Abstract: The structure design is the core of micro-nanophotonic devices and optical systems. Many artificially designed photonic structures, such as metamaterials, photonic crystals, and plasmonic nanostructures, have been widely used in high-speed visible communication, high-sensitivity sensing, and efficient energy harvesting and conversion. However, standard design methods in this field are based on simplified physical analytical model and rule-based numerical simulation method, which is a trial-and-error method, low efficiency and likely to miss the optimal design parameters. Therefore, rapidly acquiring the potential correlation between design parameters and spectral response information is the key for realizing the efficient design of photonic devices. Besides, deep learning (DL) has been developed rapidly in fields such as language recognition, machine vision, and natural language processing in the past few years. The unique advantage of DL lies in its data-driven algorithm, which allows models for discovering useful information from massive amounts of data automatically and provides a new route to solve the aforementioned design problems of photonic structures. This review starts from different application scenarios of micro-nano photonics structure design, introduces the application scope and selection basis of various DL models in the field of photonics design, and summarizes and looks forward to future opportunities and challenges in this field.