半導體供應鏈網路配置模式之敏感度分析

Abstract

隨著科技進步以及生產趨勢轉向著重少量客製化，半導體供應鏈變得非常動態且敏感，以往傳統的生產規劃方法，很難對此動態的模式加以有效地管理規劃。因而綜合考量包括服務層級、工作變異、績效表現及績效控制的網路半導體供應鏈模型被發展出來，可以有效的描述半導體供應鏈的運作，進而規劃出最佳的配置模式。但是由於半導體供應鏈的極度複雜性，求解此模式必須花費極長的時間，若在供應鏈上有擾動，重新求解將緩不濟急，且供應鏈對容忍環境變異的資訊並不易得知。因此本論文想深入探究此半導體供應鏈模型的敏感度分析，特別是限制式寬裕區間的議題。 由於模型的目標函數為不定型的二次函數，並不具有二次規劃中半正定型函數的許多良好性質，故先要對目標函數轉換後，才能進行敏感度分析。本研究利用相似性轉換的方法，解決變數間交乘項的問題；再利用分段規劃的方法，對二次係數為負的變數做片段切割逼近，最後用半正定型的片段二次函數逼近原目標函數。接著應用沃夫對偶定理的性質，進行此半導體供應鏈模型的敏感度分析。在考量各種不同擾動後，將可得到最佳配置模式下的各限制條件的寬裕區間，並從中分析得到半導體供應鏈的敏感因子。最後，可利用所得到的資訊，建立起此半導體供應鏈之監控機制。

Semiconductor supply chain operations are very sophisticated and dynamic so that they are hard to be described precisely by traditional scheduling and planning methods. In order to elucidate the behavior of these operations of semiconductor supply chain network, a new behavior model, which takes quality of services, adaptability to process varieties, engineering changes, controllability and scalability of performance metrics into consideration, was developed. This then allows us to find the optimal supply chain configuration. However, by exploring the tolerance of environment disturbance, the impact of these disturbances on optimal supply chain model can be obtained without resolving the model due to the computational complexity of the model. Thus, this thesis intends to look into the issues of range of feasibility of this semiconductor supply chain model. Unfortunately, the objective function of our supply chain model is indefinite quadratic even though we know that the positive semi-definite property is useful in the sensitivity analysis of quadratic programming. Therefore, we propose an approach that an indefinite function is initially transformed and approximated by separable programming and piecewise positive semi-definite functions such that the sensitivity analysis can be performed by applying the Wolfe-duality theory. Next, the feasibility ranges of all the constraints in the optimal configuration under different scenarios of perturbation can easily be explored. Moreover, the critical factors of the supply chain can be identified according to the sensitivity analysis results. Finally, with all these results, a possible monitoring mechanism can be developed for making quick adjustments if any inefficiency is detected.