Detecting Important Features and Predicting Yield from Defects Detected by SEM in Semiconductor Production.

Area of Science:

• Semiconductor Manufacturing
• Materials Science
• Electrical Engineering

Background:

• Optimizing semiconductor production requires accurate yield prediction based on in-process defect detection.
• Scanning Electron Microscopy (SEM) is crucial for identifying wafer defects during manufacturing.

Purpose of the Study:

• To identify optimal semiconductor layers for Scanning Electron Microscope (SEM) inspection.
• To develop a predictive model for semiconductor electrical failures using detected defects.

Main Methods:

• Odds Ratio analysis to rank inspection layers based on their predictive power for final yield.
• Gradient Boosting regression/classification model to predict electrical failures from SEM-detected defects.
• Validation of both models on two independent semiconductor datasets.

Main Results:

• A ranked list of critical semiconductor layers for SEM inspection was identified, enabling focused process control.
• A Gradient Boosting model successfully predicted electrical failures from wafer defects, confirming Odds Ratio findings.
• Both developed models effectively handled data lacunarity, enhancing prediction accuracy.

Conclusions:

• Targeted SEM inspection on identified key layers significantly improves semiconductor yield prediction.
• The developed Gradient Boosting model offers a robust method for predicting semiconductor failures, optimizing the production process.
• This research provides actionable insights for enhancing semiconductor quality control and reducing manufacturing costs.