Optical interconnects and optical memory are both relatively immature technologies, having reached the marketplace with initial products just within the past decade. But both are rapidly emerging, with many envisioned products in sight. As discussed above, some of these are strong candidates for realizing a PetaFLOPS computing capability. The three tables that follow reflect the working group's projections about the extent the technologies will continue to emerge over the next 20 years. These projections provide a basis for hardware decisions in planning the development of PetaFLOPS systems.
Tables 5.4 and 5.5 address optical interconnects, dividing them into guided interconnects and free-space interconnects. This division is necessary since the application of these two types of optical interconnects in PetaFLOPS computers is quite different; while the guided interconnects will function mostly as point-to-point interconnects, the free-space interconnects will function as broadcast and random interconnects. The first line of each table provides baseline information by giving an example of where the technology is now. Table 5.6 addresses optics for memory. The largest impediment to the introduction of optical interconnects is cost. It is expected that optical interconnects will continue to ride the learning curve in terms of costs.
