運用看板概念於半導體供應鏈存貨控制機制

Abstract

半導體產業是個結合高科技、高資本密集、激烈競爭與快速改變特性的產業，且現今半導體產業多以製造服務業自居，使得過去僅僅強調單一工廠訂單滿載產能利用率最高或產品產出量最大等績效指標最佳化不符所需，因此為了能夠同時增加產能利用率、降低系統生產成本與快速反應顧客需求等等，供應鏈上下游的整合在半導體產業有著不可替代的重要性。 本研究將原用於製造系統的看板管理概念應用在半導體供應鏈生產系統，探討以供應鏈系統總體成本最小化為目標，並考慮總體系統訂單轉單最大外包比例的限制下，應用看板管理概念對各生產階段最大在製存貨數量進行生產控制的方法，建構整合半導體產業多階段在製存貨監控模型，設定供應鏈系統中各生產階段的在製存貨監控上界。在求解上本研究所發展之結合貪婪搜尋與全域搜尋的演算法，可有效地求出各階段最佳化之在製存貨監控上界。最後並透過以接近半導體產業實際生產情況的實驗分析與模擬，以驗證本研究所提出之在製存貨監控模型適用於實際產業情況與整合半導體供應鏈系統生產之價值。 本研究透過實例分析與模擬系統加以驗證，發現為達成總體成本最小化的目標，可由產能利用率較小的後端生產階段做彈性增加在製存貨監控上界，使得前端產能較為滿載之晶圓製造階段可以降低在製存貨監控上界，由降低晶圓製造階段的總成本以降低總體系統成本。並且發現結合貪婪搜尋與全域搜尋的演算法在本研究所建構之在製存貨監控模型上，可同時改善單純用貪婪搜尋可能造成求解結果與實際最佳解相差過大，以及僅用全域搜尋求解造成求解時間過長的問題。

Semiconductor industry is becoming a service-oriented business with high technology, capital investment and fierce competition. Maximizing the capacity utilization or throughput of single production stage is not effective for the whole supply chain system. Thus, in order to improve capacity utilization, reduce total cost and make quick response to customers at the same time, there is a need to integrate each production stage in the semiconductor supply chain to achieve those targets. In this research, we apply the Kanban concept originally used in the manufacturing system to the semiconductor supply chain system to construct an integrated multi-stage WIP control model. This model is used to decide the WIP control upper limit of each stage with the objective of minimizing total cost and the constraint of the maximal demand outsourcing proportion of the whole supply chain system. A solution methodology which combines the greedy search algorithm and global search algorithm is also developed to find the optimal set of the WIP control upper limit of each stage effectively and efficiently. Our result shows that our proposed model can be adapted to the practices in the semiconductor supply chain. By the coordination in each production stage in the supply chain, the efficiency of the whole supply chain system can be improved in the performance matrices such as total cost, average system cycle time, the quoted lead time, and etc.. Besides, comparing with greedy search algorithm and global search algorithm, the solution methodology developed in this research is truly effective in terms of quality of our solution and calculation efficiency.