- Professor Kyung Min Kim’s research team in the Department of Materials Science and Engineering has developed a “programmable probabilistic neuron (PPN)” that harnesses semiconductor noise.

- The team demonstrated selective encoding of signals ranging from human activity to speech, paving the way for a key neuromorphic semiconductor technology capable of processing diverse types of signals.

A synthesis of ground surveys across Southwest China finds that soil thickness explains nearly 70% of the spatial variation in woody vegetation growth potential, far more than climate or soil nutrient concentrations. Below a 57.5-centimeter threshold, limited soil volume becomes the dominant constraint, where vegetation becomes less related to soil nutrient concentrations and more closely associated with climate variability.

Electroencephalography (EEG) foundation models are emerging as a promising approach for brain-computer interfaces, but their generalization ability remains unclear. Researchers developed EEG-FM-Compass, a benchmark reviewing 55 EEG foundation models and evaluating 12 open-source models across 13 datasets and nine brain-computer interface paradigms. The study shows that current models have transfer potential but require task-specific adaptation, while specialized models remain competitive and larger models do not necessarily generalize better.

Researchers have developed a catalyst that coordinates structure and chemistry from the atomic scale to the millimeter scale. The 3D-printed tough metallic-glass scaffold promotes fluid flow and bubble removal, while a NiMo nanoglass coating supplies active amorphous interfaces. The Gyroid-based NiMo@G catalyst required 1.438 V to reach 100 mA cm⁻² in alkaline water electrolysis and remained stable for more than 200 hours.