LAUSR

Slow light waveguides in photonic crystals are engineered using a conventional method or a deep learning (DL) method, which is data-intensive and suffers from data inconsistency, and both methods result in overlong computation time with low efficiency. In this paper, we overcome these problems by inversely optimizing the dispersion band of a photonic moiré lattice waveguide using automatic differentiation (AD). The AD framework allows the creation of a definite target band to which a selected band is optimized, and a mean square error (MSE) as an objective function between the selected and the target bands is used to efficiently compute gradients using the autograd backend of the AD library. Using a limited-memory Broyden-Fletcher-Goldfarb-Shanno minimizer algorithm, the optimization converges to the target band, with the lowest MSE value of 9.844×10-7, and a waveguide that produces the exact target band is obtained. The optimized structure supports a slow light mode with a group index of 35.3, a bandwidth of 110 nm, and a normalized-delay-bandwidth-product of 0.805, which is a 140.9% and 178.9% significant improvement if compared to conventional and DL optimization methods, respectively. The waveguide could be utilized in slow light devices for buffering.