Establishing Atomic Coherence in Twisted Oxide Membranes Containing Volatile Elements

ABSTRACT

Twisted oxide membranes represent a promising platform for exploring moiré physics and emergent quantum phenomena. However, the presence of amorphous interfacial “dead” layers in conventional oxide heterostructures impedes coherent coupling and suppresses moiré‐induced interactions. While high‐temperature thermal treatments can facilitate interfacial bonding, additional care is needed for materials containing volatile elements, where elevated temperatures may cause elemental loss. This study demonstrates the realization of an atomically coherent, chemically bonded interface in twisted NaNbO₃ membranes through controlled oxygen‐annealing treatment. Atomic‐resolution imaging and spectroscopy reveal ordered perovskite registry accompanied by systematic lattice contraction and modified electronic structure at the twisted interface, providing signatures of chemical reconstruction rather than physical adhesion. This reconstructed interface mediates highly asymmetric strain propagation in which the bottom membrane remains nearly relaxed while the top membrane accommodates substantial shear strain. By resolving the nature of the reconstructed interface, these findings establish a pathway for achieving coherent and strain‐tunable oxide moiré superlattices.

Twisted oxide membranes containing volatile elements often suffer from amorphous interfacial dead layers that suppress coupling. An oxygen‐annealing strategy establishes atomically coherent and chemically bonded interfaces in twisted NaNbO₃ membranes, enabling coherent oxide moiré architectures in volatile‐element‐containing systems.

1. Introduction

The emergence of twistronics has sparked tremendous research interest in condensed matter physics and materials science, leading to the discovery of a wealth of intriguing moiré phenomena [1, 2]. By introducing controlled angular misalignment between two crystalline layers, moiré superlattices generate modulations of local atomic registry that fundamentally influences material properties by modifying their electronic band structure and interlayer interactions. To date, most research on moiré engineering has focused on 2D van der Waals (vdW) materials, where weak vdW coupling between atomically thin layers enables rotational stacking and the formation of moiré superlattices [3]. This structural freedom has led to a series of remarkable discoveries, including flat‐band superconductivity [4, 5, 6], correlated insulating states [7, 8], emergent magnetism [9, 10, 11], and engineered band structures [12, 13, 14].

Extending twistronics beyond weakly bonded vdW systems to epitaxial complex oxides opens a vastly richer landscape that exploits the strong lattice‐charge‐spin‐orbit coupling intrinsic to transition metal oxides. Recent advances in fabricating freestanding oxide membranes have established oxides as new functional building blocks for constructing moiré heterostructures [15, 16, 17, 18]. Freed from substrate constraints through selective wet etching of a sacrificial layer or mechanical exfoliation [19, 20], these membranes can be stacked or twisted at arbitrary orientations, akin to 2D vdW materials, enabling 3D structural tunability and access to coupled charge, orbital, spin, and lattice degrees of freedom beyond conventional heteroepitaxial systems [16, 21, 22, 23, 24, 25]. Twisted oxide heterostructures exhibit robust ionic‐covalent interlayer bonding, giving rise to strain modulation and emergent moiré phenomena including flexoelectricity‐driven polar vortices, charge modulation, flat electronic bands, and unconventional magnetism [17, 26, 27, 28, 29]. Their mechanical flexibility further enables deliberate strain engineering, allowing controlled modulation of ferroic order and the creation of moiré superlattices with programmable functionalities.

Despite the promise, realizing controlled moiré structures in twisted oxide heterostructures presents far greater challenges than in vdW materials. Unlike vdW materials, where moiré superlattices can be realized through simple physical adhesion, complex oxides require the formation of strong covalent bonding across the twisted interface [29, 30, 31, 32]. Moreover, individual freestanding oxide membranes often develop a thin amorphous surface layer during release or transfer [33], originating from surface hydration, residual polymer contamination, or a partial crystalline‐to‐amorphous transition at the 2D limit. This amorphous layer inhibits the formation of strong interfacial bonding at the twisted interface, necessitating additional processing steps to remove or recrystallize the surface and recover atomic coherence [34, 35, 36]. While recent work has demonstrated clean interfaces between cubic SrTiO₃ membranes and hexagonal sapphire substrates through high‐temperature CO₂ laser annealing (≥ 1000°C) [29], such an approach may introduce stoichiometric deviations in oxides that contain volatile elements such as sodium (Na), potassium (K), lithium (Li), lead (Pb), and bismuth (Bi), which are key constituents of many ferroelectric systems. The high volatility of these elements leads to non‐stoichiometry, vacancy formation, and increased leakage current at elevated temperatures. Therefore, developing a robust alternative processing strategy is essential for achieving atomically clean interfaces in ferroelectrics containing volatile elements while preserving their structural integrity and functional properties.

