PDF Nanoscale Topotactic Phase Transformation in SrFeOx Epitaxial ... - NJU

www.advmat.de

Nanoscale Topotactic Phase Transformation in SrFeO _x Epitaxial Thin Films for High-Density Resistive

Switching Memory

Junjiang Tian, Haijun Wu, Zhen Fan,* Yang Zhang, Stephen J. Pennycook,*

Dongfeng Zheng, Zhengwei Tan, Haizhong Guo, Pu Yu, Xubing Lu, Guofu Zhou, Xingsen Gao, and Jun-Ming Liu

J. Tian, Prof. Z. Fan, D. Zheng, Z. Tan, Prof. X. Lu, Prof. X. Gao, Prof. J.-M. Liu

Institute for Advanced Materials

South China Academy of Advanced Optoelectronics South China Normal University

Guangzhou 510006, China E-mail: [email protected] J. Tian, Prof. Z. Fan, Prof. G. Zhou

Guangdong Provincial Key Laboratory of Optical Information Materials and Technology

South China Academy of Advanced Optoelectronics South China Normal University

Guangzhou 510006, China

Dr. H. Wu, Dr. Y. Zhang, Prof. S. J. Pennycook Department of Materials Science and Engineering National University of Singapore

Singapore 117575, Singapore E-mail: [email protected]

Prof. H. Guo

School of Physical Engineering Zhengzhou University Zhengzhou 450001, China Prof. P. Yu

State Key Laboratory of Low Dimensional Quantum Physics and Department of Physics

Tsinghua University Beijing 100084, China Prof. G. Zhou

National Center for International Research on Green Optoelectronics South China Normal University

Guangzhou 510006, China Prof. J.-M. Liu

Laboratory of Solid State Microstructures and Innovation Center of Advanced Microstructures

Nanjing University Nanjing 210093, China

DOI: 10.1002/adma.201903679

Oxygen stoichiometry plays a crucial role in determining the crystalline structure and physical properties of transition metal oxides (TMOs). Tuning the oxygen content via an electrochemical redox reaction can effectively manipulate the functionalities of TMOs, which is harnessed in many cutting-edge energy and information tech- nologies such as fuel cells,^[1] rechargeable batteries,^[2] supercapacitors,^[3] and memory devices.^[4] The redox reaction in certain TMOs can be enabled by a so-called topo- tactic phase transformation, manifesting itself as the insertion/release of a large amount of oxygen ions without breaking the lattice framework. For example, for an ABO₃ perovskite (PV) phase, upon forming ordered oxygen vacancy channels in its lattice, it transforms into an ABO_2.5 brownmillerite (BM) phase. Along with the structural change, many intriguing physical phenomena emerge owing to the couplings between lattice, charge, and Resistive switching (RS) memory has stayed at the forefront of next-generation

nonvolatile memory technologies. Recently, a novel class of transition metal oxides (TMOs), which exhibit reversible topotactic phase transformation between insulating brownmillerite (BM) phase and conducting perovskite (PV) phase, has emerged as promising candidate materials for RS memories.

Nevertheless, the microscopic mechanism of RS in these TMOs is still unclear.

Furthermore, RS devices with simultaneously high density and superior memory performance are yet to be reported. Here, using SrFeOx as a model system, it is directly observed that PV SrFeO₃ nanofilaments are formed and extend almost through the BM SrFeO2.5 matrix in the ON state and are rup- tured in the OFF state, unambiguously revealing a filamentary RS mechanism.

The nanofilaments are ≈10 nm in diameter, enabling to downscale Au/

SrFeO_x/SrRuO₃ RS devices to the 100 nm range for the first time. These nano- devices exhibit good performance including ON/OFF ratio as high as ≈10⁴, retention time over 10⁵ s, and endurance up to 10⁷ cycles. This study signifi- cantly advances the understanding of the RS mechanism in TMOs exhibiting topotactic phase transformation, and it also demonstrates the potential of these materials for use in high-density RS memories.

The ORCID identification number(s) for the author(s) of this article can be found under https://doi.org/10.1002/adma.201903679.

spin degrees of freedom.^[5–7] As a result, the PV-BM topotactic phase transformation has sparked a surge of research interest in recent years.

Strontium ferrite, SrFeOx (SFO), is a prominent model system for studying the topotactic phase transformation.

SrFeO₃ exhibits a cubic PV structure (PV-SFO) with a lattice constant a_C of ≈3.85 Å^[8] (lower panel in Figure 1a); whereas, the oxygen-deficient SrFeO_2.5 has a BM structure (BM-SFO), in which the octahedral FeO₆ and tetrahedral FeO₄ layers are alternately stacked along the b-axis (upper panel in Figure 1a).

The BM structure is represented by an orthorhombic unit cell with lattice parameters: a_O= ≈5.67 Å, bO= ≈15.58 Å, and c_O= ≈5.53 Å.^[9] Besides the crystalline structure, the electronic and magnetic properties of PV-SFO and BM-SFO are also distinctly different. For example, for PV-SFO, the nominal Fe⁴⁺ adopts heavily mixed 3d⁴ and 3d⁵L (L denotes a ligand hole) electronic configurations, and the ground state is a heli- magnetic metal.^[10,11] On the contrary, BM-SFO exhibits an Fe³⁺ (3d⁵) valence state and an antiferromagnetic insulating ground state.^[8] Therefore, the topotactic phase transforma- tion between PV-SFO and BM-SFO enables a dramatic tuning of transport (both electronic and ionic), magnetic, and optical properties.^[12–14]

More interestingly, PV-SFO and BM-SFO have a rather small Gibbs free energy difference (on the order of 100 meV).^[14] This allows the reversible topotactic phase transformation between

these two phases to be easily accessed. As reported previ- ously, the topotactic phase transformation could be induced by thermal annealing in oxidizing/reducing atmospheres^[14,15]

and electrochemical reactions in liquid electrolytes.^[16–18] More- over, even at ambient temperature and in the absence of liquid electrolytes, a mild electrical stimulus (several volts) is able to trigger the local topotactic phase transformation in SFO (and its analog, SrCoOx) thin films.^[19–22] This ability is particularly useful for all-solid-state resistive switching (RS) memory appli- cation, considering that the PV and BM phases exhibit very dif- ferent electrical conductivities. Furthermore, SFO has a high oxygen mobility due to the naturally formed oxygen vacancy channels,^[23,24] which is an outstanding advantage over other TMOs for RS memory application. Therefore, the SFO-based RS memories have attracted increasing attention, and supe- rior performance including low forming voltage, high ON/OFF ratio, long data retention time, narrow distribution of switching parameters, and good device-to-device uniformity has been achieved recently.^[19,20]

However, so far no direct observation of conductive fila- ments consisting of PV-SFO has been available, making the filamentary mechanism of RS (i.e., electric field-induced local topotactic phase transformation between PV-SFO and BM-SFO) still a conjecture. This also obscures the ultimate scalability of SFO-based RS memories since the sizes of the conduc- tive filaments are unknown. In addition, the downscaling of Figure 1. Crystalline structure and morphology of the SFO film. a) Schematics of lattice structures BM-SFO (upper) and PV-SFO (lower). The red dotted circles in upper panel indicate the oxygen vacancy channels in BM-SFO which are parallel to the [001]_O direction. Although BM-SFO contains oxygen vacancies while PV-SFO does not, BM-SFO is still insulating while PV-SFO is conductive because of their different electronic structures, as described in Introduction section. b) XRD θ-2θ pattern of the SFO epitaxial thin film on the SRO-buffered STO (001) substrate. c) RSM around the (103) reflection of STO along with the (1121) peak of BM-SFO. d) RSM around the (1101) peak of BM-SFO. e) AFM topography image of the SFO film. Inset shows the topography image taken after the deposition of Au nanoelectrodes.

