Session

DescriptionRecent advances in artificial intelligence are increasingly driven by specialized hardware architectures tailored to the computational structure of emerging models. This session presents nine papers addressing architectural techniques for improving the efficiency of modern AI workloads, with particular emphasis on large language models (LLMs), diffusion models, and graph neural networks (GNNs). The contributions span both algorithm–architecture co-design and microarchitectural innovations. Collectively, the papers in this session illustrate the rapid evolution of AI hardware toward workload-specific acceleration, tighter algorithm–hardware co-design, and mechanisms that balance performance, energy efficiency, and deployability across both datacenter and edge environments.