A research team from the Institute of Modern Physics of the Chinese Academy of Sciences and Lanzhou University has obtained important experimental evidence for revealing brain memory mechanisms and developing new-type neuromorphic computing. Their findings were published in Advanced Functional Materials.

Human learning and memory originate from highly dynamic connections structures between neural synapses. These structures act as intelligent switches that transmit signals and undergo dynamic remodeling, thereby enabling the encoding, storage, and retrieval of information, and serving as the biological basis for cognition and behavioral adaptation.

Biological nervous systems rely on synapses functioning as natural memristors, which process and store information through the controllable transport of ions and neurotransmitters within nanochannels. The key to the brain's ability to perform complex computations with extremely low energy consumption lies in the dynamic adjustment of synaptic connection strength based on prior activity. Replicating this effect through controllable fabrication in liquid systems is crucial for studying neural network functions and advancing the development of brain-computer interfaces and biological neuromorphic computing.

In this study, the researchers demonstrated the memristive effect in bio-inspired nanochannels using two distinct stimulation mechanisms: the divalent ion screening effect and pH-driven deprotonation.

Using the single-ion microbeam facility at the Heavy Ion Research Facility in Lanzhou (HIRFL), the researchers prepared bio-inspired nanopores. The combined action of the two mechanisms—ionic transport symmetry breaking and surface effects in the nanopores—works in synergy to produce hysteretic characteristics in ion transport. This nanofluidic memristor also simulates biological memory features, including short-term and long-term potentiation effects, as well as key synaptic functions such as paired-pulse facilitation and paired-pulse depression.

Furthermore, the researchers encoded synaptic weights dynamically, which is a core mechanism for adaptive learning behaviors in neuromorphic systems. To verify its potential applications, the researchers constructed a three-layer artificial neural network for pattern recognition. They conducted training and testing on a handwritten digit dataset and achieved a recognition accuracy of 94.6%. The nanofluidic memristor's performance is comparable to that of many solid-state memristive synapses.

This study was supported by the Ministry of Science and Technology of China, the National Natural Science Foundation of China, and the Scholarship of the University of Chinese Academy of Sciences.