Optical trapping-enhanced probes designed by a deep learning approach

Abstract

Realizing optical trapping enhancement is crucial in biomedicine, fundamental physics, and precision measurement. Taking the metamaterials with artificially engineered permittivity as photonic force probes in optical tweezers will offer unprecedented opportunities for optical trap enhancement. However, it usually involves multi-parameter optimization and requires lengthy calculations; thereby few studies remain despite decades of research on optical tweezers. Here, we introduce a deep learning (DL) model to attack this problem. The DL model can efficiently predict the maximum axial optical stiffness of Si/Si3N4 (SSN) multilayer metamaterial nanoparticles and reduce the design duration by about one order of magnitude. We experimentally demonstrate that the designed SSN nanoparticles show more than twofold and fivefold improvement in the lateral ( k x and k y ) and the axial ( k z ) optical trap stiffness on the high refractive index amorphous TiO2 microsphere. Incorporating the DL model in optical manipulation systems will expedite the design and optimization processes, providing a means for developing various photonic force probes with specialized functional behaviors.