Inverse design of optimal metaphotonic devices for dynamic structural colors using physics-informed neural networks

Phase change materials (PCMs) offer a versatile platform for dynamic optical applications due to their tunable refractive indices and phase-dependent properties, such as color modulation. However, designing multifunctional optical devices across a wide spectral range, like dynamic color printing, presents significant challenges, as traditional techniques are often limited to narrowband operations. In this research, we propose a novel approach to wide-band color printing, spanning the visible spectrum (400–800 nm), by leveraging Physics-Informed Neural Networks (PINNs) for inverse design. Our method addresses the broadband operation challenge by integrating Maxwell’s equations into the loss function of a deep multilayer neural network. Enhanced by residual connections and dropout regularization, the model improves in accuracy, generalization, and adherence to physical laws. Utilizing GeSe₃ and a dataset of 5,000 samples, our PINN predicts optimal material configurations, enabling dynamic color switching (e.g., between green and blue) through phase transitions. This approach significantly advances the inverse design process, emphasizing the need for novel, data and time efficient techniques to handle the complexity of broadband optical applications.