Engineering of Grain Boundaries in CeO2 Enabling Tailorable Resistive Switching Properties

Defect engineering in valence change memories aimed at tuning the concentration and transport of oxygen vacancies are studied extensively, however mostly focusing on contribution from individual extended defects such as single dislocations and grain boundaries. In this work, the impact of engineering large numbers of grain boundaries on resistive switching mechanisms and performances is investigated. Three different grain morphologies, that is, “random network,” “columnar scaffold,” and “island‐like,” are realized in CeO2 thin films. The devices with the three grain morphologies demonstrate vastly different resistive switching behaviors. The best overall resistive switching performance is shown in the devices with “columnar scaffold” morphology, where the vertical grain boundaries extending through the film facilitate the generation of oxygen vacancies as well as their migration under external bias. The observation of both interfacial and filamentary switching modes only in the devices with a “columnar scaffold” morphology further confirms the contribution from grain boundaries. In contrast, the “random network” or “island‐like” structures result in excessive or insufficient oxygen vacancy concentration migration paths. The research provides design guidelines for grain boundary engineering of oxide‐based resistive switching materials to tune the resistive switching performances for memory and neuromorphic computing applications.

It is widely accepted that oxygen vacancies play crucial roles in the switching process of valence change memory via two mechanisms: 1) filamentary switching mechanism and 2) interfacial switching mechanism. [2,[22][23][24] In the filamentary switching, the migration of oxygen vacancies (often perpendicular to the applied electric field) causes conducting filaments formation and rupture at particular local locations. In the interfacial switching, the drift of oxygen vacancies (parallel to the applied electric field) changes its concentration at the oxide/ electrode interfaces which facilitates the interfacial switching by modifying the Schottky barrier height and width. In both cases, the concentration and mobility of oxygen vacancies determine the switching mechanisms and properties. [25] Therefore, defect engineering during synthesis and post-synthesis have been reported to tune the switching performances of the valence change memory.
In recent years, multiple defect engineering methods tuning the properties of oxygen vacancies have been incorporated in the development of valence change memory. [26] For instance, doping foreign metal species in the oxide matrix lowers the formation energy of oxygen vacancies near the doping sites, therefore helps with controlling the generation and concentration of oxygen vacancies. [13,[27][28][29] Embedding a layer of active metals or oxygen-deficient oxides provides a reservoir of oxygen vacancies which greatly reduces the energy for vacancy generation. [9,30] Extended defects such as single dislocations, [31] grain boundaries, [32,33] and phase boundaries [34,35] lower the formation energy of oxygen vacancies and lower the migration barrier along the extended defects, leading to desired switching properties. The localized switching process is possibly facilitated by such defects, as spatially resolved RS in nm scale has been demonstrated using conductive atomic force microscopy in various studies. [36][37][38] Among all the engineering methods, utilizing extended defects has several unparalleled merits such as ultimate scalability, easy control via growth temperature, and simple fabrication. However, existing studies focus mostly on demonstrating either the contribution from single grain dislocation/boundary or performances of a certain grain morphology, [32,33] leaving the discussion on how different grain morphologies impact the RS process still a blank space.
