Spatially-Resolved Thermometry of Filamentary Nanoscale Hot Spots in TiO(2) Resistive Random Access Memories to Address Device Variability

[Image: see text] Resistive random access memories (RRAM), based on the formation and rupture of conductive nanoscale filaments, have attracted increased attention for application in neuromorphic and in-memory computing. However, this technology is, in part, limited by its variability, which originates from the stochastic formation and extreme heating of its nanoscale filaments. In this study, we used scanning thermal microscopy (SThM) to assess the effect of filament-induced heat spreading on the surface of metal oxide RRAMs with different device designs. We evaluate the variability of TiO(2) RRAM devices with area sizes of 2 × 2 and 5 × 5 μm(2). Electrical characterization shows that the variability indicated by the standard deviation of the forming voltage is ∼2 times larger for 5 × 5 μm(2) devices than for the 2 × 2 μm(2) ones. Further knowledge on the reason for this variability is gained through the SThM thermal maps. These maps show that for 2 × 2 μm(2) devices the formation of one filament, i.e., hot spot at the device surface, happens reliably at the same location, while the filament location varies for the 5 × 5 μm(2) devices. The thermal information, combined with the electrical, interfacial, and geometric characteristics of the device, provides additional insights into the operation and variability of RRAMs. This work suggests thermal engineering and characterization routes to optimize the efficiency and reliability of these devices.