Since the first report of ferroelectric response in HfO₂^[1], this material has emerged as a key material for next-generation non-volatile memories^[2,3], in particular in compositions such as Hf_0.5Zr_0.5O₂ (HZO)^[4]. Several critical factors have been identified as influencing functional properties, and these include dopants, film thickness, interfaces, stress, and defects, particularly oxygen vacancies^[5-12].

It has been demonstrated that epitaxial growth on perovskite substrates enables stabilization of the orthorhombic phase with high ferroelectric polarization^[13,14]. The lattice mismatch is very large; however, the orthorhombic phase is epitaxially stabilized through the domain-matching epitaxy (DME) mechanism^[15]. Contrary to conventional epitaxy, DME involves the formation of structurally relaxed grains during film nucleation. Epitaxial growth is enabled by the periodic insertion or absence of an extra crystal plane, which reduces the residual elastic strain. Consequently, DME strongly depends on the substrate or bottom electrode layer, as well as on the interface. Thus, interfaces play a decisive role in epitaxial HfO₂ films, and bottom electrode and/or substrate crystallographic orientation and symmetry mismatch have been shown to strongly impact ferroelectric orthorhombic phase formation^[16-18]. In the widely investigated case of SrTiO₃ (STO) as a substrate and La_0.67Sr_0.33MnO₃ (LSMO) as the bottom electrode, the orthorhombic phase of HfO₂ crystallizes with (111) orientation (3-fold symmetry) on both STO(001) (4-fold symmetry) and STO(110) (2-fold symmetry) substrates^[19,20]. Films on STO(110) have a higher orthorhombic phase content and lower monoclinic phase content than equivalent films on STO(001), indicating that residual stress in the film due to the interface mismatch is lower when using this substrate^[20]. Consequently, higher ferroelectric polarization has been reported in doped HfO₂ films grown on (110)-oriented substrates^[19-22]. However, the influence of substrate orientation on switching dynamics, retention, and breakdown electric field (E_BD) has not yet been systematically investigated.

Redox conditions during growth are also critical. Oxygen vacancies are known to influence the relative stability of competing polymorphs and can promote the formation of the ferroelectric orthorhombic phase when present in optimal concentrations^[9,20,23-25]. While excessive oxygen deficiency may lead to leakage and reliability degradation, moderate reducing conditions have been shown to enhance polarization without compromising endurance in epitaxial HfO₂ films. On the other hand, cooling conditions after deposition of epitaxial films have little effect on functional properties^[26]. In ferroelectric hafnia, fast polarization switching dynamics and long-term retention properties are also required, and in previous studies these have been linked to domain nucleation, domain wall motion, and defect-domain interactions, all of which are in turn also influenced by grain boundaries and oxygen vacancy concentrations^[27-30]. Additionally, the coexistence of ionic dynamics with the ferroelectric phase can enhance the resistive ON/OFF ratio, increasing the functionality of the material and demonstrating its potential for integration into in-memory computing platforms^[31].

Despite the extensive literature on phase stabilization and polarization enhancement in hafnia films, the impact of interface symmetry and redox growth conditions on switching kinetics and retention remains insufficiently explored in epitaxial systems. In this work, we address these questions by investigating the polarization switching behavior and retention properties of epitaxial HZO films on STO(110) and STO(001) substrates grown under different oxidation conditions. We show that switching kinetics and reliability are strongly governed by the interplay between film thickness, interface symmetry, and growth atmosphere.

