Session

DescriptionGenerative AI is reshaping the landscape of analog design, a domain long constrained by manual expertise, limited automation, and complex layout-driven trade-offs. With the advent of generative AI models, the potential for automatic analog design has grown exponentially. These AI systems, acting as "design agents," have begun to demonstrate remarkable capabilities, offering the promise of automating analog design tasks (such as topology synthesis, sizing, and layout generation) that were once considered too intricate or too dependent on human intuition to be effectively automated. However, the shift from agent to true "design expert"—a system that can generate, optimize, and innovate in analog design—requires a deeper integration of domain knowledge and reasoning. This demands rethinking AI training, moving beyond simple generation to advanced learning methods that capture the underlying physical principles and trade-offs of analog design. This transformation is not just a technical evolution; it represents a shift in analog design, where AI moves from assisting with tasks like topology synthesis and layout generation to actively driving the design process, optimizing complex trade-offs and enabling fully automated workflows that were previously unattainable.

This special session features five talks that will highlight key advancements in AI-driven analog design automation, exploring how generative AI could evolve from assisting to potentially taking a more central role in the design process. The first talk reviews modern AI/ML methodologies for AMS/RF design and how they complement traditional expertise. The second introduces an LLM- and RL-based agentic workflow enabling automated analog/RF synthesis from spec to layout. The third presents an AI-native foundation-model framework for topology discovery, performance prediction, and intent-to-silicon flows. The fourth talk highlights hierarchical RL with LLM guidance for generating valid, high-quality analog circuit topologies. The final talk showcases an LLM-enhanced framework that automates and optimizes the analog design process, bridging design intent with layout generation.