AI-powered end-to-end product lifecycle: UX-centric human-in-the-loop system boosting reviewer productivity by 82% and accelerating decision-making via real-time anomaly detection and data refinement with GPU-accelerated computer vision, edge computi

Challenging high-variability images such as ultra-high-resolution scanning electron microscope (SEM) images demonstrate extremely low signal to noise ratio, yet they are essential for lithography process control, defect detection, and optical proximity correction (OPC) modeling in semiconductor manufacturing. Traditional review methods, including manual inspection and critical dimension (CD)-based modeling, are slow, prone to human error, and limited to localized measurements. As the industry shifts toward contour-based modeling, which requires analyzing entire SEM images, there is an urgent need for fully automated, GPU-accelerated, cloud-scalable AI-driven solutions. This paper presents a novel GPU-accelerated machine learning (ML) and AI-powered digital image processing framework for automated image data refinement and defect detection. The system generates high-resolution pattern fidelity and defect maps by leveraging structural similarity with AI-enhanced adaptive reference models. Key features—such as AI-driven extrema counting, sharpness metrics, and noise level evaluation—are extracted for clustering images into quality-based groups. Cloud-scalable hierarchical clustering improves defect detection accuracy, surpassing CD-based methods and enabling real-time defect classification. A major limitation of traditional CD-based defect filtering is its reliance on multiple SEM images per process condition (dose and focus) to identify outliers. This paper addresses this challenge by introducing an AI-powered clustering approach that can detect failures even when only a single SEM image is available per process window condition, eliminating the need for rule-based defect filtering. Additionally, this method integrates AI-driven line edge roughness (LER) and line width roughness (LWR) evaluation, ensuring robust clustering across varying process conditions. A GPU-accelerated, cloud-integrated GUI streamlines SEM image analysis, enabling metrology and OPC engineers to perform HPC-optimized, fully automated data cleaning and defect detection with real-time AI-based scoring. This solution is integrated into Calibre SEMSuite™, leveraging multi-cloud deployment for enhanced scalability, usability, and performance.