In this work, we demonstrate the successful elimination of interfacial contaminants and amorphous layers and the establishment of atomically coherent interfaces in twisted NaNbO₃ membranes (t‐NNO) through systematically controlled post‐annealing under an oxygen atmosphere. Advanced microscopic and spectroscopic analyses suggest that the t‐NNO exhibits ordered ─Nb─Na─Nb─ atomic sequences and lattice modifications at the interface, confirming the formation of chemical bonding and electronic coupling, rather than weak vdW contact. This complete interfacial bonding generates asymmetric strain fields that propagate across the top membrane, whereas the bottom membrane is nearly relaxed. This study provides a fundamental understanding of how atomic‐scale interfacial bonding governs unique strain propagation in twisted oxide membranes, offering insights toward the design of adaptive, lead‐free electronic, energy‐harvesting, and quantum functional devices based on oxide heterostructures.

2. Results and Discussion

2.1. Interfacial Challenges in Twisted Oxide Membranes

Epitaxial heterostructures consisting of NaNbO₃ thin films and La_0.7Sr_0.3MnO₃ buffer layers were synthesized on (001)‐oriented single‐crystalline SrTiO₃ substrates via pulsed laser deposition (PLD) [37]. Cross‐sectional annular dark‐field (ADF) images of the epitaxial heterostructures confirm that the NaNbO₃ thin film has a thickness of approximately 5.7 nm (Figure S1). Following growth, the buffer layer was selectively etched using a diluted solution of hydrochloric acid and potassium iodide to release freestanding NaNbO₃ membranes [38]. The released membrane was then transferred onto a target substrate (e.g., Si grids) using a poly(methyl methacrylate) (PMMA) support layer, which was subsequently removed through dissolution in acetone. To evaluate the crystalline quality and phase structure, X‐ray diffraction analysis was performed, revealing high crystallinity and a single‐phase perovskite structure of the released membranes (Figure S2). A second NaNbO₃ membrane was prepared via identical procedures and stacked onto the first membrane with a relative angular offset to form twisted NaNbO₃ membranes (t‐NNO), as schematically illustrated in Figure 1a.

FIGURE 1.

Fabrication and structural analysis of twisted freestanding membranes. (a) Schematic illustration of the assembly process for twisted NaNbO₃ membranes (t‐NNO), showing selective etching of the buffer layer to release freestanding membranes and stacking with angular offset. (b) Plan‐view annular dark‐field (ADF) imaging showing the moiré superlattice of the as‐stacked t‐NNO with a twist angle of ∼47°. (c) Cross‐sectional ADF images revealing a structurally disordered interfacial layer (∼2 nm) between the two crystalline NaNbO₃ membranes, indicating interfacial roughness and discontinuity.

Plan‐view ADF imaging demonstrates the formation of a moiré superlattice of as‐prepared t‐NNO arising from controlled crystallographic misalignment, with a relative twist angle of approximate 47° (Figure 1b). The angle was determined by the fast Fourier transform (FFT) pattern of the plan‐view ADF image, in which two discrete sets of diffraction spots, corresponding to the reciprocal lattices of the top and bottom membranes, are resolved (Figure S3). The angular deviation between the two sets of diffraction spots reflects the crystallographic misorientation between the two membranes. Despite the successful twist assembly observed in plan‐view imaging, cross‐sectional ADF imaging reveals an interfacial region (Figure 1c) exhibiting structural characteristics that deviate substantially from those of the adjacent crystalline layers. While both the top and bottom membranes preserve their crystalline periodicity, each approximately 4 nm thick with a consistent pseudo‐cubic lattice spacing of d_(001)pc = 0.392 nm, an unintended layer with reduced crystalline order was observed at the interface. Low‐magnification ADF imaging further corroborates the existence of interfacial roughness and discontinuity extending laterally over 100 nm (Figure S4a). Atomic‐resolution analysis confirms a clear reduction in ADF intensity within the interlayer region (Figure S4b–d). Additional cross‐sectional analyses along the zone axes of both membranes confirm high crystalline ordering and structural integrity, although an interfacial layer of approximately 1–3 nm thickness is still observed at the interface (Figure S5). The attenuation of ADF intensity indicates the noncrystalline nature of the interface and highlights the inherent difficulty in forming structurally coherent and chemically bonded interfaces between twisted oxide membranes. Such structural discontinuities represent a key challenge, as amorphous interfaces may preclude the realization of emergent functionalities anticipated in twisted oxide membrane systems [36].

