Circuit Design Of Pavlovian Associative Memory Based On Memristor

In the past decade,many scholars have made great breakthroughs in the research work of artificial neural network.Many practical problems have been solved in many fields by virtue of the poweful intelligent characteristics of artificial neural network.However,the traditional artificial neural network has a complex structure,affected by the gradual invalidation of Moore’s Law,and subject to the von Reumann architecture,the scale of artificial neural networks has been greatly affected.Memristor,as an emerging circuit element,can replace traditional CMOS circuit as a candidate for synapse in neural networks by virtue of its excellent characteristics,such as nanometer size,nonlinearity,low power consumption,non-volatility and variable resistance..Associative memory is an important neural network and a critical biological cognitive function,The use of memristors as the major components of electronic synapses has significance reference value for the circuit implementation of associative memory neural networks.In recent years,a large number of research has been finished on the memristor implementation of Pavlovian associative memory,but the realized functions are not comprehensive.On the basis of the work in recent years,further improvement work has been completed in the memristor realization of associative memory neural network.This work of paper demonstrates a great development of memristor circuits in the direction of hardware implementation of associative memory neural networks,and provide reference for the design of memristive neural network circuit.The work of this paper is mainly divided into the following two parts:(1)An associative memory circuit with variable learning rate when a delay exists is implemented based on memristor.This work is divided into circuit implementations of Pavlovian associative memory and emotion associative memory models.(1)For Pavlov’s associative memory,in recent years,the associative memory bionic circuit has been realized mainly from four aspects-learning,forgetting,variable learning rate and hysteresis characteristics.However,previous research work did not think about the variation of learning rate when the hysteresis characteristic exists,this paper considers this situation,and designed a synaptic module,a variable learning rate module and a hysteresis module based on the amnestic resistor,and implemented three forgetting processes simultaneously.A new circuit for associative memory implementation was designed.(2)For emotion,it also has associative memory.The emotional model is an extension of the Pavlovian associative memory model.This paper improves the emotional associative memory model.Considering the change in learning rate when hysteresis is present,which was not addressed in previous work.And designs a new circuit to implement this situation with memristor.(2)Memristor circuits designed when learning,forgetting,generalization,and differentiation exist simultaneously.The work in recent years only has memristor implementation circuits for learning,generalization and differentiation.On the basis of the previous work,this paper designs a new synaptic circuit based on memristor,so that the synapse can realize the function of forgetting,and combines the realization circuit of input neuron and output neuron,design a new memristor-based associative memory circuit,Comprehensive consideration is given to situations where learning,forgetting,generalization,and differentiation co-exist.