Memristor Based Chaotic Neural Networks And Applications

In 1971,Chua,a well-known Chinese-American professor,first studied and proposed a fourth passive basic circuit element-memristor,which based on the principle of symmetry.In 2008,HP Labs successfully developed the first physical devices of memristor,which greatly promoted the development of memristor.During the 1980 s,the emergence of Hopfield network,which has important applications in many fields such as associative memory and image processing,injected new vitality into neural networks.Neural networks have highly nonlinear network structures,while chaos is a complex dynamic behavior generated by nonlinear systems.Meanwhile,chaotic neural network is a highly nonlinear dynamic system,which can optimize global search and neural calculation,and can also produce pseudorandom sequence.Memristor is a nonlinear resistance with memory,which not only can simulate the characteristics of neurons and synapses,but also can be used as a nonlinear element to generate chaotic sequences.It can further realize the integrated computing system framework for general-purpose inmemory computing by virtue of its non-volatile.Therefore,the nanoscale memristor is one of the best choices for realizing neural networks.In 2020,HP Labs developed a new third-order nano-circuit element,which realized the function of single electronic element that can mimic a neuron,paved the way for extremely compact and dense functional neuromorphic computing units and high-efficiency neuroscience model verification.Against this background,a hyperbolic tangent memristor model is proposed.Based on this model,the memristor chaotic Hopfield neural network and the memristor cell neural network are designed respectively.The nonlinear characteristics of these memristor neural network are analyzed in detail by using dynamic analysis method.In addition,the chaotic sequence generated by the memristor cell neural network is applied to information encryption.The main innovative contents of this paper are as follows:(1)Combined with the traditional definition and characteristics of memristor,a mathematical model of hyperbolic tangent memristor and its equivalent circuit are designed.Through Matlab numerical analysis and Pspice circuit simulation,the results show that the hyperbolic tangent memristor has a frequency-dependent hysteresis loop characteristics,which verifies the memristive characteristics of the model.(2)Based on the proposed memristor model,the memristor Hopfield neural network and the memristor cell neural network are designed,and their equivalent circuits are constructed,respectively.Through Matlab data analysis and Pspice circuit simulation,the nonlinear dynamic analysis of these systems is carried out,including the solution of equilibrium points and the judgment of their stability,phase diagram,time-domain waveform,Poincaré mapping,system bifurcation diagram and Lyapunov exponent spectrum,etc.It is proved that the memory chaotic neural network has rich dynamic characteristics,which can generate many different types of coexisting attractors.In addition,the complex chaotic signal generated by the neural network has good random characteristics and is suitable for intelligent information processing applications.(3)Based on the designed chaotic neural network,a digital watermarking algorithm and a face information encryption algorithm for color watermarking are proposed.To enhance the user experience of watermarking image information,a simple GUI application program is designed combining Python and Matlab to realize the function of embedding and extracting watermarks.