Doping engineering in tantalum oxide-based RRAM with enhanced resistive switching behaviors and synaptic features for neuromorphic applications

Oxide-based memristors are promising candidates for nonvolatile memory and neuromorphic computing applications. However, their performance is often hindered by challenges such as switching window, poor reliability and gradual conductance modulation. In this study, Ti-doped TaOx memristors are precisely engineered via co-sputtering to enhance the resistive switching (RS) characteristics. The oxygen vacancy content in TaOx layer can be regulated by controlling the Ti dopant concentrations via the sputtering power, thereby adjusting the performance of the device. The optimized Pt/TaOx:Ti/TiOy/Ti memristor with 8 % Ti-dopant exhibits a stable bipolar switching behavior with significantly enhanced switching window (>103), compared with the undoped counterpart (102). This device also exhibits excellent continuous conductance modulation characteristics, while the long-term potentiation/depression (LTP/LTD) synaptic characteristics have been mimicked with a great nonlinearity of 1.12/1.01, as well as a symmetricity of 43.56. These improved LTP/LTD characteristics lead to a high accuracy of 94.62 % in synaptic neural network-based image recognition tasks, much better than that of 92.45 % in the undoped device. These results highlight the precise elements doping to advance TaOx-based memristors as a promising solution for neuromorphic hardware systems.