利用EWMA結合歷史與市場預測資訊改善半導體產業之存貨管理

Abstract

限制理論(Theory of Constraints, TOC)使用需求拉動補貨(Demand-Pull)與緩衝管理(Buffer Management)管理供應鏈存貨，相較於傳統(s, S)、(s, Q)、(R, S)、(R, s, S)等補貨策略，此種方法除了不需要複雜的參數設定外，因為隨時考量了需求的即時資訊，其更能適用於動態的生產環境中。然而在半導體產業中，其產品生命週期短、需求變動大、生產前置時間長，若僅使用需求拉動補貨來管理存貨，將有可能因為無法即時針對需求變化做出反應而造成存貨過多或缺貨的風險。本研究提出一套方法，藉由結合產品需求的指數加權移動平均值(Exponentially Weighted Moving Average, EWMA)與市場對需求的滾動式預測(Rolling Forecast)資訊更加掌控產品需求的變動趨勢，並依照庫存水準制定補貨調整時機與補貨調整比例。為評估本研究提出之方法的可行性與有效性，本研究將此方法用於分析由國內某晶圓製造廠提供之實際案例資料與其餘的模擬資料。分析結果顯示本研究提出的方法相較於過去單純考量歷史需求資訊或單純考量預測資訊的補貨模式，確實能維持較低的平均存貨與較高的服務水準，讓需求拉動補貨策略更能應用於需求變異大且補貨前置時間長之產品上。

Theory of Constraints proposed using demand-pull replenishment policy and buffer management to manage supply chain inventory. Compared with traditional replenishment policy such as (s, S)、(s, Q)、(R, S)、(R, s, S), this policy better suits for a dynamic production environment by taking immediate demand information into consideration. Moreover, in this policy, there is no need to concern a complicated parameter setting. However, in the semiconductor industry, most of the products have short product life cycle, long production lead time and protean demand characteristics. In this situation, if we only use the replenishment policy and buffer adjustment method proposed by TOC, it may be unable to respond to changes in demand timely and result in the risks of too much inventory or out of stock. This study proposes a method by integrating historical demand and the rolling forecast from market demand by exponentially weighted moving average (EWMA) to observe the demand trends and decide the timing of buffer adjustment according to the inventory level. To test the validity and feasibility of the method the study proposed, we apply the method to the real historical data from Taiwanese wafer foundry plant and other simulation data. The results show that compared with the past studies concerning the historical demand information or forecasting information individually, the method we proposed has lower average inventory and higher service level, making demand-pull replenishment policy more suitable for the products with long production lead time and protean demand characteristics.