Memory chips are under pressure from every direction. Artificial intelligence workloads demand devices that can store data densely, switch almost instantaneously, and hold that information without power, all while running hot inside ever more crowded processors. A team of researchers in China now reports a step toward meeting those demands with an unlikely candidate: boron nitride, a compound better known as an inert insulating wrapper for other two-dimensional materials. In a study published in Nature Nanotechnology, the group describes the reproducible growth of four-inch wafers of single-crystal rhombohedral boron nitride, a rare stacking form of the material that behaves as a ferroelectric, and demonstrates working memory arrays built directly on top of them.
The trick behind this new class of electronics is called sliding ferroelectricity, a phenomenon that emerged from the study of atomically thin layered crystals. In conventional ferroelectrics, such as the oxides used in some sensors and emerging memories, an electric field shifts charged atoms within the crystal lattice to flip the direction of an internal polarization. In sliding ferroelectrics, the polarization is instead controlled by nudging entire atomic layers sideways relative to one another. Because the layers are held together only by weak van der Waals forces, they can slide with remarkably little resistance, which means the switching can be extremely fast and should, in principle, endure billions of cycles without the lattice damage that gradually kills conventional ferroelectric devices.
Rhombohedral boron nitride is particularly attractive for this purpose. In its common hexagonal form, hBN, the atomic layers stack in an alternating pattern that cancels out any net polarization. In the rhombohedral form, rBN, the layers stack in a repeating ABC sequence, so that the small electric dipoles formed by each boron-nitrogen pair add up in the same direction, producing a vertical polarization that can be reversed by sliding the layers. The problem has been making rBN at all. The rhombohedral phase is thermodynamically less stable than the hexagonal phase, and earlier efforts produced only small crystals or films whose stacking order drifted between regions, which is fatal for devices that depend on a uniform polarization direction.
The research team, led by Kaihui Liu of Peking University along with colleagues at several Chinese institutions, solved this problem with what they call a step-templated interfacial epitaxy strategy. They began with sapphire wafers cut along the m-plane orientation, then annealed them so that the surface reorganized into a staircase of atomic-scale steps. Onto these stepped templates they sputtered films of a nickel-boron alloy. The surface steps of the alloy act as a template: as boron nitride crystallizes out of the film, the steps force the growing layers into a single, unidirectional ABC stacking order rather than the thermodynamically favored alternating arrangement. Control experiments showed the importance of this geometry, because films grown on step-free c-plane and a-plane sapphire templates yielded predominantly the ordinary hexagonal phase, while growth at elevated temperatures also pushed the material toward the wrong polymorph.
Comprehensive structural characterization confirmed the success of the approach. Electron backscatter diffraction and low-energy electron diffraction patterns collected at randomly selected positions across the wafers showed large-scale single crystallinity, and cross-sectional electron microscopy revealed the perfect ABC stacking order maintained throughout the film thickness. Second-harmonic generation, an optical technique sensitive to the symmetry of non-centrosymmetric crystals, produced signals roughly two thousand times stronger than those of a single layer, and intensity mappings across the wafers were uniform, indicating a consistent crystal configuration from edge to edge. Crucially for any future manufacturing effort, ten different batches of wafers grown under identical conditions showed congruent quality across spectroscopic and microscopic measurements, addressing the batch-to-batch variability that has plagued attempts to industrialize two-dimensional materials.
With the wafers in hand, the researchers built ferroelectric field-effect transistors, or FeFETs, in which the sliding polarization of the rBN layer gates a semiconductor channel and encodes memory as a shift in the transistor’s switching voltage. Measurements on rBN devices with graphene channels verified the intrinsic ferroelectricity, with polarization-electric field loops showing a remnant polarization of about 1.3 microcoulombs per square centimeter and clear switching current peaks, while identical devices built with ordinary hBN showed no such hysteresis, ruling out artifacts from the device architecture itself. Piezoresponse force microscopy confirmed that reversing the applied bias reversed the local polarization, and cross-sectional electron microscopy of oppositely poled regions captured the structural signature of the layer sliding.
The performance figures are striking. The devices switched at nanosecond speeds, survived more than two billion read-write cycles without degradation, and retained their state for timescales consistent with ten-year non-volatility. Even when the transistor channels were shrunk to an ultrashort length of thirty nanometers, the devices maintained large memory windows of four volts, a parameter that determines how reliably stored bits can be distinguished, and remained stable at temperatures above 470 kelvin. The films themselves proved to be excellent insulators, with a breakdown field of roughly ten megavolts per centimeter and minuscule off-state currents, properties that matter when billions of such devices must sit side by side without leaking charge into their neighbors.
