High-Fidelity Synthetic Transmission Electron Microscopy Image Generation Using Diffusion Probabilistic Models for Data-Limited Semiconductor Metrology

Researchers have developed a method to generate high-fidelity synthetic Transmission Electron Microscopy (TEM) images using diffusion probabilistic models, and proposed a new approach to uncertainty quantification in online decision-making.

The method, detailed in a study submitted to arXiv on 23 Jun 2026^[1], uses a Denoising Diffusion Probabilistic Model (DDPM) framework to generate synthetic TEM images. This framework enables training with only 15 samples and achieves high structural similarity with real images, with MS-SSIM > 0.98. The synthetic images preserve global structural and spatial relationships, meeting FAB metrology requirements. The DDPM feature representations can also be repurposed for segmentation tasks. In a related development, a separate article submitted to arXiv on 29 May 2024^[2] proposed a new approach to uncertainty quantification and exploration in online decision-making using autoregressive sequence models. This approach views uncertainty as arising from missing future outcomes that could be revealed through action choices, and assesses uncertainty through autoregressive generation rather than sampling latent parameters from posteriors.