demonstrated that it can provide both performance and the benefits of integration. For example, 40-channel variable optical attenuator multiplexers replace 40 discrete attenuators, 39 filters and their associated piece parts. Another example is arrayed waveguide gratings, which perform multiplexing and demultiplexing functions in parallel on one chip. Over the last two years, they have gained a significant percentage of the total multiplexer/demultiplexer market because a single chip in a single package can perform either function for 40 channels in parallel. Also, the production costs for high-channel-count devices are much lower than for competing technologies. To achieve performance that is equivalent to a 40-channel arrayed waveguide grating, thin-film filters and fiber Bragg gratings would require 39 designs, production runs and packaging operations. Meanwhile, 80-channel arrayed waveguide gratings have been introduced to the market, and 256-channel devices have been demonstrated in the laboratory. Initially, arrayed waveguide gratings did not provide the same performance as the other technologies, and most believed that discrete parts had an intrinsic performance advantage over integrated devices. However, intense development to improve their performance has made them equivalent to competitive technologies in all major parameters. Because arrayed waveguide gratings process all channels in parallel, they have had a significant advantage in insertion loss for high-channel-count applications. Essentially, insertion loss is the same for 4-, 40- or 80-channel devices; the discrete variety, however, is used in series, and the losses accumulate. Figure 1. Already low, the insertion losses for arrayed waveguide gratings continue to drop. In these plots of the transmission spectrum of an ultralow-loss 40-channel, 100-GHz-spaced Gaussian arrayed waveguide grating, the insertion loss for the center channels at the ITU grid point is around 2 dB, and all 40 channels have insertion losses lower than 3 dB. Performance in isolation Another key performance parameter is isolation, which is a measure of how much light from an unwanted channel leaks into the passband of an active channel. For discrete filters, such as thin-film filters and fiber Bragg gratings, isolation is determined by nearest-neighbor crosstalk — the amount of light that leaks into a channel from its immediate neighbors. Integrated devices, including arrayed waveguide gratings, offer an extra level of isolation. Instead of dropping away as it does in a thin-film filter, the transmission curve of an arrayed waveguide grating levels off in the noise floor because these devices process channels in parallel, with all channels in one chip. Light scattered or lost because of imperfections in one channel might scatter into any other. Eighteen months ago, a noise floor around 35 dB was considered good, and adjacent channel isolation for both arrayed waveguide gratings and thin-film filters was around 25 dB. However, totaling up the contributed noise floor from the gratings’ other 37 channels dropped its cumulative isolation to only 19 dB. By applying strict process control and focused improvement programs, top vendors of the devices now achieve adjacent-channel isolations of 30 dB and noise floors lower than 50 dB. Figure 2. Ultralow-loss wide-band, or “flat top,” arrayed waveguide gratings have demonstrated insertion losses between 4 and 5 dB. Similar progress has been made in reducing polarization-dependent loss. Typical loss for a top-tier arrayed waveguide grating is now 0.1 to 0.2 dB measured across the passband. Spectral plots show low noise floor (Figures 1 and 2). When all channels are added together, the noise floor is not significant, and the total isolation of an arrayed waveguide grating is dominated by adjacent-channel isolation. These devices now exhibit total cumulative isolation as good as that from the best thin-film filters and fiber Bragg gratings. Arrayed waveguide gratings also carry significant advantages over discrete devices in regard to chromatic dispersion. Because they do not have multiple reflections and all of the wavelengths have virtually the same optical path, chromatic dispersion is very low (Figure 3). In counterpoint, multiple reflections from thin-film filters and fiber Bragg gratings create different path lengths for different wavelengths, resulting in significant pulse spreading. A typical thin-film filter can have ±30 to 40 ps/nm across the passband and up to 60 ps/nm of change over the passband. Figure 3. Chromatic dispersion, measured across the passband of an arrayed waveguide grating, is very low and very flat. The black curve represents the passband shape, and the red curve represents chromatic dispersion. Planar integration’s potential extends beyond multiplexing/demultiplexing applications. Next-generation networks require active components that enable systems with automatic and dynamic adaptability. Emerging dynamic planar light chip products use the thermo-optic effect to alter the index of the silica, delivering dynamic control. Companies such