Semiconductor Defect Detection Using GAN-Generated SEM Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for detecting semiconductor defects during manufacturing are time-consuming and require skilled personnel, and may miss defects due to reliance on visual inspection, which is prone to human error and variability.
Innovation Solution
A machine learning-based module, specifically a generative adversarial network (GAN) circuit, is used to generate scanning electron microscope (SEM) and segment (SEG) images from computer-aided design (CAD) images, enabling the prediction and localization of defects in semiconductor devices during the manufacturing process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If visual inspection method is used for defect detection, then defect classification can be performed, but detection time increases and requires skilled personnel
Solution Approach 1:
The patent replaces the manual visual inspection system with an automated machine learning system. The GAN-based circuit automatically generates SEM images and defect segmentations from CAD images, eliminating the need for skilled personnel to perform visual inspection manually. This substitution of mechanical/manual inspection with automated computational inspection resolves the contradiction by providing both high accuracy and fast processing.
Solution Approach 2:
The patent uses GANs to generate synthetic SEM images and defect segmentations that replicate real inspection data. By copying the essential features of actual defect patterns through generated images, the system enables automated defect detection without requiring physical samples or manual inspection, thus reducing time while maintaining reliability.
2Reliability
If visual inspection method is used for defect detection, then defect classification can be performed, but defects may be missed due to human error and variability
Solution Approach 1:
The patent replaces the human visual inspection system with an automated GAN-based machine learning system. This substitution eliminates human error and variability, providing consistent defect detection results. The automated system processes all defects uniformly without fatigue or subjective judgment, resolving the contradiction between reliability and system complexity.
Solution Approach 2:
The GAN-based system performs self-learning and self-improvement through automated training on defect data. The system autonomously generates defect segmentations and SEM images, then uses these to refine its detection capabilities without requiring continuous human intervention or recalibration, thereby maintaining high consistency while managing complexity through self-service mechanisms.
3Productivity
If machine learning-based module is used for defect prediction, then detection efficiency improves, but computational resources are required
Solution Approach 1:
The patent performs preliminary defect prediction by generating SEM images and defect segmentations from CAD images before actual manufacturing completion. This advance prediction allows defects to be identified early in the design or early manufacturing stages, enabling faster decision-making and reducing the need for extensive physical testing, thereby improving productivity while managing computational energy through early intervention.
Data Source
AI summary
Disclosed is an operating method of an electronic device for manufacture of a semiconductor device. The method includes receiving, at the electronic device, a computer-aided design (CAD) image for a lithography process of the semiconductor device, and generating, at the electronic device, a first scanning electron microscope (SEM) image and a first segment (SEG) image from the CAD image by using a machine learning-based module, and the first SEG image includes information about a location of a defect.