What are the Most Informative Data for Virtual Metrology? A use case on Multi-Stage Processes Fault Prediction

Abstract

In recent years, Data intensive technologies have become widespread in semiconductor manufacturing. In particular, Virtual Metrology (VM) solutions had proliferated for quality, control and sampling optimization purposes. VM solutions provide estimations of costly measures from already available data, allowing cost reduction and increased throughput. While most of the literature in VM is focused on providing the most accurate methodological approach in terms of prediction accuracy, no work has previously investigated which are the most informative data for VM. This is particularly relevant since literature is divided between VM based on Optical Emission Spectroscopy (OES) and Key Parameter Indicators (KPI) data. In this work we provide a comparison of between VM based on OES and KPIs on a real case study related to a multi-stage modeling problem.