Enhancement of ferroelectricity and orientation in solution-derived hafnia thin films through heterogeneous grain nucleation

In 2011, Böscke et al.¹ reported ferroelectricity in 10 nm-thick Si:HfO₂ films. As pointed out recently,² research on pyro- and piezoelectric properties and application potential for these materials is still in an initial stage. This is also due to the fact that thicker films—100 nm and beyond, in which the ferroelectric phase is often not stable^3–8—are favorable for these applications. In 2017, up to 390 nm-thick polar ZrO₂ films were grown using chemical solution deposition (CSD).⁹ The year after, vacuum-based methods—Y:HfO₂ via pulsed laser deposition (PLD) and sputter deposition—were also successfully used to fabricate films thicker than 100 nm.¹⁰ The limit of 1 μm thickness was achieved with Y:HfO₂ fabricated by PLD¹¹ and later with solution-deposited La:HfO₂² by some of the authors of the present paper.

We also reported piezoelectric properties with an effective piezoelectric coefficient d₃₃ of up to 7.7 pm/V. This is on the order of the values achieved with AlN or moderately Sc-doped AlN^12,13 and sparks hope for low-cost solution-derived hafnia-based films for piezoelectric devices. As encouraging as these findings are, common microstructural issues were also pointed out. The films are randomly oriented and exhibit small grains with diameters of around 20 nm and below. This is similar to most vacuum-based, non-epitaxial HfO₂ films in literature, however, the solution-based films are also porous with low relative densities (if reported) of around 80%. Note that this is a common issue observed by many different groups using different compositions, precursors, and solvent combinations.^14–22 Our own optimization efforts of processing parameters, such as drying, pyrolysis, and annealing prior to this work, are outlined in the Sec. S3 of the supplementary material (following basic experimental details introduced in Secs. S1 and S2). These preliminary experiments did not yield a noteworthy improvement of microstructure and electrical properties. This led us to conclude that this common observation of issues with microstructure and dielectric quality should not be dismissed as originating from a lack of efforts on the optimization of these basic parameters.

This paper starts with a derivation of a likely source of the problem and what routes CSD literature suggests to overcome it. We picked the approach of crystallization of a dilute solution after each spin coating cycle and demonstrate how it helps to orient the films and significantly improve ferroelectricity and breakdown (BD) field. Further microstructural investigations with x-ray reflectivity (XRR), transmission electron microscopy (TEM), and secondary ion mass spectroscopy (SIMS) provide insights into the microstructure formation.

To understand the origin of low density, it is good to review the crystallization process in CSD-derived thin films. Once the organics are removed at lower temperatures (ideally without crystallization appearing in parallel), the transformation from an amorphous into a crystalline phase takes place via nucleation and growth processes, with nucleation proceeding via the heterogeneous (i.e., from the interface or from the surface) or the homogeneous (i.e., within bulk of the film) path. In perovskite thin films, crystallization is typically nucleation limited. It has a strong influence on the final microstructure, with columnar or equiaxed grains prevailing in the case of heterogeneous or homogeneous nucleation, respectively. Heterogeneous nucleation is thermodynamically often preferable. However, when the crystallization temperature is significantly lower than the melting temperature of the film, heterogeneous and homogeneous nucleation events are equally possible and the latter prevails.^23,24

