Research And Design Of Neural Synapse Circuit Based On Memristor

Memristors have become the first choice for artificial synapses due to their synapse-like advantages such as continuously adjustable resistance,nanoscale size,and low power consumption.Associative learning and non-associative learning are the basic methods of biological learning.Using memristors as artificial synapses to simulate biological learning behavior is the basis and key to further research on neuromorphic systems.The memristor-based physical model improves its mathematical model so that it has a performance that is adapted to synaptic behavior,which can guide its design and manufacturing process to improve its bionic synaptic performance.This paper introduces a variety of memristor models,and finally selects HP memristor model and tungsten oxide forgetting memristor as synapses to simulate biological behavior and build circuits.In this paper,different memristor models are discussed in depth,and the characteristics of each model are analyzed through simulation.On this basis,a habituation circuit based on biological non-associative learning is designed using memristors as biological synapses,and the complete habituation behavior of biology is realized;a conditioned reflex circuit based on biological associative learning is designed and its functions by extension,the learning and forgetting processes of various conditioned reflexes of organisms are realized;a recognition circuit based on biological joint learning is designed,and the function expansion is carried out to realize the recognition and association functions of organisms.The research contents are as follows.Firstly,four memristor models are studied in detail,and the performance of each memristor model is analyzed through mathematical derivation and simulation.At the same time,a variety of window functions are introduced based on the nonlinear behavior of HP memristor,and various window functions are analyzed.The influence of different parameters on the performance of the memristor lays the foundation for the subsequent circuit design.Secondly,the existing synaptic circuits are summarized,the weight tunability of the existing synaptic circuits is analyzed through simulation,the synaptic circuit is improved based on the characteristics of the bridge synapse and the memristor composite circuit,and a new synaptic circuit based on the improved synaptic circuit is proposed.The analysis of synapses provides support for the subsequent simulation of biological behavior.Then,the habituation behavior based on biological non-associative learning is summarized,a dual-input habituation circuit and a single-input habituation circuit are designed based on HP memristor,and the effectiveness of the circuit is verified by simulation,including short-term habituation circuit,the complete habituation behavioral process of long-term habituation and dehabituation,paving the way for memristor-based neuromorphic systems.Finally,the conditioned reflex and associative memory learning functions based on associative learning are introduced.Based on the tungsten oxide forgetting memristor,a conditioned reflex circuit is designed.Combined with the Pavlovian dog experiment,the complete biological conditioned reflex learning and forgetting process is realized,and the function is expanded to realize the learning and forgetting of various conditioned reflex behaviors.Compared with the joint learning circuit based on the HP memristor model,the forgetting memristor simplifies the peripheral circuit and reduces the circuit complexity.Then,a recognition circuit based on forgetting memristor is proposed,which realizes the recognition of living things,and extends the function to realize the associative memory function of living things.The simulation shows that the designed circuit has better learning and forgetting ability.