Session

DescriptionThe growing diversity of AI models, from large language models (LLMs) to graph analytics and neural rendering, places increasing pressure on hardware architectures to support heterogeneous computational patterns efficiently. This session presents nine papers exploring architectural mechanisms that improve performance, efficiency, and flexibility across a wide range of emerging AI workloads. These papers demonstrate ongoing efforts to design flexible yet efficient AI hardware architectures that address the growing heterogeneity of modern machine learning workloads while maintaining high performance and energy efficiency across datacenter, edge, and mobile platforms.