機台動態分派的滿載生產函數之後設模式

Abstract

供應鏈係由數個節點工廠構成，因此供應鏈的生產管理需要兩個層次的分析工具：各節點的生產行為模式，以及節點間互動行為的鏈結。生產函數(production function)在經濟學是指生產之產出與資本、勞務等輸入因子之間的函數關係，同樣地，供應鏈管理所討論的生產函數是指生產工廠的績效表現與輸入因子間的函數關係，不過供應鏈管理所關切的績效指標有流程時間、產出量、需求滿足率等，輸入的因子有產能、在製品水準、投料率、設備可靠度、派工方法等，因此函數的形式更為複雜。其次，供應鏈管理所面對的生產單元(production unit)是工廠或工場，其複雜度與動態性遠超過工廠管理所面對的機器或工作站。由於輸出指標與輸入因子多元，系統內部有複雜的動態關係，設計供應鏈管理系統的第一個研究需求便是複雜生產系統單元的生產函數的建構方法。 供應鏈各節點如果處於非滿載狀態，其運作互動可以維持在較低的程度，而不致對整體績效有太過負面的影響。然而，如果節點處於滿載(full-load)狀態，又發生重大的生產變異事件，供應鏈所設定的績效目標將不再可預測或控制，這時必須啟動非為常設的援助措施，將系統引導回到原來的穩定狀態。半導體晶圓製造廠由於規模龐大，又有製程良率、設備可靠性、產品需求組合等各方面的不確定性，是非常複雜動態的生產系統，一旦出現滿載狀況必須採用動態派工、機台調派與推延設備維護等措施，以舒解瓶頸的工作負載。本文以晶圓製造廠為生產單元，建構滿載狀態的機台動態分派的生產函數。由於晶圓製造有很多不確定因素，生產狀態的情境對績效有顯著影響，本文採用後設模式(meta-model)建構生產函數。本文所描述、建構的生產函數可作為半導體供應鏈之廠商控管動態事件的基礎分析與設計工具。

A supply chain is consisted of multiple nodes, so supply chain management will require two modeling tools: production behavior for each node and connecting method between nodes. In economics, a production function relates throughput to capital and labor. Similarly, a production function in supply chain management describes the relationship between performance measures and production decisions in a factory. There is a need to construct production functions for complex production units before designing a system for supply chain management. If a node in supply chain encounters significant production variations during a full-load situation, its performance will become unpredictable. Remedial measures must be activated to bring the system back to steady states. Semiconductor foundry is a very complicated production system due to its large scale and the uncertainties in process yield, machines, product demand and product mixes. When dynamic events take place in a full-load situation, new bottlenecks are created and they must be mitigated by using dynamic machine assignment or other means. In this thesis, a semiconductor plant is treated as a production unit and the dynamic machine assignment is used to construct a full-load production function. Because there are lots of uncertainties in wafer fabrication and production performances are dependent on production scenarios, the meta-modeling is taken as an approach in this research work to construct the production function. The production function described in this thesis can be used as a tool to manage dynamic events for a factory in semiconductor supply chain.