Circuit Design Of Memristor-Based Bidirectional Associative Memory Neural Network And Its Application

In recent years,artificial intelligence inspired by the brain has become a major trend in the field of artificial intelligence research.Exploring the cognitive functions of the brain and constructing neural networks based on these functions of the brain have become an important direction in brain science research.This thesis will use the existing neural network as the basis to implement an efficient and universal brain-inspired associative memory neural network through hardware.It is of great significance for the development of a brain-inspired cognitive computing system with a brain-like mechanism and a human-like behavior.Associative memory is the basis of cognitive activities in the brain,and visual filling and the fusion of multi-modal information are the advanced cognitive functions that the brain can achieve.In order to implement these brain-inspired functions in hardware,a bidirectional associative memory neural network circuit based on memristor is proposed in this thesis.This circuit utilizes memristor arrays to represent the weight matrixes and carries on corresponding operations,thereby realizing the computing-in-memory and the parallel processing of information,simplifying the complexity of the circuit and increasing the processing speed.The circuit combines the online algorithm with the memristor array adjustment process,making the circuit more flexible and versatile in various tasks.Using the proposed memristor bidirectional associative memory network circuit,the function of associative memory and visual filling-in of single modal information are realized in this thesis.The circuit can not only process binary pixel images,but also grayscale images,which is closer to the processing mechanism of the brain.On this basis,by modifying the circuit structure,a memristor-based multiple bidirectional associative network circuit is proposed as well.The multiple bidirectional associative memory network circuit has input channels for multi-modal information,and can realize the fusion of multi-modal information and the association between modals.The main achievement of this thesis is the design of the memristor-based bidirectional associative memory neural network circuit,which can realize single-modal associative memory and visual filling-in,as well as the fusion and association of multi-modal information in a more brain-like manner.Compared with software algorithm implementation,this circuit has higher versatility and much faster processing speed.It has certain reference significance for constructing a highly integrated,fast processing and general brain-like information processing system.