Resistance switching memory in perovskite oxides

Contents lists available atScienceDirect

Annals of Physics

journal homepage:www.elsevier.com/locate/aop

Resistance switching memory in perovskite oxides

Z.B. Yan

^∗

, J.-M. Liu

Laboratory of Solid State Microstructures and Innovation Center of Advanced Microstructures, Nanjing University, Nanjing 210093, China

a r t i c l e i n f o

Article history:

Received 21 October 2014 Received in revised form 16 March 2015 Accepted 29 March 2015 Available online 3 April 2015

Keywords:

Resistance switching Memristor Perovskite oxide

Ferroelectric tunneling junction

a b s t r a c t

The resistance switching behavior has recently attracted great at- tentions for its application as resistive random access memories (RRAMs) due to a variety of advantages such as simple structure, high-density, high-speed and low-power. As a leading storage me- dia, the transition metal perovskite oxide owns the strong correla- tion of electrons and the stable crystal structure, which brings out multifunctionality such as ferroelectric, multiferroic, superconduc- tor, and colossal magnetoresistance/electroresistance effect, etc.

The existence of rich electronic phases, metal–insulator transition and the nonstoichiometric oxygen in perovskite oxide provides good platforms to insight into the resistive switching mechanisms.

In this review, we first introduce the general characteris- tics of the resistance switching effects, the operation methods and the storage media. Then, the experimental evidences of con- ductive filaments, the transport and switching mechanisms, and the memory performances and enhancing methods of perovskite oxide based filamentary RRAM cells have been summarized and discussed. Subsequently, the switching mechanisms and the per- formances of the uniform RRAM cells associating with the carrier trapping/detrapping and the ferroelectric polarization switching have been discussed. Finally, the advices and outlook for further investigating the resistance switching and enhancing the memory performances are given.

©2015 Elsevier Inc. All rights reserved.

∗Corresponding author.

E-mail addresses:zbyan@nju.edu.cn(Z.B. Yan),liujm@nju.edu.cn(J.-M. Liu).

http://dx.doi.org/10.1016/j.aop.2015.03.028 0003-4916/©2015 Elsevier Inc. All rights reserved.

Fig. 1. SchematicI–Vcurves of (a) unipolar and (b) bipolar switching modes. SymbolIcompdenotes the compliance current, which is adopted during the set process to prevent any permanent breakdown.

1. Introduction

1.1. Basic of resistance switching

The resistance switching (RS) behavior has attracted great attention for its application as re- sistive random access memories (RRAMs) due to a variety of advantages such as simple structure (electrode/active-layer/electrode sandwiched structures), high-density (excellent miniaturization po- tential down to<10 nm) [1], high-speed (sub-ns operation speed) [2], high-endurance (>^1012switch- ing cycles) [1], and low-power (<0.1 pJ energy consumption) [3,4]. There are two ways to activate the RS sequence [5]. One is the unipolar switching mode, in which the voltages with different magnitudes control the memory switching between high resistance state (HRS) and low resistance state (LRS). The other one is the bipolar switching mode, in which the voltages with opposite polarities control the re- versible memory switching between the HRS and LRS. The nonvolatile memory effect in RRAM cell is realized through the electrically controlled RS between the HRS and LRS. For many initially fabricated RRAM cells, a ‘‘Forming’’ process, in which a forming voltage (V_form) forces the RRAM cell to develop the initially conductive filaments with the limitation of compliance current, is usually done before the memory cell can work. InFig. 1(a) and (b) the schematicI–Vcurves of the unipolar and bipolar switch- ing modes are shown respectively. The switching from the HRS to the LRS is named as the ‘‘Set’’ process and the corresponding switching voltage is denoted asV_set. During the ‘‘Set’’ and ‘‘Forming’’ processes, a compliance current is usually used to limit the big avalanche current to avoid the damage of memory cell. In contrast, the switching event from the LRS to the HRS is denoted as the ‘‘Reset’’ process and the corresponding switching voltage isV_reset. It is noted that theV_formis usually higher than theV_set.

1.2. Storage media

The RS behavior has been observed in many electrode/active-layer/electrode sandwiched struc- tures. In a basic memory cell, the top and bottom electrodes can use the same or different materials, which can be the elementary substantial metals (Au, Pt, W, Al, Cu, Ag etc.) [4], metallic alloys (Cu–Ti, Pt–Al etc.) [6,7], graphene [8], NiSi [9], nitrides (TiN) [10], and oxides (indium-tin-oxide, Nb:SrTiO₃, SrRuO₃etc.) [11–14]. According to the role played in the RS, the electrode materials can be roughly sorted into two types. One is active electrodes (such as Ti, Cu, Agetc.), in which the migration and/or redox of electrode ions occur near the electrode/active-layer interface and contribute to the RS be- havior. The other one is inert electrodes (such as Pt and Au), in which the electrode ions are stable and do not directly take part in the RS. For the active-layer materials in the RRAM memory cell, there are many more broadened selections. It can be organic matters, silicon, graphene, amorphous oxides, crystalline binary or complex oxides, and even organic–inorganic compounds [15].

Special choices of RRAM cell materials with specific cell fabrication technologies are revealed to strongly influence the memory performances. Among many active-layer candidates, perovskite oxides

Fig. 2. Resistances measured upon various configurations. The top electrode (TE) was cut into TE-I and TE-II by the probe tip after the cell was switched from HRS to LRS by the SET process [5].

are emerging as one class of the leading materials [3,16–18]. The perovskite oxides have stable crys- tal structure, and capability of accommodating non-stoichiometric ions which allows for the lo- cal ionic migration and thermochemical reaction benefiting to the stable RS [19,20]. Besides, the strong electronic correlation in perovskite oxides brings out rich electronic phases and presents mul- tifunctionality, such as ferroelectric, multiferroic, superconductor, and colossal magnetoresistance/

electroresistance effects etc. [15,19]. The competing phenomena between various electronic phases allow the metal–insulator transition (MIT) and thus the dramatic change of resistance in response to a minute external voltage, providing an opportunity for advanced electronics [15,21]. In this short review, the RS behavior based on the transition metal perovskite oxides are focused and discussed, while our contributions in recent several years will be highlighted too.

2. Filamentary resistance switching

2.1. Evidences of conductive filaments

It has been widely evidenced that the formation and rupture of conductive filaments between the top and bottom electrodes are the most common way to produce unipolar or bipolar RS [5,20,22].

