Symmetry breaking induces room-temperature interface ferroelectricity in two-dimensional metastable mosaic-like heterostructures

Abstract

Metastable structures exhibiting unique properties hold significant potential for electronic applications. However, their inherent tendency to transform into thermodynamically more stable phases presents substantial challenges, as the stabilization of such metastable phases typically necessitates extreme conditions and complex synthesis strategy. Here, we demonstrate the synthesis of a metastable SmO₂/Sm₂O₃ mosaic-like lateral heterostructure using molten eutectic salt. While both SmO₂ and Sm₂O₃ are individually centrosymmetric, interfacial symmetry breaking at the junction simultaneously suppresses screw symmetry and inversion symmetry, resulting in a locally non-centrosymmetric configuration with broken spatial inversion. This structural asymmetry induces an intrinsic built-in dipole, a nonlinear optical response, and room-temperature interface ferroelectric polarization. Furthermore, vertically oriented interfaces provide confined pathways for directional ion transport, suppressing non-directional diffusion. Memristors based on the metastable heterostructure exhibit high performance, characterized by a maximum on/off ratio exceeding 10⁹, switching speeds as fast as 47.3 ns, and retention times surpassing 10⁴s. The switching mechanism is consistent with field-driven oxygen vacancy migration along the vertical heterointerface, regulated by reversible oxygen adsorption and desorption at the electrode. This work offers a potential route for developing high-performance memristors leveraging metastable oxide heterointerfaces with coupled ferroelectric-like polarization and resistive switching functionalities.

The authors demonstrate the synthesis of a metastable SmO₂/Sm₂O₃ mosaic-like lateral. heterostructure using molten eutectic salt. This structural asymmetry induces an intrinsic built-in dipole, a nonlinear optical response, and room temperature interface ferroelectric polarization.

Introduction

Two-dimensional (2D) materials are recognized as a significant platform for the exploration of novel devices^1–3 and investigation of metastable phases and phase transitions^4,5. When the vertical dimension is reduced to the atomic scale, surface energy and size effect reconfigure the thermodynamic landscape, thereby stabilizing metastable phases^6,7. These phases exhibit functional characteristics⁸, including photoelectric response⁹, ferromagnetism¹⁰, and ferroelectricity^11,12, which offer new opportunities for development of the next-generation digital memory and neuromorphic computing technologies^13,14. However, achieving atomic-level control during the synthesis of novel materials that are thermodynamically unstable in bulk remains a key challenge in this field. The core concept for capturing metastable phases is to hinder the system from crossing phase-transition barriers by exploiting the kinetic mechanism and external field regulation under non-equilibrium growth conditions^15,16. Researchers employ bottom-up strategies, including sol-gel processing¹⁷, chemical vapor deposition (CVD)^18,19, and molecular beam epitaxy (MBE)^20–22, to assemble fundamental units into well-defined 2D functional materials, thereby achieving targeted properties and device performance²³. For example, nonlayered 2D β-Bi₂O₃ with high hole mobility is synthesized by vapor–liquid–solid–solid (VLSS) growth mechanism via layered BiOCl intermediate²⁴. Furthermore, vacancy and strain engineering facilitate precise phase control in large-area 2D In₂Se₃ and its heterophase junctions, paving the way for next-generation memory devices²⁵. Among diverse approaches, manipulating the proportions of chemical components to generate multiple phases^26,27 and further forming heterogeneous interfaces remain an important avenue for exploring novel phenomena^28,29.

Interface^30,31 has a critical impact on a wide range of physicochemical properties^32,33, such as ferroelectricity^34,35, nonlinear optics³⁶ and resistive switching behaviors³⁷. The interface modulates ferroelectric and nonlinear optical³⁸ properties primarily by synergistic effects of breaking inversion symmetry³⁹ and strengthening local interactions^40,41. On the one hand, lattice strain⁴² and interfacial charge coupling⁴³ not only induce and stabilize ferroelectric polarization but also enable precise control over ferroelectric domain orders^11,44. On the other hand, the symmetry breaking accompanied by emergent ferroelectricity activates nonlinear optical responses⁴⁵, particularly second-harmonic generation (SHG)⁴⁶, which can be further amplified by local plasmon fields^47,48. In addition, vertically oriented interfaces provide confined pathways for directional ion transport, effectively restricting non-directional diffusion^49,50. The performance of filament-type memristors is fundamentally limited by intrinsic stochasticity associated with ion migration⁵¹ and the formation of conductive filaments (CFs)⁵². Therefore, elucidating the underlying switching mechanisms and achieving precise control over CF dynamics represent critical research directions. In this context, the engineering of oxide heterostructure interfaces through the design of well-defined vertical interfacial channels—emerges as an essential strategy^53,54. Such interfacial architectures effectively convert inherently stochastic filamentary switching into a highly controllable process governed by the vertical interface⁵⁰, thereby enabling the realization of high-performance, reliable memristive devices.

