PNNL speeds up radiation-hardened satellite chip testing with AI

Researchers at Pacific Northwest National Laboratory (PNNL) used a multi-agent AI system called SciLink to analyze radiation damage in gallium nitride and aluminum gallium nitride chips. The system processes ion-beam experiment data in hours instead of weeks and assigns confidence scores — 90% for heavy-ion-dominated damage and 60% for proton-induced defects.

• SciLink cuts radiation damage analysis from weeks to a few hours
• The system reported 90% confidence that heavy ions dominated the damage
• Ion-beam tests at the University of Tennessee mimic years of cosmic radiation
• Humans stay in the loop to check AI findings for hallucinations