Xidian University Ultrafast Photonics Team Publishes LiDAR Chip Achievement Integrating Polarization Sensing and On-Chip Optical Computing in Nature Communications

Recently, the ultrafast photonics team at the School of Optoelectronic Engineering of Xidian University has made important progress in the field of on-chip intelligent LiDAR. The team combined an optical phased array with a polarization-resolved optical neural network, enabling the LiDAR to identify materials using polarization information in the echo while acquiring the target's three-dimensional position, providing a new approach for enhancing the environmental perception capabilities of unmanned systems. The related results, titled "Optical phased array LiDAR chips towards polarization-resolved 4D sensing," were published in September 2026 in the international academic journal Nature Communications. Professor Xu Huailiang and Professor Sun Yanling of the School of Optoelectronic Engineering of Xidian University, Professor Zhang Jincheng of the School of Integrated Circuits, and Associate Researcher Fan Xiuhong of the Xi'an Microelectronics Technology Institute are co-corresponding authors of the paper, and Associate Professor Wang Zihao of the School of Optoelectronic Engineering is the sole first author. The School of Optoelectronic Engineering of Xidian University is the first affiliated institution of the paper.

Enhancing single-point information content: enabling LiDAR not only to measure position but also to identify materials

LiDAR is an important means for autonomous driving, intelligent robots, and unmanned systems to perceive their surrounding environment. By emitting laser light, receiving target echoes, measuring target distance, and reconstructing three-dimensional scenes, it provides support for navigation, obstacle avoidance, and target detection. Among these, optical phased arrays use optical modulation on a chip to achieve beam scanning without mechanical rotating components, providing an important pathway for the miniaturization and solid-state development of LiDAR.

However, how to further identify the material of a target after acquiring its position and contour remains a question worth exploring. Traditional optical phased array LiDAR mainly relies on distance and echo intensity to perceive targets; for different materials with similar appearances and close reflection intensities, this information is often insufficient to support effective discrimination. At the same time, limited by chip scale and scanning capability, the number of effective measurement points the system can acquire is limited, and improving target identification capability solely by increasing measurement points faces practical difficulties.

To this end, the team started by increasing the information content of a single measurement point and introduced polarization information into optical phased array LiDAR. When laser light interacts with a target, the polarization state of the echo is affected by the material and surface properties, thereby carrying identification information beyond distance and intensity. Extracting this information helps distinguish different material targets with similar appearances and close intensities, enabling each measurement point to reflect the material properties of the target while providing spatial position information, offering a new pathway for multi-dimensional perception under limited measurement point conditions.

System principle of the polarization-resolved optical neural network working in coordination with optical phased array LiDAR

Introducing optical computing at the receiving end: completing polarization information processing on a chip

Polarization information of light can enhance target feature recognition capability, but how to process this information quickly and efficiently is a key issue that must be addressed for its integration into chip-level LiDAR. As the number of LiDAR scanning points and information dimensions increases, echo analysis places higher demands on data transmission, computing speed, and system power consumption.

The research team integrated a polarization-resolved optical neural network chip (PolarONN) in coordination with a silicon optical phased array chip, constructing a chip-level photonic system that combines scanning ranging and material property discrimination capabilities. The transmitting end controls the scanning direction via the optical phased array, and the receiving end uses the PolarONN chip to process polarization-related echoes and map the differences therein to point-by-point material categories.

This architecture combines polarization sensing with an optical neural network, directly using polarization information in the echo for optical computing. By adjusting the amplitude and phase of optical signals, the chip uses coherent superposition of light to perform computations related to material discrimination. These computations are completed before photoelectric detection, enabling the sensing front end to undertake part of the material discrimination task and potentially reducing the data processing burden on the back end. The polarization processing and inference module in the study has the advantages of fast response and low power consumption, providing technical support for point-by-point attribute recognition under high-speed scanning conditions.

PolarONN chip structure, optical phased array chip, and experimental system

From three-dimensional contours to material properties: completing experimental validation at multiple distances and in multiple scenarios

The research team selected three representative materials—wood, stone, and steel—and conducted experiments at multiple distances, under identical geometric conditions, and with letter-shaped targets to validate the system's ranging and material identification capabilities. The experiments showed that even when the geometric features of targets are similar, polarization echoes can still provide effective information for material discrimination. The team also used "X, D, U" letter targets made of steel, wood, and stone, respectively, to demonstrate point cloud results combining spatial structure, echo intensity, and material labels. While reconstructing the three-dimensional morphology of the targets, the system assigns corresponding material categories to the measurement points, enabling the position and attributes of the targets to be presented in the same point cloud.

The "four-dimensional sensing" proposed in this study is embodied as adding point-by-point material attributes inferred from polarization information on the basis of three-dimensional spatial coordinates. As a result, point clouds can be expanded from descriptions of target appearance to expressions of target attributes, providing richer information for scene understanding.

Material identification results at multiple distances, and three-dimensional reconstruction, material labels, and echo intensity distribution of the "XDU" target

Toward intelligent unmanned systems: exploring a new path for integrating perception and computing

For robots and unmanned systems, identifying the position of obstacles is only part of environmental perception; judging the material and surface properties of targets can likewise provide a basis for subsequent decision-making. For example, in industrial scenarios, material properties facilitate target sorting and inspection; in unmanned platform operations, richer target information is expected to assist environmental understanding and interaction judgment. This study expands the connotation of a single measurement point through polarization information and uses on-chip optical computing to complete attribute discrimination, providing a new path for enhancing scene understanding capability under limited spatial sampling conditions, and is expected to serve applications such as intelligent robots, environmental perception for unmanned platforms, and industrial target detection.

From optical phased array scanning and polarization echo processing to point cloud output with material labels, the ultrafast photonics team connects integrated photonic devices, optical neural networks, and LiDAR perception, providing experimental evidence for the collaborative design of perception and computing in chip-level systems. This achievement also reflects the school's effectiveness in continuously conducting interdisciplinary research in the field of optoelectronic information and promoting collaborative innovation among multiple institutions, laying the foundation for subsequent research on high-performance intelligent optoelectronic perception technology.

This research was carried out by Xidian University in collaboration with the Xi'an Microelectronics Technology Institute, the Xi'an Institute of Optics and Precision Mechanics of the Chinese Academy of Sciences, the Xi'an Institute of Electromechanical Information Technology, and Chongqing United Microelectronics, and was supported by projects including the Ministry of Education's Basic Disciplines and Interdisciplinary Breakthrough Plan, the National Natural Science Foundation of China, and the Fundamental Research Funds for the Central Universities.