Exploring Electron Transport and Memristive Switching in Nanoscale Au/WOx/W Multijunctions Based on Anodically Oxidized Al/W Metal Layers

FULL PAPER (1 of 12) 1600512 wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Exploring Electron Transport and Memristive Switching in Nanoscale Au/WO x /W Multijunctions Based on Anodically Oxidized Al/W Metal Layers Maria Bendova , Jaromir Hubalek , and Alexander Mozalev * Dr. M. Bendova, Dr. J. Hubalek, Dr. A. Mozalev CEITEC – Central European Institute of Technology Brno University of Technology Purkynova 123 , Brno 61200 , Czech Republic E-mail: alexander[email protected].cz DOI: 10.1002/admi.201600512 size effect. [ 1–3 ] Eventual functionalization of WO x nanostructures through modifying the surface and interfacial properties by combining, doping, or decorating the pure oxide with other nanomaterials creates additional metal/semiconductor or semiconductor/semiconductor interfaces and grain boundaries capable of further infl uencing the sensing, electrical, optical, and opto-electronic properties or even introducing novel unique functionalities. [ 4–6 ] Contrarily to WO x nanostructures with no directional orientation of its morphological units, like disordered 1D nanowires or nanorods grown by vaporand liquid-phase methods, [ 1,7 ] oriented 1D WO x nanostructures, self-organized and upright-standing on a conducting substrate, without mutually intersecting with each other are expected to bring more advantages for practical applications in nanoscale electrochromic, [ 8,9 ] fi eldemission, [ 10 ] gas-sensing, [ 11 ] and photonic devices. Moreover, substrate-supported 1D nanostructures assembled in an array would enable the use of their integral properties, especially electrical, optical, or mechanical in contrast with the formation and application of single nanowires relying on their individual characteristics. [ 12,13 ] Synthesis of WO x nanostructures regularly aligned on substrates has been an active and competitive area of nanoscience and nanotechnology as the arrays of 1D structures often suffer from mechanical weakness and randomly dispersed physical contacts between neighboring units, limiting their performance in a device. Besides, creating welldefi ned nanoscale metal/semiconductor electrical and thermal interfaces and boundaries by contacting the tops of 1D nanostructures assembled in an array in a reliable and reproducible manner to benefi t from the characteristics measured along the nanostructures and across relevant interfaces has been a challenge. [ 8–11,14–17 ] Surface engineering of aluminum [ 18 ] and porous anodic alumina (PAA) fi lms grown by anodic oxidation (anodizing) of aluminum have been of increasing interest for lithography-free templating of metals, dielectrics, and semiconductors. [ 19–22 ] PAA-assisted anodization [ 23 ] of certain valve metals sputtered on substrates, including tungsten, [ 24–26 ] has proven its potential as a reproducible method for electrochemically forming self-organized vertically aligned spatially separated 1D nanostructures (nanocolumns, rods, or tubes), being nearly ideally electrically and mechanically bonded at their bottoms either to the substrate metal [ 26 ] or to a continuous solid layer of similar metal oxide that grows at the columns/substrate-metal interface. [ 24,25 ] On the way An array of semiconducting tungsten-oxide (WO x ) nanorods, 100 nm wide and 700 nm long, is synthesized via the porous-anodic-alumina-assisted anodization of tungsten on a substrate and is modifi ed by annealing in air and vacuum. The rods buried in the alumina nanopores are self-anchored to the tungsten layer while their tops are interconnected via gold electrodeposited inside and over the pores. Thus formed metal/semiconductor/metal microdevices are used for studying electron transport within the nanorods and across the multiplied nanoscale Au/WO x and W/WO x interfaces. The dominating effect of a Schottky junction that forms at the Au/WO x interface is justifi ed for the as-anodized and air-annealed nanorods tested at room temperature, which transforms into an ohmic contact at elevated temperature, whereas the bottom W/WO x interface turns out to be Schottky-like and govern the electron transport, giving a higher barrier and a set of pronounced diode-like characteristics in the as-anodized nanoarrays. The amorphous nanorods reveal bipolar resistive switching with a gradual reset due to the fi eld-driven movement of oxygen vacancies and induced modifi cations of the Au/WO x Schottky interface. The unique electrical and interfacial properties of the nanoscale Au/WO x /W multijunctions form a basis for their application in emerging resistive random access memories or 3D gas-sensing nanodevices. 