Transient Detour and Cooperative Oxygen Exchange in the Polarization Switching of Ferroelectric Hf0.5Zr0.5O2

Po-Yen Chen; Ryotaro Sahashi; Teruyasu Mizoguchi

REVIEW 4 major objections 5 minor 2 references

Transient Detour and Cooperative Oxygen Exchange in the Polarization Switching of Ferroelectric Hf0.5Zr0.5O2

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Polarization switching in ferroelectric HZO proceeds not by simple oxygen shifts but by a cooperative exchange between 3- and 4-coordinated oxygen, with a transient detour that keeps the lattice nearly strain-free; the paper establishes thi

desk verdict Useful dynamic complement to the static NEB picture, but the central mechanism rests on one MLFF trajectory and needs validation before it can be sold as definitive. read the letter →

arxiv 2607.26597 v1 pith:MNIEK2RV submitted 2026-07-29 cond-mat.mtrl-sci

Ryotaro Sahashi , Po-Yen Chen , Teruyasu Mizoguchi This is my paper · ORCID

classification cond-mat.mtrl-sci PACS 77.80.-e77.80.Fm71.15.Pd

keywords HZOferroelectricswitchingO3c/O4cexchangemachinelearningforcefieldelectric-field-inducedMDpolarizationhysteresisstrain-freeultra-thinfilmstability

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that polarization switching in ferroelectric HZO is driven by the dynamic mutual exchange of 3-coordinated (O3c) and 4-coordinated (O4c) oxygen atoms, rather than by the conventional simple-displacement S:N/S:T models. Using electric-field-induced molecular dynamics with a machine-learned force field fine-tuned on first-principles data, the authors reproduce the P-E hysteresis loop and directly observe the oxygen exchange pathway. During the O3c-to-O4c transition, oxygen atoms follow a curved 'detour' trajectory caused by transient cation-oxygen bond formation, while the reverse transition is more direct. Crucially, the paper identifies an 'internal self-compensation mechanism' in which local volumetric expansion and contraction from the two simultaneous transitions offset each other within the cell, keeping the macroscopic lattice nearly unchanged. This mechanism provides a microscopic origin for HZO's exceptional strain-free switching and its ability to retain ferroelectricity in ultra-thin films, and suggests that preserving cooperative O3c/O4c exchange pathways, rather than minimizing individual atomic displacements, is the key design principle for endurance and scalability.

What carries the argument

The central machinery is a fine-tuned machine-learned force field that predicts energies, forces, and Born effective charges on-the-fly, combined with electric-field-induced molecular dynamics under an applied field (swept 0 -> +10 -> -10 MV/cm) in an NPT ensemble. The field-induced force on each atom uses the predicted Born effective charge tensor. To quantify oxygen motion, the paper introduces the 'degree of orientation' P<uvw>, which measures how aligned oxygen displacement vectors are with crystallographic directions, with known random baselines. The trajectory analysis uses the cation cage (the rectangular cuboid of 4-coordinated cations) and the distance from its center of mass to ide

What would settle it

A direct density-functional-theory calculation (static or AIMD) of the O3c-to-O4c transition under an applied electric field that shows a straight minimum-energy path without the detour, or that shows the machine-learned force field's forces deviate substantially (e.g., >100 meV/Å) from DFT forces on the detour configurations, would falsify the proposed dynamic mechanism.

Watch

Extended reading notes

Core claim

The central claim is that polarization switching in Hf0.5Zr0.5O2 proceeds through the C:N34ex pathway: oxygen atoms remain inside the 4-coordinated cation cage without crossing the cation plane, while exchanging their coordination number between 3 and 4. The forward O3c-to-O4c transition involves a distinct detour in the x-direction, traced to the oxygen being pulled by a newly forming cation-oxygen bond. The forward and reverse transitions are not time-reversals: the initial O3c state sits close to the cation center, while the final O3c state rests farther away, explaining the asymmetry. Throughout the cycle, local tetrahedral volume expansions from O4c-to-O3c transitions are offset by cont

Load-bearing premise

The machine-learned force field, trained on 2,257 structures sampled from four equilibrium crystalline phases, extrapolates accurately to the transient, bond-breaking, strongly field-driven configurations visited along the O3c/O4c exchange pathway, which were not explicitly included in the training set.

