Circuit Design Of Associative Memory Neural Network Based On Memristor And Its Applications

In recent years,the rapid development of neural network theory has promoted the indepth study of brain science and the active exploration of brain-like artificial intelligence technology.Associative memory,as the basic function and activity of brain cognition,is the cornerstone of human reasoning and innovation.Therefore,it is also one of the research focuses in the field of brain-like artificial intelligence.According to the existing neural network model,inspired by the cognitive memory mechanism of the brain,the establishment of a neural network system for brain-like associative memory based on hardware is of great significance for the development of a highly intelligent information processing system.Based on the two classic associative memory neural network models of Hopfield neural network and chaotic neural network,this paper adopts memristor with memory function as neuron synapse and constructs memristive Hopfield neural network circuit and memristive chaotic neural network circuit respectively.The resistance adjustment rule of the memristive cross array is designed in this work,which simulates the synaptic weight adjustment of brain memory at the hardware level.The programmable characteristic of memristive synapse improves the flexibility and application value of the circuit.Based on the designed memristive Hopfield neural network circuit,the self-association of the image is realized to simulate the recognizing process of the brain,and the memristive chaotic neural network circuit realizes the hetero-association of the image to simulate the recall process of the brain.According to the theoretical knowledge of biological long and short-term memory conversion and the functional characteristics of the two networks,this paper combines the existing research results of brain-like memory mechanisms and uses the memristive Hopfield neural network as the short-term memory network and the memristive chaotic neural network as the long-term memory network.This paper designs a long-term and short-term memory autonomous conversion circuit based on these two associative memory neural networks,which realizes the transfer of the training results from the short-term memory neural network to the long-term memory network,and stores the memory information in the long-term memory network.The research results of this paper provide new research ideas and methods for hardware realization of associative memory neural networks and lay the foundation for hardware realization of larger-scale complex associative memory neural systems.Compared with software algorithms,the memristive associative memory neural network circuit can increase the speed of the network.At the same time,the flexibility of the memristive as a nerve synapse allows the circuit to handle complex and changeable application problems.