POST: Photonic Swin Transformer for Automated and Efficient Prediction of PCSEL

TL;DR

This work introduces POST, a Photonic Swin Transformer that acts as a fast and accurate surrogate for Coupled-Wave Theory simulations of Photonic Crystal Surface Emitting Lasers. By processing 32×32 single-channel dielectric patterns through a four-stage Swin Transformer, POST achieves $R^2$ values of $0.909$ for $\log Q$ and $0.779$ for $SE$, while delivering up to about $5{,}000$ predictions per second, enabling rapid exploration of large design spaces including irregular hole geometries and multi-lattice configurations. The training dataset, comprising over 25{,}000 CWT-generated samples, is augmented with rotations, flips, and translations, and can be publicly released to support cross-disciplinary research. SHAP analyses in the Fourier domain reveal that POST leverages physics-prior Fourier components, validating alignment with CWT physics and offering guidance for inverse design and accelerated optimization.

Abstract

This work designs a model named POST based on the Vision Transformer (ViT) approach. Across single, double, and even triple lattices, as well as various non-circular complex hole structures, POST enables prediction of multiple optical properties of photonic crystal layers in Photonic Crystal Surface Emitting Lasers (PCSELs) with high speed and accuracy, without requiring manual intervention, which serves as a comprehensive surrogate for the optical field simulation. In the predictions of Quality Factor (Q) and Surface-emitting Efficiency (SE) for PCSEL, the R-squared values reach 0.909 and 0.779, respectively. Additionally, it achieves nearly 5,000 predictions per second, significantly lowering simulation costs. The precision and speed of POST predictions lay a solid foundation for future ultra-complex model parameter tuning involving dozens of parameters. It can also swiftly meets designers' ad-hoc requirements for evaluating photonic crystal properties. The database used for training the POST model is derived from predictions of different photonic crystal structures using the Coupled-Wave Theory (CWT) model. This dataset will be made publicly available to foster interdisciplinary research advancements in materials science and computer science.

Figures (7)

• Figure 1: Schematic overview. This work replaces PCSEL's conventional simulation model with POST neural network prediction model, achieving a qualitative leap in the speed of design characterization.
• Figure 2: Flip-Rotate-Translate pattern effects simulation.a. A randomly generated unit cell pattern of the photonic crystal. b. The first row shows eight variants of the lattice structure in subfigure a after flipping and rotation. Rows two to nine show sixty-four additional patterns generated by horizontal or vertical translation of the original structure. cd. Simulation results of Coupled-wave Theory model for the photonic crystal lattice structure shown in b after flipping, rotation, and translation. The left and right histograms show the distributions of simulated Q and SE, respectively.
• Figure 3: Distributions of data. The histograms show the distributions of logQ and $SE$ for all samples in the original dataset.
• Figure 4: POST Network Structure. Graph a shows the architecture of the Swin Transformer (SwinT) encoder. The input single-channel image is partitioned into patches, linearly embedded, then processed through four hierarchical stages to produce the final multi-dimensional representation $\mathcal{Z}_4$. Graph b displays two consecutive Swin Transformer stacks. W-MSA and SW-MSA refer to multi-head self attention modules with regular and shifted window configurations.
• Figure 5: Training dynamics and predictive performance. Graphs a and c show the training curves of the POST's predictions for $\log{Q}$ and SE, respectively. The blue line (Train $R^2$) and red line (Test $R^2$) indicate the goodness-of-fit of the model on the training set and test set as the training epochs progress. Graphs b and d respectively display scatter plots of the model's predictive performance for $\log{Q}$ and SE. The scatter points compare the model's predicted values with the true values, while the black dashed line represents the ideal fit line used to evaluate prediction accuracy.