SFO-based RS cells to the nanometer range has never been attempted experimentally, leaving it a mystery as to whether the SFO-based RS memories are potentially useful for high-density memory application.

To address the above gaps, herein, we conduct a detailed study on phase transformation and RS behavior in SFO epi- taxial thin films. We achieve a clear visualization of the PV-SFO nanofilaments (average diameter: ≈10 nm) embedded into the BM-SFO matrix in the ON-state device, thus providing direct evidence for the filamentary RS mechanism. Furthermore, we develop Au/SFO/SrRuO₃ (SRO) nanocapacitor devices (lateral size: ≈180 nm), which exhibit excellent RS performance. Our findings show significance in understanding the RS mecha- nism in TMOs with topotactic phase transformation, and also demonstrate the great potential of these materials for use in high-density RS memories.

Figure 1b shows the X-ray diffraction (XRD) θ-2θ pattern of a representative SFO/SRO bilayer film grown on a (001)-oriented SrTiO₃ (STO) substrate. Half-order reflection peaks, such as (020), (060), and (0100) peaks from the SFO film, are observed.

These peaks stem from the alternate stacking of oxygen polyhe- dral layers,^[24,25] which is a fingerprint of the BM phase. There- fore, the phase of BM-SFO is confirmed and its out-of-plane direction is the b-axis. In addition, the (040) and (080) peaks from BM-SFO overlap with the (001) and (002) peaks from SRO, respectively. No peaks from any secondary phases are observed, confirming the phase purity of the BM-SFO film. Using Bragg’s law, the out-of-plane lattice constant of BM-SFO is calculated as b_O = ≈15.83 Å, agreeing well with the reported values.^[19]

Reciprocal space mapping (RSM) around the asymmetric (103) reflection of STO is presented in Figure 1c. As clearly seen, the (1121) peak of BM-SFO overlaps with the (103) peak of SRO, and both of them have the same H-coordinate as the (103) peak of STO. This suggests that the BM-SFO and SRO layers were grown coherently on the STO substrate with negligible in-plane lattice relaxation. To avoid the overlap with the reflections from SRO, RSM around the (1101) peak of BM-SFO was also taken, as shown in Figure 1d. This allows a more accurate calculation of the in-plane lattice constant of BM-SFO: a_PC = ≈3.91 Å (note:

subscript “PC” denotes pseudocubic, and the corresponding in-plane lattice parameters for the orthorhombic unit cell are a_O ≈ cO ≈ 2^1/2a_PC = ≈5.52 Å). The values of aO and c_O are smaller than the bulk values,^[9] indicating that the BM-SFO film is highly strained due to the epitaxial growth.

Figure 1e displays the surface topography of the BM-SFO film characterized by atomic force microscopy (AFM). The BM-SFO film surface is quite smooth with a small roughness of ≈375 pm. The inset shows the morphology of the Au nano- electrode arrays grown on the BM-SFO film, which will be used for the nanoscale electrical measurements. The triangular Au nanoelectrodes are well-ordered and uniform in size with a lat- eral length of ≈180 nm.

RS characteristics were measured for the BM-SFO film sandwiched between the SRO bottom electrode and conven- tional large-size Pt top electrodes (≈200 µm in diameter). The applied voltage is defined to be positive if a positive bias is applied to the Pt top electrode, and the sequence of applied voltage is 0 → −7 V → 0 for the electroforming and −4 V → 0 → +3 V → 0 → −4 V for the set/reset. Figure 2 presents the

typical current–voltage (I–V) characteristics of the Pt/SFO/SRO devices. In the pristine state, the device exhibits very low cur- rent of < 10⁻⁸ A at a voltage below 1 V, indicating that the as- grown BM-SFO film is relatively insulating. During the first voltage scan (0 → −7 V), the current suddenly jumps when the voltage reaches −4.8 V, pointing to an electroforming process (i.e., the formation of conductive filaments).^[26] After the elec- troforming, the device stays in a low resistance state (LRS) as the voltage reduces from −7 V to 0, and this LRS remains stable during the subsequent voltage scan from −4 V to 0. However, when a positive voltage scan (0 → +4 V) is applied, the device is reset to a high resistance state (HRS) at a voltage (V_reset) of approximately +1.8 V. When the voltage is swept back to the negative region, the transition from HRS to LRS occurs again at a voltage (V_set) of approximately −2.7 V. Therefore, bipolar RS behavior is achieved in the Pt/SFO/SRO devices via nega- tive electroforming. Positive electroforming was also attempted, but a much larger forming voltage was needed (Figure S1, Sup- porting Information). Additionally, the device after the positive electroforming always stays in the LRS and cannot be turned back into the HRS. Hence, all the RS characteristics reported hereafter are obtained via negative electroforming.

In the cyclic voltage sweeps, similar hysteretic I–V charac- teristics can be observed repeatedly with narrow distributions of V_set and V_reset (Figure S2a, Supporting Information), demon- strating good switching stability. The resistances in HRS and LRS (R_HRS and R_LRS) measured for over 20 different devices show small fluctuation (Figure S2b, Supporting Informa- tion), reflecting that our Pt/SFO/SRO devices also have good reproducibility and device-to-device uniformity. (Note: the read voltage is always −1 V hereafter unless otherwise specified.) In the retention test, both R_HRS and R_LRS exhibit negligible relaxa- tion after read voltage stressing for ≈3000 s (Figure S2c, Sup- porting Information). To measure the switching speed and endurance, voltage pulses were applied. Using write pulses of

−4 (or +3 V), the switching from HRS to LRS (or vice versa) starts at a pulse width of 10 µs, and becomes almost saturated at 1 ms (Figure S2d, Supporting Information). After switching Figure 2. RS properties of large-size Pt/SFO/SRO devices. Typical I–V char- acteristics showing bipolar RS behavior with an electroforming process.

the device with write pulses of −4 V/1 ms and +3 V/1 ms for 10⁴ cycles, R_HRS remains relatively stable while R_LRS increases sig- nificantly, leading to a decrease of the ON/OFF ratio from ≈80 to ≈3 (Figure S2e, Supporting Information).

Note that the RS performance of our large-size Pt/SFO/SRO devices, albeit not purposely optimized, is already comparable to that obtained in previous reports.^[19–22] Whether the RS per- formance could be further enhanced in nanosized devices is of interest. Prior to investigating it, we focus on unravelling the microscopic origin of RS, to be presented below.

