Machine Learning Inspired Design of Complex-Shaped GaN Subwavelength Grating Reflectors

Abstract

We present the machine learning inspired design of two types of GaN zero contrast subwavelength gratings (SWGs): a polynomial-shaped grating and, for reference, a conventional rectangular grating, using the differential evolution algorithm to optimize the designs for high broadband reflectivity and separately for large fabrication tolerance. Two wavelength regimes with 500 nm and 1.55 μm center wavelengths (λcenter) are investigated. Our results indicate that both polynomial and rectangular grating designs can achieve comparable stopband widths of 170 nm (Δλ) for 500 nm (Δλ/λcenter = 34%). For the 1.55 μm center wavelength, the 482 nm (Δλ/λcenter = 31%) stopband width of the polynomial grating is slightly less than that of the rectangular grating at 498 nm (Δλ/λcenter = 32%). We also demonstrate design for enhanced fabrication tolerance while placing a minimum constraint on the stopband width. Our results show the rectangular grating exhibits slightly higher fabrication tolerances for grating parameters than the polynomial-shaped grating in general. This work also outlines the technique we have adopted for the inverse design of these gratings.

Keywords

• Machine learning
• inverse design
• differential evolution algorithm
• subwavelength grating
• GaN
• VCSELs