Model Nanotransistors with a Hybrid Machine-Learning and Physics-Based Method

We present a hybrid modeling approach that synergistically integrates machine learning (ML) with semiconductor device physics to simulate nanotransistors. The model combines a physics-based ballistic transistor model with an ML component that predicts ballisticity, allowing flexibility in interfacing with device data. Incorporating device physics enhances the interpretability of the ML model and simplifies its training process, reducing the need for extensive training data. The model's effectiveness is demonstrated on silicon nanotransistors, achieving high accuracy with a streamlined ML component. We investigate the impact of different ML models, including Multilayer Perceptron (MLP), Recurrent Neural Network (RNN), and RandomForestRegressor (RFR), on predictive accuracy and training data requirements. The hybrid models maintain high accuracy even with a small training dataset, with the RNN-based model showing superior performance compared to the MLP and RFR models. The trained hybrid model offers significant speedup over device simulations and can be used to predict circuit characteristics based on the modeled nanotransistors. This hybrid approach leveraging ML and physics paves the way for efficient and accurate modeling of quasi-ballistic nanotransistors.