For example, the Au/BaTi_0.95Co_0.05O₃/Pt cell, as reported in Ref. [5], shows the filamentary RS. First, the Au/BaTi_0.95Co_0.05O₃/Pt cell is set to the LRS and then the top electrode (TE) is cut into two parts (denoted as TE-I and TE-II respectively), the resistance measured on TE-I is in the LRS while that on TE-II still remains at the HRS, as shown inFig. 2. This electrode-cutting experiment indicates that the

‘‘Set’’ process occurs only at local regions and verifies that the conductivity inside the whole memory cell is inhomogeneous.

Another interesting verification experiment is shown inFig. 3. By varying the TE pad area with diameter from 100 to 200µm, the switching I–V behavior of the Au/YMnO₃/Pt memory cell is not altered remarkably, which suggests a conductive filament network existing inside the memory cell [22]. Besides, the insertion of a Pt layer inside the middle of YMnO₃film leads to the multiple

‘‘Reset’’ processes and multiple current steps, as shown inFig. 3(b). This is well understood that the metal insertion into a network of conductive filaments sandwiched between the two terminal electrodes will enhance the connectivity and consequently raise the possibility to find multiple conductive filaments between the two top electrodes. Therefore, the multi-stepI–Vbehavior further confirms the filament switching mechanism.

Besides, real-time observation of the conductive filaments in some memory cells was also observed using transmission electron microscopy (TEM) and conductive atomic force microscopy (C- AFM). For instance, the three-dimensional characterization of the conductive filaments in nanoscale Cu/Al₂O₃/TiN memory device was reported using conventional C-AFM technology, as shown in Fig. 4[23].

Fig. 3. (a)I–Vcurve in switching of ten cycles in Au/YMn1−δO3(300 nm)/Pt. The olive circle dots are measured with top electrode of 200µm in diameter, while others with top electrode of 100µm in diameter. (b)I–Vdata of Au/YMn1−δO3(150 nm)/

Pt/YMn1−δO3(150 nm) [22].

Fig. 4. (a) Planar 2D C-AFM, performed on the cross-point memory element (bottom inset) in SET-state. The C-AFM is completely ineffective due to the presence of the TE shielding the CF observation. (b) Schematic of the C-AFM tomography procedure, the diamond tip is exploited to collect several slices at different heights of the CF after the removal of the TE.

(c) Over imposition of the collected 2D C-AFM slices, prior to the 3D interpolation. Note, the average space between each slice is∼0.5 nm. (d) Collection of 2D slices constituting the data set for the 3D interpolation (scale bar 80 nm). The CF appears in the middle of the active area after top electrode removal. The highly conductive features on the top-left and bottom-right corners are the exposed parts of the TiN BE, which is progressively exposed during the removal of Al2O3[23].

2.2. Switching mechanisms

A series of questions can be raised in terms of the above highlighted experiments. What is the com- position of conductive filaments, where and how does the RS occur inside the complex-oxide-based RRAM cell? These are the key points to be answered for further enhancing the RS performance. To answer these questions, we briefly review the micro-mechanisms of the filament-type RS here. It can be classified into three types, including ion migration and thermochemical reaction, MIT, and charge trapping/de-trapping, each of which will be discussed separately in the following sections [4,24,25].

2.2.1. Ion migration and thermochemical reaction

There exists often non-stoichiometry of oxygen and/or other oxide ions inside the perovskite oxide films [19,26]. Driven by an external electric field, the charged nonstoichiometric ions (such as oxygen

Fig. 5. Schematic diagrams of the Schottky barrier and the current transport across the Pt/Nb:STO junction for the HRS and LRS [33].

ions or vacancies) migrate along the field direction, leading to the local redox reactions characterized by a valence variation of the cation and as a result the response of local resistivity [27]. For this reason, the anion migration-based RRAMs are usually called valence change memories (VCMs) in literature [20]. Naturally, the conductive filaments, composed by the sub-oxides (e.g. the conductive Ti_nO_2n−1phase in insulating TiO₂matrix [26]), may be developed alongside the ion-migration paths.

Considering the random distribution of non-stoichiometric ions or defects, the presence of filaments has great randomness. It has already been demonstrated that the exact formation and rupture processes of the conductive filaments are closely related to the initial distribution of anion vacancies in oxides [28].

In the bipolar ion-migration-based RRAM cells, the voltage polarity directs the ion migration and determines the direction of the chemical reaction, i.e., reduction or oxidation. If a forward voltage drives the ion migration and forms the conductive filament by local redox, a backward voltage will partially retract the previously migrated anions, which will reverse the local redox to some extent and result in the rupture of filaments. For example, in the Ag/insulator/Pt cell and with a positive voltage applying to the Ag electrode, the Ag atoms near the Ag/insulator interface are partially oxidized into the Ag ions and then migrate toward cathode and finally reduce to Ag atoms again near the cathode, thus forming the conductive filaments [29]. As another negative voltage applies to the Ag electrode, the Ag atoms near the cathode are re-oxidized into Ag ions and migrates back toward the Ag electrode, causing the rupture of Ag filaments [29]. In this kind of the ion-migration-based cells, an active metal is used as one electrode and is also the main component of the conductive filament [3]. Usually, inert metal will be selected as another electrode to avoid the stability of RS [20].

Another kind mechanism of the ion-migration is that the excess positive charged defects (e.g. ex- cess positive ions or oxygen vacancies) migrate along the direction of electric field, which acts as dopant centers that change the Fermi level and consequently contribute to the electrical conductivity within the local region [30]. The conductive filament is assumed to consist of a suboxide phase with lo- cally enhanced density of defects, e.g. the filamentary Ti₄O₇phase inside the TiO₂film [26]. Under the application of negative voltage, the reverse electric field and the parasitic Joule heating allow the re- verse redox and then causes the rupture of filaments [30]. In this kind of ion-migration-based cells, the sufficient oxygen vacancies or excess ion with variable valence are usually necessary for the stable RS.

According to these bipolar switching mechanisms, the filament is usually ruptured near the electrode/oxide interface. Specifically, the rupture region is near the anode for then-type oxide media or near the cathode for the p-type oxide media [4,31].

In addition, it has been revealed that the ion migration without redox can also bring out bipolar filament-type RS behavior [25]. It was reported that the photocurrent across the Pt/Nb:STO junction remains constant during the RS, suggesting that only the local barrier profile of the junction is varied by electric field [32]. The investigation on the temperature dependentI–Vdata and the photocurrent spectra gives the same conclusion [33]. The migration of oxygen vacancies near the electrode/oxide interface can modulate the Schottky barrier height and the depletion width at the local regions of interface, leading to the variation of local conductivity across the electrode/oxide interface [34]. The diagram variations of Schottky barrier are schematically shown inFig. 5[33].

For the unipolar ion-migration-based RRAM cells, however, the rupture of conductive filament is not realized by the retraction of previously migrated ions or by the modulation of junction barrier.