1. Introduction Nowadays, tungsten trioxide in its stoichiometric and sub-stoichiometric forms (hereafter WO x ) is gaining much attention due to useful inherent properties, including electrochromic, photocatalytic, sensing, and photoelectrochemical, making it suitable for diverse potential applications. [ 1 ] Nanostructuring of WO x in the size order of about 100 nm is expected to substantially enhance or modify its surface properties due to the increased surface-tovolume ratio, altered surface energies, and the weak quantum Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.dewww.MaterialsViews.com This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. The copyright line of this paper was changed 2 November 2016 after initial publication. FULL PAPER 1600512 (2 of 12) wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim of utilizing the PAA-assisted metal-oxide nanostructures and relevant electrical interfaces in emerging microdevices, we have recently incorporated arrays of PAA-assisted niobium-oxide nanocolumns [ 27 ] into a 3D architecture, in which a microstructured top Au electrode was formed onto the column tops, effectively making use of the whole length of all nanocolumns and involving nearly 100% of the upper metal/oxide interfaces in gas sensing. [ 28 ] In the present work, we synthesized an array of WO x nanorods via a PAA-assisted anodization of thin tungsten layer in which, in contrast with the earlier works [ 24,25 ] the rods are well separated at their bottoms by the substrate metal and are not released from the PAA overlayer. In order to make contacts to tops of the nanorods and create patterned arrays of electrically active metal/semiconductor interfaces, we employed, for the fi rst time, point electrodeposition of gold through a photoresist mask into the alumina nanopores onto the WO x nanorods and also over the alumina surface. To explore electrical transport properties and interfacial phenomena in thus formed metal/ semiconductor/metal (MSM) microdevices, we carried out a deep investigation of electron transport along the WO x nanorods in an array and across the functional Au/WO x (upper) and W/WO x (bottom) interfaces at room and elevated temperatures, additionally modifying the fi lm morphology, crystal structure, and chemical composition by annealing in air and vacuum. This allowed us to identify the transport-limiting Schottky junctions and their contribution to the integral transport processes at various operating temperatures and correlate the parameters of a depletion layer that forms at the Au/WO x interface with the crystal structures of differently prepared and annealed WO x semiconductor nanorods incorporated in the PAA matrix. The fi ndings of this work helped experimentally justify the previously developed models of cooperative ionic transport through the respective interfaces during the fi lm growth [ 24,26 ] and formulate the ideas for use of this advanced material in nanoscale electronic, sensing, and photoelectrochemical emerging devices. 2. Results and Discussion 2.1. Film Morphology, Structure, and Composition Arrays of metal-substrate-supported WO x nanorods incorporated in a layer of nanoporous alumina were synthesized via so-called PAA-assisted anodization of a layer of tungsten sputtered on a substrate generally following the approach developed in our previous work [ 26 ] and adjusted here to meet the objectives of the present study ( Figure 1 ). An Al/W bilayer sputtered on a dielectric substrate is fi rst anodized to grow a PAA layer with well-ordered nanopores from the whole aluminum thickness (Figure 1 b), this being followed by growth of WO x nanoprotrusions into the pores (Figure 1 c). Further re-anodizing at certain conditions results in the growth of WO x at both the WO x /W and electrolyte/WO x interfaces, as shown in Figure 1 d, thus giving an array of pore-directed WO x nanorods anchored to the substrate metal (electron conductor) and deeply incorporated in the PAA layer (hereafter “PAA-inbuilt” nanorods). The situation when the tungsten-oxide nanorods grow directly from the tungsten metal without being merged in a metal-oxide layer at their bottoms has resulted from the set of technological, electrical, and electrolytic conditions developed in the previous work. [ 26 ] Scanning electron microscopy (SEM) images showing 3D views of the edges of samples having the PAA-inbuilt and PAAfree (alumina dissolved away) nanoarrays are displayed in Figure 2 a,b, respectively. SEM images of cross fractures of the PAA-inbuilt nanoarrays accompanied by schematic drawings noticing the differences at the WO x /W interfaces caused by the annealing conditions are shown in Figure 2 c,e while the surfaces of the same PAA-free samples are shown in Figure 2 d,f. In the images of PAA-inbuilt nanoarrays, some of the columns look disrupted across the fi lm fracture, which is an artefact of the breaking. The smooth outer surface of the PAA-free nanorods is justifi ed in Figure 2 b. As the fi rst step in studying functional properties of the PAA-inbuilt WO x nanoarrays, the as-anodized, air-, and vacuum-annealed samples were characterized by means of electrical measurements in a solid-state MSM assembly. To this end, a layer of gold contacting the tops of the rods was formed via an original point electrodeposition technique, as depicted in Figure 1 e, and served as an upper electrode while the tungsten underlayer was used as the bottom electrode. From the previous SEM and X-ray diffraction (XRD) results, [ 26 ] the as-anodized amorphous oxide nanorods crystallize to WO 2.9 and a mixture of WO 3 and WO 2.9 phases due to the 550 °C annealing in vacuum and in air, respectively. During the heating in vacuum, the oxygen anions composing the rods are partly released into the gas phase [ 29 ] leaving an increased concentration