Doping modulated ion hopping in tantalum oxide based resistive switching memory for linear and stable switching dynamics

Artificial intelligence (AI) has demonstrated that automated machines could eventually replace human mental labor. However, AI is only settled in restricted environments with high computing resources and power supply. Artificial neural networks are currently implemented at the software level, necessitating the constant retrieval of synaptic weight among devices. Physically composing neural networks with emerging nonvolatile memories (eNVMs) can directly map synaptic weight and accelerate AI computing. Resistive switching memory (RRAM) is a promising in-memory computing unit, but challenges regarding nonideal properties remain unsolved. In particular, nonlinear conductance update is a major issue that hinders the performance of neural networks with RRAMs. In this study, metal cation doping was introduced to tantalum oxide-based RRAM to reveal the ionic hopping behavior, which is the essence of the resistive switching phenomena. Controlled dopant concentrations and a further understanding of ionic hopping alleviated nonlinear conductance modulation up to similar to 1.46, while sustaining proper resistive switching window of similar to 10.

Automated summary

Artificial intelligence has the potential to replace human mental labor, but it currently requires high computing resources and power supply. This study introduces metal cation doping into tantalum oxide-based RRAM to reveal the ionic hopping behavior and alleviate the nonlinearity conductance modulation issue. The results show a significant improvement in the resistive switching window and conductance modulation.