半導體裝機配管工程成本估價模式之探討

Abstract

半導體裝機配管工程，又稱為二次配工程，為半導體廠房新建或擴建專案時，連接主要生產機台(例如黃光、擴散、薄膜、濕蝕刻、乾蝕刻、化學機械研磨、量測整合、離子植佈)之施工作業，其配裝成效顯著影響半導體廠產能及良率，係半導體廠關鍵之施工項目之一。實務上半導體裝機配管工程之成本估算，大都依據工程經驗及歷史資料。然而各半導體廠之產品目標不同，其使用之機台類型及數量亦具差異性，因此常見歷史統計資料具遺漏值問題，影響估價作業之準確性，所以適合應用人工智慧系統來做資料探勘。 本研究以268件半導體裝機配管工程之實際成本統計資料，分別針對半導體裝機配管工程分為11個工作項目(排氣、一般氣體、特殊氣體、真空管、製程冷卻水、基座、化學品、純水、排水、電力、低真空)建立特性參數，再透過類神經模糊系統之學習機制，建立半導體裝機配管工程成本預估模式。 本研究使用模糊適應性學習控制網路模式及快速混元基因演算法，各模組之估價平均精確率為78.8%~95.7%，較原估價方法(算術平均法)提升9.7%至33.3%。本研究在半導體裝機配管工程之規劃階段，初步獲得較精確之成本估價，可供後續研究者參考，期冀更多相關業者、工程公司投入後續相關研究。

Semiconductor hookup construction is the work for connecting the facility system and the tools of process module, such as Litho, Diffusion, Thin film, Wet etching, Dry etching, CMP, Integration, Implant module of semiconductor plant. And also one of the most critical construction works deeply influencing the capacity and yield of semiconductor plant. Practically the cost of semiconductor hookup construction is estimated mostly according to the experiences and the historic data. However, the types and quantities of the process tools used very with different production of each semiconductor plant. Therefore the common historical statistics have a problem of missing values which affects the accuracy of cost estimated. As of result, The AI system is applicable for data mining. This research establishes the characteristic parameters using 268 sets of actual cost data, including 11 working items (such as exhaust, bulk gas, specialty gas, vacuum line, process cooling water, foundation, chemical, ultra pure water, drain, power, low vacuum). It builds up a cost estimating model through the learning mechanism of the neuro-fuzzy system in order to improve precision for estimating the cost of the semiconductor hookup construction. This research use fuzzy adaptive leaning control network and fast messy genetic algorithms to estimate. The average precise rate is 78.8% ~ 95.7%, the original estimate method (arithmetic method of average) promotes 9.7% to 33.3%. This research at the stage of planning in the semiconductor hookup project, obtains the precise cost estimating initially. It may useful for the following researcher to refer. Deeply hope for the facility engineer of semiconductor and the Engineering firm could put into the following related research.