On-chip photonic neural network for multi-wavelength time-dependent signal processing

Neuromorphic photonics can greatly benefit processing of optical signals transmitted through telecom links or produced by photonic sensors, by exploiting key advantages w.r.t. electronics such as highly parallel and energy efficient linear operations. We present a compact integrated photonic neural network, composed of 96 neurons (silicon microring resonators) within a 0.15 mm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> footprint, for high-throughput optical processing concurrently in time, space and frequency domains. We experimentally test it on benchmark image classification tasks, namely MNIST and Fashion-MNIST..