Development of a Predictive and Preventive Maintenance Demonstration System for a Semiconductor Etching Tool

Abstract

Increasing yield, eliminating unexpected downtime, and reducing maintenance costs are some of the potential benefits that a predictive monitoring system can provide for semiconductor manufacturing. Prior to the implementation of a predictive based condition monitoring approach for critical semiconductor equipment and components, the technology requires further development and validation. A systematic approach for developing a predictive monitoring system is presented, including several key steps that include data pre-processing, feature extraction and selection, health assessment, and component lifetime prediction. The approach is demonstrated on a LAM 2300® Kiyo® conductor etch product, in which data was collected over an 8-month period from a GlobalFoundries Fab in Dresden, Germany. The methodology and predictive health models are demonstrated for an example component from the etching tool, the electrostatic chuck. Both trace signals and metrology data are evaluated for developing the health and prediction models for the electrostatic chuck. The health monitoring models provide a clear and increasing trend, in which the electrostatic health value can be tracked quite well over time. The systematic development framework shows promise and future works looks to further validate the prediction models, and in addition develop commercially viable implementation software from this work.