Fig. 6. Au/YMn1−δO3/Pt cell: (a)R–Tdependence of both HRS and LRS. (b) The logI– logVplot, with the green lines fitted by the ohmic law. (c) MeasuredV/^Ias a function ofI²at the LRS in high field region (0.³→0.75 V). (d) ln(^I/^V)as a function of V^1/2at the HRS in high field region (2.⁹→3.8 V) [22]. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Instead, the Joule heating assisted thermochemical reaction becomes significant [5]. In such kind of RRAM cells, the local conductive sub-lattice may be very stable and the backward voltage cannot provide enough energy to retract the migrated ions and drive the local reverse reaction. Nevertheless, the electric current in local regions can bring out sufficient Joule heat that can greatly raise the temperature at the narrowest region of the conductive filament, assisting the local reverse reaction and the rupture of filament [30,35].

For instance, the ‘‘Reset’’ process in Au/YMn₁−δO₃/Pt filament-type unipolar RRAM cell was verified to be the result of Joule heating [22]. As shown inFig. 6(a), the insulating behavior of the HRS and the metal behavior of the LRS indicate the MIT after the RS process. As current flows through the conductive filaments, the raised temperature (1^T) in the local region depends linearly on the Joule heating power (I²R), i.e.1^T ∝I²R. The raise of local temperature in turn raises the resistance of the LRS due to the metal behavior, i.e.1^R∝₁T, forming a positive feedback loop. As an approximation, one can obtain the relation of1^R∝I²[36]. Therefore, the linear dependency ofRwithI²in the LRS inFig. 6(c) evidences that the Joule heating effect dominates the electrical prosperity of the LRS in the high voltage region and causes to the switching from the LRS to the HRS. The Joule-heating assisted local reduction of Mn⁴⁺back to Mn³⁺ruptures the conductive filament in the YMn₁−δO₃film. It has been estimated that the local temperature raised by the Joule heating can be above several hundreds Kelvin to assure the reverse redox [22,37].

Furthermore, the ion migration during the RS can be inferred by analyzing theI–Vbehaviors. For example,Fig. 6(d) shows theI–Vbehavior of the HRS in Au/YMn₁−δO₃/Pt cell. The two linear depen- dences of ln(^I/^V)^withV1/2indicate two types of transport modes. One is the Ohmic behavior at low field region (V < ³.2 V), the other is the Poole–Frenkel emission mechanism at high field region (3.^2V→3.8 V) [29,30]. The Poole–Frenkel emission comes from a lowering of the Coulomb potential barrier of a trap site due to electric field [38], meaning that the transport properties of the HRS in high field region is governed by electron hopping between the trap states probably existing on struc- tural defects such as Mn and/or oxygen vacancies. The further increase of the voltage up to∼3.8 V breaks the Poole–Frenkel emission mechanism and then re-establishes the conductive Ohmic behav- ior. As an another example, the similar switching of the transport mechanisms was observed in the

Fig. 7. ((a) and (b)) Distributions of log(RLRS), log(RHRS),VSET, andVRESETafter the oxygen annealing at 800°C, evaluated from the statistics on 10⁴consecutive RESET-SET cycles. ((c) and (d)) Distributions of log(^RLRS)^{, log}(^RHRS)^,VSET, andVRESETwithout the annealing, evaluated from the statistics on 10⁵consecutive RESET-SET cycles. The binning sizes of both log(^RLRS)^{and log}(^RHRS) are 0.2, while those ofVRESETandVSETare 0.05 V and 0.5 V, respectively [22].

Au/BaTi_0.95Co_0.05O₃/Pt unipolar RRAM cell [5]. During this mechanism switching, the hopping barrier and the related trapping states must be destroyed, and hence the local ion migration occurs during the switching from the HRS to the LRS.

According to the above-mentioned ion-migration RS mechanism, the density distribution of defect sites (including the non-stoichiometric anions and oxygen vacancies) and/or the existence of valence variable elements become the key factors to determine the RS performances. For example, the annealing of BaTi_0.95Co_0.05O₃film in the oxygen air at 800°C reduces the density and changes the distributions of defects and ion vacancies, debasing the RS performances of Au/BaTi_0.95Co_0.05O₃/Pt cell. As shown inFig. 7, the distributions ofV_set,V_reset,R_HRS, andR_LRSof the Au/BaTi_0.95Co_0.05O₃/Pt cell become broader after the oxygen air annealing at 800°C, which is unfavored.

2.2.2. Metal–insulator transition

It has been demonstrated that strongly electronic correlation effects are incorporated to the realm of the oxide RRAM devices [25]. The electronic charge injection can change the partial valence of anion and acts like doping to change the electronic density, and therefore induces MIT in perovskite oxides such as (Pr, Ca)MnO₃and SrTiO₃:Cr [39–41]. The Mott-type MIT, which is induced by the electronic charge injection, can be considered as another origin of resistive switching in perovskite oxides [3,25,42]. The reconstruction of the electronic phase separation (EPS) state, driven by electric field together with its parasitical Joule heating, can also produce the MIT and present the filament-type unipolar RS behavior [24,43].

For example, La_5/8−yPr_yCa_3/8MnO₃is a typical EPS system, in which the charge-ordered insulator (COI) phase coexists with the ferromagnetic metal (FMM) phase at nanometer or sub-micrometer scale. La_5/8−yPr_yCa_3/8MnO₃exhibits the first-order-like MIT associated with the transition between the FMM and COI phases, and displays the filament-type transport behavior due to the percolation properties [43]. In this system, the electric field together with its parasitical Joule heating contributes to the nonvolatile unipolar RS [44].

Fig. 8Presents the nonvolatile RS and magnetic switching of bulk La_0.225Pr_0.4Ca_0.375MnO₃at 10 K under different temperature circulations [44]. The cooling of the sample from the temperature above the COI transition (T >120 K) leads to a LRS with bigger magnetization, while the cooling from the temperature below the COI transition but above the MIT (90 K<^T <120 K) results in a HRS with smaller magnetization at 10 K. This is because the annealing at 90 K < ^T < 120 K favors for the

Fig. 8. ((a) and (b)) Temperature dependent resistanceR(T)at 10µA and magnetizationM(T)at 100 Oe, respectively, in the sequence from path (1) to (5), i.e.,T =300→10→130→10→200→10(^K). Inset: Time dependent resistance at T =10 K, withI =15 mA and 10 mA, respectively. The violet line is the measuring current [44]. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 9. Resistance–current (R–I) characteristics at the LRS, measured in four-probe configuration by the current sweeping in the stair mode with the rate 1.33 mA/s, 1 mA/s, and 0.2 mA/s, respectively [44].

