Neural-network-assisted EKV compact modeling for complementary oxide thin-film transistors

This work reports a neural-network-assisted Enz–Krummenacher–Vittoz (EKV) compact-modeling framework for complementary oxide thin-film transistors (TFTs). The neural network is used as a bias-dependent effective-parameter generator for the equivalent mobility and onset voltage, while the drain current is calculated by the analytical EKV current core. This strategy represents defect-, contact-, and bias-dependent nonidealities of oxide TFTs through effective quantities without replacing the current equation with a black-box neural-network predictor. The model is validated using both n-type indium–gallium–zinc oxide TFTs and p-type tin monoxide (SnO) TFTs, achieving mean absolute percentage errors of 0.56% and 0.36%, respectively. The learned effective parameters further provide an EKV-constrained compact representation of bias-dependent transport capability and channel-onset behavior obtained from global fitting of the measured output-characteristic dataset, rather than serving as directly extracted material parameters. The trained model is translated into a PSpice-compatible library and applied to complementary inverter simulation, showing good agreement with measured voltage-transfer characteristics and reproducing the main transient response characteristics. These results demonstrate an accurate, analytically structured, and Simulation Program with Integrated Circuit Emphasis-compatible compact-modeling strategy that links bias-dependent oxide-TFT transport to circuit-level simulation.