A new study points to a hidden chemical divide inside hafnia ferroelectric capacitors: some microscopic regions became rough and briefly conductive under electrical stress, while other selected regions remained almost unchanged. The contrast offers a possible explanation for why nominally similar devices can show very different leakage, switching and breakdown behavior.
The work used thin films of yttrium-doped hafnium oxide, or YHO, grown on an LSMO buffer and a SrTiO3 substrate. The researchers combined surface imaging during electrical bias with conductive atomic-force microscopy, which maps current at the surface, piezoresponse-force microscopy, which detects local polarization switching, and electron microscopy with spectroscopy of the interface.
A sharp split between leaky and ferroelectric regions
In a selected leaky region, surface roughness rose from about 2 angstroms to about 1.5 nanometers by the sixth scan at a 9-volt sample bias, and continued increasing on later biased scans. Conductive AFM detected currents of 30 to 50 nanoamps only where the surface was actively deforming. Areas that had roughened during earlier scans no longer showed measurable current, pointing to a short-lived, localized conduction process rather than a uniform current across the whole region.
The comparison region behaved differently. In a ferroelectric area, roughness increased only from about 1.8 to 3.5 angstroms under a comparable scan protocol, and no measurable current appeared in the conductive-AFM maps. The difference was not simply a matter of the surface moving: the leaky area both deformed much more and carried current while that deformation was taking place.
Bias cycling also changed the surface profile in a way the authors associate with wake-up, when a local ferroelectric response becomes visible after cycling. The cycled patch rose about 2 nanometers above the pristine surface. Its height increased sharply during the first four scans and then changed only marginally, while local PFM showed a phase contrast close to 180 degrees, consistent with switchable polarization. The pristine patch showed no discernible switching in that measurement.
The interface carries the chemical clues
Spectroscopy supplied a second distinction between the regions. The manganese L3/L2 ratio, a signal that reflects the buffer layer's local chemical state, fell from 3.45 ± 0.22 in pristine regions to 2.91 ± 0.07 in regions that had become ferroelectric after cycling. The spread in the lattice parameter was also narrower in ferroelectric scans, with a standard deviation of 0.03 angstroms compared with 0.067 angstroms in leaky samples. In plain terms, the woken ferroelectric regions looked chemically and structurally more uniform than the leaky ones.
The leaky regions showed more direct signs of damage at the interface: dark contrast, blistering in the YHO, and manganese and strontium signals inside the blistered material. The authors associate these observations with electrochemical reactions at the interface, gas evolution and heating caused by conduction. Those observations are correlative, however, and do not by themselves identify the exact sequence of events.
Their proposed explanation is that the buffer layer can lose some of its ability to act as an oxygen reservoir. That may open a competing oxygen-evolution route, in which oxygen from the lattice yields molecular oxygen, an oxygen vacancy and electrons. The authors link that proposed pathway to blistering, roughening and transient leakage, while interpreting a more even redistribution of oxygen vacancies and buffer-layer oxidation as part of the route to wake-up. These are mechanistic interpretations from linked measurements, not a direct causal test.
Device tests showed the same unevenness
The electrical results varied sharply across the reported capacitor groups. All nine capacitors made from a leaky film were leaky. In a moderately leaky group, four of nine became ferroelectric after wake-up. Among eight capacitors from the best-performing film, six switched ferroelectrically, while the other two broke down before wake-up. PUND measurements extracted a remanent polarization of about 8.5 microcoulombs per square centimeter in ferroelectric devices.
The researchers then changed the post-deposition cooling conditions to 10 millibars of oxygen and a cooling rate of 6 degrees Celsius per minute. In the reported test, all 10 capacitors on one optimized film showed ferroelectric behavior. That result is a useful processing signal, but it is not yet evidence that the same yield will hold across multiple films or larger device batches.
A diagnostic still awaiting a wider test
The study suggests that watching a film change while it is electrically stressed can expose local electrochemical heterogeneity alongside device-level electrical testing. Surface height, local current and oxidation-state signatures could therefore serve as process feedback. But the evidence is confined to selected regions in a specific YHO, LSMO and SrTiO3 stack, and does not establish that the proposed mechanism applies to other hafnia systems.
That conclusion remains tentative. The comparisons are descriptive, with no formal inferential tests or confidence intervals, and the optimized result came from one film and 10 capacitors. The local PFM and current measurements also do not establish long-term endurance, retention, fatigue or lifetime improvement. The next test is whether the cooling change improves yield across independent films and whether these local chemical signatures predict extended device performance.
Paper data and sources
Original title: Operando Surface Probe Microscopy Reveals Electrochemical Origins of Reliability Variability in Hafnia Ferroelectrics
Authors: Anudeep Tullibilli, Kartick Biswas, Shubham Kumar Parate, Pavan Nukala
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-09-29
DOI: Not available
Original paper · Full text