Ultrawide-FOV full-color waveguides free of rainbow and fishbone artifacts via physics-constrained generative metagrating design
Mengguang Wang, Yong Li, Huihui Li, Fei Wu, Zeqing Yu, Huai Xia, Qiangbo Zhang, Changwei Zhang, Yiyang Liu, Huaze Xie, Chang Wang, Zhenrong ZhengAugmented reality (AR) displays are gaining prominence in consumer and industrial applications due to two-dimensional pupil-expanding waveguides, which achieve the essential dual requirements of compact form factors and large eyebox through innovative dual-axis beam replication technology. Nevertheless, these systems face an intrinsic trilemma compromising three critical parameters—efficiency, angular uniformity, and chromatic dispersion—with all three limitations becoming particularly pronounced at wide angles (>50°) across the visible spectrum, manifesting as fishbone artifacts (periodic non-uniformity patterns) and rainbow effects. To overcome these fundamental limitations, we present a physics-constrained generative adversarial network (PC-GAN) that systematically resolves the trilemma by incorporating two key physical principles: dilated pupil restraint ratio (DPRR) for uniformity control and periodic constraint theory (PCT) for dispersion management. This framework enables high-throughput discovery of metagrating designs while ensuring high-quality imaging performance in single-layer waveguides across the full visible spectrum and ultrawide field of view. Notably, the PC-GAN demonstrates its unique capability by autonomously discovering optimized snowflake-like metagrating (SLMG) geometries - fractal structures that provide breakthrough solutions for pupil expansion uniformity through their hierarchical light manipulation properties. Experimental realization of PC-GAN-optimized metagratings in disparate material systems (polymer and SiC) consistently delivered wide-angle, full-color operation in single-layer waveguides, while eliminating the characteristic fishbone artifacts and rainbow effects of conventional diffractive designs. This physics-constrained generative framework unlocks transformative potential for complex photonic systems beyond AR, enabling advanced designs ranging from ultrathin VR displays to multi-physics-optimized quantum optical devices.