Coercive Field Control in Epitaxial Ferroelectric Hf0.5Zr0.5O2 Thin Films by Nanostructure Engineering (2025)

In recent years, nonvolatile memory applications for energy-efficient data storage, e.g., FeRAM, have been the leading motivation for ferroelectrics research. (1) The discovery of ferroelectricity in nanoscale films of Si-doped hafnium oxide in 2011 enthralled the ferroelectric community by offering potential for an industry-friendly CMOS compatible ferroelectric material suited for integration with Si-based electronics. (2,3) Later, stabilization of the ferroelectricity in HfO₂ was shown using a variety of other doping elements (Zr, (4) La, (5) Y, (6) Al, (7) etc.).

A key challenge for doped HfO₂ is that several nonpolar polymorphs can be stabilized in the thin film form, such as the cubic (Fm-3m), the tetragonal (P4₂/nmc), and the monoclinic (P2₁/c) phase in HfO₂. Therefore, it is common to find studies where the ferroelectric, orthorhombic (o-), or rhombohedral (r-) structures with space groups Pca2₁ (2) or R3m, (8) respectively, appear concomitantly with the nonpolar phases. Therefore, the control of a single-phase ferroelectric material is nontrivial but highly needed. Most previous reports on ferroelectricity in hafnia-based films consider polycrystalline multiphase films, which are only partly stabilized in the ferroelectric o-phase. In such films, understanding of the factors that control the ferroelectric properties is challenging. For example, studies correlating the influence of annealing on the grain size and thus ferroelectric properties also report significant differences in the phase content alongside the investigated parameter. (9,10) By allowing for single-phase or near-single-phase stabilization, epitaxial doped-HfO₂ films offer a simpler system for understanding the factors that control the ferroelectric properties. (11) Such epitaxial films show no wake-up effect, which is desired for device applications.

In epitaxial films of o-phase doped-HfO₂, E_c can range from ∼2 to 3 MV/cm. (12) Here, various extrinsic factors have been shown to contribute to the phases formed and so to the ferroelectric properties, notably oxygen vacancies, dopants, strain, electric fields, and surface and interface effects. (13,14) More recent reports identify a rhombohedrally distorted orthorhombic (r-d o) phase with polarization pointing along [001] and a unit cell elongated along d₁₁₁. (15−19) It is understood to originate from the orthorhombic phase, where it is stabilized with additional interfacial strain from the underlying substrate. On the other hand, it forms only in epitaxial films, and its measured E_c is large (∼4 MV/cm). (16,17) Despite the demonstrated reduction of E_c with film thickness t as E_c ∼ t^–2/3 in o-phase epitaxial films, (11) the technologically relevant coercive voltage (V_c) still increases with thickness as V_c ∼ t^1/3, making further studies on understanding and reducing E_c (V_c) critical.

In this work, we aim to understand the factor(s) that controls E_c in r-d o HfO₂-based oxide thin films. This understanding is needed since different applications require different E_c values. We explored varying this laser fluence in films grown by pulsed laser deposition (PLD). Laser fluence influences the energetics of the deposition process, which can affect crystallinity, stoichiometry, grain orientation, defect types, and defect concentrations. (20,21) We aim to understand which of these is the most critical controlling factor for E_c.

We study epitaxial Hf_0.5Zr_0.5O₂ (HZO) films grown on La_0.7Sr_0.3MnO₃-buffered (LSMO) (001)-oriented SrTiO₃ (STO) substrates at laser fluences ranging from 0.5 J cm^–2 up to 1.3 J cm^–2. We selected STO as the substrate and LSMO as the bottom electrode because they have been determined to yield optimal ferroelectric response in previous studies. (22,23) We focus on the widely studied HZO composition, a solid solution of HfO₂ and ZrO₂ which has a fluorite-related crystal structure (24,25) rather than aliovalent-cation-doped HfO₂ compositions where oxygen vacancies can also form (15,26) as these could complicate the understanding.

We find that low E_c is achieved at low laser fluence. Unlike classical perovskite ferroelectrics, (20) where strain is increased at higher laser fluence, here we find no changes in strain but only in the films’ nanostructure with laser fluence. Films deposited at higher laser fluence show the emergence of grains with (11–1) orientation (in addition to (111) orientation, which forms at low fluence). Density functional theory (DFT) calculations are consistent with the formation of (11–1) oriented under higher energetic conditions, such as those provided by higher laser fluence. As shown both experimentally and from calculations, the additional orientation perturbs the film nanostructure, leading to grain tilting and associated dislocations, which can act as domain pinning sites to increase E_c, just as in perovskite ferroelectrics. (27,28) An E_c as low as 2.7 MV/cm was achieved for the lowest laser fluence studied, which produces the most crystallographically perfect material with a single grain orientation.

