Ferroelectric Undoped HfOx Capacitor With Symmetric Synaptic for Neural Network Accelerator

Jun-Dao Luo, Yu-Ying Lai, Kuo-Yu Hsiang, Chia-Feng Wu, Yun-Tien Yeh, Hao-Tung Chung, Yi-Shao Li, Kai-Chi Chuang, Wei-Shuo Li, Chun-Yu Liao, Pin-Guang Chen, Kuan-Neng Chen, Min-Hung Lee, Huang-Chung Cheng

Research output: Contribution to journalArticlepeer-review

8 Scopus citations

Abstract

A plasma-based ferroelectric undoped HfOx capacitor is fabricated using plasma-enhanced atomic layer deposition (PE-ALD), and the multilevel characteristics of synaptic behavior are investigated. A gradual ferroelectric phase transition in the material is confirmed by grazing incidence X-ray diffraction (GI-XRD) for different O-2 plasma periods. The metal-ferroelectric-metal (MFM) capacitor presents excellent remnant polarization (P-r) up to 13 mu C/cm(2) and a switching endurance of more than 10(8) cycles. Superior small potentiation/depression nonlinearity (alpha(p)/alpha(d) = -0.08 /-1.66 for voltage modulation stimulation) for synaptic training and almost symmetry training curve ( vertical bar alpha(p) - alpha(d)vertical bar) are obtained. High stability under multilevel operation with five consecutive cycles of alternating potentiation and depression is exhibited without significant polarization variation for each training step (pulse).

Original languageAmerican English
Pages (from-to)1374-1377
Number of pages4
JournalIEEE Transactions on Electron Devices
Volume68
Issue number3
DOIs
StatePublished - Mar 2021

Keywords

  • Ferroelectric
  • multilevel
  • synaptic training
  • undoped HfOx

Fingerprint

Dive into the research topics of 'Ferroelectric Undoped HfOx Capacitor With Symmetric Synaptic for Neural Network Accelerator'. Together they form a unique fingerprint.

Cite this