Editorial extensions

If this is right

If the O3c/O4c exchange is the true switching mechanism, then static minimum-energy-path analyses alone are insufficient: cooperative, transient bond formation under an applied field must be modeled to capture ferroelectric dynamics.
The internal self-compensation mechanism implies that HZO's switching is intrinsically nearly strain-free, explaining why ferroelectricity survives under strong epitaxial clamping and in ultra-thin films without external strain relief.
The design principle for endurance and scalability shifts from minimizing individual atomic displacements to preserving the integrity of the cooperative O3c/O4c exchange pathways, e.g., by avoiding defects that disrupt coordination changes.
The calculated coercive field (7.59 MV/cm) and remnant polarization (48 μC/cm2) in a defect-free ideal lattice exceed typical experimental values, supporting the view that defects such as oxygen vacancies are what reduce coercivity in real devices.
Because the forward and reverse transitions are geometrically asymmetric (detour vs. direct), the switching dynamics may be intrinsically rate-dependent and direction-dependent, which could matter for device speed and reliability modeling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A testable extension: if O3c/O4c exchange is the rate-limiting step, then isotopic tracer or time-resolved X-ray absorption experiments that track oxygen coordination changes during field cycling should observe the detour signature and the asymmetry between forward and reverse switching.
The authors' mechanism suggests that oxygen vacancy formation—which would remove O3c or O4c sites—could disrupt the cooperative exchange and be a primary cause of wake-up or fatigue; this is an editorial inference beyond the paper's defect-free simulations, but it offers a concrete avenue for future study.
The same machine-learning MD approach could probe whether other fluorite ferroelectrics (e.g., doped HfO2 or ZrO2) share the O3c/O4c exchange mechanism, or whether their switching follows different transient pathways; the paper does not make this comparison, but it is a natural next step.
The detour trajectory's dependence on field sweep rate is not discussed in the paper; if the exchange mechanism is cooperative, one would predict that faster sweeps alter the detour shape or even suppress the exchange in favor of a more direct path, a prediction that could be checked with the same simulation framework.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here's the short version: this is a useful dynamic complement to Mu et al.'s static NEB picture of HZO switching, and the detour plus internal volume compensation are genuinely new observations. But the central mechanism is supported by one MLFF trajectory, and the MLFF has not been validated on the very configurations that define the mechanism. The claim as stated is overstrong.

What the paper does well: it fine-tunes MACEField on a decent DFPT dataset (2,257 structures, four phases, energies/forces/BECs), and the reported test errors and phonon agreement are good. The P-E loop for the defect-free ideal cell is in a reasonable place (Ec 7.59 MV/cm, Pr 48 μC/cm²), and the authors explicitly argue why that differs from experiment. The O3c→O4c detour—present in one direction and absent in the reverse—is a specific, falsifiable shape of claim that can't be dismissed as a generic displacement artifact. The clamped-cell result and the local tetrahedral volume compensation give a coherent explanation for clamping resistance.

The soft spots are real. The training set samples four equilibrium phases; the switching path is out of that distribution. The ClNEB 'confirmation' uses the same force field, so it is an internal consistency check, not independent validation. There is one SQS arrangement, one trajectory, no error bars, no replicates; the coordination-number cutoff is not stated; data and code are only available 'on request.' The field sweep rate is far from experimental, and the ideal lattice gives intrinsic behavior, so 'endurance and scalability' language goes beyond the evidence.

Is this a deal-breaker? Not obviously. The mechanism is not fitted to the target result; it emerges from the MD. The asymmetry of the detour and the averaged-path reproducibility make a pure noise artifact less likely. But an out-of-distribution bias in the MLFF could produce exactly this kind of specific artifact, so the central claim should be treated as a strong hypothesis from a single simulation, not a definitive mechanism.

Recommendation: send to peer review. A good referee should push for uncertainty quantification, a second SQS arrangement, and ideally a few DFT or independent MLFF checks on the exchange path—or at least release the models and trajectories. The paper deserves referee time, but it needs revision before it supports the strength of its conclusions.

Referee Report

4 major / 5 minor

Summary. The manuscript presents electric-field-induced molecular dynamics simulations of Hf0.5Zr0.5O2 using a fine-tuned MACEField machine-learning force field that predicts energies, forces, and Born effective charges. The authors reproduce a P–E hysteresis loop and report that polarization switching proceeds via cooperative O3c↔O4c oxygen exchange along a 'C:N34ex' pathway, with an asymmetric 'detour' in the O3c→O4c direction that they attribute to transient formation of a fourth cation–oxygen bond. They further propose an 'internal self-compensation' mechanism in which local tetrahedral volume expansion and contraction cancel, explaining the small macroscopic strain change during switching and HZO's ultra-thin-film stability. The central claim is that this dynamic exchange, rather than conventional S:N/S:T displacement, is the intrinsic switching mechanism.

