High Bandwidth Memory and Computing-in-memory Technology

Cai Yimao, Dean of the School of Integrated Circuits, Boya Distinguished Professor, doctoral supervisor, and Executive Director of the Innovation Center for Advanced Integrated Circuits at Peking University. He is a recipient of the National Outstanding Youth Science Fund, and the Special Government Allowance of the State Council. He has long been engaged in research on memory and AI chips. He has directed a number of national-level projects including the National Key R&D Program and Key Programs of the National Natural Science Foundation of China. He developed novel RRAM memory and in-memory computing intelligent chips with practical applications. He has published more than 150 papers in top-tier conferences and journals including IEDM, VLSI, and Nature Electronics, with over 100 authorized Chinese patents.

As large language models become more widely used, optimizing their deployment for AI applications on edge devices has drawn growing attention. Edge devices impose strict requirements on processing speed, energy use, and computational stability. Traditional von Neumann architectures, which mainly rely on DRAM and HBM, struggle to meet these demands in edge settings. This talk will explore a compute-in-memory strategy based on 3D VNAND Flash. It aims to overcome the VNAND Flash memory wall and enable more efficient deployment of large language models on edge devices.