Introduction

Ferroelectric materials, characterised by a spontaneous polarisation that can be reversed by an external electric field, have long been a focus of research in materials science1. Their unique ability to switch polarisation provides a wide range of applications, including non-volatile memory devices2. Traditionally, perovskite oxides such as BaTiO3 and Pb(Zr,Ti)O3 have been known as typical ferroelectric materials3. However, these materials face intrinsic limitations when scaled to nanometre dimensions; for instance, the ferroelectric properties in perovskites generally degrade at film thicknesses of tens of nanometres, which hampers further miniaturisation of memory devices4. In addition, the growth techniques for these materials are less compatible with complementary metal-oxide-semiconductor (CMOS) processes, due to the high temperatures required and lattice-mismatched substrates5,6. To overcome the limitations imposed by film thicknesses, recent studies have reported an inverse size-scaling ferroelectric transition in centrosymmetric DyScO3 occurring below a critical thickness of ~5 nm, driven by surface charge transfer and symmetry breaking, as well as switchable multistate ferroelectricity in NH3-based topotactically transformed La0.7Sr0.3MnO3-δ with quasi two-dimensional ferroelectric layers thinner than 2 nm7,8.

A breakthrough in the field came with the discovery of ferroelectricity in fluorite-type oxides such as HfO2, ZrO2, and Hf0.5Zr0.5O2 (HZO)9. Unlike conventional perovskites, ferroelectricity in these materials arises from shifts of oxygen anions rather than from the displacement of cations within a crystalline structure. HfO2, already a standard high-\(\kappa\) material in CMOS processes10, demonstrated robust ferroelectric switching in Si-doped films, as first reported by Böscke et al11. This discovery not only spurred extensive investigations into the fundamental mechanisms governing ferroelectricity in these systems but also highlighted their potential for enabling devices with ultrathin active layers—down to nanometre or even just a few atomic layers—thus overcoming the critical thickness limitations inherent to perovskite ferroelectrics12.

In fluorite-type oxides, the ferroelectric phase is typically associated with a non-centrosymmetric orthorhombic structure (space group Pca21), in which displacements of oxygen ions with respect to the heavy cations (e.g., Hf and Zr) produce a net dipole moment11. Notably, these materials often exhibit polycrystalline structures, i.e., a texture of nanometre-sized grains with different crystalline phases, even if fabricated by an epitaxial-film growth technique such as pulsed laser deposition (PLD)13,14,15,16, including metastable polar and nonpolar phases due to the near-degeneracy of these phases. Consequently, unveiling these crystalline structure characteristics and their transitions, and engineering polarisation switching in the ferroelectric phase of fluorite oxides is critical not only for advancing fundamental understanding but also for optimising their performance in memory applications because these dynamics are inevitably related to the wake-up and/or fatigue phenomena12.

Scanning transmission electron microscopy (STEM) has been emerging as a powerful tool in this regard. Modern aberration-corrected STEM techniques enable direct imaging of individual atomic columns with sub-Ångström resolution, thereby providing detailed insights into the local atomic structure17. High-angle annular dark-field (HAADF) STEM is a common STEM imaging technique for materials characterisation that excels in visualising heavy elements such as cations, and hence is useful for detecting the cation-induced polarisation structures18. As for the oxygen sites, which plays a pivotal role for the fluorite-type ferroelectrics, phase-contrast-based techniques can be applied to visualise them. By means of these, recent STEM analysis of fluorite ferroelectric materials not only identified crystalline phase of individual nano-sized grains15 but also observed polarisation structures, some of which visualised the structure evolution such as polar-nonpolar phase transition and polarisation switching triggered by an electron irradiation or electric-field biasing19,20,21. In these STEM observations, it was reported that a nonpolar tetragonal phase plays an important role as an intermediate or interfacial structure during the switching, which has also been observed by in-situ grazing incidence X-ray diffraction measurements22,23. However, the detailed atomic structures and dynamics of these transition pathways, including intermediate states, remain unclear, and consequently the means to selectively facilitate or suppress specific pathways to obtain the desired ferroelectric phase and polarisation switching paths remain unknown.

In this study, we unveil the atomistic pathways of 180° and 90° polarisation switching and their intermediate structures involved in the fluorite ferroelectrics. Critically, we reveal the accessible pathways to be cation-species-dependent, by comparing the atomic-structure dynamics of freestanding ZrO2 and HZO membranes. To investigate the detailed polarisation structures and switching pathways, we employ an optimum bright-field (OBF) STEM technique24, which can visualise both oxygen and cation sites simultaneously with high sensitivity. Our approach first demonstrates that OBF STEM can effectively visualise the local atomic structure within these membranes, resolving both the polar and nonpolar phases. Building on this static analysis, we then explore the dynamic behaviour of these atomic structures under high beam-current electron irradiation. In particular, our observations in ZrO2 membranes reveal the 180° and 90° polarisation switching within the ferroelectric structure is mediated by reversible phase transitions between polar and nonpolar states. Furthermore, we perform a higher-speed OBF STEM data acquisition, which allows us to capture two different nonpolar intermediate structures along the 180° and 90° polarisation switching pathways, providing unprecedented insight into the transient states that occur during the switching process. We also perform the observation of an HZO membrane, shedding light on how the different atomic species of cation sites impacts the dynamic behaviour regarding the phase transition and polarisation switching. The energetic landscape governing these transitions and switching processes is elucidated by the density functional theory (DFT) calculations, complementing our experimental findings. In addition to the detailed structure evolution analysis during the phase transition and polarisation switching of ZrO2 and HZO, our approach offers a critical guideline to engineer these dynamics and realise the advanced fluorite ferroelectrics.

