World's first verification of space-radiation-tolerant neuromorphic AI hardware by Korean-Belgian team
A joint research team from Korea Atomic Energy Research Institute’s Advanced Radiation Research Institute, Chungbuk National University, and IMEC in Belgium announced the world’s first verification that a next‑generation AI semiconductor can operate stably in space radiation conditions. The Korean government supported the project, and the teams described the result as a milestone for space AI hardware.
The researchers built a synaptic transistor based on indium-gallium-zinc-oxide, or IGZO, a material already used in some display and semiconductor applications. They tested whether this neuromorphic device could function in the harsh radiation environment of space, where onboard AI processing could reduce reliance on downlink data and enable more autonomous spacecraft operations.
To simulate long-term exposure, the team irradiated the device with a 33 MeV high-energy proton beam at KAERI’s accelerator facility. The radiation dose was calibrated to resemble the level a device in low Earth orbit would experience over more than two decades, noting that typical small satellites and cubesats have lifespans of five to fifteen years.
Post-irradiation assessments showed some degradation: the driving current of the device declined modestly. Despite this, the core switching capability remained intact, and the synaptic plasticity—the key feature that enables neuromorphic systems to adapt by adjusting connection strengths—was preserved.
In testing AI performance, the researchers ran neuromorphic computing simulations using the MNIST handwriting dataset and achieved a pattern recognition accuracy of 92.61%. They also demonstrated a reservoir computing setup for time-series processing, confirming four-bit computational capability in a space-relevant radiation environment.
The team said the findings demonstrate that IGZO-based synaptic devices can function as neuromorphic AI hardware even under extreme high-energy radiation, and outlined plans to develop strategies to mitigate performance loss. They also aim to strengthen evaluation systems for radiation effects and expand testing to include neuromorphic circuits and logic operations.
Why this matters beyond Korea: for the United States, the work touches on space security, autonomous spacecraft, and data-handling efficiency. Onboard AI reduces the need for constant communication with ground stations, lowering latency and bandwidth demands for missions ranging from satellite constellations to deep-space probes. Radiation-hardened AI silicon is also of growing interest to NASA, the Department of Defense, and commercial operators as they pursue longer-duration missions and more capable, self-reliant spacecraft.
The collaboration highlights a growing international effort to mature space-ready AI hardware, with implications for global supply chains of radiation-tolerant semiconductors and for policy debates about resilience and autonomy in space infrastructure. It also signals potential cross-border partnerships in advanced materials and neuromorphic computing platforms as U.S. agencies and industry look to diversify sources of space-ready AI technology.