The Study On The Implementation Of The Intelligent Systems Based On The Memristor Crossbar

In 1971,Leon O.Chua determined the relationship between the electric charge and magnetic flux according to the completeness of the circuit theory.He also described the relationship with an electron device which was called memristor.The memristor was called the fourth basic circuit element besides resistor,capacitor and inductor.In 2008,HP LABS realized the memristor's preparation using a sandwiched between two platinum?Pt?electrodes of titanium dioxide thin film structure:a layer of titanium dioxide(TiO_2-x)has strong electrical conductivity because of the missing part of the oxygen atoms,the other layer of pure titanium dioxide has a very high impedance.Under the condition of the applied bias voltage,the movement of the interface between the two layers of titanium dioxide causes the change of oxygen vacancy distribution state,and then the total memristance changes.Once the applied voltage is disconnected,the interface position between the two titanium dioxide layers remains,and it does not change until the next bias voltage applied.The characteristic described is called the memory capability.The physical implementation of the memristor,which is announced by HP LABS,has aroused the widespread concern of researchers all over the world.The number of relevant research institutions at home and abroad has reached more than one hundred,such as HP LABS,SK Hynix and HRL laboratories,the University of California at Berkeley,University of Michigan,Britain imperial college,Huazhong University of Science and Technology and so on.The characteristics,mathematical model and SPICE macro model of the memristor are studied by a large number of researchers both at home and abroad.The integrability of nanoscale memristors and CMOS circuits and the physical implementation of different types of memristors are analyzed,such as HP titanium dioxide film memristor,spintronic memristor,gel memristor and organic memristor,etc.Further,the applications of the memristor at the nonvolatile memory,chaos,broadband electromagnetic radiation modulation,very large gain amplifier,etc are discussed.The Advanced Intelligent System?AIS?is speed up by the field of the software engineering including machine learning,big data analysis and cloud computing,etc.It has aroused great interest of researchers that an artificial intelligent system like a brain of human which can learn and process the information is built.However,the current computing systems face considerable challenges in processing unstructured data such as voice,images and physiological signals and so on,because the Von-Neumann bottleneck induces latency and power consumption issues.In this paper,the structures and working principles of some representative memristors are studied,and their characteristic curves are analyzed in detail through simulation experiments,so as to explore their potential applications according to the unique properties of different memristors.In chapter 1,the emergence of HP memristor,structure and working principle are mainly introduced,and the characteristics of HP memristor are analyzed in detail through the simulation results.Then the models of some representative memristors are described and its characteristics are analyzed.Finally,the combination of memristor and intelligent system is proposed based on the analysis of memristor to solve many disadvantages of intelligent system based on existing hardware architecture.In chapter 2,the radial basis function?RBF?neural network based on the memristor crossbar is presented.Spintronic memristors have received significant attention as a potential building block for control systems,and particularly,it can be laid out in a high-density grid known as a crossbar array.Meanwhile,radial basis function?RBF?neural network control algorithm can effectively improve the control performance against large uncertain systems.Therefore,based on the study of characteristics of RBF neural network and spintronic memristors,this paper proposes a RBF neural network control algorithm based on spintronic memristor crossbar array.Then analyzes its theoretical derivation process and core design idea.Finally,the system simulation model,which uses a two-link robotic manipulator as control object,is built to prove the algorithm's validity and feasibility.Simulation results show that the proposed algorithm can satisfy the effect of presupposition.In chapter 3,the implementation scheme of stochastic computing based on memristor system and material implication?IMPLY?logic are mainly described.Traditional nanometric scaling CMOS technology's higher integration density,lower voltage and current will remarkably impact the electronic circuits'performance and reduce their reliability.As an alternative to traditional binary computing,stochastic computing has unique advantages such as strong fault tolerance and low hardware cost.However,stochastic computing has a very high demand on the response time of the hardware system since it adopts a serial computing method based on bit.While the existing CMOS devices are difficult to meet the requirements of the stochastic computing.Therefore,a novel implementation scheme of stochastic computation based on memristive systems is put forward in this paper based on the merits of memristor like low power,high integration,quick response,and compatibility with CMOS technology,etc.A corresponding hardware structure is also built.Further,a stochastic computation system based on memristive combinational logic is structured and its validity is verified successfully by operating a case.Finally,the memristive stochastic computation is applied to edge detection.By comparison the simulation results of the proposed approach and the conventional approach,the memriative stochastic computation has obviously advantages in edge detection.This implementation maintains the merits of conventional stochastic computation.Moreover,it has more advantages than conventional stochastic computation at power consumption and hardware area.In chapter 4,the fuzzy logic and membership function implementation scheme based on memristor system is mainly presented.With the increasing complexity of the controlled object,higher requirements are put forward for the automatic control technology.On the basis of fuzzy mathematics theory,a new intelligent control method--fuzzy control is proposed,which is used to solve the complex nonlinear control objects difficult to be described by accurate mathematical models.At present,most of the hardware implementation methods of fuzzy control algorithm can only be realized by digital hardware devices?such as FPGA DSP,etc.?.The digital implementation method is contrary to the essence of fuzzy control concept,because the computing accuracy of these digital devices is limited.Therefore,the implementation scheme of fuzzy logic and membership function based on memristor system is proposed based on the unique advantage of the memristor's automatic continuous memory.In this paper,the Simulink model of spintronic memristor is established,and its basic characteristics are analyzed numerically.And an implementation scheme of the fuzzy membership function based on spintronic memristor crossbar array is proposed taking advantage of the physical properties of memristor devices with nanometer size and nonvolatile properties.The effectiveness and correctness of the proposed scheme were verified by comparing it with MATLAB fuzzy toolbox.Finally,the crossbar array based on the spintronic memristor was applied to the classical fuzzy control system--cold and hot water valve control system.The effectiveness of the proposed system was verified by a series of flow and temperature comparisons,error analysis and numerical simulation.In chapter 5,the implementation schemes of the combination of memristors and intelligent systems are analyzed based on the above analysis,and its significant advantages and disadvantages are summarized.Finally,the follow-up research direction is proposed.