Optical phased array LiDAR chips towards polarization-resolved 4D sensing

Abstract

Optical phased array (OPA)-based LiDAR technology provides high-speed, solid-state sensing for autonomous vehicles, robotics, and aerial systems. However, current OPA-based LiDAR platforms lack the ability to accurately recognize and classify materials, limiting their effectiveness for advanced scene understanding in robotics and machine perception. Here, we propose a method for OPA-based LiDAR that integrates a silicon photonic OPA chip with a polarisation-resolved optical neural network (PolarONN) chip to directly analyse the polarisation of backscattered signals. This integrated architecture enables a polarization-resolved point-cloud output by jointly providing 3D geometry, return intensity, and a polarization-derived attribute, and it attaches a point-wise material label inferred by the on-chip PolarONN. Compared to representative electronic classification back-ends used for polarization-based material identification, our on-chip photonic inference core provides ns-scale response with a low electrical power budget for the inference module. We experimentally demonstrate robust, real-time classification of three representative materials (wood, stone, and steel) with classification accuracies exceeding 85%, at distances up to 10 meters. Our study establishes a practical route toward real-time polarisation-resolved LiDAR perception, significantly enhancing the capabilities of next-generation autonomous sensing technologies. This study combines an optical phasedarray LiDAR with an integrated photonic neural network to map three-dimensional scenes and classify materials using polarization-resolved optical echoes.