The influence of nucleation on the final microstructure has been studied intensively in barium strontium titanate (BST), in which granular and porous microstructure is typically observed and its dielectric properties can be dramatically enhanced if columnar films are achieved.^23,25,26 Hoffmann et al.²⁶ attributed the difficulties in obtaining columnar grain growth and dense films on platinized silicon (Pt-Si) substrates to the annealing temperature of 800 °C, which is significantly lower than the melting temperature of the BST, i.e., between 1600 and 2000 °C,²⁷ depending on the ratio of strontium to barium. Pečnik et al.,²⁵ on the other hand, were successful in inducing columnar and dense microstructure in BST films on alumina substrates with enhanced properties at 900 °C. The above-mentioned granular and porous microstructure of BST is reminiscent of what is observed in HfO₂ or ZrO₂-based CSD films. The melting temperature of HfO₂ and ZrO₂ is even higher than that of BST at around 2700 °C.²⁸ Diffusion to enable grain growth or coalescence of nuclei is expected to be comparably slow just judging from this simplified comparison of annealing to melting temperature. On the contrary, for the technologically important lead zirconate titanate (PZT), with a melting temperature between 1300 and 1550 °C,²⁹ this is less of a problem and dense films with columnar structure can be achieved at 700 °C and lower temperatures with relative ease.^23,30

Several approaches can be employed to enhance heterogeneous nucleation and promote dense microstructure at low temperatures: (a) good lattice matching between a film and a substrate and/or an interface that facilitates grain nucleation; (b) utilization of nucleation layers; (c) liquid-phase assisted processing at higher temperatures; and (d) the use of dilute solutions and crystallization of every layer.²³ 

As (a)–(c) are not readily available or sufficiently understood for HfO₂ or ZrO₂, we chose approach (d) in this study. The original paper²⁶ on BST describes the idea behind as follows: If the film thickness is below the typical grain size, the energy barrier for surface/interface nucleation becomes lowered relative to the energy for volume nucleation. Schuler et al.³¹ used this idea to derive a structure zone model for CSD films. They utilized the ratio of intrinsic crystallite size (ICS), i.e., grain size measured in the final film, and single layer thickness (SLT) to predict the microstructure. A granular structure is expected when ICS/SLT < 0.4, a layered structure when 0.4 < ICS/SLT < 1 and a columnar structure when ICS/SLT > 1. Our conventional CSD process with a 0.25 M solution that we adopted from Starschich et al.^9,32 yields ICS of ∼20 nm. To promote heterogeneous nucleation and columnar microstructure, we decided to perform the crystallization after each spin coating cycle and to dilute our solution aiming at an SLT of 5 nm in the experiments described below.

Solution and film preparation were similar to what has been described in our previous work.² A 0.25 M La:HfO₂ (LHO) solution that provided an SLT of 15 nm was diluted to 0.083 M solution. With this solution, we expected to obtain the SLT value of about 5 nm, which is well-below 20 nm, i.e., the typical grain diameter found in the SEM images of our previous work² and other CSD studies of HfO₂/ZrO₂-based films.^15–19 Commercial platinized silicon substrates (SINTEF) were used and heated to 350 °C for 5 min to remove adsorbates and cooled down to room temperature prior to spin-coating. Solutions were deposited by spin coating, and crystallization was performed by rapid thermal annealing at 800 °C for 90 s in a 1:1 atmosphere of Ar and O_2. Grazing-incidence x-ray diffraction (GIXRD), x-ray diffraction (XRD), and XRR characterizations³³ were performed, as well as scanning transmission electron microscopy (STEM), electron energy loss spectroscopy (EELS), and secondary ion mass spectrometry (SIMS). For all the experimental details, please refer to supplementary material S1 and S2.

Figures 1(a) and 1(c) show the polarization vs electric field (P-E) loops of 45 nm-thick films deposited from the 0.25 M and the 0.083 M solutions using three and 11 spin-coating cycles, respectively. The former was crystallized only at the very end and represents the conventional film. On the contrary, the latter was crystallized after every spin-coating cycle in a layer-by-layer fashion and is, therefore, referred to as L/L film in further text. For both samples, a similar wake-up behavior, i.e., an initially pinched hysteresis with low remanent polarization P_r opens and the corresponding P_r value increases during progressive cycling. However, the L/L route results in a significant improvement of the remanent polarization P_r and maximum field E_max from 7 μC/cm² to 15 μC/cm² and from below 2.9 MV/cm to 4 MV/cm (beyond that field, an early breakdown occurred), respectively. The publications on solution-deposited HfO₂-based films with the largest P_r values and negligible leakage contributions by Yoneda et al.¹⁴ and Nakayama et al.¹⁵ show P-E loops measured at 3 MV/cm or below with P_r values up to 8 μC/cm². Starschich et al. applied a maximum field of 4 MV/cm,³⁴ but often less^9,32,35 when measuring their films. Reliable P_r values between 10 and 20 μC/cm² could be extracted in their work by removing the leakage contribution via dynamic leakage current compensation (DLCC^36,37). Note that the values in the present work are obtained without any compensation.

