In-sensor optoelectronic computing using electrostatically doped silicon

Complementary metal–oxide–semiconductor (CMOS) image sensors allow machines to interact with the visual world. In these sensors, capture in front-end silicon photodiode arrays is separated from back-end processing. To reduce energy cost associated transferring data between sensing and computing units, in-sensor approaches are being developed where images processed within arrays. However, such methods require electrostatically doped photodiodes photocurrents can be electrically modulated or programmed, this challenging current CMOS that use chemically photodiodes. Here we report using We fabricate thousands of dual-gate p–i–n photodiodes, which integrated into at wafer scale. With a 3 × network demonstrate processing seven different convolutional filters programmed network. A compatible complementary fabrication processes, perform by filters.