In this paper, three different grain morphologies, "random network," "columnar scaffold," and "island-like," of Cerium oxide (CeO 2 ) thin films were realized by controlling the growth temperature from 300 to 750 °C, to explore the grain boundary designs on tuning the RS properties as illustrated in Figure 1. CeO 2 is selected for this study considering its great potential in applications in catalysis, solid oxide fuel cells, and resistive switching. [39][40][41] The impact of microstructures, namely different grain morphologies, on the electrical properties in CeO 2 was studied for applications in the solid oxide fuel cells and catalysis. [42,43] Features including easy generation of oxygen vacancies, high oxygen mobility, together with the rich background information all make CeO 2 suitable for the study on how different grain morphologies would impact the RS in valence change memory. The current-voltage (I-V) characteristics of the three structures were explored to couple with the different grain boundary structures. The conduction and switching mechanisms responsible for the observed RS properties are discussed. Different switching modes were found in one of the samples during temperature-dependent I-V measurement in high vacuum condition. Such observation of different switching modes is discussed in detail and related to different switching mechanisms. A comprehensive model explaining the contribution of different designs of grain morphology is developed based on interactions between oxygen vacancy migration and grain boundary distribution. The designs of grain morphologies for tailorable RS properties could lead to better control of the RS performance in various oxide thin films for reliable memory devices. Figure 1. Illustrations of the engineering of grain boundaries by changing growth temperatures and the corresponding grain morphologies. The morphologies are named "random network" (left), "columnar scaffold" (middle), and "island-like" (right). The grain boundaries are represented by the brown curves. Lines and curves in yellow illustrate the potential migration paths for oxygen vacancies. The illustrations are designed according to the transmission electron microscopy (TEM) characterization of each sample. TE and BE refers to top electrode and bottom electrode, respectively. www.advelectronicmat.de Figure 2 shows the X-ray diffraction (XRD) θ-2θ scan of the three samples deposited at 300, 500, and 750 °C. The 300 °C sample shows no obvious CeO 2 peaks, indicating that the film is likely to be amorphous or nanocrystalline with very small grain size and randomized orientations. As the deposition temperature increases, the CeO 2 (220) peak appears in the 500 °C sample, suggesting a preferred out-of-plane direction and improved crystallinity compared to the 300 °C sample. In the 750 °C sample, the CeO 2 (002) and (004) peaks with high intensity are visible, suggesting a textured growth along the (00l) direction in the sample. Considering the large lattice mismatch between CeO 2 (a = 5.41 Å) and STO (a = 3.905 Å), the highly textured CeO 2 can form a 45° in-plane rotation with the Nb-doped SrTiO 3 (Nb:STO) substrate. Such in-plane matching relationship between CeO 2 and STO was confirmed via φ-scans in our previous work as 45° peak shifts can be observed between the CeO 2 (220) and STO (110) peaks. [35] To examine the microstructures and the morphologies of all three samples further, cross-section TEM and STEM were performed, as shown in Figure 3. From Figure 3a,d,h, the CeO 2 film deposited at 300 °C consists of high density of grains with very small sizes (≈4 nm). The inset in Figure 3a shows the diffraction patterns from the STO (001) substrates (marked by the yellow diamonds) and two incomplete rings marked by blue dashed lines suggesting the (220) and (004) orientations of the nanograins. The zoom-in TEM image shown in Figure 3d further illustrates the randomized grain orientation and the nanocrystalline nature of the sample in the marked region from Figure 3a. Although 30 nm thick CeO 2 films grown at all three temperatures were anticipated by changing the pulse number during growth according to the corresponding expected growth rates, the actual thickness of the 300 °C sample reaches ≈80 nm, suggesting high porosity and low density of the film.

Structural Characterizations
As the growth temperature increases, the 500 °C film with ≈37 nm in thickness shows much larger grains with vertical grain boundaries along the film growth direction, as shown in Figure 3b,i. The inset in Figure 3b shows similar diffraction patterns from the CeO 2 as two incomplete rings indicating the (220) and (400) orientations are marked by the dashed blue circles. Figure 3e shows the enlarged TEM image from the marked region in Figure 3b and clearly shows the two slightly tilted grain boundaries on each side of the figure and the lattices with identical orientation within the grain in the middle of the figure. According to Figure 3c,g, the 750 °C film shows continuous high-quality growth in the lower half of the film and faceted growth of CeO 2 pillars in the top half of the film. Diffraction patterns shown in the inset in Figure 3c demonstrated two sets of patterns, that is, the pattern of STO diffractions outlined by the yellow diamond shape and the pattern of CeO 2 diffractions marked by the blue diamond shape. It is noted that the two diamond shapes coincide on two end points, which are indexed as STO (200) // CeO 2 (220), and STO (200) // CeO 2 (220), on the left and right sides, respectively. This reveals the in-plane matching relationship between the CeO 2 film and STO substrate with a 45° in-plane rotation of the CeO 2 and its epitaxial growth. The zoom-in image of the tip of one pillar shown in Figure 3f shows the periodic and uniform arrangement of the lattices, indicating the high-quality epitaxial growth of the CeO 2 film at 750 °C.
