Introduction

The functional performance of ferroelectric materials is governed by their microstructure, with strain-polarization coupling being one of the characteristics widely utilized in non-volatile memory applications1,2. Defects, which vary from zero-dimensional point defects to two-dimensional interfaces, are inherent in these materials and significantly influence their functional characteristics. Among these defects, one-dimensional topological defects, such as dislocations, have recently gained attention not merely as defects, but as core elements whose configuration can be engineered to tailor material performance3,4,5.

Traditionally, dislocations have been viewed as harmful, with most defect engineering strategies merely focused on reducing their density to enhance material properties6,7. However, recent studies have revealed that strategically engineering dislocation configurations can actually improve functionality. This paradigm shift is particularly relevant in ferroelectrics, where the spatial rearrangement of dislocations is as critical as their density. A low density of randomly-distributed dislocations can suppress random strain fluctuations and promote uniform domain ordering. However, this configuration typically results in higher energy requirements for domain switching and thus a higher coercive field (Ec)8,9,10,11,12,13. In contrast, a high density of random dislocations can facilitate strain transfer and reduce switching barriers through interaction with domain wall motion, but may also generate random strain field that lead to nanoscale polarization vortex, resulting in decreased remnant polarization (Pr) and accelerated aging process14.

The profound impact of lattice control on ferroelectric properties is classically demonstrated by applying strain. In ferroelectrics like BiFeO3 (BFO), strain-induced templates can selectively manipulate FeO6 octahedral rotation modes. A striking manifestation occurs in epitaxially constrained BFO thin films under strong compressive strain, which stabilize a metastable tetragonal-like polymorph exhibiting both an exceptionally large axial ratio (c/a  ~ 1.23) and fundamentally distinct oxygen octahedral rotation patterns compared to the rhombohedral bulk phase. This “super-tetragonal” phase, first predicted theoretically in the mid-2000s, exhibits not only giant polarization but also reconstructed rotation symmetries, demonstrating dual role of strain in modifying both polar and rotational order parameters15. Similarly, in displacement-type ferroelectrics such as BaTiO3 (BTO), the interaction between strain and lattice distortions becomes particularly crucial for optimizing distortion pathways. First-principles calculations reveal that uniaxial c-axis strain promotes cooperative Ti-O displacements that linearly enhance polarization, while multiaxial (abc) strain triggers discontinuous polarization jumps through anomalous oxygen displacements16.

Building on the established foundation of strain engineering, recent advances have identified dislocation engineering as a frontier for controlling ferroelectric properties, shifting the focus from applying uniform global strain to deliberately designing nanoscale dislocation configurations17,18,19. This innovative approach enables the conversion of dislocation configurations from randomly distributed defects into purposefully crafted functional elements, providing pathway to simultaneously enhance key memory parameters: increasing Pr to broaden memory windows, tuning Ec for efficient switching, and significantly improving aging stability. By leveraging the unique, topologically protected behavior of dislocations, guided dislocation self-assembly provides a pathway to control local octahedral tilting and lattice distortions in ferroelectrics15,16,20,21. Such precise control over local structural distortions represents the central goal of modern defect engineering.

Bismuth ferrite (BiFeO3, BFO) is well-suited for memory applications due to its intrinsic properties, such as high intrinsic Pr exceeding 100 μC/cm2 and low Ec below 100 kV/cm22. However, the integration of BFO into thin-film devices presents significant challenges, primarily due to interfacial mismatch strain, which can induce defects during crystallization23. These defects can markedly impair the ferroelectric properties, leading to issues such as severe leakage current, accelerated aging in polarization, diminished Pr and increased Ec values24. Addressing these issues by defect engineering provides a strategy to concurrently optimize polarization parameters, paving the way for developing more reliable and efficient ferroelectric memory devices.

To address the limitations in ferroelectric performance of BFO thin films, one must reconsider the interactions among defect configurations and ferroelectric polarization. A key focus lies in understanding how the configurations influence ferroelectric polarization. A major challenge lies in the fabrication of BFO thin films with controlled defect configurations that can effectively regulate interactions across various levels, from atomic distortions to domain structures and overall ferroelectric polarization.

