All-integrated multidimensional optical sensing with a photonic neuromorphic processor.

Abstract

Multidimensional optical sensing is crucial in information technology and modern intelligent systems. Despite advancement in optical sensing, capturing multidimensional light field information remains challenging, typically implemented using cascaded single-dimensional sensors and discrete optoelectrical components for information decoupling. Here, we present an all-integrated multidimensional sensing chip incorporating a light field sensitizer and a photonic neural network processor. The inverse-designed sensitizer projects the multidimensional input into multiple channels; each dimension is then decoupled through the reconfigurable nonlinear neural network. We experimentally achieved 91% high accuracy for single-shot, concurrent sensing of intensity, polarization, and wavelength using a well-trained five-layer neuromorphic system. The fully on-chip system eliminates optical-electrical conversion and offline digital processing, enabling low-latency and high energy efficiency. Moreover, we achieved stabilization and recovery of high-speed signals at 100 gigabytes per second under randomly perturbed polarization and wavelengths. This work shows the potential for low-latency, energy-efficient optical sensing and complex information processing using neuromorphic integrated photonics.