Kotura Inc announced today it has been awarded a $14m contract by Sun Microsystems, to assist with photonic linking of processor cores in future supercomputers and power-limited multicore systems. Sun is carrying out the optical core-hookup work for the US military.
The idea of linking processors optically rather than electronically is to improve speed, bandwidth and power consumption in future systems, whose many powerful cores are expected to severely strain the capacity of normal electronic linkages. The funding comes, not from any normal US military lab, but from DARPA - the Pentagon brainbox apparat which likes to develop answers first and ask what the question was later.
DARPA refers to the effort as Ultraperformance Nanophotonic Intrachip Communications (UNIC), and believes it will help with "relevant computing tasks such as power-starved embedded applications, and supercomputing".
The military notioneers gave Sun $44m to develop UNIC in March. Sun have now shared the cake with Kotura, who specialise in using CMOS technology for chip-to-chip links.
“Optical interconnects increase the bandwidth between chips while reducing power consumption and chip-to-chip latency," said Sun veep Jim Mitchell.
“Kotura is well positioned to deliver silicon photonic elements that will contribute directly to Sun's research efforts on this project... we are on the verge of radically improving the speed of communications among the chips in powerful computing systems, thereby increasing their performance dramatically."
“We have a great working relationship with Sun and DARPA and are now expanding upon that,” added Kotura CTO Mehdi Asghari. He said that Kotura would be targeting "extremely small footprints and ultra low power consumption" suggesting that DARPA are interested in the "power starved embedded applications" end of UNIC here, as much as conventional supercomputing.
There are various military needs for high-performance computing in small systems without a lot of electricity to hand. In particular, synthetic-aperture radar images need a lot of processing, and it saves a lot of satellite bandwidth if this can be done onboard power- and weight-limited platforms, such as small spy satellites and reconnaissance drones. ®