摘要 Abstract 

隨著生成式 AI 快速發展，資料搬移與記憶體頻寬已成為 AI 系統效能的主要瓶頸。本演講將介紹團隊於新世代 AI 系統之高能效互連架構與異質整合研究成果，內容包含 3D 互連架構、AI 加速器設計，以及 2.5D 異質整合技術。

With the rapid advancement of generative AI, data movement and memory bandwidth have become major bottlenecks in AI system performance. This talk will introduce our recent research on energy-efficient interconnection architectures and heterogeneous integration for next-generation AI systems, including 3D interconnection architectures, AI accelerator designs, and 2.5D heterogeneous integration technologies.