Photonic computing (or optical computing) uses photons (light particles) produced by lasers or diodes for computation, rather than the electrons used in traditional electronic computers. Because light travels faster than electricity in wires and generates less heat, photonic processors promise orders-of-magnitude improvements in speed (latency) and energy efficiency (throughput per watt). It is particularly promising for linear algebra operations central to AI and deep learning.
Research dates back to the 1960s and 70s with the invention of the laser. However, practical, scalable integrated photonics for computing have only recently become feasible due to advances in silicon photonics manufacturing.
Startups like Lightmatter and Luminous Computing are developing photonic chips specifically for AI acceleration. It is seen as a potential successor or accelerator for electronic chips as Moore's Law slows down, aiming to solve the bandwidth and energy bottlenecks in data centers.