Direct Visualization of Charge Migration in Bilayer Tantalum Oxide Films by Multimodal Imaging

Inspired by biological neuromorphic computing, artificial neural networks based on crossbar arrays of bilayer tantalum oxide memristors have shown to be promising alternatives to conventional complementary metal‐oxide‐semiconductor (CMOS) architectures. In order to understand the driving mechanism in these oxide systems, tantalum oxide films are resistively switched by conductive atomic force microscopy (C‐AFM), and subsequently imaged by kelvin probe force microscopy (KPFM) and spatially resolved time‐of‐flight secondary ion mass spectrometry (ToF‐SIMS). These workflows enable induction and analysis of the resistive switching mechanism as well as control over the resistively switched region of the film. In this work it is shown that the resistive switching mechanism is driven by both current and electric field effects. Reversible oxygen motion is enabled by applying low (<1 V) electric fields, while high electric fields generate irreversible breakdown of the material (>1 V). Fully understanding oxygen motion and electrical effects in bilayer oxide memristor systems is a fundamental step toward the adoption of memristors as a neuromorphic computing technology.


Introduction
Fundamental physical principals limit future scaling of semiconductor technology according to Moore's law and lead to a prohibitive increase in energy cost for future computing.The global energy demand of computational applications is projected to increase to ≈20% by 2030. [1]This is driven by inefficiencies built into conventional complementary metal-oxidesemiconductor (CMOS) architectures.Furthermore, industrial DOI: 10.1002/aelm.202300589processing is reaching the theoretical limit of transistor size (2-3 nm), which, together with increasing challenge of heat dissipation, caused Moore's law to start failing around 2010. [2,3] This theoretical limit is dictated by the scale where transistor physics transitions from conventional to quantum physics dominating the carrier transport.As a strategy to address the growing computation needs associated with artificial intelligence, analog memory computing approaches using various tunable resistor device concepts are emerging, including ferroelectrics, phase change, electrochemical and filament forming valence change memories. [4,5]esistive switching random access memory (ReRAM) devices or memristors, have many advantages compared to other emerging non-volatile memory (NVM) device types, including extreme scalability, compatibility with CMOS, excellent retention, and the realization of up to 2048 (11 bit) analog states.
[13] Nevertheless, challenges associated with wide spread adoption of TaO x ReRAM analog NVM remain.For example, "denoising" procedures that can dramatically increase the analog state density are required after each "write" operation, which greatly complicates training. [14]hese materials demonstrate resistive switching by two main mechanisms: the valence charge mechanism (VCM) and electrochemical metallization (ECM).The ECM has been demonstrated by a transition from a high resistance state (HRS) to a low resistance state (LRS) caused by the formation of a conducting filament of metallic ions that percolates through the TaO x active layer, effectively short circuiting the otherwise insulating oxide layer. [7,10,15]This phase change is driven by a localization of the electric field by interface roughness resulting in joule heating and breakdown.ECM is not observed in in-situ scanning transition electron microscopy (STEM) experiments if the electrode/active layer interface has a low surface roughness, suggesting that conductive filaments need a localized electric field caused by interface roughness in order to form, but the threshold of roughness needed is still unclear. [8]The VCM can be observed by a smooth exponential transition from the HRS to the LRS. [8,9]This formation-free resistance change is caused by a buildup of cation vacancies at the electrode/active layer interface, which reduces the Schottky barrier height, and increases the probability of electron tunneling.Devices that operate via the VCM generally operate at a higher voltage compared to those that operate by the ECM, however the devices are more stable relative to devices driven by filament formation.