Two series of HZO films with thicknesses ranging from 4.5 to 18.1 nm were grown on STO(001) and STO(110) substrates buffered with a ~25 nm LSMO bottom electrode under an O₂ atmosphere with an oxygen pressure of 0.1 mbar (denoted as STO(001) O₂ and STO(110) O₂ series, respectively). Another series of films with thicknesses ranging from 3.6 to 14.4 nm on STO(110) was deposited under an O₂/Ar atmosphere with an argon partial pressure of 0.05 mbar and an oxygen partial pressure of 0.05 mbar (denoted as STO(110) O₂/Ar series). Both the LSMO electrode and HZO layers were deposited sequentially in a single PLD process using a KrF excimer laser. The LSMO electrodes were deposited at a substrate temperature of 700 °C under an oxygen pressure of 0.1 mbar. The HZO films were deposited at 800 °C . After deposition, the samples were cooled immediately under 0.2 mbar of oxygen. Circular Pt top electrodes (20 µm diameter and 20 nm thickness) were deposited by DC magnetron sputtering through stencil masks for electrical measurements. Structural characterization was carried out by X-ray diffraction (XRD) using Cu Kα radiation in a Bruker D8 diffractometer equipped with a two-dimensional (2D) detector. Ferroelectric characterization was performed using a TFAnalyzer3000 platform (AixACCT GmbH), with the LSMO bottom electrode grounded and the bias applied to the Pt electrodes in a bottom-top configuration. For switching spectroscopy experiments, a pre-switching pulse was applied, which was long enough to ensure polarization saturation. Afterward, a trapezoidal switching pulse with a plateau and rise/fall time of t_w was applied. After 1 s, a pulse of opposite polarity to the writing pulse was used for polarization reading. The switched polarization (∆P) was calculated from the subtraction of the values measured during X and U (i.e., X-U) or X and D (X-D), where X corresponds to the reading pulse. All polarization loops were recorded at 1 kHz. To compensate for leakage in polarization loops, the dynamic leakage current compensation was employed^[32]. To determine the breakdown electric field, current-voltage characteristics were recorded using a voltage sweep from 0 to -30 V, then to +30 V, and finally back to 0 V over a total duration of 1 s.

Figure 1A shows the XRD χ-2θ maps collected around the orthorhombic o-(111) reflection for all samples in the three series. The thickness of each HZO film is indicated above the corresponding diffraction map. In all cases, a sharp and well-defined diffraction spot corresponding to the o-(111) reflection is observed at 2θ ≈ 30°, indicating the stabilization of the orthorhombic phase. The nearly circular shape of the reflection confirms the high crystalline quality of the films and the absence of significant mosaic spread, consistent with epitaxial growth. For samples grown on STO(110), an additional, more intense reflection at 2θ ≈ 32.4°, originating from the substrate and the LSMO electrode, is also observed. As the film thickness increases beyond ~10 nm, an additional spot corresponding to the monoclinic m-(-111) reflection becomes visible for all the films. Compared to the orthorhombic reflection, this peak appears more elongated along the χ angle, indicating a larger mosaicity of the monoclinic crystallites.

To quantify the phase evolution, 2θ diffractograms were obtained by integrating the χ-2θ maps along χ within a range of ±10°, as shown in Supplementary Figure 1. From these integrated profiles, the area ratio between the monoclinic m-(-111) and orthorhombic o-(111) peaks was extracted and plotted as a function of film thickness in Figure 1B. The results reveal that for thinner films, only the orthorhombic phase is detected. As the film thickness increases, the monoclinic phase fraction also increases across all sample series. Similar epitaxial doped HfO₂ films exhibit a columnar grain microstructure, irrespective of whether the grains are orthorhombic or monoclinic, as consistently demonstrated by transmission electron microscopy in previous works^[15,21]. The found thickness dependence trend is particularly pronounced for films in the STO(110) O₂ series. The structural evolution with thickness described above is schematically represented in the sketch of Figure 1C.

Figure 2A-C presents the positive-up-negative-down (PUND) polarization-electric field (P-E) hysteresis loops measured in the pristine state for the three series of samples. The corresponding current-electric field characteristics are provided in Supplementary Figure 2. All samples exhibit well-defined ferroelectric hysteresis loops, confirming the presence of switchable ferroelectric polarization. The thickness dependence of the remanent polarization (P_r) is shown in Figure 2D. The data reveal a non-monotonic behavior, with reduced P_r observed in both very thin and thick films. The P_r increases from ~8.6 μC/cm² for the 4.5 nm film to a maximum of 23.4 μC/cm² for the 9 nm film of the STO(001) O₂ series, whereas in the STO(110) O₂/Ar series, it reaches a maximum of 34.7 μC/cm² at 7.2 nm. The reduction of P_r in thinner films can be ascribed to dead-layer effects. The comparison of the STO(001) O₂ and STO(110) O₂ series shows that the polarization is consistently higher for films grown on STO(110) across the entire optimal thickness range. This observation is in agreement with previous reports, while extending them by demonstrating that the enhancement is systematic over a broad range of film thicknesses, rather than being limited to the comparison of single thicknesses^[19,20,33]. Although Pt was selected as the top electrode due to its non-highly reactive nature, and LSMO was used because of its epitaxial compatibility with HZO, the presence of interfacial capacitance cannot be completely avoided. Such effects may occur either at the top Pt/ferroelectric-layer interface, probably due to defect accumulation^[33] or at the bottom ferroelectric-layer/LSMO interface due to cation intermixing^[34,35]. For thicker films, the remanent polarization decreases again, which is consistent with the increasing fraction of the non-ferroelectric monoclinic phase [Figure 1]. The coercive field dependence on thickness is shown in Figure 2E. A general decrease in the coercive field is observed as film thickness increases.

