Die defect detection: Datasets & Models

Manual die inspection under a microscope is slow, fatiguing, and inconsistent across operators, which is a real problem at the throughput volumes and tight tolerances of semiconductor fabrication. Subtle defects like hairline cracks or micro-contamination are easy to miss by eye, and missed defects that reach packaging or assembly become far more expensive to catch downstream. Automating this with computer vision gives you repeatable, objective inspection that runs at line speed and catches anomalies consistently regardless of shift or operator. It also produces structured defect data you can analyze to trace yield loss back to specific process steps.

A trained detection model applies the same criteria to every die, improving consistency and freeing skilled operators for higher-value work while scaling to inspect thousands of dies per hour. It integrates with existing AOI cameras, inspection microscopes, and wafer probe stations rather than requiring new capital equipment. Because every detection is logged with location and defect class, teams get traceable data that supports yield analysis and continuous process improvement across the fab.

1. Upload an image: Drop any photo into the widget above to try die defect detection for free, right on this page.
2. Preview instant results: The model detects crack, scratch, chip, contamination particle, die edge within seconds and returns an annotated image with structured predictions, so you can immediately see how well it works. Adjust the detection classes and re-run until the results fit your scenario.
3. Like what you see?: With a free Roboflow account, you can deploy this workflow with the Serverless Cloud API, customize it in your workspace, or train RF-DETR on your own images for production accuracy.
4. Get started for free: No code or training data required to begin: upload, review the results, and make it part of your stack in a single click.