<p>This article systematically reviews the cutting-edge applications, key challenges, and future directions of machine learning (ML) in metal–oxide–semiconductor field-effect transistor (MOSFET) technology. As device dimensions shrink to the nanoscale, traditional design methods face challenges such as short-channel effects, process variations, and high modeling complexity. Leveraging its powerful data-driven modeling and high-dimensional nonlinear mapping capabilities, machine learning offers a new paradigm for MOSFET performance prediction, inverse design, compact modeling, process variation analysis, and reliability assessment. The article provides a detailed analysis of specific application cases and methodological evolution of ML across several key areas, highlighting its significant advantages in improving design efficiency and optimizing device performance. Simultaneously, this review identifies core bottlenecks in current ML applications, including data dependency, model interpretability, generalization capability, and physical consistency. It also outlines future research directions, such as few-shot learning, physics-informed integration, cross-platform universal modeling, and intelligent electronic design automation integration, to promote the deep integration and collaborative innovation of machine learning and semiconductor technology.</p>