Memristor-based Synapse Learning Circuit Design

With the increasing demand for information processing,people are eager to own low-power computing ability similar to human brain.Therefore,the research of neuromorphic system has been widely concerned.However,the integration of existing semiconductor transistors is difficult to further improve,which makes it difficult for the transistor-based neuromorphic system to extend to the biological scale.The smaller electronic synaptic devices with lower energy consumption are needed.Memristor is a new passive device,which has the characteristics similar to the synapse in biological nervous system.It also has the advantages of nanometer size and low power consumption.So memristor is the best choice to realize artificial synapses.The memristor-based neuromorphic system is expected to change the existing information processing methods.In the process of continuous evolution,organisms have formed their own unique way of learning and memory.Using memristor as synaptic to realize biological learning modes can lay the foundation for further realization of memristor-based neuromorphic system.In this paper,the model of memristor is deeply analyzed,and memristive properties are explored.Based on this,the memristor-based circuits are designed to simulate biological learning behaviors.A memristive synapse circuit is proposed to simulate the habituation phenomenon in Aplysia gill-withdrawal reflex.The associative memory circuit is designed to implement the associative memory in Pavlov experiment.The recognition and recall circuits are designed to mimic the ability of human to recognize objects through features and associate features through objects.The work of this paper mainly includes four parts:Firstly,the paper introduces several classical memristor models.For the ion transport model,the HP memristor model is analyzed and its formula for calculating the memristance is derived.For the threshold model,a voltage threshold adaptive model commonly used for logic circuits and a voltage threshold model used for neural networks are analyzed.For the forgetting model,a one-dimensional forgetting model and a three-dimensional forgetting model are analyzed.The analysis of the memristor model lays the foundation for subsequent circuit design.Secondly,habituation characteristic in Aplysia gill-withdrawal reflex is studied,and a habituation circuit based on HP memristor is designed for the realization of habituation characteristic.Compared with previous studies,the circuit designed in this paper can achieve more complete habituation phenomenon.Short-term habituation,long-term habituation,dishabituation and frequency-dependent habituation are all realized by the designed circuit.The realization of habituation broadens the road for the study of neuromorphic systems based on synaptic devices.Then,the associative memory in Pavlov experiment is analyzed.Based on the improved forgetting memristor,the associative memory circuit is designed to realize the biological associative memory ability.Compared with previous associative memory circuits,the proposed circuit achieves complete associative memory function with lower complexity.The learning process and forgetting process in the Pavlov experiment can be achieved by the proposed circuit.The results of Pspice prove the correctness of the designed circuit.Finally,the recognition and recall circuits based on forgetting memristor are proposed.Through the further analysis of associative memory process,the USF learning rule is proposed.And the circuits are designed to realize the functions of object recognition and feature recall.The simulation results show that the designed circuits have the ability of learning and forgetting.