To fully address this issue, it is essential to identify the origin of the amorphous interfacial layer. Energy‐dispersive X‐ray spectroscopy (EDX) was therefore employed to elucidate the chemical characteristics of the unintended interlayer. Elemental mapping reveals the uniform distribution of the main constituents, Na (green), Nb (purple), and O (cyan),  across the crystalline membrane regions (Figure 2a). However, these constituent elements exhibit substantial depletion at the interface, accompanied by high carbon enrichment that originated from PMMA residue. A corresponding line profile analysis (Figure 2b) shows that the concentrations of Na, Nb, and O at the interface were reduced by approximately 50% relative to those in the crystalline regions. The spatial extent of this depletion corresponds to the amorphous layer thickness determined via cross‐sectional ADF imaging. As a result, the carbon signal is sharply localized at the interface, indicating that carbonaceous contamination constitutes the primary component of the disordered region. These findings suggest that the amorphous interlayer originates from residual PMMA contaminants that were not completely removed during acetone cleaning [29].

FIGURE 2.

Chemical analysis of the amorphous interlayer in as‐stacked twisted membranes. (a) Elemental mapping of Na K (green), Nb L (purple), O K (cyan) and C K (red). (b) Corresponding compositional profile across t‐NNO membranes. The interfacial region exhibits a depletion of Na, Nb, and O signals, accompanied by prominent carbon enrichment.

To further elucidate the origin of the amorphous surface layers, cross‐sectional ADF‐STEM imaging was performed on a single membrane before stacking (Figure S6). The results reveal that amorphous layers are already present on both surfaces of the membrane (gray arrows in Figure S6a): the bottom surface (exposed to the HCl/KI etchant during sacrificial layer removal) exhibits an amorphous layer of approximately 1.5 nm, and the top surface (protected by the PMMA support layer during release) shows a thinner amorphous layer of approximately 1 nm. These observations indicate that the dead layer is intrinsic to the as‐fabricated membrane and exists before stacking. The thicker amorphous layer on the bottom surface further suggests that the acid etching step contributes to the thickness of the dead layer rather than removing it. Therefore, the interfacial dead layer observed in as‐stacked membranes (Figure S6b) likely arises from a combination of factors, such as pre‐existing surface amorphization and chemical and structural modifications introduced during the release and etching processes.

2.2. Atomically Flat and Coherent Interface in Twisted Oxide Membranes

To mitigate such contamination and achieve clean interfaces, we applied a post‐annealing treatment in a tube furnace under flowing oxygen. The oxygen‐rich environment is expected to promote oxidation that facilitates the removal of carbonaceous species while potentially contributing to film densification and reduced oxygen vacancies. To determine the optimum annealing condition, systematic treatments were conducted on twisted membranes at varying annealing temperatures, followed by AFM and optical microscopy analyses to track changes in surface topography and membrane color contrast. AFM analysis reveals a continuous decrease in root‐mean‐square (RMS) roughness with increasing annealing temperature, from 960 pm in the as‐transferred t‐NNO (Figure S7a) to 186 pm after annealing at 660°C (Figure S7f). In addition, the emergence of two sets of step terraces on membranes annealed at 660°C for 2 h suggests effective interfacial reconstruction between the two membranes. Notably, extending the annealing duration results in minimal improvement, with no significant reduction in RMS roughness (Figure S7c,d). Further increasing the annealing temperature to 800°C results in a pronounced color change in the membranes (Figure S8), indicative of severe sodium loss due to its volatility at elevated temperatures. In addition, the AFM result reveals extensive pinhole formation accompanied by a pronounced increase in surface roughness, suggesting surface degradation (Figure S9). Based on these observations, an optimized annealing condition of 660°C for 2 h under continuous O₂ flow (14 mL/min) was selected, consistent with the NaNbO₃ growth temperature and lower than the typical annealing temperatures used in previous studies.