Here, we have developed an innovative strategy of dislocation engineering via controlled self-assembly to overcome the limitations of conventional defect density-control approaches in Bi(Fe,Mn)O3 (BFMO) thin films. Unlike previous studies that employed SrTiO3 (STO) substrates to achieve low dislocation densities and improve ferroelectric properties25,26,27, we selected a face-centered cubic (FCC) Ni-Cr substrate to deliberately introduce lattice and thermal expansion mismatch with the rhombohedral (R3c) BFMO thin film. This mismatch is not a defect but a design feature, serving as a source to generate a high density of dislocations for configuration control. These dislocations provide a unique way of probing dislocation-polarization interaction and the impact on ferroelectric behaviors. Furthermore, the metallic nature of Ni-Cr substrate offers inherent potential as a built-in electrode for future studies on interfacial charge behavior, while its mechanical properties also make it a promising candidate for explorations into flexible ferroelectric electronics. We then incorporated a LaNiO3 (LNO) buffer layer, which reduced BFMO/Ni-Cr lattice mismatch from ~3.8% to ~1.0%. This promoted the formation of a discontinuous-columnar grain structure, which in turn provided fundamental control over the dislocation configuration, guiding the subsequent self-assembly.

The strain energy anisotropy effect of LNO plays a pivotal role in guiding the formation of discontinuous-columnar grains in BFMO, facilitating the self-assembly of topologically-protected dislocations along grain boundaries. This interaction between dislocations and grain boundaries introduces an innovative mechanism for regulating functional properties, enabling the dissipation of strain energy through a more uniform strain field, more coherent FeO6 octahedral tilting, and enhanced macroscopic polarization homogeneity. Consequently, the LNO-buffered BFMO thin film aged for 60 days at 60 °C exhibits robust aging resistance, with only ~20% and ~35% reductions in Pr and Ec, while the LNO-free thin film suffers severe degradation of ~90% and ~80%, respectively. These findings challenge the conventional perspective of defects as purely detrimental, instead presenting a design approach focused on controlling dislocation configurations to optimize ferroelectric properties.

Results

Edge dislocation effects on ferroelectricity for Bi(Fe,Mn)O3 (BFMO)

In perovskite ferroelectric thin films, where multiple dislocation types typically coexist, edge dislocations are particularly influential due to their distinct mechanistic attributes. The characteristic extra half-plane structure in edge dislocations generates substantial strain gradients, quantified by pronounced shear stress and asymmetry normal stress near the dislocation core28,29. This anisotropic strain field breaks local symmetry and induces polarization gradients, establishing edge dislocations as effective tools to tailor ferroelectric polarization. The resulting strain-polarization coupling further enables deterministic control of dynamic responses to external stimuli, offering valuable opportunities for exploring defect-ferroelectric interactions.

To explore these effects at a fundamental level and provide a basis for their engineering, we established a 2D slice model of BFMO single crystal viewed along [100] direction, allowing us to effectively isolate and examine the intrinsic edge dislocation effects on ferroelectric polarization. The [100] direction is specifically chosen for reasons as follows: (i) the Burgers vector (b‖[100]) lies entirely within the observation plane while the dislocation line (t) aligns with [010], and (ii) it provides optimal strain field decoupling. Besides, the [100] direction enables clear separation between transverse strain (εxx) and longitudinal strain (εyy), respectively around edge dislocations, while also allowing distinct visualization of domain wall strains (primarily εxx). This geometric selectivity permits effective isolation of defect interactions and precise analysis of edge dislocation effects on polarization. We compare characteristic electric field-dependent interactions between an edge dislocation (marked as black “T” symbol, where the vertical stem of “T” indicates the extra half-plane and the horizontal bar shows the termination line of the dislocation) and a ferroelectric domain wall (where ferroelectric domain is the region with uniform polarization orientation indicated by arrows and domain wall represents boundary between adjacent domains), as shown in Fig. 1a,b. Phase-field simulations reveal the fundamental mechanisms underlying defect-polarization interaction, providing critical insights into their role in modulating ferroelectric properties.

Fig. 1: Intrinsic effects of an edge dislocation on ferroelectric polarization in 2D slice of BFMO single crystal viewed along [100] direction.
Fig. 1: Intrinsic effects of an edge dislocation on ferroelectric polarization in 2D slice of BFMO single crystal viewed along [100] direction.
Full size image

a, b Phase-field simulations of the sample without/with an edge dislocation-domain wall interaction under various applied electric fields, the coordinate system and direction of the applied electric field are defined in Fig. S1, where polarization vortices appear in b. c, d Atomic-scale edge dislocation core model by utilizing BFMO homojunction structure: (c1) pristine structure with coherently stacked BFMO layers, (c2) slip-dislocated structure modeling an edge dislocation, where the upper (→) and lower (←) BFMO layers exhibit anti-parallel in-plane displacements with slip distance of 2.645 Å, respectively, creating a mismatch at the interface (marked as black-dashed line) analogous to the core of “T”-marked edge dislocation in b. d Calculated electrostatic potential for pristine and slip-dislocated structures, revealing charge redistribution and built-in electric field modification due to dislocations. Statistical box-normal plots (DFT simulations) quantify potential fluctuations, showing enhanced local polarization screening near dislocation cores.

