Nokia Bell Labs

lab
photonic-computingneural-networkson-chip-trainingsilicon-photonicsresearch-lab

Nokia Bell Labs

Type: Industrial research lab Location: New Providence, NJ Lineage: Historic AT&T / Bell Laboratories; now part of Nokia

The Ashtiani group at Nokia Bell Labs demonstrated the first integrated photonic deep neural network with end-to-end on-chip gradient-descent backpropagation — all linear and nonlinear computations on a single silicon photonic chip. This solves the long-standing problem of on-chip activation-gradient computation in photonic neural networks.

Key Contributions

  • First end-to-end on-chip photonic backprop training — Nature 2026 (Vol 651, pp 927-932). (Ashtiani)
  • On-chip activation gradient scheme via opto-electronic cascaded IMs — solves previously unsolved problem. (Ashtiani)
  • Automatic device-variation compensation — training on the chip adapts to actual device behavior. (Ashtiani)
  • 92.5% 2D classification accuracy matching digital reference. (Ashtiani)

Research Team

  • Farshid Ashtiani (lead)
  • Mohamad Hossein Idjadi
  • Kwangwoong Kim

Mentioned In

Related Entities

  • Lightmatter — commercial complement (inference-focused)