The scalability limitations of mainstream blockchain systems like Bitcoin and Ethereum stem from their reliance on serialized transaction execution and complete ledger maintenance on full nodes, resulting in performance bottlenecks and substantial operational overhead. While sharding has emerged as a promising solution to enhance throughput and scalability through distributed data management, it introduces critical challenges in node coordination and system reliability. In this paper, we provide a comparative analysis of prominent sharding architectures, examining their respective strengths and limitations. The study delves into both intra-shard and cross-shard protocol designs, evaluating their impact on system performance and scalability. We explore recent advancements in shard reorganization, migration protocols, and shard balancing mechanisms. The discussion emphasizes the need for a holistic approach incorporating dynamic sharding, multi-level consensus, and efficient inter-shard coordination. Finally, we outline future research directions for sharding architectures, addressing emerging challenges and opportunities in this evolving field.