KAIST Unlocks Dual-Process Model for Ferroelectric Memory at Nanoscale

A team at KAIST has mapped out the precise mechanics of how information gets written into ferroelectric memory at scales so small they measure in billionths of a meter. The discovery centers on hafnium zirconium oxide, or HZO, a material that combines hafnium oxide—already standard in semiconductor manufacturing—with zirconium oxide. What makes HZO valuable is its compatibility with existing chip-making processes and its ability to hold an electrical charge without power, a property called nonvolatility that makes it attractive for next-generation memory devices.

Ferroelectric materials work by storing information in the orientation of electrical polarization—the alignment of positive and negative charges within the material. Flip that orientation, and you flip a bit from 0 to 1. When voltage is applied, tiny regions called domains, each with uniform polarization, begin to shift. New domains form while existing ones expand into surrounding space, and this cascade of change reverses the overall polarization direction. That's how data gets written. But understanding exactly how this happens at the nanoscale has proven difficult. Researchers have traditionally focused on either nucleation—the birth of new domains—or growth, the expansion of existing ones. Neither model alone captured what actually happens in hafnia-based thin films like HZO, where switching occurs across many nanometer-sized crystal grains that interact with grain boundaries and other structural features in ways that complicate the picture.

Professor Seungbum Hong's team at KAIST, working with collaborators from KAIST's School of Electrical Engineering and researchers at NaMLab in Dresden, Germany, took a different approach. Using high-resolution piezoresponse force microscopy, they directly observed how nanometer-scale domains changed as voltage increased. One nanometer is one billionth of a meter. They watched the microscopic dance in real time, then compared what they saw under magnification with the electrical behavior of actual memory devices. The results showed something both processes were happening simultaneously. New domains kept forming while existing domains kept spreading. Information wasn't written by one mechanism or the other—it was written by both at once.

To capture this dual process, the team developed what they call a simultaneous nucleation and growth model, or SNG. The model accounts for both the continuing formation of new domains and the cumulative expansion of those already present, bringing the two processes together in a single mathematical framework. Critically, the researchers linked the nanoscale changes they observed under the microscope directly to the electrical behavior of real memory devices. This connection matters because it means the model doesn't just describe what happens at the atomic level—it explains the actual performance characteristics of working memory: how fast it switches, where the writing process begins, and how it spreads across the material.

The implications ripple outward. Once researchers understand how domains form and spread, they can adjust a material's structure or modify manufacturing processes to make information writing faster and more uniform. The work could accelerate development of more reliable, energy-efficient nonvolatile memory and also inform the design of neuromorphic devices—hardware that mimics aspects of how brains process and store information—for low-power artificial intelligence applications. The research was published in Nano Letters on June 15, 2026, and selected for the journal's cover. Batzorig Buyantogtokh, a postdoctoral researcher in KAIST's Department of Materials Science and Engineering, led the work. The research was funded by South Korea's National Research Foundation through the Ministry of Science and ICT.