Short-term memory characteristics in n-type-ZnO/p-type-NiO heterojunction synaptic devices for reservoir computing

Abstract

We investigated the short-term memory characteristics of n-type-ZnO/p-type-NiO heterojunction-based memristor devices to achieve a reservoir computing system. The deposited thickness and chemical properties of each layer of the device were verified by transmission electron microscopy and energy-dispersive X-ray spectroscopy. The TiN/ZnO/NiO/Pt device exhibited the resistive switching behaviors of a typical p-n junction-based RRAM with rectification characteristics. The self-rectification characteristics of the device were attributed to the diode effect of the p-n junction and the redox reaction of TiON. Thus, stable and controllable multistate analog memory was obtained using gradual bipolar resistive switching during successive sweeping cycles. Moreover, excellent potentiation and depression were achieved with a stable dynamic range over 10 cycles by coordinated identical pulses. High MNIST pattern recognition accuracy (>94%) was obtained through a three-layer neural network (784 x 224 x 10); short-term memory dynamics were observed using DC current read sweep by controlling the delay time after the set process and conductance decay, such as paired-pulse facilitation, by varying the pulse interval. Finally, we constructed a reservoir computing system with 16 states (4-bit) distinguished by varying pulse streams ranging from [0000] to [1111] using the short-term memory effects of the TiN/ZnO/NiO/Pt device.