Session

DescriptionThis session presents a shift from manual layout and iterative tuning toward autonomous design agents and differentiable optimization for analog and RF systems. Featured works include LLM-driven PCB schematic generation, Monte Carlo tree search for analog placement, and diffusion-based priors for transistor sizing. By leveraging physics-constrained inverse design and stochastic knowledge elicitation, these approaches address the challenges due to advanced nodes. Collectively, they pave the way for a self-evolving design flow, significantly reducing the productivity gap in complex circuit synthesis while maintaining high-fidelity performance constraints.