BROWSE
Detail
Conference Paper
순열 중요도를 이용한 민감도 분석 기반 CMP 공정 MRR 예측 모델의 핵심 변수 조합 탐색
Cited - time in 
Share 
Share
- Issue Date
- 2026-06
- Citation
- 대한전자공학회 학술 대회 (하계) 2026, pp.1-4
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- Material Removal Rate (MRR) is a key performance indicator in Chemical Mechanical Planarization (CMP). This study investigates input variable sensitivity and variable reduction for MRR prediction using the 2016 PHM Challenge CMP dataset. CMP sensor data were summarized at the wafer-stage level, and 288 statistical features were used to train an XGBoost baseline model after removing zero-variance features. The baseline model achieved an R² of 0.9709 and NRMSE of 0.0461 on the evaluation dataset. Grouped permutation importance showed that usage-, pressure-, and rotation-related variables contributed substantially to MRR prediction. In the Top-k evaluation, the top seven input variables achieved an R² of 0.9691 and an NRMSE of 0.0476, comparable to the full 19-variable model. These results indicate that MRR prediction can be performed effectively using a reduced set of key input variables.
- KSP Keywords
- Baseline model, Key Performance Indicators(KPI), Material Removal Rate, Permutation importance, Statistical Features, Top-K, Variable reduction, chemical mechanical planarization, input variables, sensor data