Introduction

Ferroelectric materials with switchable polarization are the cornerstone of emerging non-volatile memory and neuromorphic computing technologies. While perovskite oxides (e.g., Pb0.8Zr0.2TiO3, BaTiO3, BiFeO3) are promising for their large intrinsic polarization1,2,3, their incompatibility with silicon processing and depolarization field with reduced dimension hinder integration. Two-dimensional (2D) ferroelectric materials (e.g., CuInP2S64, In2Se35,6, SnS7, SnTe8, and sliding 2D ferroelectric materials such as WTe29, hBN10, GaSe11) offer considerable advantages for flexible device applications and the novel assembly of heterostructures, but suffer from inadequate polarization and poor thermal stability12. Recent advances in HfO2-based systems have revived interest in CMOS-compatible ferroelectrics13,14,15,16, yet the wake-up effect associated with their metastable phases persist as critical bottlenecks17,18,19.

Aurivillius-phase oxides present a unique opportunity to bridge these gaps. Aurivillius-phase oxides, characterized by its layered perovskite structure with alternating bismuth oxide layers and perovskite-like structural layers, are distinguished by their fatigue-resistant ferroelectric switching20,21, high Curie temperature22, and remarkable oxide ion conductivity23. However, most Aurivillius ferroelectric compounds exhibit in-plane polarization24. Bi2WO6 (BWO), the simplest member of this family that consists of alternating [Bi2O2]2+ and [WO4]2- layers aligned along the c-axis, restricts spontaneous polarization to the in-plane (ab) directions25,26. This configuration fundamentally conflicts with the vertical polarization orientation required by mainstream device architectures, such as ferroelectric field-effect transistors (FeFETs) and ferroelectric tunnel junctions (FTJs).

The functional properties of ferroelectric thin films can be engineered through diverse strategies in heterostructure design. Notable approaches include the use of vertically aligned nanocomposites to create emergent multiferroicity27,28,29, the application of interphase strain to robustly enhance ferroelectric polarization30,31, and more recently, the exploitation of the unique electrostatic environment within Aurivillius frameworks to stabilize complex ferrielectric orders32. In this work, we resolve this conflict through interfacial engineering: By templating metastable WO3 domains within a BWO matrix, we simultaneously achieve perpendicular polarization and inherited thermal robustness. First-principles calculations and atomic-resolution scanning transmission electron microscopy (STEM) confirmed the presence of a WO3 structure with ferroelectric properties. This metastable WO3 phase, similar to HfO2-based ferroelectrics, belongs to the family of binary oxide ferroelectrics. Prototype ferroelectric field-effect transistors and memristors fabricated from these films exhibit robust switching ratios (>106) and thermal stability up to 350 °C. Our work expands the landscape of simple-oxide ferroelectrics, showing a structural templating strategy that can reconfigure Aurivillius oxides with controllable polarization properties.

Results and discussion

Pure Bi2WO6 thin film

We start with pure BWO films as the control experiment. In literature reports, high quality BWO films have been epitaxially grown on SrTiO3 (STO), LaAlO3 and NdGaO3 substrates, which exhibit significant ferroelectricity with purely in-plane polarization25,26,33. BWO consists of an orthorhombic phase structure with lattice constants a = 5.437 Å, b = 5.458 Å, and c = 16.430 Å34. Figure 1a shows the crystal structure of BWO as viewed from two crystallographic directions, the black solid-line box indicates the unit cell of BWO. The STO (100) substrate possesses a cubic lattice with the lattice constant a = 3.905 Å. To achieve lattice matching, epitaxial BWO films will have their c-axis in the out-of-plane direction, and their ab plane rotated 45 ° around the c-axis, thus aligning the [100] direction of BWO with the [110] direction of STO substrate25.

