Peer-reviewed publications or official reports from MIT

MIT Develops Photonic Processor for AI with 92% Accuracy in Under 0.5 Nanoseconds, Achieving 96% Training Accuracy and 120 GOPS Speed

Dec 2, 2024, 05:53 PM

Researchers at MIT have unveiled a groundbreaking photonic processor that could revolutionize artificial intelligence computations. This processor utilizes light to perform all neural network computations, achieving over 92% accuracy in under 0.5 nanoseconds while consuming ultra-low energy. The device demonstrates a 96% accuracy in training and 92% during inference, comparable to conventional hardware. The innovation is expected to significantly enhance AI training speeds and efficiency, potentially impacting various sectors including telecommunications and lidar technology. Additionally, a team from Sun Yat-sen University has proposed a new optical neural network chip framework that reaches a computation speed of 120 GOPS, further advancing the field of quantum computing and AI.