Development and Validation of Dynamic Data-Driven Device Compact Model

In this paper, we explore the potential of Machine Learning (ML) techniques to automate and expedite compact model development. We present a dynamic data-driven compact modeling approach for diodes and transistors, which involves regressing maps of DC behavior (voltage to current) and small-signal AC characteristics (voltage to capacitance). These mappings can be easily obtained using conventional laboratory setups. The validation results demonstrate that circuit simulations utilizing the dynamic data-driven model effectively capture frequency-dependent behavior, a significant improvement over the previously developed static data-driven model.