Development of smart feature selection for advanced virtual metrology

Advanced Process Control is an important research area in Semiconductor Manufacturing to improve process stability crucial for product quality. Actual research focuses on Virtual Metrology (VM) using high sophisticated Machine Learning methods to predict a predefined metrology target based on statistical models trained on historical data. Enhancing physical metrology by predictive models leads to smart future fabs. Enabling scalable VM for fab-wide deployment depends on the extraction of only the most important out of typically hundreds to thousands of available process parameters. High Density Plasma (HDP) Chemical Vapor Deposition appears to be a process area in SM predestinated for application of thus required Data Mining. In low-volume-high-mixture fabrication plants, Knowledge Discovery in Databases is extremely challenging due to complex technology mixtures and reduced availability of data for comparable process steps. In this paper, a new Feature Selection wrapper method has been developed and is investigated to reveal only the most important process parameters for smart VM. Modeling of unknown functional interrelations and selection of these features is achieved to predict the layer thickness of dielectric layers deposited onto a metallization layer of the manufactured wafers with highest accuracy and reliability.