Synthetic SEM Image Generation for Privacy-Preserving Hardware Assurance
Paper Overview
- Research area: Computer Vision (CV)
- Authors: Gijung Lee, Ronald Wilson, Damon L. Woodard
- Published: 2026-08-11
- arXiv: 2508.03801
- Privacy by distortion: Functional hardware layouts are heavily warped before training, so the synthetic layouts do not replicate any specific proprietary routing or sensitive IP.
- Two-stage synthesis: A StyleGAN generates diverse layout masks; a Pix2PixHD conditional GAN translates them into photorealistic SEM-style images that retain authentic textures and imaging noise.
- Sim-to-real transfer: A segmentation model trained exclusively on the synthetic SEM images successfully transfers to real SEM images, indicating that the synthetic distribution is sufficiently faithful.
- Performance gain over limited real data: The synthetic-only segmentation model outperforms baseline models trained on the small available real dataset.
- Reduced attack surface: Because the deployed model is trained on novel synthetic data, exposure to gradient inversion and membership inference attacks that target training data is mitigated.
- Data efficiency: Only a small set of initial example layouts is required to bootstrap the pipeline, addressing data scarcity in hardware assurance.
Summary
Hardware assurance relies on scanning electron microscopy (SEM) to verify nanoscale structures, but assembling the large, high-quality datasets required for automated analysis is impeded by time-intensive acquisition and strict intellectual property (IP) constraints on proprietary designs.
The authors propose a privacy-preserving pipeline that secures IP by heavily distorting the functional design while generating a visually realistic synthetic dataset from a small set of initial examples. A StyleGAN first learns the distribution of hardware layout masks to generate novel, macroscopically varied structures. Subsequently, a conditional GAN (Pix2PixHD) translates these masks into realistic SEM images that preserve authentic textures and noise.
Key Findings
Original Abstract (Excerpt)
> Hardware assurance relies on scanning electron microscopy (SEM) to verify nanoscale structures, but assembling the large, high-quality datasets required for automated analysis is impeded by time-intensive acquisition and strict intellectual property (IP) constraints on proprietary designs. We propose a privacy-preserving pipeline that secures IP by heavily distorting the functional design while generating a visually realistic synthetic dataset from a small set of initial examples. A StyleGAN first learns the distribution of hardware layout masks to generate novel, macroscopically varied structures. Subsequently, a conditional GAN (Pix2PixHD) translates these masks into realistic SEM images that preserve authentic textures and noise. The primary finding of this work is that a segmentation m...
---
*Auto-collected 2026-08-12*
#paper #arXiv #CV