A symmetry-reconfigurable photodiode for sensing and computing

Direct photocurrent computation at the sensor level could be used to address the stringent power constraints of smart edge applications such as microrobots and wearable electronics. However, the lack of viable device models and scalable integration methodologies restricts the implementation of such direct photocurrent computation. Here we report a symmetry-reconfigurable photodiode for sensing and computing. This device is based on silver bismuth sulfide, which is patterned between two high-work-function electrodes to form symmetrical back-to-back Schottky junctions; voltage programming can be used to induce a reversible symmetry-to-asymmetry transition via a localized electrochemical reduction of silver ions, lowering the Schottky barrier on one side of the device. As a result, the device can generate non-volatile, bipolar weights for in-sensor direct photocurrent computation. It is also fully compatible with thin-film transistor read-out circuits. We show that the approach can be used for infrared transmissive imaging as well as eye image recognition and actuator control.