The faceted growth of CeO 2 is often seen in the high temperature growth (>700 °C) conditions, especially when the film grows thicker and under high oxygen partial pressure. [35,44] Different from the surface morphologies of the 300 and 500 °C samples shown in Figure S1a,b, Supporting Information, the highest energy facet (001) with two orthogonal in-plane directions of the nanorod structure in the 750 °C was further confirmed by AFM, as shown in Figure S1c, Supporting Information, showing good agreement with the AFM results from the previous report on faceted CeO 2 growth. [45] The evolution of film morphology is thus clearly shown, following the structural design of our investigation, that is, the 300, 500, and 750 °C sample results in the different grain boundary morphologies referred to as "Random networks," "Columnar scaffold," and "island-like," respectively.

I-V Characteristics
The I-V characteristics of the devices were measured on multiple devices from each sample. The measurement configuration is shown in Figure 4a. The platinum electrodes were contacted with tungsten probe tips and a positive voltage was applied on the top electrodes. Figure S2a-c, Supporting Information, shows the I-V characteristics of 10 consecutive cycles measured on three different devices on the "random network" sample. Large variations of the I-V characteristics among the three devices and within each consecutive testing cycles show the instability of RS properties of the sample. Figure 4b shows the I-V characteristics of the "columnar scaffold" device. Noticeably, a forming process which does not require higher forming voltage and current is shown in the dark blue curve. Repeatable switching over 10 cycles showed little variation and high ON/OFF ratio (≈10 5 ) at low read voltage (≈0.5 V). The overall low current together with the current rectification are all desirable features for memory applications to prevent the sneak path problems and high energy consumption. [46] I-V curves  Figure 4c show the characteristics of the "island-like" device. 10 cycles measured on two different devices showed the most stable switching with obvious current rectification, negative differential resistance (NDR) on the negative voltage range.
Compared to the "columnar scaffold" devices, the "island-like" devices require a larger set and reset voltage (6 V) but have much lower conductance over the measured voltage range. Notably, both devices showed an interesting trait of current dip The small grain size results in randomly distributed grain boundaries. The structure of the film is therefore described as "random network." b,e) TEM and i) STEM images of the CeO 2 film deposited at 500 °C. e) the zoom-in image of the marked region in (b). The vertical grains and boundaries of this film resemble the structure of scaffolds. The film is then referred to as "columnar scaffold." c,f) TEM and g) STEM images of the CeO 2 films deposited at 750 °C. f) the zoom-in image of the marked region in (c). Faceted pillars with clear separation from each other are observed. The name "island-like" is therefore used to describe such nature of the film. Insets in (a-c) show the diffraction patterns of each selected area, the yellow diamonds mark the diffraction spots from STO substrates. Incomplete rings located within the diamond in the insets of (a) and (b) are from the polycrystalline CeO 2 film. The blue diamond in (c) shows in-plane matching between CeO 2 (220) and STO (200), suggesting a 45° in-plane rotation.
in the HRS during the negative voltage sweeping steps, which is also observed in various metal/oxide/Nb:STO systems. [47,48] Such a trait indicates the non-zero crossing of the I-V hysteresis loop, potentially due to the capacitive-coupled memristive effect as described in a previous report by Sun et al. [49] To further investigate the RS properties of the "columnar scaffold" and the "island-like" devices, endurance measurements with the configuration of read and write pulses shown in Figure 4c were performed. With 1 ms pulse width, the "columnar scaffold" devices demonstrated stable switching with large ON/OFF ratio (≈10 3 ) over 100 cycles while the "island-like" devices showed no distinguishable states throughout the measurement. The slightly inconsistent resistance of the HRS in the "columnar scaffold" device compared to its direct current (DC) I-V characteristic at 0.55 V read voltage potentially originates from the different testing protocols as the pulsed measurements are much faster than the DC sweeping. To confirm the non-volatility of the "columnar scaffold" device, a retention measurement with 0.55 V read voltage was carried out for 1000 s, as shown in Figure S3, Supporting Information. The two resistance states are well retained during the entire measurement, ensuring its functionality.

