Realistic Stripline Corner Modeling Using Surrogate Model and Topographic Fitting

This paper demonstrates a method to extract impedance-attenuation corners of a stripline with user-prescribed confidence levels. This is done using a sparse-grid-based surrogate model to quickly generate vast Monte Carlo datasets from which the impedance-attenuation distribution is calculated. Ellipses are fit to this distribution as equi-density contours to enclose a proportion of the solution data. Appropriate corners can be read off these ellipses and applied to broadband simulation. The results are compared against three measured test coupons, showing capability to analyze a PCIe Gen. 5 link. Realistic modeling of geometries and material variations is emphasized.