The construction of production performance prediction system for semiconductor manufacturing with artificial neural networks

Abstract

The major performance measurements for wafer fabrication system comprise WIP level, throughput and cycle time. These measurements are influenced by various factors, including machine breakdown, operator absence, poor dispatching rules, emergency order and material shortage. Generally, production managers use the WIP level profile of each stage to identify an abnormal situation, and then make corrective actions. However, such a measurement is reactive, not proactive. Proactive actions must effectively predict the future performance, analyze the abnormal situation, and then generate corrective actions to prevent performance from degrading. This work systematically constructs artificial neural network models to predict production performances for a semiconductor manufacturing factory. An application for a local DRAM wafer fabrication has demonstrated the accuracy of neural network models in predicting production performances.