Abstract: On-chip micro-nanophotonic devices serve as core components for realizing high-performance optical interconnects, optical computing, and optical sensing. Their level of advancement directly determines the information processing capacity and energy efficiency of integrated optical systems. However, conventional structural design relies on physical intuition and parametric sweeping, suffering from limited design space and low efficiency when optimizing structures with high degrees of freedom, multiple objectives, and complex topologies. In recent years, inverse design methods, which start from target optical performance and leverage intelligent algorithms to automatically search for optimal structures in high-dimensional design spaces, have significantly expanded the performance boundaries of devices. This paper systematically reviews the latest research progress in inverse design of on-chip micro-nanophotonic devices. It first introduces iterative optimization algorithms and deep-learning-driven design paradigms. It then provides a detailed summary of typical applications and performance breakthroughs of these methods in optical interconnects, mode control devices, polarization control devices, wavelength control devices, and augmented-reality near-eye display devices. Finally, it discusses current challenges and prospects future development directions. This review aims to provide a systematic reference for the intelligent design of on-chip micro-nanophotonic devices.