Memristor-based Fuzzy Inference System Design And Application Research

The fuzzy inference system has been considered as an intelligent system who can be constructed close to the brain computing ability.The fuzzy system,relative to the neural network,make good use of expert knowledge and requires lower samples,the reasoning process of it is easy to understand.But it also has the disadvantages such as too many manual intervention,slow reasoning and low precision,and it is difficult to realize the function of adaptive learning.Artificial neural network has a strong self-learning and association function,it is used to simulate the thinking function of human brain and need less manual intervention,higher precision.But the disadvantage is that it cannot handle and describe the fuzzy information,cannot make good use of the existing empirical knowledge,and it also has a higher requirement for samples.it can play a complementary effect if the two organically combined.The implementation of nanoscale device memristor makes it possible to realize the hardware circuit of the brain-like.The memristor is a passive component with the advantages of continuously variable,nonvolatile and fast switching characteristics.When the power is turned off,the charge that can still be "memorized" is similar to that of neuronal synapses and can be used to replace neuronal synapses in hardware implementations.The application of fuzzy systems is expanding,so researchers are keen to find a viable way to achieve a complete set of high-speed and real-time operation of the fuzzy system.In the traditional fuzzy system,the language information is processed in the digital system,the membership function is transformed into binary code when expressing the language variable.With the faster calculate speed requirement of the modern technology,the faster analog circuit form is needed.Taking max and min functions are the most important parts of fuzzy logic,such as its application in fuzzy reasoning.(1)In order to implement the brain-like hardware circuit,this paper proposes a fuzzy reasoning system based on memristor crossbar array,and the complex function is modeled to verify the correctness of the designed system.Firstly,the traditional forward neural network structure is transformed into a crossbar array structure,and used to store the fuzzy rules.Then,a hardware circuit design scheme of fuzzy reasoning system based on the structure of HP memristor is proposed.When the input data is enter,the fuzzy reasoning result can be calculated according to the output voltage of the crossbar array.Finally,the complex function is modeled by the designed reasoning system,and the numerical simulation proves the correctness of the system.(2)A fuzzy PID controller based on memristor crossbar array is designed using the proposed fuzzy inference system,which is used to control the typical object.Through the numerical simulation we can get the control surface and the control parameter curves.The control curve shows that the system can clearly exhibit each control surface.The response curves and control parameters are indicated the fuzzy control system with good performance.Compared with the traditional PID control and fuzzy PID control,the validity of the designed fuzzy controller is confirmed.(3)Based on the fuzzy logic principle of ink drop spreading,a fuzzy logic gate circuit based on the crossbar array of spin memristor is designed to store the fuzzy relation.And the correctness and feasibility of it are verified by LTSPICE circuit simulation.This fuzzy logic gate can implement the fuzzy operation of "AND","OR","NOT","XOR" with fewer steps.The fuzzy logic gate structure is simple,with the fast operation speed,the low hardware realization cost,the small volume,and the low energy consumption,and the application wide scope.Not only for the traditional analog and digital circuits,and fill the fuzzy logic gate circuit's blank,this circuit can also be extended to 3-valued and multi-valued fuzzy logic.The fuzzy logic gate provides the basis of the circuit components for the fuzzy system and the fuzzy neural network hardware implication.