A multi-stage neural network I-V and C-V BSIM-CMG model global parameter extractor for advanced GAAFET technologies

Abstract

A I-V and C-V parameter extraction methodology with various gate lengths utilizing a multi-stage neural network is proposed. This multi-stage neural network contains four networks focusing on extracting parameters from four different regions in transistor’s characteristics, enabling a machine to emulate human’s parameter extraction strategy. This methodology begins with the generation of a training dataset through Monte Carlo simulation, varying 53 selected IV and CV BSIM-CMG model parameters. With each Monte Carlo-selected parameter set, the I-V, transconductance, output conductance and C-V characteristics of seven different GAAFETs with different gate lengths ranging from 9 nm to 389 nm are generated. This multi-stage neural network is trained with the GAAFETs’ characteristics as the input and the 53 model parameters as the output. After training, TCAD-generated GAAFET I-V, conductance and C-V data with various gate lengths are used to test this neural network parameter extractor’s ability of extracting BSIM-CMG model parameters that generate data accurately fitting the TCAD IV and CV data. It is demonstrated that this parameter extraction neural network can extract BSIM-CMG model parameters’ value for GAAFETs within few seconds.