Data mining for improving a cleaning process in the semiconductorindustry

摘要

As device geometry continues to shrink, micro-contaminants have annincreasingly negative impact on yield. By diminishing the contaminationnproblem, semiconductor manufacturers will significantly improve wafernyield. This paper presents a comprehensive and successful application ofndata mining methodologies to the refinement of a new dry cleaningntechnology that utilizes a laser beam for the removal ofnmicro-contaminants. Experiments with three classification-based datanmining methods (decision tree induction, neural networks, and compositenclassifiers) have been conducted. The composite classifier architecturenhas been shown to yield higher accuracy than the accuracy of eachnindividual classifier on its own. The paper suggests that data miningnmethodologies may be particularly useful when data is scarce, and thenvarious physical and chemical parameters that affect the process exhibitnhighly complex interactions. Another implication is that on-linenmonitoring of the cleaning process using data mining may be highlyneffective