In comparison to previous studies on twisted oxide membranes, our approach is specifically designed to address the challenges associated with volatile‐element ferroelectrics. Prior work has demonstrated success in establishing interfacial bonding in non‐volatile systems such as SrTiO₃ and BaTiO₃ [21, 26, 29, 32]. For example, high‐temperature laser annealing (>1000°C) [29] and tube furnace annealing (∼900°C) [32] have been used to promote interfacial bonding. While Li et al. [21]. used a similar annealing temperature (∼650°C) and had great success for establishing twisted SrTiO₃ membranes. However, these approaches do not account for alkali volatility and are not directly applicable to Na‐containing systems. In our approach, we integrate controlled etching chemistry (HCl + KI + DI vs. DI water only), the choice of stamp material (PMMA vs. CAB; cellulose acetate butyrate), and a synthesis‐temperature‐matched post‐annealing under oxygen flow tailored to minimize Na loss, enabling the formation of a sharp, chemically bonded interface in twisted NaNbO₃.

Following the treatment, the structural characteristics of the annealed t‐NNO were examined. As shown in low‐magnification ADF images (Figure 3b), the dark regions previously observed at the interface have almost completely disappeared. Atomic‐resolution imaging and FFT pattern reveal that pseudo‐cubic lattice spacing along out‐of‐plane (d_(001)pc) remains at 0.392 nm (Figure 3c,d), while the interface becomes atomically flat and sharp. The intensity profile extracted from the ADF image indicates that the top and bottom membranes consist of 17 and 16 atomic layers, respectively, corresponding to an increase of approximately six layers in each membrane (Figure 3e) compared with the pristine state (Figure S4). To assess reproducibility, three additional samples were fabricated and annealed under identical conditions. All samples consistently exhibited well‐bonded interfaces without detectable amorphous interlayers, confirming the reliability of the annealing protocol (Figure S10). The etched membranes show a reduced thickness of approximately 4–5 nm, whereas the annealed membranes recover to ∼6 nm, indicating the recrystallization of the NNO amorphous layer. This apparent thickness restoration originates from interfacial recrystallization and the recovery of crystalline contrast from the NNO‐derived amorphous surface/dead layers formed during release and transfer, thereby restoring the ordered perovskite structure and crystalline contrast across the interface.

FIGURE 3.

Formation of an atomically coherent interface in annealed t‐NNO. (a) Schematic illustration of the post‐annealing treatment under an oxygen atmosphere for removal of interfacial contamination and structural recovery. (b–d) Cross‐sectional ADF images and corresponding FFT pattern of the t‐NNO after annealing at 660°C, showing the elimination of the amorphous interfacial layer and formation of an atomically clean interface. (e) ADF intensity profile showing a lattice spacing of 0.392 nm with 17 and 16 atomic layers in the top and bottom membranes, respectively. At the interface (red arrow), the amorphous contrast is fully eliminated, and the interfacial lattice spacing is ∼0.28 nm. (f) High‐resolution chemical profile of Na K (green) and Nb L (purple) across the interface of the annealed t‐NNO, showing the formation of an ordered ─Nb─Na─Nb─ sequence.

In particular, the interatomic spacing at the annealed interface was measured to be ∼0.28 nm, corresponding to an out‐of‐plane contraction of about 25% relative to the bulk lattice distance (Figure 3e). Additional atomic‐resolution imaging confirms consistent results of interfacial recrystallization phenomenon, consistently demonstrating out‐of‐plane lattice contraction (Figure S11). Considering the intrinsic coupling among structural, electronic, and chemical degrees of freedom in complex oxides [39], this lattice modification implies that the interface has experienced structural reconstruction rather than mere geometric reconfiguration. While the atomically flat interface indicates structural coherence, compositional and electronic structure analyses are necessary to verify whether genuine chemical bonding is established across the junction and elucidate the modified interfacial environment. Nanoscale chemical mapping of the annealed sample confirms largely uniform distributions of Na, Nb, and O across the membrane thickness, along with a significant reduction in carbon signal (Figure S12), indicating that the contamination‐rich interlayer present in the as‐stacked sample has been largely removed.

