Particle Counts Mislead Scientists: A JMP® and SAS® Integration Case Study Investigating Modeling
Kathleen Kiernan, MS, Technical Support Statistician; Diane Michelson, PhD, Statistical Trainer; Annie Zangi, MS, Research Statistician – SAS
This paper will address consequences of modeling data in real-world applications and will demonstrate methods and techniques for modeling non-normal data using the integrated capabilities of both JMP and SAS. Most commonly used statistical tests and modeling techniques are based on certain assumptions. What if the data we are analyzing is non-normal or heteroscedastic? JMP offers some methods and techniques for modeling non-normal or heteroscedastic data, while SAS offers a full range of procedures that can be used. The Box-Cox power transformation is often used to transform non-normal or heteroscedastic data. Another approach is to model a non-normal distribution. This paper will present two semiconductor case studies. During semiconductor processing, wafers are kept as clean as possible. Particles in the active area of a circuit can ultimately cause a defective chip. The first example explores a case where scientists tried to reduce particle counts occurring on wafers and found the model less effective at identifying the source of the problem than originally thought. Semiconductor chips are fabricated on large wafers, each of which can contain hundreds of chips, and wafers are processed in batches. In the second example, we will investigate misleading results that can occur when trying to find the important factors in the variance components of the process in a traditional linear models setting. We will present solutions for both of these examples using JMP and SAS integration, which can be applied to other scenarios as well.