The observed bipolar RS behavior initiated with an elec- troforming process (Figure 2) already hints that conductive filaments may play a role in RS. Moreover, the electroforming with a higher compliance current results in a lower R_LRS, which may be associated with the increase in total cross-sec- tional area of filaments with increasing compliance current (Figure S3, Supporting Information). This filamentary mecha- nism can be further backed up by electrode area- and temper- ature-dependent resistance measurements in both HRS and LRS. Figure S4 in the Supporting Information shows that both R_HRS and R_LRS decrease with increasing electrode area, but the decreasing rate of R_LRS is much slower than that of R_HRS. This implies that the conduction in LRS may rely on the conduc- tive filaments, and multiple filaments may exist.^[22] Besides, R_LRS has a much weaker temperature dependence than R_HRS, suggesting that the major conductive medium in LRS may be PV-SFO while that in HRS is BM-SFO (Figure S5, Supporting Information).^[19,21] These results provide indirect evidence for a filamentary model comprising conductive filaments of PV-SFO and an insulating matrix of BM-SFO. On the other hand, the I–V curves in LRS are asymmetric with the negative bias cor- responding to the forward direction (Figure 2), implying that a Schottky barrier is “hidden” at the SFO/SRO interface even after the filament formation.

To directly reveal the microscopic nature of the conductive filaments and the “hidden” Schottky barrier, we conducted aberration-corrected scanning transmission electron micro- scopy (STEM) for the SFO films in different states (pris- tine, electroformed, and post-reset).^[27] Note that R_LRS can be partially retained after several days (Figure S6, Supporting Information) or in vacuum (Figure S7, Supporting Informa- tion), offering the possibility for the STEM observation of the filaments. It is seen from Figure S8, Supporting Information that the Pt/SFO/SRO heterostructure exhibits sharp interfaces without inter-diffusion between different layers. In the pris- tine state, most regions of the SFO film exhibit the character- istic Z-contrast feature of superlattice stripes running along the in-plane direction (Figure 3a–c), indicating that SFO pos- sesses a BM structure with alternate stacking of the oxygen polyhedral layers. Moreover, the orientation of these stripes suggests that the b-axis of BM-SFO is along the out-of-plane direction. These observations are consistent with the XRD results (Figure 1b). Additionally, in a small fraction of regions, BM-SFO is observed with the b-axis oriented along the in- plane direction (Figure S9, Supporting Information). The appearance of such a small amount of mis-aligned BM-SFO is probably due to the balance between thermodynamic phase stability and epitaxial strain when growing the SFO film on the STO substrate.^[24]

After the electroforming, the formation of PV-SFO conductive filaments is expected. The STEM high-angle annular dark-field (HAADF) image in Figure 3d presents multiple nanofilaments (average diameter: ≈10 nm) containing no stripes, which are intercalated into the BM-SFO matrix. The crystal structure of the nanofilaments can be further identified by taking a closer look at the interface between a nanofilament and the matrix. As shown in Figure 3e–g, the nanofilament exhibits a non-layered PV-SFO structure with a uniform out-of-plane lattice spacing of ≈3.91 Å.

In the matrix, however, two alternate layers with different out- of-plane lattice spacings, i.e., ≈4.57 and ≈3.35 Å, are observed, which correspond well with the tetrahedral and octahedral layers of BM-SFO, respectively. In addition, the fast Fourier transfor- mation (FFT) patterns (insets in Figure 3e) clearly show the superlattice-free reflections from the nanofilament, in contrast to the ½(010) superlattice reflections from the matrix, further confirming the PV- and BM-SFO structures in the nanofila- ments and matrix, respectively. The PV-SFO nanofilaments can also be distinguished from the BM-SFO matrix based on the dif- ferent electronic structures, as revealed by the electron energy- loss spectroscopy (Figure S10, Supporting Information).

Another important observation from Figure 3d and Figure S11, Supporting Information is that the nanofilaments are prevented from completely extending through the matrix by some very thin (≈6.6 nm) BM-SFO gaps near the bottom inter- face. Since BM-SFO (typically n-type; bandgap: ≈2 eV^[28,29]) can form a Schottky contact with SRO, these BM-SFO gaps may thus be responsible for the above-mentioned “hidden” Schottky barrier and consequent asymmetric I–V behavior. Notably, the Schottky barrier between a very thin BM-SFO gap and SRO ena- bles the thermionic emission to occur favorably, thus making the BM-SFO gap not that insulating. Therefore, the state where a nanofilament is separated from SRO by a very thin BM-SFO gap indeed corresponds to an electrically ON state. Also because the BM-SFO gap is not that insulating, it may be unable to with- stand a sufficiently high voltage for driving the complete phase transformation from BM-SFO to PV-SFO, resulting in the for- mation of this gap (Figures S12–S14, Supporting Information).

After the reset, the PV-SFO nanofilaments are observed to be ruptured at the bottom (Figure 3h). More specifically, the bottom ends of the nanofilaments move far away from the SFO/

SRO interface (Figure 3i) and thus the BM-SFO gaps become apparently wider (average width: ≈22.5 nm; see Figure S11, Supporting Information). Meanwhile, the top ends of the nano- filaments are still touching the Pt electrode (Figure 3j). The detailed mechanism for the rupture of nanofilaments at the bottom is discussed in Section SXII, Supporting Information.

To further shed light on the filamentary RS mechanism, the local conduction behavior of the SFO film was investigated using conductive atomic force microscopy (C-AFM). For a 550 × 550 nm² area, its right half (320 × 550 nm²) was first electrically written with a tip voltage of −7 V. Then, a read scan with −2 V was applied to this area to generate a current map. As shown in Figure 4a, almost the whole pristine region exhibits negli- gible currents of < 0.1 nA. By contrast, the −7 V written region exhibits multiple conductive spots with currents of > 1 nA. The existence of multiple conductive spots in LRS can explain why R_LRS has a certain dependence on the electrode area (Figure S4, Supporting Information).

A closer inspection was carried out on a typical conductive spot along the red line in Figure 4a, by plotting the current pro- file together with the surface height change (Δh) profile (Figure 4b). The Δh profile was obtained by subtracting the surface height values along a section line in the pristine state from those along the same section line in the post-writing state (Figure S15, Sup- porting Information). A negative Δh value indicates the phase transformation from BM-SFO to PV-SFO, because the out-of-plane lattice spacing of BM-SFO (b_O/4 = ≈3.96 Å;

see Figures 1b and 3g) is larger than that of PV-SFO (a_C = ≈3.91 Å; see Figure 3g).

Comparing the upper and lower panels of Figure 4b, it is noted that the conductive spot (i.e., the current peak) is located where the Δh value is most negative. Therefore, the conductive spot may be assigned to be a filament consisting of PV-SFO. The diameter of the filament is estimated to be ≈20 nm using the full-width at half-maximum of the current peak, which is on the same order of the average diameter revealed by STEM.

The combined STEM and C-AFM results therefore demonstrate that the filament size in SFO can be very small (on the order of 10 nm), thus enabling the downscaling of SFO-based RS devices. Nevertheless, fila- ments with much larger sizes may also exist.

For example, very recently Nallagatla et al.^[30]

observed micrometer-size filaments in SFO by using X-ray absorption spectromicros- copy (XAS). In their SFO samples, signifi- cant lateral growth and merging of filaments may have occurred, thus leading to the large filaments. Alternatively, what they observed might be a dense assembly of multiple nano- filaments which appeared like a single large filament under XAS, considering that the spatial resolution of their XAS technique was 100 nm. If this is the case, their results are indeed consistent with ours (Figures 3d and 4a), where multiple PV-SFO nanofila- ments are distributed compactly and rela- tively homogenously in the BM-SFO matrix.