as NEL in Saddle Brook, N.J., and Lightwave Microsystems are shipping 40-channel variable optical attenuator multiplexers based on this technology (Figure 4). Figure 4. Each of the 40 output channels of a variable optical attenuator multiplexer can be set to any arbitrary attenuation. The dynamic range of the device is 10 dB. The minimum insertion loss at 0-dB attenuation is approximately 6 dB. These devices not only lower production costs, but also reduce 79 separate packages to one. Also, use of discrete attenuators requires that each be spliced or connected and that the fibers be properly routed to assemble the final subsystem. Although integration can streamline this configuration, a 40-channel variable optical attenuator multiplexer is considerably more complex than a simple passive optical component (Figure 5). It also must match the performance of discrete implementations. Figure 5. In a variable optical attenuator multiplexer, passive functions performed by the arrayed waveguide multiplexing filter are combined with active functions performed by the variable optical attenuators. Besides combining these functions, the module integrates electronic controls to adjust the attenuators, stabilize the waveguide temperature and communicate with the system. A challenge for designers of planar lightwave chips is polarization-dependent loss. Such loss is very low for discrete mechanical variable optical attenuators because the light remains inside the fiber. The engineering of planar devices, however, must keep polarization-dependent loss low, and this has been successful (Figure 6). Figure 6. The overall performance of variable optical attenuator multiplexers is such that major equipment companies are designing them into systems. This plot of polarization-dependent loss vs. insertion loss shows that all 40 channels of an array of attenuators have losses below 0.4 dB, out to the specification limit of 10-dB attenuation. Thermo-optic elements can perform other active functions, such as that of the dynamic gain-flattening filters, which are entering the commercial market. NEL offers an 8 x 8 thermo-optic cross connect and arrays of eight 2 x 2 thermo-optic switches. Several companies are developing integrated versions of reconfigurable optical add/drop multiplexers based on planar lightwave circuits, thermo-optic switches and arrayed waveguide gratings. Things to come At their present stage, planar integrated optics diverge significantly from the analogy of semiconductor integrated circuits. In simplified form, electronic chips obtain their power from the integration of millions of transistor gates that basically perform the same function in the same material. Optical networks require many different functions, all performed by different optimized materials. Therefore, planar integration of optics will more closely resemble microwave hybrids, where many smaller-scale integrated components are assembled on a common platform. The silica-on-silicon planar lightwave circuit is a strong candidate for an optical integration platform. In parallel with the large-scale planar integration product development described above, a completely separate set of planar lightwave circuits has emerged, including integrated fiber optic transceivers pioneered by Lucent, Nortel, NEL and Bookham Technology. Sometimes called a silicon optical bench, a small planar lightwave chip offers a platform on which to mount laser and photodetector chips, and waveguides are used to interconnect these optical elements. In some cases, active chips are prepositioned, using pick-and-place tools, and techniques such as solder bump bonding provide the precise positioning between separate optical elements. Input and output fibers are positioned using V-grooves etched into the silicon substrate. Other optical elements, such as coarse wavelength division multiplexing filters, and electronic chips, such as laser drivers, also can be added. Currently, transceiver modules are relatively small, such as might fit into a dual in-line package, whereas arrayed waveguide grating/variable optical attenuator chips are several centimeters on a side. While the scale of these is different for these two applications, planar lightwave chips are made with the same tools and can be matched to one another. The markets are large for large- and small-scale planar lightwave chips, but the promise of optical integration will be fulfilled when these two technologies are unified in a single platform to provide both active and passive optical functions in proximity with electronic control. It will be neither monolithic nor hybrid, but a combination. Each optical function can be performed in an optimized material system, but will also be capable of integration with dissimilar technologies through the platform. We are just beginning to explore the potential of such integration, and the capability of the planar lightwave circuit industry will be built up over time. Meet the authors G. Ferris Lipscomb is vice president of marketing at Lightwave Microsystems in San Jose, Calif. Marc Stiller is director of product marketing.