Answers are grounded in this paper and cite their sources.
Photonic crystal microcavity parameters directly determine their band characteristics and optical responses, while conventional forward design methods often involve repeated numerical simulations and high computational costs when solving parameter inversion problems. To address this issue, this paper proposes an artificial neural network-based method for the inversion of photonic crystal microcavity structural parameters, using photonic band features as the input and the microcavity radius as the output. The plane-wave expansion method is employed to generate photonic band data corresponding to different microcavity radii. Each sample consists of 50 TM bands at 12 k-points, forming a 600-dimensional input feature vector. A feedforward neural network trained with the Levenberg–Marquardt algorithm is then used to learn the nonlinear mapping between band features and microcavity radius. Numerical results show that the model achieves its best validation performance at the 6th epoch, with a validation MSE of 0.023263, and training is terminated at the 12th epoch due to the validation stopping criterion. The prediction results on independent test samples show that the network can effectively capture the variation trend of the microcavity radius. These results demonstrate the feasibility of using photonic band information for microcavity structural parameter inversion and provide a basis for data-driven inverse design of photonic crystal structures.
No takes yet. Share an insight, caveat, or question.
Jingxuan Xu (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: