Machine Learning Algorithm Prediction Of Graphene Field Effect Transistor Device Performance Based On TCAD Simulation

In recent years,the scale and performance of traditional silicon-based field-effect transistors gradually approach the limit with the development of integrated circuits,so it is necessary to study new alternative devices.As a new carbon material with two-dimensional planar structure,graphene has attracted extensive attention due to its excellent mechanical,optical,electrical,chemical and biological properties.Graphene field effect transistor(GFET)composed of graphene has been used in biochemical sensors,solar cells,high-speed electronic devices,touch screens and flexible printed circuits.However,due to the zero band gap characteristic of graphene,the switching current of GFET is relatively small,which hinders its development and application in the field of semiconductor devices.Therefore,how to improve the switching current ratio and other characteristics of GFET becomes the most important research.After investigation,it is found that the current research on the performance of GFET devices only focuses on the experimental or theoretical research of one configuration,but there is no comparative research on the performance of different configurations of GFET devices.This is due to the complicated process,expensive equipment,time-consuming,high cost,poor processing consistency and repeatability of GFET devices with different configuration parameters.The difference of this thesis lies in:three different configurations of GFET devices are compared and studied,and the optimal model parameters are established before the actual processing of devices by using the method of combining computer simulation and machine learning prediction,so as to avoid a series of problems in the processing technology of the above devices.Computer simulation of the performance of semiconductor devices has become one of the main ways to predict and optimize the performance of devices.At the same time,the rapid development of machine learning has shown great commercial achievements and research value in the field of semiconductor device optimization.The study proposes a machine learning algorithm prediction method based on Technology Computer Aided Design(TCAD)simulation with the prediction of GFET device performance.In other words,a method combining TCAD simulation with BP neural network is proposed to predict the performance of GFET devices.The main work of this thesis is as follows:Firstly,the electrical characteristics of GFET are simulated by TCAD software,and the effects of four model parameters on the electrical characteristics of top gate,back gate and double gate GFET are systematically studied,including the effects of gate length(L_g),gate dielectric layer thickness(t_ox),channel length(L_c)and channel doping concentration(N_c)on the electrical characteristics of GFET The simulation results,including ON/OFF current ratio(I_on/I_off)and transconductance(g_m),can be carried out.Then,based on a large number of TCAD simulation data,combined with Matlab machine learning method,aiming at the problem of GFET device performance prediction,the feature selection and optimized BP neural network algorithm are used to realize the device performance prediction.By feature selection method,the best feature subset is selected from the four-dimensional feature set to predict the performance of GFET devices.Taking the selected two-dimensional device model features as input,the ON/OFF current ratio(I_on/I_off)and maximum transconductance((g_m)_max)performance as output,the optimal BP neural network prediction model are established through experiments with the training function and the neurons hidden layer.The optimized BP neural network is used to predict the performance of GFET devices,and the MSE for I_on/I_offand(g_m)_maxdevices is very small.At the same time,the prediction time of machine learning algorithm is more than 45.5%less than TCAD simulation time,which greatly shortens the simulation time of GFET device.The research of this thesis is devoted to develop a low-cost and efficient prediction method,and has important value for optimizing the configuration and processing parameters of GFET devices.