Photonic Chips Enable Optical Neural Learning
Researchers led by Shuiying Xiang at Xidian University built photonic computing chips that let neural networks learn using light, reported in the journal Optica. The programmable system pairs a 16-channel photonic processor (272 trainable parameters) with a laser array enabling nonlinear optical spiking, delivering 1.39 TOPS/W linear, 988 GOPS/W nonlinear, and 320 ps latency while solving CartPole and inverted-pendulum tasks. The prototype may enable low-latency, energy-efficient learning for autonomous vehicles and robots.
Key Points
- 1Demonstrate photonic learning hardware with a 16-channel processor and 272 trainable parameters
- 2Eliminate electronics by performing linear and nonlinear computations entirely in the optical domain
- 3Enable low-latency, energy-efficient reinforcement learning suitable for autonomous vehicles and robots
Scoring Rationale
Strong experimental demonstration with high efficiency metrics and clear training results; prototype scale and practical deployment remain limited in scope.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