Significance. If substantiated, the work would provide a finite-temperature, dynamic atomistic mechanism for polarization switching in hafnia-based ferroelectrics, going beyond static NEB analyses and offering a microscopic rationalization of HZO's strain tolerance. The paper has clear strengths: a DFT-validated MLFF with BEC capability, explicit P–E hysteresis simulation, reproduction of experimental trends (temperature-dependent coercive field and permittivity, d111 spacing), supercell-size convergence checks, and a concrete, falsifiable pathway. However, the central mechanism rests on MLFF extrapolation to transient, bond-breaking configurations and on a single MD trajectory; these issues must be addressed before the claims are fully supported.

major comments (4)

[§2.1, §3.1] The MLFF is trained on 2,257 structures from four equilibrium phases (Pca21, P21/c, P42/nmc, Pbca) and validated on a held-out test split from the same phases plus phonon dispersions. The switching pathway, however, involves transient O3c/O4c exchange with bond breaking and formation under a strong electric field. No validation structures are drawn from this region, and the ClNEB comparison in §3.3 cannot supply independent confirmation because it uses the same MLFF. The paper should provide direct DFT (VASP) energies, forces, and BECs for representative frames along the MD switching path (e.g., 20–50 configurations before/after switching), or a DFT-NEB path for at least one O3c→O4c event. Without this, the detour and exchange mechanism may be an artifact of force-field extrapolation.
[§2.4, §3.2] The central observation is based on a single MD trajectory: one SQS arrangement, one set of initial velocities, and one field-sweep rate of 5.0 × 10⁻⁵ MV/(cm·fs). This sweep rate is extremely fast compared with experimental switching conditions, and the pathway may be rate-dependent. No replicate SQS cells, repeated runs with different random seeds, or error bars are reported. The statement in §3.2 that the results 'definitively establish' the mechanism is therefore too strong. At a minimum, repeat the switching simulation with 2–3 independent SQS cells (or different initial velocities) and report whether the O3c/O4c exchange, detour asymmetry, and self-compensation are reproduced; ideally, test at least one slower sweep rate.
[§2.3, Eq. (1)] The random baselines P<100> = 0.7008, P<110> = 0.8354, and P<111> = 0.7577 are inconsistent with the stated definition of P as <cos²α> for a fixed reference axis: for uniformly random 3D vectors, this expectation is 1/3 for every axis. The reported values appear instead to correspond to a metric based on the nearest angle among all equivalent directions in the <uvw> family. If that is the intended definition, Eq. (1) and the surrounding text must be revised to state this explicitly and to derive the baselines. Since the orientation sequence <111> → <100> → random → <111> in Figure 2 is the main evidence for the exchange mechanism, this ambiguity is load-bearing and must be resolved.
[§3.4, Figure 7] The 'internal self-compensation mechanism' is inferred from the qualitative observation of simultaneous expansion and contraction of tetrahedral volumes. No quantitative correlation, net volume balance, or error analysis is given; only 20 structures are plotted. The claim that local volume changes are 'perfectly offset' is not demonstrated. Additionally, the text overstates the macroscopic effect: §3.4 says 'without altering the macroscopic lattice constant at all', but earlier in the same section volume changes of ~1.7% and a-axis changes of ~1% are reported. Please provide statistical distributions/correlations and reconcile the wording.

minor comments (5)

[§2.1/§3.2] The O3c/O4c coordination-number assignment is never explicitly defined. Specify the cutoff distance or neighbor criterion used; the entire classification depends on it.
[§2.3] Eq. (1) appears with a typographical issue ('cos2α' rather than a superscripted square). Also clarify whether α is measured to a single fixed axis or to the closest direction in the family.
[Abstract] The phrase 'a microscopic physical origin of for HZO's exceptional ability' contains a grammatical error and should be corrected.
[Data Availability] The trained MLFF models and trajectories are only available 'upon reasonable request.' For reproducibility, please deposit the model checkpoints, training data, and representative trajectories in a public repository.
[Introduction / §3.3] The distinction from Ref. [16] (Mu et al.) should be sharpened: the paper initially describes that work as proposing a 'theoretical possibility', but then the C:N34ex pathway of [16] is directly invoked. State explicitly what is new beyond Ref. [16] (e.g., the dynamic observation, the detour, and the self-compensation) to avoid ambiguity about novelty.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the O3c/O4c exchange and detour emerge from electric-field MD after training and are not encoded in the fitting targets.

