Wide-angle emission grating using a supervised genetic optimization for LIDAR integration

Diffraction gratings are among the essential devices for spreading and shaping beams in a wide range of optoelectronic and photonic sensors and fiber optic communications. This has triggered an interest towards inverse design and optimization of the parameters using gradient-based optimization, heuristic algorithms, and machine learning models. Approaches based on complex models (such as deep neural networks) provide enhanced robustness and rely on a huge amount of data to achieve accuracy. However, the generation of these data and multi-parameter optimization can be laborious and time-consuming with the Finite Difference Time Domain (FDTD) simulation. We present an optimization approach to obtain a single grating antenna with wide-angle emission for a photonic integrated flash Light Detection And Ranging (LIDAR) system. The device is simulated using a silicon nitride material operating at a wavelength of 905 nm. Our method relies on a supervised, data-centric approach in combination with a genetic algorithm optimization. Given an optimization and several parameters, we evaluate the variables based on their correlation with the merit function and reduce the search region consequently. This approach allows faster convergence and provides a flat field of view of (56.95°, 92.82°) at Full Width Half Maximum (FWHM) in one dimension simulation.