KAIST Maps Memory Chip Pathways as GIST Advances AI Interfaces and POSTECH Opens New Program

Tiny electrode bumps offer a clue to steadier memory

KAIST researchers have reconstructed electrical pathways inside a titanium dioxide layer just 10 nanometers thick, finding that tiny bumps on a platinum electrode help guide where those pathways connect. The team removed about 0.1 nanometer of material at a time and combined 100 current maps to reveal structures hidden inside a candidate technology for future memory chips.

The findings lead a series of developments announced by South Korean science universities on November 12. Gwangju Institute of Science and Technology (GIST) reported awards for research on generating visual effects and reproducing touch, alongside a workshop with the Food and Agriculture Organization of the United Nations (FAO) on digital models for farming. POSTECH announced a graduate program connecting artificial intelligence with human cognition, ethics and public policy.

Each development addresses a different practical problem: making electronic memory behave consistently, giving people more direct ways to create and feel digital content, testing agricultural technology against real conditions, and educating researchers who can connect AI engineering with its effects on society.

At KAIST, materials science and engineering professors Hong Seung-bum and Park Sang-hee led the collaborating groups. Their central finding was repeated across five electrical pathways formed at different locations: the pathways tended to end on small protrusions rather than flat areas of the lower electrode.

Why a memory chip needs a predictable current path

Resistive random access memory, or RRAM, stores information by changing how easily electricity passes through a thin layer of material. A narrow conducting path, called a conductive filament, allows current to flow. Breaking or weakening that connection produces a state with higher electrical resistance. The two states can represent the digital values zero and one.

RRAM retains information when power is switched off. Its simple structure and rapid switching also make it attractive for research into storage devices and computing hardware inspired by networks of nerve cells in the brain.

A filament that forms differently each time creates a practical difficulty. Its location and shape affect the voltage needed to write or erase information, so variation inside the material can make the device harder to operate consistently. KAIST's experiment examined how the surface beneath that material helps shape the conducting path.

Building a three dimensional picture, one slice at a time

The KAIST team studied titanium dioxide resting on a platinum electrode. One nanometer is one billionth of a meter, placing the entire 10 nanometer layer far below what a person can inspect with an ordinary microscope.

Researchers used a conductive atomic force microscope, an instrument with a fine tip that can probe a surface and measure electrical current. A hard, single crystal diamond tip removed material under greater pressure. Scans using lower pressure then measured where current flowed at the newly exposed depth.

Repeating that process 100 times produced 100 maps. Each removal step was about one hundredth of the layer's original thickness: 0.1 nanometer out of 10 nanometers. Stacking the maps reconstructed the filament's volume, much as combining slices can reveal the inside of an object. Here, the researchers physically wore away the sample as they measured it.

The reconstructed filament resembled a funnel, broad near the upper probe and narrower toward the platinum electrode below. A constricted region, described as a neck, appeared about 0.8 nanometer beneath the probe. That detail gave the team another feature to explain alongside the filament's endpoint.

How the electrode surface steers growth

The five independently formed filaments tended to connect to protrusions on the platinum surface. An electrode that appears smooth at a larger scale can still have peaks and dips only nanometers high, and those features affect the electrical conditions immediately around them.

The researchers linked the pattern to concentrations of the electric field, the force environment that moves electrical charges. Pointed features can concentrate that field, rather like the tip of a lightning rod. Calculations using the finite difference method, which estimates a field across many small points, supported this explanation.

The simulations showed stronger fields around both the upper probe and protrusions on the lower electrode. The team interpreted those concentrations as guides influencing the filament's growth and final connection.

Titanium dioxide can contain oxygen vacancies, places where oxygen atoms are missing. These vacancies help make conducting paths possible. From the current measurements, the researchers inferred that vacancies gathered toward the filament's center near the lower electrode. They also proposed that differences between vacancy creation and movement could explain the narrow neck near the top.

What comes next for the KAIST research

Hong Seung-bum, a KAIST professor of materials science and engineering and the paper's corresponding author, described the link between the filament and the electrode as the study's central contribution:

"This study goes beyond simply observing the three-dimensional shape of conductive filaments. Its significance lies in connecting their shape and end positions to the actual nanostructure of the electrode," Hong said.

The next research task is to test whether deliberately designing electrode surfaces can make filaments form at chosen locations in working memory devices. Measurements of memory lifetime, writing speed and operating consistency will be needed to establish the practical gains. The team plans to examine different electrode materials, surface shapes, oxide compositions and operating conditions.

Park Kun-woo, a student in KAIST's integrated master's and doctoral program, was the first author. The paper, titled Three-Dimensional Visualization of Conductive Filaments in TiO2 via Conductive Atomic Force Microscopy Tomography, appeared online in Nano Letters on August 11.