of oxygen vacancies in the rods, this being accompanied by the formation of WO 2.9 phase. In case of the air-annealed nanorods, although the two phases are distributed along the whole length of the rods, [ 26 ] the WO 3 phase is likely to dominate at their tops while a higher concentration of WO 2.9 is expected in the vicinity of the bottom WO x /W interface. The quantitative XRD analysis has shown that the WO 3 Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.de www.MaterialsViews.com Figure 1. Process steps in fabricating a metal/semiconductor/metal (MSM) microdevice incorporating porous-anodic-alumina (PAA)-inbuilt WO x nanorods: a) sputter-deposition of an Al/W bilayer onto a SiO 2 /Si wafer, b) porous anodization of the Al layer, c) anodization of the W layer through the PAA nanopores, d) re-anodization of the W layer for forming PAA-inbuilt WO x nanorods, followed by annealing in air or vacuum, and e) point electrodeposition of gold for forming top electrical contacts. FULL PAPER (3 of 12) 1600512 © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim wileyonlinelibrary.com Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.dewww.MaterialsViews.com Figure 2. SEM 3D views of an edge of the sample having a) PAA-inbuilt and b) PAA-free (alumina dissolved away) WO x nanoarray; SEM images of c,e,g) cross-fractures and d,f) surfaces complemented by schematic representations of the PAA-inbuilt and PAA-free WO x nanorods annealed in vacuum (10 −4 Pa) and air at 550 °C showing relevant changes in fi lm morphology and crystal structure; h) experimental and fi tted W 4f XP core-level spectra of the air-annealed PAA-partly etched nanorods as SEM-viewed and depicted in panel (g). FULL PAPER 1600512 (4 of 12) wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim phase takes up about 70% of the rods’ volume above the original W/Al interface. [ 26 ] This can be attributed to the annealing in oxygen-rich atmosphere leading to diffusion of oxygen into the rods, establishing an equilibrium concentration of oxygen vacancies, [ 29 ] and contributing to the expansion of bottom parts of the rods (Figure 2 e). Concurrently, the diffusion of oxygen ions from the bottoms of rods into the metal substrate creates a higher concentration of oxygen vacancies there, which fi nally results in transformation into WO 2.9 phase near the oxide/ metal interface. Chemical composition of tops of the air-annealed nanorods was further confi rmed by X-ray photoelectron spectroscopy (XPS) analysis performed for the sample with partly dissolved PAA layer, as SEM imaged and depicted in Figure 2 g. The presence of carbon, tungsten, oxygen, aluminum, phosphorus, and silicon species was identifi ed in the survey surface spectrum. All binding energies were referenced to the C 1s at 285.0 eV generated by photoelectrons emitted from carbon atoms in C  H bonds. The peak-fi tting of C 1s spectra was done as described in a previous work. [ 27 ] Since the dominating contribution to oxygen and aluminum signals was obviously from the PAA surface, we did not analyze the O 1s and Al 2p narro w-scan spectra. As the C 1s, P 1s, and Si 2p spectra showed no peak already after the fi rst Ar-ion sputtercleaning cycle, we conclude that the carbon, phosphorus, and silicon species are present only in the outermost layer of organic contamination and are not incorporated in the fi lm material. Thus, we analyzed the W 4f core-level spectrum (Figure 2 h) to gain insight into chemical bonds of tungsten composing the tops of the nanorods. Three doublet peaks (W 4f 7/2 and W 4f 5/2 ) are used to reproduce the spectrum in Figure 2 h. The W 4f 5/2 peak for each species is constrained to be at a fi xed energy increment of 2.15 eV above the W 4f 7/2 peak, and the W 4f 7/2 /W 4f 5/2 peak intensity ratio is fi xed to 4:3, with full width at half maximum of 1.0 for W 6+ and 1.3 for W (6− n )+ species. The highest-intensity doublet at 35.15 eV (W 4f 7/2 ) is associated with photo electrons emitted from W 6+ ions (dominating amount). The lowest-energy doublet at 34.2 eV (W 4f 7/2 ), shifted to only −0.95 eV from the main component, may be associated with photoelectrons emitted from W 5+ (6 at%). The highest-energy low-intensity doublet at 35.9 eV (W 4f 7/2 ) can be assigned to hydroxyl groups bonded with tungsten, W  (OH) 2 . [ 26 ] The presence of W 5+ ions in the monoclinic WO 3 phase at the tops of the rods is due to a certain concentration of oxygen vacancies, capable of altering electrical properties of the MSM micro devices, as will be discussed in Section 2.2 . For statistical electrical measurements of the PAA-inbuilt WO x nanoarrays, multiplied square-shaped gold electrodes of sizes ≈100 µm × 100 µm were electrodeposited through a patterned layer of photoresist, as shown in Figure 1 e. Figure 3 shows SEM images of an air-annealed sample at various stages of assembling the MSM microdevices. At the commencement of electrodeposition, the gold ions discharge on the tops of the rods, [ 30 ] and an array of gold nanocolumns grows within the alumina pores to reach the surface of the PAA fi lm. When the gold nanocolumns come out from the pores, they widen and soon merge to a complete, pancakelike solid layer over the general level of the alumina surface (Figure 3 ). The gold layer spread over the alumina surface creates an array of nanoscale Au/WO x junctions as well as forms micropads in the MSM microcells. Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.de www.MaterialsViews.com Figure 3. SEM views showing various process stages in micromachining PAA-inbuilt MSM devices: a,c,d) panoramic views of multiplied top Au electrodes of MSM microdevices, b) surface views of a window in the photoresist mask showing the development of a top Au electrode over the PAA surface: 1 ) beginning of Au electrodeposition, 2 ) incomplete Au deposition, 3 ) complete Au deposition; e–g) fragments of cross fractures of a PAAinbuilt nanorod array with and without a top Au electrode. FULL PAPER (5 of 12) 1600512 wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 2.2. Electrical Properties To determine basic parameters of the MSM microdevices and to model their behavior with an electrical circuit, current–voltage characteristics ( I ( V ) curves) were measured and presented as a dependence between the electric current passing through the device and the corresponding voltage applied across it. Three types of MSM microcells incorporating either the as-anodized, airor vacuum-annealed WO x nanoarrays (shown in Figures 1 e and 3 ) were measured by polarizing the top Au electrode while the tungsten underlayer was always grounded. Figure 4 a compares typical I ( V ) characteristics recorded at room temperature (RT) for the three types of MSM microcells, which are presented in logarithmic scale to better visualize the differences between the devices. As the V and I are not linearly proportional, the differential resistance, R dif (d V /d I ), was calculated and plotted against the applied voltage as shown in Figure 4 b. Additionally, the same data are presented as linear I ( V ) and R dif ( V ) dependencies (Figure 4 c–e) to emphasize the asymmetric behavior of the as-anodized and air-annealed devices. One may see that the current rises at a progressively increasing rate in the voltage range of 0 to ±2 V, being asymmetric at potentials over ±1.0 V and increasing more sharply at positive polarization of the top electrode. A hysteresis is also observed at the positive biases over 1.0 V, as indicated in Figure 4 by the arrows. In addition, the as-anodized and air-annealed samples when polarized at potentials over ±1.0 V show a systematic decrease in the resistance with consecutive measurement cycles. Therefore, only the initial measurements are shown in Figure 4 . Notably, the resistance of the oxide nanoarrays decreases signifi cantly from about 10 4 to 10 −3 Ω cm 2 in the following sequence: the as-anodized → air-annealed → vacuum-annealed samples. The asymmetric I ( V ) behaviors revealed for the as-anodized and air-annealed fi lms are likely due to the formation of a rectifying Schottky junction at some of the WO x /metal interfaces. [ 31,32 ] The directionality of thermionic current may provide an insight into the metal/semiconductor contact exhibiting the rectifying behavior. [ 32 ] In the present case of RT measurements, as the forward direction corresponds to the positive polarization of the top Au electrode, the Schottky-like barrier, being the limiting electron transport feature in the system, forms at the Au/WO x interface while the W/WO x contact is ohmic. This is in agreement with the reported values of work functions and electron affi nities of the materials involved (5.4 eV for Au, [ 33 ] 4.8 eV for WO 3 , [ 34 ] 4.3–4.6 eV for W [ 33 ] ). Impedance spectra recorded for the as-anodized and airannealed devices may further confi rm the presence of a Schottky junction. Figure 5 shows a typical impedance behavior Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.dewww.MaterialsViews.com Figure 4. Experimental I ( V ) and calculated R dif ( V ) curves recorded at room temperature (RT) for the as-anodized, air-annealed, and vacuum-annealed MSM microdevices, shown as a,b) semi-logarithmic and c–e) linear scale representations for each device type. The arrows point out the direction of voltage sweep under positive bias of the top Au electrode. FULL PAPER 1600512 (6 of 12) wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim of the as-anodized and air-annealed fi lms in Bode-plot representation. The data are fi tted with a simple R s ( R p –CPE) circuit, where a constant phase element (CPE) is used to account for frequency dispersion, and the effective device capacitance ( C eff ) is calculated, being 0.033(2) and 0.11(2) µF cm −2 respectively. These values correspond to the thickness of a Schottky depletion layer, formed presumably at the tops of the nanorods, of about 54 and 16 nm respectively, providing the relative permittivity ( ε r ) of WO 3 is 20, assuming that the surface area of the nanorods is about 10% of the apparent device surface area, [ 26 ] and meaning that the depletion layer extends only into a small portion of the whole length of the rods, which is about 600 nm over the alumina/tungsten interface. The fi tted values of the parallel resistance R p belonging to the depletion layer correspond to those obtained from the I ( V ) characteristics (Figure 4 b) at 0 V. The serial resistance R s associated with the materials of the rods and electrodes is incomparably low and could be neglected. A