Discussion
First, all the samples showed RS effects in the pristine devices and the samples with different grain morphologies demonstrated vastly different I-V characteristics. A detailed analysis on the conduction mechanisms and the switching mechanisms is needed to unveil the origin of the I-V characteristic evolution.

Conduction and Switching Mechanism Analysis
The switching and conduction mechanisms of the most stable samples are therefore worth investigation. The "columnar scaffold" devices, due to their stable I-V characteristic, and several features (such as low switching voltage and fast response) suitable for memory applications, is worth the most attention. The devices are made out of three parts: Pt top electrodes, CeO 2 switching layer, and the Nb:STO substrate. To gain a basic understanding of the device performance, the electronic energy band diagram is carefully analyzed. The top electrode Pt is a high work function (Φ Pt ≈ 5.65 eV) metal. [50] CeO 2 , known as a wide bandgap (E g ≈ 3.2 eV) n-type semiconductor, has an electron affinity 3.5 eV CeO2 χ ≈ , [51,52] and a Fermi level E F located 1.2 eV below the conduction band minimum (CBM), [53] while 0.7 wt% Nb-doped STO is a degenerate n-type semiconductor with a high carrier concentration ≈10 20 cm −3 and an electron affinity χ Nb: STO ≈ 3.9 eV. [54][55][56] Consequently, as shown in Figure 5a, two barriers are formed at the Pt/CeO 2 and CeO 2 /Nb:STO interfaces with the respective nominal Schottky barrier heights φ B1 ≈ 2.15 eV and φ B2 ≈ 0.4 eV. In this case, three parts of the device can affect to the conduction and the RS process: the Pt/CeO 2 interface, the bulk CeO 2 , and the CeO 2 /Nb:STO interface. To deconvolute the contributions . I-V characteristics of the devices with "columnar scaffold," and the "island-like" morphologies. a) Measurement configuration. b) I-V characteristics of 10 consecutive cycles of the "columnar scaffold" device, c) I-V characteristics of 10 consecutive cycles of the "island-like" device. d) Testing protocol used in the endurance measurement, each of the switching pulse is followed by a read pulse with lower voltage. Endurance measurement of 100 cycles with 1 ms pulse width for e) the "columnar scaffold" and f) the "island-like" devices.
www.advelectronicmat.de from all three parts of the device, a closer look at the I-V characteristic is needed. As shown in Figure S4, Supporting Information, the Pt/CeO 2 is forward-biased when a positive voltage is applied to the Pt electrode and it is reverse-biased when a negative voltage is applied to the Pt. This is why the positive current is much higher than that in the negative voltage range. As a result, we conclude that the Pt/CeO 2 interface dominates the I-V characteristics. This conclusion is in line with several reports on metal/CeO 2 /Nb:STO systems. Foglietti et al. demonstrated the RS of metal/CeO 2−x /Nb:STO with different top electrodes, including Pt (Φ ≈ 5.5 eV), TiN (Φ ≈ 4.7 eV), and Ti (Φ ≈ 4.33 eV). [57] Similar rectifying I-V characteristics are found in the devices with Pt and TiN top electrodes, whereas the Ti/CeO 2 /Nb:STO device showed ohmic conduction, indicating ohmic contact at the Ti/CeO 2 interface. [57] Liao et al. tested the RS properties of various metal/CeO 2 /Nb:STO systems and found rectifying I-V characteristics in samples with large work function metals like Pt, Ag (Φ ≈ 4.7 eV), and Au (Φ ≈ 5.2 eV), and ohmic conduction in the pristine state devices with low work function metals such as Ti, Ta (Φ ≈ 4.2 eV), and Al (Φ ≈ 4.3 eV). [58] Both reports support our claim that the devices form Schottky barrier at the Pt/CeO 2 due to the high work function of Pt, which results in the rectifying I-V characteristics. Figure 5b shows switching characteristic and the switching sequence of the "counterclockwise" device. This switching process shows a typical counterclockwise -clockwise (CC-C) type sequence, which has SET in the positive bias and RESET in the negative bias. [59] As we demonstrate, this switching sequence is dominated by Schottky barrier modulation at the Pt/CeO 2 interface and bulk-limited conduction, and the Schottky barrier modulation can be related to charged oxygen vacancies and trapped electrons.