2.3. Chemical Bonding Between Annealed t‐NNO Membranes

To probe the interfacial chemistry at the atomic scale and verify the establishment of chemical bonding, atomic layer‐resolved EDX and EELS analyses were performed. As revealed by a high‐resolution EDX profile (Figure 3f), a sequential repetition of Na (A‐site) and Nb (B‐site) atomic columns was observed across the interface, confirming the formation of a periodic ‐(NbO₂)‐(NaO)‐(NbO₂)‐ stacking sequence. Moreover, the interfacial Na atomic position is slightly displaced toward Nb columns in the bottom membrane, a shift (Figure 3f, blue arrow) that aligns with the out‐of‐plane contraction identified in the ADF intensity profile. The continuous stacking of A‐site and B‐site layers, together with the lateral offsets, provides further evidence for the establishment of chemical bonding across the interface. Although a slight reduction in Na and Nb intensities is observed at the interface, the Na‐to‐Nb ratio remains nearly unchanged, implying that the stoichiometry is approximately preserved.

Electronic structure analysis using EELS fine‐structure measurements further corroborates the presence of interfacial bonding. As indicated by the navy and red boxes in the ADF image (Figure 4a), Nb M‐ and O K‐edge spectra were extracted from the interface and three positions below and above the interface. The Nb M‐edge exhibits a characteristic doublet in both the M ₃ and M ₂ regions (Figure 4b), consistent with the octahedral crystal field splitting observed in NbO₂ or Nb₂O₅ [40, 41]. Since the M _2,3‐edge arises from Nb 3p core levels to unoccupied 4d states, it alone cannot resolve subtle bonding variations. Thus, we focus on the O K‐edge energy‐loss near‐edge structure (ELNES), which exhibits two distinct features: peak A at ∼532 eV and peak B at ∼538 eV with an energy difference of ∼4 eV (Figure 4c). This splitting arises from crystal‐field effects reflected in the O K‐edge fine structure, corresponding to transitions into O 2p states hybridized with Nb 4d t _2g and e _g orbitals [42]. The spectra acquired from the bottom (spectra 1–3) and top (spectra 5–7) membranes exhibit typical NaNbO₃ spectral features [42], in which the intensity of peak A was generally higher than that of peak B. In contrast, this trend alters at the interface, where the intensity of peak A decreases (Figure 4c, red arrow), resulting in a spectrum showing nearly identical magnitudes for peaks A and B.

FIGURE 4.

Electronic structure analysis via electron energy loss spectroscopy (EELS). (a) ADF image of the annealed t‐NNO, with navy and red boxes marking the positions selected for EELS acquisition. (b,c) Nb M‐edge spectra, and corresponding O K‐edge ELNES, showing characteristic peak A (∼532 eV) and peak B (∼538 eV). The interfacial spectrum (red) exhibits suppression of peak A, yielding nearly equivalent intensities of peaks A and B. (d) Quantitative analysis of the intensity ratio (B/A) for O K‐edge spectra, showing an increased B/A value (∼1.0) at the interface.

To quantify this change, we calculated the intensity ratio of peak B to peak A (B/A). Quantitative analysis reveals that the B/A ratio is approximately 0.9 for both the top and bottom membranes but increases to about 1.0 at the interface (Figure 4d). The change in the O K‐edge intensity is likely associated with a modification of the local electronic environment at the strain‐modulated interface. Note that spectra 3 and 5 (Figure 4d) exhibit B/A intensity ratios comparable to those of the upper and lower layers (spectra 1, 2, 6, and 7), while differing from that of the interfacial spectrum. This finding suggests that the difference observed in the interfacial spectrum is more likely attributed to interfacial chemical bonding. Furthermore, the O K‐edge ELNES from the top membrane displays a subtle increase in peak width (FWHM) compared to the bottom membrane, which could be indicative of local variations arising from distributed shear strain (Figure S13). Based on these spectral results (Figures 3f and 4c), the observed out‐of‐plane contraction (Figure 3c,e) can be predicted to originate from the formation of interfacial chemical bonding accompanied by localized strain. Such behavior aligns with the mechanical characteristics of NNO, which exhibits relatively high ductility and compliance compared to other perovskite oxides [43]. Collectively, our imaging and spectroscopy results provide compelling evidence that the annealed t‐NNO membranes form an atomically clean and electronically active interface supporting coherent interlayer bonding.

