Modeling And Application Of Novel Memristive Neural Networks

In 1971,Chua first proposed the concept of memristor,which is a kind of nonlinear resistor with memory characteristics and is the fourth basic circuit components except for resistor,inductor,and capacitor.It was not until 2008 that HP Labs successfully developed the first practical Nano-scale physical device,opening up a new era of memristor research.The neural networks originated in the 1940s,but it was not until the 1980s that Hopfield network was put forward and applied that the research of neural networks entered a prosperous period.Neural network is a nonlinear dynamic system with strong adaptability,robustness,learning and fault tolerance.It has important research value in many fields such as pattern recognition,image processing and artificial intelligence.In addition,memristors are considered to be the best choice for achieving neuronal synapses,and the study of combining memristors with neural networks is of great significance:The constructed memristive neural network models can not only have the characteristics of chaos and associative memory,but also break through the limitations of synaptic circuit design,which provide the possibility for hardware implementation of brain-like neural network systems.Therefore,based on the theories of memristor and neural networks,this paper designs two memristor models,an improved memristive Hopfield neural network model and a novel memristive neural network model based on improved HP memristor synapses.These memristive neural networks are applied to image encryption and emotional simulation respectively,and good experimental results are obtained.The main research contents are as follows:(1)Based on the research of memristors and HP memristor model,an improved HP memristor model and an exponential flow-controlled memristor model are proposed.And the equivalent circuit of the latter is designed.The improved HP memristor model is simulated by Pspice and the equivalent circuit of the exponential flow-controlled memristor is simulated by Multisim.The experimental results prove the correctness of the proposed models.(2)A memristive Hopfield neural network model with only three neurons and chaotic properties is proposed by using the theories of memristor and neural networks to fiirther study the Hopfield network.In addition,the design of the memristor synapse is realized by the improved HP memristor,and a new memristive neural network model is proposed,which lays a foundation for the proposed and applied neural network structure with associative memory.(3)According to the basics of modern cryptography,common encryption algorithms and the unique advantages of chaos in image encryption,a discretization system of memristive neural networks with chaotic properties is presented.By combining it with Logistic mapping,a composite image encryption algorithm was designed.The algorithm was successfully applied to the image encryption and decryption by Matlab.The security of image encryption is analyzed from three aspects:histogram,correlation and key sensitivity.(4)In order to explore the practical value of memristive neural network models in intelligent behavior simulation,combined with associative memory,under the hypothetical scenario,a memristive neural network structure with associative memory is proposed to realize simplified simulation of human emotion.The circuit design and simulation experiment of the memristive neural network structure are carried out by Pspice.The experimental results not only verify the correctness of the model,but also show the process of human emotion changes,reflecting the great application potential of memristive neural networks in intelligent behavior simulation.