Li Qijun, Hu Qianlan, Zhu Shenwu, Zeng Min, Zhao Wenjie, Wu Yanqing
School of Integrated Circuits and Wuhan National High Magnetic Field Center, Huazhong University of Science and Technology, Wuhan 430074, China.
School of Integrated Circuits and Beijing Advanced Innovation Center for Integrated Circuits, Peking University, Beijing 100871, China.
Sci Adv. 2025 May 23;11(21):eadu4323. doi: 10.1126/sciadv.adu4323.
Traditional dynamic random access memory (DRAM) technology faces grand challenges in power consumption due to the constant data refresh and in density due to the physical limit for dimension scaling. Recently, two-transistor zero-capacitor (2T0C) DRAM based on amorphous indium gallium zinc oxide (IGZO) exhibits long data retention owing to its extremely low off-state leakage current. Furthermore, the low thermal budget of the IGZO channel enables the monolithic three-dimensional (3D) stacking for higher bit density beyond the planar scaling limit. However, the demonstration of 3D stacking for IGZO has been limited to 2T0C DRAM single cell. In this work, an 8 by 8 3D stacked IGZO 2T0C DRAM array was designed and manufactured with optimized electrical characteristics, exhibiting 3-bit storage with over 100-second retention time. Furthermore, a neural network model has been demonstrated to achieve an accuracy of 94.95% in image recognition, providing for promising pathway toward computing in memory to overcome the "memory wall."
传统动态随机存取存储器（DRAM）技术由于需要不断刷新数据，在功耗方面面临巨大挑战；又因尺寸缩放存在物理限制，在存储密度方面也面临挑战。最近，基于非晶铟镓锌氧化物（IGZO）的两晶体管零电容（2T0C）DRAM，因其极低的关态漏电流而具有长数据保持能力。此外，IGZO沟道的低热预算使得能够进行单片三维（3D）堆叠，以实现超出平面缩放极限的更高比特密度。然而，IGZO的3D堆叠演示仅限于2T0C DRAM单单元。在这项工作中，设计并制造了一个8×8的3D堆叠IGZO 2T0C DRAM阵列，其具有优化的电学特性，实现了3比特存储且保持时间超过100秒。此外，已证明一个神经网络模型在图像识别中达到了94.95%的准确率，为克服“内存墙”的内存计算提供了一条有前景的途径。
