Prediction of Incoming Work-In-Progress Prediction in Semiconductor Fabrication Foundry using Long Short-Term Memory Recurrent Neural Network
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UNIVERSITI MALAYSIA SARAWAK
Abstract
Description
Semiconductor process is one of the most complex processes in the manufacturing industries. In order to meet the high demands of semiconductor components, cycle time reduction in semiconductor manufacturing is crucial. One of the factors that can increase cycle time is the equipment preventive maintenance activity. Preventive maintenance activity is crucial to maintain or increase the operational efficiency and reliability of the equipment and minimises unanticipated failures due to faulty parts. Despite its criticality, preventive maintenance activity requires significantly long hours, thereby increasing cycle time. Therefore, this activity is usually performed when the equipment group is forecasted to have low incoming Work-in-Progress. However, current statistical forecasting approach has low accuracy because it lacks the ability to capture the time-dependent behaviour of the Work-in-Progress. In this work, we present an approach to Work-in-Progress arrival forecasting by using LSTM neural network. The LSTM forecasting model is trained using the historical incoming Work-in-Progress of an equipment group. The LSTM forecasting model is then required to forecast the incoming Work-in-Progress of three consecutive weeks to the same equipment group. The results of this study are compared with the results of the statistical method practised by the manufacturing production control engineers in the sponsoring company of this research. The comparison showed that the result proposed in this study is superior to the statistical method in the sponsoring company. The proposed model achieved value accuracy of 100%, 100%, 85.7% and trend accuracy of 66.7%, 100%, 50%, with respect to the three consecutive weeks of comparison.