Design of Multichannel Solitonic Neurons

We propose and analyze a new class of photonic neurons based on spatial solitons generated in photorefractive media. They are designed to operate entirely within the optical domain. By engineering single-node and multi-node multichannel architectures, we demonstrate the feasibility of constructing balanced, scalable, and reconfigurable structures capable of emulating neural behaviors such as symmetric signal splitting, plasticity, and dynamic adaptation. The optimization of geometric parameters—including soliton waveguides features, input distances, and incidence angles—proves crucial for ensuring the stability of solitonic propagation and the proper functioning of interaction nodes. The results lay the groundwork for the development of high-performance optical neural circuits, with potential applications in distributed signal processing, neuromorphic artificial intelligence, and reconfigurable optical memories.