Here, in molten eutectic salt, dissolved SmCl₃ reacts with trace oxygen species on the mica substrate. The melt’s high diffusivity and fluidity enable rapid access to local free energy minima, kinetically favoring the coexistence of metastable SmO₂ and stable Sm₂O₃ under specific conditions. The electrostatic potential difference across the interface drives electron transfer, giving rise to an in-plane electric dipole moment, as suggested by the observation of SHG signals in mosaic-like lateral heterostructure nanosheets. Meanwhile, interfacial oxygen ions in SmO₂ exhibit pronounced instability, thereby enabling ferroelectric polarization reversal via electrochemically driven oxygen-ion migration. The interface of two strong correlated oxides could serve as a tunable conducting channel mediated by oxygen vacancies (V_O), governed synergistically by polarization effects and defect chemistry, thereby offering a versatile platform for electronic devices based on rare-earth oxides, such as memristors. Exploiting this interfacial oxygen vacancy mechanism, we fabricated the vertical memristor that enables reversible oxygen adsorption/desorption by Titanium (Ti) electrode, achieving an maximum on/off ratio exceeding 10⁹ and switching times as fast as 47.3 ns. The resistive switching are consistent with interfacial oxygen vacancy dynamics, offering a compelling route toward efficient neuromorphic hardware⁵⁵.

Results

Non-centrosymmetric heterostructure model

Among rare-earth oxides, Sm₂O₃ and SmO₂ are representative of two distinct structure types of samarium oxides, exhibiting different oxidation states and crystal symmetries. Owing to effective shielding by the outer 5s²5p⁶ electrons, the 4 f electrons of Sm ion are highly localized, playing a crucial role primarily in optical transitions and magnetic interactions (Supplementary Fig. 1). The Cubic Sm₂O₃ is the thermodynamically stable phase, crystallizing in the bixbyite structure (space group Ia-3, No. 206) with lattice parameter a ≈ 10.94 Å. This structure is an ordered vacancy superstructure derived from the fluorite-type lattice, featuring a regular arrangement of vacancies on one-quarter of the oxygen sites⁵⁶ (Fig. 1a). In contrast, monoclinic SmO₂ is not thermodynamically stable under ambient conditions. It crystallizes in the P2₁/m space group and can be regarded as a symmetry-lowered variant of the fluorite structure, stabilized by Jahn–Teller distortions and oxygen-vacancy ordering (Fig. 1b). In this phase, Sm⁴⁺ is highly unstable among rare-earth elements due to the strongly positive Sm⁴⁺/ Sm³⁺ redox potential, rendering Sm⁴⁺ prone to reduction. Consequently, real samples often exhibit mixed Sm³⁺/ Sm⁴⁺ valence states or a high concentration of oxygen defects. Figure 1c presents the atomic structure model of SmO₂/Sm₂O₃ heterostructure. Although both bulk constituents are centrosymmetric, inversion symmetry is inherently broken at the interface, establishing a structural foundation for intrinsic interfacial polarization. Furthermore, the difference in the valence states of samarium drives interface charge transfer, leading to a redistribution of electrons in the vicinity of the interface under the built-in polarization field. Interfacial dipole layers effectively screen the positive charge associated with V_O, promoting local charge neutrality, while localized electronic states at the interface stabilize the released electrons. Concurrently, lattice distortion and under-coordinated atoms at interfaces significantly reduce both the defect formation energy and migration barriers. These factors collectively suggest that the conductive filament is preferentially localized at the interface. Atomic resolution high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images, also referred to as Z-contrast images, clearly resolve the two distinct phases (Fig. 1d, e).

Fig. 1. Non-centrosymmetric heterstructure model.

a, b Crystal structures of Sm₂O₃ and SmO₂, respectively. The blue rhombus represents the inversion symmetry and threefold rotational symmetry centers. c Schematics of SmO₂/Sm₂O₃ heterostructure represents inversion symmetry breaking. d, e Atomic resolution STEM images of Sm₂O₃ and SmO₂ phases along the[001] direction, respectively. Insets show schematic illustration of the unit cell. Scale bar, 2 nm.

