| 초록 |
Thin-film optical diodes at visible frequency have difficulties in simultaneously achieving both high transmission efficiency and high isolation due to the lack of materials. In this work, we use a machine learning model, Factorization Machine (FM), to design the optimal thin-film structure of visible optical diode with high transmission and isolation. We use the Fourier modal method (i.e., rigorous coupled-wave analysis, RCWA) to produce data, where it includes the structural information and associated figure-of-merit (e.g., transmission coefficient or isolation factor) of optical diodes. Based on the data, we train the FM and let the trained machine find the global optimized configuration to maximize a figure-of-merit. The RCWA finally confirms the correctness of found optimized structures. As a result, we find that the optimized thin-film metal/dielectric structures with a thickness of 130 nm on a glass substrate can have a transmission efficiency of 77 % of forward-direction and 0.8 % of backward-direction, leading to the isolation factor of ~ 19. |