Fig. 1: Growth and ferroelectric anisotropy of pure-phase BWO epitaxial films.
Fig. 1: Growth and ferroelectric anisotropy of pure-phase BWO epitaxial films.The alternative text for this image may have been generated using AI.
Full size image

a Orthorhombic BWO structure viewed along [110] and [001] axes (left), and along [100] and [001] axes (right), the black box highlights the crystal unit cell. b AFM image of BWO thin film epitaxially grown on STO, with a thickness of approximately 40 nm. c In-plane PFM phase image reveals a domain structure with four predominant in-plane polarization directions, as indicated by the red arrows. The crystallographic directions of the BWO film are marked with black arrows. d In-plane PFM phase image after local polarization switching. Schematics illustrate the relative tip–sample orientation, as well as the fast and slow scanning directions. A voltage was applied to the AFM tip while scanning within the dashed box, demonstrating reversible domain reorientation. e Out-of-plane PFM phase image. f Out-of-plane PFM response after applying a vertical electric field within the dashed region, showing no detectable switching. In panels (c–f), the white dashed boxes indicate the regions where the electric field was applied.

In our experiment, pure BWO films are grown on 0.7 wt% Nb-doped SrTiO3 (Nb:STO) (100) substrates using pulsed laser deposition (PLD). The BWO thin film depicted in Fig. 1b exhibits a thickness of approximately 40 nm. Due to the existence of epitaxial stress and the relatively small thickness, some discontinuous voids still remain on the surface. The ferroelectric property of this film was characterized using piezo-response force microscopy (PFM). Figure 1c, e presents the collected in-plane and out-of-plane PFM signals, respectively. Figure 1c clearly shows a distribution of four distinct spontaneous in-plane polarization states in the BWO thin film. In contrast, no stable out-of-plane polarization is detected (Fig. 1e), the observed signal variations are attributed solely to surface topography. In further experiments, voltage was applied to the atomic force microscope (AFM) tip during scanning within the two dashed boxes (Fig. 1d, f). Due to the trailing electric field from the tip slow-scan direction35,36,37, some of the in-plane ferroelectric polarization states were noticeably altered (Fig. 1d). Because the electrically conductive Nb:STO substrate was used, a vertical electric field also exists within the film. However, no change in the out-of-plane PFM response was observed in Fig. 1f. This indicates that the pure-phase BWO thin film lacks an out-of-plane ferroelectric polarization, corroborating with reports in previous literatures25,26,38,39.

Epitaxial textured two-component WO3/Bi2WO6 films

We design a nanocomposite Aurivillius oxide by adding WO3 component during the growth of a pure BWO film (denoted as the WO3/BWO film). In order to grow nanocomposite WO3/BWO films, we use two targets, a pure BWO target and a pure WO3 target. A 240 pulse BWO seed layer was grown initially on the substrate, followed by 10 alternating cycles of WO3 (10 pulses/cycle) and BWO (30 pulses/cycle) deposition, and subsequently terminated by 210 pulse BWO. The resulting deposition component ration of WO3: BWO is 1:7.5.

The atomic structure of as prepared nanocomposite WO3/BWO films were studied by aberration-corrected STEM, with cross-section samples cut from the as prepared film by focused ion beam (FIB). Figure 2a shows a typical STEM image of the cross-section of the film. The black region corresponds to the Nb:STO substrate, with its in-plane orientation along the [100] direction. In the film region, the darker gray contrast corresponds to the WO3 domains, which consist of W and O atoms, while the lighter gray contrast corresponds to the BWO single-crystalline framework. This contrast originates from the Z-contrast mechanism in HAADF-STEM imaging, where the presence of heavy Bi (Z = 83) atoms in BWO yields a stronger scattering signal than the WO3 phase. Both the BWO and WO3 regions exhibit a sharp interface with the Nb:STO substrate, confirming a robust epitaxial relationship. Moreover, the BWO and WO3 lattices are aligned well along the [110] in-plane orientation (Fig. 2e).