A series of thin film samples was deposited at a range of laser fluences, 0.5 J cm^–2 (HZO-0.5), 0.7 J cm^–2 (HZO-0.7), 0.9 J cm^–2 (HZO-0.9), and 1.3 J cm^–2 (HZO-1.3), with a laser frequency of 1 Hz. Since higher laser fluence also increases growth rate, we independently studied the influence of growth rate by depositing one film at a laser frequency of 5 Hz and a medium laser fluence of 0.7 J cm^–2. High-resolution X-ray diffraction (XRD) scans along the 2θ–ω axes of the HZO films are shown in Figure 1a. The diffraction peak at ∼29.9° can correspond to the (111) orientation of HZO of o-, r-, or r-d o-phase HZO. (15,17,29) We ascribe this peak to r-d o(111), for reasons discussed later. The (111) peak position remains largely unchanged across the series of samples, indicating no change in the out-of-plane lattice parameter (or strain) with laser fluence. A similar absence of strain effects from specific deposition conditions such as deposition pressure or temperature has been previously observed in epitaxial hafnia-based films. (15,30) We propose that the absence of strain dependence on laser fluence in HZO films is due to the presence of a phase transition from the tetragonal (t-) to the o-phase during cooling and a large stiffness of polar hafnia, having about ten times larger Young’s modulus than ferroelectric perovskites. (31) Thus, strain from the local nanostructure arising during growth is relaxed during this transition. This result is in stark contrast to previous reports on perovskite ferroelectrics, where strain levels were shown to be highly sensitive to changes in laser fluence. (20)

In Figure 1b, 2θ scans were performed at χ∼70° and ϕ∼45° to access an in-plane axis of a −111 plane. We find that all −111 peaks in all films share an equivalent 2θ position of ∼30.5°, thus implying d₁₁₁ > d_–111. Based on the peak positions of (111) and −111, the d-spacing was obtained (d₁₁₁ = ∼2.99 Å and d_–111 = ∼2.94 Å) using Bragg’s law. Given that the o-phase has α = β = γ = 90°, the rhombohedral distortion was calculated using 2 × arctan(d_–111/d₁₁₁), which results in α = β = γ = 89.07°, (8) thus confirming the phase is either r-HZO or r-d o-HZO. To further explore the structure of the films based on the ferroelectric switching behavior, we undertook piezo-response force microscopy (PFM) measurements. We find the presence of a significant in-plane PFM response (Figure S1) and conclude that film polarization in our films is along [001]_pc and not [111]_pc and therefore that the films are stabilized in the r-d o structure. (32)

Laser fluence also changes growth rate, hence each film’s growth rate is shown in Figure 1c. With an increase in laser fluence from 0.5 J cm^–2 to 1.3 J cm^–2, growth rate is increased by approximately 50%. Film crystallinity is evaluated by comparing the normalized intensities of the r-d o(111) peak to the STO (001) peak in the 2θ–ω XRD scans across the sample series (see Figure 1d). We observe that film crystallinity deteriorates with an increase in laser fluence and an increase in growth rate, as would be expected. For a constant fluence of 0.7 J cm^–2, an increase in laser frequency from 1 to 5 Hz (yellow data point) results in a more than 4-fold increase in growth rate. However, the reduction in r-d o(111) peak intensity is about the same as for the laser fluence increase from 0.5 to 1.3 J cm^–2 where the growth rate is increased by only 50%. This indicates that there is another factor at play for the laser fluence effect beyond just degraded crystallinity with increased kinetics.

