silicon-based semiconductor devices are encountering constraints of downscaling with significant data fidelity issues and unaffordable manufacturing cost at the impending 2-3 nm nodes, hindering their direct application for both flexible and very (ultra) large-scale integration (VLSI). [4] One way to solve this problem is by developing in-memory computing technologies. By using new materials and different mechanisms to design and implement memory cells, the "von Neumann bottleneck" can be solved, and the semiconductor industry can maintain an exciting development trend.When dealing with 21st-century applications such as real-time voice recognition, image processing, and big-data analysis, as the huge amount of involved information is usually accessed unpredictably, it may take many processor cycles to fetch the target data from the memory hierarchy and the computer may stall due to the unavailability of the data, resulting in loss of computation performance and power efficiency. Compared with the human brain, modern supercomputers have faster operation speed but much lower efficiency. The human brain transmits signals with neurons. The neural network consists of 10 11 neurons and is connected with 10 15 synapses. The synapse is the nanogap between neurons, which permits electrical or chemical signal transmission. [5,6] Benefited from the neural network structure, the human brain can deal with recognition tasks effectively. When processing massively parallel information, each synapse event consumes about 1-100 fJ. [7] The human brain has strong stability because the neural network does not collapse with the death of nerve cells. Moreover, when humans interact with new things, the neural network will automatically learn and update without the need for human readjustment. [8] In order to execute artificial intelligence algorithms with energy efficiency performance and performance comparable to that of the brain, it is hoped to simulate the function of neural networks (for example, the interconnection of biological synapses and neurons) through the inherent combination of information storage and processing on the device scale. [9] Fortunately, memristor, another resistive switching device beyond the bistable memories, may be important in computer systems before the von Neumann architecture.Memristor was conceived by Chua based on the theory of the symmetry arguments in 1971 and experimentally demonstrated by Strukov et al. [10,11] Memsietor can be defined as a two-terminal device with nonvolatile resistance which can be modulated by light, electricity, heat, and so on. [12] Unlike devices that display a Facing the exponential growth of data digital communications and the advent of artificial intelligence, there is an urgent need for information technologies with huge storage capacity and efficient computing processing. However, the traditional von Neumann architecture and silicon-based storage and computing technology will reach their limits and cannot meet the storage requirements of ultrasmall size, ultrahigh densi...