2.4. Local Strain States in Fully Bonded t‐NNO Membranes

A key question for twisted oxide heterostructures, particularly in comparison with conventional epitaxial heterostructures, is whether strain is uniformly distributed across the interface, as local strain variations can strongly influence polarization configurations. To assess strain distribution within both membranes, atomic‐resolution ADF images were acquired along the respective zone axes of the top and bottom membranes (Figure 5a,b). The peak pairs analysis (PPA) method was employed to extract and quantify the local strain states. Shear strain (E_xy ) mapping reveals that the strain is not confined to the interface but extends throughout the top membrane (Figure 5c), indicating that interfacial modulation effectively propagates across the entire film. As a result, the top membrane exhibits a shear strain of 0.9 ± 0.7% (Figure 5e). In contrast, the bottom membrane shows a minimal strain of 0 ± 0.4% (Figure 5d,f), indicating that it is nearly relaxed.

FIGURE 5.

Strain analysis of fully bonded t‐NNO. (a,b) Atomic‐resolution ADF images of top and bottom membranes in the annealed t‐NNO. (c,d) Corresponding shear strain (E_xy ) maps. Notably, the top membrane exhibits strain propagation throughout the entire region. (e) Histogram of E_xy extracted from the top membrane, showing positive shear with an average value of 0.9 ± 0.7%. (f) Histogram of E_xy extracted from the bottom membrane, showing nearly free states.

To further verify the difference in strain behavior between the top and bottom membranes, we analyzed the in‐plane (E_xx ) and out‐of‐plane (E_yy ) strain components. The results reveal pronounced strain variations in the top membrane, whereas the bottom membrane remains nearly strain‐free (Figure S14). To examine the origin of the observed shear strain in the top membrane, a comparative E_xy mapping was performed on as‐stacked t‐NNO prior to oxygen annealing. The as‐stacked sample exhibits shear strain with high spatial variability (Figure S15a,b), and the corresponding thickness‐averaged strain profile yields a mean value approaching zero (Figure S15c), indicating that the pre‐existing strain is randomly distributed without a preferred direction or systematic offset. In contrast, the annealed t‐NNO displays a spatially coherent shear strain distribution whose thickness‐averaged profile yields a statistically non‐zero mean (Figure S15d–f). These comparative results indicate a possible association between the coherent strain state observed after annealing and interfacial reconstruction, while transfer‐related artifacts cannot fully account for the observed behavior.

Additional E_xy mapping confirms consistent strain distribution, corroborating the reproducibility of strain modulation across the top membrane (Figure S16). These asymmetric results suggest that the top membrane experiences greater strain modulation than the bottom one, likely associated with the stacking configuration. During annealing, the bottom membrane is mechanically constrained between the top membrane and the underlying Si substrate, such that the interfacial reconstruction does not significantly alter its relaxed strain state. In contrast, the top membrane retains the mechanical freedom to accommodate lattice adjustments and strain redistribution because of the interfacial recrystallization. Such mechanical compliance facilitates local lattice displacements and promotes the propagation of structural modifications throughout the entire top membrane. As a result, the t‐NNO system relieves substrate‐induced mechanical constraints, thereby minimizing lattice mismatch effects and enabling the formation of a twisted structure without relying on epitaxial registry.

To elucidate the structural characteristics of the moiré superlattice, PPA was applied to a plan‐view ADF image of the annealed sample with a twist angle of ∼30° (Figure S17). The resulting strain maps reveal spatially periodic shear strain fields with alternating positive and negative domains that are commensurate with the moiré superlattice (Figure S17a,b). The measured periodicity is in good agreement with the geometric prediction based on L = a/[2sin (θ/2)], yielding L ≈ 0.76 nm for a 30° twist angle. For comparison, a rigid model consisting of two physically overlaid membranes at the same twist angle, without interfacial reconstruction, was analyzed. The corresponding strain map (Figure S17c,d) shows negligible contrast, indicating the absence of atomic displacement. This contrasts with the experimental results, which suggest that the observed moiré structure involves interfacial reconstruction associated with the formation of a chemically bonded moiré interface, rather than simple geometric superposition. Low‐magnification ADF imaging (Figure S18) further confirms that the moiré pattern is laterally homogeneous over tens of nanometers, suggesting that structural modulation associated with a spatially nonuniform contamination‐rich interlayer is minimal at the annealed interface.