Figure 1(b) shows a breakdown (BD) behavior of the two representative capacitors for both routes, by plotting current as a function of voltage. The BD voltage is defined as a voltage, where the current increases by more than two orders of magnitude with a voltage change of 0.1 V or less. The curves occasionally showed anomalies, marked by the oval in the graph, where current showed jump-like discontinuities, but returned to the previous values upon further field increase. These anomalies are considered “pre-breakdown events.” The inset of Fig. 1(b) provides a statistical analysis for 12 conventional and 13 L/L pads. The BD voltage of the conventionally deposited sample is only around 9.2 V (2 MV/cm), while the L/L samples broke down at around 14.8 V (3.3 MV/cm). Note, that the fields in the P-E measurements are larger than those of the BD fields because the P-E measurement is faster (ramp of 1 MV/cm in 188 ms) than the BD measurements using a linear voltage ramp (1 MV/cm in 100 s). Prominent pre-breakdown events occurred for nine out of 12 conventional samples but only for four out of 13 L/L samples.

The GIXRD patterns (supplementary material S4) indicate differences for both 45 nm-thick films. The intensity of the 002 peak is lower, and the 222 peak is absent for the L/L sample. However, GIXRD cannot properly assess the texture of a film. Hence, 100 nm-thick films were deposited and first analyzed by GIXRD as well to ensure that the patterns look similar to their 45 nm-thick counterparts. Figure 1(d) compares θ-2θ XRD patterns of the 100 nm-thick films from conventional and L/L route. The conventional film exhibits an almost random orientation. In contrast, strong 111 and 002 peaks indicate mixed 111 and 002 preferential orientation for the L/L sample. No other peaks than the 111, 222, and 002 are visible. In the reference pattern of the powder diffraction file of the orthorhombic Pca2₁ phase (#04-005-5597, ICDD Database PDF4+ v4.19, 2019), the 111 direction is 8–10 times as intense as the 002 direction (or 4–5 times if both 020 and 002 are considered). This means that the 002 texture is the more pronounced one. This significant fraction of crystallites with the polar 002 direction parallel to the applied external field can explain the significant improvement of P_r in L/L films. More experimental details including tilt-angle dependent XRD to justify our conclusions on texture are presented in the supplementary material S5.

From the same data, we also estimate tensile equi-biaxial in-plane strain (stress) of 0.7% (2.7 GPa) for the conventional sample and a slightly higher value of 0.8% (stress: 3.2 GPa) for the L/L sample using a modified sin²(Ψ) approach.³⁸ While we cannot rule out that this strain difference plays a role for the stabilization of the ferroelectric properties, we do not consider it a dominant impact compared to the texture change.