To gain further insight into the switching characteristics, the P-E loops were fitted using a model describing the ferroelectric polarization response^[36]:

(1) $$ \begin{equation} \begin{aligned} \mathrm{P}_{\mathrm{FE}}=\mathrm{P}_{0} * \tanh \left(\frac{\mathrm{E}-\mathrm{Ec}}{\delta}\right) \end{aligned} \end{equation} $$

where P₀ is the saturation polarization, E_c is the coercive field, E is the applied electric field, and δ represents the width of the coercive field distribution. This semi-empirical method provides a convenient measure of switching inhomogeneity represented by δ. The corresponding fits are shown in Supplementary Figure 3. The extracted values of δ are plotted as a function of film thickness in Figure 2F. The thinnest films exhibit the largest distribution, δ values reaching approximately 1.5-2.2 MV/cm, indicating a broader spread of local switching fields (enclosed by the dashed rectangle). The δ decreases with increasing thickness and then remains relatively small (~1.0-1.5 MV/cm), indicating that the switching process is more uniform.

Figure 3A-C shows the polarization time-dependent switching spectra measured using the PUND method (see Section “EXPERIMENTAL” for details) for the three series, under both positive and negative writing voltages (V_w). The leakage correction procedure is detailed in Supplementary Figure 4. The corresponding polarization-electric field loops obtained for different writing pulse widths are presented in Supplementary Figure 5. Most samples exhibit clear ferroelectric switching behavior, characterized by a rapid increase in ∆P with increasing t_w, followed by saturation at t_w > ~1 μs. STO(001) O₂ series shows relatively smaller switching amplitudes and a nearly symmetric response between positive and negative polarities, as shown in Figure 3A. In contrast, the STO(110) O₂ series exhibits a noticeably larger switching amplitude and a clear asymmetry between the two poling directions [Figure 3B]. An even stronger enhancement of ∆P, together with a more pronounced polarity dependence, is observed for the STO(110) O₂/Ar series shown in Figure 3C.

For the thinnest films (below ~5 nm), small variations in the shape of the I-V and P-V loops during the first cycles are observed [Supplementary Figure 6], although these variations remain relatively minor. In the case of the 4.5 nm STO(110) O₂/Ar series, where this variation is most noticeable, and a significant reduction in the switching time is observed after electrical cycling [Supplementary Figure 7]. This switching time reduction indicates that the redistribution of defects at the interface can influence the switching dynamics. To further analyze the switching kinetics, the experimental ∆P-t_w data were fitted over the entire experimental range using the nucleation-limited switching (NLS) model^[37], which accounts for a distribution of local switching times:

(2) $$ \begin{equation} \begin{aligned} \Delta \mathrm{P}=2 \mathrm{P}_{\mathrm{r}} \int_{-\infty}^{+\infty}\left[1-\mathrm{e}^{-\left(\frac{\mathrm{t}}{\tau}\right)^{\mathrm{n}}}\right] \mathrm{F}(\log \tau) \times \mathrm{d}(\log \tau) \end{aligned} \end{equation} $$

where n is the effective dimensionality of domain growth (fixed at n = 2), τ is the characteristic switching time, and F(logτ) is the Lorentzian distribution for the logarithm of switching time. The Lorentzian distribution can be expressed as:

(3) $$ \begin{equation} \begin{aligned} F(\log \tau)=\frac{\mathrm{A}}{\pi}\left[\frac{\mathrm{w}}{\left(\log \tau-\log \mathrm{t}_{1}\right)^{2}+\mathrm{w}^{2}}\right] \end{aligned} \end{equation} $$

where A is a normalization constant, w accounts for the width of the distribution, and logt₁ represents the central value of the distribution.

The extracted distribution functions are shown in Figure 3D-F, while the corresponding fitting parameters are summarized in Supplementary Tables 1-3. For STO(001) O₂ series, the switching time distribution is relatively narrow and symmetric for the two polarities, indicating similar switching dynamics under positive and negative fields. In contrast, STO(110) O₂ series exhibits broader distributions and a more pronounced and systematic difference between positive and negative switching times. This asymmetry becomes even more important for the STO(110) O₂/Ar series, where the switching distributions indicate faster switching for one polarity.

