Impact of the Crystal Phase of ZrO₂ on Charge Trapping Memtransistor as Synaptic Device for Neural Network Application

IEEE Journal of the Electron Devices Society

Wan-Hsuan Chung¹  Yu-Che Chou¹  Chao-Hsin Chien¹  Chien-Wei Tsai¹  Chin-Ya Yi¹ 

[1] Institute of Electronics, National Chiao Tung University, Hsinchu, Taiwan;

关键词: Germanium;    high-κ dielectrics;    multilayer perceptron;    neural network hardware;    synaptic device;    zirconium oxide;   

DOI  :  10.1109/JEDS.2020.2993859

来源: DOAJ

【 摘 要 】

In this work, we investigated the effects of the crystal phase of ZrO₂ on charge trapping memtransistors (CTMTs) as synaptic devices for neural network applications. The ZrO₂ deposited through thermal (t-ZrO₂) atomic layer deposition (ALD) and plasma (p-ZrO₂) ALD were analyzed using an X-ray diffractometer, which indicated that the t-ZrO₂ consisted of pure cubic phase, whereas p-ZrO₂ consisted of both cubic and tetragonal phases. Through X-ray photoelectron spectroscopy analysis, we then constructed the energy band diagram of the gate stacks. The ΔEC of tand p-ZrO₂ with respect to tunneling and blocking Al₂O₃ were 1.84 and 1.19 eV respectively. Because of the relatively large ΔEC of t-ZrO₂, the window of the flat band voltage (VFB) shift extracted from charge trapping capacitors was enlarged by 591.9 mV more than the one using p-ZrO₂ as the charge trapping layer. Retention was also improved by 10.4% after 105 s in the t-ZrO₂ case. Finally, we fabricated the CTMTs with the gate stack of the t-ZrO₂ case and demonstrated their characteristics as synaptic devices. With the optimization of pulse schemes, we reduced the nonlinear factors of depression (αd) and potentiation (αp) from -6.72 and 6.47 to 0.03 and 0.01 respectively, enlarged the ON/OFF ratio from 15.6 to 70.4 and increased the recognition accuracy from 27.6% to 86.5% simultaneously.

【 授权许可】

Unknown