All-photonic reservoir computing for dynamic image classification via spatial-to-temporal encoding

We present an all-photonic reservoir computing device that integrates the reservoir and readout layers on a silicon photonic chip, enabling end-to-end optical processing of time-series data with ultralow latency. The photonic reservoir computing achieves real-time high-speed image processing by incorporating a photonic spatial-to-temporal encoding technique, which converts visual information into a time-domain signal at a rate of 25 Gigasamples per second. We experimentally demonstrated that the proposed photonic approach can recognize microsecond-scale switching events that are challenging to capture using conventional electronic systems. This approach offers a scalable solution for photonic processors targeting time-critical tasks such as anomaly detection and real-time sensing applications.