Efficient yield optimization of a 100 GHz filter using the novel combination of a performance guided random walk and NLPLS-based PCE

Yield optimization can be a crucial step in highvolume passive microwave component design, however it is typically a very expensive process, as a multitude of full-wave electromagnetic analyses are required for statistical analysis at each step of the optimization process. Metamodeling techniques have been shown to be useful solutions to this problem. In this paper yield optimization of a 43-variable passive 100 GHz ridged waveguide filter is done using the novel combination of a performance-guided random walk and Non-Linear Partial Least Squares based Polynomial Chaos Expansion (NLPLS-based PCE). It is shown that a substantial improvement of the yield is achieved with reduced model evaluations using this method.