Combining Time Series Data and Snapshot Data for Situation Aware Dispatching in Semiconductor Manufacturing

Dispatch rules are commonly used to schedule lots in the semiconductor industry. Previous studies have indicated that adapting dispatch rules can improve overall factory performance. Machine learning has proven useful in learning the relationship between manufacturing situations and dispatch rules. However, using only snapshot data at a given point in time to generate features for these models does not account for trends in the manufacturing situation, which can be represented as time series data. To address this issue, the proposed method generates features from time series data and combines them with features from snapshot data to train machine learning models for dispatch rule prediction. The results demonstrate the effectiveness of this methodology, as the combination of features from both types of data achieves the highest prediction accuracy. Simulation results show that this approach can adapt the dispatch rule according to the manufacturing situation and achieve a comparable factory performance.