Semiconductor Defect Localization Using Probability-Guided Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current examination processes for semiconductor specimens face challenges in efficiently detecting defects, particularly in small structures with low signal-to-noise ratios, and require high-resolution tools that can only inspect a small area in a reasonable time, leading to inefficiencies in time and cost.
Innovation Solution
A system and method that utilizes a deep neural network to analyze low-resolution images, build a probability map of defect locations, and iteratively refine the inspection by correlating defectivity data with specimen thickness, directing high-resolution tools to areas with high defect probability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution examination tools are used to detect defects in small structures, then measurement precision is improved, but productivity deteriorates because the tools can only inspect a small area in a reasonable time
Solution Approach 1:
The patent segments the specimen into multiple regions and processes them differently: low-resolution imaging is applied to the entire specimen to identify candidate defect areas, while high-resolution imaging is applied only to selected candidate areas. This segmentation allows the system to maintain high measurement precision for defect detection while improving productivity by limiting expensive high-resolution imaging to only necessary regions.
Solution Approach 2:
The patent applies local quality by using different imaging resolutions for different regions of the specimen. Low-resolution imaging covers the entire specimen to provide broad coverage, while high-resolution imaging is concentrated on specific candidate defect areas. This local differentiation optimizes both detection precision (where needed) and inspection throughput (across the whole specimen).
2Productivity
If low-resolution imaging is used to cover large areas, then productivity is improved, but measurement precision deteriorates due to low signal-to-noise ratio in small structures
Solution Approach 1:
The patent performs preliminary low-resolution imaging of the entire specimen to identify candidate defect areas before applying high-resolution imaging. This preliminary action allows the system to quickly screen large areas (maintaining productivity) while using the results to guide subsequent high-resolution imaging to only those areas where defects are likely present (restoring measurement precision where needed).
Solution Approach 2:
The patent introduces an intermediary processing step between low-resolution and high-resolution imaging: a neural network analyzes the low-resolution images to predict and identify candidate defect areas. This intermediary acts as a bridge, translating low-resolution data into targeted high-resolution imaging decisions, thereby maintaining both productivity and measurement precision.
3Measurement precision
If iterative refinement with neural network analysis is performed, then measurement precision is improved by correlating defectivity data with thickness, but device complexity increases
Solution Approach 1:
The patent implements a multi-functional neural network system that performs multiple tasks: analyzing low-resolution images, predicting defect locations, correlating with thickness data, and guiding high-resolution imaging. This universal system handles diverse data types (images, thickness measurements) and performs multiple functions (detection, prediction, correlation, guidance) within a single framework, improving measurement precision while managing complexity through functional integration.
Solution Approach 2:
The patent employs feedback mechanisms where the neural network's defect predictions are correlated with thickness data, and the results feed back into refining the defect location analysis. This iterative feedback loop continuously improves measurement precision by incorporating multiple data sources and refinement cycles, while the automated nature of the feedback reduces the operational complexity for users.
Data Source
Figure 1
Figure 2~3
Figure 3A~4
AI summary
There is provided a method and a system configured to obtain an image of a one or more first areas of a semiconductor specimen acquired by an examination tool, determine data Datt, informative of defectivity in the one or more first areas, determine one or more second areas of the semiconductor specimen for which presence of a defect is suspected based at least on an evolution of Datt, or of data correlated to Datt, in the one or more first areas, and select the one or more second areas for inspection by the examination tool.