The Korea Atomic Energy Research Institute (KAERI) announced on the 24th that a joint research team from its Advanced Radiation Technology Institute, Chungbuk National University, and Belgium’s IMEC has, for the first time in the world, verified next-generation artificial intelligence (AI) semiconductor technology that operates reliably even in space radiation environments. IMEC is a nonprofit, comprehensive semiconductor research institute jointly established by Belgium, France, and the Netherlands.

With the rapid recent advances in space exploration technologies, securing “radiation-hardened” characteristics that enable semiconductor devices handling artificial intelligence (AI) and big data analytics to withstand the harsh radiation conditions of space has emerged as a key challenge.

In response, the research team fabricated synaptic transistors based on indium, gallium, and zinc oxide, which are next-generation semiconductor materials, and validated the feasibility of using AI semiconductors in space environments. A synaptic transistor is a device that mimics the “synapse,” the junction responsible for signal transmission between neurons in the human brain, and performs high-efficiency AI computation with low power consumption.

After fabricating the devices and evaluating their characteristics, the research team irradiated them with a 33 MeV (mega-electronvolt, a unit of energy) high-energy proton beam using KAERI’s proton accelerator. The radiation dose of the irradiated beam was set to a level equivalent to more than 20 years of exposure to space radiation in low Earth orbit (the typical lifespan of a low Earth orbit satellite is about 5–15 years). A subsequent re-evaluation of the device characteristics showed some performance degradation, such as a partial decrease in drive current, but confirmed that the core switching operation of the semiconductor and the synaptic plasticity (the ability to modulate neuronal connection strength), which is essential for neuromorphic devices, were stably maintained. In particular, in handwritten digit recognition tests conducted to verify AI computation efficiency under radiation exposure, the devices recorded a high pattern recognition accuracy of 92.6%.

The research team stated, “This study proves that the device can sufficiently function as a neuromorphic computing system even under the extreme condition of high-energy radiation,” adding, “We plan to expand the research by further studying technical strategies to compensate for performance degradation and develop it into a core technology in the field of AI semiconductors for aerospace applications.”

Popular News