QR-Code Structure for Beam Shaping and Polarization Control via Machine Learning for Chip-Integrated Quantum Applications

While many research fields are driven to higher levels of performance with photonic integrated circuits, the forward design of these systems faces certain limitations. This paper presents a machine learning based approach to design binary metasurfaces to facilitate beam shaping, angle, and polarization state. We implement Lumerical FDTD and Non-Dominated Sorting Genetic Algorithm III (NSGA-III) to optimize the topology of the outcoupling structure composed of subwavelength pixels, allowing a higher degree of control over the emitted light field. The generated pattern is shown to maintain the desired beam shape and angle while modulating the right/left circular and linear polarization states, allowing scalability of the design for different wavelengths without large distortion of the field properties and promising low fabrication complexity.