Korean AI chip endures space radiation, preserves neuromorphic computing functions
A joint team of Korean researchers and international partners has demonstrated that a domestically developed AI semiconductor can operate stably in space-like radiation. The project was led by Korea Institute of Nuclear Research and Collaboration with Chungbuk National University and IMEC of Belgium, under the oversight and support of the Ministry of Science and ICT and KAERI’s Advanced Radiation Research Center.
Researchers fabricated an indium-gallium-zinc-oxide (IGZO) based synapse-mimicking transistor and exposed it to a 33 MeV high-energy proton beam to simulate the harsh radiation environment of space. The radiation dose used in testing was equivalent to more than 20 years of exposure in Earth’s low orbit.
The study found that, although the device exhibited some degradation in drive current, its core functions remained intact: the switching operation and the ability to modulate synaptic strength—essential for neuromorphic computing—were preserved. In neuromorphic AI tests under radiation, the system achieved a handwriting recognition accuracy of 92.61 percent.
The researchers also demonstrated a reservoir computing configuration, showing four-bit arithmetic capability and the processing of time-series data, which are important for practical AI workloads in space applications. They plan to pursue further technical strategies to mitigate performance loss and to strengthen the radiation-effect analysis framework.
Oh Dae-hyun, a policy official at the Ministry of Science and ICT, described the results as a meaningful milestone that suggests AI systems can operate in extreme environments like space. He said Korea will continue supporting efforts to secure core technologies for aerospace AI semiconductors.
The results were published in the March issue of the Journal of Semiconductor Processing and Materials. The work reflects growing global interest in radiation-hardened AI hardware, a factor that could influence future satellite autonomy, deep-space missions, and defense-related space technologies.
Why this matters to the United States: space programs and commercial ventures in the U.S. increasingly rely on onboard AI for autonomous navigation, data analysis, and fault detection. Demonstrations of radiation-tolerant neuromorphic AI hardware could shorten the development cycle for space-ready AI chips, affect supply chains for radiation-hardened components, and spur potential collaboration with international partners on rugged AI architectures suitable for satellites, probes, and other space assets. The involvement of IMEC and the publication in a peer-reviewed journal also highlight international collaboration pathways that could intersect with U.S. research and development efforts in aerospace hardware and semiconductor technology.