Reliable Multistate RRAM Devices for Reconfigurable CAM and IMC Applications

Reliable Multistate RRAM Devices for Reconfigurable CAM and IMC Applications

This work presents a reliable multistate RRAM device based on a Cu/Ta2O5/WO<inline-formula> <tex-math notation="LaTeX">${}_{\text {3-x}}$ </tex-math></inline-formula>/Pt structure, utilizing fully CMOS-compatible materials. The device demonstrates four distinct resi...

Full description

Saved in:
Bibliographic Details
Main Authors: Shengpeng Xing, Zijian Wang, Zhen Wang, Pengtao Li, Xuemeng Fan, Ziyang Zhang, Guobin Zhang, Jianhao Kan, Qi Luo, Shuai Zhong, Yishu Zhang
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of the Electron Devices Society
Subjects:
RRAM
multistate
reconfigurable memory
content-addressable memory
in-memory computing
Online Access:https://ieeexplore.ieee.org/document/10969850/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This work presents a reliable multistate RRAM device based on a Cu/Ta2O5/WO<inline-formula> <tex-math notation="LaTeX">${}_{\text {3-x}}$ </tex-math></inline-formula>/Pt structure, utilizing fully CMOS-compatible materials. The device demonstrates four distinct resistive states under varying switching voltages, achieving a swift response time of 25 ns and an on/off ratio exceeding <inline-formula> <tex-math notation="LaTeX">$10{^{{4}}}$ </tex-math></inline-formula>. Additionally, it demonstrates a robust data retention time exceeding <inline-formula> <tex-math notation="LaTeX">$10^{6}$ </tex-math></inline-formula> seconds and endures more than <inline-formula> <tex-math notation="LaTeX">$10^{4}$ </tex-math></inline-formula> pulses in endurance tests. Statistical analysis conducted over 100 cycles across ten devices reveals consistent resistance characteristics, with variations maintained below 10%. Leveraging these advantages, the RRAM devices were integrated with MOS transistors to construct a 4T2R unit-based array, enabling reconfigurable applications such as analog voltage-based content-addressable memory (CAM) and in-memory computing (IMC) accelerators. Notably, the proposed solution reduces energy consumption by over 20% in CAM applications and significantly enhances energy efficiency for fingerprint recognition tasks through convolution operations, achieving more than three times the energy efficiency compared to conventional GPU and CPU systems while maintaining an accuracy of 98%.
ISSN:2168-6734

Similar Items