Researchers integrated a silicon photonic optical phased array chip with an optical neural network chip to classify materials in real time using LiDAR, according to a study published in Nature Communications. The system links an optical phased array with a polarization-resolved optical neural network chip, known as PolarONN, to analyze the polarization states of backscattered light signals directly on the hardware.
Optical phased array platforms provide solid-state, high-speed sensing for autonomous vehicles, aerial systems, and robotics, yet existing hardware cannot identify surface materials. The new integrated architecture addresses this limit by outputting a polarization-resolved point cloud. The chip produces three-dimensional geometry, return intensity, and polarization attributes while appending a point-wise material label inferred by the PolarONN core.
The on-chip photonic inference module delivers nanosecond-scale response times while maintaining a low electrical power budget, contrasting with conventional electronic classification back-ends used for polarization analysis. In bench experiments, the team demonstrated real-time classification across three representative materials—wood, stone, and steel. The setup maintained classification accuracy above 85 percent at operational distances reaching up to 10 meters.
Scientists at Xidian University led the study alongside contributors from the Xi'an Microelectronic Technology Institute, Jilin University, the Xi'an Institute of Electromechanical Information Technology, and the Chinese Academy of Sciences. Technical teams at the Chongqing United Microelectronics Centre provided architecture guidance. The work received backing through grant 62405229 from the National Natural Science Foundation of China, grant JYB2025XDXM105 from the Ministry of Education of China, and grant XJSJ25007 from the Central Universities fund. Nature Communications received the paper on September 4, 2025, and published it on September 12, 2026.
