Search

Memristive devices are promising candidates for synaptic memories in neuromorphic computing systems, but the insufficient reliability of the analog behavior has been a challenge. Lateral oxide scaling with bottom-up technologies such as thermal oxidation, coupled with new device architectures, can improve the reliability. However, the effect of thermal oxide growth on analog device behavior remains unclear. In this study, we fabricate a variety of resistive switching thermal oxides on tantalum electrode surfaces and develop a method for rapid full factorial electrical analysis. The analysis uses 2500 unique control parameter combinations of current limits and RESET-stop-voltages to compare device behavior across millions of modulated current–voltage sweeps. We clarify the mechanisms that shift the device behavior from abrupt towards gradual SET transitions, a desirable characteristic for emulating analog plasticity. We found that a mildly negative differential resistance and an increased internal series resistance at the metal–oxide interface contribute to the stabilization of the gradual SET transitions. These findings highlight the importance of fine-tuning the bottom-up oxide growth for improving switching performance. [ABSTRACT FROM AUTHOR]

Long-term retention is one of the major challenges concerning the reliability of redox-based resistive switching random access memories based on the valence change mechanism (VCM). The stability of the programmed state has to be ensured over several years, leaving a sufficient read window between the states, which is even more challenging at large statistics. Thus, the underlying physical mechanisms have to be understood and experimental data have to be evaluated accurately. Here, it shows that the retention behavior of the high resistive state (HRS) is more complex than that of the low resistive state and requires a different evaluation method. In this work, we experimentally investigate the retention behavior of 5M VCM devices via accelerated life testing and show the difficulties of commonly used evaluation methods in view of the HRS. Subsequently, we present a new evaluation method focusing on the standard deviation of the HRS current distribution. Hereby, an activation energy for the degradation process can be extracted, which is essential for the prediction of the devices' behavior under operating conditions. Furthermore, we reproduce the experimentally observed behavior with our 3D Kinetic Monte Carlo simulation model. We confirm the plausibility of our evaluation method and are able to connect the calculated activation energy to the migration barriers of oxygen vacancies that we implemented in the model and that we believe play a key role in the experimentally observed degradation process. [ABSTRACT FROM AUTHOR]