Another observation is that the surface height decreases in the whole −7 V written region (see lower panel of Figure 4b and Figure S15b, Supporting Information), sug- gesting that local BM-SFO → PV-SFO phase transformation may widely occur during the tip scanning with −7 V. A PV-SFO filament may first nucleate beneath the tip where the electric field is the most intense and then start to grow. Only when the PV-SFO filament extends almost through the film, this posi- tion is electrically switched ON. This explains Figure 3. Direct observation of PV-SFO conductive filaments in the BM-SFO matrix. a) STEM-

HAADF image of the pristine SFO film. b) Atomically resolved STEM-HAADF image showing the BM-SFO structure with the b-axis (indicated by white arrow) parallel to the out-of-plane direction, and c) corresponding FFT pattern, where the white circle indicates the superlattice reflection. d) STEM-HAADF image of the SFO film in the electroformed state, showing some typical PV-SFO nanofilaments almost extending through the BM-SFO matrix. e) Atomically resolved STEM-HAADF image from the interface between PV-SFO and BM-SFO, with FFT pat- terns from both regions. f) Enlarged image from panel (e). g) Out-of-plane lattice spacing map around the interface. h) STEM-HAADF image of the SFO film in the post-reset state, showing some typical PV-SFO nanofilaments which are ruptured at the bottom. Atomically resolved STEM-HAADF images from the i) bottom and j) top ends of a ruptured nanofilament.

why the current peak locates where the surface height decreases most significantly rather than in the whole −7 V written region.

It is also noteworthy that the current peak does not appear where the absolute surface height is lowest (see Figure 4a and Figure S15b, Supporting Information), indicating that the phase transformation rather than the thickness reduction is the origin for the dramatically enhanced conductance.

After re-writing the right region with +4 V, almost all the con- ductive spots disappear (see Figure 4c). Hence, while the −7 V writing leads to LRS, the +4 V writing can turn this region back to HRS. Moreover, the surface heights in this region are largely recovered after the +4 V writing (Figure S15c, Supporting Information). The reversible changes in both conductance and surface height demonstrate that the local phase transformation between BM-SFO and PV-SFO can be reversibly controlled by the electric field.

The above STEM and C-AFM measurements reveal that RS in SFO is caused by the formation/rupture of conductive fila- ments involving the nanoscale topotactic phase transformation between PV-SFO and BM-SFO. The RS mechanism can be better understood with the schematics illustrating the probable electrochemical and charge transport processes, as shown in Figure 5. In brief, during the electroforming, the PV-SFO fila- ments may nucleate at the top interface due to oxygen incorpo- ration from the ambient to SFO,^[31,32] and subsequently these filaments may grow vertically because the oxygen ions can migrate from top to bottom under large negative voltages.^[33]

Eventually, the filaments extend almost through the matrix with very thin BM-SFO gaps separating the filaments and the SRO layer (Figures 5b and 3d). Because the current can flow easily along the PV-SFO filaments and the BM-SFO/SRO Schottky barrier with a relatively low barrier height favors thermionic emission, an LRS is thus achieved. In addition, the forward- bias (reverse-bias) thermionic emission occurs in the negative (positive) voltage region (Figure 5e,f), resulting in the asym- metric I–V behavior (Figure 2). In the reset process, the oxygen extraction may take place at the top interface,^[31,32] and subse- quently the oxygen ions may migrate from bottom to top,^[33]

causing the filaments to be ruptured at the bottom (Figures 5c and 3h). As a result, the BM-SFO gaps become wider and thus the Schottky barrier height increases, leading to the HRS. In the HRS, reverse-bias thermionic emission across the Schottky

barrier dominates the conduction under positive voltage, while under negative voltage conduction is determined by hopping in the gap (Figure 5g,h). In the set process, the incorporated oxygen ions may migrate and reconnect the ruptured conduc- tive filaments, giving rise to the LRS again. A more detailed dis- cussion on the electrochemical and charge transport processes underlying the RS phenomenon is given in Sections SXIV and SXV of the Supporting Information.

In the scenario described above, oxygen transfer (incorporation/extraction) between the ambient and SFO is key to the formation/rupture of filaments. To verify this, we com- pared the I–V characteristics measured in air and vacuum for the same Pt/SFO/SRO device. While the device in air shows the bipolar RS behavior, that in vacuum exhibits no current jump during the electroforming and remains in the HRS with negligible I–V hysteresis during the subsequent voltage sweep (Figure S7, Supporting Information). One can thus deduce that the oxygen incorporation from the ambient into SFO indeed occurs and is indispensable for the filament formation (note: even with a top electrode, oxygen is able to penetrate the electrode to enable oxygen incorporation^[32]). This compara- tive result also rules out the drift of the oxygen ions between the SFO and SRO layers as the major cause for the observed RS; otherwise, RS would depend weakly on the oxygen in the ambient and the electroforming and set would take place at pos- itive voltages.^[19,21] In addition, we checked the surface potential variations associated with the oxygen incorporation/extraction, using scanning Kelvin probe microscopy (SKPM). If the oxygen incorporation (extraction) occurs, the amount of oxygen vacan- cies will decrease (increase) near the surface, leading to a reduc- tion (enhancement) in the surface potential. Our SKPM results show that the surface potential is reduced by the −7 V writing while enhanced by the +7 V writing (Figure S17, Supporting Information), consistent with the scenario of the oxygen incor- poration/extraction. For the vertical filament growth, assuming that the oxygen ions can migrate through the whole film within a critical time of ≈1 ms at an applied voltage of −4 V (Figure S2, Supporting Information), the oxygen mobility is estimated to be ≈1.4 × 10⁻⁸ cm² V⁻¹ s⁻¹, which is at least three orders of magnitude higher than the reported room tempera- ture (RT) values.^[34,35] Therefore, the electric field-driven vertical oxygen migration may be aided by Joule heating.^[36,37] The Joule Figure 4. Electrical manipulation of conductive filaments. a) C-AFM current map (read at −2 V) measured for a 550 × 550 nm² area after writing the right region (320 × 550 nm²) with −7 V. b) Current (upper) and surface height change (lower) profiles along the red line in panel (a). c) C-AFM current map measured after re-writing the right region with +4 V.

heating, together with the existing oxygen vacancy channels, may also enhance the in-plane oxygen diffusion, thus contrib- uting to lateral filament growth.^[38]

With the above understanding of the RS mechanism, we believe that the nanoscale topotactic phase transformation between BM-SFO and PV-SFO can enable us to aggressively downscale the SFO-based RS memory cells. The inset in Figure 1e shows well-ordered triangular Au nanoelectrodes on the SFO film, constituting the Au/SFO/SRO nanodevices for the C-AFM-based I–V measurements (see schematic set-up in the inset of Figure 6a). As seen from the I–V characteristics

displayed in Figure 6a, the Au/SFO/SRO nanodevices exhibit bipolar RS behavior akin to the large-size devices. The different top electrodes lead to similar RS behavior (Figure S18, Sup- porting Information), further supporting our conclusion that RS originates from the SFO layer. The most significant dif- ference between nanodevices and large-size devices is that the ON/OFF ratios in the nanodevices can reach ≈10⁴, much larger than those in the large-size devices (see comparison between Figure 2 and Figure 6a). Obtaining such high ON/OFF ratios in the nanodevices is a consequence of R_LRS increasing much more slowly than R_HRS as the electrode area decreases Figure 5. Filamentary RS mechanism. Schematics illustrating a) the pristine SFO film with the BM-SFO matrix, b) the formation of PV-SFO conduc- tive filaments after the electroforming, and c) the rupture of PV-SFO conductive filaments after the reset. The red arrows indicate the directions of oxygen transfer and migration. “TE” denotes the top electrode. d) Fits of hysteretic I–V curves in different voltage regions by using different conduction mechanisms. Schematic energy band diagrams illustrating e) forward-bias thermionic emission in LRS at negative voltages, f) reverse-bias thermionic emission in LRS at positive voltages, g) reverse-bias thermionic emission in HRS at positive voltages, and h) hopping in HRS at negative voltages.

