Analog Design and Machine Learning: A Review

Analog and mixed-signal (AMS) integrated circuit (IC) design remains a critical yet challenging aspect within electronic design automation (EDA), primarily due to the inherent complexity, nonlinear behavior, and increasing variability associated with advanced semiconductor technologies. Traditional manual and intuition-driven methodologies for AMS circuit design, which rely heavily on iterative simulation loops and extensive designer experience, face significant limitations concerning efficiency, scalability, and reproducibility. Recently, machine learning (ML) techniques have emerged as powerful tools to address these challenges, offering significant enhancements in modeling, abstraction, optimization, and automation capabilities for AMS circuits. This review systematically examines recent advancements in ML-driven methodologies applied to analog circuit design, specifically focusing on modeling techniques such as Bayesian inference and neural-network-based surrogate models, optimization and sizing strategies, specification-driven predictive design, and AI-assisted design automation for layout generation. Through an extensive survey of existing literature, we analyze the effectiveness, strengths, and limitations of various ML approaches, identifying key trends and gaps within the current research landscape. Finally, the paper outlines potential future research directions aimed at advancing ML integration in analog IC design, emphasizing the need for improved explainability, data availability, methodological rigor, and end-to-end automation.