Session

DescriptionThis session heralds the dawn of 'Active Memory', where the boundary between storage and computation dissolves. By embedding intelligence -- from fine-tuned LLMs to in-storage Hyperdimensional Computing -- directly into the hardware fabric, these works tackle the 'Data Movement Wall' head-on. Spanning Mixture-of-Experts (MoE) acceleration, Ethereum analytics, and MRAM-based BNNs, this session redefines memory as a cognitive engine. It marks a paradigm shift toward autonomous, energy-efficient silicon capable of processing massive AI workloads at the source, unlocking performance for next-generation intelligent systems.