(Figure S4, Supporting Information), which is caused by an increase in the effective number density of filaments in the LRS with decreasing electrode area.

The statistical distributions of V_set and V_reset obtained from the cycling test are plotted in Figure 6b. The V_set and V_reset are distributed in −4 to −6.5 V and +1.9 to +4.1 V, respectively, which are higher than the corresponding values observed in the large- size devices (see comparison between Figure 6b and Figure S2a, Supporting Information). These differences may be caused by different electrode areas, work functions,^[39] and kinetics of interfacial oxygen transfer^[40,41] of the top electrodes (i.e., Au and Pt). Nevertheless, the distributions of V_set and V_reset in our nano- devices are still narrow, suggesting good switching stability.

We further measured the RS characteristics for over 20 dif- ferent nanodevices, and their R_HRS and R_LRS values as well as ON/OFF ratios are statistically shown in Figure 6c. The ON/

OFF ratio fluctuates in a small range of 10³–10⁴, indicating the high yield and good uniformity of our nanodevices.

Figure 6d presents the retention properties of the Au/SFO/

SRO nanodevices. There is no significant degradation in R_HRS or R_LRS over a long period of 10⁵ s. Using the linear extrapola- tion method, an ON/OFF ratio of ≈10⁴ may be retained after 10 years, attesting to the nonvolatility of our nanodevices.

To test the switching speed, the switching characteristics as a function of write pulse width were measured with fixed pulse amplitudes of −8 and +4 V. As shown in Figure 6e, the switching from HRS to LRS or vice versa is completed at a pulse width of 0.1 ms. Notably, the switching in nano devices is faster and more abrupt than that in large-size devices

(see comparison between Figure 6e and Figure S2d, Supporting Information). The nanodevices contain fewer filaments, which may mean that shorter times are required for their formation and rupture.^[42]

Figure 6f shows the endurance characteristics of the Au/

SFO/SRO nanodevices measured by applying write pulses of

−8 V/0.1 ms and +4 V/0.1 ms repeatedly. Both RHRS and R_LRS remain relatively stable during the cycling, and an ON/OFF ratio of ≈10³ is sustained for up to 10⁷ cycles. Clearly, the nano- devices exhibit significantly improved endurance compared with the large-size devices (see comparison between Figure 6f and Figure S2e, Supporting Information). In our large-size devices, the endurance failure may be caused by the inability of an increasing number of filaments to be re-formed during the cycling.^[43,44] In the nanodevices, however, there are much fewer filaments and the occurrence of formation/rupture of fila- ments may become more deterministic during the cycling,^[45,46]

thereby improving the endurance.

For filamentary-type RS memories, traditional materials being extensively studied by both academia and industry are undoubtedly binary oxides (MOx, where M is a transition metal). The major obstacle to putting these materials into prac- tical application is the low reliability, which is mainly caused by:

i) lack of knowledge of optimal oxygen stoichiometry in MO_x and consequent blindness in selecting fabrication parameters, and ii) difficulty to control the formation/rupture of filaments due to the complex nature of the filaments in these materials.

To address these issues, researchers have paid great effort to develop new materials.

Figure 6. RS properties of Au/SFO/SRO nanodevices. a) Typical I–V characteristics with an inset showing the set-up for I–V measurements using C-AFM. b) Cumulative probability plots of V_set and V_reset for 50 cycles of I–V sweep. Inset shows the I–V curves of the 1st and 50th cycles. c) R_HRS and RLRS and corresponding ON/OFF ratios measured for 23 different devices. d) Retention data for both HRS and LRS. e) Write pulse width-dependent resistance evolution. The device is initially set in the LRS (HRS) before the +4 V (−8 V) pulse writing. f) Fatigue characteristics measured by applying

−8 V/0.1 ms and +4 V/0.1 ms write pulses repeatedly.

Table 1 summarizes the performance of filament-type RS memories based on traditional and emerging materials. In terms of their comprehensive performance, SFO stands out among the emerging materials and also compares favorably with commercial flash memory.^[61] However, currently SFO cannot yet compete with some well-established binary oxides, like HfOx and TaOx. It is nevertheless emphasized that the performance reported in this work is by no means the upper limit for the SFO-based devices. For example, the switching speed, which depends on the kinetics of topotactic phase trans- formation, may be further improved by i) engineering the top electrode/SFO interface to facilitate the interfacial oxygen transfer^[62] and ii) modifying the orientations of oxygen vacancy channels in the SFO bulk to facilitate the oxygen migration.^[20]

Moreover, compared with that in traditional binary oxides, the formation/rupture of filaments in SFO seems more control- lable because the mechanism based on nanoscale topotactic phase transformation between BM-SFO and PV-BFO is more deterministic and robust. We therefore believe that the ultimate

performance and reliability of SFO can surpass those of tradi- tional binary oxides.

In addition, assuming a cross-bar structure with a cell size of 4F² (F denotes the minimum feature size) and specifying F to be 10 nm (i.e., the average size of nanofilaments), the den- sity of SFO-based memory cells would reach the high value of

≈1.5 Tbit in.⁻².

Last but not least, the oxygen stoichiometry of the SFO films is well targeted (i.e., SrFeO_2.5) and these films can be facilely fabricated. Additionally, the SFO-based devices are compat- ible with standard complementary metal-oxide-semiconductor (CMOS) technology, since the substrate materialSrTiO₃, in single-crystal form, has been successfully integrated onto silicon wafers using a molecular beam epitaxy technique.^[63]

Given the high density, good performance, facile fabrication, and CMOS-compatibility, the SFO-based RS memory seems to be a promising alternative to commercial flash memory.

In summary, the RS effect associated with the electric field- induced local topotactic phase transformation has been studied Table 1. Device performance of filament-type RS memories based on different materials.

