Atomic‐Scale Phase Transformation in Perovskite LaCoOx Resistive Switching Memristive Devices

Resistive random‐access memory (RRAM) is considered the next‐generation nonvolatile memory owing to its simplicity, low power consumption, and high storage density. Resistive switching (RS) occurs in a wide range of materials among the transition metal oxides. Herein, an epitaxial ternary metal oxide layer, LaCoOx (LCO), grown on Nb‐doped SrTiO3 substrates, is utilized as an RRAM device. When voltage is applied, it exhibits excellent RS behavior. More than 900 cycles are obtained, and the retention time reaches up to 104 s. To investigate the RS behavior, high‐resolution transmission electron microscopy and atomic‐scale scanning transmission electron microscopy are used to observe the structural evolution and oxygen ion migration in LCO. The structure exhibits a perovskite–brownmillerite topotactic phase transformation from LaCoO2.5 or LaCoO2.67 to the LaCoO3 conductive regions. The reversible transition between the low‐resistance states and high‐resistance states enables the RS mechanism. Additionally, the valence states are confirmed using high‐resolution X‐ray photoelectron spectroscopy. This study not only illustrates the oxygen‐ion migration mechanism of LCO but also demonstrates its suitability for RRAM applications.


Introduction
[3] In recent years, flash memory has dominated because of its high storage density and high erase speed.However, conventional flash memory faces challenges as the technology scales down, such as scalability limits and increased power consumption.In light of these limitations, alternative NVM solutions have emerged.Resistive random-access memory (RRAM) and other contenders such as ferroelectric random-access memory, [4] phasechange random-access memory, [5] and magnetic random-access memory [6] are promising candidates for traditional flash memory.Compared with other NVM technologies, RRAM stands out because it harnesses the resistive switching (RS) phenomenon, enabling it to store data through changes in electrical resistance.9][10] RRAM devices typically feature a metal-insulator-metal sandwich structure, with the RS layer playing a crucial role in their performance and reliability.Resistance switching has been observed in various materials, including transition metal oxides, [11][12][13] ferromagnetic materials, [14,15] low-dimensional materials, [16,17] and ABO 3 perovskite (PV) oxides. [18,19]Typically, the most prevalent mechanism involves the utilization of filament-based mechanisms, where the generation and rupture of conductive filaments induce resistance transitions. [20,21][24][25][26] Topotactic phase transformation refers to a change in the arrangement of atoms or ions within a crystal lattice while preserving the framework.Certain topotactic phase transformations can result in materials with altered magnetic, electrical, or optical properties, making them potentially useful for memristor applications.In this context, ABO 3 PV phases have emerged as promising memory candidates owing to their ordered crystal structure and uniform switching behavior, which offer stable and controlled RS behavior.Unlike filament-based mechanisms, which can introduce variability in set voltages and compromise device stability, ABO 3 PV phases provide consistent and reliable switching operations.This predictability translates into enhanced data retention times, which is crucial for memory applications requiring prolonged storage intervals.29] Recently, lanthanum cobalt oxide (LCO), LaCoO 3-δ (where δ = 0-0.5),has attracted extensive attention due to its extraordinary strip-like structure. [30]33][34] Moreover, the valence state of Co ions changes from Co 3þ to Co 2þ because of the transferred electrons filling the Co 3 d states, which play a critical role in the magnetic ordering of cobaltites, such as LaCoO 2.5 and LaCoO 2.67 .Importantly, this transformation changes physical and chemical properties, including metal-insulator transitions.LaCoO 2.5 stacks with alternating octahedral CoO 6 and tetrahedral CoO 4 layers, LaCoO 2.67 consists of two alternating octahedral CoO 6 layers and one tetrahedral CoO 4 layer, and LaCoO 3 stacks with octahedral CoO 6 layers.Detailed crystal structure information on LCO is presented in Figure S1 and Table S1, Supporting Information. [35]ccording to the previous literature, the ratio of Co─O tetrahedral layers significantly affects the insulating properties of the film due to the broader bandgaps. [36,37]This phenomenon is particularly noticeable in the BM-LCO (LaCoO 2.5 ) and LaCoO 2.67 films, where a higher Co─O tetrahedral ratio leads to pronounced insulation characteristics and V  O (oxygen vacancy)ordering.In contrast, the PV-LCO (LaCoO 3 ) phase displays comparatively lower resistance, demonstrating a more conductive nature.The structural variation between these phases directly influences their distinct resistance values, thereby enabling the resistance-switching behavior essential for RRAM applications.
