DARPA funds Sun's optical chip chatter magic
Well, well, well, Sun Microsystems worked DARPA over for some SPARC funding after all. The server maker today celebrated a $44m contract to produce optical chip-to-chip connections technology for the Defense Department's R&D unit.
The official line on this arrangement has Sun taking $44.29m over five and a half years to produce "microchip interconnectivity via on-chip optical networks enabled by Silicon photonics and proximity communication." This research fits into DARPA's (Defense Advanced Research Project Agency) Ultraperformance Nanophotonic Intrachip Communication program. Because at DARPA "performance" simply isn't good enough.
We're told that "The project presents a unique opportunity to develop supercomputers through interconnecting an array of low-cost chips, with the potential to overcome the fundamental cost and performance limits of scaling up today's large computer systems. By providing unprecedented high bandwidth, low latency, and low power interconnections between the parallel computing chips in such an array, this research project will help enable a broad class of companies and organizations to utilize applications with high compute and communication requirements, such as energy exploration, biotechnology and weather modeling."
Away from this statement, there's quite an interesting back story.
Sun, you might remember, fought with IBM and Cray for a piece of a massive DARPA contract around future supercomputing systems. Ultimately, Sun lost out to IBM and Cray, but it seems that DARPA still wants to fund some of the key bits of Sun's proposal.
While the exact nature of Sun's old proposal remains secret, we've learned that it involved the second generation of multi-core Rock chips - a flavor of the UltraSPARC processor family. Sun planned to link these chips via the proximity communications technology mentioned earlier.
Ivan Sutherland - the father of computer graphics - developed the proximity communications idea in his work at Sun Labs. The basic concept behind the technology involves marrying two processors together via a direct connection to speed the flow of information between the processors. We've discussed the technology at length here.
Sun's DARPA proposal also involved some optical networking components where it would replace on-board wires with optical interconnects - again to speed the flow of data across a large system.
Of late, Intel and IBM have talked more than Sun about their optical work with both companies producing optical switches and other manners of silicon photonics devices.
When Sun lost the DARPA bid and then subsequently delayed the release of its first Rock chip to 2009, we wondered about its long-term commitment to these more advanced projects. Sun could well have decided that the UltraSPARC work, as many have argued, was too costly for a company of its size.
DARPA's funding of at least the communications pieces of the future Sun systems seems to confirm that Sun has no plans of backing down from future Rock system designs. The end result should be modest-sized servers that boast what we would consider today as supercomputer class machines. ®