Gated recurrent unit (GRU)-based deep learning method for spectrum estimation and inverse modeling in plasmonic devices

In this research, we propose a deep learning model employing gated recurrent units (GRUs) for the transmission spectrum prediction, and computational inverse designing of all-optical plasmonic devices (AOPDs), which may a crucial role in the precise fabrication process of photonic integrated circuits (PICs). This shift from a conventional simulation-based design to a GRU-based model offers a significant advantage in terms of accuracy and precision. The technique facilitates the intricate design process by simulating the propagation of Surface Plasmon Polaritons (SPPs) within the plasmonic structures of AOPDs. The forward modeling approach presented here demonstrates a substantial improvement in computational efficiency over the finite-difference time-domain method, adeptly forecasting transmission spectra characterized through the analysis of various geometrical parameters. The inverse modeling process infers the necessary design parameters to produce specific transmission spectra, markedly expediting the design process and eclipsing the time-intensive nature of traditional optimization methods. With a prediction loss (MSE) of 0.168 and 0.9217, this research substantiates the efficacy of GRUs in streamlining the forward and inverse design processes of AOPDs for integration into PICs.