Artificial intelligence–driven design, optimization, and control of multi-core and multi-mode optical fibers: a comprehensive review

Abstract

The adoption of space-division multiplexing (SDM) technologies like different cores and different modes in a single fiber core has been spurred by the fast expansion of global data traffic, which has driven optical communication systems toward their basic capacity limits. Although these fiber architectures significantly increase capacity, complex, high-dimensional trade-offs between inter-core crosstalk, differential mode group delay, effective mode area, bandwidth, bending loss, and fabrication tolerance make their practical implementation difficult. Large design spaces are becoming increasingly inappropriate for traditional numerical simulation and parameter-sweeping methods due to their computational demands. Powerful alternatives for fiber modeling, inverse design, and performance optimization have been made possible by recent developments in artificial intelligence (AI) algorithms, machine learning (ML) design, and neural learning networks (NLNs) architecture. This review examines cutting-edge AI-driven methods for few-mode multi-core fibers, multimode and hollow-core fiber systems, and homogeneous and heterogeneous MCFs. Particle swarm optimization, digital twin models, hybrid ML–physics frameworks, and neural network-based regressors have all shown notable computation time reductions while attaining ultra-low inter-core crosstalk, minimum value of effective mode area, controlled modal analysis behaviour, and enhanced transmission capacity.