Inverse design of photonic crystal laser using deep learning to control wavelength and Q-Factor

This paper presents a novel deep-learning-based inverse design framework for photonic crystal lasers. The proposed approach employs neural networks to directly learn the complex and highly nonlinear mapping between the geometric parameters of photonic crystal structures and key laser characteristics, including resonant wavelength and quality factor. Unlike conventional design methods that rely on computationally expensive and iterative electromagnetic simulations, the proposed framework enables rapid generation of optimized designs with significantly reduced computational cost and high accuracy. In this method, the neural network first produces a continuous representation of the photonic crystal geometry, which is subsequently converted into a binary, fabrication-ready structure through a thresholding process. Numerical results demonstrate that the optimized designs closely match the target wavelength and Q-factor while exhibiting physically meaningful features such as effective optical confinement and favorable geometric symmetry. Training convergence analysis confirms the stability and robustness of the learning process, and symmetry evaluation reveals that the network implicitly favors structures with balanced field distributions, even without explicitly imposed constraints. These findings highlight the ability of deep learning not only to optimize device performance but also to capture essential physical principles governing photonic crystal lasers.