時間相依投料環境下晶圓製造廠之短期產出預測

Abstract

短期產出預測（Short-Term Output Prediction）在晶圓製造廠中扮演重要角色。給定生產線上的即時在製品分佈資訊與未來的投料計劃，便能以準確的短期產出預測幫助管理人員監控生產進度，並提供未來投料計劃與現場排程派工上的參考。然而，晶圓製造是一個高複雜度的生產系統，其包含了迴流及當機等特性，且所面對的產品組合（Product Mix）在市場需求下亦會不斷地調整，造成了一個非穩態（Non-Steady State）的晶圓製造生產環境，其中各項生產績效指標將隨時間而變化，因此傳統長期平均績效方法無法有效描述此一變化。本研究將發展一套以車流理論為基礎的晶圓動態產出預測模型，能夠處理隨時間變動的生產環境。透過能描述相異類別工件具有交互作用的速率函數，可構成一個單站多種工件類別的動態產出預測模型，有能力處理產品組合隨時間變化的情形。此模型亦可彈性擴充至多種型式的生產線，例如具有重工及迴流的晶圓生產線。經模擬案例分析結果顯示，此模型在短期產出績效的預測上有良好的效果。

Short-term output prediction plays an important role in wafer fabrication operations. Given the initial WIP profile and future release schedule, an accurate output prediction could help production managers to monitor production status and provide useful information in making decisions. Wafer fabrication, however, is a very complicated production system. Some special characteristics, such as reentrant, unpredictable machine breakdown, and the product mix changing from time to time, lead it to be a production environment with time-dependent performance. Therefore, traditional approaches considering long-term average performance fail to give a satisfactory output prediction. This study proposes a traffic-flow-based model to predict outputs dynamically, especially with time-variant inputs. A single work station model, which is able to handle multiple products, is proposed. This model is based on a multi-class velocity function capable of dealing the interaction of different products and product mix changes. The single-station multi-class model can be flexibly and easily extended to different production systems with multiple stations, reentrant and/or rework. Simulation studies demonstrate that the proposed model provides good predicting results, especially for high-volume production systems.