“…B y mimicking the human brain's architecture, neuromorphic computing can address complex nonstructural cognitive tasks, including learning, memory, decision-making and reasoning. 1−5 However, many neuromorphic computing frameworks, such as artificial neural networks (ANN), 6,7 spiking neural networks (SNN), 8−10 convolutional neural networks (CNN) 1,11,12 and recurrent neural networks (RNN), 13−15 place specific requirements on device characteristics to meet the computation complexity. 16 Memristors, as one of the main drivers of neural network hardware implementation, 7,12,16−18 are divided into volatile and nonvolatile types for optimizing synapses and neurons.…”