Fig. 2: Atomic-scale characterization of textured two-component WO3/BWO films.
Fig. 2: Atomic-scale characterization of textured two-component WO3/BWO films.The alternative text for this image may have been generated using AI.
Full size image

a HAADF-STEM image of the nanocomposite film. The dark region at the top represents Nb:STO (001) substrate, while the gray region in the middle corresponds to the WO3/BWO film, with thickness of approximately 20 nm. The film consists of a BWO matrix with embedded crystalline WO3 domains. The epitaxial growth direction is [001], with the in-plane direction perpendicular to the epitaxy being [110] for both WO3 and BWO. b–d FFT patterns from the regions in (a): b Nb:STO substrate, with the (100) and (001) diffraction spots highlighted by yellow and red circles, respectively; c BWO matrix, with the (111) and (002) spots marked with yellow and red circles, white arrows indicate the primary superstructure spots; d WO3 domain, showing consistent orientation with BWO. e Magnified view of the interfacial transition region between BWO and WO3, revealing an atomically sharp interface and coherent lattice continuity across the heterojunction. f Magnified HAADF-STEM image of the WO3 region, highlighting its atomic arrangement. g–i EDS elemental maps of W, Bi, and O corresponding to the region in (f). Scale bars in (e–i), 2 nm.

In the BWO region, a layered lattice structure is clearly visible, with alternating bismuth-oxygen (Bi-O) and tungsten-oxygen (W-O) layers along the c-axis. The alternation of light and dark contrast is attributed to the differing electron scattering factors of the various atomic layers. In contrast, the layered structure is not discernible in the WO3 domain. Figure 2f provides a higher magnification atomic-scale characterization of a WO3 region, corresponding to the energy-dispersive X-ray spectroscopy (EDS) analysis of tungsten (W), bismuth (Bi), and oxygen (O) shown in Fig. 2g–i. The EDS spectrum confirms the presence of W atoms and absence of Bi atoms in the nanocrystal region. On the other hand, high-resolution XPS spectra (Supplementary Fig. 1) reveals O, Bi, and W components in the nanocomposite thin film, with their binding energies closely match those of pristine BWO phase, with no detectable W peaks corresponding to reduced tungsten species (W4+/W5+), indicating that the nanocrystal is a pure WO3 phase. Raman spectroscopy analysis (Supplementary Fig. 2) reveals characteristic vibrational peaks of BWO (such as the symmetric stretching mode of WO6 octahedra at 797 cm-1)40 without notable shift even in the presence of WO3. This indicates that the primary lattice structure of BWO remains unchanged by the incorporation of WO3.

Figure 2b–d presents the fast Fourier transform (FFT) patterns derived from the Nb:STO substrate, BWO matrix, and WO3 domain in Fig. 2a, respectively. Both the BWO and WO3 FFT patterns exhibit a consistent 45 ° rotation relative to the STO substrate, confirming their shared epitaxial relationship. The FFT pattern from the BWO region exhibit excellent agreement with the known parameters of pristine BWO structure (Supplementary Fig. 3a). The WO3 FFT pattern shows closely aligned symmetry and orientation with that of BWO, consistent with their structural similarity, while differing in the absence of BWO-specific superlattice spots. This shared epitaxial alignment, distinct from the STO substrate orientation, provides definitive evidence that the WO3 phase grows via a structural templating effect mediated by the BWO matrix rather than through direct nucleation on the substrate. These results indicate that our interfacial templating strategy fundamentally alters the local chemistry, surpassing mere topotactic alignment, as manifested by the distinct lattice modulation characteristics: the presence of superlattice spots in BWO, originating from its commensurate Aurivillius superlattice, stands in sharp contrast to their absence in the structurally simplified, cation-homogeneous WO3 phase.

To further identify the atomic structure of the WO3 phase, we conducted structural characterizations in different crystallographic directions. Figure 3a, b depict the BWO and WO3 regions from the same film in Fig. 2, but the TEM sample was cut along a different crystallographic orientation, rotated 45 ° as compared with Fig. 2. Here the out-of-plane direction is aligned along [001] and the in-plane direction along [100], respectively. In Fig. 3a, the BWO structure distinctly shows a layered configuration, featuring alternating Bi-O layers and W-O layers along the c-axis. The inserted theoretical crystal structure corresponds well with the atomic structure observed in the STEM images. In contrast, the WO3 region lacks layering but displaying a square lattice geometry (Fig. 3b). Figure 3c, d shows the FFT analysis corresponding to Fig. 3a, b, where Fig. 3c exhibits precise congruence with the reference BWO crystallographic parameters (Supplementary Fig. 3c). The FFT patterns of the WO3 phase in Fig. 3d closely resemble those of BWO, apart from the absence of 3 × 2 superlattice spots caused by the periodic modulation. This observation provides direct evidence that the WO3 phase grows via a structural templating effect inherited from the BWO matrix. To validate the general applicability of the structural templating mechanism, we synthesized ultrathin films by employing the identical epitaxial growth mode with a reduced deposition dosage. The 8 nm-thick textured nanocomposite WO3/BWO thin film exhibits structural and compositional invariance across its engineered interfaces, as evidenced by cross-sectional STEM analysis (Supplementary Fig. 4). Importantly, the WO3 phase forms a semi-coherent epitaxial relationship with the BWO template, demonstrating the robustness of this templating strategy in governing phase evolution across dimensional scales, further validating the epitaxial growth and crystallographic alignment of the nanocomposite film.

