Optical Photonic Processors Compute Deep Neural Networks at the Speed of Light in Pilot Trials
Replacing copper wires with silicon waveguide lasers allows matrix multiplications to occur with near-zero heat generation and extreme bandwidth.
Ask AI About This Story
Real-time contextual insights, local impact & analysis
In what engineers describe as the post-electronic computing paradigm, photonic integrated circuits (PICs) utilize beams of laser light to perform matrix tensor math for large language models. Operating at lightspeed with negligible thermal dissipation, optical compute promises to reduce data center power demands dramatically while scaling model inference bandwidth.