Answers are grounded in this paper and cite their sources.
Design variables in fixed-topology RF circuits are coupled, and multiple performance constraints and limited simulation budgets increase the difficulty of optimization. To examine how circuit information can inform decisions about variable combinations, this study proposes a stage-wise collaborative optimization framework guided by a large language model (LLM). The framework combines circuit information, parameter descriptions, and simulation feedback to select active variable subsets. A Gaussian process with expected improvement (GP/EI) searches parameter values, while Advanced Design System (ADS) evaluates complete circuit designs. A parameter registry and circuit adapters provide unified handling of geometric dimensions and lumped component values, allowing different circuits and models to reuse the same optimization workflow. Comparisons on two-stage common-source and cascode low-noise amplifiers (LNAs) show that, under the same simulation budget, all LLM configurations produce more feasible candidate designs on average than full-parameter search. They also achieve higher mean final best feasible gain in the two-stage common-source LNA. Compared with stage-wise random subset selection, none of the LLM configurations achieved a higher mean final best feasible gain; however, some yielded feasible designs with lower mean in-band gain ripple and larger mean output matching margins. These results indicate that, for the circuits and budgets tested, the benefits of LLM-guided variable selection depend on the performance metric. By translating circuit analysis into executable and comparable variable-selection decisions, this study provides experimental evidence and a reusable implementation framework for exploring how circuit information and numerical optimization can work together.
No takes yet. Share an insight, caveat, or question.
Niu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: