Abstract: This survey systematically reviews the progress in the paradigm evolution and interdisciplinary applications of Domain-Specific Architectures (DSAs) in the field of custom computing. Addressing the limitations of general-purpose computing architectures in energy efficiency, real-time performance, and task-specific performance, this research focuses on how hardware-software co-designed DSAs break through traditional computational bottlenecks through paradigm iterations. By analyzing architectures such as ASICs, FPGAs, and neuromorphic chips over the past decade and quantitatively comparing their performance gains across different application scenarios, we identify three major trends in the evolution of DSAs: enhanced dynamic reconfigurability of hardware, deep integration of cross-layer optimization (circuit-algorithm-application), and widespread adoption of the Domain-Driven Design (DDD) methodology. In interdisciplinary applications, DSAs have significantly advanced fields such as biomedical, autonomous driving, and edge intelligence (inference on resource-constrained devices), addressing challenges related to high concurrency and real-time requirements. Current challenges lie in issues such as the lack of design automation toolchains, insufficient cross-platform portability, and high verification costs. Future research needs to integrate architectural innovations with interdisciplinary knowledge, develop standardized design frameworks, and explore the potential of DSAs in emerging scenarios such as quantum-photonics-classical hybrid computing and brain-computer interfaces. This survey provides theoretical and practical references for research on the evolution of domain-specific computing ecosystems.