[19p-Z10-1~15] 3.11 Photonic structures and phenomena

[19p-Z10-8] Designing L3 nanocavity with machine learning by asymmetrically shifting the air holes

〇(B)Akari Fukuda¹, Takashi Asano², Taro Kawakatsu¹, Susumu Noda², Yasushi Takahashi¹ (1.Osaka Pref. Univ., 2.Kyoto Univ.)

Keywords:photonic crystal, Machine Learning

Q-value is an important performance index in the application of two-dimensional photonic crystal nanocavities. In recent years, machine learning has made it possible to optimize many structural parameters that have been neglected in the past, and dramatic improvements in design Q-values have been achieved. However, asymmetric structures are also interesting from the viewpoint of structural freedom. However, asymmetric structures are also interesting from the viewpoint of structural freedom, and it is not obvious whether symmetric resonator design is experimentally superior or not, since symmetry cannot be perfectly maintained in actual structures due to fabrication fluctuations. In addition, it is important to investigate whether optimization by machine learning is possible for asymmetric structures, since symmetry is lost when waveguides are added to the resonator in actual applications. In this paper, we report an attempt to learn the asymmetric structure of an L3 nanocavity and to achieve a high Q-value based on the learning.