ML-Based Prediction of Dual-Channel Core Gate Junctionless FET Device Parameters Using XGBoost

Abstract

This study investigates the application of machine learning technique especially the ensemble learning category algorithm, that is, ‘Extreme Gradient Boosting (XGBoost)’ for making predictions of the various characteristics of Dual Channel Core Gate Junctionless Field Effect Transistors (DCCG-JLFET). Using data generated from the Technology Computer Aided Design (TCAD) simulations, the machine learning model is trained to predict the behavior of Dual Channel Core Gate Junctionless Field Effect Transistors based on various physical parameters. The objective of the model is to reveal the relationships and establish relationships among various parameters including drain current (I_DS) and various short channel effects like subthreshold slope (SS), threshold voltage (V_th), ON current (I_ON) and OFF current (I_OFF). Comparative analysis reveals that the ML model achieves an accuracy of 98.7% for current voltage curve prediction. Also, scatter plots reveal MSE of 5.96 × 10⁻⁹ for I_DS, 6.98 × 10⁻⁸ for V_th, 3.24 × 10⁻⁹ for I_ON, 4.85 × 10⁻⁹ for I_OFF, and 9.84 × 10⁻⁸ for SS and RMSE of 7.72 × 10⁻⁵ for I_DS, 2.64 × 10⁻⁴ for V_th, 5.69 × 10⁻⁵ for I_ON, 6.96 × 10⁻⁵ for I_OFF, and 3.14 × 10⁻⁴ for SS and R²-score of 0.91 for I_DS, 0.99 for V_th, 0.96 for I_ON, 0.99 for I_OFF, and 0.97 for SS when compared to TCAD Simulations. This ML approach can be effectively applied in optimizing and designing semiconductor devices.