COI phase and reconstructs the EPS with a totally reduced FMM fraction at 10 K. Due to the filament- type transport properties, the Joule heating near the local region of the conductive filaments can bring out local annealing process, which will similarly lead to reconstruction of the EPS state at local regions and display the unipolar RS behavior. For example, the unipolar RS was observed under the alternative applications of 100 and 15 mA pulses, as shown in the inset ofFig. 8(a).

Fig. 9Shows the resistance–current curves of La_0.225Pr_0.4Ca_0.375MnO₃ under different current sweeping rates [44] With increasing sweeping rate, the resistance peak (i.e. MIT peak) shifts to a higher current region, confirming the key role of the Joule heating in the reconstruction of the EPS in La_0.225Pr_0.4Ca_0.375MnO₃.

2.2.3. Charge trapping/de-trapping

Another filament-type RS mechanism is the carrier trapping and de-trapping at the defects inside the oxides or near the electrode/oxide interface, at which the active and positively charged trapping centers can be developed to trap electrons [45–50]. Different from the anion migration mechanism, application of external electrical voltage with different polarities causes the electrons to be trapped into or de-trapped from the trapping sites, which varies the local energy barrier and leads to the filamentary RS. For example, Fujii et al. [46] reports the colossal electroresistance effect and the absence of hysteresis in theC–Vrelationship in the heteroepitaxial SrRuO₃/^SrTi1−xNb_xO₃Schottky junctions. They suggest that the electron charging in or discharging from a self-trap depending on the bias polarity in the Schottky junctions lead to the resistance switching through additional tunneling paths rather than the change in barrier potential profile [46].

In this type RRAM cells, the distributions of the trapping sites and the trapping depth strongly influence the RS performances [45,46,51]. The detailed discussion on the carrier trapping/detrapping mechanism will present in Section3.1.

2.3. Memory performances

Owning to the abrupt capture/rupture of the nanometer-scale conductive filaments, the filament type RRAM cells have many merits of performances, such as the large off/on ratio, fast speed, and high density. Besides, by controlling the compliance current level, one RRAM cell has the ability to stay at multiple resistance states that can be used as multi-bit storage [52]. Additionally, for the ion migration related filament-type RRAM cells, the resistance states are stable and have good retentions.

Despite these merits of the performances, there is a serious drawback in the filament-type RRAM cells [53,54]. Due to the random distributions of defects and/or ion vacancies, and because of the in- completely reversible redox induced by the anion-migration or Joule heating, the formed filament is not disrupted by a constant voltage and at a fixed position during the endurance testing, which com- plicates the filament formation/rupture processes and causes the evolution of filamentary network in an un-controllable way. As a result, the RS performance is debased and the serious frustrations of the RS parameters, such asV_set,V_reset,R_HRS, andR_LRS, are usually observed in many filament-type RRAM cells [55–57]. Similar with other resistive switching oxides, the perovskite oxide based filament-type RRAM cells also exist these drawbacks in performance. However, the strong electronic correlation and the capability of accommodating non-stoichiometric ions in perovskite oxides provide the more possible ways to improve the switching mechanisms.

2.4. Performances enhancing methods

How to enhance the switching performances of the filament-type RRAM cells? The key answer exists in stabilizing the filament morphology and fine-control of the rupture positions. For this purpose, various methods, including the ion doping, defect and interface engineering, optimization of device structure, and operational and technological tactics, have been proposed and tried over the past decades [35,49]. For example, in the silicon oxide resistive random access memory device, the electroforming voltages of is significantly reduced more silicon oxide sidewall area is formed, and an external series resistance may be helpful enable device survival beyond 10⁷pulse cycles [58]. The ultraviolet illumination on the Pt/ZnO/Pt cell significantly reduces the variations of resistance in high resistance state, set voltage and reset voltage by 58%, 33%, and 25%, respectively [59].

2.4.1. Ion doping

It is well known that the ion doping (or self doping) for perovskite oxides not only impacts the concentration and mobility of the oxygen vacancies but also changes the carrier density and electronic phases, due to strongly electronic correlation [60–62]. Many experimental and first-principles studies have revealed that the ion doping can effectively enhances the RS performances [60]. For example, the Pr and Ca co-doped LaMnO₃shows the EPS phenomenon and enhances the colossal electroresistance (CER) and colossal magnetoresistance (CMR) effects [21]. In the Bi₁−xBa_xFeO₃−δ/SrTi_0.99Nb_0.01O₃

Similarly, the interface engineering can also alter the density and distribution of the defects and/or vary the charge trapping depth near the junction interface, improving the electrical migration ability of ions/carriers and therefore providing an effective approach toward enhancing the RS performances [65–67]. For example, the hand-made In/Nb:STO junction displays an Ohmic behavior while the In/Nb:STO junction fabricated by magnetic sputtering technology presents a Schottky emission behavior. [68]. Another example is that the spontaneously formed ZnWO_x bilayer in the ZnO/W interface via the interfacial engineering results in an asymmetricI–Vbehavior and leads to the enhancement of the RS performances [69]. It was reported that the heavily dopedn-type silicon nanotips modulate the number of defect states at the surface of La_0.7Sr_0.3MnO₃and as result enhances the off/on ratio [70].

2.4.3. Optimization of device structure

The optimization of device structure is another important approach to enhance the RS perfor- mances. Simultaneously the excellent RS uniformity, electroforming-free, and self-rectifying function- ality in the Pt/^Ta2O₅/^HfO2/TiN structure was reported by the structure optimization [49]. The off/on ratio, uniformity, and stability of the ZnO RS memory arrays by introducing ceria (CeO₂) quantum dots at the ZnO nano-rod surface was reported too [71]. The uniformity of Al/Cu_xO/Cu RS memory by in- troducing a thin phase-change Ge₂Sb₂Te₅(GST) layer at the Al/Cu_xO interface, as shown inFig. 10, was demonstrated [72]. The optimization of device structure not only engineers the interfacial properties but also optimizes and stabilizes the filamentary morphology.

2.4.4. Operational and technological tactics

As an electrical voltage applies to the RRAM cell with the HRS state, the electrode/insulator/

electrode structure also composes a parasitic capacitor. During the switching from the HRS to the LRS, additional current that comes from the discharging of parasitic capacitor will flow through the filament and influences the RS performance. Previous investigations revealed that the charging/

discharging of the parasitic capacitance and the direct stressing of voltage on the RRAM cell will dam- age the uniformity and reduce the lifetime of RRAM cells [58,73,74]. Hence, a specially designed circuit based on the RRAM cells is helpful for reducing these undesirable effects. For example, a serial connec- tion of a resistance to the RRAM cell can reduce the voltage stress on the cell and limit the maximum current flowing through the cell, which improves the RS performance and elongates the lifetime of cell [75,76].

