Large-Scale Nanofiber Manufacturing| Ephaptic Coupling in Ultralow-Power lon-Gel NanofberArtifcial Synapses for Enhanced Working Memory

Prof. Xiao Kai from Southern University of Science and Technology Adv. Mater.: Ion-Gel Nanofiber Artificial Synapses Achieve Ultralow-Power Working Memory Enhancement

In recent years, the efficient information processing characteristics of biological neural networks have inspired research in brain-like computing, and the realization of neuromorphic computing systems similar to biological systems has become a research hotspot. The information transmission in biological nervous systems essentially relies on ion movement. The field potential generated by ion transmembrane transport in neurons can not only efficiently transmit information (with energy consumption of only about 10 fJ per synaptic event) but also achieve globally coordinated functions without direct connections through the ephaptic coupling effect, which is crucial for memory formation and cognition. However, existing neuromorphic devices mostly focus on simulating biological synapses at the functional level while paying insufficient attention to ion transport mechanisms, making it challenging for these devices to achieve the complexity and connectivity of biological systems.

Recently, the research group of Associate Professor Xiao Kai from Southern University of Science and Technology and collaborators proposed a bionic synaptic device based on a double-layer heterogeneous ion-gel nanofiber network by learning from the ion communication strategy of biological neural systems. By utilizing the ion potential relaxation behavior during charging and discharging to mimic the pulsed discharge function of synapses, they achieved the array construction of ion neuromorphic devices for low-energy brain-like computing. This structure reproduces synaptic plasticity through selective ion trapping and asymmetric migration mechanisms, with an energy consumption of only 6 fJ, comparable to biological synaptic efficiency. A reservoir computing system based on ion short-term memory and nonlinearity achieved 88% accuracy in MNIST handwritten digit recognition, demonstrating the efficiency of ions in edge computing. More importantly, by leveraging the unique global sharing characteristics of ions, the device array verified an ephaptic coupling effect similar to biological neurons, achieving over 94% brain-like accuracy in n-back working memory tasks, opening a new path for constructing high-efficiency, high-complexity brain-like intelligent systems.The study, titled "Ephaptic Coupling in Ultralow-Power Ion-Gel Nanofiber Artificial Synapses for Enhanced Working Memory," was published in the latest issue of Advanced Materials.

1.1 Construction of Ion-Bionic Synaptic Devices
The research team used electrospinning technology to fabricate an ion-gel nanofiber network, enabling large-scale integration on flexible substrates and significantly improving ion transport efficiency. The device's double-layer structure consists of: an upper ion storage layer rich in free cations and anions to simulate the presynaptic membrane, and a lower ion-gating layer containing polyanionic electrolytes to simulate the postsynaptic membrane. This design allows the device to form robust synaptic voltage responses under pulsed signal stimulation and store/transmit information through excitatory postsynaptic potentials (EPSPs), similar to biological synapses. Additionally, the electrospun fiber membrane combines flexibility and stability, avoiding the drawbacks of traditional hydrogels, such as easy evaporation and contamination.

Figure 1. Intelligent life neural system and ion neuromorphic system.

1.2 Ion Transport Mechanism and Electrical Response Characteristics
When current stimulation is applied, ions in the gel accumulate at the electrode surface to form an electric double-layer capacitor. Through the charging/discharging of this capacitor and the slow ion recovery caused by asymmetric ion transport, the memory characteristics of biological synapses are achieved. The high surface area of the fiber structure enables more efficient ion migration under current stimulation, realizing ultralow-energy information transmission.

Figure 2. Electrical response characteristics of ion-bionic synaptic devices.

1.3 Simulation of Synaptic Plasticity
Based on the hysteresis of ion transport, the device successfully simulated biological synaptic properties such as short-term plasticity, long-term plasticity, and experiential learning, laying the foundation for constructing neuromorphic computing systems.

Figure 3. Synaptic plasticity of ion-bionic synaptic devices.

1.4 Reservoir Computing System Based on Ion Synaptic Devices
A reservoir computing system was constructed based on the short-term memory and nonlinearity of ion synaptic devices. In the MNIST handwritten digit recognition task, the system achieved 88% classification accuracy by leveraging the nonlinear mapping function of ions combined with linear regression algorithms. With only 10 reservoir nodes, the accuracy still exceeded 70%, demonstrating its potential for edge learning with low resource requirements.

Figure 4. Reservoir computing system based on ion synaptic devices.

1.5 Ephaptic Coupling Effect and Cognitive Task Implementation
Inspired by the cross-regional collaboration in biological neural systems, the study further explored the spatiotemporal integration capability of ion synaptic device arrays. An array of six synaptic devices without hard-wired connections successfully simulated the ephaptic coupling effect of biological neural networks through global coupling formed by ion migration. In the n-back task, it reproduced the dynamic process of working memory with a test accuracy of 94%, showcasing the associative memory ability of brain-like systems.

Figure 5. Ion-bionic synaptic array for cognitive tasks.

1.6 Conclusion
The ion-gel nanofiber bionic synaptic device not only efficiently simulates biological synaptic functions but also achieves complex interconnections between devices through electric field coupling mechanisms, advancing the development of brain-like neural networks. This provides a new direction for the design of future brain-like intelligent devices and opens broad prospects for improving the energy efficiency and complexity of neuromorphic computing systems.

The paper lists Southern University of Science and Technology as the first affiliation, with Associate Professor Xiao Kai, Assistant Professor Liu Quanying, and Postdoctoral Fellow Wang Li (now at Nanjing Tech University since February this year) as co-corresponding authors. Master's student Chen Yuanxia is the first author. This work was supported by the National Key R&D Program, National Natural Science Foundation of China, and Guangdong Provincial Key Laboratory projects.