Results

Visualisation of the static polarisation structure at the atomic scale

The projected atomic-structure models along the [001] zone-axis of nonpolar tetragonal (space group P42/nmc), polar orthorhombic (space group Pca21), and anti-polar orthorhombic (space group Pbca) phases in the fluorite ZrO2 and HZO are shown in Fig. 1a. In the tetragonal phase, cation (Hf/Zr) sites form a cubic-like lattice along this projection and oxygen sites are located in the centre of this cation lattice, giving a nonpolar structure as shown in Fig. 1a. In the orthorhombic phase, on the other hand, the distance in the [100] direction between the tetragonal (100) and (200) cation planes alternates from narrow (2.36 Å) to wide (2.69 Å), in this projection. Moreover, in the wide layer, the oxygen site shifts off-centre within the wide cation lattice in the [010] direction, while the oxygen sites in the narrow layer remain centred. This polarisation due to oxygen shifts causes the ferroelectricity in (H)ZO (where we use (H)ZO as shorthand to refer to both ZrO2 and HZO). These fluorite-type oxides have distinct polarised atomic layers in between oxygen-centred nonpolar spacers, making the polar domain stable even down to a single atomic layer in width25. In the anti-polar orthorhombic phase, the direction of the oxygen shift alternates as shown in Fig. 1a, resulting in the anti-polar behaviour.

Fig. 1: Atomic-structure model and atomic-resolution OBF STEM images of polar/non-polar phases in (H)ZO membranes.
Fig. 1: Atomic-structure model and atomic-resolution OBF STEM images of polar/non-polar phases in (H)ZO membranes.
Full size image

a Atomic-structure models of nonpolar tetragonal, polar orthorhombic, and anti-polar orthorhombic phases along the projection perpendicular to the (001) plane. Cation (Zr/Hf) and oxygen sites are shown as green and red spheres, respectively. The projected cation lattice distances are also shown, using values obtained from DFT calculations. Experimental OBF STEM images of (b) tetragonal in a ZrO2 membrane (5 nm thick), (c) polar orthorhombic in a HZO membrane (5 nm thick), and (d) anti-polar orthorhombic phases in a ZrO2 membrane (10 nm thick). Yellow arrows indicate the polarisation direction of each atomic layer in the orthorhombic phases. The atomic-structure models are also overlaid. Scale bar: 1 nm.

We first analysed these features inside the (H)ZO membranes using OBF STEM. The OBF STEM technique can visualise atomic structures including both heavy and light elements with high sensitivity, making it powerful for probing the oxygen-induced polarisation structures described above. For the structure analysis of ZrO2 and HZO, nanometre-thickness membranes were fabricated using the pulsed laser deposition (PLD) method and then transferred onto TEM specimen grids, with details described in the Method section. Figure 1b–d show the resulting experimental OBF images of each phase inside the ZrO2 (5 and 10 nm thicknesses) and HZO (5 nm thickness) membranes, visualising both the Hf/Zr and O atomic sites. As shown in Fig. 1b, in the nonpolar tetragonal phase observed in the 5 nm ZrO2 membrane, the higher intensity spots, which correspond to the Hf/Zr sites, are arranged in a square lattice in projection, and lower intensity spots, corresponding to the oxygen sites, are located in the centre of this square cation lattice. On the other hand, in the orthorhombic phases, the oxygen sites show a dumbbell-like elongated intensity distribution compared with the tetragonal phase and are located in the off-centred position inside the wider-spaced layers of the cation lattice as shown in Fig. 1c, d. It is also evident that the oxygen shift is aligned along one direction in the polar orthorhombic phase (Fig. 1c), which was observed in the 5 nm thick HZO membrane, whereas the oxygen shift direction alternates 180-degree in the anti-polar phase observed in the 10 nm ZrO2 membrane (Fig. 1d). These results show that the OBF STEM technique is promising for probing the atomic-scale oxygen-triggered polarisation structures in the fluorite oxides. It is noted that recent atomic-structure investigations have revealed that perovskite oxides can also exhibit ferroelectricity due to the oxygen displacements relative to the cation sublattice despite the traditional picture of cation-induced polarisation26, indicating that observations of the atomic displacements of both cations and anions are essential for analysing the ferroelectricity.

Reversible phase transition and polarisation switching under high electron dose illumination

In electron microscopy, electron illumination can cause dynamic structure changes inside materials such as phase transitions27,28, grain boundary migration29, and polarisation switching30. In the case of fluorite ferroelectric materials such as (H)ZO, it is reported that electron irradiation can trigger the phase transition and/or polarisation switching19. Observing these structure evolutions and their transition pathway at the atomic scale is critical for understanding the ferroelectric mechanism and how it might be manipulated and optimised for practical applications. Here, we use OBF STEM with a high illumination beam current to capture the dynamic behaviour of our ferroelectric membrane sample at the atomic-scale, where a mechanism to induce these dynamic events via the electron beams is discussed in the Supplementary Note 4. In typical STEM measurements, such as shown in Fig. 1, the electron beam current is around 10-20 pA. Here, we deliberately use a 100 pA probe current to stimulate the phase transition and record as a movie, a sequence of images from the same field-of-view (FOV). We have confirmed a clear threshold of the beam current to trigger the dynamics, providing a strategy to choose the proper current for the experiments performed in the present study as shown in Supplementary Note 6.

To establish a baseline, we first use a relatively low probe current of 15 pA. Figure 2a, b shows the corresponding OBF STEM image of the nonpolar tetragonal grain of the 5 nm thickness ZrO2 membrane, as discussed in Fig. 1. We obtained thirty images sequentially with a scanning speed of 500 ns/pixel with a 2048 \(\times\) 2048 pixels scan (equivalent to three seconds per frame or 0.33 frames per second) from the same field-of-view, resulting in 2.7 \(\times\) 103 e−/Å2/frame as a total dose per area. We then averaged these images after aligning them to correct for stage drift. During the sequential acquisition, the tetragonal grain retained its original atomic arrangement as shown in Supplementary Movie S1, indicating that the electron irradiation at 15 pA did not change the atomic structure.

