Title: Zeolite‐Based Memristive Synapse with Ultralow Sub‐10‐fJ Energy Consumption for Neuromorphic Computation
Abstract: The development of neuromorphic computation faces the appreciable challenge of implementing hardware with energy consumption on the level of a femtojoule per synaptic event to be comparable with the energy consumption of human brain. Controllable ultrathin conductive filaments are needed to achieve such extremely low energy consumption in memristive synapses but their formation is difficult to control owing to their stochastic morphology and unexpected overgrowth. Herein, a zeolite‐based memristive synapse is demonstrated for the first time, in which Ag exchange in the sub‐nanometer pore closely resembles synaptic Ca2+ dynamics across biomembrane channel. Particularly, the confined ultrasmall pore and low Ag ion migration barrier give the zeolite‐based memristive synapse ultralow energy consumption below 10 fJ per synaptic spike, on par with the biological counterpart. Experimental results reveal that the gradual memristive effect is attributed to the dimension modulation of Ag clusters. In addition to emulating inherent cognitive functions through electrical stimulations, the experience‐dependent transition of short‐term plasticity to long‐term plasticity using a chemical modulation method is achieved by treating the initial Ag quantity as a learning experience. The proposed memristors can be used to develop highly efficient memristive neural networks and are considered as a candidate for application in neuromorphic computation.