Kim et.al. showed that Si doping can improve the on/off ratio of TaO x devices at only 2.7 at% Si. [10] An increased on/off ratio in a filamentary device is caused by a thicker filament in the LRS and a wider depleted gap between the filament and electrode in the HRS; both of which suggests a higher oxygen ion mobility.For silicon-doped TaO x devices, this higher oxygen mobility was justified by Si induced ion conductivity channels.These weak bonding channels are caused by strong bonding in proximity to the Si ions, allowing for weaker bonding away from the Si ions and faster oxygen ion motion in the material overall.These ion conductivity channels can be simulated by the atomic structure, utilizing density functional theory (DFT), but this method can be inconclusive in amorphous oxide materials, since it is challenging to accurately simulate amorphous structures.Though doping presents an enticing path to tuning resistive switching in oxide films, the VCM and ECM mechanisms must be understood in TaO x in order to fully understand the effects of doping.
Characterization of memristive oxides requires a means to electrically inducing and detecting ion motion.Atomic force microscopy (AFM) is a prime candidate to both induce and detect local ion motion.AFM has been used to investigate resistive switching along with local IV measurements. [16,17]Conductive atomic force microscopy (C-AFM) has also been used to induce and measure low and high conductance states in oxide films. [16]n thin films, kelvin probe force microscopy (KPFM) can also detect ion motion through changes in contact potential difference (CPD) between the tip and sample surface, though interpretation of the corresponding results is complicated by a number of auxiliary phenomena, such as charge trapping and surface potential (ionic motion).These phenomena cannot be investigated by AFM means alone.In order to study them here, we use spatially resolved time-of-flight secondary ion mass spectrome-try (ToF-SIMS) to characterize possible chemical changes in tantalum oxide devices.[20][21][22][23] Previous work using a combined workflow using C-AFM/ToF-SIMS demonstrated that is was possible to track oxygen motion induced by C-AFM in a SiO x active layer, but showed that C-AFM in ambient conditions is prone to driving moisture into the active layer. [24]Therefore, to capture true oxygen motion a C-AFM/ToF-SIMS workflow needs to be conducted in the same high vacuum environment to allow for induction and measurement of the ionic motion in the active layers without the effects of ambient oxygen or moisture.This experimental workflow also allows for a sputter cleaning of the surface prior to C-AFM measurements to ensure good probe contact with the active layers.
In-situ STEM characterization of a 40 nm TaO x device operation was analyzed by Ma et.al. [25] A 10-20 nm wide oxygen deficient filament formed between the top and bottom TiN electrodes.Lateral oxygen movement by temperature-driven Fick's diffusion was suggested to be the driving mechanism for filament formation, rather than vertical diffusion in the active layer.Oxygen was not observed to move in or out of the electrode in the EDS maps shown.This study contributes evidence of how tantalum-rich filaments form in a single layer TaO x sandwich device structure and shows that lateral diffusion of oxygen is possible in these systems.
In this study we utilize multimodal chemical imaging to investigate charge trapping and ionic motion effects in poled amorphous Ta 2 O 5 /Ta bilayer films.To accomplish this, films were dynamically poled using C-AFM and subsequently analyzed by KPFM and ToF-SIMS in order to measure changes in CPD and ionic motion, respectively.These experiments help illuminate the mechanisms that contribute to resistive switching in tantalum oxide films.KPFM showed a nonlinear CPD response.Under vacuum, low applied voltage (< 1 V) during C-AFM measurements show reversable ionic motion that could be observed in ToF-SIMS measurements.Larger applied voltage (> 1 V) generated irreversible breakdown of the film, which was also corroborated by ToF-SIMS measurements.These results provide insight into the different mechanisms that are present in these bilayer oxide films and enable better understanding of resistive switching in these materials.

