Fixed-Attention Mechanism for Deep-Learning-Assisted Design of High-Degree-of-Freedom 3D Metamaterials

Abstract

The traditional design approaches for high-degree-of-freedom metamaterials have been computationally intensive and, in many cases, even intractable due to the vast design space. In this work, we introduce a novel fixed-attention mechanism into a deep learning framework to address the computational challenges of metamaterial design. We consider a 3D plasmonic structure composed of gold nanorods characterized by geometric parameters and demonstrate that a Long Short-Term Memory network with a fixed-attention mechanism can improve the prediction accuracy by 48.09% compared to networks without attention. Additionally, we successfully apply this framework for the inverse design of plasmonic metamaterials. Our approach significantly reduces computational costs, opening the door for efficient real-time optimization of complex nanostructures.