Virtual Metrology of WAT Value With Machine Learning Based Method

Abstract

Plasma etch is a key process during semiconductor manufacturing, materials are removed from the surface of semiconducting, typically silicon. The etch process is highly complicated because of chemical and electrical interactions, which makes the etch process notoriously difficult to model and control. Traditional metrology depends on expensive equipments and exhausts a lot of time, which has become new bottleneck of high-volume and high-throughput manufacturing at advanced nodes. In this work, we focus on the virtual metrology technique which could be used for prediction of WAT value based on the key parameters in etch process. By using random forest (RF) and stepwise regression (SR) method, the predicted values of NRV_Idsat reached a high R2 compared with true measurements. It reveals high efficiency of the proposed method and great potentials in the applications of virtual metrology in real manufacturing.