Autoencoder Enables Rapid FinFET Device Modeling

The development of advanced semiconductor devices like FinFETs requires accurate and efficient modeling to facilitate design and simulation. Traditional methods can be time-consuming and complex. This research introduces a machine learning framework that significantly accelerates FinFET modeling by leveraging an autoencoder. The process begins by calibrating a BSIM-CMG model to generate a comprehensive dataset of current-voltage (ID-VG) characteristics. This rich dataset is then used to train an autoencoder, which is adept at compressing high-dimensional data into a much smaller, latent space. A key innovation in this approach is the explicit inclusion of parameters like drain-to-source voltage (VDS) as an input feature, which enhances the model's ability to capture bias-dependent variations in device behavior. Once trained, the autoencoder can effectively reconstruct full I-V curves from its compressed representation. More importantly, it can directly extract crucial device metrics such as threshold voltage (VTH), subthreshold slope (SS), and peak transconductance (gm) with high accuracy. This data-driven compact modeling approach, built from actual characterization data, demonstrates that machine learning can provide a powerful and rapid tool for device characterization, modeling, and subsequent circuit-level simulations, even with minimal training data.