Figure 4A shows the thickness dependence of the characteristic switching time extracted from the NLS fitting of the switching spectroscopy data presented in Figure 3. The values are plotted for both positive and negative voltage polarities for the three sets of samples. For ultrathin films (thickness below ~6 nm), the characteristic switching time exhibits a pronounced dependence on the polarity of the applied electric field. In STO(110) O₂ series, the extracted values change from log(t₁) = -7.32 for positive polarity and -6.68 for negative polarity for the 4.5 nm film. A similar asymmetry is also observed for the STO(110) O₂/Ar series, where the 3.6 nm film shows -7.36 for positive polarity and log(t₁) ≈ -6.75 for negative polarity.

As the film thickness increases, above ~9 nm, the characteristic switching time gradually shifts toward smaller values, and the difference between the two polarities becomes significantly reduced. In the STO(110) O₂ series, the switching time changes from log(t₁) ≈ -6.6 in the 5-9 nm range to approximately -7.2 for films with thicknesses of 13.6-18.1 nm for the negative polarity (V_w < 0). The difference between positive and negative switching times decreases from log(t₁) ≈ 0.60 at 4.5 nm to less than ~0.15 for films thicker than ~13 nm. A similar trend is observed for the STO(001) O₂ and STO(110) O₂/Ar series. The observed reduction in switching time asymmetry for thicker films indicates that this is governed by the presence of defects at the interfaces, as discussed above^[33-35] and further supported by the reduction in the imprint electric field with thickness [Supplementary Figure 8]. Figure 4B shows the thickness dependence of the w parameter obtained from the NLS fitting. The w values remain within a relatively narrow range of ~0.37-0.47 for all samples. Slightly lower values for the thinnest and thickest films are mainly attributed to the lower polarization of these films.

Figures 5A-C show the Weibull statistical distribution of the E_BD for the three sets of samples. Representative current-electric field curves used to determine the breakdown field for the different film thicknesses are shown in Supplementary Figure 9, except for the 3.6 nm film, which showed larger conductivity. The E_BD was defined as the electric field at which a sudden increase in leakage current indicates dielectric failure. This, in the present case, occurs for negative polarity. The Weibull relationship is expressed as^[38,39]:

(4) $$ \begin{equation} \begin{aligned} \operatorname{In}[-\operatorname{In}(1-\mathrm{F}(\mathrm{i}))]=\beta \ln \mathrm{E}-\beta \ln \mathrm{E}_{0} \end{aligned} \end{equation} $$

where F(i) represents the cumulative breakdown probability, E is the applied electric field, E₀ is the characteristic breakdown electric field, and β is the Weibull slope describing the dispersion of breakdown events. As shown in Figures 5A-C, the plots of ln[-ln(1 - F(i))] vs. ln(E) exhibit linear behavior for all samples. The characteristic breakdown fields extracted from the Weibull analysis are summarized in Figure 5D as a function of film thickness. β values lie between ~15 and ~31 for all samples. A clear monotonic decrease in E₀ with increasing thickness is observed for all three series. The characteristic E₀ decreases from ~55 MV/cm for the 4.5 nm film to about 15 MV/cm for the 18.1 nm films. In brief, the overall thickness dependence remains very similar across the three series, indicating that the dielectric breakdown strength is primarily governed by the film thickness. More careful data inspection allows us to infer that E_BD is slightly lower in O₂/Ar-grown films (see Supplementary Figure 10), suggesting a higher concentration of defects, most likely oxygen vacancies. These defects promote dielectric breakdown at lower electric fields, indicating that defect density acts as a secondary factor governing the breakdown strength.

Figure 6 summarizes the retention performance of the three series, evaluated from the remanent polarization extrapolated to 10 years and normalized to the initial polarization value, as a function of film thickness. The corresponding PUND loops and raw retention measurements used for the extrapolation are provided in Supplementary Figure 11. For films with thickness in the 5-10 nm range, both STO(001) O₂ and STO(110) O₂ series exhibit relatively stable retention, with more than 50% of the initial polarization preserved. In contrast, STO(110) O₂/Ar series shows a more pronounced asymmetry between the positive and negative polarization states. In particular, the retained polarization decreases to approximately 50%-60% for the negative polarization state, while the positive state remains comparable to the other series (≈70%-77%).