Besides, phenomenological theory and analytical model are developed to describe the scaling be- haviors and the relationships between the RS parameters includingR_HRS,R_LRS, Reset current and volt- age, and Set voltage and compliance current [57,77,78]. By using a quantitative switching probability model, the key controlling parameters such as the Set and Reset voltage pulses can be dynamically estimated to switch the resistance state in the subsequent operation. This provides important guid- ance in yielding improvement and memory evaluation [67]. For instance, using an unified prediction model, a program to dynamically adjust the operation voltages, which enhances the endurance of the RS device, is available now [79].

Fig. 10. TheI–Vcurves of SET process in the (a) Al/CuxO/Cu and (b) Al/^GST/^CuxO/Cu devices. Statistical distributions of the (c)RLRSandRHRS, and (d)VsetandVreset[72].

3. Uniform resistance switching

Beside the filamentary RRAM cells, uniform-type RRAM cells have been developed, in which the variation of the cell resistance is uniform and usually realized by modulating the interfacial barrier of the heterojunction. The modulation mechanisms can be mainly classified into the carrier trapping/detrapping and the modulation of ferroelectric polarization, as addressed in the following.

3.1. Carrier trapping and detrapping

Both the migration of oxygen vacancies and the carrier trapping/detrapping, as discussed in Sec- tion2.1, can modulate the Schottky barrier at the local regions of the heterostructure and lead to the filamentary resistance switching. In fact, modulation of the Schottky barrier is possibly uniform at the junction interface, which strongly depends on the fabrication technology of the heterojunc- tion. The flat interface structure may possibly lead to the uniform switching, while the accidental interface structure may prefer the filamentary switching. Although the uniform motion of defects or oxygen vacancies is too slow to produce the nanosecond switches, as proposed [80], the charge trap- ping/detrapping near the junction interface can lead to uniform resistance switching, as verified in the Au/Nb:STO junction [11].

For example, we measured the interface resistance of the Au/Nb:STO junction using the three- probe measuring configuration. As shown inFig. 11(a), the rectifyingI–Vbehavior of the HRS shows the Schottky barrier at the junction, while the linearI–V behavior of the LRS indicates the collapse of the Schottky barrier, suggesting that the electric field modulates the Schottky barrier and as result leads to the RS. Besides, theR_HRSandR_LRSof the Au/DMO/Nb:STO/Au device decrease with increasing Au pad area, as shown inFig. 11(b), which indicates the good scaling behavior and suggests the uni- form rather than filamentary RS. The remarkable capacitance memory effect in the Nb:STO/Au junc- tion further confirms the uniform switching nature, since the junction capacitance depends on the mean Schottky barrier width but insensitive to the formation/rupture of local conductive filaments

Fig. 11. (a)I–Vmeabehaviors of the HRS and the LRS of the Au/Nb:STO junction, measured by using 3-probes configuration on the In/ Nb:STO /Au stack. Inset: Sqrt (−Vmea) dependent ln(−I)plot of the HRS. (b) The grounded Au pad area dependent resistances of the HRS and the LRS of the Au/DMO/Nb:STO/Au device respectively, with the configuration shown inFig. 1(a).

(c) Schematic diagrams for the detrapping and trapping electrons in the Nb:STO/Au junction. The electron trapping (detrapping) causes to the broadening (narrowing) of barrier and consequently forms to the HRS (LRS) respectively. (d)Vwrite-dependent resistance hysteresis of In/Nb:STO/Au, with Au deposited under the Ar pressure of 15 Pa, 20 Pa, and 30 Pa, and the depositing current of 4 mA and 2 mA, respectively [5].

[33,46]. Due to uniform modulation of the barrier width, the HRS associates with the low capacitance state (LCS) and the LRS associates with the high capacitance state (HCS) [45,68,81].

To verify that the migration of oxygen vacancies is not the dominate switching mechanism, dif- ferent Au/Nb:STO junctions are fabricated under several different Ar atmosphere pressures and de- positing currents. The variation of depositing condition changes the spatial defect density but scarcely alters the oxygen vacancy density near the junction interface due to the deposition under an atmo- sphere without oxygen ambient but with a small depositing current.Fig. 11(d) shows that the re- sistance hysteresis behavior of the Au/Nb:STO junction is sensitive to the Ar atmosphere pressures, indicating that the space defects other than the oxygen vacancies seriously affect the junction RS be- havior and the space defects related carrier trapping/detrapping near the interface dominate the uni- form RS behavior.

Fig. 11(c) Shows the schematic diagrams of the junction barrier for the electronic trapping and detrapping in the Nb:STO/Au junction. Application of a positive voltage (negative voltage applies to the Nb:STO and the Au electrode is grounded) lowers the barrier height of the Nb:STO/Au junction and some detrapped electrons are possibly flowing through toward the Au electrode, leaving the positively charged empty traps nearby the interface [45,68]. Such positively charged traps reduce the build-in potential and the depletion width, and thus lead to the LRS in order to maintain the Fermi energy balance between the Nb:STO and Au [81]. On the contrary, as a negative voltage is applied to the LRS, the junction barrier is raised and hinders the electrons to flow. However, the electrons are possibly injected into the empty traps by the negative voltage, which causes the traps to be neutral and then resumes the depletion width, i.e. the resume of the HRS. The electron trapping (detrapping) causes the broadening (narrowing) of the barrier and consequently the HRS (LRS) respectively [7,82–84].

Fig. 12. Schematic band diagrams for the electrode/ferroelectric/electrode heterostructure [85].

3.2. Modulation of ferroelectric polarization

The ferroelectric tunnel junctions (FTJs), consisting of two electrodes sandwiched with a ferroelec- tric tunnel barrier, are attracting great attention for the nonvolatile memory applications in recent years [85–90]. In the FTJs, the polarization at the ferroelectric/metal interface strongly modulates the junction barrier profile and varies the tunnel transmission, giving rise to giant tunnel electroresis- tance effect and opening the path to applications as high-density ferroelectric memories with non- destructive readout.

