臺灣博碩士論文加值系統

在半導體製造系統中，主要的績效衡量指標為在製品量、流動量、生產週
期，許多不同的因素如當機、操作不當、投料與派工法則不佳、警急插單
、物料短缺等皆會對上述的績效造成影響。生產管理者往往利用在製品量
的分配情形，來辨別系統的異常狀況，進而採取必要的改正措施；然而，
如此方式並無法於事前防範異常的產生，可謂被動形式；主動形式應能預
測未來績效的可能變化，找出績效可能下降的原因，並採取對應措施，防
範異常現象於事前。因此，本研究的目的便在於利用類神經網路為工具，
構建生產績效的預測模式，並以某DRAM廠之資料作為驗證，以確認本模式
的實際預測成效。

The major performance measurements for any wafer fab manufactur-
ing system comprise of WIP level, Move volume and cycle time.
Different factors including machine breakdown, improper
operation, poor releasing and dispatching rules, emergency
order, and materials shortage, influence such measurements.
Production managers use the WIP level profile of each stage to
identify an abnormal situation, making necessary corrective
actions. However, such a measure is a reactive action not a
proactive one. A proactive action must predict the future
performance, identify the abnormal situation, understand why it
occurs and generate corrective actions to prevent a decrease in
abnormal performance.Therefore, this work presents a production
performance prediction model using artificial neural networks.
An illustrative example in which data are collected from a local
DRAM wafer fab demonstrates the accuracy of neural network
models in predicting wafer fab performance.