In this study, we fabricated RRAM devices with epitaxial LaCoO 3-δ (where δ = 0-0.5)thin films grown on conductive Nb-doped SrTiO 3 (Nb-STO) substrates using an Ag metal layer as the top electrode.Three distinct series of LCO, LaCoO 3 , LaCoO 2.67 , and LaCoO 2.5 were prepared.When voltage is applied, oxygen ions can be injected into the oxygen-deficient layer (LaCoO 2.5 or LaCoO 2.67 ), leading to the formation of an oxygenrich LaCoO 3 phase.The reversible transformation between different phases in the switching layer enables the LaCoO x thin films to exhibit low-resistance state (LRS) and high-resistance state (HRS), making them suitable for information storage applications.This transition was clearly evident in our high-resolution transmission electron microscopy (TEM), atomic-scale scanning TEM (STEM) imaging, and energy-dispersive spectrometry (EDS) analyses, with dark stripes indicating the presence of an oxygen-deficient layer.Moreover, we validated the Co oxidation states before and after the electrical measurements using high-resolution X-Ray photoelectron spectroscopy (HRXPS) techniques.It is remarkable that prior research has not explored this specific cobaltite for its memristive properties.Our study has revealed that LaCoO x exhibits promising retention characteristics and demonstrates good reproducibility in its switching behavior, indicating its potential utility in memristive applications.This phenomenon of topotactic phase transformation not only revealed different phase structural evolutions of LaCoO x but also proved to be a promising candidate for novel memory materials.

Results and Discussion
Figure 1 shows a fundamental analysis of the initial RRAM device based on LCO. Figure 1a shows a schematic diagram of the Pt/Ag/LCO/Nb-STO RRAM cell.The fabrication process involved the following steps: first, 20 nm thin films of LCO were deposited on a 0.5 wt% Nb-STO substrate by pulsed laser deposition (PLD).Next, a 130 nm-thick Ag top electrode with a protective Pt layer was deposited using an electron-beam evaporation system (E-gun).The top electrode is patterned using a metal shadow mask with a diameter of 80 μm.We also fabricated devices utilizing Au/Ti as a top electrode in Figure S2-S4, Supporting Information.In our experiments, both LaCoO 2.5 and LaCoO 3 switching layers were explored.The Au/Ti/ LaCoO 2.5 /Nb-STO devices revealed topotactic phase transformation after electrical treatments.In addition, we also demonstrated the structural characteristics and RS properties of Au/Ti/ LaCoO 3 /Nb-STO devices in Figure S3-S4, Supporting Information.Figure 1b shows a cross-sectional STEM image of the Pt/Ag/LCO/Nb-STO device, confirming the excellent contact between LCO and Nb-STO.For further analysis of the elemental composition, EDS was utilized, and the results are shown in Figure 1c, where different colors represent the corresponding components (Ag, La, Co, O, Sr, and Ti).In Figure 1d, the X-Ray diffraction (XRD) spectrum reveals distinct components corresponding to LaCoO 2.67 and LaCoO 2.5 films, each exhibiting unique characteristics.The presence of LaCoO 2.67 components is identified by diffraction peaks of the (300) and (600) planes grown on a (001) Nb-STO substrate.Additionally, threefold superstructure peaks such as (400) and (500) were observed, indicating the ordering of oxygen defects. [38]In contrast, the pristine LaCoO 2.5 film displays twofold superstructure peaks of (006) and (0010).Based on XRD analysis, the dielectric layers of LaCoO 2.67 align with the monoclinic structure, whereas LaCoO 2.5 exhibits an orthorhombic structure.The Nb-STO substrate exhibited a high crystal quality with a constant distance between the two Nb-STO peaks.Figure S5, Supporting Information, shows the peaks corresponding to the (001) and (002) diffractions, indicating PV-structured LaCoO 3 . [39]o explore the RRAM characteristics, thickness-dependent LaCoO 2.5 and LaCoO 2.67 films were adopted.Figure 2a-b illustrates the RS characteristics of the Pt/Ag/LaCoO 2.5 / Nb-STO devices for 20 nm and 40 nm-thick dielectric layers, respectively.Figure 2a shows the first, 100th, and 300th sweep processes.We follow the voltage sweep steps: 1) 0 !À10 V; 2) À10 !0 V; 3) 0 !