Metaplasticity‐Enabled Graphene Quantum Dot Devices for Mitigating Catastrophic Forgetting in Artificial Neural Networks
Xuemeng Fan,Anzhe Chen,Zongwen Li,Zhihao Gong,Zijian Wang,Guobin Zhang,Pengtao Li,Yang Xu,Hua Wang,Changhong Wang,Xiaolei Zhu,Rong Zhao,Bin Yu,Yishu Zhang
DOI: https://doi.org/10.1002/adma.202411237
IF: 29.4
2024-12-10
Advanced Materials
Abstract:A two‐terminal metaplastic resistive memory device is developed by combining graphene quantum dots (GQDs) with iron oxide (FeOx). This research contributes to the field of neuromorphic computing by providing a scalable and reconfigurable hardware solution that effectively integrates metaplasticity into artificial synaptic devices, addressing the problem of catastrophic forgetting in current neural network architectures. The limitations of deep neural networks in continuous learning stem from oversimplifying the complexities of biological neural circuits, often neglecting the dynamic balance between memory stability and learning plasticity. In this study, artificial synaptic devices enhanced with graphene quantum dots (GQDs) that exhibit metaplasticity is introduced, a higher‐order form of synaptic plasticity that facilitates the dynamic regulation of memory and learning processes similar to those observed in biological systems. The GQDs‐assisted devices utilize interface‐mediated modifications in asymmetric conductive pathways, replicating classical synaptic plasticity mechanisms. This allows for repeatable and linearly programmable adjustments to future weight changes linked to historical weights. Incorporating metaplasticity is essential for achieving generalization within deep neural networks, which enables them to adapt more fluidly to new information while retaining previously acquired knowledge. The GQDs‐device‐based system achieved a 97% accuracy on the fourth MNIST dataset task, while consistently achieving performance levels above 94% on prior tasks. This performance substantiates the feasibility of directly transferring metaplasticity principles to deep neural networks, thereby addressing the challenges associated with catastrophic forgetting. These findings present a promising hardware solution for developing neuromorphic systems with robust and sustained learning capabilities that can effectively bridge the gap between artificial and biological neural networks.
materials science, multidisciplinary,chemistry, physical,physics, applied, condensed matter,nanoscience & nanotechnology