New memory hierarchy and technologies in AI era

Physical and agentic AI systems demand computing platforms that deliver real-time intelligence under stringent power, latency, area, and cost constraints. Memory capacity and bandwidth are increasingly the bottlenecks limiting responsiveness and decision quality. As traditional SRAM and DRAM approach scaling limits at advanced nodes, innovation across the entire memory hierarchy becomes essential. This talk will explore how emerging memories can complement or replace SRAM with lower leakage, non-volatility, and denser bit cells, and advanced DRAM and 3D-stacked architectures to deliver the capacity and bandwidth needed for increasingly demanding AI workloads.