Realization of a metastable ferroelectric heterostructure

Figure 2a presents a schematic illustration of the samarium oxide nanosheet growth process, inspired by the molten salt synthesis method^57,58. The synthesis of samarium oxide nanosheets was carried out in a home-built chemical vapor deposition (CVD) system, with mica as growth substrate and SmCl₃·6H₂O as samarium source. It is worth noting that the introduction of Fe₂O₃ could drive oxygen atom transfer reactions and effectively enhance the oxidation process⁵⁹ (Supplementary Fig. 2). Due to the hygroscopicity of SmCl₃ powder, atmospheric moisture significantly influences the sample during mixing and grinding. Both the nanosheets’ surface roughness and crystalline morphological regularity are sensitive to relative humidity. The atomically flat surface of mica has a small migration barrier of adatoms and accelerates ion diffusion rates and consequently promotes nucleation and atomic-level growth⁶⁰. Figure 2b illustrates that the atomic scheme of the nucleation and growth process in eutectic salt-assisted growth, including eutectic melting and nucleation (i), growth (ii) and substrate cleaning (iii). From the perspective of fluid-phase synthesis and energy landscape dynamics, SmCl₃ dissolves and undergoes chemical reactions with trace oxygen-containing species present in the melt to form various samarium–oxygen species. Under non-equilibrium conditions, the high fluidity and ionic diffusivity of the melt give rise to localized concentration supersaturation of samarium-oxygen clusters, facilitating rapid access to local minima on the free energy landscape⁶¹. This kinetic environment preferentially promotes nucleation and growth of the metastable SmO₂ phase, due to a lower nucleation barrier and more favorable early-stage kinetics, rather than the thermodynamically stable oxide-an interpretation consistent with classical nucleation theory. Optical microscopy (OM) images of the as-grown Sm₂O₃ nanosheets on mica reveal that the majority of nanosheets exhibit a rectangular morphology (Fig. 2c). Atomic force microscopy (AFM) imaging of a representative Sm₂O₃ nanosheet confirms a flat surface, with a corresponding height profile measuring 8.9 nm (Fig. 2d).

Fig. 2. Realization of a metastable ferroelectric heterostructure.

a Schematic illustration of vdW epitaxial growth of nanosheets. b Schematics of the nucleation and growth process of 2D nanosheets during synthesis. c An OM image of an as-grown nanosheet. Scale bar, 10 μm. d An AFM image with the thickness profile (corresponding to the black dotted line crossing the edge) of a typical nanosheet. Scale bar, 2 μm. e XRD of SmO₂/Sm₂O₃ heterostructure on Si/SiO₂ substrates. f SHG signal of the nanosheets with a 1064 nm laser. The insert shows the SHG intensity vs laser power in logarithmic scale. g Temperature dependence of SHG intensity. h The polarization-resolved SHG signal of the nanosheets.

X-ray diffraction (XRD) was employed to analyze the crystalline structure of SmO₂ and Sm₂O₃ phases (Fig. 2e). The typical diffraction peaks at 2θ = ~ 8.96°, 17.8°, 26.8°, 36.1°, 45.4° and 55.2° are assigned to the (001) to (006) planes Sm₂O₃ cubic phase, respectively, while the typical diffraction peaks at 2θ = ~ 22.9° and 65.5° correspond to the SmO₂ monoclinic phase (010) and (300). X-ray photoelectron spectroscopy (XPS) confirms the mixed valence states at the heterointerface, as detailed in Supplementary Fig. 3. To further investigate the crystal symmetry, a significant SHG signal is detected upon irradiation with a 1064 nm laser (Fig. 2f), providing evidence of a non-centrosymmetric structure⁶². Furthermore, we measured the SHG intensity of the SmO₂/Sm₂O₃ nanosheets as a function of incident laser power and plotted on a log-log scale (Fig. 2f, inset). The linear fitting yields a slope of 1.85, which closely agrees with the theoretical value of 2. This indicates that the SHG signal scales quadratically with the excitation laser power⁶³, consistent with the expected behavior for a second-order nonlinear optical process. The temperature-dependent SHG intensity exhibits a sharp drop at approximately 672 K, reflecting the disappearance of non-centrosymmetric structural order (Fig. 2g). The SHG signal showed negligible variation upon cooling from 400 K to 300 K. In contrast, a substantial reduction in the SHG signal was detected upon cooling from 1000 K to 300 K. The signal failed to return to its initial state, which is consistent with an apparently irreversible phase transformation (Supplementary Fig. 4). Control measurements on different substrates showed similar SHG response (Supplementary Fig. 5), confirming that the observed signal originates from the heterostructure. In addition, the angular dependence of SHG intensity exhibits a two-fold anisotropic petal pattern, confirming the expected symmetry of the heterostructure (Fig. 2h). Collectively, these results demonstrate robust second-order nonlinear optical activity arising from symmetry breaking at the SmO₂/Sm₂O₃ heterointerface.

