Session

DescriptionThis section explores how next-generation hardware architectures are breaking the memory wall to accelerate large language models, computer vision, and neuromorphic computing for emerging AI systems. The first paper proposes a neuromorphic (spike-based) processor, which leverages spike-tracing bipolar-integrate-and-fire neurons. The second paper proposes an integrated image sensing and near-sensor MRAM-based processing architecture. The last four papers present cutting-edge research contributions addressing algorithm-hardware co-design for vision transformers (ViTs) and large language models (LLMs) on edge devices, pushing the boundaries of edge AI, offering new directions for efficient deployment of vision and language models in resource-constrained environments.