Material system Cell size [µm²] ON/OFF ratio Write voltage [V] Switching speed [µs] Retention [s] Endurance [cycles]

Oxides

HfOx[47] 1.6 × 10⁻³ >10 1–2 5 × 10⁻³ >10⁴

@ 200 °C

>10¹⁰

TaOx[48] 9 × 10⁻⁴ >10 4.5–7 10⁻² >10⁴

@ 250 °C

>10¹²

NiOx[49] 0.2 >10 0.5–1.5 10⁻² >10⁷

@ RT

>10⁶

WOx[50] 3 × 10⁻² ≈10 2–3.3 5 × 10⁻² >10⁴

@ 100 °C

>10⁷

TiOx[51] 4 × 10² >10² 2–3 5 × 10⁻³ >10⁶

@ 85 °C

>2 × 10⁶

ZnOx[52] ≈10⁵ ≈10⁷ 2–3 5 × 10⁻³ >10⁷

@ RT

>10²

SrTiO3[53] ≈10³ >10² 1–2 10² >10⁵

@ RT

>10⁶

BaTiO3[54] ≈8 × 10³ >10⁴ 1–5 7 × 10⁻² >7 × 10⁴

@ RT

>10⁵

BiFeO3[55] ≈3 × 10⁴ >10 4–6.5 NA >5 × 10⁴

@ RT

>10⁴

Emerging materials

Graphene oxide^[56] ≈8 × 10³ >10 <1 NA >10⁴

@ RT

>10²

Boron nitride^[57] ≈3 × 10⁴ ≈10² <1 NA >3 × 10³

@ RT

>5 × 10²

Black phosphorous^[58] ≈2 × 10⁵ >10³ <2 NA >3 × 10⁵

@ RT

NA

MoS2[59] ∼3 × 10³ ≈10² 3–4 NA >10⁵

@ RT

>10³

CH3NH3PbI_3−xClx[60] ≈2 × 10³ ≈3 <1 NA >10⁴

@ RT

>10²

This work

SrFeOx ≈2 × 10⁻² ≈10⁴ 4–7 10² >10⁵

@ RT

>10⁷

in SFO epitaxial thin films. It is unambiguously revealed that the RS arises from the formation/rupture of PV-SFO nanofilaments in the BM-SFO matrix, mediated by the electric field-controlled oxygen transfer and migration. This greatly deepens our under- standing of the RS mechanism in SFO and, more generally, in TMOs exhibiting topotactic phase transformations. In addition, the ultra-small size of PV-SFO nanofilaments (average diam- eter: ≈10 nm) enables us to develop Au/SFO/SRO nanoscale RS devices (lateral size: ≈180 nm). These nanodevices show superb RS performance including good device-to-device uniformity, large ON/OFF ratio of ≈10⁴, long retention time of > 10⁵ s, high endurance of > 10⁷ cycles, and relatively fast switching speed of ≈100 µs, demonstrating their great potential for use in high- density information storage devices. It is also expected that applications of the easily accessible topotactic phase transforma- tion in SFO (and analogous TMOs) can be extended to a wide range of fields, like electric-field control of magnetism, energy storage/conversion, redox catalysis, and gas sensing.

Experimental Section

Film Deposition: ≈70 nm thick SrFeO2.5 (BM-SFO) epitaxial thin films on ≈40 nm thick SRO bottom electrode layers were grown on (001)-oriented STO substrates via pulsed laser deposition (PLD) using a KrF excimer laser (λ = 248 nm). The SRO layers were first deposited at a substrate temperature of 650 °C and an oxygen pressure of 20 Pa.

Subsequently, the SFO films were grown at an oxygen pressure of 1 Pa with the substrate temperature unchanged. During the growth of both SRO and SFO films, the laser energy was fixed at ≈1.2 J cm⁻² with a repetition rate of 5 Hz. After the deposition, the samples were cooled down to room temperature at a cooling rate of 10 °C min⁻¹ in 1 Pa oxygen environment. For large-size Pt/SFO/SRO devices, circular Pt electrodes with a diameter of 200 µm were deposited via PLD at high vacuum through a shadow mask. For Au/SFO/SRO nanodevices, triangular Au nanodots with ≈180 nm in length were deposited via thermal evaporation at high vacuum using well-ordered polystyrene spheres as templates. Au rather than Pt was used for nanoelectrodes because Au was compatible with the thermal evaporation method.

Structural and Microstructural Characterizations: The crystalline structures of SFO films were characterized by XRD and RSM (PANalytical X’Pert PRO). STEM was conducted on a JEOL ARM200F equipped with a cold field emission gun and ASCOR probe corrector. The thin TEM specimens were prepared by a focused ion beam machine (FEI Versa 3D).

Macroscopic Electrical Measurements: The RS characteristics of large- size Pt/SFO/SRO devices were investigated using a source meter (Keithley 6430) and a home-built probe station. In the pulse switching and endurance measurements, a ferroelectric workstation (Precision Multiferroic, Radiant) was used to generate the pulses.

Nanoscale Morphological and Electrical Characterizations: The AFM-based investigations were performed using a scanning probe microscope (Cypher, Asylum Research) with conductive Pt-coated silicon probes (EFM Arrow, Nanoworld). Three different modes, i.e., AFM, C-AFM, and SKPM, were used to characterize the topography, local current, and surface potential, respectively. The RS characteristics of Au/

SFO/SRO nanodevices were measured either by C-AFM only or by an integrated set-up connecting the Keithley 6430 source meter to the AFM tip. In the SKPM measurement, the tip lift height was kept as 50 nm and the AC drive amplitude was 1 V.

Supporting Information

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

Acknowledgements

J.T. and H.W. contributed equally to this work. The authors thank National Key Research Program of China (nos. 2016YFA0201002 and 2016YFA300101), State Key Program for Basic Researches of China (no.

2015CB921202), National Natural Science Foundation of China (nos.

51602110, 11674108, and 51431006), Guangdong Innovative Research Team Program (no. 2013C102), Science and Technology Project of Guangdong Province (nos. 2016B090918083, 2017B030301007, and 2015B090927006), Natural Science Foundation of Guangdong Province (no. 2016A030308019), and Science and Technology Project of Shenzhen Municipal Science and Technology Innovation Committee (GQYCZZ20150721150406). H.W. acknowledges Lee Kuan Yew Postdoctoral Fellowship through a Singapore Ministry of Education Tier 1 grant (R-284-000-212-114). H.W. and S.J.P. acknowledge the support by the Ministry of Education, Singapore under its Tier 2 Grant (grant no. MOE2017-T2-1-129). X.G., X.L., and Z.F. acknowledge the Project for Guangdong Province Universities and Colleges Pearl River Scholar Funded Scheme 2014, 2016, and 2018, respectively.

Conflict of Interest

The authors declare no conflict of interest.

Keywords

nanofilaments, resistive switching, SrFeOx, topotactic phase transformation

Received: June 10, 2019 Revised: October 3, 2019 Published online: October 22, 2019

[1] B. C. H. Steele, A. Heinzel, Nature 2012, 414, 345.

[2] P. Poizot, S. Laruelle, S. Grugeon, L. Dupont, J. M. Tarascon, Nature 2000, 407, 496.

[3] X. Lang, A. Hirata, T. Fujita, M. Chen, Nat. Nanotechnol. 2011, 6, 232.

[4] X. Leng, J. Pereiro, J. Strle, G. Dubuis, A. T. Bollinger, A. Gozar, J. Wu, N. Litombe, C. Panagopoulos, D. Pavuna, I. Božovic´, npj Quantum Mater. 2017, 2, 35.

[5] N. Biškup, J. Salafranca, V. Mehta, M. P. Oxley, Y. Suzuki, S. J. Pennycook, S. T. Pantelides, M. Varela, Phys. Rev. Lett. 2014, 112, 087202.