The results demonstrate that when the applied electric field (E) is below the critical pinning strength (E < Ec, where Ec is the coercive field), the strain field generated by the edge dislocation creates a significant energy barrier that strongly pins the domain wall at the dislocation core. This pinning effect primarily stems from the strong coupling between the local strain field induced by the dislocation and change in polarization across the domain wall. As the E value increases beyond the critical value (E > Ec), the domain wall undergoes a dynamic de-pinning process, characterized by local deformation near the dislocation core and eventual release through characteristic structural transformations. Notably, in the defect-free BFMO (Fig. 1a), polarization vortices emerge, only when the E value reaches a critical threshold value. In contrast, in BFMO containing an edge dislocation (Fig. 1b), distinct polarization vortices are readily detectable near the dislocation even in the absence of E. Without defects, the formation of polarization vortices relies entirely on E overcoming the energy barrier of the uniform polarization state. However, with an edge dislocation, the strong localized strain field introduced by an edge dislocation couples significantly with the ferroelectric order parameter, markedly promoting formation of vortex through strain-polarization interactions. More importantly, this localized strain field provides an additional topological protection, enhancing the stability of the formed polarization vortices. This process aligns with classical domain wall pinning process but highlights the unique strain effects from dislocations. These insights not only deepen our understanding of defect-polarization interactions but also establish the foundational principles for the dislocation engineering approaches demonstrated in this work30.

To further explore this mechanism at the atomic scale, DFT calculations were performed (Fig. 1c) to simulate atomic-scale edge dislocation configurations, which directly correlate with the edge dislocation observed in Fig. 1b. Comparing the pristine (c1) and slip-dislocated (c2) BFMO systems, where the latter was constructed by cleaving the BFMO crystal along the (001) plane, the dislocated configuration (c2) mimics the core characteristic of edge dislocation through a relative in-plane shear displacement (marked by ←/→ arrows) across a defined interface (indicated by the black-dashed line). This structure mismatch reflects the extra half-plane characteristic of an edge dislocation, indicated by the black “T” symbol in Fig. 1b, where the black-dashed interface line corresponds to the horizontal bar of the “T” notation. The anti-parallel displacements explicitly represent Burgers vector components of neighboring dislocation cores, introducing a shear-like distortion at the interface that may influence local polarization. Fig. S2 reveals the corresponding electrostatic potential distributions with the statistical-box normal plots (Fig. 1d) to show the potential fluctuation. The significantly enhanced electrostatic potential fluctuations in the slip-dislocated structure are evidenced by extended box ranges and flattened normal distribution profiles, which fundamentally reflect charge redistribution and built-in electric field modification. Additionally, the differential charge density analysis (Fig. S3) reveals distinct localized charge accumulation around O sites and depletion near Fe cations, demonstrating that these variations originate from the significant electron redistribution primarily from Fe-3d to O-2p orbitals. This charge transfer, driven by FeO6 octahedral tilting [arrows in Fig. 1(c2)], couples with ionic displacement-induced dipole re-alignment to stabilize the observed electronic configuration. The resulting symmetry-breaking shear distortion creates polarization disorder, providing direct evidence for how an edge dislocation fundamentally modulates ferroelectric behavior through local structural and electronic modifications.

The influence of edge dislocations on the domain structure (Fig. 2a,b), ferroelectric polarization (Fig. 2c) and local strain field (Fig. 2d) in BFMO single crystal were investigated using phase-field simulations based on Ginzberg-Landau equations. In the absence of edge dislocations, the saturated polarization (Ps) forms a stable equilibrium configuration with well-defined straight 90° domain walls (Fig. 2a). In contrast, phase-field simulations of an active slip system (Fig. 2b) containing randomly-distributed edge dislocations (Burgers vector |b| = 2ai, where i = 1, 2, 3) reveal that domain walls become distorted and broadened near dislocation cores, forming kinked, bulged configurations. This results in significantly enhanced fluctuations of local ferroelectric polarization. The corresponding polarization switching dynamics under super-coercive electric field was systematically investigated to elucidate the pinning effects of edge dislocations on domain switching (Fig. 2c). Compared to the pristine one, BFMO containing edge dislocations exhibits reduced remnant polarization (Pr) and increased coercive field (Ec), highlighting the strong pinning imposed by dislocations on domain wall motion. Further analysis of the local strain field (Fig. 2d) shows that dislocations introduce significant elastic distortions that influence domain behavior, causing domain walls to bend along the isoenergy contours of the strain field, where elastic strain energy is minimized. Without dislocations, only minor transverse strain (εxx) variations are observed around the domain walls, while the longitudinal strain (εyy) remains essentially unchanged. The introduction of edge dislocations with specific geometric configurations (Burgers vector b‖[100], dislocation line t‖[010]) generates significant εyy and coupled εxx at the dislocation cores. This complex strain field presents a fundamental challenge for ferroelectric modulation.

