Abstract: Ion based synaptic devices (ISDs) are one of the excellent candidates for neuromorphic computing. However, most of ISDs utilized additional ion sources to supply ions for adjusting the conductance of the device channel, which might hinder the large-scale integration for fabricating hierarchical artificial neural network. Here a high-performance monolayer MoS₂ ISD is demonstrated using Na⁺ ions doped in MoS₂ lattice as ion sources. Benefited from the Na⁺ ions and S vacancy defects in the MoS₂ lattice, the device not only exhibits various synaptic plasticity (long- and short-term plasticity) and typical biological features (pain-perceptual nociceptors and associative learning), but also has a low synaptic event response voltage (100 mV) and a low energy consumption (0.92 pJ) for a synaptic event. A dissociation-adsorption-migration-binding model is proposed to elaborate the resistance switching mechanism, which is corroborated by density functional theory calculations and characterizations. In addition, an artificial neural network (ANN) based on MoS₂ ISDs is simulated for the recognition of the MNIST handwritten digits. The deviation of the recognition accuracy is less than 8% compared to the ideal floating-point numeric precision. These results provide a new strategy for fabricating high-performance ISDs for neuromorphic computing.

Key words: 2D materials, Monolayer, Molybdenum disulfide, Synapses, Neuromorphic computing