The out-of-plane high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image of the SmO₂/Sm₂O₃ heterostructure reveals a ordered atomic arrangement near the interface and shows a homogeneous distribution of elements (Fig. 3a and Supplementary Fig. 6). Given that the Sm₂O₃ phase predominates in the nanosheet, with a minor SmO₂ phase embedded within the Sm₂O₃ matrix, the selected-area electron diffraction SAED pattern (Supplementary Fig. 7) exhibits diffraction spots of the Sm₂O₃ phase, suggesting that the SmO₂ signal is masked by the stronger Sm₂O₃ contribution. Nevertheless, Z-contrast imaging clearly resolves the interface between the (100) plane of SmO₂ and the (110) plane of Sm₂O₃, with an interplanar spacing of 0.2824 nm (Supplementary Fig. 8). As shown in Fig. 3b, the cross-sectional HAADF-STEM image shows the atomic structure of the Sm₂O₃ region within SmO₂/Sm₂O₃ heterostructue. The corresponding fast Fourier transform (FFT) pattern, acquired from the (001) plane of SmO₂ and Sm₂O₃, aligns well with the simulated SAED pattern, confirming the coexistence of both phases (Fig. 3c). Notably, it can be clearly found that the atomic structure of the overlapping surface of SmO₂/Sm₂O₃ heterostructure differs from the pure Sm₂O₃ (Fig. 3d and Supplementary Fig. 9). Based on these observations, we constructed a corresponding horizontal heterostructure model (Fig. 3e). Due to the close interplanar spacing of SmO₂ (100) and Sm₂O₃ (110), the lattice can match well and form heterostructure, which explains the observation of indexed diffraction spots from both phase in the same FFT image. Figure 3f shows the cross-sectional STEM image with a thickness of ~42.7 nm, along with energy-dispersive X-ray spectroscopy (EDS) elemental mappings, which reveal a largely uniform distribution of Sm and O throughout the entire nanosheet.

Fig. 3. STEM analysis of a metastable ferroelectric heterostructure.

a HAADF-STEM image of the interface of SmO₂/Sm₂O₃ heterostructure along the [001] direction. Scale bar, 5 nm. b Cross-sectional STEM image of the Sm₂O₃ region of heterostructure. Scale bar, 2 nm. c Experimental and simulated FFT images of SmO₂ and Sm₂O₃. The diffraction spots (marked by circles) exhibit a one-to-one correspondence. Scale bar: 5 nm⁻¹. d Cross-sectional STEM image of the heterostructure region. Scale bar, 2 nm. e Atomic model and FFT image of SmO₂/Sm₂O₃ heterostructure. f TEM-EDS elemental mapping of SmO₂/Sm₂O₃ heterostructure with a thickness of ~ 42.7 nm. Scale bar, 50 nm.

Ferroelectric properties of a metastable heterostructure

To evaluate the ferroelectric properties, polarization-electric field (P-E) hysteresis loop measurement was performed. The loop exhibits a hysteretic switching response, with the black arrow indicating the polarization switching current. From these measurements, a double remnant polarization (2P_r) of ~2.97 μC·cm^-2 and a coercive field (E_c) of ~ 0.22 MV·cm^-1 were extracted (Supplementary Fig. 10). Figure 4a illustrates the piezoresponse force microscopy (PFM) measurement configuration, where an AC voltage is applied to the conductive tip to probe the local electromechanical response and polarization orientation within the SmO₂/Sm₂O₃ nanosheets. To probe ferroelectric switching, box-in-box patterns were written with opposite biases (±7 V) at room temperature on an approximately 16 nm-thick samarium oxide nanosheet (Fig. 4b,c). The resulting out-of-plane PFM images clearly exhibit phase and amplitude contrast with good retention, suggesting switchable ferroelectric domains. The corresponding PFM phase and amplitude hysteresis loops further support this behavior, showing an approximately 180° phase flip and a well-defined butterfly shape (Fig. 4d). In addition, in-plane PFM phase and amplitude images acquired from another 13 nm-thick nanosheet also exhibit characteristic ferroelectric hysteresis (Fig. 4e). Reproducibility of these PFM responses was verified across multiple flakes and devices and spatially resolved pfm images demonstration that the PFM response appears to be associated with the interface (Supplementary Fig. 11, 12). Collectively, these macroscopic and local electrical measurements are consistent with the presence of switchable ferroelectric response in the SmO₂/Sm₂O₃ heterostructure.