The average nanostructure of the HZO films was further analyzed by performing ω scan (rocking curves) at the (111) peak position (see Figure 1e). The rocking curves consist of a sharp peak, indicating a highly oriented crystalline region, and a broad peak, indicating the presence of some misoriented regions. Gray dotted lines indicate the baselines used for the integral breadths of the broad (B) and sharp (S) peaks, whose ratios provide a measure of grain misalignment. In Figure 1f, we plot the ratio of B and S as a function of laser fluence. For the films grown at 1 Hz laser frequency, we observe less grain misalignment for lower laser fluence. The HZO film grown with 5 Hz laser frequency shows only a slightly reduced grain misalignment than the one deposited at 1 Hz, despite the 4-fold higher growth rate than the 1 Hz sample (Figure. 1c). This is in agreement with Figure 1d, where the intensity of the r-d o(111) peak was shown to be reduced to a lesser extent than for laser energy increase. This points again to another factor at play for the laser fluence effect other than just increased kinetics for degrading grain alignment/crystallinity.

Oxygen content and cation stoichiometry are critical factors in setting ferroelectric properties in HZO films. (4,33−35) To investigate the effect of laser fluence on film stoichiometry, we performed X-ray photoelectron spectroscopy (XPS) measurements using direct vacuum transfer (see Figure S2). We found no significant change between films grown at laser fluences of 0.5 J cm^–2 and 1.3 J cm^–2. This lack of stoichoimetric variation with fluence is consistent with reports on perovskite ferroelectric BaTiO₃. (20) The absence of composition changes with laser fluence is likely due to the similar mass, size, and the same valence of Hf and Zr ions, as well as the controlled oxygen content during the slow cooling process in high oxygen pressure (see Methods).

We now explore the influence of laser fluence on the grain size and associated defects. Here, an extensive scanning transmission electron microscopy (STEM) analysis was performed. Based on multiple cross-sectional HAADF images (zone axis [110] STO, ϕ = 45°), the average lateral grain size was evaluated using Fourier-filtered images (Figure 2a). Fourier filtering is performed from fast Fourier transform (FFT) of HAADF images by selecting all peaks corresponding to the first order frequency of the HZO(111) lattice parameter, applying masks, and performing inverse FFT. This procedure enhances the contrast between grains, therefore allowing us to extract grain size. The HAADF images, each with a total lateral scale of 320 nm, provided sufficient data for this determination. As expected, the average lateral grain size was found to be larger, 12.3 nm, for lower laser energy film, HZO-0.5, compared to higher laser energy film, HZO-1.3, where it is 7.7 nm (Figure 2b). We further analyzed the same cross-sectional HAADF images to observe dislocations at the grain boundaries. Figure 2c shows an inverse fast Fourier transform (IFFT), visualizing horizontal dislocations. The dislocation density for HZO-0.5 is ∼2 × 10¹² dislocations/cm², while it is ∼3 × 10¹² dislocations/cm² for HZO-1.3. While this is only a broad estimate of dislocation density, it nevertheless shows the higher density of dislocations for higher laser fluence, consistent with the higher density of grain boundaries. The dislocation density values also broadly agree with those reported previously in r-d o-phase in doped-HfO₂ work. (17)

To obtain more information about the dependence of grain tilting on laser fluence, TEM cross-section along zone axis [010] STO, ϕ = 0°, was performed to inspect interface quality between LSMO|HZO and grain boundaries. For HZO-0.5 (Figure S3a) an atomically smooth LSMO|HZO interface is observed along with a thin t-phase interfacial layer (8) (inset: blue solid box). Also, minor grain tilting is observed in the body of the HZO film, consistent with the very minor B peak observed in the XRD spectrum (Figure 1e,f). The interfacial layer is tensile strained to the substrate, resulting in the in-plane lattice parameter of ∼3.91 Å, similar to the value observed in the literature. (8) For HZO-1.3, the interface in HZO-1.3 is rougher, with an irregular thickness of a t-phase interfacial layer (inset: red solid box, Figure S3b). This is consistent with the more energetic particles impinging on the LSMO surface at higher laser fluence (36) and occurs at the same time as the increased grain tilting (Figure S3b), with tilt angles of up to 2.6°, broadly consistent with the full width half-maximum of the X-ray B peak (Figure 1e,f).

Having shown that higher laser fluence yields smaller grains, more grain tilting, and associated dislocations, we now explore how laser fluence influences grain alignment. To do this, we further investigated cross-sectional HAADF iamges of HZO-0.5 and HZO-1.3, now along the zone axis [110] STO, ϕ = 45°. High-magnification images and their FFT images are shown in Figure 3a,b and lower-magnification images and their FFT images in Figure 3c–e.

