Evaluación de algoritmos de optimización de primer orden para la optimización topológica robusta de dispositivos nanofotónicos

Abstract

Silicon photonics is an emerging area with constant growth in the last decades. Photonic devices show potential applications to improve computing systems, telecommunications and other areas. Nevertheless, it is still a challenge to integrate a great number of fundamental photonic devices in a chip with small area and low loss. In this work we designed two fundamental photonic devices: (i) bend and (ii) 2-channel wavelengthdemultiplexer (WDM). The designs were done on a 2µm × 2µm area following a robust topology optimization based strategy. We evaluated and comparated five first-order optimization algorithms: (i) Limitedmemory Broyden–Fletcher–Goldfarb–Shanno with boundaries (L-BFGS-B), (ii) Method of Moving asymptotes (MMA), (iii) Covariance Matrix Adapatation Evolution Strategy (G-CMA-ES), (iv) Gradient Particle Swarm Optimization (G-PSO) and (v) Gradient Genetic Algorithm (G-GA). The last three algorithms are variations of their more standard versions (CMA-ES, PSO and GA) where the computation of the gradient is included to guide the optimization process. The best optimized designs show: (i) transmission greater than 90 % and robustness to over/under-etching, (ii) a gray percentage of less than 2 % and (iii) their performance is broadband consistent with smooth changes (1500-1600 nm bend and 1250-1600 nm WDM) even after deleting non-convex regions. These results are promising for (i) the integration of WDM devices in an lower area than state of the art (<2.8µm × 2.8µm) and (ii) the design of bends with lower los than intuitive-traditional designs of 1µm radius.

Date of Award	2022
Original language	Spanish (Peru)
Supervisor	Jorge Gonzalez Reaño (Asesor)