Fig. 4. Ferroelectric properties of a metastable heterostructure.

a Schematic image of the nanosheet transferred onto Au substrate. b, c The PFM phase and amplitude images of the box-in-box pattern domain. Scale bars, 2 μm. d,e The out-of-plane and in-plane PFM phase and amplitude hysteresis loops of 2D nanosheet. f Averaged electrostatic potential of the P→ polarized in SmO₂/Sm₂O₃. g Energy evolution during the heterostructure optimization process. h HAADF-STEM image of the phase transformation of SmO₂.

Our analysis of the rightward-polarized state (P → ) is presented in Fig. 4f, which shows the in-plane averaged electrostatic potential across the SmO₂/Sm₂O₃ heterostructure. Within each region, the negligible variation in the electrostatic potential minimum indicates that neither SmO₂ nor Sm₂O₃ possesses substantial intrinsic spontaneous polarization. However, a pronounced interfacial potential drop is evident: the electrostatic potential at the left monoclinic SmO₂ surface is approximately 1.23 eV lower than that at the right Sm₂O₃ surface. This potential gradient gives rise to a built-in electric field across the heterostructure, driven by interfacial polarization and the associated charge redistribution. The corresponding calculated electric dipole moment is 1.14 |e | ·Å. To investigate the thermodynamic stability of samarium oxides at heterointerfaces, a theoretical model was constructed by stacking four layers of monoclinic SmO₂ onto four layers of cubic Sm₂O₃, with the initial interface aligned using the lattice constant of the cubic phase. As shown in Fig.4g, the iterative structural optimization reveals that SmO₂ spontaneously transforms into Sm₂O₃, accompanied by a continuous decrease in the system’s total energy and the concurrent release of a significant amount of oxygen molecules from lattice interstitial sites. These results unequivocally demonstrate the intrinsic thermodynamic instability of SmO₂ under these conditions. HAADF-STEM imaging demonstrates that continuous irradiation by a high-energy electron beam progressively reduces Sm⁴⁺ ions to Sm³⁺, concomitant with the release of oxygen. Simultaneously, the thermodynamically stable Sm₂O₃ phase gradually engulfs the adjacent metastable SmO₂ layer, resulting in a pronounced migration of the phase boundary. These in situ observations provide direct experimental evidence of the thermodynamic instability of SmO₂ (Fig. 4h and Supplementary Fig. 13). Evidence from the contrast variations and lattice fringe mismatches in the TEM images confirms that SmO₂ nanodomains do not form a continuous top layer, but are instead embedded throughout the Sm₂O₃ nanosheet matrix.

Memristors based on a metastable heterostructure

To investigate the electrical properties of the as-grown samarium-oxygen nanosheets, we fabricated memristors with a vertical configuration on Si/SiO₂ substrate, which employed a top electrode of titanium (Ti) and bottom electrodes of either gold (Au) or graphite (Gr) (Fig. 5a and Supplementary Fig. 14). Depending on the bottom-electrode material, the memristors exhibit different resistance switching (RS) performance (Supplementary Fig. 15). In Au/Ti/SmO₂/Sm₂O₃/Gr memristor, a typical bipolar abrupt RS behavior is exhibited by the current–voltage (I–V) curves over continuous cycles (Fig. 5b). A clear “SET” process is observed at approximately 4 V, characterized by a rapid transition from the HRS to the LRS, which the device maintains its LRS throughout the subsequent voltage sweep from 7.5 V to 0 V. Conversely, a “RESET” process occurs upon application of a reverse bias from 0 V to -3 V, switching the device back to its HRS. Notably, the off-state current reaches as low as 10^-13A at -0.1 V, which is comparable to the values reported in recent oxide-based memristors^64–66. Furthermore, both the HRS and LRS exhibit stable resistance values over the experimental timescale of 10⁴s (Fig. 5c), indicating good retention ability. The endurance characteristics of the device were systematically evaluated over 200 consecutive switching cycles with a compliance current (CC) of 1 μA (Supplementary Fig. 16); the corresponding resistance distributions in both the high-resistance state (HRS) and the low-resistance state (LRS) are presented in Fig. 5d,e, respectively. Throughout the entire cycling test, the device consistently maintained a stable on/off ratio of no less than 10⁵. To further improve the cyclic on/off ratio, endurance tests were performed on the graphene-based devices at a compliance current (CC) of 50 μA. The experimental results show that the cycle life was limited to only 85 cycles. Statistical analysis of the HRS and LRS distributions reveals that, compared with devices operating at a lower CC, the elevated current level yields an on/off ratio of at least 10⁷. However, this enhancement was achieved at the cost of reduced cycling stability, which is consistent with the trade-off between the on/off ratio and endurance observed in oxide-based memristors. We attribute this degradation in device longevity to the excessive Joule heating generated at higher current levels. Preliminary thermal analysis of the conductive filament indicates that at a power of 10 μW, the conservative estimate of the peak temperature is approximately 574 K, which remains below the threshold of 672 K for irreversible phase change (Supplementary Fig. 17). This suggests that the Joule heat generated when working under the aforementioned low-current conditions exerts an acceptable thermal impact on device reliability.