full rationale

The paper's central result is an emergent electric-field-induced MD trajectory, not a fitted target. The MACEField model was trained on DFT energies, forces, and Born effective charges for 2,257 SQS structures from four equilibrium HZO phases, then validated on held-out DFT data, phonon dispersions, and experimental dielectric/coercive-field trends. The claimed O3c/O4c mutual exchange, the asymmetric 'detour,' and the internal volume-compensation mechanism are read off the MD trajectories after training; the training data and loss function do not encode the C:N34ex pathway, the detour, or the strain-compensation conclusion. The P-E hysteresis loop, coercive field, and remnant polarization are dynamical outputs rather than training labels, so the central claim is not equivalent to its inputs by construction. The ClNEB overlays in Figures 3-4 use the same fine-tuned MACEField potential, so they provide an internal-consistency check rather than independent DFT validation of the pathway; this is a robustness limitation, not a definitional circularity. The self-citations for the electric-field-induced force expression (Eq. 3) supply a standard method and are revalidated here against DFT BECs and experimental trends, and the HZO-specific switching mechanism is not imported from those papers. The degree-of-orientation metric defines its origin from the O4c cation cage, so some distance changes are geometric consequences of that definition, but the central claim of cooperative coordination exchange and transient detour does not reduce to that definition. No load-bearing uniqueness theorem or ansatz is smuggled in via self-citation, and no equation reduces to its own input.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claims rest on a learned potential and a linear BEC field-coupling approximation. The model has many fitted weights and hand-chosen hyperparameters; the coordination classification cutoff is unstated, and the results use one SQS arrangement and one field sweep. No new physical entities are introduced.

free parameters (5)

MLFF weights (MACEField fine-tuned from MACE-MP-0a large) = not disclosed; trained on 2,257 DFPT structures
All conclusions depend on this learned potential; hyperparameters (loss weights 100 BEC / 10 energy / 10 forces, rmax 6.0 Å, EMA, SWA) were chosen by hand.
Electric-field sweep rate = 5.0 × 10^-5 MV/(cm·fs)
Chosen to make switching observable in MD; affects coercive field and may affect whether the detour pathway is populated.
Coordination-number assignment (O3c/O4c) = not stated
The central claim classifies oxygen as 3- or 4-coordinated, but the bond-length/bond-order cutoff used for this classification is not reported.
SQS cation arrangement = one random solid-solution configuration
Hf/Zr disorder is represented by a single SQS realization; results may depend on the specific arrangement.
MD temperature and thermostat/barostat settings = not specified in text
NPT ensemble is stated, but target temperature, pressure, timestep, and thermostat/barostat parameters are not reported; these affect the dynamics and the hysteresis loop.

assumptions (5)

domain assumption DFT-PBEsol is an adequate reference for HZO energetics, forces, and BECs.
All training labels come from PBEsol VASP calculations; functional approximations may bias switching barriers and BECs.
domain assumption Born effective charges from the MLFF give the full electric-field response via linear coupling in Eq. (3).
The external force is e E · Z*; field-induced charge redistribution, nonlinear polarization, and finite-field effects are neglected.
domain assumption The MACE-MP-0a foundation model plus 2,257 fine-tuning structures generalizes to transition states and high-field configurations.
Validation is on held-out structures from the same four phases; the switching path includes transient structures outside the training distribution.
domain assumption A defect-free, ideal Pca21 lattice represents the switching mechanism in device-relevant HZO.
Real films contain defects, interfaces, grain boundaries, and electrodes, which may alter the pathway.
domain assumption The C:N34ex pathway nomenclature and geometric criteria from Ref. [16] are accepted.
The paper maps its trajectories onto the previously proposed C:N34ex pathway without re-deriving the geometric criteria.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Transient Detour and Cooperative Oxygen Exchange in the Polarization Switching of Ferroelectric Hf0.5Zr0.5O2." pith.science (2026). https://pith.science/paper/MNIEK2RV

