Design And Dynamic Analysis Of Memristor-based Neural Networks

Memristor,which is regarded as the fourth circuit element after resistor,resistance,and inductance,is postulated by Leon Chua in 1971 originally and developed by HP Labs in 2008.Memristor,with its function is analogous to the biological synapse,can consist of a memristor-based neural network.Inheriting the low power consumption and nano volume of memristor,memristor-based neural network has higher performance.In order to ensure the stability and efficiency in application,it is meaningful to investigate the dynamic behaviors of memristor-based neural network.In this thesis,a math model of memristor is used to design the memristor-based neural network and its dynamic behaviors is studied.The main work is as follow:1)Research on design of memristor-based neural networkIn this thesis,by using a math model of mermristor with piece linear characteristic,two classes of memristor-based neural networks are designed.One is memrisotr-based WTA neural network(MWNN),which is consisted of memristor,MOSFETs,resistors,capacitors and other circuit elements.Another is a drive-response memristor-based recurrent neural networks with leakage delay and time-varying delay,which is consisted of memristor,MOSFETs,resistors,capacitors and amplifiers,and other circuit elements.The state equations of these two systems are obtained.2)Research on dynamic analysis of memrisotr-based WTA neural networkDynamic analysis of MWNN is presented,and sufficient conditions for WTA point to exist and network convergence are obtained.A BP-MWNN classifier system,which is consisted of a BP neural network and a MWNN,is designed to diagnose erythematosquamous diseases.Finally,illustrative examples are given to demonstrate our obtained results.3)Research on dynamic analysis of drive-response memrisotr-based recurrent neural networkAn extensive rigorous(Q,S,R)-?-dissipativity analysis of the drive-response memristor-based recurrent neural networks with leakage and time-varying delay is presented.The dynamics of the system has been modeled by the theories of differential inclusion and set-valued map.Also,by using the free-weighting matrices technique,an appropriate Lyapunov Krasovskii function and linear matrix inequality(LMI)technique,several relaxation criteria for the strict(Q,S,R)-?-dissipativity are postulated.Finally,some numerical examples are provided to demonstrate the effectiveness of the proposed theoretical results...