Fig. 5. Memristors based on a metastable heterostructure.

a Structure schematic of Au/Ti/SmO₂/Sm₂O₃/Au and Au/Ti/SmO₂/Sm₂O₃/Gr memristors. b Current–voltage (I–V) switching curves of a Au/Ti/SmO₂/Sm₂O₃/Gr memristor. The arrows indicate the direction of the scanning voltage. c Retention time characteristic of Ti/SmO₂/Sm₂O₃/Gr memristor after SET and RESET operations. d High/low resistance states (H/LRS) of Au/Ti/SmO₂/Sm₂O₃/Gr memristor under 200 cycles(CC = 1μA). e Relative probability of the resistance levels at the LRS and HRS(CC = 1μA). f Statistical distribution of on/off ratio for multiple devices. g Set behavior of Au/Ti/SmO₂/Sm₂O₃/Gr memristor at 3.2 V (1 μs) pulse. The corresponding off switching curve in Supplementary Fig.8. h Benchmark plot comparing the on/off ratio and switching time among the reported memristors and our memristors. Supplementary Table 1 provides the detailed references.

A memristor fabricated with gold bottom electrode (BE) exhibits a slightly lower SET voltage (V_set ~ 3.3 V) and a higher switching ratio reaching 1.2 × 10⁹ (Supplementary Fig. 18). However, the stability of switching operation of Ti/SmO₂/Sm₂O₃/Au memristor is in need of further optimization, as evidenced by cycle-to-cycle variability in performance and a reduced endurance of merely 10 cycles. This limited endurance may be tentatively attributed to the inferior interfacial contact between the Au bottom electrode and oxide nanosheet. The statistical distribution of the on/off ratio for multiple devices, differentiating between their On and Off states, is clearly observable within ranges from 10⁴ to 10¹⁰ (Fig. 5f and Supplementary Figs. 19, 20). While the mechanically exfoliated graphite provides a comparatively smooth and uniform surface, the thermally evaporated gold electrode exhibits a rougher interface with lower resistance. The difference in bottom electrode results in different average switch ratio distributions for the two types of memristors. Besides, switch speeds of the memristors are also measured by applying a 3.2 V (1μs) pulse. The switching time is measured as the time from 10% to reach 90% of the maximum current (Fig. 5g). The current response confirming a transition to LRS, shows the highest speed with a switching time of 47.3 ns, demonstrating nanosecond-scale switching speed. In contrast, a −5V (1 μs) pulse indicates an LRS-to-HRS transition with an actual switching time of 57 ns (Supplementary Fig. 21). A performance benchmarking (Fig. 5h) shows that this device has a large on/off ratio and the same level of switching time compared with other reported approaches.

The resistance switching in our SmO₂/Sm₂O₃ heterostructure arises from the coupled interplay of interfacial polarization and oxygen-vacancy dynamics. To elucidate the switching mechanism in our devices, we have developed a model to delineate the formation and disruption of conductive pathways (Supplementary Fig. 22). Theoretical calculations reveal that Sm₂O₃ is a wide-bandgap insulator (E_g = 3.9 eV) with a valence band maximum (VBM) at -4.44 eV, while SmO₂ possesses a narrow bandgap (E_g = 0.46 eV) and a deeply situated VBM at −7.40 eV, conferring heavily doped semiconductor-like electronic properties(Supplementary Fig. 23). In contrast to the ground-state [Sm] (4 f ⁶6s²), both Sm³⁺ and Sm⁴⁺ ions exhibit a highly localized electronic configuration in its +3 and +4 states, where the 4 f electrons are effectively shielded by the 5s²5p⁶ electron shells and strongly confined near the atomic nucleus⁶⁷. Strong Coulomb repulsion and Hund’s rules⁶⁸ favor high-spin configurations for the 4 f electrons, further suppressing their delocalization. These mechanisms lead to high insulation and low electrical conductivity in samarium oxides. It is worth noting that the SmO₂/Sm₂O₃ heterostructure constitutes the fundamental structural unit of the resistive switching mechanism. Lattice mismatch disrupts inversion symmetry, and local strain fields induced by differences in thermal expansion cause the crystal structure to distort. This process lowers the formation energy of oxygen vacancies, thereby reducing the insulation of the system. Furthermore, the difference in Fermi levels between the two phases results in charge transfer and band bending, establishing an electrostatic potential gradient and inducing interfacial polarization. Consequently, the interface serves as a preferred polar-confined channel for the accumulation and migration of oxygen vacancies. These two effects of the interface structure promote the formation of lattice defects (primarily oxygen vacancies) and facilitate their preferential transport at the interface under an electric field⁶⁹. This implies that the interface serves not only as the site for polarization but also constitutes an endogenous reservoir for oxygen vacancies, thereby establishing it as the preferred pathway for ionic transport⁷⁰.

