An iterative outlier screening procedure for reliability data using mixed effect modeling

We present an outlier screening procedure that has been used to remove outliers in NBTI (Negative Bias Temperature Instability) data, using the statistical approach of Mixed Effect Modeling. Our method has proven very useful with its inherent values on two aspects. First, it removes outliers that are considered to be deviating from the regression line at individual stress conditions. Second, it may avoid pitfalls in the traditional univariate method where valid data-points are incorrectly detected as outliers, especially with very few samples per stress condition.