To verify the conduction mechanisms of the devices, the positive branch of the I-V curves (forward-biased) were fitted by various models including thermionic emission, Poole-Frenkel emission ( Figure S5a, Supporting Information), and space-charge-limited conduction (SCLC) (Figure S4b, Supporting Information). The positive sides of the I-V curves for the "columnar scaffold" were fitted by a forward-biased thermionic emission model, as shown in Figure 5c. The thermionic emission model corresponding to the Pt/CeO 2 interface in the forward-bias condition, is described in Equation (1). [60] exp e xp Here, φ B is the Schottky barrier height (SBH), V is the applied voltage, k is the Boltzmann constant, the constant term qm k h π = * * is known as the Richardson's constant, where q is the unit charge, m * the effective electron mass, and h is the Planck constant. An effective electron mass m * ≈ 0.5m 0 in CeO 2 is adapted according to a previous study by Kim et al. [53] The Schottky barrier heights of the low-resistance state (LRS) and high-resistance state (HRS), shown in Table 1, are extracted by deriving the interception of the fitting lines with the y-axis in the ln( J )-V plot. Fittings of I-V characteristics of the "island-like" device with mentioned conduction models are shown in Figure S6a-d, Supporting Information, with extracted Schottky barrier heights from the forward-biased thermionic emission model shown in table S1, Supporting Information. The "island-like" device showed similar fitting results as the "columnar scaffold" device, indicating similarities in the overall conduction mechanism and switching mechanism.
The Schottky barrier height and width can be modulated by the oxygen vacancies in CeO 2 , as they are considered donor dopants and provide additional carriers. [61,62] The modulation of the Schottky barrier can be realized in two ways: One is via the generation and migration of oxygen vacancies and the resulting accumulation and depletion of the oxygen vacancies at the oxide/metal interface, the other one is via charge trapping/de-trapping, where defects (e.g., grain boundaries in top electrodes, the structural defects on the film surface) act as charge traps. The Schottky barrier height and width are reduced when the oxygen vacancies/electrons are accumulated/de-trapped and are increased when the oxygen vacancies/ electrons are depleted/trapped. [63,64] Despite multiple investigations on metal/CeO 2 /Nb:STO systems, the discussions on the potential switching mechanisms are inconsistent. Gao et al. observed structural changes from CeO 2 to Ce 2 O 3 due to oxygen vacancy generation and migration during in situ biasing TEM measurements. [39] Zhang et al. reported the RS phenomenon of Au/epi-CeO 2 /Nb:STO and proposed oxygen vacancy migration and charge trapping/de-trapping at the CeO 2 /Nb:STO interface as the switching mechanism. [63] This is unlikely to be the dominant switching mechanism due to the ohmic contact at the CeO 2 /Nb:STO interface. Foglietti et al. proposed that conducting channel formation at the metal/CeO 2 interface modulates the effective Schottky barrier height and results in the RS in Pt/CeO 2 /Nb:STO and TiN/CeO 2 /Nb:STO systems as they verified the ohmic conduction at the CeO 2 /Nb:STO interface. [ at the Pt/oxide interface in a Pt/SrTiO 3 /Nb:STO cell, which provides strong evidence of an identical switching mechanism in the case of Pt/CeO 2 /Nb:STO. [16] Further evidence indicates that moisture also plays a role in the switching process as all three samples showed consistent changes of I-V characteristics when measured in dry and humid Ar flow (data not shown here). It was also reported that charge trapping and de-trapping mechanism is related to protons/moisture. [65] Therefore, it is possible that multiple sources contribute to the overall switching process.

