Neural Net Based Prognostics for an Industrial Semiconductor Fabrication System

Modern semiconductor fabrication machinery has the capability to generate huge volumes of health data, well beyond the capability of the typical process engineer to discern the subtle clues it contains. This data holds the key to performing periodic maintenance on an as-needed basis, rather than on a schedule. Soft computing techniques such as neural networks can allow the engineer to use this data to detect the need for maintenance. This paper discusses useful tools to accomplish the above goal and describes the results of proof-of-principle experiments, which prior to the receipt of actual data, which will eventually lead to prototype testing on the actual semiconductor fabrication systems