3.2.1. Modulation mechanism

At the ferroelectric/electrode interface, the presence of positively (negatively) polarized charges will attract (repel) electrons and thus form a negative (positive) charge-screening layer in the electrode. This effect will build up an additional positive (negative) electrostatic potential at the ferroelectric/electrode interface and thus modulate the junction barrier. The magnitude of additional electrostatic potential increases with the value ofλ/ε^{, here}λstands for the screening length andε the dielectric constant of the electrode [85]. The electronic band structure of the electrode/ultrathin- ferroelectric/electrode under the reversed polarization directions is shown inFig. 12. For discussion convenience, the value ofλ/εat the left electrode/ferroelectric interface is proposed to be smaller than that at the right interface. Thus, as the polarization points to the right, the increase of the barrier height at the left interface is smaller than the decrease of the barrier height at the right interface, which leads to the downward distortion of electronic band and the decrease of average barrier height (Φ⁻). As the polarization is reversed, the asymmetry of the electronic potential profile is reversed, which results in the upward distortion of electronic band and the increase of average barrier height (Φ+) on the contrary. Therefore, the direction of polarization modulates the junction barrier and leads to the bipolar resistance switching.

Many FTJs have been investigated and successful polarization modulation of the resistance states have been confirmed, including not only Co/^BaTiO3/La_0.7Sr_0.3MnO₃[91], Pt/BiFeO₃/SrRuO₃[92], PFM- tip/Pb(^Zr0.²Ti_0.8)^O3/^La0.⁷Sr_0.3MnO₃[86], AFM-tip/BaTiO₃/La_0.7Sr_0.3MnO₃[93] etc.

An interesting case is the Pt/BaTiO₃/Nb:STO FTJ, in which the nonvolatile capacitance and resis- tance memory were observed simultaneously [87]. Because the polarization not only modulates the barrier height but also varies the depletion width of the BaTiO₃/Nb:STO junction. Another interest- ing case is the La_0.7Sr_0.3MnO₃/BaTiO₃/La_0.5Ca_0.5MnO₃heterostructure [88], where the polarization of BaTiO₃changes the screening charges that result in the electronic or hole doping in La_0.5Ca_0.5MnO₃. Due to the strong phase competition between antiferromagnetic insulator (AFI) phase or ferromag- netic metal (FMM) phase, a slight electronic or hole doping in La_0.5Ca_0.5MnO₃ would change the ground state of La_0.5Ca_0.5MnO₃, i.e. the FMM or the AFI. Therefore, the polarization not only mod- ulates the FTJ barrier but also triggers the metal–insulator transition in La_0.5Ca_0.5MnO₃.

Fig. 13. (a) Contour plot of the electroresistance (ER) versus voltageVand ferroelectric film thicknessd. The transition regions between direct DT, FNT, TI are sketched by the thick dashed lines; (b) a cross section through the contour plot atV =1.5 V (dashed–dotted line) andV= −0.2 V (solid and dashed line), i.e., the thickness dependence of the ER [98].

In addition, except for the junction barrier modulation mechanism, the polarization switching induced modulation of charges as a function of the ferroelectric domain wall geometry can also lead to the resistive switching in ferroelectric film [94]. Both experimental measurements and theoretical analysis reveal the existence of ‘strongly’ charged ferroelectric domain walls that break polarization continuity, but are stable and conduct steadily through a quasi-two-dimensional electron gas [95].

The carrier density and the conductivity of the domain wall are tunable by many orders of magnitude through the tilt of domain wall [96]. Under the application of an electric field, the nucleation and growth of ferroelectric domains change the domain walls geometry of ferroelectric film, and as a result modulate the conductivity of the ferroelectric film [93,97].

3.2.2. Transport mechanisms

The transport mechanism is also a critical issue for the PS behavior. As proposed, there are three possible mechanisms responsible for the current through ultrathin ferroelectrics: direct tunneling (DT), Fowler–Nordheim tunneling (FNT), and thermionic emission (TI), as shown inFig. 13[98]. For the ultrathin ferroelectrics (thicknessd <4 nm), the transport property is dominated by the direct tunneling at low bias voltage while probably controlled by the Fowler–Nordheim tunneling at high bias voltage. As the ferroelectric barrier thickness increases, the direct tunneling current decreases exponentially while the thermionic emission mechanism dominates the transport behavior.

3.2.3. Coexistence of the multiple RS mechanisms

For realistic FTJs, in particular for those FTJs including transition metal oxides as the tunneling layers, it seems that no a single but general mechanism alone can be responsible for all observed phe- nomena. The modulation of electronic band may be quite complex. Besides the polarization switching,

80 60 40 20 0 10¹¹ 10¹⁰ 10⁹ 10⁸ 10⁷ 10⁶

200 250 300 350 400

9

6 3

45 30 15 0 Temperature (K)

Resistance (Ω)Pr(C/cm2) Dfp biRH/RL

a

b

Fig. 14. Temperature dependentPr,RH,RL, andRH/RLof Au/BiFeO3/SrRuO3[14].

some other mechanisms such as the carrier trapping/detrapping at the defect sites and/or the migra- tion of oxygen vacancies near the electrode/ferroelectric interface may take part in the RS process simultaneously via distorting the junction barrier and inducing the bipolar RS behavior [14,89]. For instance, the bipolar resistance switching in Au/BiFeO₃/SrRuO₃structure was evidenced to come from more than one switching mechanism, as discussed below.

Fig. 14(a) and (b) shows the temperature dependent remnant polarization (P_r), resistances of the HRS and LRS, and off/on ratio (R_H/^RL) of Au/BiFeO₃/SrRuO₃structure respectively [14] TheP_r is ex- tracted from the Polarization-Voltage data measured under the fixed voltage cycle of 0 →V_max →

−V_max →0 at different temperature. Because the coercive field increases with decreasing the tem- perature and the sample is polarized under a same maximum voltage at different temperatures,P_r decreases with decreasing temperature. But the off/on ratio has a maximum value at∼323 K. This in- dicates that in addition to the polarization modulation mechanism, other mechanisms such as the oxy- gen vacancies migration and/or the carrier trapping/detrapping mechanisms play an important role in the RS at high temperature. If the polarization switching dominates the RS behavior, theP_rand the off/on ratio should have the similar temperature dependences because of the relationship ofR_H/^RL∝ 1Φbi∝_{Pr, where1Φbi}is the variation of build-in potential induced by the polarization [99,100].

Another evidence is the sweep-rate dependentI–Vcurves at different temperatures, as shown in Fig. 15[14]. At low temperature (e.g. 173 K), theI–V curve rotates clockwise and the current peak distinctly increases with the sweep-rate, which is the typical consequence of charging/discharging behavior induced by the polarization switching. At high temperature (e.g. 373 K), however, theI–V curve rotates anticlockwise and the peak current distinctly decreases with the sweep-rate on the contrary, which indicates that some slow mechanisms dominate the RS behavior at high temperature.

Therefore, in this case, the oxygen vacancies migration and/or the carrier trapping/detrapping mechanisms possibly play an important role in the RS behavior especially at high temperature, although the polarization modulation mechanism is substantial too.

