Dongjae Shin1,Jingxian Li1,Karsten Beckmann2,Anton Ievlev3,Yiyang Li1
University of Michigan1,State University of New York Polytechnic Institute2,Oak Ridge National Laboratory3
Dongjae Shin1,Jingxian Li1,Karsten Beckmann2,Anton Ievlev3,Yiyang Li1
University of Michigan1,State University of New York Polytechnic Institute2,Oak Ridge National Laboratory3
Resistive memory are highly promising candidates for neuromorphic computing due to their ability to retain a constant resistance state over time. Hafnium oxide-based memristors in particular have shown exceptional retention stability, exceeding 10 years at 85<sup>o</sup>C. It is believed that this retention stability is a result of slow oxygen vacancy migration; however, there exists substantial discrepancy between the experimentally measured device retention time and the characteristic oxygen diffusion time of HfO<sub>2</sub>. In this work, we measure the oxygen <sup>18</sup>O tracer diffusion of different HfO<sub>x</sub> thin films with different metal-to-oxygen ratios. We found that sub-stoichiometric hafnium oxide exhibits ~1-3 orders of magnitude lower diffusivity compared to stoichiometric HfO<sub>2</sub>. We show that the retention time of Hafnia memristors can be approximated by the diffusion time when we use this lower oxygen diffusivity of the substoichiometric oxide, resolving the question of why Hafnia memristors are able to retain state.