Fig. 3: Structural verification of WO3/BWO nanocomposite films.
Fig. 3: Structural verification of WO3/BWO nanocomposite films.The alternative text for this image may have been generated using AI.
Full size image

a Atomic characterization of the BWO region, highlighting its layered structure along the [001] direction. The inset illustrates the known structural model of BWO, showing a strong correspondence with the experimentally observed atomic arrangement. b Atomic characterization of the WO3 region, which exhibits a non-layered structure along the [001] direction. c, d FFT analysis derived from the atomic structures in (a) and (b), with yellow and red circles identifying the (200) and (006) diffraction spots, respectively. Scale bars in (a, b), 2 nm.

Notably, this WO3 structure does not match to any existing WO3 phase in literature41,42,43, signifying the stabilization of a metastable polymorph. While previous studies have demonstrated that epitaxial strain from perovskite substrates can homogeneously distorts the bulk WO3 structure44,45, the formation mechanism here is fundamentally different. This distinction can be understood by analogy ferroelectric HfO2—which adopts a non-ferroelectric monoclinic P21/c phase at ambient conditions, but exhibits ferroelectricity in its metastable orthorhombic Pca21 phase18,46. Crucially, BWO shares this Pca21 ferroelectric space group. This structural templating mechanism is further supported by X-ray diffraction (XRD) and reciprocal space mapping (RSM) analyses. Conventional θ–2θ scans (Supplementary Fig. 5) reveal no significant peak shifts between pure BWO and WO3/BWO nanocomposite films, which is consistent with the presence of nanosized WO3 domains whose diffraction is masked by the dominant BWO matrix. In contrast, a pure WO3 film grown directly on Nb:STO (001) under identical conditions exhibits distinctly different XRD features, corresponding to a conventional monoclinic phase47,48, which confirms that the metastable polar phase observed in the nanocomposite is not stabilized by the substrate alone. RSM measurements provide further insight into the strain state of the heterostructure. Around the STO (001) reflection (Supplementary Figs. 6a,b), the WO3/BWO film shows a systematic shift toward lower qz, indicating out-of-plane expansion. Consistently, the RSM around the STO (103) reflection (Supplementary Figs. 6c,d) reveals a shift toward higher qx accompanied by lower qz, confirming in-plane compression coupled with out-of-plane elongation. These RSM results are in excellent agreement with cross-sectional STEM observations (Supplementary Fig. 7), collectively demonstrating that the BWO framework acts as an atomic-scale template, guiding the WO3 into a semi-coherent, non-bulk-like configuration stabilized by the host lattice.

The BWO structural template provides a shared epitaxial orientation and a local chemical environment, which directs the WO3 phase to nucleate and grow into nanodomains with specific crystallographic characteristics, rather than forming randomly oriented or thermodynamically stable phases. The primary role of the STO substrate is thus to enable the epitaxial growth of a high-quality BWO template, which in turn architecturally guides the formation of this polar WO3 phase.

The origin of ferroelectricity

The atomically-resolved TEM images suggest a straightforward structural model as shown in Fig. 4a. In this model, each W atom is coordinated by five oxygen atoms, forming a pyramid shape, distorted pentahedral as the structural unit. There are two different types of oxygen atoms in this structure model. First, four O atoms are located roughly in the same plane with the W atoms (the bottom plane of the pyramids), where each O atom connects two pyramid units, as marked by the dotted rectangle and circle in Fig. 4a. Second, there is another O atom sitting at the top site of the pyramid, which bonds solely to the W atom below it (marked by red arrow in Fig. 4a). In total this configuration gives a WO3 stoichiometry, agreeing with our experiments. These WO5 pyramids form a lattice precisely align with experimental TEM images from different views. First-principles density functional theory (DFT) calculations verify the structural stability of this model.

