封閉迴圈式主生產排程再排程之研究--以晶圓製造廠為例

Abstract

面臨外部市場、內部現場變異很大之環境下，封閉迴圈式生產計劃與 控制系統是建立與維護一套實際可行之生產計劃必然的發展方向。然而過 去有關這類研究在架構上有二點不夠週全之處：(1) 變異資訊未因應用性 質再加以分類與處理；(2) 因應之評估標準定義模糊不顯著。因此，本研 究針對半導體製造業，在及時化預測式生產方式下，建立兩模組：(1) 混 合巨觀與微觀流動時間估計模組；(2)再排程模組。 過去有關迴歸導向流動時間估計之研究，都以巨觀為主。巨觀流動時 間估計模式簡單，但估計誤差大；反之，微觀流動時間估計模式複雜，但 估計誤差小。巨觀與微觀流動時間估計模式各有其優點與限制，因此本研 究提出混合巨觀與微觀流動時間估計模式，在流動時間估計誤差與模式複 雜度之間取得平衡。本文並比較混合巨觀與微觀流動時間估計模式與傳統 流動時間估計模式之優劣，結果顯示混合巨觀與微觀流動時間估計模式優 於後者。 關於再重排模組方面，其包含了再重排時機決定、再排程評估、排程 修正三個階段。對於再重排時機與評估，本研究在混合重排方式下，建立 一總投料變化量指標，來評估臨機性再重排是處於正常狀況或重大變動， 以便在客戶服務水準與排程穩定性之間取得一平衡點。至於在排程修正方 面，本研究在總預計下料量與投料期限已知之情況下，提出了瓶頸機器負 荷平準 (BLL)投料法則與修正式加入最短等候列 (JSQ)平行瓶頸機器指派 法則，使產品組合、多部瓶頸機器在投料期限內形成三維負荷平準，最後 產生細部投料排程與瓶頸機器指派時程，使各產品每日產出變異最低。

Exposing rapid change of external market and internal shop, closed-loop production plan and control system is a necessary development direction to build and maintain a set of practical and feasible production plan. However, previous related researches still had two incomplete points. The first one is variance information are not further classified and processed according to the application nature. The second one is that the rescheduling evaluation criteria and following revision procedures are not so ambiguous. Therefore, this paper proposes two modules: mixed macro and micro flow time estimation model, and rescheduling model namely, for wafer fabrication under Just- In-Time, Make-To–Forecast environment. Previous investigations involving flowtime estimation of regression approach have extensively applied the macro flowtime estimation concept. The ‘macro’ flowtime estimation has less model complexity and larger estimation error. In contrast, the 'micro’ flowtime estimation approach has larger model complexity and less estimation error. Each approach has its own merits and limitations. Therefore, in this study, we examine the feasibility of mixing the macro and micro flowtime estimation models. Comparing the proposed model with variety of macro and micro models reveals that the mixed macro and micro flowtime estimation model can achieve a balance between flowtime estimation error and model complexity. The rescheduling module contains rescheduling frequency, evaluation, and revision. For rescheduling frequency and evaluation, this paper builds a releasing variation index to evaluate whether environment is under normal condition or catastrophe. It maintains a balance between service level and schedule stability under hybrid rescheduling policy. For rescheduling revision, assuming total releasing quantity and releasing interval are known, Bottleneck Load Leveling (BLL) releasing policy and modified Joint the Shortest Queues (JSQ) bottleneck machines assignment rule are proposed to reach three dimensional load leveling among product mix, multiple bottleneck machines, and releasing interval. Finally, detailed releasing schedule and bottleneck machines assignment table are generated to minimize the variance of daily move out.