Optical AI Fits Three Neural Network Layers Inside One Piece of Glass
Credit: GizmoGuru

Optical AI Fits Three Neural Network Layers Inside One Piece of Glass

A glass device processes four colors of light, demonstrating a compact approach to optical AI that avoids aligning separate processing layers.
Nov 21, 2026
1 min read
0.00 MPXR

Researchers have built an optical neural network with three processing layers inside a single piece of silica glass, tackling an awkward engineering problem: keeping separate optical components precisely aligned.

The study, posted November 20 in Photonics Research, uses ultrashort laser pulses to modify the glass internally. Those modifications create layers that shape passing light, allowing the material itself to perform classification tasks.

Four visible wavelengths provide separate processing channels within the same physical volume. In experiments, the device classified 20 categories drawn from three datasets with 80% accuracy overall. Its strongest individual channel reached 86%.

The approach adds a practical manufacturing advance to work on neural networks made of light. Integrating layers inside one solid structure avoids the repeated positioning required when optical layers sit apart, potentially making deeper systems easier to assemble and maintain.

The researchers also report 96.7% accuracy for a theoretical extension using additional spectral channels. That figure was not demonstrated by the fabricated device.

This remains specialized classification hardware. Its broader significance is a clearer route toward compact optical computing, where improvements in fabrication must accompany progress in training AI on light-based hardware.

About the Writer

More from Mindplex

Keep reading

Three more ideas worth your time.

Browse MindBytes

Discussion

Join the discussion

Sign in to share a response with the community.

Type @ to mention someone Type / or use + to add a block Highlight text, then choose Link
Loading editor

Comments cannot be edited after posting because they become part of the reputation record. Give yours a quick review first.