Fig. 2: Influence of randomly-distributed edge dislocations on domain structure, ferroelectric polarization and local strain distribution in BFMO single crystal viewed along [100] directions.
Fig. 2: Influence of randomly-distributed edge dislocations on domain structure, ferroelectric polarization and local strain distribution in BFMO single crystal viewed along [100] directions.
Full size image

a,b Domain structures without/with randomly-distributed dislocations as shown by phase-field simulations, respectively, where dislocations lead to domain wall broadening and bulging, as demonstrated in enlarged insets. c Phase-field simulations derived polarization-electric field (P-E) loops, where randomly-distributed dislocations lead to reduced Pr and increased Ec. d εxx (d1) and εyy (d2) for BFMO without dislocations, εxx (d3) and εyy (d4) for BFMO emerged with randomly-distributed dislocations, where dislocations induce significant elastic distortions, bending domain walls along isoenergy contours of the strain field to minimize elastic energy, thereby altering switching dynamics. e Atomic-scale schematic illustrating strain origins (εxx, εyy) at edge dislocation-ferroelectric domain wall interactions, where (e1) and (e2) correspond to (d1)–(d2) and (d3)–(d4), respectively.

As shown in Fig. 2(e1), in the absence of dislocations, the εxx observed at domain walls originates from cooperative lattice distortions that accommodate polarization switching between adjacent ferroelectric domains, primarily manifesting as transverse lattice stretching or compression. While, the negligible εyy reflects minimal polarization-induced variation in the out-of-plane lattice parameters. Upon introducing edge dislocations with b‖[100] and t‖[010] in Fig. 2(e2), the characteristic “extra half-plane” disrupts lattice periodicity, generating substantial longitudinal compression or tension at the dislocation core. This strong localized distortion, through Poisson coupling, further induces εxx, creating a three-dimensional strain field with significantly greater intensity than domain wall induced local strains.

These results demonstrate that the configuration of edge dislocations exerts a much stronger localized influence on ferroelectric polarization than domain walls alone, as its associated three-dimensional strain field directly impacts both longitudinal and transverse directions.

Structural mechanisms of BFMO thin films

Building on the principle that ferroelectric polarization is governed by dislocation configuration, we designed a heterostructure (Fig. 3) to actively engineer these arrangements. The introduction of a LaNiO3 (LNO) buffer layer between the BFMO thin film and Ni-Cr substrate fundamentally modifies the interfacial constraints (Fig. 3a,b). In the Au/BFMO/Ni-Cr structure (Fig. 3a), direct contact between BFMO and Ni-Cr results in a severe lattice mismatch (Δa/a₀ ~ 3.8%). By contrast, the Au/BFMO/LNO/Ni-Cr (Fig. 3b) reduces this mismatch to Δa/a₀ ~ 1.0% through LNO’s intermediate lattice parameter, enabling near-coherent hetero-epitaxial growth through optimized lattice parameter matching. XRD analysis confirms that both BFMO thin films possess phase-pure polycrystalline perovskite structures, consistent with the reference (#PDF 01-072-7680, Fig. S4). With the introduction of LNO, the BFMO thin film exhibits significantly enhanced intensity and sharpness of diffraction peaks, particularly (012), (104), and (110), compared to the LNO-free one. This improvement stems from the better lattice matching between LNO and BFMO, which lowers the nucleation energy barrier and promotes hetero-epitaxial growth of BFMO31,32,33.

Fig. 3: Schematic diagrams of BFMO thin film structures and stress distribution analysis.
Fig. 3: Schematic diagrams of BFMO thin film structures and stress distribution analysis.
Full size image

a Au/BFMO/Ni-Cr, b Au/BFMO/LNO/Ni-Cr. c BFMO thin film structure modeling as an Euler-Bernoulli beam on a Winkler foundation (Ni-Cr and LNO/Ni-Cr), where N is the axial force, representing the stress acting between thin film and foundation, M is the corresponding bending moment, h and H is the thickness of the thin film and foundation, respectively, and ω(x) is the thin film deflection. d–f Stress/deflection versus region/thickness according to the schematic in c. The LNO buffer layer reduces lattice mismatch, improves strain energy dissipation, and enables multi-directional strain delocalization.

