The study on critical dimension target prediction for etch process : IE: Industrial Engineering

Foundry FAB manufactures a variety of semiconductor products with adverse mixture of process flows. For process control, it is a challenge to define some critical dimension measurement items target of each unique product characteristic before new tape out (NTO). After some wafers pilot run completed, a proper measurement target was defined. The engineers have to waste valuable resources for measurement retargeting and expense extra process time to validate these changes satisfied.As we know, the mask layout pattern density (PD) has highly correlated with measurement target. However, traditional regression model results can not satisfy advanced processes requirements. In this study, we proposed new factors, including local pattern density, line density and traditional global pattern density (GPD) into model. Furthermore, the regression model was refined with machine learning, k-NN (k-th Near Neighbor), to enhance the prediction accuracy for NTO measurement target. The simulation result showed average prediction accuracy come up to 85% above, compared with previous 61% only. Even some layers accuracy achieved 95% above.