Scaling from single devices to arrays is where most exotic memory concepts stumble, but the wafer format allowed the team to take that step. They constructed FeFET arrays across the rBN wafers using molybdenum disulfide channels, achieving high integration density with on/off current ratios of around one million, large enough for robust readout in dense circuits. The thickness of the rBN layer provided an additional design knob: thicker layers produced larger memory windows, with devices using 10.6-nanometer films showing windows of 1.7 volts and those using 40.3-nanometer films reaching 4.4 volts. That tunability, combined with the thermal robustness, suggests the material could serve not only as standalone non-volatile memory but also as the basis for artificial synapses in in-memory computing architectures, where the same devices that store weights also perform the multiply-and-accumulate operations that dominate neural network workloads.
What makes the result resonate beyond the laboratory is the manufacturing story. Two-dimensional materials have dazzled physicists for two decades, yet almost none have made the leap from millimeter-scale flakes to the four-inch, six-inch, and larger wafers that fabs require. By demonstrating batch production with consistent quality, the researchers have converted sliding ferroelectricity from a curiosity assembled under tweezers into a platform compatible with wafer-scale processing. The step-templated epitaxy itself is a generalizable idea, hinting that surface geometry can be used to stabilize other metastable stacking orders in layered materials, potentially unlocking ferroelectric versions of semiconductors and metals as well. Challenges certainly remain, including integration with existing silicon processes, contact engineering, and scaling the growth to larger substrates, but the combination of nanosecond switching, billion-cycle endurance, decade-long retention, and high-temperature stability in a wafer-scale format marks a genuine milestone. If the momentum continues, the humble boron nitride layer that once served only as a passive substrate may find itself at the active heart of the memory chips that power the next generation of artificial intelligence.
Subject of Research: Wafer-scale synthesis of single-crystal rhombohedral boron nitride for sliding ferroelectric memory devices
Article Title: Single-crystal rhombohedral boron nitride wafers for integrated sliding ferroelectric memory
Article References: Qi, J., Gu, T., Tu, J., Yang, Q., Guo, Q., Guo, L., Wu, P., Chen, L., Zhang, M., Zhao, C., Tian, E., Yang, Q., Xie, K., Yang, F., Song, B., Zhang, L., Zheng, X., Lu, X., Wang, E., … Liu, K. (2026). Single-crystal rhombohedral boron nitride wafers for integrated sliding ferroelectric memory. Nature Nanotechnology. https://doi.org/10.1038/s41565-026-02280-4
Image Credits: AI Generated
DOI: 10.1038/s41565-026-02280-4
Keywords: rhombohedral boron nitride, sliding ferroelectricity, two-dimensional materials, ferroelectric memory, FeFET, epitaxy, non-volatile memory, nanosecond switching, wafer-scale synthesis, in-memory computing, sapphire template, Nature Nanotechnology
Cite Scienmag News
APA
MLA
Chicago
Denise Maddox. (September 30, 2026). Four-Inch Wafers of Sliding Ferroelectric Boron Nitride Bring Atom-Thin Memory Closer to Chips. Scienmag. https://scienmag.com/four-inch-wafers-of-sliding-ferroelectric-boron-nitride-bring-atom-thin-memory-closer-to-chips/
Denise Maddox. “Four-Inch Wafers of Sliding Ferroelectric Boron Nitride Bring Atom-Thin Memory Closer to Chips.” Scienmag, 30 September 2026, https://scienmag.com/four-inch-wafers-of-sliding-ferroelectric-boron-nitride-bring-atom-thin-memory-closer-to-chips/. Accessed 30 September 2026.
Denise Maddox. “Four-Inch Wafers of Sliding Ferroelectric Boron Nitride Bring Atom-Thin Memory Closer to Chips.” Scienmag. September 30, 2026. https://scienmag.com/four-inch-wafers-of-sliding-ferroelectric-boron-nitride-bring-atom-thin-memory-closer-to-chips/
Copy citation
Download RIS
Tags: 2D ferroelectric materials for memory devicesatom-thin memory technologydevelopment of ultra-thin ferroelectric memory arraysepitaxyFeFETferroelectric boron nitrideferroelectric memoryferroelectric switching mechanisms in layered crystalshigh-density data storage with boron nitridein-memory computingintegration of ferroelectric materials in semiconductor chipslarge-scale wafer growth of boron nitridenanosecond switchingNature Nanotechnologynon-volatile memoryrhombohedral boron nitridesapphire templatesliding ferroelectricitysliding ferroelectricity in 2D materialstwo-dimensional materialsvan der Waals forces in 2D ferroelectricswafer-scale synthesis
