Design And Application Of Neural Network Circuit Based On Memristor

Since HP Labs proposed the model and physical object of the memristor in 2008,scientists have never stopped the research on the memristor.The memristor has a variable resistance,non-volatility,nanometer size,and memory.Unique advantages such as performance and low power consumption have become the favored objects of scientists in various fields,and have a wide range of applications in artificial intelligence,chaotic circuits,secure communications,and artificial neural networks.Because the characteristics of the memristor are very similar to the function of biological synapses,with low power consumption and small size,it is most suitable to use the memristor for the simulation of biological synaptic functions.Compared with the traditional circuit constructed with CMOS transistors in terms of reducing the complexity of the circuit and reducing the power consumption,it has great prospects for the use of memristors to replace synapses to construct neural network circuits to simulate human brain functions.This thesis analyzes the theoretical characteristics of the memristor,introduces several memristor models,briefly summarizes the preparation method and process of the memristor,and creates a synaptic circuit for simulation in view of the similarity between the memristor and biological synapses,And carried out physical verification,created the memristor-based identification circuit,recall circuit,competition circuit and logic gate circuit,performed simulation and analysis,and carried out physical welding to verify the theoretical results.The research content of this thesis includes the following four parts:(1)The method and process of the physical preparation of the memristor were introduced,and the typical V-I hysteresis curve was obtained by testing the physical memristor,which proved the feasibility of the physical memristor prepared in this laboratory.(2)Create a synaptic circuit based on the memristor and simulate it on PSPICE software,perform physical welding of the synaptic circuit,replace the synaptic function with the memristor prepared in our laboratory,test and debug,and prove the memristive The device can simulate synaptic function.(3)Create and simulate the recognition circuit and recall circuit based on the memristor.After the recognition circuit is learned,the fruit can be identified by a certain characteristic of the fruit.After the recall circuit is learned,the characteristic of the fruit can be detected.This is the learning process of the circuit,and forgetting will occur after a period of time.The welding and debugging of the hardware circuit of the identification circuit and the recall circuit proved that the circuit has a good learning and forgetting function,laying a good foundation for the artificial neural network to simulate more complicated human behavior.(4)Create a synaptic circuit based on a memristor applied to a competing circuit for data storage,combine three winner-takes-all circuits to complete the competition process,and perform hardware circuit welding,testing and debugging for the winner-takes-all circuit,Combine each part into a competitive neural network circuit,which can better perform image recognition applications.(5)Apply memristors to logic gate circuits,construct and simulate memristor-based AND gates and OR gates,perform welding,testing and debugging of physical circuits,and prove the feasibility of applying memristors in the field of logic gate circuits.