AI-enabled Inverse Design of Harmonic Terminated mmWave PAs and Frequency Doublers

Harmonic termination is fundamental to high-efficiency large-signal RF and mmWave circuits, enabling optimal operation of power amplifiers (PAs) and frequency multipliers. Achieving these terminations typically requires complex multi-order passive networks and combiners . Conventional design relies on heuristic templates, extensive full-wave simulations, and iterative manual tuning—an approach that becomes impractical when suitable templates do not exist. This paper introduces an AI-enabled inverse-design methodology for synthesizing multi-frequency networks directly from desired S-parameter responses, eliminating reliance on predefined structures. We demonstrate the approach through two mmWave examples in 90-nm SiGe BiCMOS: (i) a PA with engineered second-harmonic termination delivering 18.6 dBm output power and 32\% peak efficiency at 43 GHz, and (ii) a frequency doubler achieving 11.6 dBm output power and 16\% collector efficiency at 80 GHz. By removing any templates and enabling direct synthesis of broadband and harmonic-engineered networks, this framework opens new design spaces for next-generation large-signal RF/mmWave systems.