Fig. 2: Direct observation of reversible phase transition and polarisation switching in ZrO2.
Fig. 2: Direct observation of reversible phase transition and polarisation switching in ZrO2.
Full size image

a OBF STEM image of a tetragonal grain in the ZrO2 membrane. Scale bar: 1 nm. b OBF STEM image obtained by cropping and magnifying the cyan-dotted rectangular region shown in (a). The atomic-structure model is overlaid. Scale bar: 1 nm. Sequentially obtained OBF STEM images which show (c) (0–72 s) the phase transition between tetragonal and orthorhombic phases, (d) (99–108 s) 180-degree polarisation switching, and (e) (216–225 s) 90-degree polarisation switching. The time count is set so that the first image acquisition is 0 s just after increasing the probe current to 100 pA. Scale bar: 1 nm. f Time-series showing the ratio of the nonpolar and polar phases in the whole nanocrystal. Yellow, red, green and blue bars indicate the ratio of the phase polarised along [0−10], [010], [100], and [−100] directions, respectively. The relationship between the colours and the directions are also indicated in the right-hand side panel of (c). The nonpolar tetragonal phase is shown as grey bars. The events shown in (c–e) are highlighted on this timeline.

We then increased the probe current to a high current of 100 pA by changing monochromator focus value only (while keeping all other electron-optical parameters and resultant observation condition the same) and obtained another image sequence from the same initially tetragonal grain observed above, with results shown in Fig. 2c–e. Here, the electron dose per area was also increased to 1.8 \(\times\) 104 e−/Å2/frame. At 69 s, the tetragonal grain changed into the orthorhombic phase that has polarised oxygen shifting off centre relative to the cation lattice along the tetragonal [010] direction, which is evident in the OBF STEM image (see also Supplementary Movie S2). After an extra scan, this structure changed back to the original tetragonal structure reversibly.

After several scans taking a further ~30 s, the tetragonal grain changed back to an orthorhombic phase (at 102 seconds), however, this time the polarisation structure is still along tetragonal [010] direction but has different polarisation ordering from what was observed in Fig. 2c. After a further beam exposure (at 105 s), this grain remained in the orthorhombic phase but a 180-degree polarisation switching occurred in the two leftmost atomic layers shown in Fig. 2d. Before and after this polarisation switching, the arrangement of the cation lattice remained unchanged, but the oxygen sites are shifted locally relative to the Zr cation lattice. After an additional scan, this orthorhombic phase changed back to the tetragonal phase reversibly, with the oxygen located at the centre of the cation lattice. As shown in the Supplementary Movie S3, the 180-degree polarisation switching and phase transition between tetragonal and orthorhombic phases were observed multiple times and were reversed in the subsequent scans.

After additional electron exposure, 90-degree polarisation switching also started to be triggered as shown in Fig. 2e and Supplementary Movie S4. The grain had a tetragonal atomic structure at 216 s and then, after a single scan, became a polarised orthorhombic structure where the oxygen sites are shifted in the tetragonal [100] direction with reference to the cation sites (horizontal direction in the figure orientation). At 222 s, the orthorhombic structure has changed into an orthorhombic structure polarised along tetragonal [010] direction, showing a different type of switching path from previous results. As discussed in Fig. 1, the polar orthorhombic structure has a uniform cation-cation distance along the polarisation direction and alternately wide and narrow distances perpendicular to it. During this 90-degree switching process, not only the oxygen sites but also the cation sites change their positions, requiring a larger shift of the atomic positions than that required in the 180-degree switching. The distinct change of cation lattice spacing and resulting lattice strain are also confirmed by the lattice strain measurement based on the simultaneously obtained ADF images as described in Supplementary Note 2. Here it is clearly visualised that the orthorhombic phase has two different sublattices strained alternately perpendicularly to the polarisation direction, while the tetragonal phase has uniform strain distributions. Furthermore, it is visualised that the grain contains the tetragonal phase near the perimeter of the grain despite the bulk area having transformed to the orthorhombic phase in ZrO2. This is confirmed via the oxygen position determined approximately from OBF STEM images and strain distribution determined from the ADF STEM images, as shown in Supplementary Figs. S3 and S5.

From the time-series of OBF STEM images, a ratio of the volume of polar and nonpolar phases within the whole ZrO2 nanocrystal was also counted as a function of time, as shown in Fig. 2f. It can be seen that the initially tetragonal nanocrystal takes some time, (i.e. accumulated electron dose), before the first reversible phase transition between tetragonal and orthorhombic phases occurs (Fig. 2c) and further time/dose for the first 180-degree switching event to occur (Fig. 2d) and even further time/dose for the 90-degree switching events to occur, as discussed above. This suggests qualitatively that the activation energy required is lowest for the tetragonal to orthorhombic phase transition, higher for the 180-degree switching event and highest for the 90-degree switching event.

