Fully Automated Schematic and Layout Design of Millimeter-Wave CMOS Power Amplifiers Using Prior Knowledge-Guided Decomposition and Machine Learning-Assisted Optimization

Authors:

Abstract
The RF CMOS technology is preferably used in the development of mass-market millimeter-wave (mmWave) integrated circuits because of its cost-effectiveness and integration capabilities. However, conventional design methods limited by fixed topologies struggle to achieve optimal performance, power, and area (PPA) trade-offs. A fully automated method is proposed for rapid mmWave CMOS power amplifier (PA) design that uses prior knowledge-guided decomposition and machine learningassisted optimization (MLAO). Following topology determination through schematic simulation, the proposed approach decomposes the entire circuit into multiple optimizable subproblems using microwave network theory. Then these subproblems are solved sequentially and iteratively. In particular, a multibranch MLAO algorithm is applied to on-chip transformers to enable broadband matching and high-fidelity impedance synthesis. Postlayout simulation results for three different design cases spanning 17 to 41 GHz demonstrate that the proposed method significantly accelerates the design process compared to conventional approaches while maintaining a high PPA.