Methods for assessing empirical model parameters and calibration pattern measurements

Assessing an empirical model for ILT or OPC on a full-chip scale is a non-trivial task because the model's fit to calibration input data must be balanced against its robust prediction on wafer prints. When a model does not fit the calibration measurements well, we face the difficult choice between readjusting model parameters and re-measuring wafer CDs of calibration patterns. On the other hand, when a model does fit very well, we will still likely have the nagging suspicion that an overfitting might have occurred. Here we define a few objective and quantitative methods for model assessment. Both theoretical foundation and practical use are presented.