Fig. 4: Theoretical calculations on ferroelectricity in WO3.
Fig. 4: Theoretical calculations on ferroelectricity in WO3.The alternative text for this image may have been generated using AI.
Full size image

a, b The geometric structure of the ferroelectric (FE) and paraelectric (PE) phases WO3. The black arrow represents the direction of polarization (P). Two types of oxygen atoms are distinguished: O atoms coplanar with W atoms (bottom plane of pyramids, dotted rectangle and circle) and apical O atom bonded solely to the W below it (red arrow). c The variation of out-of-plane polarization is described by O atom displacement. d Two ferroelectric states of WO3 are presented. FE’ represents a structure with an oxygen atom migration (marked in cyan), and FE” denotes the pristine ferroelectric phase. Both structures are modeled using a 2 × 2 × 1 supercell (outlined by the dashed rectangle). Structural schematic: W atoms depicted as gray spheres; O atoms represented by cyan and red spheres. e The energy barrier of the migration of an oxygen atom. FE’ and FE” correspond to the two structures in (d).

Remarkably, we find that such an atomic configuration is naturally ferroelectric (FE). In Fig. 4a, all the pyramid units are pointing upward. Structural analysis reveals a Cm space group configuration that breaks out-of-plane symmetry compared to the centrosymmetric WO6 octahedral arrangement, generating a substantial out-of-plane spontaneous polarization of ~80 μC/cm2 (Fig. 4c). This polarization magnitude approaches that of prototypical perovskite ferroelectrics, including BiFeO3 (100 μC/cm2)49,50 and significantly exceeds that of BaTiO3 (26 μC/cm2)51,52. More interestingly, the structural model naturally suggests the coexistence of a paraelectric phase (PE), where half of the pyramid units are flipped (Fig. 4b). The reversed polarization in neighboring layers results in zero net polarization in this phase. If all the pyramids are flipped, it will result in the switching between the ferroelectric states (from FE’ to FE”).

First-principles Nudged Elastic Band (NEB) simulations employing a 2 × 2 × 1 supercell unveil a multi-stage polarization reversal pathway. Figure 4e illustrates the energy profile for dipole switching, where Fig. 4d shows the two states before and after flipping from FE” to FE’. The computed activation energy barrier for individual oxygen migration approximates 330 meV, indicating practical feasibility for electric-field-induced polarization switching. This energy landscape suggests that the ferroelectric transition proceeds through a concerted mechanism involving sequential atomic displacements rather than simultaneous collective motion. Additionally, the ferroelectricity is achieved here through oxygen sublattice engineering in a simpler binary oxide system, different from conventional perovskites. The combination of giant spontaneous polarization and moderate switching barriers positions this WO3 phase as a promising candidate for non-volatile memory applications.

To further corroborate the consistency between the computational and experimental results, we performed diffraction pattern simulations along the [\(\bar{1}\)10] and [010] crystallographic directions for the theoretically optimized structure. As shown in Supplementary Fig. 3b, d, the simulated FFT patterns exhibit excellent agreement with the experimental data presented in Figs. 2d and 3d, validating the structural fidelity of the computational models and their alignment with the observed lattice configurations.

