Virtual Metrology Model Robustness Against Chamber Condition Variation by Using Deep Learning

In these days, deep learning (DL) [1] is attracting a lot of interest due to their highly discriminative representations that have outperformed many state-of-the-art techniques, mainly in the field of computer vision (CV). In this article, we exhibit a new DL method that exploits the semiconductor domain knowledge and extracts the informative features from optical emission spectroscopy (OES) data, that have both wavelengths and time spatial information. The virtual metrology (VM) is the target functionality, as it is difficult to get a robust model by conventional methods. As a comparison, two other well-known DL methods were also evaluated. The evaluation was executed on a real industrial dataset related to the etch process, which is one of the most important semiconductor manufacturing processes.