We also performed this (non)polar-phase-ratio statistical analysis under higher-dose electron illumination over several ZrO2 grains, as shown in Supplementary Fig. S8, and find the dynamic behaviour and frequency of the switching events are consistent with what was shown in Fig. 2 (see Supplementary Note 5 for details). It has been observed by Lomenzo et al using band-excitation piezoresponse force microscopy (BE-PFM), that the nano-sized grains in ZrO2 independently undergo the reversible tetragonal-orthorhombic phase transition under applied electric field, which is responsible for the anti-ferroelectric nature of ZrO2 such as the double hysteresis loop in the P-E measurement31. Given the switching observed in the present study is also driven by an electric field induced by electron irradiation (see the Supplementary Note 4), it can be said that the reversible phase-transition and switching events visualised here at the atomic-level, provide a direct atomic-scale counterpart to the electric-field-driven dynamics reported in the literature using BE-PFM. We also demonstrate how the switching dynamics can be manipulated via different electron-probe scanning strategies as discussed in Supplementary Note 6. Unconventional scan patterns and/or electron dose modulation using a high-speed electrostatic beam-blanker, beyond the uniform raster scan of the STEM probe, may allow for the further manipulation of switching dynamics32,33,34.

Intermediate structures during the polarisation switching

In the previous section, we observed reversible phase transitions and polarisation switching in the ZrO2 membrane by means of OBF STEM with a high probe current. To provide a detailed understanding of the polarisation switching pathway, we performed OBF STEM measurements with a higher-speed scan. With a scan speed of 200 ns/pixel and sampling of 512\(\times\)512 pixels, we obtained sequential OBF images at 0.0629 seconds/frame or 16 frames-per-second. Figure 3a shows the tetragonal grain observed under an initial 15 pA probe current illumination (5.5 \(\times\) 102 e-/Å2/frame), which shows the nonpolar oxygen-centred atomic configuration consistent with the result shown in Fig. 2a. We focus on the single atomic layer along the tetragonal [100] direction as highlighted by the cyan dotted rectangular in Fig. 3a.

Fig. 3: Observations of intermediate structures during the 180- and 90-degree polarisation switching in ZrO2.
Fig. 3: Observations of intermediate structures during the 180- and 90-degree polarisation switching in ZrO2.
Full size image

a OBF STEM image of a tetragonal grain obtained under the initial lower probe current (15 pA). Scale bar: 1 nm. b Time-series of OBF STEM images taken with higher probe current (150 pA), probing the 180-degree switching (16 fps or 0.0629 seconds/image). Each frame is cropped from the region indicated in (a). The assigned phases are tetragonal (T), right-polarised orthorhombic (O(→)), left-polarised orthorhombic (O(←)), and nonpolar orthorhombic (O’), respectively. The domain wall between O(→) and O(←) is highlighted by dotted orange lines. Structure models are overlaid, and one of the oxygen sites is highlighted with yellow-dotted circles or arrows. Scale bar: 1 nm. c Simulated OBF STEM images of each phase observed in (b) with overlaid structure models, where green and red spheres represent Zr and O sites, respectively. The half-red and half-white spheres indicate the O sites with a half occupancy. d OBF STEM image of another tetragonal grain obtained under the initial lower probe current (15 pA). Scale bar: 1 nm. e Time-series of OBF STEM images, probing the 90-degree polarisation switching. Frame rate and probe current illumination are the same as (b) and each frame is cropped from the area highlighted in (d). The polarisation directions are indicated by coloured arrows. Schematics of atomic structures are also shown for each frame, corresponding to the area featured in (e). Scale bar: 1 nm. f Schematic of the structure evolution for 180- and 90-degree switching, which is divided into initial, intermediate, and final structures, showing the different intermediate phases. Grey boxes indicate cation sublattices of the orthorhombic and tetragonal phases. Coloured dots and arrows specify the displaced oxygen sites and their polarisation direction. Nonpolar oxygen sites are denoted with grey dots, which indicate the nonpolar phase with oxygen sites centred in the cation lattice (e.g., tetragonal phase and nonpolar spacer layer in orthorhombic phase) and the nonpolar O’ phase observed in (b).

Following the initial low current OBF STEM image of the tetragonal grain, we increased the illumination current to 150 pA (5.5 \(\times\) 103 e−/Å2/frame) and took a higher-speed time-series OBF STEM images. Figure 3b and Supplementary Movie S5 show the sequence of OBF STEM images, focusing on the single atomic layer inside the grain, where a more extended field-of-view is shown in the Supplementary Note 10. As observed in the lower probe current (15 pA) condition, the layer shows the tetragonal structure at 0.0 s. At 0.1887 s, while the atomic layer mostly has the tetragonal structure, it can be seen that some of the oxygen sites in the right-hand side have started to shift away from the centre of the cation lattice, showing a mixture of the two phases (see Supplementary Fig. S12 for details). It is noted that the oxygen-site displacements with respect to the Zr sites can be clearly observed by intensity line profiles of the OBF images as discussed in Supplementary Note 7. At 0.3774 s, the oxygen sites have shifted completely in one direction parallel to tetragonal [100] and the atomic layer has changed to the polarised orthorhombic structure as observed in previous sections. After nine scans (at 0.9435 s), the oxygen sites on the right-hand side of the layer started to shift in another direction along tetragonal [−100] whereas the left-hand side of the layer still shows the polarisation toward the [100] direction. After a further scan (at 1.0064 s) the left polarised domain propagated into the left-hand direction. Here, the area between the right- (O(→)) and left- (O(←)) polarised domains has a transient atomic structure (O’) during the 180-degree polarisation switching, where the image intensity at the oxygen site is lower than that of the polar orthorhombic phase and distributed either side of the vertical tetragonal (100) and (200) planes of the cation lattice.

