Rapid Yield Optimization of Compact Microwave Couplers By Means Of Variable-Fidelity Response Features

摘要

In this paper, a technique for rapid yield optimization of miniaturized microwave couplers is presented. Our approach utilizes local surrogates constructed based on a response feature paradigm and variable-fidelity EM simulations of the structure under design. Upon appropriate correction using a single high-fidelity EM analysis results, the feature-based model obtained from low-fidelity simulations is utilized as a prediction tool to carry out sequential Monte Carlo analyses and generate an improved-yield design. The process is iterated with the surrogate reset after each iteration. The proposed procedure permits expedited yield optimization at the overall computational cost of around a dozen of high-fidelity analyses as demonstrated using a compact rat-race coupler operating at 1 GHz. The results obtained with variable-fidelity feature-based surrogates are validated through direct Monte Carlo analysis involving high-fidelity EM coupler model.