半導體設備備用零件存貨預測模式之研究

Abstract

半導體製造業是一種要求高精密度且技術密集的工業，由於機台的快速更新，促使新設備與零件不斷的大量推陳出新，為了使加工能持續且穩定的進行，減低機台不定時的當機所造成的損失，必須準備足夠的備用零件；然而，目前晶圓廠備用零件的庫存金額經常佔總庫存成本之極高的比例，其庫存量的大小對晶圓廠的生產力有很大的影響。為了解決因備用零件庫存過多造成庫存持有成本提高，或是庫存過低發生缺料而造成生產中斷將導致公司產生當機成本，對備用零件需求預測大部份現存的方法都是以經驗法則作判斷，但是因為其準確偏低，導致為了避免各機台因為缺少備用零件而當機，紛紛在製造現場堆積大量存貨和零件，造成資金的積壓與呆料的情形普遍存在。 本研究提出一個半導體設備備用零件存貨預測模式的研究架構，主要針對設備備用零件之失效機率型態，建立備用零件存貨預測模式，在統計預測模式構建上，考慮機台與備用零件的特性，定義其失效型態為一符合韋伯分配之機率函數，進而找出零件失效情形在浴缸曲線上的動態變化，利用備用零件需求之歷史資料，經由統計分析技術求算備用零件之預估備料量，此模式將提供設備備用零件存貨水準的資訊，作為訂定零件需求之安全存量的參考，同時，使半導體產業能參考此一模式架構針對備用零件之需求作有效的管理。

In high precision required semiconductor manufacturing, the machine's rapid updating prompts the continual mass outdating of the new equipment and parts. In order to continuous and steady progress of process, it is necessary to prepare enough spare parts reduction the loss caused by the machine's abrupt breakdown; however, the stock of parts in the wafer fab factory often occupy the extremely high proportion in the total stock cost, the stock amount affect the production of the wafer fab factory largely. To solve the raising of the stock cost caused by the excess stock of spare parts, or the machine's breakdown cost caused by the production interrupt made by the shortage of parts lower stock. Most of the existing methods for parts demand prediction by experience. But low precision make cash idle and parts excessive, to enable prevent machine from "down" owing to the lack of spare parts, and pile up mass stock and parts in the factory. In this research, we have developed the structure of the stock forecasting model for the semiconductor equipment's parts; namely, it construct a forecasting model of the parts stock for the failure probability model of the equipment's stock parts. In build to the statistical forecasting model, considering the characteristics of the machines and spare parts, we define the failure model follow weibull distribution, then find the dynamic change of the bathtub curve the parts is failure, and use the historical data of the demand of spare parts to prediction demand of the spare parts through the statistical analysis technology. This model will provide the data for the stock level of spare parts, to which was referred when we decide the safe stock; at the same time, it can invoke the semiconductor manufacturing to refer to the model structure to control the spare parts demand.