Researchers have developed a Denoising Diffusion Probabilistic Model (DDPM) to generate high-fidelity synthetic Transmission Electron Microscopy (TEM) images for semiconductor metrology. This approach addresses the scarcity of real TEM data, which is limited by destructive sample preparation, slow imaging, and high costs. The DDPM framework utilizes a progressive patch-based training strategy and integrates techniques like TrivialAugment adaptation and RePaint-style inpainting to produce synthetic images that preserve structural and spatial realism, achieving high MS-SSIM scores and expert validation. Beyond image generation, the model's features are also repurposed for image segmentation tasks, aiding in defect detection and metrology. AI
IMPACT Enables more robust ML training for semiconductor defect detection and metrology by overcoming data scarcity.
RANK_REASON The cluster contains a research paper detailing a new method for synthetic data generation using diffusion models.
- Denoising Diffusion Probabilistic Model
- machine learning
- semiconductor metrology
- Transmission Electron Microscopy
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
