Computing Device Signatures in Resistive-Switching Memory Materials—Utilization of Machine Learning

The growth of conductive filaments in resistive-switching memory (RSM) materials depends on the geometry of conductive filaments and switching behaviors, whose complexity cannot be captured by traditional modelling approaches. Here we apply the concept of machine learning to modelling conductive-filament growth in these systems. The approach enables the computation of device signatures without knowing the conductive-filament geometries and switching behaviors, and is thus ideal for RSM thin films, doped layers and other material systems. Computation of device signatures is consistent with experimental data and discloses switching behaviors not captured by conventional models. A state-of-the-art performance on challenging device-signature learning tasks was achieved, and it provides a new method for computing device signatures based on machine learning. This intuitive approach combined with a simple tool, enables researchers with little computing experience to carry out realistic modelling.