Wafer Defect Prediction via Interpretable SEM Mode Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect prediction methods in semiconductor manufacturing, such as those using parameter-based approaches or machine learning models, suffer from low accuracy and inability to provide information on the causes of defects, particularly in stochastic printing failures like microbridges and missing contacts during EUV lithography.
Innovation Solution
A machine learning model is trained on SEM images to decompose them into interpretable modes like critical dimension, shift, and ellipticity, allowing for the identification of defect causes by evaluating contributions from these modes, enabling early prediction and correction of defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If parameter-based approaches or machine learning models are used for defect prediction, then prediction speed is improved, but prediction accuracy and ability to identify defect causes deteriorate
Solution Approach 1:
The patent segments the defect prediction problem into multiple interpretable modes (e.g., focus, exposure, alignment, morphology) that can be independently analyzed. Each mode represents a specific physical characteristic or defect category, allowing the system to evaluate contributions from different failure mechanisms separately while maintaining high prediction accuracy through comprehensive coverage of all relevant defect causes.
2Measurement precision
If complex machine learning models are used for defect prediction, then prediction accuracy is improved, but interpretability and ability to identify root causes deteriorate
Solution Approach 1:
The patent introduces interpretable mode images as intermediary representations between the raw SEM image and the final prediction. These mode images serve as mediators that preserve both the predictive information needed for accurate defect detection and the interpretability needed to identify root causes. Each mode image corresponds to a specific physical characteristic, enabling traceability from prediction back to the underlying defect mechanism.
3Productivity
If wafer-level analysis is performed to improve manufacturing yield, then productivity is improved, but computational complexity and analysis time increase
Solution Approach 1:
The patent applies partial action by focusing the wafer-level analysis on specific interpretable modes that are most relevant to common defect types and manufacturing concerns. Rather than performing exhaustive analysis of all possible defect modes across the entire wafer, the system selectively evaluates the most critical modes, reducing computational complexity while maintaining the ability to identify and address the most significant yield-impacting defects.
Data Source
AI summary
A method and apparatus for analyzing an input electron microscope image of a first area on a first wafer are disclosed. The method comprises obtaining a plurality of mode images from the input electron microscope image corresponding to a plurality of interpretable modes. The method further comprises evaluating the plurality of mode images, and determining, based on evaluation results, contributions from the plurality of interpretable modes to the input electron microscope image. The method also comprises predicting one or more characteristics in the first area on the first wafer based on the determined contributions. In some embodiments, a method and apparatus for performing an automatic root cause analysis based on an input electron microscope image of a wafer are also disclosed.