Photonic Chips Boost Real-Time Learning in Neural Nets

New chips enable fast, energy-efficient, and compact computation for reinforcement learning tasks.

Published on Mar. 6, 2026

Got story updates? Submit your updates here. ›

Researchers have developed photonic computing chips that can perform both linear and non-linear computation in the optical domain, overcoming key limitations of previous photonic spiking neural systems. The two-chip system includes a 16-channel photonic neuromorphic chip with 272 trainable parameters, enabling it to process multiple streams of optical signals at once and adjust many connections through learning. The system was able to quickly learn through trial and error on reinforcement learning tasks, showing potential for applications like autonomous driving and embodied intelligence.

Why it matters

This breakthrough in photonic computing chips could lead to significant improvements in the speed, energy efficiency, and compactness of neural networks used for real-time learning and decision-making in autonomous systems. By performing both linear and nonlinear computations entirely in the optical domain, the new chips avoid the delays and inefficiencies of converting between optical and electronic signals.

The details

The researchers developed a two-chip system, including a 16 x 16 Mach-Zehnder interferometer mesh chip tailored for spiking neural networks and a chip containing a distributed feedback laser array with a saturable absorber optimized for low-threshold nonlinear spiking activation. They also created a hardware-software collaborative training and inference framework to account for chip-level variations. Testing on reinforcement learning tasks like balancing a pole on a cart and swinging a pendulum upright showed the system achieved near-perfect performance, with only a 1.5-2% drop in accuracy compared to a software-only approach. The chips demonstrated extremely fast (320 ps latency), energy-efficient (up to 987.65 GOPS/W), and compact (up to 533.33 GOPS/mm^2) computation.

  • The research was published in March 2026.

The players

Shuiying Xiang

The research team leader from Xidian University in China.

Optica Publishing Group

The publisher of the journal where the research was published.

Got photos? Submit your photos here. ›

What they’re saying

“Photonic spiking neural systems use brief optical pulses, or spikes, to emulate neural signaling, but they can typically only process the linear parts of computation using light. Previously, the nonlinear steps that make learning and decision making possible required the signal to be converted back into electronic signals. This adds delay and undercuts the speed and energy advantages of photonics.”

— Shuiying Xiang, Research team leader (Optica)

“We used this system to demonstrate reinforcement learning, supported by a hardware and software collaborative framework that trains and runs the neural network. The system was able to learn quickly through trial and error, showing potential as a fast, low-latency solution that could be used for applications such as autonomous driving and embodied intelligence.”

— Shuiying Xiang, Research team leader (Optica)

What’s next

The researchers would like to design and fabricate an even larger-scale (128-channel) fully functional photonic spiking neural network chip to solve more complex reinforcement learning tasks such as neuromorphic autonomous navigation.

The takeaway

This breakthrough in photonic computing chips represents a significant advance in the speed, energy efficiency, and compactness of neural networks for real-time learning and decision-making, with potential applications in autonomous systems and robotics.