Results and Discussion
In order to simulate the energy needed for oxygen and tantalum to diffuse, an amorphous Ta 2 O 5 structure was created and ion activation energies for diffusion processes were simulated using plane-wave density functional theory (DFT) and molecular dynamics (MD), respectively.Plane-wave DFT calculations were carried out using VASP and the projector augmented wave (PAW) pseudopotentials for electron-ion interactions. [26]xchange-correlation interactions were accounted for using the generalized gradient approximation (GGA) functional of Perdew-Burke-Ernzerhof (PBE). [27]Starting with crystalline Ta 2 O 5 , we used a supercell structure containing 168 atoms (i.e., Ta 48 O 120 [28,29] ) with optimized lattice constants of x = 14.635Å, y = 12.377 Å, and z = 11.761Å.During structural optimization, both atoms and cell volume were allowed to relax until the residual forces fell below 0.02 eV Å −1 , with a cutoff energy of 520 eV and Gamma-point only k-point sampling.Next a melt-and-quench approach was used in ab-initio molecular dynamics (AIMD) simulations to generate an amorphous Ta 48 O 120 structure and its amorphous derivative structures with different oxygen vacancy concentrations, including Ta 48 O 119 , Ta 48 O 96 , and Ta 48 O 84 . [30,31]We note that this model has shown very good results, agreeing with experiments based on cathodoluminescence (CL), electron energy loss spectroscopy (EELS) and scanning microwave impedance microscopy.For the amorphous Ta 48 O 119 (see Figure S1, Supporting Information), we then ran AIMD using a Nose-Hoover thermostat for 5 different temperatures, ranging from 300-2000 K and used the mean square displacement (MSD) of Ta and O via Einstein's relation to find the respective diffusion coefficients.Using the Arrhenius equation, we then determined the activation energy for the diffusion process.Oxygen vacancy migration for the amorphous model was found to be ≈1.2 eV, which is in accord with other simulation results. [31]The diffusion barrier for Ta was considerably higher, ≈4 eV.The origin of the relatively low activation energy of oxygen vacancy migration has been previously detailed by Hur. [33]s highlighted, the large mobility of oxygen vacancies predicted by a switching model can be explained by the relatively low activation barrier of positively charged oxygen vacancies.As described below, this is in good agreement with our experimental findings.
Inducing and measuring electric field driven oxygen vacancy migration was first done via C-AFM/KPFM.After using C-AFM to apply a range of electric fields to the surface between +/− 4 V, a large change in CPD is observed in the subsequent KPFM scan.A slight swelling of the film is observed in Figure 1 (< 1 nm) after applying a positive C-AFM bias, while negative applied bias resulted in virtually no topography change.The cause of the film swelling under positive bias is unclear, but is most likely associated with pulling negative charge to the surface.At tip voltages between 4 and −4 V, only the 4 V region showed measurable current readings during the C-AFM scans, but significant changes in CPD were observed in the subsequent KPFM scans at all applied voltages.The change in CPD is observed in Figure 1d,e,f for pristine and silicon irradiated regions.An increase in CPD is observed after a positive electric field is applied and naturally, the CPD decreases when a negative field is applied.This effect is due to electrons being pushed and pulled to the near surface region and being trapped.Trapped electrons at the surface of the material, result in a higher work function because of the negative space charge induced at the surface.Conversely, a negative electrical bias applied to the AFM probe depletes the near surface region of electrons, resulting in a positive space charge and lower work function.The asymmetry in the CPD difference for positive versus negative polarity is not clearly understood but could be due to the asymmetry in the number of surface states versus bulk states.The regions exposed by silicon are not as susceptible to CPD change by the negative applied bias which suggests less positive space charge.To elucidate the effect of the silicon irradiation, SRIM simulations (Figure S2, Supporting Information) were performed and the silicon doping was estimated by normalizing the implanted Si concentration to the areal dose applied to the samples.As shown in Figure S3 (Supporting Information), the concentration of silicon in the Ta 2 O 5 /Ta layers is negligible (< 1.4 at.%Si) and thus well below the doping levels reported by Kim et al (2.7 at.%Si). [10]We attribute the changes in the observed KPFM to ion-induced defects caused by electronic and nuclear energy loss, namely damage accumulation induced by the ion beam, which produces deeper traps and thus less mobile charge.The charge trapping effects are significant and appear to dominate over ion motion effects as the CPD dissipates noticeably over 14 hours (Figure S4, Supporting Information).Our calculated activation energy for thermal driven diffusion in an amorphous TaO x system shows that oxygen vacancies require less energy to diffuse compared to Ta ions and although electric field driven ion migration is different than thermally driven ion migration, it is possible that oxygen vacancies are being pulled or pushed through the film by this applied AFM tip polling.In order to separate charge trapping and ionic motion effects, ToF-SIMS was employed to directly measure ion contrast in the films.