To further understand the stress environment, we developed a mechanical model treating BFMO thin film (estimated thickness h ≈ 600 nm) as an Euler-Bernoulli beam on a Winkler foundation (Ni-Cr and LNO/Ni-Cr, estimated thickness H ≈ 50 μm), where the axial force (N) and induced bending moment (M) collectively determine the deflection profile ω(x) (Fig. 3c)34. For simplicity, the structure is treated as one-dimensional and symmetric, with interfacial stress originating from the BFMO-foundation boundary and acting uniformly at BFMO. This symmetric stress distribution induces upward buckling of BFMO. Leveraging the symmetry, we restrict our analysis to the right half-region (x ≥ 0), with the symmetry axis fixed at x = 0 and the region of interest spanning x є [0, L]. The deflection profile ω(x) of BFMO thin films on the Winkler foundation is governed by Eq. (1).

$${EI}{\omega }^{{\prime} {\prime} {\prime} }\left(x\right)+N{\omega }^{{\prime} {\prime} }\left(x\right)+k\omega=0$$
(1)
$$k=\frac{E}{H}$$
(2)

where E is the Young’s modulus of BFMO, I is the moment of inertia, N characterizes the interfacial stress, and k is the elastic constant of the foundation. For the thin-sheet foundation, k is simplified by Eq. (2). We derive the analytical solution under the boundary condition of an applied moment M at x = L. This model directly correlates lattice matching at the interface with stress relaxation on the macroscopic scale, establishing a multiscale framework for understanding how interfacial constraints can be tailored to engineer defects.

Building upon the mechanical model described in Fig. 3c, we perform a systematic comparison of stress distribution within BFMO thin films (Fig. 3d–f). It reveals that the BFMO grown on LNO/Ni-Cr exhibits a distinct multi-directional load transfer pathway, forming a characteristic “square star-shaped” configuration (Fig. 3e). This contrasts sharply with the uniaxial, striped stress distribution observed in BFMO directly on Ni-Cr (Fig. 3d). This fourfold symmetric stress distribution not only provides clear orientation guidance for engineering defects, but also demonstrates improved strain energy dissipation through modified geometric patterning, mirroring reported stress-relief strategies35,36,37,38. Besides, the BFMO on LNO/Ni-Cr exhibits smoother deflection gradient with reduced slope dω/dx (Fig. 3f) across the whole thin film, potentially enabling multi-directional strain delocalization. This behavior, which creates a more favorable environment for controlled dislocation self-assembly, merits further investigation through Geometric Phase Analysis (GPA).

To compare the crystallinity of BFMO thin films grown on Ni-Cr and LNO/Ni-Cr substrates, bright-field (BF) and dark-field (DF) TEM imaging were performed (Fig. 4a–c). Both thin films demonstrate excellent compositional homogeneity across their thicknesses (Fig. S5). The thicknesses of BFMO and LNO layer are approximately 650 nm and 100 nm, respectively. BF-TEM and DF-TEM of BFMO thin film on LNO/Ni-Cr reveal discontinuous-columnar grains (Fig. 4b–c, S6, S7), demonstrating epitaxial registry between BFMO and LNO, which is consistent with previous reports39,40,41. The vertical growth of discontinuous-columnar grains establishes a longitudinal grain boundary network spanning the film thickness. Local selected-area electron diffraction (SAED) of region (c1) (Fig. 4c) shows superlattice diffraction spots with two sets of Bravais lattices, attributed to FeO6 octahedral tilting, which is a hallmark of the high-symmetry R3c phase in BFO derivatives42,43. HRTEM images (Fig. 4d,e) reveal well-defined lattice fringes with interplanar spacings (e.g., (012), (104), (110), (10\(\bar{2}\))) matching well with the reference perovskite structure (PDF #01-072-7680). SAED confirms the crystallographic orientations: [2\(\bar{2}\)1] for BFMO/Ni-Cr and [010] for BFMO/LNO/Ni-Cr, both consistent with the R3c symmetry.

Fig. 4: Microstructural characterization of BFMO thin films.
Fig. 4: Microstructural characterization of BFMO thin films.
Full size image

a,b Bright-field TEM images of BFMO on Ni-Cr and LNO/Ni-Cr with yellow-solid square (a1) represents a randomly selected small grain. c Dark-field TEM image of BFMO on LNO/Ni-Cr, where white-dashed (c1) lines marking discontinuous-columnar grains, orange-solid circle (c2) demarcating grain-boundary-adjacent region and inset shows SAED pattern of area (c1) with blue and red circles highlighting two Bravais lattices. d–f HRTEM images of area (a1), (c1) and (c2), respectively. g–i Inverse Fast Fourier Transform (IFFT) (color bar: strain variation scales proportionally with color depth from white to green) corresponding to d, e and white-dashed square in f, where red-dashed line marking a grain boundary. j,k Fast Fourier Transform (FFT images (in the left top) and Geometric Phase Analysis (GPA) in different faces (the other three) corresponding to g and h. In contrast to the randomly-distributed dislocations and a non-uniform strain state in BFMO/Ni-Cr, the BFMO/LNO/Ni-Cr thin film exhibits dislocation self-assembly along grain boundaries with a uniform strain state.