[6] R. Mishra, Y.-M. Kim, J. Salafranca, S. K. Kim, S. H. Chang, A. Bhattacharya, D. D. Fong, S. J. Pennycook, S. T. Pantelides, A. Y. Borisevich, Nano Lett. 2014, 14, 2694.

[7] J. H. Jang, Y.-M. Kim, Q. He, R. Mishra, L. Qiao, M. D. Biegalski, A. R. Lupini, S. T. Pantelides, S. J. Pennycook, S. V. Kalinin, A. Y. Borisevich, ACS Nano 2017, 11, 6942.

[8] J. P. Hodges, S. Short, J. D. Jorgensen, X. Xiong, B. Dabrowski, S. M. Mini, C. W. Kimball, J. Solid State Chem. 2000, 151, 190.

[9] C. Greaves, A. J. Jacobson, B. C. Tofield, B. E. F. Fender, Acta Crystallogr. B 1975, 31, 641.

[10] A. E. Bocquet, A. Fujimori, T. Mizokawa, T. Saitoh, H. Namatame, S. Suga, N. Kimizuka, Y. Takeda, M. Takano, Phys. Rev. B 1992, 45, 1561.

[11] S. Ishiwata, M. Tokunaga, Y. Kaneko, D. Okuyama, Y. Tokunaga, S. Wakimoto, K. Kakurai, T. Arima, Y. Taguchi, Y. Tokura, Phys. Rev.

B 2011, 84, 054427.

[12] J. Zhao, Y. Luo, J.-O. Wang, H. Qian, C. Liu, X. He, Q. Zhang, H. Huang, B. Zhang, S. Li, E. Guo, C. Ge, T. Yang, X. Li, M. He, L. Gu, K.-J. Jin, K. Ibrahim, H. Guo, Sci. China Mater. 2019, 62, 1162.

[13] P. Adler, A. Lebon, V. Damljanovic´, C. Ulrich, C. Bernhard, A. V. Boris, A. Maljuk, C. T. Lin, B. Keimer, Phys. Rev. B 2006, 73, 094451.

[14] A. Khare, D. Shin, T. S. Yoo, M. Kim, T. D. Kang, J. Lee, S. Roh, I.-H. Jung, J. Hwang, S. W. Kim, T. W. Noh, H. Ohta, W. S. Choi, Adv. Mater. 2017, 29, 1606566.

[15] H. Jeen, W. S. Choi, M. D. Biegalski, C. M. Folkman, I. C. Tung, D. D. Fong, J. W. Freeland, D. Shin, H. Ohta, M. F. Chisholm, H. N. Lee, Nat. Mater. 2013, 12, 1057.

[16] A. Nemudry, M. Weiss, I. Gainutdinov, V. Boldyrev, R. Schöllhorn, Chem. Mater. 1998, 10, 2403.

[17] N. Lu, P. Zhang, Q. Zhang, R. Qiao, Q. He, H.-B. Li, Y. Wang, J. Guo, D. Zhang, Z. Duan, Z. Li, M. Wang, S. Yang, M. Yan, E. Arenholz, S. Zhou, W. Yang, L. Gu, C.-W. Nan, J. Wu, Y. Tokura, P. Yu, Nature 2017, 546, 124.

[18] C. Ge, C. Liu, Q. Zhou, Q. Zhang, J. Du, J. Li, C. Wang, L. Gu, G. Yang, K. Jin, Adv. Mater. 2019, 31, 1900379.

[19] S. K. Acharya, R. V. Nallagatla, O. Togibasa, B. W. Lee, C. Liu, C. U. Jung, B. H. Park, J.-Y. Park, Y. Cho, D.-W. Kim, J. Jo, D.-H. Kwon, M. Kim, C. S. Hwang, S. C. Chae, ACS Appl. Mater.

Interfaces 2016, 8, 7902.

[20] S. K. Acharya, J. Jo, N. V. Raveendra, U. Dash, M. Kim, H. Baik, S. Lee, B. H. Park, J. S. Lee, S. C. Chae, C. S. Hwang, C. U. Jung, Nanoscale 2017, 9, 10502.

[21] O. T. Tambunan, K. J. Parwanta, S. K. Acharya, B. W. Lee, C. U. Jung, Y. S. Kim, B. H. Park, H. Jeong, J.-Y. Park, M. R. Cho, Y. D. Park, W. S. Choi, D.-W. Kim, H. Jin, S. Lee, S. J. Song, S.-J. Kang, M. Kim, C. S. Hwang, Appl. Phys. Lett. 2014, 105, 063507.

[22] H.-B. Li, N. Lu, Q. Zhang, Y. Wang, D. Feng, T. Chen, S. Yang, Z. Duan, Z. Li, Y. Shi, W. Wang, W.-H. Wang, K. Jin, H. Liu, J. Ma, L. Gu, C. Nan, P. Yu, Nat. Commun. 2017, 8, 2156.

[23] W. Paulus, H. Schober, S. Eibl, M. Johnson, T. Berthier, O. Hernandez, M. Ceretti, M. Plazanet, K. Conder, C. Lamberti, J. Am. Chem. Soc. 2008, 130, 16080.

[24] A. Khare, J. Lee, J. Park, G.-Y. Kim, S.-Y. Choi, T. Katase, S. Roh, T. S. Yoo, J. Hwang, H. Ohta, J. Son, W. S. Choi, ACS Appl. Mater.

Interfaces 2018, 10, 4831.

[25] J. Zhao, H. Guo, X. He, Q. Zhang, L. Gu, X. Li, K.-J. Jin, T. Yang, C. Ge, Y. Luo, M. He, Y. Long, J.-O. Wang, H. Qian, C. Wang, H. Lu, G. Yang, K. Ibrahim, ACS Appl. Mater. Interfaces 2018, 10, 10211.

[26] R. Waser, M. Aono, Nat. Mater. 2007, 6, 833.

[27] D.-H. Kwon, K. M. Kim, J. H. Jang, J. M. Jeon, M. H. Lee, G. H. Kim, X.-S. Li, G.-S. Park, B. Lee, S. Han, M. Kim, C. S. Hwang, Nat.

Nanotechnol. 2010, 5, 148.

[28] J. Young, J. M. Rondinelli, Phys. Rev. B 2015, 92, 174111.

[29] R. J. D. Tilley, Defects in Solids, John Wiley & Sons, New York 2008.

[30] V. R. Nallagatla, T. Heisig, C. Baeumer, V. Feyer, M. Jugovac, G. Zamborlini, C. M. Schneider, R. Waser, M. Kim, C. U. Jung, R. Dittmann, Adv. Mater. 2019, 31, 1903391.

[31] Q. Lu, B. Yildiz, Nano Lett. 2016, 16, 1186.

[32] D. S. Jeong, H. Schroeder, U. Breuer, R. Waser, J. Appl. Phys. 2008, 104, 123716.

[33] B. Cui, P. Werner, T. Ma, X. Zhong, Z. Wang, J. M. Taylor, Y. Zhuang, S. S. P. Parkin, Nat. Commun. 2018, 9, 3055.

[34] O. A. Savinskaya, A. P. Nemudry, N. Z. Lyakhov, Inorg. Mater. 2007, 43, 1350.

