On the Use of Machine Learning in the Semiconductor Industry: Examples and Case Studies

• Activity Number: 232
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Theresa Utlaut*+ and Kevin Anderson
• Companies: Intel Corporation and Intel Corporation
• Keywords: machine learning ; semiconductor

Abstract

Semiconductor manufacturers must wring all available improvements in the quality and yield of their products. Some questions an engineer or statistician must ask when confronted with optimizing a 400-operation process are: At which operation to begin? What factors in that operation influence what responses in the process? Have all available opportunities for optimization been exhausted? Huge quantities of data are generated on production processed in a semiconductor facility. However, most of these data are observational in nature. Complex structure, multicollinearities, and sparsity are rampant in these data, and compromise the ability of classical statistical methods to answer the engineer's questions. Machine Learning can be defined as using algorithms that improve their performance by the analysis of data rather than relying on the analyst. Popular methods include Binary Recursive Partitioning, Stochastic Gradient Boosting, and Random Forests. This presentation will offer some theory behind these methods, and discuss their application in semiconductor manufacturing using software named IDEAL. Case studies will demonstrate the utility of these approaches in data analysis.