Abstract: As processor design continues to grow in complexity, development cycles are increasingly compressed, and application demands evolve rapidly, processor design automation is shifting from auxiliary generation for localized implementation tasks toward a closed-loop intelligent design paradigm that spans architecture design, RTL generation, verification and debugging, and optimization feedback. Existing research has progressed from early structured approaches based on rules, generators, and hardware description languages to more intelligent methods that incorporate machine learning, reinforcement learning, large language models, and agents, thereby advancing automation across multiple stages of processor design, including specification understanding, solution generation, code implementation, verification and repair, and performance evaluation. This paper systematically reviews the development trajectory of processor design automation, summarizing the representative ideas and capability boundaries of three stages: the rule- and programming-language-based paradigm, the small-model and reinforcement-learning-based paradigm, and the large-model and agent-based paradigm. It further surveys research progress along two main threads: automatic generation from architecture to RTL code, and closed-loop design, covering generators, intermediate representations (IRs), composable transformations, multimodal specification alignment, EDA agents, and analytical, data-driven, and structured hybrid evaluation models. Finally, it identifies key challenges, including high-quality domain-specific data, multimodal specification modeling, tool-feedback-driven closed-loop design, early-stage architectural evaluation, and cross-stage collaborative optimization, with the aim of providing a useful reference for future research on processor design automation.