Perpendicular ferroelectricity of the textured nanocomposite WO3/Bi2WO6 film

Indeed, the most remarkable finding with our textured nanocomposite WO3/BWO films is their robust ferroelectricity with perpendicular polarization. We employed PFM for a comprehensive analysis of their ferroelectric response. The conductive Nb:STO substrate allows us to ground the substrate and ensure the application of a vertical electric field to the thin films for out-of-plane polarization measurements. A pre-designed badminton pattern was utilized during the PFM scanning process (Fig. 5f), where white regions correspond to the application of a positive voltage (+8 V) by the PFM tip, and black regions represent a negative voltage (−8 V). During point-by-point scanning of the thin film surface with the AFM tip, voltages were applied according to this pattern. Figure 5a, b shows the resulting out-of-plane PFM amplitude and phase after the electric field was applied. Point-by-point scanning across the film surface resulted in distinct ferroelectric domains, as observed both in the out-of-plane PFM amplitude and phase images. The amplitude image shows minimized polarization at domain walls, which are formed between regions of opposite polarization. Meanwhile, the phase image reveals a clear 180 ° phase shift between adjacent domains, with black areas indicating upward polarization and yellow areas representing downward polarization. Analysis of the PFM amplitude and phase data (Fig. 5c) along the red dashed lines in Fig. 5a, b clearly indicates minimized polarization amplitude at the domain walls, with a phase difference of 180 ° between the adjacent domains, thereby confirming the excellent out-of-plane ferroelectric properties of the fabricated thin films. Supplementary Fig. 8 also demonstrates comparable out-of-plane polarization switching characteristics in the 8 nm thick nanocomposite film, further illustrating the low-dimensional nature of the ferroelectric thin films constructed in this work.

Fig. 5: PFM characterization of the textured nanocomposite WO3/BWO film.
Fig. 5: PFM characterization of the textured nanocomposite WO3/BWO film.The alternative text for this image may have been generated using AI.
Full size image

a, b Vertical PFM amplitude and phase images following +8 V and −8 V poling. The red dashed line across both images indicates the profile shown in (c). d, e Lateral PFM amplitude and phase images after applying +8 V and −8 V poling voltages. Scale bar: 5 μm. f Predefined badminton-shaped pattern used for voltage application. g PFM switching spectroscopy over ten cycles. h PFM hysteresis loop under varying AC driving voltages. i PFM hysteresis loop with variable temperature. These comprehensive measurements elucidate the film’s robust perpendicular ferroelectricity.

In addition to out-of-plane measurements, in-plane PFM amplitude and phase images were also recorded (Fig. 5d, e). The images demonstrate variations in in-plane ferroelectric polarization, primarily induced by the trailing electric effect of the PFM tip during the scanning process. This observation indicates that the thin films exhibit both out-of-plane and in-plane ferroelectric polarization. Notably, higher-resolution imaging (Supplementary Fig. 9) confirms that the characteristic in-plane ferroelectric domain structure of the BWO matrix remains clearly resolvable at finer scan scales, demonstrating its preservation within the nanocomposite architecture. Furthermore, to evaluate the stability and reliability of the ferroelectric switching, we conducted polarization switching curve tests at a single PFM tip point (Fig. 5g). Ten consecutive measurements revealed highly stable and repeatable ferroelectric hysteresis and butterfly loops, with a consistent phase difference of 180 ° and a switching threshold voltage of approximately 4 V. These results highlight the robust ferroelectric switching performance of the thin films.

To eliminate potential artifacts and verify the authenticity of the observed ferroelectric properties, we performed alternating current (AC) voltage-dependent polarization switching measurements (Fig. 5h). When the applied AC voltage exceeds 3 V, close to the threshold voltage for ferroelectric switching, the polarization hysteresis loop collapsed, this behavior confirms the genuine nature of the out-of-plane ferroelectric polarization in the textured nanocomposite WO3/BWO film53. Additionally, temperature-dependent polarization curve tests were conducted to assess the films’ thermal stability (Fig. 5i). Measurements were performed from room temperature up to the maximum operating temperature of 250 °C, where the polarization switching curves exhibited robust stability despite minor jitter caused by temperature-induced contact instability between the tip and the sample. These results collectively demonstrate the exceptional durability and performance of our textured nanocomposite WO3/BWO thin films under varying thermal conditions.

Performance of ferroelectric FET and memristor devices made from WO3/Bi2WO6 films

Our textured nanocomposite WO3/BWO thin films exhibit also macroscopic ferroelectricity, as measured by a ferroelectric analyzer. Figure 6a displays the measured ferroelectric hysteresis loop, from which the key ferroelectric parameters were quantitatively extracted. Specifically, at an applied electric field of 4 MV/cm, the film exhibits a remnant polarization (Pr) of approximately 10 μC/cm2, accompanied by a coercive field (Ec) of 2.5 MV/cm. Since the out-of-plane ferroelectric polarization in these films primarily originates from the WO3 domains, the measured value is lower than that of the pure ferroelectric WO3 calculated in Fig. 4c (80 μC/cm2). The macroscopic ferroelectric hysteresis loop serves as direct evidence for the presence of out-of-plane polarization in the nanocomposite films, suggesting that these ferroelectric thin films hold significant potential for applications in electronic devices.

