LightGen Photonic Chip Enables Large-Scale AI
Researchers at Shanghai Jiao Tong University and Tsinghua University published a Science paper presenting LightGen, an all-optical processor that integrates over two million photonic neurons on a quarter-square-inch device. The chip uses an "optical latent space" with metasurfaces and fiber arrays to process full-resolution images and reportedly achieves more than 100× the speed of an Nvidia A100 in laboratory tests. LightGen performed high-resolution semantic image generation and 3D manipulation but currently requires external laser setups and specialized manufacturing.
Key Points
- 1Integrates over two million photonic neurons on a quarter-square-inch chip, enabling large-scale optical inference.
- 2Uses an optical latent space with metasurfaces and fiber arrays to process full-resolution images without patching.
- 3Offers reportedly >100× throughput versus Nvidia A100, indicating potential for energy-efficient generative-AI acceleration.
Scoring Rationale
High novelty and peer-reviewed Science publication, but prototype reliance on external lasers limits immediate deployment.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems