Repeated finite-element method (FEM) simulations of ring-core photonic crystal fibers are accurate but computationally expensive, particularly during large parameter sweeps and optimization. This study presents a sustainable surrogate-modeling framework that couples FEM-generated data with a feed-forward neural network for rapid prediction of the effective mode area, \({A}_{\text{e}\text{f}\text{f}}\) . The model was developed using 1950 samples described by the ring-core radius, operating wavelength, and effective refractive index. Its predictive behavior was evaluated through training-loss, prediction, residual, and statistical analyses, while computational sustainability was assessed in terms of runtime, electricity consumption, and estimated CO2 emissions. A single FEM case required approximately 6 h, 720 Wh, and 0.36 kg of CO2, whereas one surrogate inference required approximately 6 s, 0.2 Wh, and 0.0001 kg of CO2. These results represent a reduction of approximately three orders of magnitude in runtime and nearly four orders of magnitude in estimated carbon emissions. The principal innovation of this work is the combined evaluation of predictive performance and environmental cost within a single RC-PCF design workflow. The proposed surrogate therefore provides a rapid tool for preliminary screening, parameter exploration, and optimization, while FEM remains necessary for final design verification.
Graphical abstract
