Voltage modulated long-term plasticity in perovskite heterostructured memristive synaptic devices with high-performance neuromorphic computing

The development of neuromorphic computing is expected to enable the computer to realize the integration of storage and computation. The development of memristors provides hardware support possibilities for the development of neuromorphic computing. In this work, we have prepared a (La0.67, Sr0.33)MnO3/BaTiO3-based memristor with good forward and reverse memristor function and multilevel resistive tunability, including an increased resistance state at forward voltage and a decreased resistance state at reverse voltage. This is mainly due to the barriers of the ferroelectric dielectric layer and its ferroelectric polarization under the electric field, and the migration of oxygen vacancy under the electric field. The devices also successfully implement the synaptic simulations of short-term plasticity, long-term plasticity, excitatory postsynaptic current, paired-pulse facilitation, spike-rate-dependent plasticity, and spike-timing-dependent plasticity and reimplement these synaptic simulations by varying the amplitude and pulse width of the applied voltage. We have also achieved a classification accuracy of 96.7% for the given handwritten digit data by an artificial neural network with supervised learning. The high classification accuracy is attributed to the good nonlinearity of the device in terms of continuous conductance decreased (0.91) and increased (0.58). Our results are expected to provide a good reference value for neural devices.