Light-Based AI Chip Breakthrough Promises Faster Training

TheWkly Analysis

Engineers at the University of Pennsylvania unveiled a photonic chip that can train neural networks using light, potentially at “lightspeed” compared to standard electronic chips. The breakthrough harnesses photons to perform core AI computations (matrix multiplications), drastically cutting power consumption and latency. While small-scale now, it exemplifies a shift toward optical computing to handle soaring AI demands. Neural networks require massive computational horsepower, often burning significant energy. By processing with light, the chip avoids resistive heating and can parallelize tasks. Researchers tested a modest AI model on this photonic chip, finding it both quick and efficient. Although scaling and cost remain challenges, experts see it as a big leap toward more sustainable, high-speed AI.