Application Of Memristive Neural Network Circuit In Image Recognition And Associative Memory

Artificial neural network is one of the most popular research directions in the field of artificial intelligence,successfully solving many practical problems in various industries.With the advent of the era of big data,the amount of data that neural networks need to process continues to increase,making it difficult for computers based on the von Neumann architecture to cope with the growing demand for data processing.Memristors have unique advantages such as nonvolatility,high storage density,and low power consumption,and can be used to store synaptic weights in neural networks.Therefore,applying memristors to the field of neural networks can achieve an integrated architecture of storage and computing,providing a feasible solution to data processing problems.In recent years,research on neural networks based on memristors has achieved fruitful results in image recognition,simulated associative memory,and other aspects,causing widespread attention by researchers.Because the memristor has the characteristics of resistance value switchability and low power consumption,the memristor neural network circuit can adjust the weight value in the training process.When the circuit carries out image recognition tasks,it can efficiently complete the recognition task of a variety of different images.In addition,associative memory is an important cognitive function in the biological brain,and it is also an important topic in the research and application of memristor neural networks.Therefore,studying the application of neural network circuits based on memristors in the fields of image recognition and associative memory has certain significance.The main content and innovations of this article are as follows:(1)The classic 1M,2M,4M and 5M synaptic circuits are analyzed and compared,and the advantages and disadvantages of each synaptic circuit in terms of weight range,number of memristor,circuit area and implementation difficulty are summarized,which provides a basis for the design of new memristor synaptic circuits.(2)A new type of memristor synaptic circuit is designed,which has the advantages of less use of control voltages,and memristors,wide weight range and linear programming process.In the further research,firstly the corresponding neuron circuit and activation function circuit are designed based on the synaptic circuit;Secondly,a neural network circuit that can recognize three types of character images was implemented combining the Widrow-Hoff algorithm;Finally,the correctness of the overall scheme was verified through SPICE simulation.(3)Based on the digital logic gate composed of the above-mentioned memristor synaptic circuit and memristor,a memristor neuron circuit and control circuit are designed,and then a memristor neural network circuit used to simulate Pavlov’s associative memory experiment process is realized.This circuit can accurately realize the learning,food forgetting and bell forgetting processes in Pavlov’s associative memory experiment.SPICE simulation analysis has verified the effectiveness and correctness of the circuit.