結合深度神經網路和隨機森林對化學研磨機台的研磨率之預測

Abstract

近年來半導體工業精密製程蓬勃發展，然而龐大的製程需要精密的控管才能有效提升良率並節省成本。然而晶圓片在化學研磨製程中的良率與研磨率息息相關，因此對於化學機械研磨製程(CMP) 中研磨率的預測為相當重要的課題。在本篇論文中，我們提出利用深度神經網路學習的方法結合隨機森林來進行預測，它比單一模型預測更穩定，更能準確地預測晶圓片的研磨率。

In recent years, the semiconductor industry is very flourishing. However, a cumbersome process requires sophisticated control to achieve the high yield rate and keep the cost down actually. And the yield of the wafer is related to the removal rate in the Chemical Mechanical Planarization process. Therefore, the prediction of the CMP removal rate is the important issues. In this paper, we combine the deep neural network learning and random forest to predict the removal rate, and it is more stable and gets less loss than using the only one method.