KAIST researchers used multi-modal atomic force microscopy to visualize co-localized changes in conductance, oxygen-vacancy concentration, and surface potential in titanium dioxide thin films, revealing the key resistive-memory mechanisms governed by coupled ionic and electronic processes

Research scheme. Conductive atomic force microscopy (C-AFM) was used to induce resistive switching on a 10-nm-thick TiO₂ thin film by applying forming and reset operations and to map the resulting changes in local current. The same sites were then revisited to acquire electrochemical strain microscopy (ESM) and Kelvin probe force microscopy (KPFM) signals, enabling correlative analysis of spatial relationships between ionic and electronic processes associated with resistive switching.

Professor Seungbum Hong’s group at KAIST demonstrated a co-localized, multi-modal atomic force microscopy (AFM) approach that maps how current paths, oxygen-vacancy defects, and surface potential evolve at the same nanoscale sites during resistive switching, providing new insight into ReRAM operation and reliability.

As demand for next-generation artificial intelligence hardware and high-speed data storage grows, energy efficiency has become critical. One major direction is to place data storage and computation closer together to reduce the energy and time costs of data exchange. In this context, resistive random-access memory (ReRAM) has drawn attention as a promising alternative because it stores information through electrical signals that depend on a material’s resistance. However, the variability of resistance during operation—and the resulting reliability problem—remains a key obstacle. In particular, it has been difficult to directly observe, at the nanoscale, which material properties change during switching and how those changes create or rupture the current pathways.

To address this gap, the researchers used AFM, which scans a sharp probe across a surface and reconstructs local information by measuring minute forces between the tip and the specimen. Importantly, AFM is not limited to surface topography. By switching measurement modes, it can spatially visualize where currents flow (a conductance map), how oxygen-vacancy concentration changes (a defect distribution map), and how the local electrical state varies due to accumulated charge (a surface-potential map). These co-localized maps enable coordinate-matched comparison of current paths, defects, and potential, allowing correlative analysis of coupled ionic and electronic dynamics during resistive switching.

Using this capability, the team applied voltage through an AFM tip to a titanium dioxide (TiO₂) thin film, locally implementing a forming (data storage) and a reset step (data erase). After forming, regions with increased current also showed signals consistent with higher oxygen-vacancy concentration. After reset, regions with reduced current tended to show decreased oxygen-vacancy-related signals. This indicates that conductive pathways are spatially linked to the local creation and annihilation of oxygen vacancies.

The group further proposed two mechanisms that may contribute to the conductance drop after reset. First, under the reset electric field, electrons injected from the AFM tip can accumulate near the oxide surface, altering tip–sample electrostatic interactions and shifting measured AFM signals; the fact that this component relaxes relatively quickly over time supports an electronic contribution. The injected electrons may also become trapped near oxygen vacancies or remain as excess surface charge, hindering electron transport and thereby reducing conductance.

Second, oxygen-ion injections and migration can drive recombination that fills oxygen vacancies, rupturing conductive paths and producing a high-resistance state; in vacancy-rich regions created during forming, oxygen-ion motion during the subsequent reset can proceed more readily, enabling efficient path rupture and stabilizing high-resistance retention. However, the authors also noted that reliable operation requires keeping ion injections within a beneficial window, since excessive ion injections can destabilize the film.

Overall, the study provides a nanoscale, co-localized analysis of multiple properties and offers a diagnostic methodology to support the design of more reliable memory and neuromorphic devices.