神经形态工程学
材料科学
铁电性
可操作性
终端(电信)
范德瓦尔斯力
光电子学
计算机科学
人工神经网络
人工智能
量子力学
分子
物理
电信
软件工程
电介质
作者
Fei Xiao,Dongxin Tan,Liang Tian,Zheng‐Dong Luo,Qiyu Yang,Xuetao Gan,Dawei Zhang,Marin Alexe,Zhufei Chu,Junpeng Zhang,Fei Xue,Wenchao Chen,Yinshui Xia,Yan Liu,Yue Hao,Genquan Han
标识
DOI:10.1002/adfm.202523298
摘要
Abstract Ferroelectric‐based non‐volatile memories (NVMs), which can merge logic and memory functionalities within a single device, represent a promising solution for future in‐memory computing architecture. Although various types of ferroelectric‐based NVM are demonstrated, a device concept integrating compact structure, reconfigurable operation and superior performance that can meet the demands of the emerging in‐memory computing remains to be explored. Here, a new type of device termed ferroelectric field‐effect memtransistor (FeFEMT) is reported, which features a multifunctional device operation scheme using the proposed multi‐terminal operability. In this device, multi‐level non‐volatile conductance switching based on the ferroelectric polarisation switching and the multidomain structure can be achieved with either gate‐ or drain‐voltage control. The device operates with control voltage inputs from both drain and gate terminals, enabling a single‐transistor structure two‐input Boolean logic‐in‐memory gate. Furthermore, the drain‐voltage controlled multidomain structure enables the FeFEMT to function as a two‐terminal synaptic transistor, showing excellent analogue conductance switching with over 7‐bit conductance states. The demonstrated synaptic properties consequently facilitate the simulation of implementation of a convolutional neural network, offering a superior inference performance. These findings highlight the prospects of the FeFEMTs as a suitable device technology for diverse and high‐efficiency in‐memory computing hardware platform.
科研通智能强力驱动
Strongly Powered by AbleSci AI