Physics-Informed Convolutional Framework for Predicting Quantum Dot Scattering Patterns

This study presents a method based on energy distribution patterns to generate kernel filters of various sizes for image convolution, aimed at improving the accuracy and similarity of quantum dot (QD) scattering simulation results. First, optical simulations of different photomask patterns are conducted using LightTools (LT) to obtain QD scattering energy distribution data. The energy distribution is then divided into 3×3, 5×5, 7×7, and 9×9 matrices, with the energy ratios of each region calculated to generate the corresponding kernel filters. These kernel filters are applied to the original photomask image, and convolution operations are performed to simulate the blurring effect caused by QD scattering. The resulting images are compared with those from LT simulations. Structural similarity index (SSIM), mean squared error (MSE), and standard deviation are used to quantify the differences, between the kernel filter method and the actual simulation results. Additionally, the effect of different convolution iterations and kernel sizes on image similarity is analyzed to identify the optimal convergence region. The results demonstrate that the proposed kernel filter method effectively simulates QD scattering energy distribution, improving image similarity evaluation metrics. This approach offers a feasible and accurate solution, which can be further applied in optical design and image processing to enhance the precision and efficiency of optical simulations.