The limitation of traditional Von Neumann architecture could be resolved by machine learning training in neuromorphic computing. However, the nonlinearity characteristic during conductance modulation in memristor severely restricts its application in neuromorphic computing. To improve the analog switching including linearity of the hafnium oxide memristors, Ti metal layer has been inserted on hafnium oxide by using the redox reaction on the interface to achieve more gradual switching. Furthermore, a new multilayer structure device is fabricated utilizing the Ti/HfO_x interface characteristic, which enhanced the electrical characteristic and the conductance modulation linearity to 98.4%, amplitude and symmetry have also been improved. The conductive mechanism of segmented growth of conductive filaments by adjusting the oxygen vacancy concentration gradient was analyzed and further characterized by XPS and TEM. The results of this paper will exalt conductance modulation linearity of hafnium oxide memristors for application to neuromorphic computing systems.

传统冯诺依曼架构的局限性可以通过神经形态计算中的机器学习训练来解决。然而，忆阻器电导调制过程中的非线性特性严重限制了其在神经形态计算中的应用。为了改善模拟开关，包括氧化铪忆阻器的线性度，钛金属层已经通过界面上的氧化还原反应插入到氧化铪上，以实现更渐进的开关。此外，利用 Ti/HfO _x制造了一种新的多层结构器件。界面特性，使电特性和电导调制线性度提高到98.4%，幅度和对称性也得到了改善。分析了通过调节氧空位浓度梯度实现导电细丝分段生长的导电机制，并通过XPS和TEM进一步表征。本文的结果将提升氧化铪忆阻器在神经形态计算系统中的电导调制线性度。