Search papers, labs, and topics across Lattice.
This paper introduces a Denoising Diffusion Probabilistic Model (DDPM) framework specifically designed for generating high-fidelity Transmission Electron Microscopy (TEM) images in scenarios with limited data availability. By employing a progressive patch-based training strategy and integrating techniques such as TrivialAugment adaptation and classifier guidance, the model can generate full images from as few as 15 training samples while maintaining critical TEM-specific details and noise characteristics. The resulting synthetic images achieve an MS-SSIM score exceeding 0.98, demonstrating their utility for downstream machine learning applications in defect detection and metrology.
Achieving high-fidelity TEM image synthesis with as few as 15 samples could revolutionize data availability in semiconductor metrology.
Advanced semiconductor nodes drastically increased demand for Transmission Electron Microscopy (TEM), yet destructive sample preparation, slow imaging and high costs severely limit the availability of diverse datasets needed for downstream machine learning (ML). Synthetic data generation is becoming essential, but current generative models often miss TEM-specific noise, structural detail, and stochastic variability crucial for evaluation. We present a Denoising Diffusion Probabilistic Model (DDPM) framework for synthetic TEM image generation under extreme data scarcity. A progressive patch-based training strategy scales from low-resolution patches to full images, enabling from-scratch training with only 15 samples. We integrate a custom TrivialAugment adaptation, cross-process domain transfer, classifier guidance, and RePaint-style inpainting, culminating in full-image generation that preserves global structural and spatial relationships in compliance with FAB metrology requirements. Beyond synthesis, we repurpose DDPM feature representations for segmentation, partitioning encoder feature maps to obtain coherent region masks. Our synthetic images achieve up to MS-SSIM > 0.98 and qualitative expert assessment consistent with structural similarity results, facilitating downstream ML training for defect detection, segmentation, and metrology while preserving statistical and physical realism.