Observation of Filamentary and Interfacial Switching Modes
To further investigate the RS properties of the "columnar scaffold" and "island-like" devices, temperature-dependent measurements were carried out in a vacuum chamber. Contrary to the original expectation for I-V characteristics with the same switching mode as that measured in atmosphere and room temperature, completely different switching mechanism was found in the high vacuum and temperature-dependent measurements. Figure 6a shows the temperature-dependent measurement results for the "columnar scaffold" device at six different temperatures (120 to 300 K) under high vacuum condition (10 −7 Torr). The device was written back to the HRS after the measurements in atmospheric and room temperature conditions. A forming process shown by the black curve was observed during the first cycle measured at 120 K. Different from the common forming process, the forming process observed in this case (indicated by arrow No.1 in Figure 6a) only set the device in the HRS. Interestingly, the switching polarity for vacuum measurement (clockwise-counterclockwise or C-CC) is the opposite to the atmospheric results (CC-C), indicating a different switching mechanism. A further SET process happened when negative bias was applied to the top electrode, as indicated by arrow No. 3. The RESET or the erase process was observed when a positive bias was applied on the Pt electrode. Such a switching sequence has been reported in various material systems including TaO x , [66] and TiO x . [67] The possible reason for the positive RESET voltage is closely related to the oxygen vacancy activity and redox reactions at the Pt/CeO 2 interface. The abrupt current increase during the SET process suggests filamentary switching in the device, where the conducting filaments are formed due to the migration of oxygen vacancies under external bias. [7,19] During the RESET or the erase process a gradual current decrease originates from the dissolving of the conducting filaments due to the migration of oxygen vacancies. [68,69] Both the SET and RESET processes observed are evidence for filamentary switching, which is a completely different switching mechanism from the atmospheric I-V characteristics. The following cycles tested at different temperatures from 120 to 300 K all showed stable switching with only slight variation in the ON/OFF ratio, showing the validity of the filamentary switching in the device. Figure 6b shows the interesting phenomena where two completely different types of switching were found in the same device under different measurement conditions. It should be noted that such a change of switching behavior is irreversible. Such observation of two switching modes in a single device has been reported in various systems and is related to the variation of local concentration of oxygen vacancies and defects and power dissipation. [59,70,71] The I-V curve of the filamentary switching in the positive and negative bias ranges are best fitted using the SCLC model. In Figure 6c, the RESET process in the filamentary switching mode can be fitted linearly in several sections. The red and green sections in the LRS and the blue section in HRS have a slope ≈1, suggesting Ohmic conduction in the LRS and low field region in the HRS. The higher field regime in the HRS has a slope ≈2, indicating trap-filled SCLC region. [2,72] Figure 6d shows the SET process where the LRS has a slope of ≈1 indicating that Ohmic conduction and the transition between HRS and LRS is realized by a direct current jump. This conduction mechanism and transitions in the SET and RESET process have been reported in many filamentary switching systems, and are considered the fingerprints of filamentary switching. [73] Different from the "columnar scaffold" device, temperature-dependent I-V characterization of the "island-like" devices