Multi-structural attractor generation and FPGA implementation of reconfigurable memristor-based dual-pathway Hopfield neural networks
Chaos, Solitons and Fractals, cilt.211, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 211
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.chaos.2026.118823
- Dergi Adı: Chaos, Solitons and Fractals
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, zbMATH
- Anahtar Kelimeler: FPGA, Hopfield neural network (HNN), Memristive, Multi-structural attractors
- Ankara Üniversitesi Adresli: Evet
Özet
Owing to their nonlinear and memory properties that are similar to those of biological synapses, memristors are extensively employed for synapse emulation. In this work, a family of four-neuron Hopfield neural networks built upon reconfigurable memristors (RM-HNN) is introduced. The N2RM-HNN model is first established by embedding the reconfigurable memristor into the synaptic connections of the sensory cortex–motor cortex pathway. By flexibly configuring the internal reconfigurable function P(φ) and the extension parameters R and L, which respectively govern the number of regulatory directions of synaptic plasticity in the positive and negative directions, multi-structural attractors with scroll numbers strictly following the R+L+2 rule can be generated by this model, and directional bias of attractors along the phase space can be realized under asymmetric parameters. Meanwhile, dynamic behaviors such as hidden chaotic attractors, multistable coexistence and amplitude decoupling control are analyzed. Subsequently, to verify the universality of the reconfigurable memristive coupling mechanism across different brain functional pathways and to reveal the dynamic differentiation characteristics between the pathways, the N3RM-HNN model is constructed by coupling the memristor to the synapses of the thalamocortical pathway under the same memristor and network architecture. Not only is the generation rule of multi-structural attractors consistent with that of the N2RM-HNN model reproduced by this model, but also the dynamic differentiation characteristics of the dual pathways under variations in synaptic coupling strength are revealed — the thalamocortical pathway is maintained in a stable chaotic state to simulate the sensory relay function, where the stable chaotic dynamics ensure reliable and undisturbed transmission of sensory information, while a flexible transition from periodic state to hyperchaotic state is exhibited by the cortical pathway to simulate motor command generation, where the rich dynamical repertoire supports the rapid integration and flexible switching required for diverse motor outputs. Synaptic plasticity of different brain functional pathways is simulated by the two models using reconfigurable memristors, and multi-structural hidden attractors ranging from 2 to 6 scrolls as well as multiple sets of coexisting attractors can be generated. Finally, the hardware feasibility of the proposed models is verified on the Field-Programmable Gate Array (FPGA) platform.