To understand the cause of improved leakage current and breakdown fields in L/L films, XRR experiments were conducted. Figures 2(a) and 2(b) show the XRR results for both 45 nm-thick films, and Table I summarizes the corresponding fit parameters. We find an enhancement of overall (relative) density from 8.0 g/cm³ (83%) for the film produced by the conventional route to 8.4 g/cm³ (87%) in the L/L film. For the latter, several peaks are observed in addition to regular oscillations. These are explained by an alternation of layers with higher and lower densities (so-called “superstructure”). A likely origin of this periodicity is the deposition scheme, which is a repetition of eleven spin-coating cycles. The sketches in Fig. 2(c) show the films' stacks used for the respective fits. The roughness of the different LHO interfaces is in the range of 0.8 nm and comparable for both samples (Table I). The fitted thickness is 46.5 nm (equaling three times 15.5 nm per spin-coating and drying cycle) for the conventionally deposited film and 48.3 nm for the L/L film. The SLT of 4.4 nm is obtained in the L/L sample, which results from a sub-layer structure consisting of 1.9 nm “bulk layer” (bulk) with 7.9 g/cm³ density and a 2.5 nm thick “capping layer” (cap) or 8.7 g/cm³. We estimate an uncertainty of 0.1–0.2 g/cm³ for this superstructure fit. For the single-layer fits, the uncertainty is generally expected to be below 0.1 g/cm³. The density values are compared to literature values of different phases in supplementary material S6. supplementary material S7 explains the choice of the terms “cap” and “bulk” and that the formation of a superstructure with the L/L route has no significant substrate dependence.

To gain further insight into the origin of the superstructure found in XRR, STEM, and SIMS, analyses were performed for the L/L film. Density alternations are indeed also visible in high-angle annular dark-field [HAADF, see Fig. 3(a)] STEM data. In HAADF, lighter color represents regions with higher density, if we assume that the lamella thickness does not change significantly. As the integrated HAADF intensity data in Fig. 3(b) show, the oscillations resemble the sub-layer thicknesses shown in Table I, while such prominent oscillations are not observed for the conventional sample [Figs. 3(c) and 3(d)]. For the interested reader, supplementary material S8 complements these data in conjunction with annular bright-field and fast Fourier transforms as well as EELS.

To verify that the density alternations stem from different local atomic densities and not variations in the atomic composition of atoms like Hf and La, we employed low-energy SIMS, which offers higher sensitivity and depth resolution. Figures 3(e) and 3(f) show the elemental composition for La and Hf through the evolution of La⁺, LaO⁺, Hf⁺, and HfO⁺ signals together with that of Al⁺ to monitor when sputtering reaches the sapphire substrate (chosen for SIMS because a smooth substrate is needed for optimal depth resolution). After about $1.4×104$ s, the sapphire substrate is reached as the dashed vertical line in Fig. 3(e) indicates. While the general course with thickness does not show any significant difference for La- and Hf-related species, all traces exhibit periodic modulations of the intensity. Figure 3(f) shows a magnification for the La-O⁺ and the Hf-O⁺ signals. Maxima and minima of both curves coincide, which show that there is no gradient in the La/Hf ratio. Eleven of these periods can be found (see dashed vertical lines), which is in agreement with the number of spin-coating cycles performed for the L/L sample. Therefore, we conclude that the oscillations in STEM intensity originate from the density modulations found in XRR and not from changes in the La:Hf ratio and reflect different sputter yields during the SIMS measurement with direct impact on the detected counts of the secondary ions (sputtering yield).

Having gathered structural and chemical data, we want to judge our findings in the frame of the zone model by Schuler et al.,³¹ briefly reviewed at the beginning of this paper. XRR and STEM evidence the superstructure character of the L/L films stack. XRD shows that a significant degree of orientation was achieved. Nonetheless, the film looks like composed of small crystallites in the STEM images. Therefore, we conclude that the film exhibits a “layered structure,”³¹ as a transition from a granular to a columnar microstructure. As heterogeneous nucleation most probably starts from the bottom interface (substrate/film), the bottom layer is expected to be denser.

