Compatible resistive switching mechanisms in Ni/SiOx/ITO and application to neuromorphic systems

In this work, we studied the switching mechanisms of Ni/SiOx/ITO devices before and after experiencing a reversible switching. And we also investigated its application for neuromorphic computing systems. First, we checked the composition of the device and progressed the electrical measurements. The device had two operating properties that are affected by an external applied voltage, and thus, we divided it into two distinct I-V curves to experimentally investigate the features. Even if they originated from the same RRAM devices, they exhibited different electrical measurements such as a retention, threshold voltages, and the conductance ratio. The I-V curve with an abrupt switching showed a good retention time of 7000 s and a wide conductance ratio of about similar to 41. On the other hand, the other I-V curve that shows progressive operation displayed a low retention time of 5000 s and a narrow conductance ratio of about similar to 3.43. We discussed the different results that were identified on the same device and concluded that it was caused by a change of switching mechanisms induced by a reversible switching in negative polarity. The overshoot current and large fluctuation in threshold voltage were used as evidence for our discussions. After identifying the electrical features, we progressed the biological processes such as potentiation/depression, paired-pulse facilitation (PPF) and spike-timing-dependent plasticity (STDP) to implement the neural networks. (C) 2022 Elsevier B.V. All rights reserved.

Automated summary

In this work, the switching mechanisms of Ni/SiOx/ITO devices before and after reversible switching were studied, and their application in neuromorphic computing systems was investigated. Different electrical measurements and behaviors were observed in the RRAM devices, which were attributed to a change in the switching mechanisms induced by reversible switching in negative polarity. The identified electrical features were further used to implement neural networks.