Simulations Of Hopfield Neural Networks Based On Ferroelectric Synaptic Transistors

Compared with the traditional von Neumann architecture,the neural network data processing structure of the biological brain has the characteristics of flexibility,adaptability,and low energy consumption,playing an important role in fields such as learning and decision-making.Therefore,developing computing architectures that imitate neural networks has profound significance.Among them,the researches of artificial neural networks based on memristor simulating synaptic structures have gradually become hot spots.In a number of memristor devices,ferroelectric synaptic transistors can adjust the state of ferroelectric domains in the gate dielectric layer by applying electric pulses to the third terminal,and realize the adjustment of channel conductance used to simulate synaptic weights.This work uses ferroelectric synaptic transistors as synapses and utilizes relevant peripheral devices of neurons to design Hopfield neural network structures and constructs the simulation model,achieving the functions of networks solving optimization problems and associative memory.Moreover,based on ferroelectric synaptic transistors,this work further proposes improved associative learning neural networks and constructs the simulation model,overcoming the problems of classical Hopfield neural networks being unable to train in real-time and requiring multiple iterations in associative learning tasks.Specifically,it includes:1.Based on the electrical characteristics of ferroelectric synaptic transistors,Hopfield neural networks based on ferroelectric synaptic transistors are designed,and the simulation model of the networks is constructed.2.The simulation model of Hopfield neural network based on ferroelectric synaptic transistor is used to solve the optimization problems which includes continuous optimization problem such as function minimization problem,combinatorial optimization problems such as max-cut problem,traveling salesman problem.And in solving the max-cut problem,simulated annealing algorithms are combined based on the third terminal regulation characteristics of ferroelectric synapses.3.The simulation model of Hopfield neural network based on ferroelectric synaptic transistor is used to realize associative memory of digital images.The working characteristics of this network for associative memory of single digital images and multiple digital images are explored.4.Based on ferroelectric synaptic transistors,improved associative learning neural networks are further proposed and the simulation model is constructed,which overcomes the problems of classical Hopfield neural networks not being able to train in real-time and requiring multiple iterations in some associative recalling tasks,and realizes associative learning and recall of digital images.