The grain orientations were identified from d-spacing measurements from the localized FFTs as shown in Figure 3b. FFTs from HZO-0.5 (blue box from Figure 3b) and HZO-1.3 (middle red box from Figure 3b) give an out-of-plane d-spacing of 2.99 Å and in-plane d-spacing of 2.93 Å, which agrees with XRD results from Figure 1a,b. On the other hand, the far right red dotted boxed FFT from Figure 3b demonstrates an out-of-plane d-spacing of 2.93 Å and in-plane d-spacing of 2.99 Å, which is indicative of a (11–1)-oriented grain. Therefore, we assign both (111) and (11–1) grain orientations for HZO-1.3, while only a single (111)-oriented r-d o-grain is assigned for HZO-0.5. The absence of a distinct 2θ peak related to (11–1)-oriented grains in XRD is due to several factors: the lower fraction of (11–1) relative to the (111) grains, the overlap of expected peaks from the two orientations, the lower structure factor of (11–1) compared to (111), the tilt of the (11–1) grains out-of-plane relative to the (001) direction of the substrate, and the increased mosaicity of the (11–1)-oriented grains (Figure S4). To provide more statistical information about the grain sizes and orientations in the two samples, low-magnification images of the two samples were obtained (Figure 3c,d) with FFTs shown in Figure 3e. It is immediately apparent that smaller grains are present for HZO-1.3 (Figure 3d) compared to HZO-0.5 (Figure 3c), also in agreement with the grain size analysis from Figure 2a,b. Also, the grain orientations (Figure 3e) are consistent with those measured at higher magnification (Figure 3b).

Next, DFT calculations were undertaken to understand the relative stability of the (111) and (11–1) orientations. Calculations were performed on the r-d o-phase where α = β = γ = 89.07°, from which the (111) and (11–1) oriented HZO cells were obtained. The surface energy (E_s) for the r-d o-phase showed E_s(111) = 1.08 J m^–2 and E_s(11–1) = 1.15 Jm^–2, indicating that the (11–1) oriented surface is slightly less favored than the (111) oriented surface, and so the (111) oriented grains are more stable (Figure 3f). Under higher energetic conditions, such as high laser fluence (HZO-1.3 film), the presence of the additional (11–1) orientation is enabled. To understand whether the presence of the (11–1) oriented grains impacst the ferroelectric properties, the energy barrier and polarization were calculated along the ferroelectric switching paths of (111) and (11–1) oriented HZO grains. No significant differences in either P_s or barrier height were found indicating that E_c would not be increased simply because of switching of the (11–1) domains.

The calculated interplanar distances for the (111) and (11–1) orientations align well with the HRTEM and XRD results, yielding d₁₁₁ = 2.997 Å and d_11–1 = 2.934 Å. The detailed structural parameters of the (111) and (11–1) oriented HZO cells can be found in Table S1. Because of the slight differences in the lattice vectors and angles between the (111) and (11–1) orientations, the tilting between the boundaries and associated dislocations is expected at the (111)/(11–1) grain boundary (Figure 3g). Indeed, the DFT calculations predicted a grain tilt of about 3°, which broadly agrees with the experimental observations (Figure 1e,f, Figure S3b). The formation of the additional (11–1) grains which are intrinsically stabilized at higher laser energy is consistent with the finer grain size of HZO-1.3 compared to the HZO-0.5 (Figures 2b and 3c,d).

Finally, on the microstructural side, energy dispersive spectroscopy (EDS) was performed based on the HAADF images (Figures S5–S6). A homogeneous distribution of cations and sharp interfaces was observed in both HZO-0.5 and HZO-1.3 films, corroborating the wide-range XPS results.

We now investigate the influence of laser fluence on the ferroelectric properties of HZO films on a macroscopic scale. For electrical measurements, 50-μm diameter electrodes were patterned using UV lithography, and tungsten (W) was sputtered as the top electrode. Figure 4a–c shows the current–electric-field (I–E) relation measured by different protocols: dynamic hysteresis measurement (DHM), dynamic leakage current compensation (DLCC), and positive-up–negative-down (PUND). DHM shows all current contributions, including the dielectric, ferroelectric, and leakage currents. DLCC corrects for the leakage current following a built-in software function. (37) PUND includes only the ferroelectric contribution. All films show wake-up-free ferroelectric behavior. All HZO films demonstrate clear ferroelectric switching peaks at both negative and positive bias (see Figure. 4a–c). The coercive peak values are extracted from the current maxima positions to avoid the influence of leakage. The loops are off-centered by ∼−0.5 MV/cm because of the asymmetric LSMO|HZO|W device stack. (38) The E_c was calculated as the average of fields of current maxima on the positive and negative bias. Corresponding polarization–electric field (P–E) and I–E loops for all of the HZO samples with different measurement protocols can be found in Figures S7–S10.

