Quantitative, Dynamic TaOx Memristor/Resistive Random Access Memory Model
Quantitative, Dynamic TaOx Memristor/Resistive Random Access Memory Model
复制标题
DOI：
10.1021/acsaelm.9b00792
复制
发表时间：
2020-03-24
影响因子：
4.7
通讯作者：
Lu, Wei D.
中科院分区：
文献类型：
--
作者：
Lee, Seung Hwan;Moon, John;Lu, Wei D.
氧化物基忆阻器是一种双端器件，其电阻可以通过施加刺激的历史来调制。忆阻器作为存储器（电阻式随机存取存储器）和神经形态计算应用的突触装置已被广泛研究。了解忆阻器的内部动态对于持续的器件优化和大规模实现至关重要。然而，从初始成形过程开始，能够以自一致的方式定量描述动态电阻开关（RS，例如，设置/复位循环）行为的模型仍然缺失。在这项工作中，我们提出了一个Ta2O5/TaOx器件模型，该模型可以可靠地预测成形和重复设置和重置周期中的所有关键RS属性。我们的模型表明，形成过程源于电场聚焦和初始不均匀氧空位（V-O）缺陷分布的局部加热效应。该模型可以定量捕获广泛的器件行为，包括在设置/复位周期中V-O分布的循环、多电平存储和两种不同的灯丝生长过程。特别是，发现具有低编程电流的体型掺杂效应可以产生具有大动态范围的线性电导变化，这对于神经形态计算应用是非常理想的。仿真结果还与1R和1T1R结构的直流和脉冲实验测量结果进行了比较，结果吻合良好。
Oxide-based memristors are two-terminal devices whose resistance can be modulated by the history of applied stimulation. Memristors have been extensively studied as memory (as resistive random access memory) and synaptic devices for neuromorphic computing applications. Understanding the internal dynamics of memristors is essential for continued device optimization and large-scale implementation. However, a model that can quantitatively describe the dynamic resistive switching (RS, e.g., set/reset cycling) behavior in a self-consistent manner, starting from the initial forming process, is still missing. In this work, we present a Ta2O5/TaOx device model that can reliably predict all key RS properties during forming and repeated set and reset cycles. Our model revealed that the forming process originates from electric field focusing and localized heating effects from the initial nonuniform oxygen vacancy (V-O) defect distribution. A broad range of device behaviors, including cycling of the V-O distribution during set/reset cycles, multilevel storage, and two different filament growth processes, can be quantitatively captured by the model. In particular, a bulk-type doping effect with low programming current was found to produce linear conductance changes with a large dynamic range that can be highly desirable for neuromorphic computing applications. The simulation results were also compared with experimental dc and pulse measurements in 1R and 1T1R structures and showed excellent agreements.