Support came from the Ministry of Science and ICT and the National Research Foundation of Korea's core research program, as well as a strategic research collaboration with Samsung Electronics' Device Solutions Division. The work puts electrode surface geometry among the variables researchers can investigate when trying to reduce variation in RRAM operation.

GIST turns written ideas into editable visual effects

GIST's Human-Centered Intelligent Systems Laboratory, led by School of Artificial Intelligence professor Kim Seungjun, presented four papers at ACM UIST 2026 in Detroit from November 2 to 5. The conference focuses on technologies that shape how people interact with computers.

The laboratory received a Best Paper Award for SpeLL, a system that creates a draft of a three dimensional visual effect from a user's written description. Its output can then be modified and edited with generative AI.

That editable draft is the useful step between describing an idea and refining its appearance. Rather than ending with a fixed result, the process gives the user material to revise. A written request becomes a starting point for further creative decisions.

SpeLL addresses a different part of the computing experience from KAIST's memory experiment. The KAIST researchers studied physical structures governing current inside a material; GIST's system changes how a person communicates a desired visual result to software.

Touch becomes another channel for digital interaction

Two papers produced jointly by GIST researchers and MIT examined haptic interfaces for virtual reality and everyday objects. Haptics means communicating through touch, using physical sensations as part of an interaction with a computer or digital environment.

GIST also received the Participants' Choice Best Poster Award for Surf-Roll, a technology designed to convey sensations of slippery or sticky surfaces to the fingertips. Those properties concern how a surface feels when touched or moved across, rather than simply what it looks like.

The projects extend the laboratory's work beyond visual output. SpeLL helps a user shape an effect through language, while the haptic studies investigate how an interface can deliver information through the body. For a person using virtual reality, that makes touch another part of the experience alongside images.

GIST and FAO examine digital models for agriculture

On November 11, GIST hosted an Expert Workshop for Designing Agricultural Digital Twins with FAO and Sunchon National University. More than 20 experts from South Korea and abroad met at GIST College Building A to discuss AI, agriculture, food supply chains and international cooperation.

A digital twin is a computer representation connected to a real object or process through data. In agriculture and food supply chains, the aim is to represent actual conditions closely enough to help examine operations and possible changes.

Researchers from GIST and Sunchon National University, together with representatives of domestic companies, presented development work and technology demonstrations. A panel then discussed possible applications, barriers to adoption and priorities for projects.

The participating institutions plan to define joint research topics, select areas for priority testing and assess practical use in agricultural settings step by step. Their immediate work centers on choosing what to validate and where. That gives the collaboration a concrete sequence between a technical demonstration and use in the field.

POSTECH connects AI engineering with people and policy

POSTECH has established the Human-centered AI & Society graduate program, known as HAIS, within its Division of Humanities and Social Sciences. The first students will enter in the first semester of the 2027 academic year.

Faculty studying AI engineering will participate alongside colleagues in philosophy and ethics, economics and policy, society and culture, and human cognition. The program brings technical development and questions about human behavior and institutions into the same educational structure.

Its two tracks are Human & Cognition and Society & Governance. Four specialized modules cover media, culture and narrative; mind, brain and cognition; applied economics and policy; and AI ethics and institutional frameworks.

The curriculum creates room to study both how people think and how rules and public decisions shape AI use. Those subjects connect directly to the practical choices raised by the other university projects: how people direct creative software, experience digital environments and decide where agricultural AI should first be tested.

POSTECH scheduled an admissions information session with online and in person participation for November 17. Applications for the first intake are scheduled for December 1 through 11.

Dates from publication to the first student intake

The announcements bring together completed research, conference recognition, a planning workshop and an upcoming admissions process. The sequence stretches from the KAIST paper's publication to POSTECH's first intake in 2027.

• August 11: KAIST's filament imaging paper was published online in Nano Letters.
• November 2 to 5, 2026: ACM UIST took place in Detroit, where GIST presented its research and received two awards.
• November 11: GIST, FAO and Sunchon National University held the agricultural digital twin workshop.
• November 12: The universities announced their research developments and program plans.
• November 17: POSTECH scheduled its HAIS admissions information session.
• December 1 to 11: POSTECH scheduled applications for the first HAIS intake.
• First semester of 2027: The new graduate program will welcome its first students.

Key Points

• KAIST reconstructed current pathways in a 10 nanometer memory material using 100 maps taken at successive depths.
• Five filaments tended to connect to tiny protrusions on a platinum electrode, a pattern supported by electric field simulations.
• Further experiments will test surface designs and measure their effects on working memory devices.
• GIST received awards for SpeLL, which generates editable visual effects, and Surf-Roll, which conveys surface sensations to fingertips.
• GIST, FAO and Sunchon National University are selecting research and testing priorities for agricultural digital twins.
• POSTECH's HAIS graduate program will begin in 2027, with applications scheduled for December 1 to 11.