During the 180-degree domain switching in the fluorite oxide materials such as (H)ZO, it has been suggested that the domain wall has a nonpolar interlayer between two oppositely polarised domains, which has orthorhombic Pbcm structure for the head-to-head boundary and tetragonal P42/nmc structure for the tail-to-tail boundary35. To investigate this, we compare our experimental OBF STEM images with simulated images of nonpolar tetragonal, polar orthorhombic (polarised to right and left directions respectively), and non-polar orthorhombic phase (space group Pbcm) within single layers, where the polarised oxygen shift occurs when the structure has polar phase, as shown in Fig. 3c. In the crystallography literature, the nonpolar orthorhombic Pbcm structure in fluorite-type materials is considered as a mixture of disordered domains of ferroelectric Pca21 phase with two opposite polarisation directions, providing a centrosymmetric and nonpolar phase as an average whose oxygen sites are split into two half-occupied sites distributed either side of the cation planes, as shown in Fig. 3c36,37. In some theoretical calculations of the switching pathway, the Pbcm structure is assumed to have non-split oxygen sites located in a straight line with the cation planes35,38. Instead, we employ the former Pbcm structure due to its better description of what was observed in the OBF image. Here, in contrast to the tetragonal and polar orthorhombic phases, in the nonpolar orthorhombic Pbcm structure, the oxygen site splits into two sites with a half occupancy (shown as half-red and half-white spheres in the structure model)39, appearing as even lower intensity spots in the simulated OBF image. The experimental OBF image at 1.0064 s shown in Fig. 3b corresponds approximately to the simulated intensity distribution for the nonpolar orthorhombic phase around the head-to-head domain boundary region. After another scan (at 1.0693 s) the left-polarised domain extends into the left-hand side of the layer and a few oxygen sites at the left edge of FOV show diffuse but centred intensity, which is interpreted as the mixture of T and O phases similar to what was observed at 0.1887 seconds. At 1.6983 seconds this atomic layer turns back completely to the tetragonal phase, where the oxygen sites are located in the centre of the cation lattice. This result shows that the oxygen site is bridging across the cation lattice plane for the 180-degree switching via the nonpolar orthorhombic Pbcm phase.

We next show higher-speed OBF STEM imaging of 90-degree polarisation switching. Figure 3d shows another tetragonal grain in the same ZrO2 membrane, whose atomic structure was identified by low probe current OBF STEM imaging. This grain first showed 180-degree domain switching after increasing the probe current and then started 90-degree switching subsequently as also shown in Fig. 2e. Figure 3e shows the time-series of 90-degree switching, starting from the vertically polarised structure, whose time is counted as the 0.0 second frame (see also Supplementary Movie S6). In this image, the oxygen sites are shifted along the tetragonal [010] direction (vertically with respect to the cation sites). In the next scan (0.0629 s), these oxygen sites are shifted to the centre of cation lattice, showing the tetragonal phase. At 0.1258 seconds, the oxygen site within a single layer has shifted to the [100] direction (right-hand side), showing the locally left-polarised orthorhombic phase within that monolayer. After another scan at 0.1887 seconds, oxygen sites within the same layer shifted back to the centre of the cation lattice, corresponding to the tetragonal phase again. We illustrate this sequential structure evolution as a series of atomic-structure models below each OBF STEM image, corresponding to the region in the cyan dotted box as highlighted in Fig. 3e. This result shows that the oxygen sites are moving around inside the cation sublattice during the 90-degree polarisation switching: the shifted oxygen sites within the polar orthorhombic phase move to the centre, and then shift perpendicularly to the other side, indicating that the 90-degree switching occurs via the nonpolar oxygen-centred tetragonal phase.

From these results, our OBF STEM observations show that the oxygen site behaves differently between the 180- and 90-degree domain switching via distinct nonpolar intermediate states, as shown schematically in Fig. 3f. In the case of 180-degree switching, several oxygen sites collectively cross the cation lattice plane and form the nonpolar intermediate phase which extends across a few nanometres. In contrast, in the case of the 90-degree switching, each oxygen site is confined and moves locally within the individual cation sublattice.

Effect of different cation species

We observed the polarisation switching pathways in the pure ZrO2 membrane in the previous sections. In addition to the ZrO2, ferroelectric fluorite oxides with other cation species are of significant interest because the other cation species may modify the ferroelectric properties11,40, with the potential to engineer advanced ferroelectric devices. Here we compare ZrO2 with a HZO membrane, which is a well-known and promising ferroelectric material. In this material, the Hf and Zr are occupying cation sites, which can be considered as a binary of ZrO2 and HfO2, exhibiting robust ferroelectricity with a stable orthorhombic phase41.

We first obtained an OBF STEM image of a tetragonal grain in the HZO membrane with a lower probe current of 15 pA, which is shown in Fig. 4a. As well as the ZrO2 membrane, the oxygen sites are located at the centre of the cation lattice, showing the nonpolar structure. We then repeated the higher probe current experiments using the same approach and experimental conditions used for the ZrO2 membrane. However, after increasing the probe current to 100 pA, the tetragonal grain immediately changed into the polar orthorhombic phase with just a single scan (denoted as 0 second). Furthermore, unlike the ZrO2 membrane, the orthorhombic phase did not change back to the tetragonal phase, nor was 180-degree polarisation switching observed even after more than 200 s of irradiation with 100 pA probe current. At 213 seconds, we observed a 180-degree switching of just a single atomic layer. However, phase transitions back to the tetragonal phase and 90-degree switching were not observed. This result indicates that the reversible nature of the phase transition between nonpolar tetragonal and polar orthorhombic phases is absent in the HZO membrane (at least under this illumination condition). In terms of the polarisation switching, the 180-degree type can still occur but even this is rare compared with the pure ZrO2 membrane, and the 90-degree switching and the phase transition to tetragonal phase did not occur at all. This switching property was consistent over multiple grains as discussed in Supplementary Note 8, showing the robustness of the ferroelectric phase in HZO.

