Semiconductor Wafer Processing With Sparse Defect Image Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in semiconductor manufacturing is the detection of smaller defects in wafers due to shrinking dimensions, which can cause device failures and affect electrical parameters, necessitating efficient inspection methods to enhance yield and quality.
Innovation Solution
A method utilizing sparse coding and competitive feature exclusion (COMPFEX) procedures to encode semiconductor wafer images, followed by hebbian strengthening (HEBS) to amplify relevant features, enabling efficient retrieval of similar images from an indexed database for defect detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional inspection methods are used for semiconductor wafers, then the inspection process is simple, but the detection accuracy of smaller defects deteriorates
Solution Approach 1:
The inspection process is segmented into multiple stages: initial inspection to identify candidate defects, retrieval of similar images from database, and detailed analysis of retrieved images. This segmentation allows the system to focus computational resources on potential defects rather than processing entire wafer images uniformly, thereby improving defect detection accuracy while managing complexity.
Solution Approach 2:
The system performs preliminary actions by building and maintaining a database of defect images in advance. When a new wafer is inspected, the system can quickly retrieve similar defect images from this pre-built database, avoiding the need to analyze all possible defect patterns from scratch. This preliminary preparation significantly enhances detection accuracy for small defects without proportionally increasing real-time inspection complexity.
2Productivity
If the wafer dimensions are shrunk to increase device density, then the productivity increases, but the difficulty of detecting defects worsens
Solution Approach 1:
The system creates copies of known defect images from the database and uses them as reference patterns for detecting similar defects on current wafers. By comparing actual defect candidates against these copied reference images, the system can reliably detect small defects on shrunk wafers even when the defects are smaller than the optical resolution limit, thus maintaining detection capability despite increased device density.
Solution Approach 2:
The inspection system dynamically adjusts parameters such as image magnification, contrast enhancement, and matching thresholds based on the detected defect size and type. When small defects are detected on shrunk wafers, the system changes imaging parameters to enhance visibility and adjusts matching parameters to improve sensitivity, thereby overcoming the detection difficulty imposed by reduced wafer dimensions.
3Productivity
If the inspection process is automated to improve yield, then the productivity increases, but the device complexity increases
Solution Approach 1:
The inspection system performs self-service by automatically retrieving similar defect images from its own database and performing the comparison analysis without requiring external expert intervention. The system autonomously identifies defect candidates, queries the database for matching patterns, and generates inspection results, thereby automating the inspection process and increasing throughput while the modular architecture manages system complexity.
Solution Approach 2:
The system implements feedback mechanisms where inspection results are continuously fed back into the database to expand the collection of known defect patterns. This feedback loop enables the system to learn from past inspections and improve future detection accuracy automatically. The feedback-driven learning reduces the need for manual system reconfiguration, managing complexity while enhancing automated inspection capabilities.
Data Source
AI summary
Methods for processing a semiconductor wafer are provided. Patches are encoded with a set of weightings to obtain an encoding matrix. Images corresponding to the query image are retrieved according to the encoding matrix. An inspection result is generated according to the retrieved images. A semiconductor process is performed on the semiconductor wafer when the inspection result is normal.