Fig. 6: Macroscopic ferroelectric polarization and ferroelectric device characteristics of textured nanocomposite WO3/BWO thin films.
Fig. 6: Macroscopic ferroelectric polarization and ferroelectric device characteristics of textured nanocomposite WO3/BWO thin films.The alternative text for this image may have been generated using AI.
Full size image

a Ferroelectric hysteresis loop illustrating the polarization switching behavior of the nanocomposite film. b Schematic of the fabricated FeFET device structure. c Transfer characteristics of the FeFET demonstrating the on/off switching behavior. d Cross-sectional schematic of the WO3/BWO based ferroelectric memristor. e Sweep I-V curves (−5 V to +5 V) for the WO3/BWO based memristor. f Temperature-dependent I-V characteristics (−6 V to +5 V) across operational thermal regimes (100–350 °C).

To further showcase the potential device performance based on our films, we deposited a 5 nm thick HfO2 insulating layer on the as-grown films using atomic layer deposition (to minimize leakage effects during testing, the leakage currents primarily originate from oxygen vacancy defects introduced during the growth of the oxide films, as well as conducting pathways formed along interface within the textured film.) and subsequently transferred a monolayer MoS2 via a dry transfer method. Figure 6b illustrates the schematic diagram of the complete device structure, for which a FeFET characterization was conducted. Figure 6c presents the double-sweep transfer characteristics of the device, obtained by scanning the back-gate voltage (VGS) from −1.5 V to +5 V and then back to −1.5 V at a fixed drain voltage (VDS) of 1 V. The FeFET exhibits a high on/off current ratio of 3 × 106, demonstrating its excellent performance. Then, we deposited gold electrodes on the film surface to investigate the memristive effect induced by out-of-plane ferroelectric polarization. Figure 6d shows the schematic of the ferroelectric memristor device, while Fig. 6e presents the current-voltage (I-V) characteristics under voltage sweeps ranging from –5 V to +5 V. The device transits from the initial high-resistance state (HRS) to a low-resistance state (LRS) after a 0 to +5 V sweep, maintains LRS during the subsequent +5 V to –5 V scan, and returns to HRS from −5 V to 0 V. We further conducted high-temperature tests demonstrating robust memristive behavior across a wide temperature range from 100 to 350 °C (Fig. 6f), highlighting the potential of our ferroelectric thin films for high-temperature memory applications. This comprehensive analysis highlights the promising capabilities of the fabricated textured nanocomposite WO3/BWO films for advanced electronic applications, supported by both macroscopic ferroelectric measurements and device-level performance evaluations.

In summary, we demonstrate a structural engineering strategy to achieve out-of-plane ferroelectricity in a metastable WO3 phase derived from the Aurivillius oxide BWO framework. Through a combination experiments and first-principles calculations, we have confirmed the existence of this previously unreported ferroelectric WO3 phase. The resulting textured nanocomposite WO3/BWO thin films exhibit both out-of-plane and in-plane ferroelectric polarization, showcasing stable and repeatable switching behavior. Prototype FeFETs and ferroelectric memristors fabricated from the films exhibit remarkable switching ratios (>106) coupled with robust thermal stability up to 350 °C. This work not only expands the family of binary ferroelectric oxides but also provides a generalizable approach for polarization enhancement in Aurivillius oxides and other layered ferroelectrics.

Methods

Thin film growth

Bi2WO6, WO3/Bi2WO6, and WO3 thin films were epitaxially grown on (001)-oriented Nb:SrTiO3 (0.7% wt. doped) substrates by pulsed laser deposition using a 248 nm KrF excimer laser. The deposition was carried out at a substrate temperature of 750 °C under an oxygen environment of 400 mtorr. The laser energy density and repetition rate were maintained at approximately 1.4 J cm–2 and 2 Hz, respectively. After deposition, the samples were cooled down to room temperature in 7.5 torr of O2 with a cooling rate of 5 °C/min.

