Autoencoder Defect Detection for Semiconductor Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect detection methods struggle with accurately identifying defects in semiconductor devices with complex, arbitrary design shapes due to the enormous number of variations in small region combinations, making it difficult to prepare and apply appropriate models for inspection.
Innovation Solution
A system and method using an autoencoder trained on multiple images at different locations, dividing input images into narrow sub-regions, and comparing them with output images to detect defects by calculating the degree of discrepancy, facilitating high-accuracy inspection without requiring design data or golden images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the size of the small region is increased to reduce the number of models, then the number of variations of combinations of complicated shapes becomes enormous, but model formation becomes difficult
Solution Approach 1:
The patent divides the image into multiple small regions and generates a separate estimation model for each small region. This segmentation approach allows the system to handle complex patterns by breaking them down into simpler, manageable units, where each unit has its own dedicated model rather than requiring one model for the entire complex image.
Solution Approach 2:
The patent changes the parameter of region size to be sufficiently small so that the geometric shape within each small region can be considered simple. This parameter change enables the use of simple estimation models for each small region while still being able to detect defects in complex overall patterns through the aggregation of multiple small region analyses.
2Difficulty of detecting and measuring
If the size of the pattern in the small region is decreased to simplify the shape, then the number of models becomes enormous, but it becomes difficult to prepare and apply models
Solution Approach 1:
The patent creates a library of estimation models that can be universally applied to different small regions. Each model in the library is designed to handle specific simple geometric shapes, and the system selects and applies the appropriate model from the library to each small region based on its characteristics, rather than creating a unique model for every possible region configuration.
Solution Approach 2:
The patent performs preliminary preparation by creating a library of estimation models in advance, categorized by geometric shape types. This preliminary action allows the system to quickly select and apply pre-prepared models during defect detection, avoiding the need to create models on-demand for each small region.
3Measurement precision
If estimation models are generated in small region units for complex patterns, then defect detection accuracy may improve, but the time and resources required for model preparation increase significantly
Solution Approach 1:
By segmenting the complex pattern detection task into multiple small region analyses, the system achieves high defect detection accuracy for complex patterns while reducing model preparation complexity. Each small region has its own simple estimation model, but collectively they provide comprehensive coverage of the complex overall pattern.
Solution Approach 2:
The patent changes the parameter of region size to be sufficiently small, which simplifies the geometric shapes within each region and enables the use of simple estimation models. This parameter change reduces model preparation time and complexity while maintaining the ability to detect defects in complex overall patterns through the aggregation of multiple small region results.
Data Source
AI summary
The purpose of this disclosure is to generate a reference image on the basis of a proper model even for a sample such as a semiconductor device including many patterns and to perform a defect inspection using the reference image. This disclosure proposes one or more computer systems for identifying defects in a received input image. The one or more computer systems include a training device including an autoencoder that has been trained beforehand by inputting multiple images at different positions in a training image. The one or more computer systems divide the input image into multiple input images, input same to the autoencoder, and compare images output from the autoencoder with the input images.