Flexible Schottky photodiodes for fJ-level energy neuromorphic computing and nociceptor-mediated synaptic simulation

In this work, we have fabricated a flexible Schottky photodiode, introducing tunneling/trapping structure to achieve dual-modal plasticity. Under no-injured stimuli, normal synaptic functions can be programmed with a very low switching energy of 720 fJ, which is lower than other state-of-the-art artificial nociceptors. Combined with a convolutional neural network, it successfully achieves 94% maximum accuracy in image classification. Under injured stimuli, the device clearly exhibits nociceptor-mediated features, including threshold, no adaptation, relaxation, and reconfigurability. Moreover, under the bending condition, the device also exhibits programmable dual-mode plasticity characteristics, which indicates the potential feasibility of our device for performing the damage alarm task in both dynamic and unstructured environments.