Deep Learning Framework for Quantum-Aware Modeling and Inverse Parameter Extraction in CNTFETs

Abstract

This paper presents a machine learning (ML) framework for the modeling and inverse parameter extraction of carbon nanotube field-effect transistors (CNTFETs) with channel lengths below 100 nm. A high-fidelity dataset is generated using the non-equilibrium Green’s function (NEGF) formalism combined with acoustic phonon (AP) scattering to capture critical quantum transport effects. For forward modeling, a deep operator network (DeepONet) is employed to predict current-voltage (I-V) characteristics from structural and bias parameters, achieving significant improvements in inference efficiency while maintaining high accuracy. Furthermore, the DeepONet enables efficient multi-objective optimization, rapidly identifying optimal solutions that balance conflicting performance metrics. For inverse parameter extraction, a two-stage model is developed based on principal component analysis (PCA) and a multilayer perceptron (MLP), enabling the recovery of key device parameters from high-dimensional I-V data. A tandem model with a hybrid loss function combining inverse and forward consistency is leveraged to alleviate the non-uniqueness problem. The proposed framework enables rapid, quantum-aware device simulation and direct inverse design, providing a scalable solution for CNTFET development, design optimization, and process monitoring.