OFE2 uses integrated diffraction modules to perform feature extraction — the computational step that identifies patterns in images, audio, or sensor data — at the speed of light. Processing at 12.5 GHz, it handles real-time recognition tasks that electronic chips accomplish with far more energy.
The chip combines optical diffraction with data preparation modules on a single integrated platform. This matters because feature extraction is the most computationally expensive part of many AI pipelines. Doing it optically frees electronic processors to handle the remaining computation.
OFE2 complements LightGen in China's post-silicon computing strategy. Where LightGen targets generation tasks, OFE2 targets recognition and classification. Together, they suggest a future where AI's most demanding computations migrate from electronics to photonics, with electronic chips handling only the glue logic.