The C-AFM and ToF-SIMS workflow described above was employed to induce and image ionic motion effects respectively.This workflow also enables induction of electric field driven ion motion by C-AFM and direct measurement of the ion motion by ToF-SIMS in the same area of the film without breaking vacuum.The ToF-SIMS depth profile of the 50×50um C-AFM area shows that oxygen ions appear to be the primary species that are moved by the generated electric field.This oxygen migration can be seen below in Figure 2, where the Cs 2 O + ion count (Figure 2a) at the Ta 2 O 5 /Ta interface is presented.Figure 2b shows a sum of oxygen ion and normalized tantalum oxide ion counts (O + ion count, TaO + normalized with the Ta + ion count and TaO 2 + ion count normalized to the Ta + ion count).This C-AFM/SIMS data agrees with the previously calculated thermal diffusion activation energy for oxygen vacancies and tantalum in an amorphous Ta 2 O 5 system.This thickening of the Ta 2 O 5 layer under applied positive polling is also consistent with in situ STEM work done by Myong-Jae Lee et.al. [8] where oxygen is observed moving in the opposite direction to the applied electric field, effectively increasing the Ta 2 O 5 layer thickness.The trace and retrace images from the C-AFM measurement (Figure S5, Supporting Information) show an increase in current from the trace to the retrace, suggesting this thickening of the Ta 2 O 5 layer was enough to increase the conductivity of the film stack.The ion profiles of other relevant ions in the film stack are plotted in Figure S6 (Supporting Information).Unlike oxygen, the other ions in the stack (Ta + , TaO + and TiN + ) do not seem to move under these low electric fields, suggesting that the barrier for oxygen motion under applied electric fields is lower compared to the other ions in the system.
A second C-AFM/ToF-SIMS experiment was done in order to investigate reversibility of observed ion motion.The first C-AFM scan of this experiment was done to induce reversable ionic motion without breaking down the film stack, employing low applied voltages of 0 to 1 in 0.25 V increments.The current maps from this first C-AFM scan can be seen in Figure S7 (Supporting Information).The pristine RMS surface roughness, taken from the top 1/5th of the C-AFM scan where no AFM tip bias was applied was 9.61 nm.The RMS roughness of the subsequent regions corresponding to tip biases of the 0.25, 0.5, 0.75, and 1 V regions were 6.09, 3.28, 6.14, and 11.01 nm, respectively (Figure S8, Supporting Information).The total roughness of the entire scan was 17.19 nm.This low surface roughness suggests the film did not experience dielectric breakdown or filamentary resistive switching.A second C-AFM scan was done that overlapped with the first, but the scan direction was rotated 90 degrees.This second scan attempts to reverse the ionic motion induced by the first scan by applying low negative voltages of 0 to −1 V in 0.25 V increments.Subsequent ToF-SIMS measurements show that applied electrical bias moved small amounts of oxygen from the Ta/TiN interface into the Ta layer with an applied positive bias (Figure 3a).In order to visualize oxygen motion through the material stack, the ion counts for tantalum oxide species (sum of TaO + , TaO 2 + , Ta 2 O 5 + ) were normalized with the respective tantalum species (sum of Ta + , Ta 2 + ).These visualizations are shown in Figure 3.The average normalized oxygen ion count in the TiN layer under the 0.5 to 1 V bias region was found to be 19.5% lower compared to the total average oxygen ion count in the unbiased TiN layer.Subsequently, the oxygen ion count in the area above this oxygen deficient region was found to be 7.5% higher compared to the total oxygen ion count in the Ta layer.This change in oxygen concentration in the 0.5 to 1 V biased area is attributed to oxygen movement toward the positively charged electrode.Depth profiles of this biased region compared to a pristine region can be seen in Figure S9 (Supporting Information).No oxygen motion is observed in Figure 3 at 0.25 V which suggests that this is below the activation voltage needed to initiate this oxygen motion.Above 0.25 V, oxygen can be seen migrating toward the positive surface.This redistribution of oxygen through the film stack is enough to noticeably change the conductivity of the film stack.This change in conductivity can be seen by the difference between the current trace and retrace of the C-AFM scans that induced the oxygen motion (Figures S7, S8, Supporting Information).3a) that is reversed in Figure 3b).
