.png){kind=icon}
.png){kind=icon}
Return
Input-guidance diffusion model for unknown defect patterns detection in wafer bin map
.png){kind=icon}
PRE
AI
DOI：10.1016/j.aei.2024.103078.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The detection of unknown defect patterns in wafer bin map (WBM) is crucial for maintaining high production yields in semiconductor manufacturing. Existing methods often fail to handle these unknown defect patterns because of their reliance on high-quality labeled data. In this study, we propose a WBM input-guidance diffusion model (WigDM) for detecting unknown defect patterns using unlabeled data. WigDM leverages the specific guidance of input samples to enhance reconstruction quality, allowing effective differentiation between known and unknown patterns. Furthermore, existing unknown detection scores are designed for natural image data and thus fail to accurately reflect the characteristics of WBM. Therefore, the proposed method introduces a new datadriven unknown detection score for WBM dataset. Experimental results on WM-811K and MixedWM38 datasets demonstrate the superiority of the proposed WigDM, outperforming existing methods in 32 out of 36 unknown defect pattern scenarios. The proposed method reduces the reliance on labeled data, offering a new solution for defect pattern detection in semiconductor manufacturing.
Keywords:
Semiconductor manufacturing
Wafer bin map
Unknown defect pattern detection
Diffusion models
Journal
IF:
9.9
Papers:
4.0K
Citations:
1.7W