Pattern Inspection System Using ML Image Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing pattern inspection systems for semiconductors require extensive manual effort and time to create learning data for machine learning, which hampers production line operations and is inefficient in handling variations in SEM photography conditions, circuit shapes, and manufacturing process fluctuations.
Innovation Solution
A pattern inspection system that utilizes a storage unit to store pattern images and data, along with an image selection unit to choose learning pattern images based on pattern data and photography conditions, reducing the need for extensive learning data and shortening the learning time by selecting minimum necessary data for target inspection performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive learning data is prepared to cover variations in SEM photography conditions, circuit shapes, and manufacturing process fluctuations, then inspection accuracy is improved, but the time and effort required for true value creation increases significantly
Solution Approach 1:
The system performs preliminary analysis of design data to predict pattern variations and pre-selects representative learning data before actual inspection. By anticipating the types of variations that will occur (photography conditions, circuit shapes, manufacturing fluctuations), the system prepares only the necessary learning data in advance, avoiding the need to collect and manually annotate extensive data covering all possible variations.
Solution Approach 2:
The system generates synthetic learning data by copying and transforming design data (GDS files) to simulate various photography conditions and process variations. Instead of collecting real-world images under all possible conditions, the system creates virtual copies of design patterns with applied transformations (scaling, rotation, noise addition) that mimic actual inspection variations, significantly reducing the need for physical sample preparation and manual annotation.
2Reliability
If manual creation of true values for each learning data is performed to ensure accuracy, then inspection reliability is improved, but productivity decreases due to the time-consuming nature of manual work
Solution Approach 1:
The system automatically generates true values for learning data by processing design data through the same inspection algorithms that will be trained. The machine learning model uses design data (which contains ground truth information about pattern locations and characteristics) to automatically create labeled training samples without human intervention. This self-service approach maintains reliability because the true values are derived from authoritative design specifications rather than manual annotation, while dramatically improving productivity by eliminating the manual true value creation process entirely.
3Measurement precision
If a large amount of learning data is collected to achieve target inspection performance, then model accuracy is improved, but the learning time and computational resources required increase to several weeks or months
Solution Approach 1:
The system extracts only the essential features and variations from design data that are most relevant for training the inspection model. Instead of using all possible design data equally, the system identifies and extracts key pattern variations (critical dimensions, shape variations, density changes) that have the greatest impact on inspection accuracy. This selective extraction reduces the effective learning data volume while maintaining model accuracy, thereby reducing learning time and computational requirements.
Solution Approach 2:
The system transforms design data into multiple variations by applying parameter changes (scaling factors, rotation angles, noise levels, focus variations) to simulate different inspection conditions. Rather than collecting extensive real-world data under all possible conditions, the system efficiently generates diverse training samples by systematically varying parameters of a smaller base dataset, achieving comprehensive coverage with less total data and faster training.
Data Source
AI summary
A pattern inspection system inspects an image of an inspection target pattern of an electronic device using an identifier constituted by machine learning, based on the image of the inspection target pattern of the electronic device and data used to manufacture the inspection target pattern. The system includes a storage unit which stores a plurality of pattern images of the electronic device and pattern data used to manufacture a pattern of the electronic device, and an image selection unit which selects a learning pattern image used in the machine learning from the plurality of pattern images, based on the pattern data and the pattern image stored in the storage unit.