Piezoresponse force microscopy (PFM)

PFM measurements were performed using an Oxford Instruments Cypher (E)S atomic force microscope, equipped with conductive Pt/Ir-coated silicon probes. The local electromechanical response was characterized in both out-of-plane and in-plane directions using AC excitation amplitudes of 800 mV at frequencies of 350 kHz and 650 kHz, respectively. Litho-PFM mode was utilized to locally pole the material and write domain patterns by applying a DC bias through the scanning tip, enabling the assessment of polarization switching. Spectroscopic PFM (SS-PFM) measurements were performed in dual AC resonance tracking (DART) mode. In this mode, an AC excitation signal was superimposed on a series of DC triangular and sawtooth waveform voltages to track the bias-dependent electromechanical response.

Scanning transmission electron microscopy (STEM)

Cross-sectional specimens for STEM were prepared by focused ion beam (FIB) lift-out and subsequently thinned using a Gatan 691 PIPS for final ion milling. High-angle annular dark-field (HAADF) STEM imaging was performed on an aberration-corrected (scanning) transmission electron microscope (JEOL, Grand ARM 300) operated at an accelerating voltage of 300 kV.

Materials characterization

X-ray θ–2θ scan and reciprocal space mappings were performed using a high-resolution X-ray diffractometer (Bruker, D8 Advance). Raman spectroscopy was performed on a Horiba LabRam HR Evolution system with a 532 nm laser excitation. X-ray photoelectron spectroscopy (XPS) measurements were conducted on a Thermo Fisher ESCALAB XI+ spectrometer using a monochromatic Al Kα X-ray source. All binding energies were calibrated relative to the adventitious carbon C 1 s peak at 284.8 eV.

Functional property measurements

The P-E hysteresis loops and I-E curves were measured on a ferroelectric analyzer (tf analysis 3000E).

Device Fabrication

For two-terminal ferroelectric memristors fabrication, AZ5214 photoresist was spin-coated onto the sample surface at a rate of 4000 rpm and subsequently baked at 110 °C for 2 min. The patterning of the top gate was achieved through UV-lithography (MA6, Karl Suss), followed by the deposition of 30 nm Au metal via electron beam evaporation. The lift-off process was then employed to obtain the patterned top electrodes.

For three-terminal ferroelectric transistors fabrication, 5 nm thick hafnium oxide dielectric layer was grown on the bottom-gate at 150 °C using atomic layer deposition (ALD). Following this, the monolayer MoS2 film was carefully transferred onto the HfO2-WO3/Bi2WO6-Nb:SrTiO3 substrate. Next, the MoS2 channels were defined through UV-lithography and reactive ion etching (RIE), employing the Plasma Lab 80 Plus instrument from Oxford Instruments Company. To fabricate the device contacts, UV-lithography was utilized to define the contact regions, which were subsequently metallized with a 30 nm Au through electron beam evaporation.

Device characterization

Characterization of electrical and optoelectronic properties was performed using a probe station (Janis ST-500-1-(4CX)) equipped with Agilent semiconductor analyzers (models 4156 C and B1500A).

DFT calculation details

Density functional theory (DFT) calculations were carried out using the generalized gradient approximation (GGA) for the exchange-correlation potential, the projector augmented wave (PAW) method, and a plane-wave basis set as implemented in the Vienna ab-initio simulation package (VASP). The lattice constants of WO3 were fixed to the experimentally determined values: a = 5.34 Å, b = 5.34 Å, c = 5.90 Å, respectively. An 8 × 8 × 1 k-mesh was used to sample the first Brillouin zone. The kinetic energy cut-off for the plane-wave basis was set to 400 eV for the geometric properties and energy calculations. The atomic positions of the structures were fully relaxed until the residual force per atom was less than 0.001 eV/Å. A second-order Methfessel–Paxton smearing method with a sigma value of 0.05 eV was adopted during calculations. The ferroelectric switching in FE and PE phases, respectively, was investigated utilizing the nudged-elastic-band (NEB) method.