Improving Accuracy of Rational Macromodels under Realistic Loading Conditions

Data-driven rational fitting algorithms are the methods of choice for generating behavioral circuit models for complex multiport components. These approaches generate the circuit model by minimizing the deviation of one of its network functions (e.g., its scattering matrix) from the reference response. In this paper, we show that this commonly-employed fitting condition, even when met with high accuracy, is not sufficient to guarantee the reliability of the macromodel when it is used as building block in larger electrical interconnections. We address this issue by suitably modifying the cost function that drives the rational fitting process, and we outline how to modify the Vector Fitting (VF) iteration accordingly. The effectiveness of the resulting scheme is tested over a relevant Power Integrity application.