comparison of the RT data relevant to Schottky junction at the Au/WO x interface of the MSM microcells tested here (Figure 1 e) with those obtained for a depletion layer formed at a solution/WO x interface of similarly prepared anodic fi lms (as in Figure 1 d) reported in a previous work [ 26 ] is shown in Table S1 (Supporting Information). Comparable capacitance values and the trend of resistance drop in the order as-anodized → air-annealed → vacuum-annealed nanorods are obvious for both fi lm assemblies. The sequence may be due to the differences in the crystal structure of the rod material: [ 26 ] the as-anodized nanorods are amorphous whereas the airand vacuum-annealed nanorods comprise, respectively, WO 3 and WO 2.9 crystal phases at their tops (Figure 2 ). The 2–3 orders of magnitude higher resistance of the as-anodized nanorods relative to the air-annealed rods cannot be explained only by the thicker depletion layer, but rather by its amorphous nature because amorphous oxides usually exhibit lower electron conductivity relative to their crystalline counterparts. [ 35 ] Additionally, the concentration of oxygen vacancies, serving as the electron donors, may be lower at the tops of the asanodized nanorods or the Schottky barrier could be higher, as indicated by the thicker depletion layer at the Au/WO x interface. The resistance of the vacuum-annealed nanorods comprising WO 2.9 phase is 4 orders of magnitude lower than that of the air-annealed nanorods having WO 3 at their tops. This is in agreement with the nature of WO 2.9 phase, being an n-type degenerate semiconductor, [ 36 ] and is supported by the reported values of conductivities of differently doped phases of WO 3 (10–10 −4 S cm −1 ) [ 1 ] and WO 2.9 (10 3 S cm −1 ). [ 36 ] To gain a deeper insight into the Schottky-junction nature of the respective interfaces of the MSM microcells incorporating the as-anodized and air-annealed nanorods, temperature-dependent I ( V ) curves were recorded, as shown in Figure 6 together with corresponding R dif ( V ) plots. The narrower voltage range of 0 ± 1.0 V was chosen to minimize the effect of resistance decrease due to the higher voltage applied, as observed during the RT measurements. Three main effects of elevated temperature on the electrical behavior of both types of the MSM devices are worth mentioning. (1) The resistance of the nanoarrays decreases systematically by several orders of magnitude with raising the temperature, as shown in Figure 6 h by plotting the temperature dependence of R dif at 0 V, the resistance drop being more pronounced for the as-anodized samples. The decreasing R dif value, regardless of the bias polarity, is associated with semiconducting properties of the WO x nanorod arrays. (2) The I ( V ) curves exhibit increasingly asymmetric temperature-dependent electron transport properties, being inverse with respect to the higher-voltage RT behavior (Figure 4 ), i.e., with a steeper current rise when the upper electrode is negatively biased. Over about 100 °C, the diode-like I ( V ) behavior becomes signifi cantly enhanced in the as-anodized nanorods, with a current difference of 2 orders of magnitude between the positive and negative bias polarities, compared with only twice as different values of current that fl ows through the airannealed nanorods. (3) The onset forward potential is going to zero with increased temperature, as shown in Figure 6 g for the as-anodized sample. For the air-annealed fi lm, the exact position of the onset potential cannot be defi ned and does not change much with operating temperature, taking the range of −0.2 to −0.1 V. The asymmetric I ( V ) behavior of the as-anodized fi lms at elevated temperature, contrasting with the inverse behavior at RT, being dramatically enhanced beyond 100 °C and showing a substantially earlier and sharper current rise, implies that, when operating temperature is raised, a Schottky barrier is created at the opposite, i.e., W/WO x interface, which forms between the substrate metal (W) and the bottom parts of the oxide nanorods. Considering the relatively low work function of W (4.3–4.6 eV) [ 33 ] and the observed high thermal sensitivity of the W/WO x interface, we assume that the barrier is relatively weak and unstable. The reverse current begins to gradually increase rather soon with increasing the temperature owing to temperature-stimulated thermionic emission, while the rectifying effect of the W/WO x junction strengthens (Figure 6 a). This also suggests that the upper (Au/WO x ) interface, which forms between the gold electrode and the tops of the rods, gives an ohmic-like contact or, alternatively, a Schottky-like junction that is, however, too low to compete with the barrier at the W/WO x interface at elevated temperature. Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.de www.MaterialsViews.com Figure 5. The Bode-plot representation of impedance spectroscopy measurements of the MSM devices incorporating the as-anodized and air-annealed WO x nanorods recorded at RT at 0 V DC bias. The black curves are the fi tted spectra; the arrows indicate the corresponding y -axis. FULL PAPER (7 of 12) 1600512 wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim An intermediate situation arises with the air-annealed nanorods, where the Schottky barrier is lower than that in the as-anodized samples, manifested by the lower resistance and a smaller extent of rectifi cation. The asymmetric I ( V ) behavior at elevated temperature is also determined by the W/WO x Schottky junction, as assumed from the sharper current rise when the negative bias is applied to the top (Au) electrode (Figure 6 d). In order to confi rm the dominating contribution of the W/WO x Schottky interface for both the as-anodized and airannealed samples at elevated temperature and possibly estimate the corresponding Schottky-barrier heights, we plotted the I ( V ) temperature dependencies as log I vs log V and log I vs V 1/2 for the different polarizations of the top (Au) electrodes (see Figure S1 in the Supporting Information). A qualitative comparison of features present in I ( V ) characteristics for an ideal Schottky W/WO x interface with those revealed in the experimental temperature-dependent I ( V ) curves for the two fi lm types is given in Table S2 in the Supporting Information. As the log I vs V 1/2 plots at positive polarization of the top (Au) electrode, which is the reverse direction, show linear dependencies above ≈0.15 V (Figure S1b,f, Supporting Information), we consider the Schottky emission as the main conduction mechanism in the given voltage range, expressed by the following equation [ 32,37 ] φπε =×× −− × ⎛ ⎝ ⎜⎞ ⎠ ⎟ IAT qq dV kT exp 4 R *2 BR ( 1) where I R is the reverse current density, A * is the effective Richardson constant, φ B is the Schottky barrier height, ε is the permittivity of WO 3 , and d is the depletion layer thickness. From this relation, a linear dependence in the log I R vs V R 1/2 representation is expected, which is indeed obtained for the as-anodized and air-annealed fi lms. We may rewrite Equation ( 1) as follows φπε =−×− × ⎛ ⎝ ⎜⎞ ⎠ ⎟ ln ln 1 4 R 2 * BR I TAT q k q dV ( 2) and create so-called Richardson plots, ln I / T 2 vs 1/ T , for selected voltages, which are shown in Figure 6 c,f for the as-anodized and air-annealed fi lms and ideally are linear dependencies. By multiplying their slope by k / q and plotting against V 1/2 (see insets of Figure 6 c,f), one can obtain the Schottky-barrier height, φ B , as the y -axis intercept and can also get the thickness of the depletion layer, d , from the slope of the plots in the inset graphs: slope = −( q /4 π ε d ) 1/2 . Thus, the air-annealed sample gives a φ B -value of 0.25 eV and a d -value of 3.6 nm. However, the as-annealed fi lms show rather two slopes in the Richardson plots, the higher-temperature one being considered more reliable, leading to a φ B -value of 1.78 eV and a d -value of 5.6 nm. Considerations of this calculation at the reversed bias polarity yielding relevant φ B -values, as well as additional calculations at the forward bias, and a discussion of possible contribution of Frenkel–Poole emission transport [ 32,38 ] are given in the Supporting Information. In spite of some uncertainties in fi tting the data, introduced by the thermal instability of the Au/WO x interface, Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.dewww.MaterialsViews.com Figure 6. Experimental a,d) I ( V ) and b,e) R dif ( V ) curves recorded at elevated temperature for the MSM microdevices incorporating the a,b) as-anodized and d,e) air-annealed PAA-inbuilt WO x nanorods. c,f) Richardson plots (ln I / T 2 vs 1000/ T ) calculated from the currents at the positive bias on the top Au electrode (shown in panels (a) and (d)) for the as-anodized and air-annealed devices, respectively, to reveal the Schottky barrier height (shown as inserts in panels (c) and (f)). g) The onset potential vs operating temperature for the as-anodized WO x nanorods (as estimated from log–log representations of the I ( V ) curves at the negatively biased top electrode, shown in the Supporting Information). h) Resistance vs operating temperature at 0 V bias voltage for the as-anodized and air-annealed WO x nanorods. FULL PAPER 1600512 (8 of 12) wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim the following conclusions can be drawn. (1) With raising the temperature, in both the as-anodized and air-annealed fi lms, the bottom (W/WO x ) interface behaves like a Schottky junction, contrasting with the RT situation when a Schottky barrier forms at the upper (Au/WO x ) interface. (2) The φ B -value calculated for the as-anodized fi lms (0.5–2.2 eV) is obviously higher than that obtained for the air-annealed fi lms (0.2–0.3 eV). The d -value for the bottom W/WO x depletion layer is about 5 nm for both fi lm types. A well-defi ned Schottky junction may not form at the Au/WO x interface at elevated temperature because the work function of a mixture of amorphous WO 3 , WO 2 , and Al 2 O 3 composing the upper parts of the as-anodized rods [ 26 ] may be comparable to, or exceeding, that of Au (5.4 eV), [ 33 ] so that the formation of an ohmic contact is quite feasible. Moreover, the Au/WO x interface is likely to give an impeded contact and may have an uneven depth profi le owing to the point electrodeposition of gold and possible partial mixing of the materials at the interface. At the W/WO x interface, due to the absence of extraneous bodies, an intimate (true) contact is formed by the strongest short-range forces. Further, due