新興晶元代工廠生產規劃與排程系統之構建

Abstract

晶圓代工廠在擴充(ramp-up)時期的產能變異、產出目標改變、新製程開發、機台適 應問題、產品組合易變動等現象，都造成其生產規劃與排程的困擾。有鑑於此，本研究擬 以階層式理念建立完整之生產排程規劃系統，包括：訂單承接、主生產排程、細部排程等 規劃模式。 首先，本研究利用產能估算來構建晶圓代工廠的訂單承接模式，用以決定最佳之實際 生產產品組合。其次，依工廠擴充的進度，將主生產排程規劃幅度劃分為若干個環境固定 之階段別(stage)，再運用數學規劃模式，訂定各階段之系統最適在製品量、估算各產品 各層級(layer)流動時間，並以層級拆解的方式，確切求算其生產週期時間。另一方面， 配合動態產能需求規劃來訂定投科時間表，並依生產週期時間推估產出時點及設定交期。 最後，將前述規劃成果細分至各產品子層級別，規劃各產品各層級每日應完成作業數，以 控制產品平順化產出。 實驗結果顯示，本規劃系統估算各工作站利用率與模式相差僅約0.78%；系統最適在 製品量平均相差7.89%，其中第三階段更低達1.63%。若以產品層級別觀之，檢定結果也顯 示與模擬並無顯著差異。 新興晶圓廠在擴充時期內的生產規劃模式，必須兼具「考量環境變異」與「迅速規劃 」的功能，才足以符合其需求。而實驗證明，本研究可在無法使用模擬工具情況下，提供 一套有效且迅速執行的規劃系統。

A good production planning in ramp-up fab is difficult to be built due to the variation of production environment, such as the increase of capacity and throughput target, the changes of product mix and the development of new process technology. In this study, we design a complete planning system containing three models with the hierarchical conception. First, the order-accepting model is used to determine the optimal product mix with the considerations of the gained profit and the capacity limitation. Second, we define each planning period as a planning stage, different from others, in which the production environment is unchanged. The length of every planning period may be not the same. For each stage, we design a heuristic with mathematical programming to derive the utilization rate of each workstation and the ideal system WIP level so as to achieve the throughput target. Then, we scheme the wafer start time under the fixed-WIP releasing policy with the dynamic capacity requirement plannign (DCRP). Besides, according to the layer-splitting heuristic, production cycle time for each photolithography layer of each product type can be estimated exactly. The wafer out time thus can be computed for setting the due-date. At last, we use the sub-layer as the planning object in the detail production scheduling model. With this design, the daily moves for each sub-layer of each product type can be planned to level the production flow. To satisfy the requirement of a ramp-up fab, we have to deal with not only the variation of production environment but also the function of quick planning. From the experiments, we find that the calculation of utilization rate for each workstation is very close to simulation results, about 0.78% difference only. Besides, the expected system WIP deviate by 7.89% from those predicted by simulation model, even 1.63% only in the third planning period. These results mean that the planning system we proposed above is effective and efficient for a ramp-up fab without any simulation tool.