þ8 V; and 4) þ8 !0 V.The device was initially in a HRS and switched to a LRS during the voltage sweep from 0 to À10 V, called the "set" process.Unlike other materials, LCO dielectric layers do not require a forming voltage. [40,41]This electroforming-free behavior is owing to the high concentration of oxygen vacancies, which migrate under an electric field and result in changes in the resistive state of the device. [42,43]ubsequently, when we applied positive voltage to 8 V, the device "reset" back to HRS.During the "set" operation, we observed relatively small changes in current from HRS to LRS.We believe that this similar to rectification could be attributed to interface energy barriers within the material, which could restrict the current flow and result in a less pronounced switching behavior.However, during the "reset" process, we observed a more distinct on/off ratio, suggesting the presence of RS.As we attempt to "set" the device with a positive bias, as shown by the red curve (arrows 5 and 6) in Figure S6, Supporting Information, we observe a slight change in current, which remains within the same order of magnitude.This result indicates that our LCO devices can only use negative voltage to "set" and positive voltage to "reset" to facilitate the resistance switching.After measuring more than 300 cycles, the on/off ratio remained similar to the measurement in the first cycle.For the 40 nm-thick dielectric layer, the voltage scan range was expanded, as shown in Figure 2b.The device exhibited stable switching behavior for more than 900 cycles.Furthermore, we investigated LaCoO 2.67 films for their RS behavior.The LaCoO 2.67 switching layer also exhibited RS characteristics, as shown in Figure 2c.To confirm stability and repeatability, 600 continuous set and reset cycles were performed.According to a previous report, [39] LaCoO 2.5 film exhibits a higher resistivity than LaCoO 2.67 by more than one order of magnitude.This inherent material property is reflected in our devices, where LaCoO 2.67 demonstrates a lower operating voltage and smaller RS window compared to LaCoO 2.5 .This was attributed to the lower concentration of oxygen vacancies in LaCoO 2.67 , resulting in a smaller change in the resistive state upon switching, as shown in Figure 2d.To ensure the consistency of the RS behavior, more than 20 devices with LaCoO 2.5 switching layer were measured, and all devices exhibited similar average resistive contrasts, as shown in Figure 2e.We also collected the distribution of HRS/LRS for seven different LaCoO 2.67 dielectric layer devices, as shown in Figure S8, Supporting Information.To understand the effect of the dielectric layer thickness on the resistance switching of RRAM devices, we fabricated LaCoO 2.5 layers with thicknesses of 10 nm, 20 nm, 27 nm, and 40 nm, as shown in Figure 2f.RRAM devices with a thickness of 20 nm exhibited a larger on/off ratio.The on/off ratio is a critical parameter in RS devices, as it directly influences the device's performance regarding better signal integrity and reliability.A thin dielectric layer may limit the integrity of the conductive filament formation, while a thick dielectric layer may lead to too dispersed conductive filaments.In this study, we select 20 nm as the primary research device to enhance the device's utility for memory applications.Figure 2g, I refer to 20 nm-thickness LaCoO 2.5 devices.Figure 2g shows the cumulative probabilities of the HRS and LRS values measured at a read voltage of 0.3 V.In this analysis, the HRS exhibits a larger coefficient of variation (0.55) compared to the LRS (0.23), indicating that the distribution of the LRS was narrow, whereas that of the HRS exhibited relatively large variations.In Figure 2h, the resistance states were extracted using a reading voltage of 0.3 V during continuous switching cycling for 600 cycles.The endurance test demonstrated remarkable durability with DC operation and a good on/off ratio between the HRS and LRS.Moreover, the resistance-state retention ability was measured for more than 10 4 s in the degradation test using a constant read voltage (0.8 V) on the device, as shown in Figure 2i.We also provided the IV curve analysis at varying sweep rates and temperatures in Figure S7, Supporting Information.In our experiment, we observed that higher rates caused a small decrease in the on/off ratio, as shown in Figure S7a, Supporting Information.Furthermore, our device exhibits thermal stability in its resistive states across a range of temperatures (25, 50, 80 °C, and then back to 25 °C), as depicted in Figure S7b, Supporting Information.