Fig. 4: Observation of non-reversible phase transition and polarisation switching in an HZO membrane.
Fig. 4: Observation of non-reversible phase transition and polarisation switching in an HZO membrane.
Full size image

a OBF STEM image of a tetragonal grain in the HZO membrane. b OBF STEM image of the polar orthorhombic phase observed in the same grain as (a) just after increasing the probe current into 100 pA. c, d Cropped and magnified OBF STEM images obtained from the cyan dotted area in (a, b), respectively. Polarisation directions are indicated by coloured arrows. e OBF STEM image cropped from the same area as (c, d) after 213 s electron illumination with the 100 pA probe current. Scale bars: 1 nm.

In addition to the HZO, we also performed observations of a Y-doped HfO2 (Hf0.93Y0.07O2; HYO) membrane. To a tetragonal grain in the HYO specimen, we illuminated with electron beams following the same procedure as the ZrO2 and HZO. However, despite a much longer exposure time of 810 s with the 100 pA beam current, the tetragonal phase did not show a phase transition, nor did it change with a beam current of 200 pA, as described in Supplementary Note 9. This demonstrates a higher phase stability of the tetragonal phase in HYO compared with ZrO2 and HZO.

Energetic landscape of structural transitions

Transition between nonpolar tetragonal and polar orthorhombic phases

To clarify the different structural behaviour of the two membranes, ZrO2 and HZO, under electron irradiation, we performed density functional theory (DFT) calculations. As summarised in Fig. 5a and Table 1, we calculated the relative energy of the tetragonal phase with respect to the orthorhombic phase in ZrO2 and HfO2, and the energy barrier to polarisation switching and phase transitions in both materials. It is technically difficult to evaluate the energy landscape of HZO with DFT calculations due to the likely random arrangement of Hf and Zr on the cation site and the periodicity constraints of the calculation. Hence, we calculated energies for pure ZrO2 and HfO2 to obtain insights into HZO, which can be considered as a mixture of ZrO2 and HfO241. As shown in Fig. 5b, we also schematically summarise the transition/switching pathways accessible in the ZrO2 and HfO2 membranes by indicating which path can be triggered in each material as per the following discussion. As shown in Table 1, the relative energy of the tetragonal phase with respect to the orthorhombic phase of HfO2 is higher than that of ZrO2, indicating that the tetragonal phase is more unstable in HfO2. As shown in Fig. 5a and Table 1, whereas the energy barrier of the phase transition from tetragonal to orthorhombic is comparable between ZrO2 and HfO2, the barrier from orthorhombic to tetragonal shows a much higher value for HfO2 than for ZrO2. These results agree well with the experimental OBF STEM observations: the phase transition between these two phases was observed to be reversible in ZrO2 but irreversible in HZO, that is, in HZO the tetragonal phase can transition to the orthorhombic phase but not vice versa. The difference in the phase stability is also manifested by the distinct lattice-strain distributions of ZrO2 and HZO as discussed in the Supplementary Note 2. This result corresponds to our previous findings that the HZO membrane with a 5 nm thickness mostly has orthorhombic grains while the ZrO2 membrane mainly has tetragonal grains, even though both films have approximately the same thickness13,14.

Fig. 5: DFT-calculated energy landscapes of phase transition and polarisation switching in ZrO2 and HfO2.
Fig. 5: DFT-calculated energy landscapes of phase transition and polarisation switching in ZrO2 and HfO2.
Full size image

a Energy diagrams of O → T phase transition, 180-degree polarisation switching, and 90-degree polarisation switching in ZrO2 and HfO2. Two types of pathways are calculated for 180-degree switching, namely cross-type and inner-type. b Schematic diagram showing the possible transition pathways between nonpolar tetragonal phase and polar orthorhombic phases along four different polarisation directions. The dotted arrows in HfO2 indicate the possible but relatively rare pathway.

Table 1 DFT-calculated energy barriers of phase transition and polarisation switching in ZrO2 and HfO2
Full size table

While a polar-nonpolar phase transition between the tetragonal and orthorhombic phases is investigated in the present study, a phase transition from the orthorhombic to a nonpolar monoclinic phase has been reported in the ZrO2 during electron irradiation in the literature20. In that case, a lattice-deformation process and ferroelasticity play a critical role in attaining the reversible polar-nonpolar phase transition. It is also observed that the monoclinic structure in the HfO2 can have a polarised structure in the presence of twin defects without breaking centrosymmetry42.

180-degree polarisation switching

For the 180-degree polarisation switching, we observed that the oxygen sites shifted by crossing the cation lattice planes. In the literature describing DFT calculations, two mechanisms are suggested for the 180-degree switching, namely the crossing pathway and the inner pathway43. We therefore also calculated the energy diagram of these two paths using the nudged elastic band (NEB) method, which is able to evaluate the ferroelectric transition pathway44. In the crossing pathway, the oxygen sites cross between the cation atoms in the (100) and (200) cation lattice planes. Whereas in the inner pathway, the oxygen sites pass through the centre of the cation lattice. From the energy diagram and barrier of the two 180-degree switching mechanisms shown in Fig. 5a, though the minimum energy levels in the middle (50%) of switching paths are comparable for both types, the energy barrier of the crossing pathway is much higher than that of the inner pathway in both ZrO2 and HfO2. Our calculations are based on the bulk model, where the calculated structure is periodic and atomic shifting occurs everywhere simultaneously in the bulk. On the other hand, calculations in the literature using the domain wall model, where the large atomic cell is prepared and polarisation switching occurs locally within a single atomic layer, show that the crossing pathway has a lower energy barrier than that for the bulk model35,43. It can be said that the energy barrier can be lowered by taking an aperiodic structure locally around the domain wall to initiate the polarisation switching.