The applied bias dictates the migration direction of these interfacial oxygen vacancies. The initial positive bias aligns the polarization, creating a localized accumulation of oxygen vacancies along the SmO₂/Sm₂O₃ interface toward the top electrode, progressively aggregating into a conductive filament that bridges the two electrodes. This corresponds to the SET process and the transition to the low-resistance state (LRS). Conversely, the negative bias induces polarization reversal, which destabilizes the filament and promotes oxygen vacancy redistribution. Oxygen vacancies are driven away from the filament region, leading to its partial dissolution and the recovery of the high-resistance state (HRS) during the RESET process. Through the application of alternating bias voltages, the SET and RESET operations are iteratively performed, enabling reversible resistance switching. The resistance switching mechanism, which hinges on the migration of O^2- and V_O, is commonly referred to as the valence change mechanism (VCM)⁷¹. This framework unifies the two seemingly distinct mechanisms: the polarization and the filamentary switching. The polarization modulates the local electrostatic potential and migration barriers, effectively guiding the preferential pathways for oxygen vacancy movement, while the resulting filament provides the macroscopic resistive readout.

This mechanism is substantiated by experimental findings. Cross-sectional EELS was performed at the Ti electrode interface in both HRS and LRS (Supplementary Fig. 24). Near the interface, the Ti L-edge spectra evolve from a doublet in the HRS to a four-peak structure in the LRS, indicating a locally increased Ti oxidation state upon switching. Spatially resolved O K-edge intensity further reveals a pronounced oxygen depletion near the interface in the LRS(Supplementary Fig. 25). The spatial overlap between the Ti valence change and the oxygen-depleted region provides direct evidence that the conductive filament is formed via field-driven oxygen migration, accompanied by reversible redox reactions at the Ti electrode. To directly visualize the conduction pathways, we performed conductive atomic force microscopy (C-AFM) current mapping on the same scanning area in both HRS and LRS (Supplementary Fig. 26). In the HRS, the current map reveals a non-uniform distribution with scattered high-current spots, suggesting preferential leakage paths within the nanosheet. Upon switching to the LRS, sharp and highly localized high-current filaments emerge, with intensities markedly higher than those in the HRS. Notably, these filaments predominantly appear at positions corresponding to the boundary between distinct conductivity regions observed in the HRS, suggesting that the conductive filament is preferentially localized near the SmO₂/Sm₂O₃ heterointerface.

Discussion

We have demonstrated the synthesis of metastable SmO₂/Sm₂O₃ lateral heterostructures via eutectic salt-assisted chemical vapor deposition, where interfacial symmetry breaking induces strong nonlinear optical response and ferroelectricity. On one hand, the heterostructure-based memristors exhibit a maximum on/off ratio exceeding 10⁹, switching speeds as fast as 47.3 ns, and retention over 10⁴s, surpassing many existing resistive switching systems. On the other hand, the switching mechanism is governed by field-driven oxygen vacancy migration along the vertical heterointerface, effectively confining filament formation and enhancing device controllability. These results establish a promising materials platform for high-performance memristors and neuromorphic hardware. For practical application, scalable and reproducible synthesis of such metastable heterostructures remains a challenge. Future efforts should focus on improving the uniformity of large-area growth to enable wafer-scale integration and the reliable implementation of memristive synapses for neuromorphic computing.

