Optical technology advancements continue to drive incredible increases in bandwidth transport capabilities that have, in turn, driven new system requirements. For example, advances in wavelength-division multiplexing are pushing core transport capacity forward, eliminating trunk bandwidth bottlenecks. In short-reach applications, 10-Gbit/second interfaces will soon become commonplace.

These ultrahigh-bandwidth links increase capacity requirements for core switches, routers and multiplexers. That is particularly true in the optical-switch marketplace, where density, bandwidth and power consumption are all key issues. These systems, which are typically implemented across several racks, need even faster backplanes and intraswitch links than the inter-switch-or transport links-they manage.

Optical switches steer data from one optical port or one individual lambda flow to the next. All-optical approaches are available, but electronic cross-connects remain prevalent. If the equipment offers data service, then electronic-based packet processing, switching and monitoring is required. In the terabit-class switch or router, line cards perform the optical-electrical conversion and the packet-processing functions and forward data to a switching subsystem.

Hard-pressed I/O

The switch, in turn, moves the data to a separate line card for eventual egress from the system. Today's electrical I/O is hard pressed to support the bandwidth needed using widely available board materials. Now there is an answer: Parallel optics, based on vertical-cavity surface-emitting laser (VCSEL) arrays, provides a way through the bandwidth/ reach limitations of legacy copper backplanes when used "inside the skin" of these switches.

Point-to-point optical links have been used inside systems for specialty high-speed interconnects. Board space and connector density limit the deployable link bandwidth. In contrast, parallel optics offer up to a twelvefold increase in density over serial approaches. These modules use VCSEL arrays and PIN diodes coupled to fiber ribbons to provide ultrahigh-speed, very short-reach links that are required in backplanes for such applications as rack-to-rack, switch-to-switch and switch-to-aggregation device interconnects. However, early implementations had difficulty ramping to the needed volume. Difficulties included reliable supply of laser arrays, multi-element optical alignment, perfection of high-speed electronics and module packaging. Those difficulties are being overcome through innovative solutions along a variety of fronts. Parallel VCSEL technology has matured to the point that it offers a credible alternative for system designers facing the challenge of implementing intrasystem or intrachassis links from tens of gigabits to terabits.

Parallel optics must meet demanding requirements to become the first choice of system designers implementing high-capacity links. Performance both in terms of throughput and link quality is key. The components must use a minimum of the scarce available system resources. The technology needs to fit into the system in terms of electrical interfaces and its mechanical interaction with panels and connectors. The individual devices must be based on a highly robust and manufacturable design to support production ramps. Lastly, and most importantly, the assembly techniques required to incorporate the optics into the particular system must have high yield, since these types of cards are usually very expensive, even before the optics are attached.

Optimum parallel-optical performance requirements include more than just individual channel bit rate. Applications range from 1.25 Gbits/s to over 2.5 Gbits/s. In addition, link/jitter and power budgets can be demanding. This leaves no margin for sloppy subcomponent selection. The lasers, the optical components and the driver/receiver circuits-and even the circuit cards-must all meet exacting standards. Parallel-optical suppliers should be able to demonstrate performance in every dimension and have deep technical expertise at every subcomponent layer.

Second-generation parallel-optical solutions significantly reduce system requirements as compared with the previous approaches. Here, too, significant differences exist. In general, newer modules consume half the real estate of the first generation, although some are even smaller. Reductions in power consumption and heat dissipation are equally, if not more, important. Again, second-generation solutions are about twice as efficient. Even among recent solutions, there are wide disparities in this area. Modules are available that help solve the heat and power problems.

Lastly, and most importantly, parallel optics suppliers must themselves use-and must enable the use of-good manufacturing approaches. The module design must be such that it can be built in high volume. Electrical bending of the signal from the card plane to the optical plane is one technique that supports high-volume optics.

See related chart