Bioinspired activation of silent synapses in layered materials for extensible neuromorphic computing

Activation of silent synapses is of great significance for the extension of neural plasticity related to learning and memory. Inspired by the activation of silent synapses via receptor insertion in neural synapses, we propose an efficient method for activating artificial synapses through the intercalation of Sn in layered a-MoO3. Sn intercalation is capable of switching on the response of layered a-MoO3 to the stimuli of visible and near infrared light by decreasing the bandgap. This mimics the receptor insertion process in silent neural synapses. The Sn-intercalated MoO3 (Sn-MoO3) exhibits persistent photoconductivity due to the donor impurity induced by Sn intercalation. This enables the two-terminal Sn-MoO3 device promising optoelectronic synapse with an ultrahigh paired pulse facilitation (PPF) up to 199.5%. On-demand activation and tunable synaptic plasticity endow the device great potentials for extensible neuromorphic computing. Superior performance of the extensible artificial neural network (ANN) based on the Sn-MoO3 synapses are demonstrated in pattern recognition. Impressively, the recognition accuracy increases from 89.7% to 94.8% by activating more nodes into the ANN. This is consistent with the recognition process of physical neural network during brain development. The intercalation engineering of MoO3 may provide inspirations for the design of high-performance neuromorphic computing architectures.& COPY; 2023 The Authors. Published by Elsevier B.V. on behalf of The Chinese Ceramic Society. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Automated summary

Activation of artificial synapses through Sn intercalation in a-MoO3 allows for efficient response to light stimuli and persistent photoconductivity. The resulting Sn-MoO3 device exhibits high paired pulse facilitation and tunable synaptic plasticity, showing great potential for neuromorphic computing. The extensible artificial neural network (ANN) based on Sn-MoO3 synapses demonstrates superior performance in pattern recognition.