Designing metal-dielectric nanoantenna for unidirectional scattering via Bayesian optimization

Recently, using various intelligent approaches to achieve the efficient inverse design of photonic nanostructures with predefined and appropriate functionalities has attracted considerable attention. We propose a method to design subwavelength metal-dielectric nanoantennas and optimize the scattering directionality using a Bayesian optimization approach. The nanoantennas consisted of three gold disks separated by two dielectric layers. The geometrical parameters were optimized in an intelligent and fully automatic process. We showed that with the aid of the machine learning method, strong forward scattering or backward scattering at a specific wavelength could be efficiently achieved. We further showed that unidirectional scattering in opposite directions at two separate wavelengths can be designed. Moreover, it is possible to exchange the forward and backward directionality at two target wavelengths. The multipole decomposition approach was applied to analyze the multipole moments of the scattering field up to the third order. In the optimized unidirectional nanoantennas the electric and magnetic dipole moments satisfied the Kerker or anti-Kerker conditions at the wavelengths of interest. Our results demonstrated the possibility of automatically designing nanoantennas for specific applications via a machine learning scheme.