Atomic-scale STEM imaging was employed to investigate further the RS behavior of the Pt/Ag/LCO/Nb-STO RRAM devices.Before the electrical measurements, Figure 3a reveals a welldefined boundary between the LCO thin film and the Nb-STO interface.A low-magnification TEM and STEM macrographs in Figure S9, Supporting Information, provide a broad sample view.Dark stripes with a periodicity of 3 a 0 (lattice constant) were observed, corresponding to increased interatomic distances, which resulted in a reduced intensity overlap between the La and La atoms.This contrast modulation is frequently observed in oxides with ordered oxygen vacancies.In addition, in the LCO dielectric layer, oxygen vacancy stripes exist in both the vertical and horizontal directions.Figure 3b presents an enlarged atomic-scale STEM image of the region with dark stripes parallel to the LCO/STO interface.Green spheres represent La, whereas purple and red spheres represent Co 3þ and Co 2þ , respectively.In addition, the corresponding fast Fourier transform diffraction (FFT-DP) pattern demonstrated the structure of the initial LCO film before RS.Additional superlattice diffraction spots were observed, confirming the stacking of oxygen vacancy channels, as labeled by the two yellow stripes with diffraction vectors of 1/3 (001) and 2/3(001), corresponding to the LaCoO 2.67 phase.In other regions (Figure 3c), the dark stripes run perpendicular to the LCO/STO interface.The period remains three unit cells.The stacked structure of two CoO 4 tetrahedra and one CoO 6 octahedron was visible in the LaCoO 2.67 matrix.The atomic model in Figure 3d shows a larger La-La interatomic distance in the higher spin states (4.69 Å) than in the other layer (3.46 Å) for every twice unit cells in LaCoO 2.5 and the La-La interatomic distance in the higher spin states (4.63 Å) larger than in the other two layers (3.51 Å) for every third unit cells in LaCoO 2.67 .The larger interatomic distances of the higher spin states were attributed to the increased ionic radius of the Co 3þ ion and the increased Co─O distance.The La-La interatomic distance in LaCoO 3 is 4.11 Å.After the electrical measurements, Figure 3e shows an atomic-scale STEM image revealing the structural transformation of LaCoO 2.67 into a combination of LaCoO 3 and LaCoO 2.5 phases.The formation of LaCoO 3 phases indicates that the devices switch to LRS.The image clearly shows distinct regions where the oxygen atoms migrated and redistributed within the lattice, forming both LaCoO 3 and LaCoO 2.5 domains.The EDS results in Figure S10, Supporting Information, indicate a relative increase in oxygen concentration within the switching layer after electrical treatments (points 3 and 4), which means the formation of LaCoO 3 with a higher oxygen content.This demonstrates that a highly oxygenated PV phase was achieved.Furthermore, the analysis did not reveal the presence of Ag after structure transformation.This could be attributed to the limited availability of defect sites within this single-phase material.Hence, we believe that Ag filaments are not the mechanism behind the switching behavior in our study.
Additionally, Figure 3f-g shows the appearance of newly generated oxygen vacancies, with some predominantly running horizontally (parallel to the substrate) and others running vertically.The corresponding FFT-DP analysis indicates that certain regions revert to the original BM-LCO structure after switching, as indicated by the yellow stripe with a diffraction vector of 1/2 (001).These oxygen ions were injected into the oxygen-deficient layer of CoO 4 via an electric field, leading to a topotactic transition from LaCoO 2.67 to LaCoO 3 , as shown in Figure 3h.Additionally, Figure S11, Supporting Information, illustrates the topotactic phase transformation of the LaCoO 2.5 switching layer after electrical treatment, resulting in regions partially consisting of PV LaCoO 3 , along with other regions, including the BM phase.The STEM results indicated that the switching mechanism of the Pt/Ag/LCO/Nb-STO RRAM device originated from the topotactic phase transformation between BM-LCO and PV-LCO.