3.3. Memory performance

Comparing with the filamentary RRAM cells, the uniform-type RRAM cells show the promis- ing properties including the stable off/on ratio, multi-bit storage, good endurance, and high uni- form operation voltages. Besides, the uniform-type RRAM cells have great potential applications as

Fig. 15. Sweep-rate dependentI–Vcurves of Au/BiFeO3/SrRuO3at several temperatures [14].

memcapacitive devices. For example, the Au/^DyMnO3/Nb:STO/Au stack is an uniform-type RRAM cell, which shows the high RS performance, as shown inFig. 16[11]. The nesting resistance hys- teresis inFig. 16(b) indicates the existence of intermediate states, which can be used as multi-bit storage. A voltage pulse train with different magnitudes and pulse numbers can be used to control these intermediate resistance states, as shown inFig. 16(c). The capacitance hysteresis inFig. 16(d) indicates the nonvolatile capacitance memory with a variation ratio of 1000. Another example is the Pt/BaTiO₃/Nb:STO FTJ [87]. The FTJ has an off/on resistance ratio above 10⁴, an retention longer than 10 years, an endurance more than 10⁴cycles, and also owns the capacitance memory.

The ferroelectric polarization switching has very stable and uniform switching performances. Fer- roelectric random access memory (FeRAMs) based on thick ferroelectric films are shown to have very high endurance (10¹⁴cycles){Park,1999#13 553}. Hence, the right combination of ferroelectric/

electrode materials is very possible to reach this goal in FTJs. Many ferroelectric oxides are found in the perovskite oxides. Therefore, the perovskite oxides provide good platforms to insight into the re- sistive switching mechanisms and to achieve the competitive advantage in the resistive switching performance.

Despite these merits of memory performance, there are obvious deficiencies that hinder actual applications of the uniform-type RRAM devices. For the carrier trapping/detrapping type RRAM cells, the HRS and the LRS usually have remarkable relaxations, which greatly worsen the retention of the RRAM cell [11]. To overcome this shortcoming, an effective interfacial engineering will be necessary to control the trapping depth near the junction interfaces. For the FTJ cells, the polarization modulated tunneling current is very small and the investigations on the endurance are still needed for supporting any trial to actual applications. To resolve these problems, substantial effort on the highly qualified ultrathin ferroelectric films is needed.

4. Perspectives

Despite a great of progress in understanding the RS mechanisms has been made and many merits of memory performances have been revealed in the filament-type and uniform-type RRAM cells, there is high potential for improvement both from technological and fundamental perspectives.

Fig. 16. Resistance and capacitance switching performances of the Au/DyMnO3/Nb:STO/Au stack at room temperature.

(a) Upper panel: Schematic of the stack and the measurement configuration. Lower panel: Schematic of the measuring method for the resistive hysteresis. (b)Vwrite-dependent resistive hysteresis, withVwrite: 0→ −10 V→Vmax→0 and readout at 0.1 V.

(c)Vwritepulse number dependent resistive switching from the LRS, with fixed pulse width of 20 ms andVwrite =4.^{5, 5, 5.5,} and 6.5 (V) respectively, readout at 0.1 V. (d)Vwrite-dependent capacitive hysteresis between−10 and 10 V, readout at 50 kHz and 50 mV AC voltage. (e) Retention data of the HRS and the LRS respectively, readout at 0.1 V. (f) Time dependent resistance before and after the stressing of±9.6 V pulses with width of∼8 ns, readout at 0.1 V. (g) Endurance data for 10⁸consecutive switching cycles, operated by±10 V pulses (100 ms-width, compliance current: 10 mA) and readout at 0.5 V [11].

For the filament-type RRAM cells, the fine control of the filament morphology becomes the key point to enhance the RS performance. To reduce the randomness of filament formation, the ion doping, interface engineering, and/or the structure design and optimization in the nanometer scale perovskite oxides are still very promising research methods. The electrically modulated electronic phase transition together with the electrochemistry and ion migration in the nanometer scale perovskite oxides provides a good platform to investigate the effective ways for enhancing the RS performances.

Besides, investigations on how the pining layer in the electrode/ferroelectric junctions affects the RS performances of FTJs are still scarce. For thick ferroelectric films, the endurance of the polarization switching has reached up to 10¹⁴cycles, while the reported endurance of the RRAM cells based on FTJs is limited to 10⁶cycles [87,101]. It is known that defects including the oxygen vacancies near the electrode/ferroelectric interface will form a pining layer, which will degrade the endurance of the polarization switching. However, the defects can be also electrically migrated and/or become the charges trapping centers, which will present other mechanisms for stable bipolar resistance switching. Therefore, it is very valuable to comprehensively understand how the interfacial defects of electrode/ferroelectric junction affect the performances of FTJ RRAM cells.

In addition, Sluka et al. reported that quasi-two-dimensional electron gas exists in the 90°^head- to-head charged domain walls (CDW) of ferroelectric BaTiO₃, which has a steady conductivity∼₁₀⁹ times higher than the bulk DC conductivity of BaTiO₃[95]. Thus, the current in FTJs may possibly flow through the CDW rather than direct tunneling through the barrier. Furthermore, the CDW can be electrically displaced, erased, and recreated in the ferroelectric films [95]. Hence, via a special CDW engineering in ferroelectric films, the nonvolatile RS in FTJs may be realized by electrical modulation of the CDW. By adopting this switching mechanism in FTJs, the On and Off currents will be much higher than the tunneling currents and the very small thickness of the ferroelectric films may be no longer a necessity, which will greatly benefit to real RRAM applications and deserve for future investigation.

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

[6]D.-S. Ko, S.-I. Kim, T.-Y. Ahn, S.-D. Kim, Y.-H. Oh, Y.-W. Kim, Appl. Phys. Lett. 101 (2012) 053502.

[7]A. Sawa, T. Fujii, M. Kawasaki, Y. Tokura, Appl. Phys. Lett. 85 (2004) 4073.

[8]X. Wang, W. Xie, J.-B. Xu, Adv. Mater. (2014).

[9]X.A. Tran, W.G. Zhu, B. Gao, J.F. Kang, W.J. Liu, Z. Fang, Z.R. Wang, Y.C. Yeo, B.Y. Nguyen, M.F. Li, H.Y. Yu, IEEE Electron Device Lett. 33 (2012) 585–587.

[10]IEEE Electron Device Lett. 34 (2013).

[11]Z.B. Yan, J.M. Liu, Sci. Rep. 3 (2013) 2482.

[12]Z. Wen, L. You, J. Wang, A. Li, D. Wu, Appl. Phys. Lett. 103 (2013) 132913.

[13]Appl. Phys. Lett. (2014).