Such modulation facilitates strain–coupling interactions distinct from those in substrate‐clamped epitaxial films, with the resulting local distortions and moiré‐induced strain patterns, which potentially influence polarization switching [28, 44, 45], flexoelectricity [46, 47], and energy‐storage behavior [20]. Owing to the rich polymorphism, complex oxygen octahedral tilt patterns, and the coexisting ferroelectric and antiferroelectric orders [38] in NNO, its strong coupling to strain may give rise to emerging strain‐mediated topological polar textures, such as coupled ferroelectric and antiferroelectric configurations, that are distinct from those observed in twisted BaTiO₃ and SrTiO₃ heterostructures. Asymmetric strain fields extending throughout the top region may enable tunable ferroelectric and antiferroelectric phase transitions, potentially yielding reversible double hysteresis loops and offering prospects for high‐energy‐density, lead‐free capacitor applications with improved environmental sustainability. In addition, the strain modulation in twisted moiré heterostructures perturbs local polarization and electronic structures, providing new opportunities for tailoring electromechanical and optoelectronic functionalities in oxide membranes. While this study focuses on elucidating the fundamental interfacial bonding mechanisms in twisted oxide membranes rather than demonstrating direct device applications, the insights gained here highlight the potential of twist engineering as a powerful route to tune interfacial structure, strain, and functionality in complex oxides. Furthermore, the optimal annealing temperature for forming coherent, atomically bonded moiré interfaces is likely comparable to the film growth temperature of each compound. This provides a practical starting point for process optimization, particularly for systems containing volatile elements.

3. Conclusion

In conclusion, we resolve a long‐standing challenge in volatile‐element‐containing oxide twistronics by eliminating the interfacial amorphous “dead” layer that commonly forms in oxide membranes. Through identifying the origin of interfacial disorders in twisted freestanding NaNbO₃ and developing a practical oxidative annealing process, we achieve complete removal of the carbon‐containing amorphous phase formed during aqueous processing. This treatment produces an atomically sharp, chemically bonded interface, with interlayer coupling directly confirmed by atomic‐resolution imaging and spectroscopy. Furthermore, the observation of distinct strain states in the top and bottom membranes reveals asymmetric strain propagation due to the lattice softness of NaNbO₃, offering insights into engineering strain distributions in twisted oxides. These results overcome a critical bottleneck in oxide moiré fabrication and open new opportunities for designing chemically coherent, strain‐tunable architectures hosting emergent ferroic and electronic phenomena, while establishing a transferable guideline for extending coherent twisted oxide membrane fabrication to a broader class of materials, particularly those containing volatile elements.

4. Experimental Section/Methods

4.1. Freestanding Membranes Synthesis

Epitaxial heterostructures consisting of NaNbO₃ films with La_0.7Sr_0.3MnO₃ sacrificial layers were synthesized on (001)‐oriented single‐crystalline SrTiO₃ substrates via pulsed laser deposition. The La_0.7Sr_0.3MnO₃ sacrificial layer was synthesized in an oxygen partial pressure of 200 mTorr, at a growth temperature of 730°C, a laser fluence of 1.7 J/cm², and a repetition rate of 3 Hz, using a 3.7 mm² imaged laser spot. The synthesis of NaNbO₃ films was conducted in an oxygen partial pressure of 210 mTorr, at a growth temperature of 660°C, a laser fluence of 2.1 J/cm², and a repetition rate of 2 Hz, using a 4.58 mm² imaged laser spot. For membrane release and transfer, a 600 nm thick poly(methyl methacrylate) (PMMA) layer was first spin‐coated on top of the heterostructure and cured at 135°C. The etching conditions were optimized to ensure complete removal of the LSMO layer while minimizing damage to the NaNbO₃ membrane. A diluted etchant solution was prepared from potassium iodide (5 g), hydrochloric acid (0.5 mL of 6 M), and deionized water (50 mL), yielding an approximate HCl concentration of 0.06 M. The etching was carried out at room temperature, with the PMMA‐coated sample immersed in the HCl/KI solution for 4 days. Upon completion of etching, the La_0.7Sr_0.3MnO₃ sacrificial layer was completely dissolved, allowing the PMMA/NaNbO₃ bilayer to detach from the SrTiO₃ substrate. The released membrane was transferred onto perforated silicon nitride (SiN_x) windows with circular holes on a silicon (Si) frame (Norcada Inc.). The PMMA support was subsequently removed by immersion in acetone at 80°C followed by rinsing in isopropanol, yielding freestanding NaNbO₃ membranes suspended across the apertures. This process was repeated twice to prepare two separate freestanding membranes, which were then stacked with a relative angular offset to form twisted NaNbO₃ membranes. Deterministic control of the twist angle during stacking was achieved using a motorized transfer stage equipped with independent rotational degrees of freedom. Prior to stacking, the membranes were aligned using pre‐patterned chromium alignment markers, after which a target twist angle was introduced by controlled rotation of the transfer arm under optical microscopy. Finally, the transferred twisted NaNbO₃ membranes were annealed under a dynamic oxygen flow (14 mL/min) at 660°C for 2 h.

