KAIST Develops Semiconductor Neuron That Mimics Brain's Intrinsic Plasticity

Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have developed a "Frequency Switching Neuristor," a semiconductor device that mimics the biological brain's ability to autonomously adjust its sensitivity. This breakthrough allows AI hardware to implement intrinsic plasticity, enabling a system to reorganize itself and maintain performance even after sustaining partial damage.

The device achieves this biological mimicry by combining two types of Mott memristors within a single hardware architecture. The system pairs a volatile Mott memristor to handle momentary reactions with a non-volatile memristor to provide long-term memory. This combination allows the neuristor to remember past activity and adjust its spiking frequency based on repeated exposure to stimuli, effectively becoming more or less sensitive depending on the context.

The Gap in AI Hardware

Traditional AI semiconductors typically focus on regulating synapses—the connections between neurons—to process information. However, biological neurons are more flexible, utilizing intrinsic plasticity to adapt their own internal states independently of their connections. This research, led by Professor Kyung Min Kim and published in Advanced Materials, seeks to bridge this gap by creating a single semiconductor device capable of autonomous state-memory and adaptation.

Efficiency and Resilience

The implications for hardware efficiency are significant. Simulations conducted by the KAIST team demonstrate that the Frequency Switching Neuristor achieves the same performance levels as conventional neural networks while consuming 27.7% less energy. Beyond power savings, the technology introduces a level of structural resilience previously unseen in rigid semiconductor circuits. Because the neurons can adapt their sensitivity, the network can reorganize its processing flow to restore performance if specific neurons are damaged.

"This study implemented intrinsic plasticity, a core function of the brain, in a single semiconductor device, thereby advancing the energy efficiency and stability of AI hardware to a new level," stated Professor Kyung Min Kim.

Future Applications

This advancement provides a critical foundation for the development of more stable edge computing and autonomous driving systems, where hardware must operate reliably in unpredictable environments. By enabling devices to adapt to external stimuli and recover from circuit failures in real-time, the neuristor moves AI hardware closer to the robustness of organic intelligence. Future development will likely focus on scaling these devices into larger, integrated neuromorphic chips.