[35] J. T. Mefford, X. Rong, A. M. Abakumov, W. G. Hardin, S. Dai, A. M. Kolpak, K. P. Johnston, K. J. Stevenson, Nat. Commun. 2016, 7, 11053.

[36] D. Cooper, C. Baeumer, N. Bernier, A. Marchewka, C. La Torre, R. E. Dunin-Borkowski, S. Menzel, R. Waser, R. Dittmann, Adv.

Mater. 2017, 29, 1700212.

[37] C. Baeumer, C. Schmitz, A. H. H. Ramadan, H. Du, K. Skaja, V. Feyer, P. Müller, B. Arndt, C.-L. Jia, J. Mayer, R. A. De Souza, C. M. Schneider, R. Waser, R. Dittmann, Nat. Commun. 2015, 6, 8610.

[38] L. Yao, S. Inkinen, S. van Dijken, Nat. Commun. 2017, 8, 14544.

[39] W.-G. Kim, S.-W. Rhee, Microelectron. Eng. 2010, 87, 98.

[40] C. B. Lee, B. S. Kang, A. Benayad, M. J. Lee, S. E. Ahn, K. H. Kim, G. Stefanovich, Y. Park, I. K. Yoo, Appl. Phys. Lett. 2008, 93, 042115.

[41] M.-C. Chen, T.-C. Chang, C.-T. Tsai, S.-Y. Huang, S.-C. Chen, C.-W. Hu, S. M. Sze, M.-J. Tsai, Appl. Phys. Lett. 2010, 96, 262110.

[42] M.-J. Lee, S. Han, S. H. Jeon, B. H. Park, B. S. Kang, S.-E. Ahn, K. H. Kim, C. B. Lee, C. J. Kim, I.-K. Yoo, D. H. Seo, X.-S. Li, J.-B. Park, J.-H. Lee, Y. Park, Nano Lett. 2009, 9, 1476.

[43] Y. Y. Chen, R. Degraeve, S. Clima, B. Govoreanu, L. Goux, A. Fantini, G. S. Kar, G. Pourtois, G. Groeseneken, D. J. Wouters, M. Jurczak, in Int. Electron Devices Meeting, 2012 IEEE International (Ed: V. Misra), IEEE, Piscataway, NJ 2012, pp. 20.3.1–20.3.4.

[44] H. Y. Lee, Y. S. Chen, P. S. Chen, P. Y. Gu, Y. Y. Hsu, S. M. Wang, W. H. Liu, C. H. Tsai, S. S. Sheu, P. C. Chiang, W. P. Lin, C. H. Lin, W. S. Chen, F. T. Chen, C. H. Lien, M. Tsai, in Int. Electron Devices Meeting, 2010 IEEE International (Ed: M. Ieong), IEEE, Piscataway, NJ 2010, pp. 460–463.

[45] S. H. Jo, K.-H. Kim, W. Lu, Nano Lett. 2009, 9, 496.

[46] C.-H. Tu, C.-C. Chang, C.-H. Wang, H.-C. Fang, M. R. S. Huang, Y.-C. Li, H.-J. Chang, C.-H. Lu, Y.-C. Chen, R.-C. Wang, Y. Tzeng, C.-P. Liu, J. Alloys Compd. 2014, 615, 754.

[47] Y. Y. Chen, L. Goux, S. Clima, B. Govoreanu, R. Degraeve, G. Kar, A. Fantini, G. Groeseneken, D. Wouters, M. Jurczak, IEEE Trans.

Electron Devices 2013, 60, 1114.

[48] M.-J. Lee, C. B. Lee, D. Lee, S. R. Lee, M. Chang, J. H. Hur, Y.-B. Kim, C.-J. Kim, D. H. Seo, S. Seo, U. I. Chung, I.-K. Yoo, K. Kim, Nat.

Mater. 2011, 10, 625.

[49] I. G. Baek, M. S. Lee, S. Seo, M. J. Lee, D. H. Seo, D. Suh, J. C. Park, S. O. Park, H. S. Kim, I. K. Yoo, U. Chung, J. T. Moon, in Int. Electron Devices Meeting, IEDM Technical Digest, 2004 IEEE International (Ed: J. Welser), IEEE, Piscataway, NJ 2004, pp. 587–590.

[50] W. C. Chien, Y. C. Chen, E. K. Lai, Y. D. Yao, P. Lin, S. F. Horng, J. Gong, T. H. Chou, H. M. Lin, M. N. Chang, Y. H. Shih, K. Y. Hsieh, R. Liu, C. Lu, IEEE Electron Device Lett. 2010, 31, 126.

[51] C. Yoshida, K. Tsunoda, H. Noshiro, Y. Sugiyama, Appl. Phys. Lett.

2007, 91, 223510.

[52] Y. C. Yang, F. Pan, Q. Liu, M. Liu, F. Zeng, Nano Lett. 2009, 9, 1636.

[53] H. Nili, S. Walia, S. Balendhran, D. B. Strukov, M. Bhaskaran, S. Sriram, Adv. Funct. Mater. 2014, 24, 6741.

[54] Z. Yan, Y. Guo, G. Zhang, J. M. Liu, Adv. Mater. 2011, 23, 1351.

[55] X. Ou, Y. Shuai, W. Luo, P. F. Siles, R. Kögler, J. Fiedler, H. Reuther, S. Zhou, R. Hübner, S. Facsko, M. Helm, T. Mikolajick, O. G. Schmidt, H. Schmidt, ACS Appl. Mater. Interfaces 2013, 5, 12764.

[56] C. L. He, F. Zhuge, X. F. Zhou, M. Li, G. C. Zhou, Y. W. Liu, J. Z. Wang, B. Chen, W. J. Su, Z. P. Liu, Y. H. Wu, P. Cui, R.-W. Li, Appl. Phys. Lett. 2009, 95, 232101.

[57] K. Qian, R. Y. Tay, V. C. Nguyen, J. Wang, G. Cai, T. Chen, E. H. T. Teo, P. S. Lee, Adv. Funct. Mater. 2016, 26, 2176.

[58] C. Hao, F. Wen, J. Xiang, S. Yuan, B. Yang, L. Li, W. Wang, Z. Zeng, L. Wang, Z. Liu, Y. Tian, Adv. Funct. Mater. 2016, 26, 2016.

[59] M. M. Rehman, G. U. Siddiqui, J. Z. Gul, S.-W. Kim, J. H. Lim, K. H. Choi, Sci. Rep. 2016, 6, 36195.

[60] E. J. Yoo, M. Lyu, J.-H. Yun, C. J. Kang, Y. J. Choi, L. Wang, Adv.

Mater. 2015, 27, 6170.

[61] R. Bez, E. Camerlenghi, A. Modelli, A. Visconti, Proc. IEEE 2003, 91, 489.

[62] Q. H. Qin, L. Äkäslompolo, N. Tuomisto, L. Yao, S. Majumdar, J. Vijayakumar, A. Casiraghi, S. Inkinen, B. Chen, A. Zugarramurdi, M. Puska, S. van Dijken, Adv. Mater. 2016, 28, 6852.

[63] H. Li, X. Hu, Y. Wei, Z. Yu, X. Zhang, R. Droopad, A. A. Demkov, J. Edwards, K. Moore, W. Ooms, J. Kulik, P. Fejes, J. Appl. Phys.

2003, 93, 4521.