A novel hybrid implementation of memristive neuron networks

Memristive neural networks (MNNs) are neuromorphic circuits that provide an effective framework for investigating the complex behaviors exhibited by biological neural systems. In general, physical validation of MNN dynamics relies on purely analog or digital circuits. Here, we introduce a novel hybrid analog–digital architecture that exploits the strengths of both domains. Specifically, we implement two coupled Hindmarsh–Rose (HR) neurons interconnected through a second-order memristor. The proposed architecture operates via an ADC-DAC closed-loop interface, integrating analog neuron circuits to capture continuous-time neuronal dynamics, and a digital memristor implementation on a digital signal processor (DSP) board, enabling flexible tuning and reconfigurable memristor parameters. The dynamics of the coupled systems are studied using bifurcation and Lyapunov exponent diagrams, revealing chaotic behavior, transient chaos, coexisting attractors, and phase synchronization. Although digital processing causes some delay, the experimental attractors from the hybrid platform closely match the numerical simulations. This agreement is quantitatively confirmed through parametric, spectral, and topological metrics, including the correlation of the power spectral densities, the phase-space overlap, and the correlation dimension, demonstrating the value of this architecture as a scalable platform for neuromorphic applications.