AI on the factory floor,
with no network to build

Semiconductor fabs are among the most closed environments in manufacturing. Nothing connects out, and nothing leaves. That is exactly what makes AI hard to deploy there, and exactly what makes it valuable.

Semiconductor fabs are among the most closed environments in manufacturing. Nothing connects out, and nothing leaves. That is exactly what makes AI hard to deploy there, and exactly what makes it valuable.

The Challenge

SK hynix is one of the world's leading memory semiconductor manufacturers. They needed AI-based vision inspection for product quality control on the manufacturing line. Two things made cloud inference impossible: production data cannot leave the facility, and running AI through a server would have meant building network infrastructure the line was never designed to carry.

What we built

We deployed the vision model directly on-device, running on the camera at the inspection point. No server, no cloud call, no data leaving the line.

The Result

Quality inspection running fully on-device
No additional network infrastructure required on the line
No confidential production data leaves the facility

Have a model that cannot go to the cloud?

Have a model that cannot go to the cloud?