Energy and synchronization of multifunctional loop neural networks

Multifunctional thermo-photoelectric neurons are newly proposed intelligent artificial neurons with expanded functionalities. They are capable of detecting and encoding multimodal thermo-photoelectric signals. This study constructs three loop neural networks by bridging photosensitive, thermosensitive, and thermo-photoelectric neurons with linear resistors, inductive coils, and memristors, respectively. These networks are designated as Voltage, Magnetic, and Memory Coupling Functional Networks (VCFN, MACFN, and MECFN). In the adaptive VCFN model, the system's diverse energy cannot be effectively balanced by voltage coupling, resulting in an inability to achieve a stable synchronization state. MACFN and MECFN models perform energy pumping and extraction among neurons, resulting in a self-organizing behavior. This behavior spontaneously adjusts the membrane sequences of neurons within the model, ultimately achieving a stable equilibrium of the system's intrinsic field energy. The acceleration of this process can be facilitated by an increase in the coupling gain ratio. The MECFN model quickly achieves phase synchronization, and due to its adaptive mechanism, the phase synchronization state is further stabilized. This indicates memristors can serve as bridges for communication and encoding between heterogeneous functional neural networks. The effects of chaotic currents on isolated neurons and models were also considered. High-intensity chaotic currents will cause resonance phenomena in neurons under single-peaking oscillation mode, exhibiting chaotic geometric characteristics but without changing their periodic characteristics. In neurons with multi-spiking oscillation modes, chaotic currents may induce extremely narrow chaotic windows. VCFN, MACFN, and MECFN models demonstrate good robustness in spike periodic modes but are more susceptible to external chaotic current interference in chaotic modes. Appropriate intensities of chaotic current stimulation can promote synchronization in models, while excessive intensities may suppress it. This research provides insights into the synchronization processes of multifunctional neural networks and the design of multimodal and high-robustness intelligent sensor systems, providing a theoretical foundation for advancements in artificial intelligence.