Next Generation Memory Solutions for AI Era: From Datacenter to Edge Device

"Next Generation Memory Solutions for AI Era: From Datacenter to Edge Device"

As the AI landscape evolves from Generative AI to Agentic AI, system requirements are shifting. Model size growth is slowing, but token usage is increasing rapidly, driving new demands for cost-efficient LLM deployment. Meeting these demands requires memory solutions that deliver both capacity and performance. This talk examines how memory architectures from datacenters to edge devices must adapt, focusing on high-bandwidth memory (HBM), Processing-in-Memory (PIM), CXL-based expansion, and NAND flash–based innovations. By aligning these technologies with the workload patterns of LLMs, we outline pathways toward scalable and efficient infrastructure for the Agentic AI era.