Semiconductor Image Processing for Defect Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Recent semiconductor devices are miniaturized, making it difficult to accurately extract defects from wafer surfaces due to diffraction patterns caused by periodic structures, and the varying shapes, sizes, colors, and luminance of defects in images captured.
Innovation Solution
A semiconductor image processing apparatus comprising a processing circuitry that identifies feature amounts in input images using an identifier, learns a model to infer these amounts, and performs additional learning using both simulated and real images to enhance defect extraction, employing a self-learning unit, self-feedback unit, and image segmentation models to classify and emphasize defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If semiconductor devices are miniaturized, then device density and integration are improved, but defect extraction accuracy deteriorates due to diffraction patterns from periodic structures
Solution Approach 1:
The patent segments the image processing task into multiple specialized models: a defect extraction model that isolates defects from diffraction patterns, a diffraction pattern extraction model that separates periodic structures, and an identifier model that classifies defect types. This segmentation allows each model to specialize in specific features, improving defect extraction accuracy in miniaturized devices where diffraction patterns interfere with defect detection
Solution Approach 2:
The patent introduces intermediary processing steps including image preprocessing to reduce diffraction pattern effects, and intermediate feature extraction stages that separate defect signals from diffraction artifacts before final classification. These intermediary steps act as mediators between the raw captured image and the final defect identification, enabling accurate defect extraction despite miniaturization-induced diffraction
2Adaptability or versatility
If multiple fabrication processes are used, then device functionality is improved, but defect variety and image complexity increase making accurate extraction difficult
Solution Approach 1:
The patent develops a universal defect extraction system that handles multiple defect types from various fabrication processes through a single integrated framework. The identifier model is trained to recognize diverse defect categories including particles, scratches, and fabrication artifacts, making the system versatile across different fabrication processes while managing image complexity through unified multi-class classification
Solution Approach 2:
The patent employs parameter changes in the form of adjustable model configurations and training parameters to adapt to different fabrication process variations. The system can modify extraction thresholds, classification parameters, and model weights based on the specific fabrication process being analyzed, enabling accurate defect extraction across diverse processes without requiring separate specialized systems for each
3Ease of operation
If traditional image processing methods are used, then processing simplicity is maintained, but defect identification accuracy deteriorates in the presence of diffraction patterns and varying defect characteristics
Solution Approach 1:
The patent implements self-service through automated model training and selection mechanisms. The system automatically trains extraction and identification models on process-specific data, selects appropriate model configurations based on the fabrication process type, and adapts to new defect patterns without requiring manual reconfiguration. This maintains ease of operation while achieving high accuracy through automated adaptation to complex defect scenarios
Data Source
AI summary
A semiconductor image processing apparatus including a processing circuitry, the processing circuitry configured to identify a label corresponding to a feature amount included in an input image by using an identifier, learn a model for inferring the feature amount included in the input image and learns the identifier, and perform additional learning of the model based on the input image and the learned identifier.