Photonic spiking AI boosts intelligent routing

By AI, Created 12:20 PM UTC, May 22, 2026, /AGP/ – A new Opto-Electronic Science paper says a photonic spiking reinforcement learning system can improve routing for data centers and 6G networks while cutting latency and power demands. The researchers tested the approach on a fat-tree SDN setup and say it beat Dijkstra on throughput, delay, packet loss and load balance.

Why it matters: - Data centers, satellite networks and 6G systems need routing that reacts quickly without burning extra power. - Conventional electronic neural-network hardware can add latency and energy cost that do not fit large-scale, high-speed networks. - Photonic computing and spiking neural networks promise lower latency and higher efficiency for real-time routing decisions.

What happened: - Opto-Electronic Science published a paper titled “Photonic spiking reinforcement learning for intelligent routing.” - The paper proposes a software-defined networking routing architecture that combines spiking neural networks, proximal policy optimization, and photonic computing chips. - The team evaluated the system on a fat-tree data center topology. - The paper lists DOI 10.29026/oes.2026.260005.

The details: - The architecture uses a three-layer design with a data plane, control plane and intelligent decision-making plane. - A photonic hardware-software framework uses an MZI-based photonic synapse chip and a DFB-SA-based photonic spiking neuron chip. - The spiking Actor network runs on photonic hardware to improve speed and energy efficiency while keeping training stable. - Tests on 640 state-action pairs found the hardware-software framework matched the pure software algorithm in inference accuracy. - Under varying traffic loads, the photonic spiking PPO routing beat Dijkstra on throughput, packet loss rate, average latency and load balance. - The system also showed strong robustness and fast re-convergence after topology changes such as link failures.

Between the lines: - The result is a sign that optical hardware may move from lab demos toward control tasks inside communications networks. - The claim of first full integration suggests the work is positioning itself as an early reference point, not just an incremental routing tweak. - The biggest practical appeal is not raw novelty alone. It is the combination of real-time adaptation, lower latency and better energy efficiency in one routing stack.

What’s next: - The authors say the approach has potential in large-scale data centers, computing power networks, satellite Internet and future 6G networks. - The paper frames the work as a foundation for space-air-ground integrated network optimization. - Wider deployment will depend on whether the photonic hardware can scale beyond the tested setup and maintain performance in operational networks.

The bottom line: - The study argues that photonic spiking reinforcement learning could become a new routing model for networks that need both speed and efficiency.