Scientists at the US Department of Energy's Pacific Northwest National Laboratory (PNNL) are using a team of artificial-intelligence agents to analyse radiation damage in microchips more quickly, aiming to accelerate the development of electronics that can keep working in space and nuclear environments.
PNNL announced the work on October 26. Its advance is in speeding up the scientific analysis used to understand radiation damage and judge how materials and electronics might be improved.
What the team did
Researchers at the University of Tennessee used an ion beam to expose chips made from layers of aluminium gallium nitride and gallium nitride to concentrated radiation-like conditions, reproducing in a short time certain effects that build up over years in space.
Data from techniques including atom probe tomography and scanning transmission electron microscopy then revealed changes inside the materials.
PNNL's AI system, called SciLink, brought that evidence together. It coordinates several AI agents under three lead roles : experimental analysis, campaign planning and computational simulation. The system compared results across methods, proposed explanations for the damage, assessed the strength of its conclusions and suggested further experiments. PNNL says it completed this combined analysis in hours instead of the days or weeks it could take using conventional workflows.
However, the AI did not itself irradiate or test the chips.
Confidence scores and human oversight
The system reported 90 percent confidence that ions dominated the damage and 60 percent confidence in its conclusions about damage caused by protons. These figures represent the confidence scores from the AI workflow, not independently validated probabilities that the conclusions are correct. The gap between the figures gives scientists a reason to look more closely at the less certain finding.
PNNL says its researchers designed the experiments and checked the AI's output for errors, implausible calculations and unsupported reasoning, sometimes called hallucinations.
Why radiation hardening matters
Radiation can harm chips in three main ways, according to NASA. A single energetic particle can flip a stored bit or trigger a damaging surge of current. Repeated exposure can gradually change a chip’s electrical properties. Particles can also knock atoms out of place in the semiconductor itself. Engineers need to know which kind of damage occurred and where, which is the information the PNNL team wants to extract faster.
Gallium nitride and aluminium gallium nitride are wide-bandgap semiconductors used in power electronics and high-frequency devices. PNNL identifies them as relevant to satellite communications, power systems and radar. They are not immune to radiation.
Satellites and reactors
Satellite electronics must run for years without repair. Shielding adds mass and backup systems add cost and complexity, trade-offs highlighted by PNNL’s Ben Kowash, who leads the laboratory’s radiation-hardening work. Faster analysis could help researchers compare candidate materials and designs sooner.
Nuclear facilities also rely on electronics in harsh conditions. In June 2024, the US Department of Energy reported that Oak Ridge National Laboratory researchers tested gallium nitride transistors near a research reactor core at Ohio State University. They survived three days, including seven hours at 90 percent reactor power, and tolerated a dose 100 times higher than standard silicon devices at 125°C. That was a separate experiment and not a SciLink result. Ion-beam tests do not automatically show suitability near a reactor, where neutron, gamma-ray and temperature conditions differ.
