Failure analysis and prediction using neural networks in the chip manufacturing process
Emilia Șipoș, Laura-Nicoleta Ivanciu · 2017
This paper describes a complex software instrument, able to perform failure analysis and to provide an accurate prediction of manufacturing failures. The novelty consists of the association between a basic quality management tool (Pareto diagram) and computational intelligence techniques (neural networks). The data is obtained from the electronic chip manufacturing process, and is used as input for both Pareto diagram and neural network. Software instruments which perform failure analysis using basic quality management tools lack the prediction part. On the other hand, neural networks are widely used for prediction. However, a software instrument which provides both analysis and prediction options, for the chip manufacturing process has not yet been developed, to our best knowledge. The Pareto diagram is generated using failure types specific to the chip manufacturing process and updated for every new entry. The prediction is made for Automatic Test System (ATS) failures, using a nonlinear autoregressive neural network, with the total number of products per day as external input. By analyzing the predicted ATS value and the impact that removing a certain type of failure has on the overall number of failures, the company can set priority directions for failure elimination, as part of the Six Sigma strategy.