In our bulk-model based calculations, we also made a comparison between the two crossing-type pathways with and without a symmetry constraint, as shown in Suppl Figure S6, where the one without symmetry has been employed for Table 1. In the presence of a symmetry constraint, the two oxygen sites, which are visible along the projection direction along [001] in the polarised layer, pass through the cation lattice at the same time, whereas without the symmetry constraint, the oxygen sites pass through one-by-one as seen in the NEB calculation. This indicates that the transient structures during the switching can relax by taking lower-symmetry structures, particularly regarding switching oxygen sites. According to the above discussion, the experimentally observed 180-degree switching shown in Fig. 3b can be interpreted as follows; the oxygen sites passed through the cation lattice one-by-one to lower the energy barrier and this oxygen configuration may be non-uniform along the [001] (film thickness) direction as the oxygen sites are distributed in a disordered manner in the nonpolar Pbcm structure36, making the depth-projected oxygen sites look similar to the O’ phase with half-occupied oxygen sites as shown in Fig. 3c. It is noted that there is likely to be native oxygen vacancies in the specimen observed in our study. Recent quantitative analysis via multislice electron ptychography reported that the oxygen sites near the 180-degree domain wall in ferroelectric ZrO2 may have a nonstoichiometric (i.e., excess or deficient) occupancy to compensate for bound polarisation charges at the domain wall45, suggesting that the switching and domain wall propagation can promote a migration of oxygen ions.

90-degree polarisation switching

For the 90-degree polarisation switching in ZrO2, the energy barrier and whole energy diagram show higher values than those of the O → T transition, indicating that the direct 90-degree switching is not energetically preferred and this switching occurs via the O → T transition as observed in the OBF STEM experiment (Fig. 3e), which is also consistent with the STEM observation and DFT calculation in the literature46. In the presence of Hf, 180-degree switching has a higher energy barrier than that of ZrO2 for both inner- and cross-types as shown in Fig. 5 and Table 1. Moreover, 90-degree switching also has a higher barrier as well as the O → T phase transition, which makes the 90-degree switching less favourable even via the tetragonal phase in HfO2. These DFT results are consistent with the experimental results which show that the HZO membrane has less frequent 180-degree switching than that in ZrO2, and the 90-degree switching was not observed at all.

Effect of different cations in fluorite ferroelectrics

Our results indicate that different cation species and/or dopants may substantially affect the switching behaviour and stability of the ferroelectric phases in fluorite oxide materials. In the case of (H)ZO, DFT calculations suggest that the ionic radius of Hf and Zr cations, and the bonding between them and oxygen, may be an important factor: Hf (76 pm) has a smaller ionic radius than Zr (78 pm)47, resulting in smaller lattice constants of HfO2 compared with ZrO248, which is also confirmed in our DFT calculations (Supplementary Table S1). In addition, it is evaluated that the Hf-O bonding is stronger than that of Zr-O by the difference of their cohesive energy49, whereby the oxygen sites are less constrained by a softer bonding within a larger cation lattice in ZrO2 compared with HfO2, making it easier for oxygen ions to shift when switching in ZrO238. In the same way, other species of dopant such as Si, Y, and La at cation sites may also alter the stability of each phase and corresponding switching behaviour as the switching barrier is engineered by the strength of chemical bonds in molecular-type ferroelectrics50.

In the case of Y doping, for example, we have observed that a tetragonal phase is very stable and the switching events are not triggered despite applying significantly higher beam currents for a longer time than for the ZrO2 and HZO, as shown in Figure S14. This is a significant difference when compared with the reversible phase transition between tetragonal and orthorhombic phases observed in ZrO2 and the irreversible and immediate phase transition from tetragonal to orthorhombic phases observed in HZO. When comparing the behaviour of HYO, ZO and HZO, we must consider the cation valency. Whereas Hf4+ and Zr4+ ions are both tetravalent, Y3+ is a trivalent cation and is accompanied with an oxygen vacancy when doped into HfO2 or ZrO2. In the field of structural materials that need better hardness properties, it is well known that Y-doping and resulting oxygen vacancies stabilise the high-temperature phases such as tetragonal and cubic structures in the fluorite ceramics51. On the other hand, it has also been reported that the charged oxygen defects stabilise not only these nonpolar phases52 but also the ferroelectric orthorhombic phase53,54. Which crystalline phase can be stabilised sensitively depends on the concentration of the dopant and oxygen vacancies40,55,56, where the available dopant concentration window is much narrower than that in the HZO41. It is also noted that acceptor-donor co-doping via trivalent and pentavalent cations (such as La3+ and Ta5+) is also reported to be able to balance robustness of the ferroelectricity and the switching speed57. Our experimental result regarding the HYO specimen, which are quite different from ZrO2 and HZO, suggests that the valency and oxygen vacancy also play a crucial role in modifying the accessible phase and dynamics.

Our results highlight the pivotal role of the cation species in tuning the hardness of ferroelectricity in fluorite oxides, whereby targeted doping and/or cationic composition may realise a more desirable (e.g., more energy-efficient yet durable) switching response critical for next-generation ferroelectric devices.