4.2. Structural Characterizations

High‐resolution X‐ray diffraction (HRXRD) measurements were carried out using a Rigaku SmartLab diffractometer to obtain θ–2θ line scans and rocking curves. Tapping mode atomic force microscopy was used to image the topography with an MFP‐3D Origin+ AFM (Asylum Research) using a reflective aluminum‐coated tip (BudgetSensors, Tap300Al‐G, force constant ≈ 40 N m⁻¹). Cross‐sectional TEM specimens were fabricated using a focused ion beam (FIB) system (Helios NanoLab 450, FEI, Nanolab Technologies). Atomic‐resolution imaging was conducted using annular dark‐field scanning transmission electron microscopy (ADF‐STEM) on a Cs‐corrected microscope (NeoARM200CF, JEOL, Japan) operated at 200 kV. The probe convergence and detector collection semi‐angles were set to ∼28 mrad and 68–270 mrad, respectively. Strain analysis was carried out using peak pairs analysis (PPA, HREM Research) applied to the ADF images. To suppress scanning distortion, a series of ADF frames were collected with a rapid acquisition time of 1 s and aligned with drift correction. Bragg filtering was applied by selecting the two primary reflections corresponding to the (100)_pc and (001)_pc lattice directions as reference vectors. Peak positions were subsequently identified in the averaged ADF images, from which the relative displacement fields (u_x , u_y ) were computed. The strain tensors were defined as follows:

$$E_{xx}=\frac{\partial u_x}{\partial x}{E}_{yy}=\frac{\partial u_y}{\partial y}{E}_{xy}=\frac{1}{2}\left(\frac{\partial u_x}{\partial y}+\frac{\partial u_y}{\partial x}\right)$$

4.3. Chemical Characterizations

Elemental mapping was performed by energy‐dispersive X‐ray spectroscopy (EDX) using a JED‐2300T integrated into the ADF‐STEM system. For EELS measurements, the core‐loss EELS spectra of the Nb M‐ and O K‐edge were obtained with a scan step of 0.2 Å across membranes. The energy dispersion was 0.03 eV/ch and the dwell time 0.05 s/pix with 10 passes using hybrid‐pixel electron detector (Stela, Gatan, USA). Background subtraction was performed prior to core‐loss signal extraction by fitting and removing the power‐law background intensity. Min‐max normalization between 0 and 1 was subsequently applied to the O K‐edge spectrum over an energy‐loss range of 525–555 eV and to the Nb M‐edge spectrum over 350–395 eV. To evaluate the B/A peak intensity ratio, the local maxima corresponding to peak A (∼532 eV) and peak B (∼538 eV) were identified within the O K ELNES. To minimize the artifacts from spectral noise, intensity values were then extracted by averaging the signal over a 0.05 eV energy window centered at each peak position, and the B/A ratio was calculated from these averaged values. Error bars associated with the B/A ratio indicate local statistical variations, derived from the standard deviation of three spectral measurements collected along individual atomic rows of the same sample.

Author Contributions

M.C., R.X., and Y‐H.K. conceived this work. R.G. and R.X. prepared the twisted freestanding membranes and conducted XRD and AFM characterization. Y.‐H.K., M.‐H.J., Y.Z., and Y.‐M.K. prepared TEM lamella specimens. Y.‐H.K. conducted microscopy experiments for structural and chemical characterization. Y.‐H.K. wrote the initial manuscript with contributions from R.X. and M.C. All authors participated in editing the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File: adma74365‐sup‐0001‐SuppMat.docx.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.