Figure 4d displays E_c as a function of laser fluence. E_c was observed to increase from ∼2.7 to ∼3.3 MV/cm with an increase in laser fluence from 0.5 J cm^–2 to 1.3 J cm^–2 from DHM. The trend of increasing E_c with laser fluence is consistent across all three measurement protocols. Notably, the sample grown at 5 Hz exhibits the same E_c value as the one deposited at 1 Hz with the same laser fluence of 0.7 J cm^–2, despite its poorer crystallinity (overall level of defects in the sample, Figure 1d). This indicates that factors other than overall crystalline perfection influence E_c.

Figure 4e shows the dependence of P_r on laser fluence based on positive-up (PU) measurements. PU measurements were chosen due to variations in leakage current at negative bias (see Figures S8–S9). The higher leakage current in films deposited at lower fluence might arise from their higher polarization value leading to more accumulation of bound charges and enhanced carrier injection. HZO-1.3 demonstrates much lower P_r (<3 μCcm^–2) compared to HZO-0.5 (∼8 μCcm^–2), at 5 MV/cm. The marked reduction in P_r for the 5 Hz deposition rate (∼3 μC cm^–2) is consistent with the strong reduction in crystallinity (Figure 1d). We recall 5 Hz increases the deposition rate by 4-fold (Figure 1c). We also recall that 5 Hz does not produce more grain tilting (Figure 1e,f and Figure S3a) and so microstructure plays less of a role in controlling P_r than random defects do. Figure 4f shows the dependence of E_c and P_r (from PU measurement) on the B/S peak ratio. A trend of higher E_c and reduced P_r is observed with higher B/S ratio, i.e., more grain tilting.

Overall, we have shown that high laser fluence leads to increased E_c. This occurs when two grain orientations, (111) and (11–1), are stabilized. At the same time, no change in strain (Figure 1a,b and Figure 3d), composition (Figure S2), or switching barrier height (Figure 3f) takes place. We also showed that E_c is not reduced by defects associated with the higher growth rate of 5 Hz (Figure 4d). Thus, we can deduce that the increased E_c with laser fluence is dominated by the grain tilting effect of the (111)/(11–1) grain boundaries (Figure 4f), along with associated grain boundary dislocations (Figure 2c). While the grain boundary defects also lead to a decrease in P_r, the defects associated with the higher growth rate appear more significant for controlling P_r (as seen from the 5 Hz point in Figure 4e). Overall, E_c is controlled by domain wall pinning by dislocations associated with tilted grains, similar to the case of perovskite ferroelectrics. (39−41)

While o-phase films can exhibit higher P_r values, (11) similar values of P_r (sub-10 μCcm^–2) were achieved in r-d o-phase epitaxial hafnia-based films of similar thickness, (17) but the E_c values observed previously are usually higher than 4 MV/cm. (15,17) However, in this work, in films of ∼10 nm thickness, by tuning the laser energy, we demonstrated a further reduction in E_c (by ∼30%) to sub-3 MV/cm (∼2.7 MV/cm). Tuning E_c at a specific thickness is an important step toward the integration of ferroelectrics in memory and logic devices for next generation electronics. (42,43)

In epitaxial rhombohedrally distorted orthorhombic (r-d o) Hf_0.5Zr_0.5O₂ (HZO) films where composition, strain, and thickness are kept constant, we have demonstrated a link between nanostructure and coercive field (E_c). We showed that increasing laser fluence leads to increased nanostructural disorder via the emergence of a (11–1) grain orientation, which emerges in addition to the (111) orientation, which is present at low laser fluence. The (11–1) orientation is shown to produce in-plane grain tilting and dislocations at the low angle grain boundaries, and these imperfections are consistent with increased domain pinning and increased E_c. Low laser fluence, which produces a higher proportion of (111) orientation and high crystallinity, leads to a roughly 30% reduction in E_c compared to literature values on the r-d o-phase (to ∼ 2.7 MV/cm at 0.5 J cm^–2). Overall, our work shows the importance of controlling nanostructure to tune the ferroelectric properties of HZO films, enabling performance optimization of hafnia for memory devices.