Collaborative Research by Assistant Professor Ge Dongjiao of the Faculty Published in National Science Review: Rippled Graphene Pores Achieve a New Breakthrough in Neuromorphic Computing


Recently, a cross-disciplinary collaborative research outcome co-authored by Dr. Ge Dongjiao, Assistant Professor at the Faculty of Data Science, City University of Macau, as co-first author and co-corresponding author, was officially published in the top international academic journal National Science Review (NSR). The study is titled “Rippled graphene pores as fluidic memristive devices with synaptic and neuromorphic functionalities.” Notably, when the preprint of this work was previously released on arXiv, it had already attracted widespread attention from academia and was reported and reposted by professional science and technology media such as USA Graphene.

Breaking Conventional Perceptions: Micrometer Pores Can Also Possess “Memory”

Fluidic memristors use ion transport in nanopores to emulate information storage and processing in biological neural systems, making them a frontier hotspot in the field of brain-inspired computing. However, conventional theory holds that nanoconfinement is a prerequisite for achieving ionic memory effects, which are usually obtained by restricting pore diameters to the nanometer or even angstrom scale. Such extreme spatial confinement not only makes device fabrication difficult and costly, but also readily leads to pore clogging and poor stability.

To address this pain point, a joint research team comprising the University of Macau, City University of Macau, Tsinghua University, and other institutions proposed a new design approach. Under normal circumstances, micrometer-scale pores are generally believed to exhibit only resistive ionic transport characteristics. However, the team found that if the smooth edges of solid-state micropores are modified with closely stacked “rippled graphene,” they can exhibit a significant ionic memory effect. The study reveals that this occurs because ions are strongly spatially confined at the edges of rippled graphene, causing their kinetic processes to slow down. This discovery breaks the bottleneck of traditional nanoconfinement and demonstrates that efficient ionic memory can also be achieved by regulating the nanomorphology of pore walls, rather than merely shrinking pore diameters.

Data Science Empowerment: High-Precision Brain-Inspired Computing and Neural Signal Analysis

In this study, Dr. Ge Dongjiao was mainly responsible for the core work on machine learning modeling and neuromorphic computing applications. Using the long-term memory and synaptic plasticity of the memristive devices, Dr. Ge constructed a temporal spike encoding and recognition framework for image data (using the MNIST and CIFAR-10 datasets as examples), demonstrating from a data-driven perspective the high fidelity of the devices in long-duration, large-scale event-stream information storage and transmission. In addition, using the short-term memory and plasticity of the devices, as well as their excellent long-term operational stability, Dr. Ge further constructed a collaborative analysis system combining physical reservoir computing based on multi-device parallel circuits and a multi-scale convolutional neural network (Multi-scale CNN). This system was trained and online-decoded directly based on real experimental conductance data, overcoming the performance degradation bottleneck from simulation to reality (Sim-to-Real) in existing studies, and achieving high-precision real-time identification of various neuronal firing patterns and interneuronal synchronization states.

Future Outlook: Scalably Integrated Ionic Brain-Inspired Chips

This study not only reveals the important regulatory role of the nanomorphology of pore walls in ion transport, but also demonstrates the great potential of the device in scalable manufacturing and ionic circuit integration. Because the preparation and transfer processes of graphene films are already quite mature, and device fabrication is simple and easy with a success rate of over 90%, this rippled graphene pore array is expected to enable wafer-scale manufacturing and be further assembled into fluidic crossbar arrays for more complex parallel information processing. This work provides a new degree of freedom for the design of fluidic memristors, greatly promotes the development of the field of ionic neuromorphic computing, and opens a new path for the future development of brain-inspired intelligent devices highly compatible with biological systems. The publication of this research outcome fully demonstrates the international influence of the Faculty of Data Science at City University of Macau in interdisciplinary research and its strong research strength in the integration of artificial intelligence and frontier materials science.

Original paper link: https://doi.org/10.1093/nsr/nwag704