When a subsequent negative bias (−0.25 V) is applied to the adjacent region, the oxygen migrates into the TiN electrode interface.This can be seen by a 9.1% increase in the oxygen ion counts under the biased area in column 2. Though this oxygen movement is subtle and different from the oxygen movement observed in the first C-AFM/ToF-SIMS experiment (Figure 2), it is consistent with expectations that oxygen moves in this Ta 2 O 5 /Ta bilayer system under small (< 1 V) applied electric fields.Migration of TiO + into the TiN electrode at every negative applied bias can be seen in Figure S11 (Supporting Information) along with breakdown of the film from −0.5 to −1 V.This suggests oxygen motion away from a negatively charged surface.The average oxygen counts as a function of depth for columns 1 and 2 are shown in Figure 3, along with corresponding schematics of the overall oxygen movement.
During the second C-AFM scan of the oxygen motion reversibility C-AFM/ToF-SIMS experiment (negative bias applied to the surface), breakdown of the film began to occur in the −0.5 V region and persisted for the remainder of the scan.Breakdown occurring at −0.5 V and not at +0.5 V demonstrates an asymmetric electric field polarity effect in this system.This effect can be seen in point current-voltage measurements of an oxygen cleaned area, where a negative applied electric field results in a larger current compared to an equal positive electric field (Figure S12, Supporting Information).Relevant ion profiles from this experiment can be seen in Figure S13 (Supporting Information).
Upon applying higher voltages (+/− 4 V), these systems undergo irreversible breakdown where the applied energy is higher than the activation energy for diffusion of tantalum and oxygen, resulting in atomic motion for all species present in the system.When the voltage increases to the point of breakdown, a third mechanism driven by joule heating emerges.Larger voltages lead to breakdown of the entire film stack, where the breakdown is localized to directly under the AFM tip since a straight line between the AFM tip and bottom electrode will predominantly be the shortest circuit distance.The migration of O 2− is demonstrated by the TaO + ion count shown in Figure 4.A swelling of the film at positive applied bias can also be seen in the topog-raphy post poling (Figure 4b), suggesting a similar charge trapping mechanism to the ambient C-AFM/KPFM experiment discussed earlier.There was no significant variation in ion counts between positive and negative biased areas (Figure S14, Supporting Information), suggesting the breakdown mechanism is primarily current driven.This breakdown is the driving mechanism of filament formation in these amorphous tantalum oxide films.Filaments are prone to breaking when an opposite electric field polarity is applied, returning the film to its high resistance state.Therefore, electric field effects must be, at least partially, involved in filamentary resistive switching as well.
When applying these larger electric fields to the surface (C-AFM), this breakdown mechanism is easily observed by significant changes in topography (average RMS surface roughness of 29.58 nm) and significant ion migration through the stack (Figure S16, Supporting Information).An increase of TaO + concentration in the Ta layer of the film stack correlates to an increase in oxygen in the Ta layer.The oxygen appears to build up at the Ta 2 O 5 /Ta interface, which supports previous proposed mechanisms for filament induced resistive switching.The TaO + depth profile of the biased and pristine regions is shown in Figure 4a, next to the depth profile of oxygen species in the film stack.