Figure 7 summarizes the relationship between the remanent polarization and switching time, dielectric breakdown strength, and retention. The numbers next to the data points indicate the corresponding film thickness in nm, allowing the role of thickness to be directly visualized in the correlations. Figure 7A shows the dependence of the characteristic switching time on the remanent polarization for positive polarity. A general correlation between polarization magnitude and switching time is observed. Films with low polarization tend to exhibit relatively faster switching dynamics. Note that the obtained low switching times are very close to the time constant of the experimental setup, measured in detail in ref.^[40], thus influenced by that.

Figure 7B shows the absence of direct correlation between remanent polarization and the dielectric breakdown field. E_BD increases with polarization for low E_BD values, while it decreases for high E_BD values. This non-monotonic behavior reflects the thickness dependence of polarization, which first increases and then decreases with increasing film thickness. Therefore, it can be concluded that E_BD is mainly determined by film thickness. The correlation between polarization and retention stability is shown in Figure 7C. The normalized retention P_r(10⁵ s)/P₀ smoothly increases as the polarization increases. Note that here retention is evaluated by P_r at 10⁵ s, rather than the value extrapolated to 10 years, as done in Figure 6. This approach avoids introducing additional uncertainty associated with long-term extrapolation and provides a direct comparison of the experimentally measured retention across samples. This behavior indicates that larger polarization stabilizes the ferroelectric state and suppresses depolarization-driven back-switching, leading to improved retention. The results obtained in samples grown under O₂/Ar atmosphere do not show systematically decreased retention. This behavior suggests that higher oxygen vacancy concentrations under reducing conditions tailor the stabilization of the orthorhombic phase, increasing polarization [Figure 2C], but do not introduce additional depolarization mechanisms that reduce retention. The corresponding dependences for negative switching time and retention for negative polarity are shown in Supplementary Figure 12.

To summarize, it is observed that as the film thickness increases, the fraction of the monoclinic phase gradually increases, while the remanent polarization initially increases and then decreases as the orthorhombic ferroelectric phase becomes less stable in thicker films. The switching becomes faster in films containing a larger content of the monoclinic phase, indicating that grain boundaries between monoclinic and orthorhombic grains may act as nucleation centers, thus increasing the switching speed [Figure 7D]^[40]. The dielectric breakdown field decreases monotonically with thickness. Figure 7E shows the dependence of E_BD on thickness in a log-log plot, including data fitting following the relation E_BD ∝ t^-n for the three series^[41]. In all cases, the extracted exponent value is close to n ≈ 1. An exponent near 1 suggests dielectric failure governed by interface-limited carrier injection and/or interfacial redox processes, rather than by bulk effects or the density of orthorhombic/monoclinic grain boundaries^[41]. The retention stability improves with polarization and reaches optimal values in the intermediate thickness regime. The improved retention is consistent with a reduction in the depolarization electric fields (E_dep)^[42-44]. Thus, when the fraction of the orthorhombic phase decreases at the expense of the monoclinic one, in-plane depolarization fields (E_dep) are promoted [Figure 7F]. These fields ultimately favor back-switching of the out-of-plane polarization, leading to poorer retention, as suggested in previous studies of La-doped HfO₂ films^[45]. These competing trends highlight the existence of an optimal window where large polarization, stable retention, and acceptable breakdown strength can be simultaneously achieved. In the present films, this optimal region occurs at intermediate thicknesses of approximately 5-10 nm.

In this work, we investigated the influence of substrate orientation, growth atmosphere, and thickness on the switching behavior and reliability of epitaxial HZO films. Structural analysis shows that the ferroelectric orthorhombic phase is preferentially stabilized for thinner films, while increasing thickness promotes the formation of the monoclinic phase. The largest remanent polarization (above 30 µC/cm²) is obtained for films in the intermediate thickness range (5-10 nm). Switching spectroscopy reveals polarity-dependent switching kinetics in thinner films, attributed to built-in electric fields. Reliability measurements further show that the breakdown field decreases with thickness, while retention remains stable in the intermediate thickness range. Thus, detrimental properties are found in the thinnest films, due to a larger contribution of interface-related defects, and in the thicker films, where the orthorhombic phase ratio decreases. Switching speed is found to be favored by the presence of the monoclinic phase. The presented results provide guidelines for the design of ferroelectric tunnel junctions and other resistive switching devices, where the coexistence of oxygen vacancies and ferroelectricity can lead to enhanced device performance^[31]. Overall, these results demonstrate that phase stabilization and interface asymmetry govern the switching dynamics and reliability of epitaxial HZO films, providing guidelines for optimizing ferroelectric hafnia-based devices.