Probabilistic Modeling and Machine Learning for Preventative Maintenance Prediction in Semiconductor Manufacturing

Machinery within semiconductor manufacturing facilities exhibits high degrees of integration and complexity, making it susceptible to unexpected failures. These failures pose the risk of disrupting all manufacturing operations, indirectly impacting profit and overall production output. To avoid such problems, predictive models can be used to forecast the next tool state and anticipate tool failure. Using these predictions, preventative maintenance measures can take place to avoid such disruptions to the supply chain. This paper introduces two approaches for the preventative maintenance prediction problem, a probabilistic approach and a machine learning approach. The probabilistic approach leverages a Markov chain to model the probabilities for the set of outcomes given the current state of a tool. These probabilities are calculated by a conditional probability formula with historical data on each tool's past operational states. Additionally, the preventative maintenance prediction model is expanded to incorporate features from the semiconductor manufacturing plant to be used as input into machine learning algorithms. These additional features further refine the predictive capabilities of the preventative maintenance prediction model by improving model robustness.