興建一座半導體晶圓製造廠往往需要至少五至七佰億新台幣,加上晶圓製造的加工途程,具有高度的迴流(Reentrant)特性,造成物料搬運活動非常頻繁。因此,晶圓製造廠的佈置方式以及自動化搬運系統的配置在初期規劃階段就顯得格外的重要。目前晶圓製造廠的佈置方式是以脊椎式(Spine)佈置為主,其目的是為了尋找一個運輸成本最低、物流最順暢,並且能夠減少搬運車碰撞與振動的最好佈置方式。 本研究結合遺傳演算法(Genetic Algorithm, GA)以及空間填滿曲線(Space-Filling Curve, SFC)的優點,藉由田口實驗設計的方法,求得演算法的最佳參數組合,同時針對12個脊椎式設施佈置的案例加以測試。每個案例各有四種不同的物流方向,包含順時針單向、雙向、順時針單向加一捷徑、雙向加一捷徑等,探討其搬運成本的差異。最後將本研究與其它演算法所得結果做一比較與分析,結果發現本演算法的求解品質績效為96%以上,在求解20個工作站案例中,所得到的最佳解已經優於目前已知的最佳解。
The construction cost for the wafer lab. of a semiconductor manufacturing is about NT$ 50 to 70 billons. Due to the higher reentrant characteristics of the manufacturing routing, the automated material handling system and the layout of the manufacturing fab. will determine the manufacturing efficiency at the earlier planning stage. Current of the wafer fab. manufacturing layouts are mostly based on the spine layout concepts. The spine layout is to allocate the material handling system at the center (like the spine of human body) of the manufactory facilities, the rest of the bays are assigned to areas along the spine. This layout can reduce the material handling distance (time) and decrease the collision and vibration of the material handling systems. This study will combine the genetic algorithm (GA) and the space filling curve (SFC) method to solve the facility layout problem of the wafer fab. manufacturing system. We also employ the Taquchi method to find out the best combination of the design parameters. The layout improvement heuristics are evaluated on 12 study cases of fab. layouts with 8 to 20 bays. Four different material flow directions (clockwise, bi-direction, clockwise with a short cut, and bi-direction with a short cut) will be analyzed to compare the material handling cost. The results showed that the quality of solution is more than 96%. In 20 bays case, the solutions obtained are better than previous results.