Leveraging Machine Learning for Capacity and Cost on a Complex Toolset: A Case Study

In this case study, we introduce two ML techniques, Long Short-Term Memory (LSTM) and an optimized Random Forest (RF), to address challenges related to capacity and cost, by addressing problems of unscheduled downtime and Process Time (PT) variation in the case of a complex chamber processing tool. We show that by using these ML techniques, traditional methods of Predictive Maintenance (PdM) and PT analysis can be enhanced with new insights and lead to significant productivity improvements. We demonstrate that, with these methods, by detecting states and attributes of the tool, trends in the tool's behavior can be more effectively identified to reduce its unscheduled downtime and improve its run-rate, thereby resulting in significant capacity and cost improvements. This is achieved by reducing the variability of availability; extending the Mean Time Between Failures (MTBF); and removing variability in PT between lots and chambers.