Discussion

Using the OBF STEM technique, we observed directly the evolution of atomic structures during electron illumination in nanometre-thick ZrO2 and Hf0.5Zr0.5O2 (HZO) freestanding membranes, revealing at the atomic level the dynamics of polar-nonpolar phase transitions and polarisation switching. Importantly, we find this to be cation dependent. In the ZrO2 membrane, it was observed that the nonpolar tetragonal phase can be reversibly transformed into the polar orthorhombic phase, where the time-series OBF STEM datasets allowed us to visualise reversible 180- and 90-degree polarisation switching. Furthermore, we clarified the detailed switching pathways by combining higher-speed scanning with OBF STEM: 180-degree polarisation switching occurs via a nonpolar orthorhombic domain wall, while 90-degree switching is facilitated by the nonpolar tetragonal phase, where the dynamic displacement of the oxygen site plays a critical role. Applying the same observation scheme to a Hf0.5Zr0.5O2 (HZO) membrane and complementing the experiments with density functional theory (DFT) calculations, it was revealed that the presence of Hf cation impacts the stability of polar/nonpolar phases and the energy barriers to the phase transition and polarisation switching—rendering the polar orthorhombic phase more energetically favourable in HZO because of the presence of Hf. We note that the tetragonal-phase-mediated 90-degree polarisation switching has been reported in a previous STEM study of ZrO246. In the present work, together with the 90-degree polarisation switching, we also reveal dynamic 180-degree polarisation switching. Moreover, we uncover the distinct intermediate atomic structures involved in the 180- and 90-degree switching transitions and show how they differ, both structurally and energetically, at the atomic scale. We demonstrate qualitative control over the frequency and prevalence of these transitions by adjusting electron dose. Furthermore, we show this can be controlled by varying spatial and temporal application of the electron beam and hence the electric field. This suggests avenues for manipulating switching behaviour by atomic-scale application of electric fields. Importantly, we also show how different cation compositions affect the strain associated with the different phases and the probability of these transitions occurring, providing broad insights into the atomic-level mechanisms controlling the behaviour of fluorite ferroelectrics and opening pathways for compositional engineering of the switching behaviour. These findings provide fundamental insights into the atomic scale pathways of polarisation dynamics and phase transition phenomena in ferroelectric fluorite-type materials and the role of the cation, thereby enhancing their potential for advanced ferroelectric device applications.

Methods

Fabrication of freestanding (H)ZO membranes

For the (H)ZO membrane fabrication, we firstly obtained an epitaxially grown (H)ZO/La0.7Sr0.3MnO3 (LSMO)/SrTiO3 (STO) (001) thin film, at which the (H)ZO and LSMO films are fabricated on an STO (001) substrate by the pulsed laser deposition (PLD) method with the Coherent COMPex Pro 205 KrF excimer laser (wavelength was 248 nm). The LSMO layer was first deposited at 650–700 °C under an oxygen partial pressure of 100 mTorr. During the deposition, the LSMO ceramic target was ablated with a laser density of 1.2 J/cm2 and at a repetition frequency of 5 Hz. (H)ZO film was subsequently deposited on the LSMO layer without breaking the vacuum of the PLD chamber. The (H)ZO layer was deposited at 800 °C and under an oxygen partial pressure of 75 mTorr, where the target was pulsed with a laser density of 1.4 J/cm2 at 2 Hz. After the deposition, the specimen was cooled down to room temperature under an oxygen partial pressure of 75 mTorr. The thickness of (H)ZO layer was evaluated by X-ray reflectivity (XRR) measurements. The freestanding (H)ZO membrane was obtained by etching the LSMO layer in 5 wt.% hydrochloric acid (HCl) and solution with 0.05 M potassium iodide (KI). After the etching and exfoliation, the membranes were transferred onto the TEM grids, which have a holey carbon supporting film. We fabricated freestanding five- and ten-nanometre-thick ZrO2 membranes and five-nanometre-thick HZO membrane, respectively, where the detailed fabrication process can be found elsewhere13,14.

Atomic-resolution structure analysis of (H)ZO membranes

Atomic-resolution STEM datasets were acquired using a Thermo Fisher Scientific Spectra φ FEG TEM equipped with a monochromator and probe and image aberration correctors. All images were collected at an accelerating voltage of 300 kV and probe forming semi-angle of 25 mrad. During the experimental observation, the probe current was adjusted by changing the focus of the monochromator, which is performed by a lens underneath the electron gun and can modulate the flux of electrons going through the optical system with all other parameters staying the same. For the OBF STEM imaging, we acquired four quadrant segmented detector datasets using the Panther detector and then reconstructed OBF images via an algorithm that can be referred to in the literature24. The ADF images were obtained with an ADF detector simultaneously with the Panther detector datasets. The camera length was set to be 115 mm, resulting in a collection angle of 11-43 mrad for the Panther detector and 48-200 mrad for the ADF detector. A second order Butterworth low-pass filter58 was applied to the experimental OBF and ADF images. For the OBF STEM image simulation, firstly the quadrant segmented detector images were calculated by the MuSTEM package59, being subsequently reconstructed into the OBF image via the same procedure as the experiments. The MuSTEM calculation is based on the multislice algorithm, where the full dynamical scattering effect is incorporated. The crystalline structures used in the image simulation were referred to from the DFT-relaxed crystalline structure files used in our present study (tetragonal and polar orthorhombic phases) and from the literature (anti-polar orthorhombic phase)60.

DFT calculation

The DFT calculations were performed to optimise the crystal structures and investigate the energy landscape using the projector augmented-wave (PAW) method implemented in the VASP code. The exchange-correlation term was treated with GGA-PBE. For the PAW potentials, 2 s and 2p electrons for O, 4 s, 4p, 4 d, and 5 s electrons for Zr, and 5p, 5 d, and 6 s electrons for Hf were explicitly treated as valence electrons. The plane-wave cutoff energy was set to 550 eV. Integration in reciprocal space was performed with a grid spacing of 0.3 Å-1. Structure optimisation was conducted until all residual forces acting on each atom were less than 0.01 eV/Å. The energy barriers of phase transitions were calculated using the nudged elastic band (NEB) method with five images between initial and final states, where the intermediate structures may be biased according to the symmetry of the initial and final ones. For the calculation without symmetry constraint, all atoms in the intermediate structures were displaced by 0 to 0.1 Å in random directions to reduce the symmetry to P1 before performing the structure optimisation, which can mitigate the bias to the intermediate states.