Afternoon Tutorial

"Multivariate Analysis for Microelectronics Manufacturing: the Right Tools for the Job"

 
Sunday, October 5 • 1PM-5PM
Instructors: Dr. Craig Garvin and Courtney Fowler, INFICON

As multivariate analysis techniques become an accepted component of microelectronic manufacturing FDC systems, the core question moves from whether to use these techniques to which one to use.  In this tutorial we present both data-centric and goal-oriented approaches to choosing analysis techniques.  The data-centric approach focuses on available process outcome information. Depending on the level of outcome information available: none, qualitative, or quantitative, different techniques can be used. Multivariate analysis is typically motivated by the goals of accurately modeling key outcomes and identifying root cause for any deviation from normal behavior. We will also show how these goals cannot always be met simultaneously.  Depending on whether root cause analysis or model fidelity is the more important, different analysis techniques are appropriate.

The tutorial will consist of four sections:

• First, data framing and feature extraction will be covered.  In almost all cases, a step of transforming the time series data obtained into a small set of well-chosen variables is required before any analysis is used.  The degree to which this step is performed correctly has a major impact on analysis outcome, independent on chosen analysis technique.  Additionally, feature extraction in the semiconductor environment presents unique challenges that are not well addressed in the general statistical literature. 
• Second, a lightly mathematical overview of the basic classes of multivariate analysis will be presented.  We will discuss supervised and unsupervised learning techniques, and further breakdown supervised techniques between classification and regression. 
• The third section will address relative merits of various techniques, and answer the question what is the right tool for this job. 
• Finally, we will address the larger question of data content and the limits of machine learning.    

Real world examples of both successfully completed projects and works in progress will be included.

About the Instructors

Dr. Craig Garvin joined INFICON in 2003 where he is currently Senior Modeling and Controls Scientist. Dr. Garvin holds a Ph.D. in electrical engineering from the University of Michigan, where he was a member of the EMACS (electronics manufacturing and control systems) research groups, whose many alumni now populate the semiconductor industry. His research focused on novel sensors and data analysis techniques for the semiconductor industry. Prior to joining INFICON, he served as algorithm development lead for KLA-Tencor’s CATALYST run-to-run control framework.

Courtney Fowler joined INFICON, Inc. in January of 2008 as a Senior FDC Applications Engineer. She brings over 10 years of semiconductor process experience in the areas of Ion Implantation, Rapid Thermal Process, Diffusion, Thin Films, as well as Fault Detection and Classification.  Ms. Fowler holds a Bachelor of Science degree in Mathematics from The University of Texas and is currently pursuing a Master of Science in Chemical Engineering, emphasis Advanced Process Control.

More information about INFICON can be found at www.inficon.com.

Registration Information

You must be a registered symposium attendee to register for the tutorials. The combined fee for both tutorials $75 if registered on or before Sept. 1 . After that date, a registration fee of $125 will apply. To register for this event, check the corresponding "Tutorial" box on the Symposium Registration form. Please see the AEC/APC Symposium refund/cancellation policy for refund and cancellation deadlines.