Methods

Synthesis

The synthesis was carried out in a 2-inch-diameter quartz tube by eutectic salt assisted CVD. Freshly cleaved fluorophlogopite mica [KMg₃(AlSi₃O₁₀)F₂] sheets were placed in the middle of the tube furnace to serve as epitaxial substrates. The 6 mg SmCl₃·6H₂O powder (99.99%, Aladdin), 2 mg α-Fe₂O₃ (99.99%, 3 A) and 20 mg NaCl-KCl eutectic (NaCl: KCl = 1: 1) was uniformly ground as the precursors and loaded in mica. High-purity Ar gas (99.999%) with a flow rate of 100 sccm was used as the carrier gas. Then, the furnace was heated from room temperature to 980 K in 20 min and maintained at this temperature for another 25 min for sample growth. After the reaction completed, the tube furnace naturally cooled down to room temperature. The substrates were washed with deionized water and acetone several times to remove the salts.

Sample transfer and characterizations

For transferring, poly(propylene carbonate) (PPC) was first spin-coated onto the mica for carrying the samples. After baking on a hot plate with 100 °C for 30 s, poly(methyl methacrylate) (PMMA) solution was spin-coated onto the PPC and the coated samples were heated on a hot plate at 100 °C for 20 min. The PPC/PMMA films with nanosheets were separated from the mica in deionized water and transferred to the needed substrates, followed by drying it at room temperature. Finally, the PPC/PMMA covering the samples was removed using hot acetone (70 °C for 10 min). OM (BX51M, Olympus) and AFM (Dimension Icon, Bruker) were used for morphology and thickness analysis under ambient air. SHG imaging was conducted on a home-made setup with a 1064 nm supercontinuum laser (SC-PRO, Ysl-photonics). TEM, STEM, SAED and EDS mapping were performed on JEM-F200 and JEOL-ARM200F with a cold-field emission gun. PFM was carried out using the same AFM system (Dimension Icon, Bruker) in off-resonance mode, with a conductive tip of spring constant ~ 80 Nm^-1 driven at a frequency of 15 kHz.

Device fabrication and characterizations

The memristors were fabricated on Si/SiO₂ substrates with MIM structure. Nanosheets were first transferred onto the pre-prepared bottom electrodes (5/20 nm Cr/Au) or multilayer graphite exfoliated from single crystal block. Subsequently, the top electrodes (5/60 Ti/Au) were patterned by ultraviolet lithography using AZ5214 as the photoresist and deposited via electron-beam evaporation and thermal evaporation apparatus. A B1500 semiconductor parameter analyser equipped with a probe station (Lake Shore, TTP4) is used to measure electrical characterizations of memristors. All the electrical measurements were conducted at room temperature unless otherwise specified.

Theory calculations

The calculations of density functional theory are done with the projector augmented plane-wave basis, which is implemented in Vienna ab-initio simulation package⁷². And the plane-waves are cut-off at 550 eV. The exchange-correlations of electrons are described by the generalized gradient approximations with the form proposed by Perdew, Burke, and Ernzerhof⁷³. To improve the descriptions on the correlation interactions, the Hubbard U scheme^74,75 is employed with Ueff = 5.0 eV applied on the f-shell of Sm. Vacuum layer of 20 Å is set in z-direction which is supposed to be nonperiodic to eliminate the interactions between images. Dipole correction is considered. The energy converge criterion for solving self-consistent Kohn-Sham equations is 10^-5eV. The Brillouin zone is sampled with resolutions better than 0.03 Å^-1, using the scheme of Monkhorst-Pack⁷⁶. The atomic model of SmO₂ and Sm₂O₃ was constructed using VESTA⁷⁷. The z-component of electronic polarization per cell is defined with:

$$p_z=\int d^{3}r{r}_z\left({\rho }_{ion}\left(r\right)-{\rho }_e\left(r\right)\right)$$ 1

where ${\rho }_{ion}$ and ${\rho }_e$ are the charge density of ion cores and electrons, and the integration is done over unit cell.

Author contributions

This project was supervised and directed by C.X., S.L., J.H. and Y.W.. S.Y. and L.G. contribute equally to this work. Y.W. conceived this work. Y.W. and S.Y. designed the experiments. S.Y., H.Z. (Hao Zhu), J.Z. and S.L. conducted the device fabrication. C.C., H.Z. (Hui Zeng) and J.C. conducted electrical measurements. S.Y., L.G. and Y.T. performed the material characterization. S.Y., L.Y., G.Z. and H.Y. conducted the density functional theory calculation. S.Y., R.C. and S.L. performed the material preparation. All authors contributed to the discussion and analysis of the results. S.Y., S. L., J.H. and Y.W. wrote the manuscript. All authors contributed to the discussion and analysis of the results.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Natural Science Foundation of China (Nos. 12274119, 92464303, U23A20364, 12574128, 62274121) and the Natural Science Foundation of Wuhan (Grants 2024040701010049).

Data availability

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

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77566-1.