KAIST researchers have directly imaged how hafnium zirconium oxide (HZO) ferroelectric memory writes bits, and the answer contradicts the standard picture: new nanoscale domains keep forming at the same time as existing domains expand, rather than one process following the other. The team, led by Professor Seungbum Hong of KAIST's Department of Materials Science and Engineering, published the work in Nano Letters (Batzorig Buyantogtokh et al., DOI: 10.1021/acs.nanolett.6c01580).

The material at the center of the study matters commercially. HZO blends hafnium oxide — already ubiquitous in semiconductor fabs as a gate dielectric — with zirconium oxide, and it survives conventional CMOS process flows. That compatibility makes hafnia-based ferroelectrics a leading candidate for nonvolatile memory that keeps data without power, and more recently for neuromorphic devices aimed at low-power AI hardware.

What did the researchers actually see?

Ferroelectrics store 0s and 1s by flipping polarization: when a voltage reverses the alignment of positive and negative charges inside the material, the new orientation persists after the voltage disappears. The flip happens through domains — nanometer-scale regions of uniform polarization. In HZO thin films, switching plays out across many nanometer-sized crystal grains, and grain boundaries complicate the picture.

Existing models split into two camps. Nucleation-centric models treat switching as the creation of new domains; growth-centric models treat it as the expansion of domains already in place. Because both processes interact with grain structure in hafnia films, neither camp alone matched device behavior.

Hong's group, working with Professor Byung Jin Cho's team at KAIST's School of Electrical Engineering and researchers at NaMLab/TU Dresden in Germany, used high-resolution piezoresponse force microscopy (PFM) to watch domains evolve as applied voltage increased, then compared the nanoscale movies against the electrical switching curves of complete devices. The result: nucleation and propagation proceed simultaneously. Bits are written by two concurrent mechanisms — fresh changes starting at multiple sites while changes already underway spread outward.

"When information is written in a ferroelectric material, which retains its electrical state even without power, small changes begin at multiple sites while those already underway spread into the surrounding regions," Hong said.

Why a unified model matters for memory design

The team formalized the observation in a simultaneous nucleation and growth (SNG) model, which folds both processes into one framework. Crucially, the model links what the microscope sees — where domains appear and how fast they advance — to the current-voltage behavior engineers measure on real capacitors.

That linkage gives chip designers a predictive handle rather than a phenomenological fit. Because the model explains where writing begins and how it spreads, materials teams can tune film structure and fabrication conditions to make switching faster and more uniform across a wafer — the two parameters that determine write speed and endurance margins in any ferroelectric memory product.

"This study reveals how these two processes work together to write information, providing a new basis for designing faster and more reliable next-generation memory," Hong said.

The findings also point toward more energy-efficient nonvolatile memory and neuromorphic devices, which mimic aspects of the brain's combined processing and storage and target low-power AI hardware.

The work remains a materials-science result, not a product announcement: no timeline for commercial HZO memory follows from it. But by grounding switching dynamics in directly observed domain behavior, the KAIST-led collaboration gives fabs and device engineers a validated model to optimize — a prerequisite for turning hafnia ferroelectrics from a promising lab material into qualified silicon.