To determine the conduction mechanism, we analyzed the I-V curves of the switching processes, as shown in Figure 4a.The I-V curves were transformed into ln-ln plots for the set and reset processes, as depicted in Figure 4b,c.In the set process, the slopes of the LRS and HRS indicate a space-charge limited conduction (SCLC) mechanism.The SCLC model considers three conductive regions: 1) a low-negative-bias region with an approximate slope of 1.3, representing Ohmic conduction.In this region, the current is mainly contributed by free carriers and not by injected electrons from the electrode.Because of the thermal excitation of these carriers, the conduction current is proportional to the electric field.2) The transition region, where the slope increases to ≈2.62, corresponds to Child's law.With an increase in the applied voltage, the injected carriers became predominant.3) High-voltage region, where the current increased sharply and the slope changed to ≈4.41.According to the SCLC mechanism, [44] the electron trap is associated with oxygen vacancies, and the resistance gradually decreases as the vacancies are filled with electrons, resulting in a transition from the HRS to the LRS.In the reset process, ohmic conduction dominates in lower voltage regions, whereas Schottky emission dominates under high electric fields.In Figure 4d, the slope of the HRS within the 0-0.4V range is close to 1, indicating that Ohmic conduction is the dominant mechanism.Conversely, the curves of both LRS and HRS beyond 0.4 V exhibit the Schottky emission mechanism, as evidenced by the linear relationship between ln(I) and the square root of the applied voltage (V 1/2 ) shown in Figure 4e,f.These findings suggest that the conduction mechanism at high voltages in both HRS and LRS is Schottky emission.According to previous studies, the electric field applied to the metal-insulator interface reduces the potential barrier height, facilitating electron flow across the barrier.This results in the current flow governed by the Richardson-Schottky law. [45,46]igure 4g shows the Co 2p X-Ray photoelectron spectroscopy (XPS) spectra of the films in different phases.The peaks at binding energies of 778.23 and 793.40 eV correspond to Co 3p 3/2 and Co 3p 1/2 , respectively.The higher binding energy peaks are associated with Co 2þ ions, whereas the lower binding energy peaks correspond to Co 3þ ions in the LCO film.The satellite peaks at approximately 790 and 785 eV represent the Co 3þ and Co 2þ states, respectively.Only a Co 3þ satellite peak is evident for the LaCoO 3 films, whereas both Co 3þ and Co 2þ satellite peaks are detected for LaCoO 2.67 .These characteristics were consistent with the nominal valence states of Co in the PV-LCO and BM-LCO films. [34]The chemical states of the cycled device were examined by HRXPS.In Figure 4h, the lower binding energy peak (blue line) at 528.91 eV was assigned to lattice oxygen, whereas the higher binding energy peak (red line) at 529.71 eV was attributed to adsorbed oxygen.Figure 4i shows the peaks at binding energies of 833.90 and 850.65 eV, which are associated with La 3d 5/2 and La 3d 3/2 , respectively.As shown in Figure 4j, we determined the binding energies and full width at half maximum (FWHM) and calculated the peak area ratios of Co 2þ to Co 3þ .The Co 2þ /Co 3þ ratios in the initial and cycled LCO films were estimated to be 5.25 and 2.25, respectively.According to I-V curve fitting and HRXPS analysis, oxygen injection and extraction achieved switching between fully oxygen-coordinated LaCoO 3 and oxygen-deficient LaCoO 2.5 or LaCoO 2.67 in the switching layer.This reversible transformation enables switching between HRS and LRS in the RRAM device, allowing information storage and retrieval.
As mentioned above, we provided evidence for the switching mechanism of the Ag/LCO/Nb-STO devices through curve fitting, atomic-resolution TEM, XPS, and annular dark-field STEM (ADF-STEM) analyses.Figure 5a shows the crystal structures of PV LaCoO 3 , BM LaCoO 2.5 , and LaCoO 2.67 .The switching layer of LaCoO 2.5 consists of alternating layers of one CoO 4 oxygen-deficient layer and one CoO 6 fully oxygen-coordinated layer, while LaCoO 2.67 consists of one CoO 4 oxygen-deficient layer and two CoO 6 fully oxygen-coordinated layers.The phase transition from the BM phase to the PV phase involves the rearrangement of the stacks of CoO 4 tetrahedral and CoO 6 octahedral layers into alternating stacks of CoO 6 octahedral layers.Schematic illustrations for the switching mechanism are presented in Figure 5b-i.The first case was LaCoO 2.5 switching layer, as shown in Figure 5b-e.In Figure 5b, during the setvoltage operation, LaCoO 3 (PV-LCO) conductive filaments are likely to form at the upper interface as oxygen ions from the surrounding ambient environment are incorporated into the CoO 4 oxygen-deficient layer.This causes coordination expansion from CoO 4 to CoO 6 , rather than oxygen movement within LaCoO 2.5 , as shown in Figure 5c.This is indicated by the formation of conductive filaments starting at the upper interface.Subsequently, filaments of relatively low resistance can grow vertically downward as oxygen ions migrate from the top interface to the bottom under a large negative voltage application to the top electrode.Eventually, as shown in Figure 5d, the filaments spread throughout the matrix; however, thin gaps (BM-LCO) separated the filaments from the LCO layer.According to a previous study, [47] the Schottky barrier between a very thin BM-LCO gap and Nb-STO enables the thermionic emission to occur and makes the BM-LCO gap not so insulating.Therefore, although a narrow BM-LCO gap exists, electrons can pass through this thin barrier, and the devices can switch to an ON state.The microstructure of the LCO was retained in the PV structure before the reset process.During the reset process, oxygen extraction occurred at the upper interface, prompting oxygen ions to migrate from the bottom to the top.This led to the gradual transformation of PV-LCO back to the BM-LCO structure.This resulted in filament rupture at the bottom, as depicted in Figure 5e.STEM images investigated the formation of the conductive region in Figure S11, Supporting Information.In Figure S11d-e, Supporting Information, the switching layer was initially composed of pristine LaCoO 2.5 .After the RS behavior, as shown in Figure S11f,g, Supporting Information, the multiple conductive regions consisting of LaCoO 3 almost extend through the switching layer, which is marked with the white dashed line.It keeps 2 nm BM-LCO gaps near the Nb-STO bottom electrode.By applying a negative bias, we observed the initiation of a structural transition toward PV-LCO starting from the top region of the device, as shown in Figure S11c, Supporting Information.This phase analysis result suggests that the movement of oxygen ions commences from the top and progresses downward, facilitating the transformation from an oxygen-deficient phase to the oxygenrich phase of LCO.During the set process, the incorporated oxygen ions can migrate and re-establish the PV-LCO conductive regions, thereby restoring the LRS.Conversely, in the reset process, oxygen extraction led to oxygen ions migration from the bottom to the top.The PV-LCO gradually transformed into a BM-LCO structure.In the following cycles, the microstructure of the LCO can be reversibly changed between BM/PV during the set and reset operations.Finally, the switching layer contains a residual conductive region at the top.Consequently, the Schottky barrier height increased, leading to a HRS.
When the initial state was LaCoO 2.67 , it exhibited higher electrical conductivity compared to LaCoO 2.5 .Figure 5f-i shows the switching mechanism diagram.The TEM/STEM images in Figure S9, Supporting Information, illustrate the initial stages of the topotactic phase transformation of LaCoO 2.67 switching layer.When a negative voltage is applied, the switching layer partially transforms into LaCoO 3 , as illustrated in Figure 5g.Conversely, applying a positive voltage causes the switching layer to change back into the original LaCoO 2.67 phase, as shown in Figure 5h. Figure S9c, Supporting Information, also demonstrates that the LaCoO 2.67 phase is retained in the bottom region of the dielectric layer (Figure S9f,g, Supporting Information), supporting our claim that the switching layer returns to the LaCoO 2.67 phase during the reset process.After multiple switching cycles, as shown in Figure 3e and 5i, oxygen vacancies can also be injected into LaCoO 2.67 when a positive bias is applied, forming the LaCoO 2.5 phase.This transformation was verified using in situ TEM in Figure S12, Supporting Information.Initially, in Figure S12a, Supporting Information, a switching layer corresponds to LaCoO 2.67 .As the bias increases (Figure S12b-e, Supporting Information), the LaCoO 3 phase appears at the top.When the bias increased to À7 V, the conductive regions spread throughout the matrix.After in situ electrical experiment (0 to À7 V with a voltage step of 0.05 V s À1 ), some regions near LaCoO 3 consisted of pristine LaCoO 2.5 .The corresponding FFT-DP patterns are displayed in Figure S12g-i, Supporting Information.Figure S13, Supporting Information, presents the results of electron energy loss spectroscopy conducted before and after the electrical treatment.A shift in the Co L 3 peak toward a lower energy and a higher L 3 /L 2 ratio may indicate a reduction in the oxidation state of the cobalt ions.When Co is in the Co 2þ oxidation state, the occupancy of the 3d electron states is higher than that of the Co 3þ state, which has a higher population at the L 3 level than at the L 2 level. [48,49]he simultaneous formation of LaCoO 2.5 and LaCoO 3 within the LaCoO 2.67 phase can be attributed to their relative thermodynamic stability.LaCoO 2.5 and LaCoO 3 were both more stable than LaCoO 2.67 .When oxygen ions migrate out of the LaCoO 2.67 phase, they leave behind vacancies, creating regions with oxygen deficiency.This results in the formation of two distinct regions within the material: one with oxygen migrating into the LaCoO 2.67 matrix, which promotes the transition to LaCoO 3 and another that favors the formation of LaCoO 2.5 .These observations provide valuable insights into the modifications that occur in the electronic structure and oxidation state of Co during cycling.In this work, the volumetric structural switching behavior grew from the top to the bottom, resembling a filament connection between the top and bottom electrodes, contributing to the conductivity.The structural variation between the BM/PV phases directly influences their distinct resistance values, thereby enabling the resistance-switching behavior essential for RRAM applications.

