Comprehensively Modulated\nSub-Attojoule Operated Optoelectronic\nSynapses for Image Encryption and Inpainting

High-performance optoelectronic synaptic transistors\nplay a crucial\nrole in developing and emulating artificial visual systems. However,\ndue to the predominant use of single-structure material modulation\nin optimizing optoelectronic synapses, their energy consumption significantly\ntrails behind that of electronic synapses by several orders of magnitude.\nHerein, polymer dielectric layers and optimized contact strategies\nare adopted to realize the ultralow consumption optoelectronic synapses.\nIntegration of polyimide dielectric significantly enhances photogenerated\ncharge carrier dissociation, leading to substantial improvements in\nphotoresponsivity (1.5 × 10<sup>6</sup> A·W<sup>–1</sup>), photodetectivity (6.9 × 10<sup>12</sup> Jones), and external\nquantum efficiency (4.0 × 10<sup>8</sup>%). Additionally, optimized\ncontact properties augment their appeal for ultralow energy consumption\nin optoelectronic synapse applications. Excitatory postsynaptic current\nis triggered at an incredibly low voltage of 5 μV and boosts\nan impressively low energy consumption of 0.05 aJ, ranking among the\nbest-reported results in this field. Next, we demonstrate an integrated\nsystem combining the MoS<sub>2</sub> optoelectronic synapses with\na recurrent neural network enabling 100% accurate recognition of optical\nsignals, particularly in scenarios with aJ-leveled energy consumption.\nFinally, an image encryption system has been developed, in which images\nare encrypted by photoelectronic conversion of synapse arrays with\nrandom voltage settings and decrypted according to the recurrent neural\nnetwork-based accuracy. More importantly, once partially damaged images\nare encrypted, through the decryption image inpainting can be realized\ndue to the high accuracy. The proposed innovative approach holds promise\nfor advancing artificial intelligence applications with improved energy\nefficiency, information security, and computational capabilities.