showed no filamentary switching at all. As shown in Figure S7a-d, Supporting Information, the I-V curves measured under high vacuum and at various temperatures showed diode-like behavior with low overall current, indicating the low defect concentration and insulating nature of the film. As the measurement temperature increases, the overall I-V characteristics show a slight increase in the current, suggesting the increasing contribution from the carrier injection and the interfacial Schottky barrier. The absence of filamentary switching in the "island-like" devices can be related to the higher oxygen vacancy formation energy and the resulting lower oxygen vacancy concentration and lower oxygen vacancy mobility due to the lower grain boundary (as fast-diffusion path) density compared with the "columnar scaffold" device. [74]

Structure-Property Relation and a Comprehensive Model for the Observation of Filamentary and Interfacial Switching Modes
To unveil the detailed mechanisms behind the observation of two switching modes, a comprehensive model is developed, as shown in Figure 7. Since the oxygen vacancy concentration and distribution determines the switching mode, it is necessary to examine the source of oxygen vacancies in the device under different measurement conditions. The oxygen vacancies in the switching layer can come from two different sources: intrinsic and extrinsic. Intrinsic oxygen vacancies are formed during the growth of CeO 2 because of the thermal equilibrium of the Frenkel defect. In addition, extended defects like grain boundaries in the CeO 2 film can also introduce more oxygen vacancies at the grain boundary sites to mitigate the structural distortions and lower the energy of the grain boundaries. [74] Extrinsic oxygen vacancies refer to those that are generated during the electrical characterization via the reaction O Oxygen gas is generated when positive bias is applied to the top electrode, leaving positively charged oxygen vacancies and electrons in the CeO 2 switching layer. In the pristine Pt/ CeO 2 /Nb:STO device, the oxygen vacancies and oxygen ions are intrinsic defects formed during the growth of CeO 2 . The existence of grain boundaries also results in the increase of the oxygen vacancies concentration to relax the structural distortion. [74] The concentration of oxygen vacancies is therefore higher near the interface and the grain boundaries, as shown  Figure 7c shows the RESET process where the oxygen vacancies are attracted to the top electrode (Pt) and combined with the oxygen ions via the reverse reaction, leading to the depletion of oxygen vacancies near the Pt/CeO 2 interface and the restoration of the Schottky barrier height. In contrast to the atmospheric measurement, vacuum condition has much lower oxygen partial pressure in the environment, which facilitates the reaction towards the generation of oxygen vacancies and oxygen molecules. In this case, the forming process during vacuum introduces high concentrations of oxygen vacancies and these oxygen vacancies migrate along the grain boundaries and form conducting filaments. As shown in Figure 7d, partially formed filaments are present in the switching layer as the forming process only leads to the HRS in the filamentary switching mode. Figure 7e shows the SET process when negative bias is applied to the Pt top electrode. During the SET process, the already constructed partial filaments as virtual cathodes stay intact because of the neutral charge state of the oxygen vacancies in the filaments. [75] Consequently, the charged oxygen vacancies drift toward the top electrode and form complete filaments along the grain boundaries, resulting in the LRS. In the RESET process, shown in Figure 7f, the conducting filaments are ruptured due to Joule heating and the oxygen vacancies migrate away from the filaments, leaving the device in the HRS.
