Updated
Updated · Newswise · Jul 17
CUHK Team Fabricates 4 Million 500-nm Optical Neurons in 15 Minutes, Reaching 99% Accuracy
Updated
Updated · Newswise · Jul 17

CUHK Team Fabricates 4 Million 500-nm Optical Neurons in 15 Minutes, Reaching 99% Accuracy

3 articles · Updated · Newswise · Jul 17

Summary

  • A CUHK-led team built four million 500-nm optical neurons on a millimeter-scale chip in 15 minutes using a randomized multi-focus two-photon lithography platform.
  • The system uses holographic light-field control for parallel scanning, tackling a key bottleneck for visible-wavelength optical neural networks that need millions of wavelength-scale neurons.
  • Tests delivered 97% to 99% classification accuracy in hand-drawn figure recognition, human action recognition and face keypoint detection, using a digital readout layer with as few as 1,000 weights.
  • The researchers said the 3D-printed diffractive layer acts as a task-agnostic optical encoder, allowing new tasks to be handled by retraining only the lightweight digital layer.
  • Unlike slower, costlier conventional lithography, the method is designed for UV nanoimprinting and tiled writing, opening a path to mass-produced optical vision processors for LiDAR, biomedical diagnostics and human-computer interaction.

Insights

With this record-setting optical chip, is China positioned to dominate the next generation of AI hardware?
A 15-minute print creates a brain-like chip. Is this the end for traditional silicon manufacturing?
As optical AI promises near-zero power use, how will it reshape our energy-hungry digital world?

Record-Breaking 10,000 Neurons/mm² Optical Neural Networks: China’s Leap in Fast, Scalable AI Chip Manufacturing

Overview

On July 17, 2026, Chinese researchers announced a major breakthrough in optical neural network (ONN) technology by unveiling a rapid fabrication method for high-density ONNs, published in 'Nature Photonics'. This development marks a pivotal moment for optical computing, as it addresses long-standing challenges in scaling up ONN fabrication and integration. The new technique enables the creation of ONNs with an unprecedented neuron density, paving the way for more energy-efficient and high-speed AI hardware. This achievement is set to accelerate advancements in artificial intelligence by making large-scale, powerful optical computing systems more practical and accessible.

...