Abstract

The continued scaling of AI hardware increasingly relies on innovations at the device and integration levels rather than on planar CMOS alone. Monolithic three-dimensional (M3D) integration provides a versatile platform to vertically stack logic and emerging memory devices using BEOL-compatible processes, enabling high bandwidth, reduced footprint, and heterogeneous material integration on a single chip. In this presentation, I will introduce our recent progress on M3D architecture of emerging memories, including 2T0C gain cell, ferroelectric gain cell, and MRAM, integrated with RRAM compute-in-memory (CIM) chip. First, CNT/IGZO hybrid-polarity gain cell with counteractive coupling and high read current is introduced. Then the fully vertical ferroelectric gain cell co-designed with attention mechanism computation flow is demonstrated. Furthermore, we will discuss the M3D integration of RRAM CIM and BEOL magnetic tunnel junction (MTJ)/IGZO-transistor for one-shot learning task. Beyond the demonstrated devices, this talk aims to highlight M3D as an open and extensible integration framework, inviting joint efforts to advance next-generation 3D integrated AI hardware. 

Biography

Yuyan Wang, Yuyan WANG is an Associate Professor at Tsinghua University. She received her Ph.D. in Materials Science and Engineering from Tsinghua University in 2015. From 2015 to 2020, she worked as an Assistant Professor at Beihang University. In 2017, she was awarded the Humboldt Research Fellowship, supporting her postdoctoral research at the University of Regensburg and Technical University of Munich in Germany (2017-2019). Her research focuses on emerging memory technologies and related materials, including spintronic and ferroelectric materials and devices. She has authored over 60 SCI papers, including papers on Nature Nanotechnology, Nature Electronics, Nature Communications, Science Advances, Advanced Materials, Physical Review Letters, and so on.