A single neural network global I-V and C-V parameter extractor for BSIM-CMG compact model

Abstract

A global I-V and C-V BSIM-CMG parameter extraction methodology based on deep learning is proposed. 100 k training datasets were generated through Monte Carlo simulation varying 28 IV and CV model parameters in the industry-standard BSIM-CMG FinFET model. For each of the 100 k Monte Carlo-selected BSIM-CMG parameter dataset, the I_D-V_G and C_GG-V_G characteristics of seven Monte Carlo-selected gate lengths ranging from 14 nm to 110 nm were generated as the input to train the parameter extraction neural network. The neural network outputs for training are the 28 model parameters’ values. The neural network's capability to extract BSIM-CMG model parameters that accurately fit TCAD-generated I_D-V_G and C_GG-V_G data over a range of gate lengths was demonstrated. This marks the first time a deep learning compact model parameter extraction flow, employing a single neural network for both I-V and C-V parameters and for a range of gate length, is presented.