神经形态工程学
符号
Padé逼近
人工神经网络
算法
计算机科学
材料科学
电气工程
人工智能
数学
算术
工程类
应用数学
作者
Yifan Yang,Zichong Zhang,P Jiang,Rui Su,Menghua Huang,Tsau Young Lin,Xiangshui Miao,Xingsheng Wang
标识
DOI:10.1109/led.2023.3336396
摘要
The Hf0.5Zr0.5O2-based ferroelectric tunnel junction (FTJ) devices are fabricated and further enhanced by a proposed “thermal rewake-up” (TR) operation at different high temperatures. The measured electrical characteristics show that TR operation can significantly enhance the ferroelectric remnant polarization (Pr) up to $29.1~\mu \text{C}$ /cm2, and can boost the FTJ resistance ON/OFF ratio from 9.8 to 34.9, meantime achieving $10^{{5}}$ cycles endurance and $10^{{4}}$ s data retention. Furthermore, the neural network adopting TR FTJ devices as electronic synapses exhibits improved accuracy from 95.76% to 98.81% and enhanced tolerance to noise in the task of handwritten digit recognition. Hence, this letter demonstrates a feasible approach to improve the switching behavior of FTJ devices in applications of binary/multi-level memories and neuromorphic computing.
科研通智能强力驱动
Strongly Powered by AbleSci AI