[14]Y.B. Lin, Z.B. Yan, X.B. Lu, Z.X. Lu, M. Zeng, Y. Chen, X.S. Gao, J.G. Wan, J.Y. Dai, J.M. Liu, Appl. Phys. Lett. 104 (2014) 143503.

[15]D.S. Jeong, R. Thomas, R.S. Katiyar, J.F. Scott, H. Kohlstedt, A. Petraru, C.S. Hwang, Rep. Progr. Phys. 75 (2012) 076502.

[16]A. Sawa, Mater. Today 11 (2008) 28–36.

[17]S. Menzel, M. Waters, A. Marchewka, U. Böttger, R. Dittmann, R. Waser, Adv. Funct. Mater. 21 (2011) 4487–4492.

[18]J.C. Hou, S.S. Nonnenmann, W. Qin, D.A. Bonnell, Adv. Funct. Mater. 24 (2014) 4113–4118.

[19]Y.L. Jin, K.J. Jin, C. Ge, H.B. Lu, G.Z. Yang, Modern Phys. Lett. B 27 (2013) 13.

[20]R. Waser, R. Dittmann, G. Staikov, K. Szot, Adv. Mater. 21 (2009) 2632–2663.

[21]L. Gorkov, V. Kresin, Phys. Rep. 400 (2004) 149–208.

[22]Z.B. Yan, S.Z. Li, K.F. Wang, J.-M. Liu, Appl. Phys. Lett. 96 (2010) 012103.

[23]U. Celano, L. Goux, A. Belmonte, K. Opsomer, A. Franquet, A. Schulze, C. Detavernier, O. Richard, H. Bender, M. Jurczak, W. Vandervorst, Nano Lett. (2014).

[24]Z.B. Yan, K.F. Wang, S.Z. Li, S.J. Luo, J.M. Liu, Appl. Phys. Lett. 95 (2009) 143502.

[25]H.S. Lee, S.G. Choi, H.H. Park, M.J. Rozenberg, Sci. Rep. 3 (2013) 1704.

[26]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. 5 (2010) 148–153.

[27]E. Strelcov, Y. Kim, S. Jesse, Y. Cao, I.N. Ivanov, I.I. Kravchenko, C.H. Wang, Y.C. Teng, L.Q. Chen, Y.H. Chu, S.V. Kalinin, Nano Lett. 13 (2013) 3455–3462.

[28]S. Kim, S.J. Kim, K.M. Kim, S.R. Lee, M. Chang, E. Cho, Y.B. Kim, C.J. Kim, U. In Chung, I.K. Yoo, Sci. Rep. 3 (2013) 1680.

[29]Y. Yang, P. Gao, S. Gaba, T. Chang, X. Pan, W. Lu, Nature Commun. 3 (2012) 732.

[30]S. Larentis, F. Nardi, S. Balatti, D.C. Gilmer, D. Ielmini, IEEE Trans. Electron Devices 59 (2012) 2468–2475.

[31]K. Nagashima, T. Yanagida, M. Kanai, U. Celano, S. Rahong, G. Meng, F. Zhuge, Y. He, B. Ho Park, T. Kawai, Appl. Phys. Lett.

103 (2013) 173506.

[32]D.S. Shang, J.R. Sun, L. Shi, B.G. Shen, Appl. Phys. Lett. 93 (2008) 102106.

[33]E. Lee, M. Gwon, D.-W. Kim, H. Kim, Appl. Phys. Lett. 98 (2011) 132905.

[34]D.S. Shang, J.R. Sun, L. Shi, Z.H. Wang, B.G. Shen, Appl. Phys. Lett. 93 (2008) 172119.

[35]F. Pan, S. Gao, C. Chen, C. Song, F. Zeng, Mater. Sci. Eng. R 83 (2014) 1–59.

[36]S.B. Lee, S.C. Chae, S.H. Chang, T.W. Noh, Appl. Phys. Lett. 94 (2009) 173504.

[37]J.P. Strachan, D.B. Strukov, J. Borghetti, J. Joshua Yang, G. Medeiros-Ribeiro, R. Stanley Williams, Nanotechnology 22 (2011) 254015.

[38]J.S. Choi, J.S. Kim, I.R. Hwang, S.H. Hong, S.H. Jeon, S.O. Kang, B.H. Park, D.C. Kim, M.J. Lee, S. Seo, Appl. Phys. Lett. 95 (2009) 022109.

[39]D. Kim, Y. Kim, C. Lee, Phys. Rev. B 74 (2006) 174430.

[40]G.I. Meijer, U. Staub, M. Janousch, S.L. Johnson, B. Delley, T. Neisius, Phys. Rev. B 72 (2005) 155102.

[41]A. Asamitsu, Y. Tomioka, H. Kuwahara, Y. Tokura, Nature 388 (1997) 50–52.

[42]R. Fors, S. Khartsev, A. Grishin, Phys. Rev. B 71 (2005) 045305.

[43]H. Sakai, Y. Tokura, Appl. Phys. Lett. 92 (2008) 102514.

[44]Z.B. Yan, K.F. Wang, S.Z. Li, S.J. Luo, J.M. Liu, Appl. Phys. Lett. 95 (2009).

[45]X.G. Chen, X.B. Ma, Y.B. Yang, L.P. Chen, G.C. Xiong, G.J. Lian, Y.C. Yang, J.B. Yang, Appl. Phys. Lett. 98 (2011) 122102.

[46]T. Fujii, M. Kawasaki, A. Sawa, Y. Kawazoe, H. Akoh, Y. Tokura, Phys. Rev. B 75 (2007) 165101.

[47]B. Cheng, Z. Ouyang, C. Chen, Y. Xiao, S. Lei, Sci. Rep. 3 (2013).

[48]Y. Shuai, S. Zhou, D. Bürger, M. Helm, H. Schmidt, J. Appl. Phys. 109 (2011) 124117.

[49]J.H. Yoon, S.J. Song, I.-H. Yoo, J.Y. Seok, K.J. Yoon, D.E. Kwon, T.H. Park, C.S. Hwang, Adv. Funct. Mater. (2014).

[50]J.H. Hur, M.-J. Lee, C.B. Lee, Y.-B. Kim, C.-J. Kim, Phys. Rev. B 82 (2010) 155321.

[51]X.B. Yan, Y.D. Xia, H.N. Xu, X. Gao, H.T. Li, R. Li, J. Yin, Z.G. Liu, Appl. Phys. Lett. 97 (2010) 112101–112103.

[52]X. Yang, A.B.K. Chen, B. Joon Choi, I.W. Chen, Appl. Phys. Lett. 102 (2013) 043502.

[53]D. Ielmini, IEEE Electron Device Lett. 31 (2010) 552–554.