Electrically Controllable Neuromodulation Emulated by 2D Weight-Tunable Memristor for Neuromorphic Application

Two-dimensional (2D) semiconductors have emerged as a promising material for low-power and high-performance electronics because of the intrinsic atomic thickness and the exceptional properties maintaining even with ultimate scaling. Besides, the competitive ability to electrostatically control the electrochemical potential allows us to design band-modulated 2D heterostructures for implementing a variety of gate-tunable electronic devices. Such a unique capability of 2D materials can also offer great potential for realizing an energy-efficient artificial synapse with high controllability. Nevertheless, the artificial synapse utilizing functionally unique properties has rarely been demonstrated, as appropriate materials and structures with robust memristive switching characteristics and an adequately integrated device architecture are not available.<br/> Here, we report a functionally advanced artificial synaptic architecture, a three-terminal device consisting of a defect-controlled molybdenum disulfide (MoS₂) memristor on hexagonal boron nitride (<i>h</i>-BN), termed as a ‘weight tunable memristor’. Through the precise defect control of MoS₂ channel, the device exhibits low power switching phenomena even without applying gate voltages, which cannot be implemented in previously reported memtransistors utilizing gate dielectric as a pre-synaptic component. One more step, owing to the electrostatically controlled space charge limited current in the ultrathin channel, the device exhibits gate-controlled memristive switching characteristics. The device can implement essential synaptic characteristics, such as short-term plasticity and long-term plasticity. Notably, by electrostatic tuning with a gate terminal, we can additionally regulate the degree and tuning rate of the synaptic weight independent of the programming impulses from source and drain terminals, with sub-1 FJ pulse input. Moreover, the changed states are within stable region for 1500 consecutive pulses. These capabilities eventually enable the accelerated consolidation and conversion of synaptic plasticity, functionally analogous to the synapse with an additional neuromodulator in biological neural networks. Furthermore, such acceleration improves the recognizing accuracy and reduces learning step in MNIST pattern recognition, with considerable power-saving benefits. Our demonstration represents an important step toward highly networked and energy-efficient neuromorphic electronics.