@misc{pith2026260726597,
  author       = {Pith},
  title        = {Pith review of: Transient Detour and Cooperative Oxygen Exchange in the Polarization Switching of Ferroelectric Hf0.5Zr0.5O2},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MNIEK2RV}},
  note         = {Machine review of arXiv:2607.26597}
}

read the original abstract

Hafnium zirconium oxide (HZO) has attracted significant attention as a core material for next-generation non-volatile memories due to its excellent ferroelectricity in the ultra-thin film regime and its CMOS process compatibility. However, the exploration of its polarization switching mechanism has predominantly relied on static energy barrier analyses, leaving the transient bond formation and cooperative dynamic mechanisms under actual electric field driving unresolved. In this study, we performed Electric-Field-Induced MD simulations on a defect-free ideal HZO lattice using a fine-tuned machine learning force field (MACEField). As a result, we successfully reproduced the P-E hysteresis loop dynamically and demonstrated that the polarization switching in HZO is driven not by conventional simple displacement models (S:N/S:T models), but by the dynamic mutual exchange of 3-coordinated oxygen (O3c) and 4-coordinated oxygen (O4c). Analysis of the oxygen atom displacement trajectories revealed that this pathway is accompanied by a unique "detour" behavior originating from transient cation-oxygen bond formation. Furthermore, we identified an "internal self-compensation mechanism" in which the local volumetric expansion and contraction accompanying the coordination number changes are effectively offset within the cell. These findings provide, from a dynamic perspective, a microscopic physical origin of for HZO's exceptional ability to sustain stable polarization switching without macroscopic strain, a property that has long distinguished HZO from conventional perovskite ferroelectrics yet lacked atomistic explanation. These findings suggest that preserving the integrity of cooperative O3c/O4c exchange pathways, rather than minimizing individual atomic displacements, is the key design principle for endurance and scalability in next-generation ferroelectric memories.

Figures

Figures reproduced from arXiv: 2607.26597 by the authors.

**Figure 1.** Polarization switching in HZO: (a) P-E hysteresis loop, and oxygen atom displacement vectors during the (b) O4c → O3c process and (c) O3c → O4c process, and (d) schematic diagram showing the local structures of the O3c and O4c states. In (b) and (c), the origin of each vector is the center of mass of the four surrounding cations to which the oxygen atom belongs. The arrows are color-coded with a gradient from 1 (red… view at source ↗

**Figure 2.** Displacement trajectories (pole figures) and dynamic transition of orientation for oxygen atoms during polarization switching. Pole figures of oxygen atom displacement trajectories viewed from the [100] direction during the (a) O4c → O3c process and (b) O3c → O4c process. For a total of 20 structures (10 structures each before and after the polarization switching point where the sign of the polarization value flips)… view at source ↗

**Figure 3.** Displacement trajectories of individual oxygen atoms within the cation cage during the (a) O4c → O3c process and (b) O3c → O4c process. The trajectories of individual oxygen atoms are shown as thin blue lines, their average pathway (average structure of MD) as a thick blue line, and the trajectories calculated by the ClNEB method as black dashed lines [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Relationship between the cation–oxygen bond length and the y distance of the oxygen atom during the (a) O4c → O3c process and (b) O3c → O4c process. The plots show the specific cation–oxygen pair that exhibits the largest change (accompanied by bond breaking and format…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

2 extracted references

[1]

S. Wu, R. Cao, H. Jiang, Y. Li, X. Zhang, Y. Yang, Y. Wang, Y. Wei, and Q. Liu, Reliable ferroelectricity down to cryogenic temperature in wake-up free Hf 0.5 Zr 0.5 O 2 thin films by thermal atomic layer deposition, Journal of Semiconductors 45, 032301 (2024)

2024
[2]

Wang et al., Stable ferroelectric properties of Hf 0.5 Zr 0.5 O 2 thin films within a broad working temperature range, Jpn

D. Wang et al., Stable ferroelectric properties of Hf 0.5 Zr 0.5 O 2 thin films within a broad working temperature range, Jpn. J. Appl. Phys. 58, 090910 (2019). Figure S 6. Various plots under geometrically constrained conditions (bc-axis clamped state): (a) P-E hysteresis loop under the bc-axis clamped state, and the three-tier plot showing the evolution...

2019

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

S. Wu, R. Cao, H. Jiang, Y. Li, X. Zhang, Y. Yang, Y. Wang, Y. Wei, and Q. Liu, Reliable ferroelectricity down to cryogenic temperature in wake-up free Hf 0.5 Zr 0.5 O 2 thin films by thermal atomic layer deposition, Journal of Semiconductors 45, 032301 (2024)

2024

[2] [2]

Wang et al., Stable ferroelectric properties of Hf 0.5 Zr 0.5 O 2 thin films within a broad working temperature range, Jpn

D. Wang et al., Stable ferroelectric properties of Hf 0.5 Zr 0.5 O 2 thin films within a broad working temperature range, Jpn. J. Appl. Phys. 58, 090910 (2019). Figure S 6. Various plots under geometrically constrained conditions (bc-axis clamped state): (a) P-E hysteresis loop under the bc-axis clamped state, and the three-tier plot showing the evolution...

2019