Fast prediction of photonic crystal microcavity frequency based on multilayer feedforward neural network

Abstract

Photonic crystal microcavities play an important role in modern integrated optical devices. However, traditional electromagnetic simulations based on the Plane Wave Expansion (PWE) method or Finite-Difference Time-Domain (FDTD) method consume significant computational resources and time, which severely restricts the efficiency of reverse design for device parameters. This paper proposes a deep learning surrogate model based on a Multilayer Feedforward Neural Network (MLP) for the fast prediction of the normalized defect state frequency of a two-dimensional triangular lattice silicon-based point defect microcavity. By scanning the parameter space of the background dielectric pillar radius and the defect pillar radius, we constructed a dataset containing 1041 sets of high-quality physical samples. The experimental results show that the model exhibits extremely high prediction accuracy on the test set, with a Mean Absolute Error (MAE) of less than 0.002. Furthermore, for a batch of 1000 samples, it reduces the total prediction time from approximately 20 hours using traditional numerical simulations to less than 0.1 seconds, achieving an overall computational acceleration of over 700,000 times. This research provides an efficient and reliable computational paradigm for the rapid and intelligent design of optoelectronic devices.