半導體廠機台預警管理之研究

Abstract

半導體廠具有投資金額龐大，製造流程複雜，設備當機不易預測等特性。晶片從投產到產品加工完成約需2個月。若機台無法維持正常運作，加工出來的晶片不是重工(rework)就是報廢(scrap)。業界經常利用機台預警系統，監控機台的參數，希望能在機台故障之前及時檢修保養。然而太保守的預警門檻設定會造成不必要停機，影響產能。本研究利用成本分析的方法，改善預警系統區間設定與備份件檢修順序的決策，期能幫助半導體廠節省成本並提升產能。

Semiconductor manufacturing involves some characteristics: capitally intensive, long cycle time, and unexpected machine breakdown. The cycle time for producing a wafer job takes about two months. A wafer job, even almost finished, may have to be scrapped due to an unexpected machine malfunction. Therefore, practitioners often establish a machine alarm system by setting thresholds for operation signals of machine. Threshold setting is a trade-off decision. Using an easy-to-alarm threshold tends to increase the production yield at the price of losing capacity. This research develops a cost model for two decision-makings in a machine alarm system: (1) how to determine the alarm threshold for a monitored signal, (2) how to determine the repairing sequence while a machine has been alarmed. Track, a significant type of machine, is used as a testing example for the cost model. Test results indicate that the cost saving due to a better alarm threshold may be much higher than the total salaries of maintenance engineers.