The work is supported by $1m (£626,000) in funding as phase one of a $4m co-operative agreement contract with the US Defense Advanced Research Projects Agency (DARPA).
‘Computers are becoming faster and able to tackle more complex challenges, but they are also consuming larger amounts of electrical power,’ said Dr Paul Franzon, a professor of electrical and computer engineering at NC State and lead researcher under the DARPA co-operative agreement. ‘For example, the amount of computing power in unmanned vehicles is limited by the power supply they can carry.’
To address this challenge, the NC State research team will design a multiprocessor computer system that has a computing-to-power ratio of 75 gigaflops per Watt.
At this ratio, the system must be able to process 75 billion floating-point operations per second for every Watt of power used. By contrast, the most efficient processors currently on the market have a computing-to-power ratio of 16 gigaflops per Watt.
‘We plan to use 3DIC technology and heterogeneous computing techniques to develop what will essentially be a highly efficient multiprocessor that would fit in a matchbox,’ Franzon said in a statement.
Franzon is the primary investigator under the co-operative agreement, which has negotiated options through 2017.
Comment: Engineers must adapt to AI or fall behind
A fascinating piece and nice to see a broad discussion beyond GenAI and the hype bandwagon. AI (all flavours) like many things invented or used by...