The IFFT reconstructions (Fig. 4g–i) highlight characteristic edge dislocations, marked by white “T” symbols, where an extra atomic half-plane terminates within the crystal lattice, which is a common feature in perovskite ferroelectrics44. Critically, a stark contrast in their configuration is observed: the BFMO on Ni-Cr exhibits a high density of randomly-distributed dislocations, whereas in BFMO/LNO/Ni-Cr, they are preferentially assembled along a discontinuous-columnar grain boundary marked with the red-dashed line in Fig. 4f. This transition from disorder to order directly correlates with the macroscopic strain state. As evidenced by GPA (Fig. 4j,k), the randomly-distributed dislocations in BFMO/Ni-Cr induce a pronounced spatial variation in εxy and εyy, while the self-assembled dislocations in BFMO/LNO/Ni-Cr give rise to a uniform strain distribution. We attribute this to the templating effect of LNO layer during growth. The LNO promotes near-single-crystal order within individual grains. Under the influence of the anisotropic stress distribution (Fig. 3e), this structure actively guides dislocations to migrate and self-assemble along discontinuous-columnar grain boundaries (Fig. S8), using the grain boundaries as effective sinks for defect isolation.

Electrical properties of BFMO thin films

High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) was employed to map spontaneous polarization vectors via tracking B-site cation displacements (Fig. 5a,d) for BFMO thin films on Ni-Cr and LNO/Ni-Cr. The corresponding atomic brightness contrast directly visualizes spatial variations in polarization amplitude and angle (Fig. 5b,c,e,f). Regions from Fig. 4(a1) and (c1) were specifically selected to investigate the local influence of dislocation configurations on polarization (Fig. S9). In BFMO/Ni-Cr, which contains a high density of randomly-distributed dislocations (Fig. 4g), the polarization vectors exhibit significant spatial disorder (Fig. 5a). This disorder arises from irregular B-site cation displacements relative to the A-site lattice, where localized lattice distortions near edge dislocations disrupt the coherent ionic off-centering, creating competing polarization variants evident as random brightness contrast (Fig. 5b,c). In stark contrast, the BFMO/LNO/Ni-Cr thin film, featuring a microstructure with dislocations self-assembly along discontinuous-columnar grain boundaries (Fig. 4i), exhibits remarkable uniformity in both polarization vector alignment and angular distribution (Fig. 5d–f). This direct correlation between the engineered dislocation configuration and the enhanced polarization homogeneity demonstrates that defect engineering is a key to optimize ferroelectric performance. The reduction of randomly-distributed dislocations and their self-assembly along discontinuous-columnar grain boundaries minimizes the random local strain fluctuations that disrupt long-range polarization ordering.

Fig. 5: Polarization analysis of BFMO thin films corresponding to regions of a randomly selected small grain marked Fig. 4(a1) and discontinuous-columnar grain marked Fig. 4(c1).
Fig. 5: Polarization analysis of BFMO thin films corresponding to regions of a randomly selected small grain marked Fig. 4(a1) and discontinuous-columnar grain marked Fig. 4(c1).
Full size image

a–d Spontaneous polarization vectors (derived from B-site cation displacements) overlaid on HAADF-STEM images (Fig. S6) for BFMO on Ni-Cr and LNO/Ni-Cr, where BFMO on LNO/Ni-Cr exhibits uniformity in both polarization vector alignment and angular distribution, confirming enhanced homogeneity in polarization compared to the disordered configuration on Ni-Cr, b–e atomic brightness contrast maps showing polarization amplitude, c–f atomic brightness contrast maps showing polarization angle. g P-E hysteresis loops of BFMO on Ni-Cr measured at 10 kHz (@RT). h,i P-E hysteresis loops of BFMO on Ni-Cr and LNO/Ni-Cr measured at 10 kHz before and after aging (@RT, 60, 120, and 180 °C), which demonstrates superior aging resistance of the LNO-buffered sample.

