晶圓製造廠生產週期時間估算模式

Abstract

晶圓製造廠所追求的生產目標不外乎是高產出率、短生產週期時間、高良率及準時交貨等，而上述目標均與生產週期時間有關，故生產週期時間的掌握在晶圓製造廠中佔有舉足輕重的角色。有鑑於此，本研究應用區塊基礎式週期時間演算法(Block Based Cycle Time Forecasting Algorithm; BBCT)推估產品在各工作站之生產週期時間。 首先，以區塊基礎式週期時間演算法為基礎，透過考量工作站不同的加工批量特性、及前後機台產出速率相異等因素交互影響，估算出產品在各工作站之等候時間，包含：（1.）因機台負荷因素所形成之等候時間－以M/M/c等候模式估算、及（2.）等待形成批次機台可加工批量之集批時間－以區塊基礎式週期時間演算法估算之。 其次，因製程中批次機台之釋料往往造成下游序列機台的負荷瞬間增加，形成尖峰負荷，此種尖峰負荷所引發之等候機台加工時間（簡稱第二等候時間）必定大於由平均負荷所估算得知之等候時間。因此，吾人以BBCT為基礎，推估工件由批次機台流至各序列機台時所引發之第二等候時間。以及由序列機台流經批次機台所引發之集批時間，以得到各工作站之週期時間。 經實例驗證顯示：本文所求得之各工作站週期時間與模擬結果比較，平均誤差均能保持在6﹪，而運算時間卻僅需數分鐘。而應用此結果推估各工作站之在製品量，結果顯示本文所得之各工作站在製品量與模擬結果之平均相差0.2個lot，由此成效可知，本文所構建之生產週期估算模式可快速地提供生產規劃及控制系統精確的參考值。

The production target for the wafer fabrication includes high throughput、short cycle time、high yield and on time delivery,…, etc. All these targets are related with the cycle time. Precisely and quickly estimating the length of cycle times for each product type and the flow time for each workstation are important. In this study, the model for estimating the cycle time for each product type going through each workstation is developed based on the block based cycle time estimating algorithm (BBCT). To calculate the product's cycle time, the BBCT considers two factors：the variety of batch sizes on workstations and the difference in throughput rates among workstations. With such an idea, the average waiting time for a product waiting at a workstation includes (1) the waiting time resulted from the workload on the workstation, which is determined with the M/M/s queueing model and (2) the waiting time related to the batch forming, which is determined by the BBCT. However, the lot releasing from a workstation often causes the peak workload on the downstream serial workstations. The waiting time caused because of the peak workload (also named the second waiting time) will be longer than the waiting time derived with the average workload. Hence, when deriving the cycle time for a workstation, both the second waiting time resulting from the lots flowing from batch workstation to serial workstation and the batch forming time resulting from the lots flowing from serial workstation to batch workstation are considered. With the workstation's cycle time, Little's law is applied to calculate the workstation's WIP level. The case study shows that there only exists about 6% difference in average between the workstation flow time to the simulation results. Meanwhile, the calculation time of this model takes only a few minutes instead of a few hours for the simulation model. Result also shows that estimated the workstation WIP level is different from the simulation result about 0.2 lots in average only. Thus, the model developed for the cycle time estimation can quickly provide the good enough reference digits to production planning and production activity control system.