- Nucleation and growth occur simultaneously in new 'SNG model'

- Findings expected to boost low-power non-volatile memory and AI neuromorphic semiconductor performance

KAIST researchers who conducted the study: Dr. Buyantogtokh Batzorig, Professor Hong Seung-bum and doctoral candidate Kim Seung-hun. [Provided by KAIST]
KAIST researchers who conducted the study: Dr. Buyantogtokh Batzorig, Professor Hong Seung-bum and doctoral candidate Kim Seung-hun. [Provided by KAIST]

What happens at the nanometer scale in the instant a memory chip encodes a "0" or a "1"? South Korean researchers have found that the answer involves two processes unfolding at once: new data-storage regions form while existing ones simultaneously spread outward. The discovery is expected to serve as a key principle for improving the performance of next-generation low-power memory chips and AI neuromorphic semiconductors.

KAIST announced Thursday that a research team led by Professor Hong Seung-bum of the Department of Materials Science and Engineering, working jointly with a team led by Professor Cho Byung-jin of the School of Electrical Engineering and researchers at TU Dresden in Germany, has mapped the data-writing process of hafnium-zirconium oxide (HZO) — a next-generation memory material — at the nanoscale.

The research team focused on ferroelectric materials. A ferroelectric material retains its internal electrical orientation, known as polarization, even after power is cut. Because the direction of polarization can be switched to store binary data, ferroelectrics are considered a core material for non-volatile memory, which preserves data without a continuous power supply.

HZO in particular achieves ferroelectricity by mixing zirconium into hafnium oxide, which is already widely used in conventional semiconductor manufacturing. Its high compatibility with existing fabrication processes has made it a leading candidate for next-generation memory.

When voltage is applied to a ferroelectric material, nanoscale regions called "domains" — areas sharing the same polarization direction — undergo change. The academic community has traditionally explained memory operation by treating two processes separately: "nucleation," in which new domains form, and "growth," in which existing domains expand outward.

In thin films composed of countless nanoscale crystals, such as HZO, however, the two processes become intricately entangled, and existing models have struggled to fully explain how real memory devices actually operate.

The research team used high-resolution microscopy to directly observe how domains change under varying voltages, then compared those observations against the electrical characteristics of actual memory devices.

As a result, they confirmed that while data is being written, new domains continuously nucleate at multiple sites at the same time as existing domains expand into surrounding areas. Nucleation and growth do not proceed separately during data writing — they happen simultaneously.

The cover of Nano Letters, the international academic journal in which the study was published. [Provided by KAIST]
The cover of Nano Letters, the international academic journal in which the study was published. [Provided by KAIST]

The research team formalized this finding as the "Simultaneous Nucleation and Growth (SNG) model."

The study is notable for bridging nanoscale observations made under a microscope with the electrical operating characteristics of actual memory devices under a single unified principle. It enables researchers to predict not only how quickly data is written, but also where the writing process begins and how it propagates.

The team believes the SNG model can be used to optimize electrode interfaces and manufacturing process conditions, allowing precise control over domain nucleation and propagation. This, they say, opens a path toward developing faster, more stable and lower-power non-volatile memory, ferroelectric transistors and AI neuromorphic devices.

"We confirmed that when data is written to a ferroelectric material, small changes begin anew at multiple sites simultaneously while changes already underway continue to spread outward," Professor Hong said. "Having established the principle by which these two dynamics together complete the data-writing process, this will serve as a new benchmark for designing faster and more reliable next-generation memory."

The findings were selected as the cover paper of Nano Letters.


nbgkoo@heraldcorp.com