期刊
APPLIED MATERIALS TODAY
卷 23, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.apmt.2021.101024
关键词
Synaptic devices; In-Sn-O memtransistors; Steep slope; Precise weight update
资金
- National Natural Science Foundation of China [61975241, 51673214, 52073031]
- National Key Research and Development Program of China [2017YFA0206600, 2016YFA0202703]
- Fundamental Research Funds for the Central Universities [E0EG6801x2]
- Beijing Nova Program [Z191100001119047]
- Hundred Talents Program of the Chinese Academy of Sciences
A multi-gate In-SnO memtransistor with steep-slope design has been demonstrated to simulate synaptic functions and artificial neural networks, with an efficient algorithm developed for weight update control. Image recognition training using the Modified National Institute of Standards and Technology (MNIST) dataset achieved a recognition accuracy of up to 90%, showing promise for developing neuromorphic computing.
Multi-gate architectures in synaptic transistor are promising to implement the modulation on transport properties of channel and electrical performances of device. Here, we demonstrate a steep-slope In-SnO memtransistor with multi-gate design to simulate the synaptic functions with readily programmable plasticity. The working mechanism of field-driven modulating oxygen vacancies in In-Sn-O channel has been elaborately explored. Furthermore, artificial neutral networks (ANNs) are simulated by constructing the In-Sn-O memtransistors in a crossbar array, and 3528 statistical data points for long-term potentiation are gathered from the memtransistors to explore the regularity of the conductance state. An efficient algorithm is developed to precisely control the weight update, which is used to construct an XOR gate function. The simulated ANNs for image recognition training with the Modified National Institute of Standards and Technology (MNIST) dataset have also been achieved. The recognition accuracy of the simulation can reach as high as 90%. The proposed weight update method provides a new strategy for developing neuromorphic computing. (c) 2021 Elsevier Ltd. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据