KAIST researchers develop brain-inspired neuromorphic semiconductor

Researchers at KAIST, led by Professor Kyung-min Kim of the Department of Materials Science and Engineering, have developed a new neuromorphic neuron semiconductor technology that utilizes natural semiconductor noise to process information, mimicking the human brain.

Traditionally, noise in semiconductors is viewed as an interference that must be eliminated to ensure signal accuracy. However, the research team leveraged the inherent irregularity of noise—similar to how biological neurons respond stochastically to stimuli—to create a ‘Programmable Probabilistic Neuron’ (PPN). Using memristor devices, the team can adjust resistance states to control noise characteristics, converting electrical noise into probabilistic spikes.

In experimental tests, the technology demonstrated high efficiency across different frequency ranges. By adjusting the memristor's state, the artificial neurons were optimized for both low-frequency human activity signals (achieving 94.8% accuracy in motion recognition) and high-frequency audio signals (achieving 95.0% accuracy in speech recognition). This capability allows a single hardware component to be reconfigured for various signal types, offering significant potential for low-power edge AI systems in wearable devices and voice sensors.