Skip to Main Content
In semiconductor manufacturing, in order to guarantee an optimal production flow it is necessary to perform a quick and correct equipment repair when an error message occurs. Since most equipment types are very complex, maintenance engineers are provided with manuals of troubleshooting flow charts. These manuals offer guidelines for finding the cause of the problem. Since such manuals are often static, clumsy and difficult to extend, it might be hard for maintenance engineers to efficiently perform cause-effect testing. For this reason, we employed a Bayesian network model that is developed from troubleshooting flow charts, which is able to overcome these deficiencies. The network is built as a self-learning diagnostic system. Troubleshooting sessions are used to train the network, so that the order of potential root causes is dynamically updated by actual maintenance experience. An Expectation Maximization (EM) algorithm is used to update the network. Furthermore, by ordering symptoms according to a mutual information criterion, it is possible to provide maintenance engineers with a ranking of the most informative and efficient tests to run.