Optical Computing: Random-Projection Optical Neural Networks Printed at Record Speeds | Newswise

Researchers at The Chinese University of Hong Kong have achieved a significant milestone in optical computing. Professor Shih-Chi Chen and Professor Chaoran Huang led a team that developed a high-throughput randomized multi-focus two-photon lithography platform. This new method allows for the fabrication of four million 500-nanometer neurons on a millimeter-scale chip in only 15 minutes.

Traditional hardware often struggles to match the power and speed demands of modern machine vision tasks. Optical neural networks offer a path forward by processing data at the speed of light while keeping energy usage low. Previously, creating these networks at visible wavelengths proved difficult due to the sheer number of neurons required, making traditional manufacturing methods slow and expensive.

The team's approach uses a task-agnostic optical encoder. It performs random projections through a 3D-printed diffractive layer. When paired with a compact camera and a lightweight digital readout layer, the system achieves 97 to 99 percent classification accuracy in tasks like facial keypoint detection and human action recognition.

This fabrication technique is compatible with ultra-low-cost UV nanoimprinting, which allows for mass production. By moving past the slow pace of conventional lithography, this technology provides a practical route to integrating optical vision processors into daily devices. Future work aims to expand the operational range from ultraviolet to infrared regimes and increase device size to centimeter-scale applications.

This development addresses the gap between prototyping and large-scale manufacturing. It positions optical computing as a viable option for industries such as biomedical diagnostics, LiDAR, and human-computer interaction.