to a specifi c chemical composition of the bottom parts of the rods, which are mainly substoichiometric WO 3 , the work function of the material may also differ from the reported values. This may result in the formation of a more pronounced Schottky-like junction at the W/ WO x interface at elevated temperature. A similar phenomenon of forming a Schottky contact at a W/WO x interface, even at RT, has been reported by Kim et al. [ 39 ] in course of studying electrical properties of thin WO 3 fi lms. Additionally, surface states present at the interfaces, which might be also infl uenced by the residues of adventitious carbon (at the Au/WO x interface), may also contribute to the formation of a potential barrier, which may impact electron transport in the MSM microcells. [ 31,32,40 ] The substantially lower W/WO x Schottky barrier in the airannealed fi lms relative to the as-anodized fi lms may result from chemical and crystallographic changes in the material composing the bottom parts of the rods. The diffusion of oxygen ions into the metal during the heating makes the bottoms of the rods more oxygen-depleted relative to the as-anodized fi lms, thus leading to the formation of WO 2.9 phase and a more ohmic-like contact at the W/WO x interface. This is obvious from both the low Schottky barrier and the weak rectifi cation behavior at elevated temperature of the air-annealed fi lms compared to the as-anodized fi lms. Moreover, the depletion layer that forms at the bottom interface of both the as-anodized and air-annealed fi lms at elevated temperature, about 5 nm thick, is much thinner than that formed at the tops of the rods at RT, being 54 and 16 nm, respectively. We assume that this is caused by a relatively higher concentration of oxygen vacancies at the bases of the rods with respect to their tops, [ 26 ] rather than by a higher Schottky barrier formed at the upper (Au/WO x ) interface at RT compared with that of the bottom (W/WO x ) interface at elevated temperature. This is especially true for the as-anodized fi lms, in which a relatively high W/WO x Schottky barrier forms. [ 40,41 ] Additionally to the contact phenomena at the top and bottom interfaces, the conductance along the nanorods contributes to the overall electron transport in the fi lms. Beside the inner part of the rods having inherent oxygen vacancy concentration, the electronically unsharp Al 2 O 3 /WO x interfaces having a mixed composition with a gradient in the alumina content [ 24,26 ] may be considered as the shells of the rods infl uencing the conductivity, for example, through their dielectric nature or by changing the concentration of electron traps at the surface of the rods. [ 42 ] In the case of the PAA-inbuilt as-anodized and air-annealed nanorod arrays, the effect of the Schottky interfaces on the overall electron conductance is most essential, as becomes evident from the RT capacitance measurements and the asymmetric I ( V ) characteristics. The situation, however, may change if the alumina surrounding the rods is removed via chemical dissolution. The “naked” rods may gain higher or lower electron conductance and may demonstrate a different type of I ( V ) characteristics, with a higher impact of the conductance along the rods on the overall electron transport in the fi lms. This issue may be addressed in a future work. From a practical view-point, the nanorod arrays developed here may sustain a current density as high as 100 A cm −2 without causing an electric breakdown or blistering damage of the materials composing the fi lms. The achievement is explained by the parallel conduction model, which posits that a Schottky barrier is essentially a collection of a very large number of nanoscale diodes in parallel (≈7 × 10 8 cm −2 ), with the net I ( V ) characteristic of the junction collecting the currents of about 0.1 µA carried by each metal-oxide nanorod. 2.3. Memristive Electrical Switching The as-anodized WO x nanorods assembled in the Au/WO x /W microcells described in the previous section reveal resistive switching properties at RT with the following major characteristics. (1) The devices are switched from the virgin state into the low-resistance state (LRS) during I ( V ) cycling to increased voltage applied under both the positive and negative polarization of the top Au electrode ( Figure 7 a). The value of forming voltage, represented by an abrupt current jump to the current compliance limit, is slightly higher in the case of the positive bias, ranging from 2.0 to 2.5 V. Additionally, the asymmetric character of the I ( V ) curves due to the presence of an Au/WO x Schottky barrier at RT, as described above, becomes better pronounced with extending the voltage range. (2) After the forming step, bipolar resistive switching is observed with a reset process taking place at the positive polarity at the top Au electrode, manifested by a gradual decrease in measured current, leading to transformation of the device into a high-resistance state (HRS), followed by a consecutive set process at the negative bias polarity typically represented by an abrupt current jump (Figure 7 b). The switching polarity is independent of the voltage bias polarity of the forming step. (3) The value of fi lm resistance in the HRS is dependent on the maximum voltage applied during the reset process (Figure 7 c), whereas consecutive I ( V ) sweeps to the same maximum voltage do not lead to any further resistance rise. However, the HRS resistance cannot be tuned by the maximum voltage in a defi ned manner, and it varies by up to two orders of magnitude within a group of 15 tested devices. (4) A higher value of set voltage is required to switch a higherresistance HRS into LRS, as depicted in Figure 7 d. Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.de www.MaterialsViews.com FULL PAPER (9 of 12) 1600512 wileyonlinelibrary.com © 2016 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim This kind of resistive switching behavior can be explained as being due to the electric-fi eld-driven movement of oxygen vacancies and thus induced changes in the Schottky junction that forms at the Au/WO x interface and dominates the transport properties at RT, [ 29,38,39,41,43 ] as depicted in detail in Figure 8 . The nature of the forming process appears to be bias-polarity dependent. In case of the positive bias polarity during the forming step (Figure 8 a), the Schottky junction at the Au/WO x interface is in forward direction, resulting in almost uniform potential distribution along the rods. Oxygen vacancies are repelled from the top electrode and start to accumulate [ 29 ] near the oxygen-defi cient WO x phase present at the bottoms of the nanorods. [ 26 ] A conducting fi lament is created in this way reaching the upper Au/WO x interface, connecting electrically the two electrodes. The fi lament may be composed of extended defects in the amorphous nanorods, like vacancy chains or shear planes; [ 29 ] eventually a conducting Magnéli phase may crystallize [ 44 ] due to the Joule heat locally generated via vacancy and electron drift. [ 43 ] Another situation arises at the negative voltage bias polarity of the forming step (Figure 8 b). In such case, the Au/WO x Schottky junction is in reverse direction, and most of the applied voltage drops across the depletion layer at the Au/WO x interface, being about 50 nm thick as estimated from the impedance measurements, leading to oxygen-vacancy drift mainly within the depletion layer. A conducting fi lament is likely to form in the same way as under the positive forming bias, altering the barrier and thus modifying the electron transport properties of the oxide nanorods. In most of the tested devices, the resistances of the two LRSs were very similar, ranging from about 0.005 to 0.02 Ω cm 2 , although one may expect a bigger difference. As can be seen, the active interface for the resistive switching after the two forming tactics is still the Au/WO x junction, and this explains the identical direction of the bipolar resistive switching being independent of the voltage bias polarity of the forming step. The subsequent transition from LRS to HRS, happening by a reset process at the positive bias polarity reveals characteristics typical for a gradual reset. Recently, the origin and kinetics of such a reset have been reviewed and thoroughly investigated with TaO x -based structures. [ 43 ] The conclusion was that the gradual reset is related to the temperature-accelerated oxygen-vacancy drift and diffusion, combined with a moderate sensitivity of electron transport Adv. Mater. Interfaces 2016, 3, 1600512 www.advmatinterfaces.dewww.MaterialsViews.com Figure 7. Experimental I ( V ) curves showing memristive electrical switching in the MSM microdevices incorporating the as-anodized PAA-inbuilt WO x nanorods (RT, top (Au) electrodes are biased). a) Bias-polarity independent forming cycles for four individual microdevices. b) Bipolar resistive switching, which is independent of the polarity of the forming step, with a gradual reset developing at the positive polarization, followed by a set process at the negative bias. c) Three consecutive reset processes for tuning the HRS resistance by changing the value of the positive voltage bias. d) Set processes for several devices having different resistances in their HRS. The set voltage is dependent upon the HRS resistance value, as depicted in the inset graph. Figure 8. Schematic illustrations of the forming processes during the a) positive and b) negative polarization of the top Au electrode of the MSM microdevices employing the as-anodized PAA-inbuilt nanorods. A conducting fi lament is created either a) within the whole length of a nanorod, growing from the bottom at the positive bias polarity applied at the top electrode or b) only in the upper part of a nanorod within the depletion layer of the original Au/WO x Schottky junction. The light blue circles represent oxygen vacancies in the fi lms in the virgin state, denoting the oxygen-depleted area at the bottoms of the nanorods, whereas the deepblue circles show oxygen vacancies created during the forming processes. The corresponding motion of electrons and oxygen vacancies due to the drift in electric fi eld is marked by arrows. The band diagrams at the right show the voltage-related changes of the Schottky barrier at the Au/WO x interface at a zero bias (black curves) and at the commencement of the voltage bias rise (colored curves). c) Bipolar resistive switching takes place at the Au/WO x interface, regardless of the extent of fi lament growth.