上海科技大学知识管理系统

Linear Weighting Method(LWM)1 and Multi-Objective Particle Swarm Optimization (MOPSO)2 Method are

two popular approaches for addressing multi-objective problems. LWM suffers from strong subjectivity, making

it challenging to ensure that each objective is adequately optimized. Meanwhile, MOPSO demands substantial

computational resources, is prone to getting trapped in local optima, and its convergence accuracy as well as the

uniformity of solution set distribution remain areas for improvement. In the context of inverse design for photonic

devices, the problem of dual objectives is particularly prevalent. By employing a two-step inverse design method,3

we efficiently designed a 50% : 50% directional coupler tailored for operation in the O-band within a relatively

short timeframe, requiring only two iterations of a conventional Particle Swarm Optimization algorithm(PSO).4

This coupler boasts an exceptionally low Insertion Loss(IL) of 0.034dB and a minute Power Splitting Ratio Error

(PSRE) of 1.19 × 10−6 . Furthermore, it features a compact footprint of 138um2 , 5 underscoring its practicality

and efficiency.