An alternative explanation for the occurrence of denser (capping) layers (often termed as “crust”) is based on chemical origins as described in detail in Refs. 18, 39, and 40. Several different underlying causes were proposed, including evaporation of the solvent at the surface, gradient formation of the cation ratio due to differences in the solubility and differences in network forming abilities of precursors, as well as carbonate formation from adsorbed CO₂ during or after annealing. With the repetitive procedure of CSD, these phenomena will also lead to a superstructure of alternating densities. Such chemical effects can also explain the occurrence of lower- or higher-density capping layers. However, if only chemical effects would play a role, the occurrence of a superstructure would not be coupled to the observation of an oriented film. A summary of papers showing superstructure formation in HfO₂/ZrO₂-based films is provided in supplementary material S9 for the interested reader.

Figure 4 compares these two models from the literature. Additional XRR experiments (see supplementary material S7) with different deposition schemes and thermal treatments indicate that both nucleation-related (Schuler's “layered structure”) and chemical effects (caps that are sometimes referred to as “crust”) can be observed. Indications for a “layered structure” are: (1) As Fig. 4(a) shows, the pores of two subsequent layers are often aligned in vertical direction similar to what Refs. 23 and 26 show for the transition from granular to layered structure in BST. (2) A significant degree of preferential orientation is observed. For chemical effects, the following indications were observed: (1) Superstructure occurs already in amorphous phase [Fig. S8(b)]. It becomes more pronounced with higher drying temperatures and prolonged times. (2) The experiments in Fig. S7(a) and Table S2 show that a ∼2.5 nm thick cap also occurs on top of a 13 nm thick bulk layer (0.25 M non-dilute solution) and that this cap is denser than the bulk.

We want to note that there are indications that the two above models might not capture the full ensemble of mechanism behind the superstructure. For instance, Fig. S7(b) shows that when omitting the cap layer at the very top of the L/L film stack during fitting, some features of the XRR pattern are better resembled than with the top cap in place. Future work to complement our studies is needed to further detangle the complex mechanisms behind microstructure formation in HfO₂-based solution-derived thin films.

In summary, we demonstrated a way of improving the density, preferential orientation, and ferroelectric properties of solution-derived La:HfO₂ thin films. P_r was doubled from 7 to 15 μC/cm², while the breakdown field increased by more than 50%. A dilute spin-coating solution of 0.083 M instead of 0.25 M was used to reduce the deposition per spin coating cycle from ≈15 nm to ≈4.5 nm, and annealing was performed every spinning instead of every three cycles. This reduction of film thickness per anneal below the typical grain radius enhanced heterogeneous nucleation compared to homogeneous nucleation. A larger degree of preferential orientation of the L/L films is evident from both XRD and STEM and is in good agreement with the observation of a doubled P_r. XRR and STEM further show that a “capping layer” of around 2.5 nm thickness forms between subsequently annealed CSD layers. This layer is denser than the corresponding “bulk” layer below it, but still not well-above 90% of the theoretical density of LHO. While our experiments show indications for the development of a “layered structure” according to the microstructure zone model by Schuler et al.,³¹ we also find evidence for chemical effects similar to what Page et al.^18,39,40 reported. The main improvement of orientation and density clearly stems from nucleation-related effects, i.e., a partial heterogeneous nucleation of our films. This highlights the importance of orientation control in thin films of the novel HfO₂/ZrO₂-based ferroelectric materials. We believe that our findings will motivate further studies of the remaining routes to promote heterogeneous growth in these novel ferroelectric thin films.

See the supplementary material for additional data of GIXRD, tilt-angle dependent XRD, XRR, STEM, EELS, and a discussion of a theoretical density and chemical effects can be obtained online.

Fonds National de la Recherche Luxembourg is acknowledged for the financial support of N.A., T.G., and E.D. within the frame of the project MASSENA (FNR-PRIDE/15/10935404) and of T.S. through the project CO-FERMAT (FNR/P12/4853155/Kreisel). T.S. is also grateful for the financial support by LIST via the self-funded project SF_MRT_CSDFO. Stephanie Girod and Aymen Mahjoub are acknowledged for their help with electrode patterning and electrode deposition. The authors also wish to thank Brigita Kmet for STEM sample preparation.