應用模式預測控制以穩定具變動特性的半導體供應鏈之投料與存貨

Abstract

由於半導體產業之市場變動快速，產業之競爭壓力大。相關企業間的競爭也從傳統公司間的競爭，轉變成供應鏈與供應鏈間的競爭，故穩定且高效率的供應鏈運作模式對產業供應鏈中的相關企業而言，為保持競爭力的重要條件之一。半導體供應鏈屬於多階層、多廠區的供應鏈結構，其中包含: IC設計、晶圓製造、晶圓測試、晶片封裝以及最終測試等階段。由於各階段生產過程需要經過數百道不等的加工步驟。故加工過程中的產品等級與良率分佈將隨著人為因素、機台因素、材料因素與製程因素的影響而有不確定性的變動。不同階段的不確定性變動程度也會隨生產過程之複雜度不同而有所改變，而穩定且有效率的供應鏈管理可以為半導體製造公司或整合元件製造公司每年節省龐大之支出，而如何有效率地整合與管理供應商、製造商、倉儲配送等階段，並保持供應鏈中各階段之投料量、存貨量於穩定之狀態並於對的時間生產足量的產品以面對下游客戶的需求變化便是本研究將探討之主要課題。 本研究以模式預測控制(Model Predictive Control)建構具有變動特性之半導體供應鏈模式。模式預測控制具有容易建構、高調整彈性、即時更新系統資料之優點。利用此方法對於半導體供應鏈之各階段、各產區之投料量、存貨量的變動進行控制。同時，本研究將會依照供應鏈的採購策略、供應鏈管理方法、需求變動的變化而設計八種不同的實驗，並應用模式預測控制於此八項實驗中，觀察八項實驗的存貨與投料之變動率以及是否可以保持高服務水準。而藉由本實驗之結果可以發現，模式預測控制可以有效地處理系統狀態與目標狀態的差異量，並將複雜的半導體供應鏈之存貨與投料保持於穩定的狀態，且同時可維持高服務水準。

As semiconductor industry faces more and more fierce competition, the competition originally between companies gradually is changed between supply chains. Actually, Supply chain management (SCM) in Semicondctor industry is an essential way to keep companies to keep their competitioness. Especially for Integrated Device Manufacteur (IDM), there are multiple manufacteuring stages of semiconductor manufacturing process. And IDM need to monitor their performance in different stages. Supply chain in semiconductor manufacturing is characterized by integrating dynamics, nonlinearity and high levels of stochasticity. Moreover, the semiconductor supply chain has vertical distintegration character, and each member of the supply chain for different goals which leads to inefficiency of supply chain performance and difficulty in monitoring it.

In this research, we use model predictive control (MPC) to do material purcurement, and manage inventory, WIP, material input for each different stage. MPC is a control technique in which a finite optimal control problem is solved at each sampling instant to obtain the control input. And MPC method is a successful way for semiconductor supply chain management to track inventory targets and satisfy stochastic customer demandes given uncertainty on yield rate and binning rate in factories. We design eight scenarios by purcurement type, supply chain management method and customer demand variation to compare the performance in the level of variation of inventory, control input and in the satisfication rate of service level. The way to measure the performance will by some indexes which can help us to quantize the performance. Through this research, we are able to understand MPC is an efficient way to minimizing the variation of inventory and control input and keep high service level at the same time.