Research on Construction and Simulation Methods of Memristor-based CNN

This paper proposes a method for constructing and simulating a 3×4 structure convolutional neural network (CNN) based on memristors. By combining the nonlinear characteristics of memristors with the local connection characteristics of CNNs, a novel neuromorphic computing architecture is designed. Multisim is used for circuit - level simulation to verify the feasibility of the memristor cross - array, and Python is employed to implement the complete CNN training process. The experimental results show that this structure achieves an accuracy of 92.3% in the MNIST handwritten digit recognition task, and the power consumption is reduced by approximately 42% compared with traditional CMOS implementations. This research provides new ideas and methods for the application of memristors in neuromorphic computing.