This paper explores a machine learning (ML)-based method for the inverse design of meta-optical structures for photonic beam engineering. A data-driven approach models a grating meta-structure to create Gaussian, focused, and collimated excitation beams for lab-on-chip applications. Using feedforward deep neural networks (DNNs) and convolutional neural networks (CNNs), the model predicts meta-surface design parameters (repetition period, height, and size of scatterers) based on a desired electromagnetic field outcome. The trained neural network achieves a high correlation coefficient (0.996) in predicting diffraction profiles, offering a faster alternative to conventional optimization methods.

Robin Singh, Anu Agarwal, Brian W. Anthony