Optical Interconnects

Some years ago, Gene Amdahl was mounting a second startup. His first, Amdahl Computer was reasonably successful and subsequently sold to Fujitsu. The second startup was called Trilogy, it was an effort to build macro-scale processors that occupied an entire silicon wafer rather than just a chip. Though Trilogy came and went, it was a big deal at the time. The reason why the company is now just a historical footnote is that they ran into an insurmountable technology issue. They had no way to package the wafer. Specifically there was no way to get signal from one side of the wafer to the other. At the time, optical communications technology was still in its infancy. Trilogy approached Corning about developing an optical package but, of course, it was well beyond the capability of the optical technology as well.

Over a decade later, the issue came up again. As processor chips became denser, not only clock skew but interconnect densities started to become a problem, there was just not enough linear space at the edge to keep up with larger and more complex chips. Again, the obvious solution was to move to optical I/O. After another decade plus period of development, IBM has announced that the technology is ready to go. What this means is a few things. First, Moore’s Law gets another one of those needed breakthrough’s to keep it on-track. Second, with chips communicating optically, it is likely that you will now star seeing optical cabling inside computing devices rather than just a T3 fiber line leading to the building where they are housed. Third, rather than fiber reaching into the home (FTTH) or premises, you might see optical pathways directly from the device reaching out to the network.

In the optical networking world, there is a concept called “transparency” keeping the signal optical for a long as possible. Before the development of optical amplifiers, optical signals had to be periodically converted to electronics and regenerated as an optical signal. Given the processing and rise time of the electronics, it was much like flying a plane across the country but have to land every few miles and change planes. Transparency meant that the plane no longer had to land until it was near its destination. Optical I/O, optical output from computing devices may mean that the optical signal will eventually no longer have to drive to the local airport to catch its plane but can take off from right inside your PC. The results will be data rates that make today’s connections look like dial up. This scenario is probably another decade or so away, but an important threshold has been crossed.

This may also have some tendency to roll back the current "mobility boom" in favor of wired connections. Although wired connections tend to have inherently more bandwith than wireless, the wireless industry has been quite clever regarding signal processing and compressing more and more data into the spectrum available. Transparency right up to the processor may be more than can be matched by increased data compression presuming there is some application that could make use of it.