Conclusions
The tapering of Moore's Law in the coming decade, combined with ever increasing demands on data processing, will greatly accelerate the energy consumption of information technology (already the fastest growing consumer of energy worldwide) in order to maintain growth in computing power.To sustain continued increase in computing power at a significantly reduced energy consumption, fundamental materials research must be undertaken to underpin and enable transition to new computing principles and new types of computing devices.In this work we investigate the switching mechanism in tantalum oxide memristive devices by directly visualizing ion migration via a unique workflow combining C-AFM and ToF-SIMS.We found that by applying an electric field across a Ta 2 O 5 /Ta film stack results in an interplay of several mechanisms including: nonlinear charge trapping, reversable oxygen motion and irreversible breakdown.The degree to which these mechanisms dominate depends on the amount of current generated by the applied voltage.If an electric field is applied and accompanied by a low current through the system, charge trapping is the dominant mechanism.If the electric field is excessive, a large current is generated, and the film breaks down by joule heating and ions from the film and electrodes start to intermix.At modest voltages (<1 V) a mechanism associated with field assisted the migration of oxygen/oxygen vacancies in the film can cause changes in conductivity but does not irreversibly breakdown the film.This type of electric field driven oxygen diffusion is more stable if it can be controlled.In the future, more effort will be placed into understanding this sub-breakdown ionic motion and its potential in resistive switching devices.Changes in film thickness and/or chemistry will also be investigated in order to vary the resistance of the films.This change in resistance will enable clearer separation of the electric field effects from the current effects.Molecular dynamic simulations of these structures could also be useful.Simulating the energy needed to migrate each ion type through TaO x layer could define the operational window for these devices where the lower boundary of the operational window would be defined by the minimum energy needed to migrate oxygen and upper boundary would be the minimum energy needed to migrate tantalum.

Experimental Section
In order to investigate the driving mechanism of resistive switching in bilayer memristive oxide devices, thin films of tantalum oxide + tantalum on a titanium nitride electrode (Ta 2 O 5 (10 nm) / Ta (15 nm) / TiN (20 nm)) were deposited on an 8-inch P-type silicon substrate via a combination of reactive DC magnetron sputtering and inert gas sputtering (see [32] for overview of reactive sputtering of Ta 2 O 5 ).The entire deposition was done without breaking vacuum.The 15 nm Ta layer acts as a gettering layer, which removes oxygen from the neighboring Ta 2 O 5 layer.The functional bilayer of Ta 2 O 5 /Ta was shown to be ≈25 nm total (10 nm/15 nm each, respectively) and no crystalline peaks were identified in the structure as measured by X-ray reflectivity (XRR) and X-Ray diffraction (XRD), respectively (Figure S17, Supporting Information).XRD and XRR measurements were conducted on a PANalytical X'Pert Pro MRD equipped with a proportional Xenon counter.For the XRD and XRR measurements, the X-ray beam was generated at 45 kV/40 mA, and the X-ray beam wavelength,  was  = 1.5406Å (Cu K 1 radiation).The step size (Δ2) was 0.01°and the exposure time at each step was 20 seconds.In XRR, the x-axis was converted from scattering (diffraction) angle, 2 to q z , where q z = (4sin)/.A functioning film stack without the top electrode enables experimental flexibility of how an electric field can be applied to the stack.An atomic force microscopy tip was used as a top electrode to apply an electric field across the film stack in specific areas.This dynamic way of applying an electric field to the film enables induction of the resistive switching mechanism in a designated area.While the uniformity of an electric field between electrodes is dependent on the interface roughness, the uniformity of the field generated during a C-AFM scan is dependent on parameters that determine surface contact, such as scan speed, tip shape and surface roughness.
