Revolutionizing Edge Computing With Superconducting Deep Neural Networks
ECE Assistant Professors Marco Colangelo and Francesco Restuccia were awarded a two-year $500,000 DARPA grant for “Superconducting Deep Neural Networks at the Edge.” They aim to reduce end-to-end processing latency by at least three orders of magnitude compared to existing edge computing paradigms without impacting AI performance.
Edge computing through AI techniques is fundamental to achieve situational awareness in many real-life scenarios. Existing edge computing paradigms exhibit networking and AI processing latency that are orders of magnitude more than critical mobile applications can tolerate. The key unaddressed technical challenge is obtaining a drastic reduction in computing and networking latency without impacting the AI performance. In the proposed FLEX project, the PIs will research new Superconducting Deep Neural Networks (SuperDNNs), which are implemented at the waveform level by directly encoding DNN inputs on the waveform and by implementing a physical DNN that processes waveforms directly coming from the radio frequency frontend. Overall, the target is to reduce the end-to-end processing latency by at least three orders of magnitude compared to existing edge computing paradigms.