Above all, the three samples showed vastly different RS properties and the grain morphologies impact the switching process. First, the "random network" devices demonstrate unstable switching and the lowest resistance in both LRS and HRS among all three samples. From the structural perspective, the polycrystalline structure of the "random network" film deposited at 300 °C yields high concentrations of oxygen vacancies, making the film more conductive. The high density of grain boundaries and their random distribution makes the migration of oxygen vacancies less controlled and therefore result in an unrepeatable switching. Second, the "island-like" devices showed the most stable I-V characteristics and highest resistance during DC measurement. The overall highest resistivity in the "island-like" devices deposited at 750 °C is in line with previous reports due to low oxygen vacancy concentrations and fewer carriers. [76] Despite its desirable features in the I-V characteristics, such as 1) stable switching, 2) low current in both HRS and LRS for low energy consumption and 3) large current rectification, the slow response during pulse measurement makes it non-ideal for memory or neuromorphic computing applications. With much fewer grain boundaries, as the high growth temperature enables energetically favored grain growth, the "island-like" devices have less oxygen vacancies and less oxygen vacancy transport paths (grain boundaries) compared with the lower temperature grown devices. The RS process, relying heavily on the transport of oxygen vacancies, is therefore inhibited by two factors: 1) the low concentration of oxygen vacancies, 2) the slow oxygen diffusivity in the bulk due to the Figure 7. Schematics of the different switching mechanisms in the "columnar scaffold" device. a) Pristine device. b) SET process of the device measured in atmosphere. c) RESET process in atmosphere. The abundance of oxygen in atmosphere is indicated by the clusters of two adjacent red circles. d) The forming process during measurement in vacuum (10 −7 Torr), the oxygen ions inside the film are attracted to the Pt electrode and turned into oxygen molecules leaving high concentration of oxygen vacancies in the CeO 2 switching layer. e) SET process of the device measured in vacuum, conducting filaments are formed. f) RESET process in vacuum, the conducting filaments are ruptured. Green circles represent positively charged oxygen vacancies, red circles represent negatively charged oxygen ions. Pairs of red circles represent oxygen molecules.
www.advelectronicmat.de absence of grain boundaries. [77] Last, among all three samples, "columnar scaffold" devices deposited at 500 °C demonstrated the best overall performances including fast response, large ON/OFF ratio, low write (2.5 V) and read (≈0.5 V) voltages, and stable switching. As discussed previously in the switching mechanism section, the vertical grain boundaries extending through the film serve as transport paths for oxygen vacancies that lead to the fast switching response, and the ideal concentration of oxygen vacancies contributes to the switching process, rendering the "columnar scaffold" devices promising for memory and computing applications.

Conclusion
In summary, we have investigated the RS properties of different designs of grain morphologies in CeO 2 films including "random network" (deposited at 300 °C), "columnar scaffold" (deposited at 500 °C), and "island-like" (deposited at 750 °C). Among the three samples prepared, the sample with "columnar scaffold" vertical grain boundaries demonstrated the best resistive switching properties, which can be attributed to an ideal concentration of oxygen vacancies and fast transport along grain boundaries through the film thickness. The structuralelectrical correlation is verified experimentally by XRD, TEM, and electrical measurements. The engineering of grain boundaries, realized simply by varying the growth temperature, can be applied to other thin film oxide RRAM systems, paving a way for the development of state-of-the-art tailorable highperformance memristor systems.

Experimental Section
Pulsed Laser Deposition: Fluence F = 3 J cm −2 , growth time t = 12 min for all samples, frequency f = 5 Hz, oxygen pressure p = 50 mTorr, growth temperatures for all samples specified in the context. After deposition, the heater was cooled down to 30 °C at 10 °C min −1 in 200 Torr oxygen, but without continuous flow.
Structural Characterizations: X-ray diffraction θ-2θ measurements were performed on all samples using Empyrean from Malvern Panalytical. Surface morphologies of the samples were examined using atomic force microscope (Dimension Icon) from Bruker. Samples for transmission electron microscopy were prepared manually by pasting, grinding with 600 grits sand papers and silica diamond papers in the sequence of 15, 6, 3, and 1 µm. The sample was then dimpled and underwent an ion-milling process using a PIPS II precision ion polishing system. Transmission electron microscopy (TEM), scanning transmission electron microscopy (STEM), and energy dispersive X-ray spectroscopy (EDX) were performed in a Talos F200X G2 TEM with a gun brightness of 200 kV.
Electrical Characterizations: Devices using circular Pt electrodes with ≈100 nm thickness and ≈200 µm diameter were fabricated via sputtering with shadow masks at room temperature. Current-voltage characteristics of all devices were measured in a 1) manual probe station at room temperature and 2) in a XYZ low-temperature probe station at temperatures ranging from 120 to 300 K. For both probe stations, a Keysight B2912A source/measure unit was used.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.