Conclusions
In conclusion, we investigated the switching mechanism of Pt/Ag/LCO/Nb-STO RRAM devices using LaCoO x as a dielectric switching layer.We observed the structural evolution through ADF-STEM imaging, confirming the topotactic phase transformation from LaCoO 2.5 or LaCoO 2.67 with a CoO 4 oxygendeficient layer to LaCoO 3 with CoO 6 fully oxygen-coordinated layers by electrical control.The valence changes observed in the XPS analysis support the different chemical states of the phases.A Co 3þ satellite peak is evident for the PV-LCO films, whereas Co 2þ satellite peaks are detected for the BM-LCO films.Furthermore, the device exhibited excellent electrical behavior.The RRAM device exhibited remarkable cycling stability with over 900 cycles and a retention time exceeding 10 4 s.Moreover, LaCoO 2.67 demonstrated a lower operating voltage than LaCoO 2.5 .Overall, our findings contribute to the understanding of the switching mechanism in RRAM devices based on LCO and provide insights into the utilization of LCO as a viable material for future memory applications.

Experimental Section
The Fabrication Process of the RRAM Devices: The LCO epitaxial thin films were deposited using PLD with a KrF excimer laser (λ = 248 nm) onto a 0.05 wt% Nb-STO (001) substrate at an oxygen pressure of 100 mTorr and a temperature of 650 °C.After growth, the LaCoO 2.67 samples were obtained through annealing at 500 °C under an oxygen pressure of 1 mTorr for 30 min, while the LaCoO 2.5 samples were obtained during cooling at high vacuum conditions. [50,51]LCO thin films with thicknesses of 10, 20, 27, and 40 nm were fabricated.Subsequently, a 130 nm Ag top electrode and 30 nm Pt layer were deposited onto the LCO thin film using an E-gun at a deposition rate of 0.5 Å s À1 .The electrode was patterned using a diameter of 80 μm metal shadow mask to define the desired device geometry.
Measurement of Electrical Properties: The electrical properties of the fabricated RRAM devices were characterized using semiconductor parameter analyzers (Agilent 4145 B and Agilent B1500A).All the electrical measurements were conducted using a biased top electrode and a grounded Nb-STO bottom electrode.I-V curves were measured in direct current voltage sweep mode with a voltage step of 0.05 V s À1 and a compliance current of 20 mA.The endurance properties were determined from the measured results after 900 cycles.Retention measurements were performed at a reading voltage of 0.8 V for a duration of 10 4 s.
Microstructural Characterization of Materials: The microstructural evolution of the LCO thin films was characterized using TEM and ADF-STEM.TEM lamellar specimens were prepared using a TESCAN GAIA3 dualbeam focused ion beam (FIB) system, controlling the thickness to below 100 nm to obtain high-resolution images, as shown in Figure S14, Supporting Information.The prepared lamellar samples were then transferred onto Cu grids using glass tips.TEM imaging and EDS elemental analyses were performed using a field-emission TEM instrument (JEOL-F200) operated at an accelerating voltage of 200 kV coupled with an Oxford EDS 100 TLE system for elemental analysis.The in situ TEM specimens were also fabricated using the FIB system and transferred onto an electrical chip using a glass tip.Platinum (Pt) wires were deposited onto the sample using FIB to establish connections between the electrodes of the sample and chip, as illustrated in Figure S15, Supporting Information.Subsequently, the prepared sample was inserted into a specialized in situ TEM holder (Protochips Aduro 300).By applying a bias through the specimen, the switching behavior was observed using TEM (JOEL-F200).In addition, atomic-scale ADF-STEM analysis and valence state confirmation were performed using a STEM instrument with a Cs-corrector (JEOL JEM-ARM200F) at an accelerating voltage of 200 kV.The chemical states of the LCO thin films were analyzed using high-resolution XPS with a ULVAC-PHI PHI Quantera II system.

