Free-Form Plasmonic Meta-Optics via Machine Learning

Advanced neural networks and optimization algorithms have enabled a paradigm shift in the discovery and design of structured photonic materials and devices. In this talk we report a diverse set of inversely designed meta-optical structures, devices, and systems for wavefront control, imaging, computing, and nonlinearity.<br/><br/>Flat optics, represented by metamaterials and metasurfaces, foresees a promising route to ultra-compact optical devices. Conventional designs of such meta-structures start with a certain structure as the prototype, followed by extensive parametric sweeps to accommodate the requirements of phase and amplitude of the emerging light. Regardless of how computation-consuming the process is, a predefined structure can hardly realize the independent control over polarization, frequency, and spatial channels, which hinders the potential of metasurfaces to be multifunctional. Besides, achieving complicated and multiple functions calls for designing meta-systems with multiple cascading layers of metasurfaces, which introduces exponential complexity. We have developed a series of deep-learning enabled generative frameworks for the inverse design of plasmonic structures in response to on-demand optical properties, with extended case studies and experimental demonstrations. Moreover, we further present a hybrid deep learning framework for designing multilayer meta-systems with multifunctional capabilities, as well as nonlinear metasurfaces for the generation of new spectral components and active control of light waves.<br/><br/>Metasurfaces composed of meta-molecules with spatially variant building blocks, such as gradient metasurfaces, are drawing substantial attention due to their unconventional controllability of the amplitude, phase, and frequency of light. However, the intricate mechanisms and the large degrees of freedom of the multi-element systems impede an effective strategy for the design and optimization of meta-molecules. We propose a hybrid artificial intelligence-based framework consolidating compositional pattern-producing networks and cooperative coevolution to resolve the inverse design of meta-molecules in metasurfaces. The efficacy and reliability of the design strategy are confirmed through experimental validations. This framework reveals a promising candidate approach to expedite the design of free-form, large-scale metasurfaces in a systematic manner.<br/><br/>We further introduce a deep learning framework for the discovery and design of a multilayer multifunctional meta-system, which is too complicated to be accomplished through conventional design processes. As solid illustrations, we report three examples designed by our framework: a polarization-multiplexed dual-functional beam generator, a second order differentiator for all-optical computing, and a space-polarization-wavelength multiplexed hologram. These functions are barely achievable by single-layer metasurfaces, and the devices are hardly approachable by conventional design means. The designed devices are constituted of arbitrary patterns, which indicates the extremely high degrees of freedom of the structures involved.<br/><br/>The machine learning methods developed here is applicable to the inverse design of other photonic components and systems, including photonic crystals, chip-scale silicon devices, quantum-optical devices, and nonlinear optical materials. This design methodology is also significant to other disciplines of natural sciences, such as the design of nano materials, searching for new topological insulators, planning of chemical syntheses, prediction of protein structures, and many more.