A domestic research team has developed next-generation ultra-low-power semiconductor technology that reproduces the brain's ability to flexibly adapt to circumstances.
Kim Kyung-min, a professor in the Department of Materials Science and Engineering at the Korea Advanced Institute of Science and Technology (KAIST), said on the 28th that the team developed a "frequency-switching neuristor" that mimics "intrinsic plasticity," in which neurons (nerve cells) remember past activity and adjust their own response properties. The frequency-switching neuristor is an artificial neuron device that self-regulates the frequency of signals, like people becoming less startled as they grow accustomed to a stimulus or, conversely, becoming more sensitive through repeated training.
The human brain processes information not only by regulating consolidations (synapses) that send and receive signals, but also through "intrinsic plasticity," an adaptive ability in which individual nerve cells become more sensitive or less sensitive on their own depending on the situation. However, existing artificial intelligence (AI) semiconductors have struggled to emulate this flexibility of the brain.
The team combined a "volatile Mott memristor," which reacts briefly and then returns to its original state, with a "non-volatile memristor," which remembers traces of input signals for a long time, to implement a device that allows neurons to freely control how often they fire signals.
When they applied the developed technology and ran simulations, they achieved the same performance even with 27.7% lower energy consumption than conventional neural networks by leveraging the neurons' own memory function. They also observed resilience in which the network reconfigured itself through intrinsic plasticity to recover performance even if some neurons were damaged.
Kim Kyung-min said, "This study implements intrinsic plasticity, a core function of the brain, in a single semiconductor device, raising the energy efficiency and stability of AI hardware to a new level," adding, "This technology, which can remember its own state and adapt and repair itself even when damaged, could be used as a key device for systems that require long-term stability, such as edge computing and autonomous driving."
The findings were published online in the international journal in materials, "Advanced Materials," on Aug. 18.
References
Advanced Materials (2025), DOI: https://doi.org/10.1002/adma.202502255