運用季節性時間序列預測模式在半導體製造業提升製程控制品質之研究－以化學機械研磨製程為例

Abstract

經由對目前半導體業界常用的EWMA控制器進行觀察後發現，對於更換關鍵零組件或耗材後所形成研磨速率的激烈變化，現行的控制法則仍有可改善的空間。主流的改善方向採取將控制法則複雜化來進行，但是這將直接提高了企業軟體開發與系統維護的成本。本研究試圖將關鍵零組件或耗材生命時間導入預測函數，並配合採用季節性時間序列預測模式，將Run-to-Run控制器的控制核心簡單化，搭配Excel程式即可運算出控制所需要的回饋參數。在Mean Squared Error與製程能力指標的分析比較結果之下，以關鍵零組件或耗材生命時間為自變數的季節性時間預測模式控制器，其控制結果表現良好，能夠達到半導體製造業對於先進製程控制Run-to-Run系統的要求。

Through the observation of commonly EWMA controllers in the semiconductor industry currently found the space of improvement at the existing controllers for the intense variation of polishing rate after replacing the key component parts or consumables. The major improvement direction is to take the complexity control rule and it increases the costs of software development and system maintenance for the enterprises directly. This study attempts to implement the life time of the key component parts or consumables in prediction function. By the seasonal time series forecasting model, it will simplify the core of Run-to-Run controller. And we can calculate the requested feedback parameter of control with Excel program. Based on the analysis and comparison results of Mean Squared Error and Process Capability Indices, the control results of seasonal time series forecasting model controller as the variables of the life times of the key component parts or consumables perform well. It can achieve the requirements of the semiconductor manufacturing industry for advanced process control of Run-to-Run system.