Role of correlation in systematic variation modeling

Systematic process variation is typically modeled using process corners where parameter range is set at the time of corner generation. Recent efforts to reduce design guard bands has led to considerable interest in more accurate random variable (RV) based models of systematic process variation, typically known as Universal Variation Model (UVM) or Die to Die (D2D) model. In this work, we show the important, and so far unrecognized, role that the fine structure of parameter correlations plays in accurate modeling of electrical measurement (E-Test) distributions. We also provide an intuitive reasoning to help understand why simple correlation coefficient is insufficient. Finally, we develop a statistical model, compatible with simulation tools, that accurately captures the fine structure of parameter correlations.