Transmittance Prediction and Inverse Design of Microring Resonator Channel Dropping Filters with Deep Learning

Abstract

We propose a method to use deep learning to achieve transmittance prediction and inverse design of microring resonator channel dropping filters. We transform the transmittance prediction and inverse design into model training problems, which learn and approximate the intrinsic interactions from the geometric parameter space to transmittance space and the transmittance space to geometric parameter space. The test loss and mean square error for the transmittance prediction case are 3.94888×10-2 and 4.68901×10-3, respectively; the test loss and mean square error for the inverse design case are 7.27015×10-3 and 4.0029×10-4, respectively. The numerical results suggest that the models developed by deep learning can make an efficient prediction of the transmittance and achieve excellent performance of the inverse design for microring resonator channel dropping filters, validating the effectiveness and feasibility of the approach we propose. With generalization ability within the given design space, the well-trained models can produce fast and accurate results without the need for time-consuming numerical calculations or case-by-case design.