Flexible memristors based on ZrO2/ZnO heterojunctions for neuromorphic computing

A brain-like neural network constructed by memristors has attracted wide attention, which is expected to break the limitations of traditional von Neumann design. In the work, a flexible artificial synaptic device was prepared by radio-frequency magnetron sputtering, and the resistive switching characteristics were investigated. The transition between the high and low resistance states of this device is primarily attributed to the formation and fracture of conductive filaments. The device can achieve typical synaptic behaviors, including short-term/long-term plasticity, paired pulse facilitation and peak time-dependent plasticity. Importantly, the device still has stable synaptic properties under bending conditions. In addition, the convolutional neural network (CNN) constructed by the device has good accuracy for MNIST handwritten digit dataset and Fashion-MNIST clothing dataset. These results provide a feasible method for creating an effective neuromorphic network in the future. • The device effectively mimics synaptic plasticity, including STP, LTP, PPF and STDP. • The ZrO 2 /ZnO memristor device has good synaptic plasticity in the bending state and demonstrates good flexibility. • The device has a stable recognition rate of over 96 % for handwritten digits in both bent and extended states.