Bayesian Inference Technique for Data Mining for Yield Enhancement in Semiconductor Manufacturing Data

Abstract

The yield management in semiconductor manufacturing is one of interesting area that data mining approaches find useful applications. The abundant steps and complex workflows during wafer manufacturing automatically generate large volumes of data and, hence, engineers who rely on personal domain knowledge cannot find possible root cases of defects quickly and effectively. The complexities involved in semiconductor manufacturing have always delayed the dream of creating a reliable process to produce 100% yield. Although the manufacturing recipes are carefully designed and revised to maximize yield, yield is still affected by errors that are reported by systematic factors (e.g., defective tools or interactions between tools) or random factors (e.g., dust particles).Furthermore, experiments have shown that most insidious and dangerous defects come from the interactions between components of a complex system - that cannot be detected by human diagnostic at individual developer level. Although, generally, selecting the process Tool, chamber set and recipe name, eventuate based on a series of previous experience, however these practical intuitions don’t have any seat in computerized process mining for defect detection. This study aims to develop a framework for data mining and knowledge discovery from database that consists of three phases: data preparation, data dimension reduction and the model construction and evaluation based on Bayesian Variable Selection (BVS) to figure the effect of practical intuitions and investigate the huge amount of semiconductor manufacturing data and infer possible causes of faults and manufacturing process variations. The proposed approach has been validated by a simulation and an empirical study, eventually replicated Cross-validation has emerged as the preferred method to estimate the accuracy of proposed approach on a particular data set and the results have shown its practical viability.