This defect engineering strategy is fundamentally validated by combining it with DFT calculations (Fig. 1c–d, S3). The calculations confirm that a single edge dislocation induces a localized symmetry-breaking distortion, primarily manifested as FeO6 octahedral tilting and the resultant charge redistribution, which collectively stabilize a disordered polarization configuration. In BFMO/Ni-Cr, the collective effect of a high density of randomly-distributed dislocations creates overlapping disruptive fields, accounting for the severe macroscopic polarization disorder. Therefore, the reduction in randomly-distributed dislocations and their self-assembly achieved in the LNO-buffered BFMO thin film effectively mitigate this collective detrimental impact, directly enabling the observed polarization homogeneity.

To characterize the polarization behavior of the fresh BFMO thin films, we measured P-E loops at their respective breakdown strengths (Fig. 5g). The BFMO thin film on LNO/Ni-Cr exhibits reduced coercive field (Ec) with markedly higher Pr and Pm at room temperature. This indicates a lower domain-switching barrier, which aligns with the reduced pinning from the controlled dislocation configuration (Fig. 4i), in contrast to the strong pinning imposed by the high density of randomly-distributed dislocations in BFMO/Ni-Cr (Fig. 4g). Additionally, it also helps in reducing the leakage current density in BFMO on LNO/Ni-Cr, being one order of magnitude lower than that of the BFMO on Ni-Cr (Fig. S10).

Having established the superior initial ferroelectric properties of the LNO-buffered BFMO thin film, we evaluated the stability under accelerated aging conditions at multiple temperatures (RT, 60, 120, and 180 °C) for 60 days (Fig. 5h,i). The performance of the two thin films diverges significantly. The BFMO/Ni-Cr thin film suffered severe ferroelectric degradation and ultimately failed at elevated temperatures, while the BFMO/LNO/Ni-Cr thin film demonstrated remarkably superior aging resistance, maintaining functionality even at 180 °C. A notable contrast occurs at 60 °C. The BFMO thin film on Ni-Cr suffers a severe ~90% reduction in Pr and ~80% in Ec, whereas the LNO-buffered thin film maintains a stable state with only ~20% and ~35% reductions, respectively. We attribute this dramatic difference to their distinct dislocation configurations (Fig. 4). In BFMO/Ni-Cr, the high density of randomly-distributed dislocations creates heterogeneous and thermally unstable pinning sites that are easily de-pinned, leading to rapid polarization degradation. In contrast, the reduced randomly-distributed dislocations and their self-assembly along discontinuous-columnar grain boundaries in BFMO/LNO/Ni-Cr creates a more uniform and coherent microstructure. This engineered configuration minimizes random local strain fluctuations and establishes a stable pinning environment that is robust against thermal stress, thereby ensuring the observed aging resistance.

In situ piezoelectric force microscope (PFM) was performed to probe the domain stability and switching reversibility (Fig. S11). After partial poling with opposite voltages, the BFMO/LNO/Ni-Cr thin film exhibited robust stability, with the PFM phase and amplitude of the switched domains remaining nearly unchanged for over 40 min. In stark contrast, the PFM signals from BFMO/Ni-Cr showed significant degradation within 30 min, indicating substantial back-switching of the ferroelectric domains. This distinct relaxation behavior is directly linked to their dislocation configurations. The dislocation self-assembly along grain boundaries in BFMO/LNO/Ni-Cr creates a stable energy landscape that effectively pins the domain walls, suppressing back-switching. Conversely, the high density of randomly-distributed dislocations in BFMO/Ni-Cr generates a heterogeneous and unstable pinning environment, which facilitates the depolarization fields that drive domain back-switching.

Discussion

In this work, we have established a dislocation engineering paradigm via controlled self-assembly for designing high-performance BFMO ferroelectric thin films. By employing LNO buffer layer to tailor the interfacial constraints, we guided the formation of discontinuous-columnar grain structure and the self-assembly of dislocations along grain boundaries. This engineered microstructure characterized by a low density of randomly-distributed dislocations and ordered self-assembly characteristics, directly resulted in enhanced polarization homogeneity and superior aging stability. Our findings demonstrate that controlling the spatial configuration of dislocations, rather than merely reducing their density, is the key to unlocking superior ferroelectric performance. This defect engineering approach establishes a design principle for functional oxide thin films, paving the way for next-generation memory and electronic devices.

