9 Apr 2025

Photonic processor runs ResNet, BERT and an Atari reinforcement learning agent

AI and autonomous agentsPhotonics and optical computing
AI illustration · Illustrative scene, not a photograph of the specific device or experiment.

What was demonstrated

Lightmatter researchers described a photonic AI processor that executed full models, including ResNet, BERT and the DeepMind Atari deep reinforcement learning algorithm, and reported near-electronic precision for many of these workloads.

Why it matters

Earlier photonic computing demonstrations were limited to simplified benchmark tasks. Running standard models at useful precision is a step towards photonic accelerators competing with electronic ones.

Caveats

  • The work is from a company developing the product; independent benchmarking is not yet available.
  • The abstract claims near-electronic precision for many, not all, workloads.
  • Energy efficiency and throughput relative to current GPUs at system level need independent verification.

Technology profiles

Sources

  1. peer reviewed paper
    Universal photonic artificial intelligence acceleration
    Sufi R. Ahmed, Reza Baghdadi, Mikhail Bernadskiy et al.. Nature, 9 Apr 2025. DOI: 10.1038/s41586-025-08854-x.

Last reviewed 20 Oct 2026.