End-to-End AI and Transfer Learning-Based Power Amplifier Synthesis Considering Layout and PVT Effects

Designing robust mm-wave circuits has become increasingly challenging as modern applications demand high performance across wide operating conditions. Recent AI-driven methodologies have accelerated mm-wave circuit design, however, most still target a single design point or rely on costly multi-corner simulation, limiting their efficiency and usefulness for system-level co-design. This work presents an end-to-end AI-driven synthesis methodology that unifies nominal optimization, data-efficient process, voltage, and temperature (PVT) modeling, and layout-aware verification within a single methodology. The framework employs machine learning-based surrogate models for inductors and transformers to avoid full-wave electromagnetic simulations, a transfer-learning strategy that reduces PVT simulation cost, and a multi-objective evolutionary algorithm that generates layout-verified Pareto-optimal fronts capturing key circuit performance trade-offs. The approach is validated through the synthesis and measurement of a 30-GHz power amplifier in 65-nm CMOS. The fabricated design exhibits competitive performances, demonstrating the effectiveness of the proposed AI-driven optimization methodology.