Figure 1 .
Figure 1.The basic analysis of the initial state of LCO devices.a) The schematic diagram and the process flow in the fabrication of the Pt/Ag/LCO/ Nb-STO RRAM devices.The bias is applied to the Pt/Ag top electrode, and the Nb-STO substrate is grounded.b) The cross-sectional STEM image of pristine LCO thin films RRAM devices.c) The EDS mapping of Ag, La, Co, O, Sr, and Ti.d) The XRD spectrum reveals distinct peaks associated with LaCoO 2.67 and LaCoO 2.5 , representing different components within the LCO thin film.

Figure 2 .
Figure 2. The electrical properties of Pt/Ag/LCO/Nb-STO devices.a) I-V characteristic curves corresponding to the following voltage sweep steps for the 20 nm-thick LaCoO 2.5 switching layer: 1) 0 !À10 V; 2) À10 !0 V; 3) 0 !þ8 V; and 4) þ8 !0 V.The red, yellow, and green curves indicate the first, 100, and last 300 cycling tests, respectively.b) I-V characteristic curves for the 40 nm-thick LaCoO 2.5 switching layer.c) I-V characteristic curves for the LaCoO 2.67 switching layer.(d) Switching characteristics for two different dielectric layers.e) Distributions of the HRS/LRS from 20 devices with 20 nmthick LaCoO 2.5 switching layers.f ) Switching layer thickness dependence of resistance shows that 20 nm has the best switching window.g) The cumulative probability of resistance.(h) Endurance measurement at the reading voltage of 0.3 V. i) Retention measurements for both HRS and LRS.

Figure 3 .
Figure 3. Atomic-scale STEM images of Pt/Ag/LCO 2.67 /Nb-STO RRAM devices and the schematic diagrams of LCO structures.a) Atomic-scale STEM observation before RS.b,c) Enlarged atomic-scale STEM images of LaCoO 2.67 phase, which consisted of one tetrahedral CoO 4 layer with two octahedral CoO 6 layers, where green spheres represent La, and purple and red represent Co.Both horizontal and vertical layers can be observed.d) Atomic model for the vertical modulations of the atomic projection along the (100) plane in the LCO films.e) The ex situ atomic-scale STEM observation after RS in LRS.f,g) Atomic-scale STEM image of LaCoO 2.5 phase, consisting of one tetrahedral CoO 4 layer with one octahedral CoO 6 layer.h) Some regions were PV type, which means LaCoO 3 phase.According to the FFT-DP, the yellow arrows indicate the positions of superstructure spots, referred to as LaCoO 2.67 and LaCoO 2.5 .These images show the structural evolution from LaCoO 2.67 to LaCoO 3 conductive regions.

Figure 4 .
Figure 4. Electrical measurements of the conducting mechanism of the device and HRXPS spectra analysis of the phase transformation of the LCO 2.67 dielectric layer: a) I-V curve for LCO dielectric layer.Log (I)-Log (V ) plots of the b) set and c) reset processes.d) Ohmic conduction mechanism.Log(I)-V 1/2 plot fitting well with the e) HRS and f ) LRS of the reset process, which indicated that the conduction mechanism is Schottky emission.g) Co 2p 3/2 and 2p 1/2 peak for the LCO thin films before and after cycling.The peaks associated with Co 2þ and Co 3þ are indicated.The XPS spectra of h) O 1s and i) La 3d of the films, respectively.j) The table of Co 2p 3/2 and 2p 1/2 spectra of binding energy and FWHM value.

Figure 5 .
Figure 5. Schematic diagram of the RS in the Ag/LCO/Nb-STO device: a) crystal structures of LaCoO 3 , LaCoO 2.67 , and LaCoO 2.5 , respectively.b-e) The switching mechanism based on LaCoO 2.5 switching layer.f-i) The switching mechanism based on LaCoO 2.67 switching layer.The pink mark represents LaCoO 3 regions.The red spheres represent oxygen ions.