Bayesian optimization of photonic nanojets generated by multilayer dielectric structures

We employ a Bayesian optimization (BO) approach coupled with electromagnetic solver to optimize the photonic nanojets (PNJs) respectively generated by a 2D cuboid and a 3D concentric cylinder. The operating wavelength is selected atλ = 632.8 nm. By optimizing a five-layer cuboid, we achieved a narrow PNJ with waist at the full width at half-maximum (FWHM) about w ~ 0.2λ and an ultra-long PNJ with a beam length 123λ , respectively. Whereas by optimizing 3D concentric cylinder, we achieved a narrow PNJ with waist at FWHM about w ~ 0.22λand ultra-long PNJ with a beam length 78λ , respectively. Furthermore, we explore the physics behind the PNJ phenomenon, such as the fundamentals of extreme PNJ generation and the relation between the objective intensity profile and the refractive index gradients across the structure. The proposed approach presents an alternative approach to the design of various photonic nanostructures and may provide conventionally-inaccessible physical insight to their functionalities.