ML-TCAD: Perspectives and Challenges on Accelerating Transistor Modeling using ML

Technology CAD (TCAD) demonstrates great capabilities in solving complex problems and remains an essential tool for transistor modeling. Despite its high accuracy, realized through convoluted physics-based simulations, TCAD calculations are exceedingly slow and require a high-cost license. Computational speed limitation is the central bottleneck for large-scale design-space exploration. Thus, it is crucial to accelerate transistor simulations and potentially eliminate TCAD from the loop. Machine learning (ML) has proven to be extremely efficient in complex data processing, which enables ML-based transistor simulation. In this paper, we explore the unique opportunities provided by ML-based modeling, and its associated challenges.