In order to investigate the effects of charge trapping on the mechanism of resistive switching, selected areas were irradiated with Si ++ using a Raith VELION FIB-SEM at doses of 5, 10 and 50 ions nm −2 with an accelerating voltage of 35 KeV (70 KeV landing energy).KPFM images were taken in ambient conditions using an Oxford Cypher AFM, measuring topography during the trace scan and driving the tip to the null voltage (voltage where the attractive/repulsive force equals zero) between the tip and substrate during the retrace scan.A 15 μm ambient C-AFM scan was done at a range of voltages (+4 to −4 V) across the boundary of each irradiated area in an attempt to switch the film.C-AFM scans were measured across the boundaries of the irradiated areas to probe both the Si irradiated regions and pristine, unexposed, regions of the film for reference.The topography and current maps from the ambient C-AFM scans can be found in Figure S18 (Supporting Information) and the average change in CPD from the applied C-AFM can be seen in Figure S19 (Supporting Information).
To directly probe electrically driven ionic motion in this Ta 2 O 5 /Ta/TiN system, C-AFM measurements were done under high vacuum (10 −8 mbar) followed by ToF-SIMS in the same location.The instrument used (A ToF.SIMS 5 by iONTOF from Munster, Germany), enables measurements of conductive properties of the film stack by C-AFM while subsequently depth profiling through the area of interest with ToF-SIMS without the film leaving the vacuum chamber.This C-AFM to SIMS workflow in vacuum enables the induction of ion motion via the C-AFM and subsequent measurement of the resulting ion distribution with the ToF-SIMS without breaking vacuum.An oxygen sputter cleaning was done prior to all C-AFM biasing at 1 keV for 20 s in order to remove surface contaminants and ensure good contact to the Ta 2 O 5 active layer.Oxygen surface cleaning was critical to generate any C-AFM or local current/voltage measurements.A ToF-SIMS depth profile of a 1 keV oxygen beam sputtering the Ta 2 O 5 /Ta/TiN system

Figure 2 .
Figure 2. ToF-SIMS depth profile showing the Y-Z slice of Cs 2 O + ion counts a) and the sum of the O + ion count, TaO + normalized with the Ta + ion count and TaO 2+ ion count normalized to the Ta + ion count b).A 500 mV electric field was applied to the surface using C-AFM in the region between the dotted lines.Oxygen buildup can be seen at the Ta 2 O 5 /Ta interface.Cartoons of this SIMS data c,d) shows the direction the oxygen ions are pulled to the surface from the induced electric field.

Figure 3 .
Figure 3. Shows YZ depth profiles of the Ta 2 O 5 /Ta/TiN interface for column 1 a) and column 2 b).The ions counts potted include: The O + ion count, TaO + normalized with the Ta + count, TaO 2 + ion count normalized to the Ta + ion count and the Ta 2 O 3 + ion count normalized with the Ta 2 + ion count.The green lines indicate the border of each 10 μm bias area (these YZ slices correspond to column 1 and 2 in Figure S10, Supporting Information).The cartoons below the depth profiles c) and d) show the rough movement of oxygen across the biased regions.The applied voltages are shown above each bias area where the bottom row indicates the voltage applied in the first scan and the top bias indicated the voltage applied during the second scan.(An absence of oxygen can be seen at the bottom of Figure3a) that is reversed in Figure3b).

Figure 4 .
Figure 4.A comparison of the TaO + depth profile in the biased area versus the pristine area a), the resulting topography change b) by AFM and a YZ slice of ion counts of: TaO + , CsO + , Cs 2 O + , CsTaO 2 + across the biased region highlighting oxygen depletion the Ta 2 O 5 layer c) and the oxygen insertion in the Ta layer d).Line scans of the topography image b) can be seen in Figure S15 (Supporting Information).Note that both c) and d) are the same dataset with different scale.