Human experts working alongside agentic AI are building the foundation for lighter, longer-lasting satellites. Scientists at Pacific Northwest National Laboratory (PNNL) are accelerating the development of “radiation-hardened” (rad-hard) materials using Super Intelligence (SI) and automation.These materials are built to withstand high-radiation environments, such as space and nuclear reactors. Radiation-induced failures in these components can delay or cancel missions, making radiation hardening important for decade-long operations.“Space is a highly energetic environment, with heavy ions and other particles bombarding our devices. Agentic workflows come at a time when there is a need for better rad-hard testing techniques,” said scientist Erin Barker. SciLink AI systemResearchers are studying layered aluminum gallium nitride and gallium nitride chips that power satellites, communications, radar, and power systems. Before a microchip ever tastes the void of space, it is manufactured in a clean facility where even a single stray speck of dust could mean instant failure. Once launched into space on a satellite, that delicate microchip enters a dangerous environment filled with constant, invisible radiation. Galactic cosmic rays, heavy ions, and near-light-speed electrons pepper the cosmos.One bad exposure can scramble a satellite’s memory, send a spacecraft off course, or completely kill an expensive weather or communications mission.Testing how materials handle space environments has long been a slow and tedious process. To prevent these orbital catastrophes, PNNL researchers are changing how we shield space electronics. This new work accelerated the development of radiation-hardened microchips using a multi-agent AI system named SciLink. This intelligent software hierarchy evaluates microscopic radiation damage in just a few hours. It could replace analytical work that previously took human teams weeks.Physical shielding or backup systems protect electronics, but extra weight raises launch costs and reduces overall mission efficiency. Therefore, making the microchips themselves radiation-resistant may be a more practical solution.Protecting satellite microchipsAchieving this requires testing advanced chips made from aluminum gallium nitride and gallium nitride under extreme simulated conditions.Using ion-beam experiments at the University of Tennessee to mimic years of cosmic radiation in hours, PNNL researchers generated massive, complex datasets. Normally, integrating data from instruments like atom probe tomography and scanning transmission electron microscopy takes weeks.SciLink organizes its AI agents into a corporate hierarchy led by three “project leads” overseeing experimental analysis, planning, and computational simulation. In interactive sessions, these AI leads debate hypotheses, synthesize multi-instrument data, evaluate damage mechanisms, and propose follow-up experiments. “The system allows us to take into account all the measurements to piece together what has happened, to understand the damage that has occurred and why it happened,” said Jenna Pope, a PNNL scientist. “It’s like having several people who did different forms of testing come together and discuss the results — only it’s an agent pulling together the information instantly.”The AI also assigns numerical confidence scores to every conclusion. In recent tests, SciLink reported a 90 percent confidence score that heavy ions dominated the primary microchip damage. It also signaled a cautious 60 percent confidence regarding proton-induced defects, letting researchers know exactly where to dig deeper.Despite the blinding speed of agentic workflows, experts stress that humans remain firmly in the loop to design experiments, run calculations, and double-check AI findings for hallucinations. And even scientist Barker asserts, “Intelligent agents are a tremendous help, but you can’t just turn your brain off.”