headerSearch form

Changing the World through Creative Research

In-sensor optoelectronic computing using electrostatically doped silicon

Journal
Nature Electronics
Date
2022.08.23
Abstract

Complementary metal-oxide-semiconductor (CMOS) image sensors are a centerpiece of machines that visually interact with the world. While present CMOS imagers separate image capture in front-end silicon photodiode arrays from image processing in digital back-ends, an approach to at least partially process images within the photodiode array itself is emerging to minimize the energy cost for shuttling data between sensing and computing. Electrical modulation of photocurrents is a requisite for such in-sensor computing, which was indeed recently demonstrated with electrostatically doped diodes whose photocurrents are tuned by gating, but their build was non-silicon materials. CMOS imagers are currently incapable of in-sensor computing, as the photocurrents of their chemically-doped silicon diodes cannot be electrically modified. Here we report in-sensor computing with an array of electrostatically doped silicon p-i-n photodiodes with dual-gate controls. Such electrically programmable silicon photodiode array can be integrated with the rest CMOS imager electronics, and this implementation strategy within the mainstream silicon electronics framework could expedite the bringing of the in-sensor computing to the real-world. Our wafer-scale fabrication of thousands of electrostatically doped silicon p-i-n photodiodes shows their suitability for integration with CMOS electronics. We then demonstrate in-sensor processing of serial optical images oncoming to a 3 × 3 array of these photodiodes, which is electrically programmed into a variety of convolution filters.

Reference
Nature Electronics, 5, 51-525 (2022)
DOI
http://dx.doi.org/10.1038/s41928-022-00819-6