SEM Image Anchoring to Design for Array Using GAN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor defect review systems face challenges in accurately aligning scanning electron microscope (SEM) images with design targets due to limitations in cell size and uncertainty, leading to incorrect defect location reporting and increased complexity in identifying yield-relevant defects as design rules shrink.
Innovation Solution
The method involves using a generative adversarial network (GAN) to select anchor points from pixel-to-design alignment image patches generated by an optical inspection system, allowing for precise alignment of SEM images with a design clip, reducing location uncertainty from ±125 nm to ±25 nm, and enabling accurate defect detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional alignment methods are used with fixed anchor points, then the alignment process is simple, but alignment accuracy deteriorates due to uncertainty accumulation (±125 nm)
Solution Approach 1:
The patent transforms the static alignment process into a dynamic one by implementing iterative refinement. The system starts with a rough alignment using provided anchor points, then dynamically identifies new anchor points from repeating patterns in the SEM image, and repeatedly refines the alignment by alternating between image acquisition and design matching until convergence is achieved. This dynamic approach reduces uncertainty from ±125 nm to ±25 nm or better.
Solution Approach 2:
The patent introduces repeating patterns in the SEM image as intermediary reference features between the provided anchor points and the target defect location. These repeating patterns serve as intermediate alignment references that bridge the gap between coarse anchor-point-based alignment and fine target-location alignment, enabling sub-25 nm accuracy without requiring anchor points at every location.
2Productivity
If design rules are shrunk to increase device density, then productivity is improved, but measurement precision deteriorates due to smaller defect sizes and increased noise
Solution Approach 1:
The patent performs preliminary alignment to sub-25 nm accuracy using the iterative repeating pattern recognition method before attempting to detect and characterize defects. This preliminary high-precision positioning ensures that when defects are detected, their locations are accurately known, enabling reliable detection of small defects in shrunk design rules without being obscured by alignment uncertainty or stage noise.
3Measurement precision
If anchor points are provided for all layers by the semiconductor manufacturer, then alignment accuracy is improved, but device complexity increases due to additional data processing requirements
Solution Approach 1:
The patent extracts repeating patterns directly from the SEM image data at the target location and uses these extracted patterns as local alignment references. This eliminates the need to process and store anchor point data for all layers, as the system only processes the relevant repeating patterns in the immediate vicinity of each defect target. The anchor points provided by the manufacturer are used only for initial rough positioning, not for final alignment.
Data Source
AI summary
A scanning electron microscope receives a results file for a wafer from an optical inspection system. The results file includes an anchor point on the wafer. A defect review image at the anchor point on the wafer is generated using the scanning electron microscope. A design clip is aligned to the defect review image at the anchor point thereby generating an aligned defect review image. The aligned defect review image is used for defect detection.