4.7 Article

Three-terminal vertical ferroelectric synaptic barristor enabled by HZO/ graphene heterostructure with rebound depolarization

期刊

JOURNAL OF ALLOYS AND COMPOUNDS
卷 965, 期 -, 页码 -

出版社

ELSEVIER SCIENCE SA
DOI: 10.1016/j.jallcom.2023.171247

关键词

Neuromorphic computing; Artificial synapse; Synaptic barristor; Ferroelectric HZO; Rebound depolarization

向作者/读者索取更多资源

This study introduces a novel three-terminal vertical device called ferroelectric synaptic barristor (FSB), which can function as an energy-efficient artificial synapse for neuromorphic computing. By using electrostatic gating, the FSB can simultaneously change the polarization degree and direction of the thin film and the Schottky barrier and trapping degree at the graphene interface. The FSB exhibits stable long-term potentiation/depression and has better accuracy, fast learning ability, and low energy consumption.
Ferroelectric switching devices employing doped hafnium oxide (HfO2)-based thin films have expanded their range of electronic applications from densely-packed memory to brain-inspired artificial synapses. Here, we introduce a novel and distinctive three-terminal vertical device using a heterogeneous stack of hafnium-zirconium-oxide (HZO) thin film and graphene, called ferroelectric synaptic barristor (FSB), which can func-tion as a scalable artificial synapse for energy-efficient neuromorphic computing. Electrostatic gating in the FSB can simultaneously reorient the polarization degree and the direction of HZO as well as the Schottky barrier and the trapping degree at the graphene interface. With these peculiar controlling factors, this process mimics crucial synaptic characteristics with rebound depolarization (RD), which converts an incoming inhibitory pre-spike into cell excitation. Using the RD function mediated by both ferroelectric switching and trap-mediated charge transport, the FSB exhibited a stable long-term potentiation/depression for 5000 identical pulse schemes with excellent linearity, high yield (47/48 cells), and a good state-retaining ability for 5000 s. The learning ability and energy consumption based on a single neural network were evaluated using MNIST handwritten digits and fashion patterns. The FSB with RD exhibited better accuracy (90.03% and 74.65% for handwritten digits and fashion patterns, respectively), fast and low-power learning ability than that without RD. Our FSB device can bring the advantages of conventional ferroelectric synaptic devices while mitigating their ingrained technical and operational issues.

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。

评论

主要评分

4.7
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据