Researchers at the Korea Advanced Institute of Science and Technology (KAIST) used three-dimensional imaging to map a nanoscale current pathway inside a capacitor for resistive random-access memory (RRAM) device. They found that it connected to a tiny protrusion on the bottom electrode. The finding suggests that shaping the electrode could help control where these pathways form.
KAIST (President Choongsik Bae) announced the findings on November 12. Professor Seungbum Hong led the study in collaboration with Professor Sang-Hee Ko Park’s team. Both professors are in KAIST’s Department of Materials Science and Engineering.
RRAM stores information by forming or breaking a pathway for electric current. When a narrow pathway known as a conductive filament forms in a thin, poorly conducting layer, current flows readily. When the filament breaks, little current flows. These two resistance states represent the digital values “0” and “1.”
RRAM retains information when the power is turned off, has a simple structure, and can switch quickly. Researchers are also exploring its use in AI hardware that mimics aspects of how the brain processes information.
Conductive filaments can vary in shape and location each time they form. These differences can affect how much current flows and make the voltages needed to write or erase data less predictable. Understanding how filaments form could help make switching more consistent.
To map the filament, the team removed the memory material in thin layers and measured the current across each newly exposed surface. They stacked the current maps to reconstruct the filament in three dimensions. The technique, called conductive atomic force microscopy (C-AFM) tomography, uses the same basic idea as a medical CT scan. It combines cross-sectional images to build a three-dimensional view.
In C-AFM, a sharp, conductive probe scans a surface while measuring electric current. The team used a conductive diamond probe as the device’s top electrode. During imaging, they alternated between scans that removed material and scans that mapped the current across the exposed surface.
The device consisted of a 10-nanometer (nm) layer of titanium dioxide (TiO₂) on a platinum (Pt) bottom electrode. After forming a conductive filament in the layer, the team removed an average of 0.1 nm of material per removal scan over 100 scans.
The filament was generally wider near the probe and tapered toward the bottom electrode. Just below the probe, the team observed a constriction where the filament abruptly narrowed.
The reconstructed filament connected to a small protrusion on the bottom electrode. Although the electrode surface appears flat, it has small variations in height at the nanoscale.
The team observed the same pattern at five separate sites on the film. At each site, the filament ended at a raised area of the platinum electrode.
The researchers attribute the pattern to stronger electric fields at the sharp probe tip and at protrusions on the bottom electrode. They propose that field concentration at both ends helps guide the filament as it grows.
Computer simulations showed that the electric field intensified around protrusions on the bottom electrode. This result supports the team’s explanation for why the filaments connected at those spots.
The team also examined how current flows within the filament. In titanium dioxide, oxygen ions can move out of their usual positions, leaving empty sites known as oxygen vacancies. These vacancies help form the conductive filament.
The current maps suggest that oxygen vacancies cluster more tightly around the filament’s center near the bottom electrode. The team also proposed that the narrowing near the probe reflects a balance between the formation of new vacancies and their downward drift in the electric field.
The electrode’s nanoscale texture could therefore become a design variable for resistive memory.
If engineers can control the shape and position of the protrusions, they could guide filaments to form at selected sites and make the voltages needed to write and erase data more consistent.
The team plans to test how filament formation changes with different electrode materials and surface shapes, storage-layer compositions, and operating conditions.
“Beyond observing the three-dimensional shape of conductive filaments, this study connects their shape and end positions to the actual nanostructure of the electrode,” said Professor Seungbum Hong. He added that engineering the electrode surface could help make RRAM operate more reliably.
Kunwoo Park, a student in the integrated master’s and doctoral program in the Department of Materials Science and Engineering at KAIST, was the first author. Professor Seungbum Hong was the corresponding author, and Professor Sang-Hee Ko Park was a co-author. The findings were published online in Nano Letters on August 11, 2026.
※ Paper title: Three-Dimensional Visualization of Conductive Filaments in TiO₂ via Conductive Atomic Force Microscopy Tomography, DOI: https://doi.org/10.1021/acs.nanolett.6c02423
This work was supported by a National Research Foundation of Korea (NRF) grant (RS-2026-25487626). Samsung Electronics Co., Ltd. also supported the work through an industry–academia cooperation program.