Cycle time prediction in wafer fabrication line by applying data mining methods

Abstract

Wafer fabrication is considered the most complex and costly challenge in the semiconductors industry. Cycle Time (CT), which denotes flow time, is one of its key performance measures. This work develops CT prediction models by applying Machine Learning (ML) and Data Mining (DM) methods. The models can assist in improving manufacturing and supply chain efficiency. They rely on historical production line data taken from the fab's Manufacturing Execution System (MES), and include wafer lot processing details of various operations. The prediction is done for an average CT of a single lot, processed through a single operation step. Two types of classification techniques are used. The best fitted Decision Trees (DT) model achieves 76.5% accuracy, and the best Neural Network (NN) model (two hidden layers) achieves 87.6% accuracy. The significance of this study is in establishing dynamic CT prediction models, which can be used to predict CT of a single operation step, a line segment or a complete production line.