類神經網路於預測半導體機台異常值之應用

Abstract

先進的半導體製程皆是由非常精密且複雜的機台所生產。對於製程精密度的要求使得這些生產機台的控制參數(Recipe)也是動輒成千上百，其中只要有些許的關鍵值偏差，就有可能造成製程偏離(Excursion)，使得辛苦生產的晶圓良率降低，甚至報廢。 所以，半導體廠商很早就注意到機台異常監控的重要，只是其重要性隨著製程技術的提升而愈來愈高。對於機台監控的精細程度要求也愈高，從每一批次FOUP(Front Opening Unified Pod)檢查一次參數、到每一片晶圓檢查一次、甚至到12吋廠時要求到每一秒或每0.1秒都要去檢查。而這種檢查機制必須依靠資訊系統才能做到，半導體產業給這類軟體系統一個通用的名字，稱為FDC (Fault Detection and Classification)。 FDC能夠及時探測到機台參數偏差，當參數偏離原值、並超出設定的區間範圍時發出警報。如果能夠結合FDC開發出一套預警機制，事先預測機台在未來可能發生的異常值，就可以依據預測結果事先做出預防措施，以防止事態變嚴重，進而促進製程良率與產能的提升。

Advanced semiconductor processes are produced by very sophisticated and complex machines. Processes precision requirements make these production machine control parameters (Recipe) is often hundreds or even thousands, of them as long as the deviation of some critical value, it may cause the process to deviate from (Excursion), making the hard production of wafers good rate, or even scrapped.

Therefore, semiconductor manufacturers have long noted the exception monitoring of machines is important. The demand has higher precision monitoring system is becoming more importance when the devices are shrunk into smaller sizes. The high quality and high solution checking mechanism must rely on advanced information system, such as FDC (Fault Detection and Classification).

FDC can timely detect the machine parameter deviations when the parameters deviate from the original value and exceed the range of the specification. If early warning mechanism may predict abnormal value, the precautionary measures might prevent the failure from becoming more severe, thus contributing to the process yield and throughput enhancement.