Methods

Mn-doped BiFeO3 (BFMO) thin film preparation

The 3 mol. % Mn-doped BiFeO3 (BFMO) thin film was synthesized via sol-gel method combined with layer-by-layer annealing. This optimal concentration was determined through preliminary experiments to effectively suppress leakage current density while significantly enhancing ferroelectric properties (Fig. S12). High-purity bismuth nitrate (99.99%, Aladdin, with 8 mol. % excess added to compensate for Bi volatilization during annealing), ion nitrate (99.99%, Aladdin), and manganese acetate (99.0%, Aladdin) were stoichiometrically weighed. The solvent system consisted of ethylene glycol (99.8%), acetylacetone (99.6%), and acetic acid (99.5%) in a 1: 3: 8 volume ratio. Bismuth nitrate was dissolved in the mixture of ethylene glycol and acetate under stirring at 70 °C, while iron nitrate was premixed with acetylacetone at room temperature. The iron solution was incorporated into the bismuth solution, followed by addition of manganese acetate. After aging for 48 h, the sol was spin-coated onto 10 × 10 mm2 Ni-Cr substrate at 4000 rpm for 30 s and dried at 250 °C for 180 s. Each layer underwent rapid thermal processing (RTP) immediately after deposition. The RTP steps for each layer were as follows: (i) preheating at 200 °C for 60 s, (ii) pyrolysis at 350 °C for 600 s, and (iii) annealing at 525 °C for 480 s. During the annealing of each layer, the underlying crystallized layers experience a repeated thermal cycle. This process provides necessary driving force for the reorganization and propagation of structural defects, such as dislocations, across the film thickness, originating from the lattice and thermal mismatch at the progressive crystallization interface between successive layers. Both the deposition and annealing cycle were repeated 10 times to achieve the desired thickness.

Fabrication of a LaNiO3 (LNO) buffer layer

The LNO buffer layer was deposited between BFMO and Ni-Cr substrate using lanthanum nitrate (99.999%, Aladdin) and nickel acetate (99.9%, Aladdin). Precursors were homogenized in the mixture of 2-methoxyethanol and acetic acid under rotation at 60 °C for 2 h, followed by 24 h aging at room temperature. Spin-coating at 3000 rpm for 30 s and drying at 300 °C for 180 s were performed with the following RTP steps: (i) preheating at 200 °C for 60 s, (ii) pyrolysis at 350 °C for 1500 s, and (iii) annealing at 650 °C for 480 s. Both the deposition and heat treatment cycle were repeated 3 times.

Structure characterization

The crystal structures were characterized by using X-ray diffractometer (XRD) with Cu Kα radiation (λ = 1.5406 Å) (6°/min, Bruker D8 Discover, Germany). The microstructure of the samples was observed by a transmission electron microscope (TEM, FEI Talos F200X, USA). High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were taken by a condenser spherical aberration-corrected transmission electron microscope (STEM, Thermo Spectra 200, USA).

Electrical measurements

The response of domains to electric field after a certain relaxation time was studied using a piezoresponse force microscopy (PFM, Bruker Dimension Edge, Germany) with the writing voltage of ± 10 V. For the electrical measurements, Au top electrodes were sputtered onto the top surface with a shadow mask (~1 mm in diameter) to obtain the Au/BFMO/Ni-Cr and Au/BFMO/LNO/Ni-Cr structures. The polarization-electric field (P-E) loops and leakage current density were derived from the ferroelectric measurement system (Radiant RTI-Multiferroic, USA). For comparing the polarization stability, the BFMO thin films were first characterized at RT as the fresh reference, then aged at RT for 60 days, and the corresponding P-E loops were tested. Furthermore, P-E loops were also measured at elevated temperatures (e.g., 60, 120 and 180 °C) to further evaluate their stability.

Phase-field simulations of the evolution of polarization behaviors

In the phase-field simulations, the ferroelectric polarization was simulated by solving the time-dependent (Ginzburg-Landau) (TDGL) equation for the temporal evolution of the polarization vector field. More details can be found in the Supplementary Information.

DFT simulations

All calculations were carried out using the Vienna Ab initio Simulation Package (VASP). We employed the projector augmented wave (PAW) method to simulate the electronic behavior of BFMO, with the generalized gradient approximation (GGA) and the Perdew-Burke-Ernzerhof (PBE) functional describing electron exchange-correlation (XC) interactions depending on the calculation type. More details can be found in the Supplementary Information.

Modeling BFMO as Euler-Bernoulli beam on a winkler foundation (Ni-Cr or LNO/ Ni-Cr)

The mechanical behavior of BFMO was modeled as an Euler-Bernoulli beam resting on a Winkler foundation (Ni-Cr and LNO/Ni-Cr substrates). The governing differential equation was numerically solved using a custom-developed PYTHON-based framework. More details can be found in the Supplementary Information.

Calculation of polarization vector

To analyze the positions and shapes of atomic columns in STEM images, we used ATOMAP, a free and open-source software tool based on 2D elliptical Gaussian fitting. The method involves iteratively fitting the most intense atomic columns, subtracting them from the image, and then extracting the positions of the remaining atoms. More details can be found in the Supplementary Information.