Resistive switching in TiO2 nanocolumn arrays electrochemically grown

Resistive switching in metal oxides, especially in TiO2, has been intensively investigated for potential application in non-volatile memory microdevices. As one of the working mechanisms, a conducting filament consisting of a substoichiometric oxide phase is created within the oxide layer. With the aim of investigating the filament formation in spatially confined elements, we fabricate arrays of self-ordered TiO2 nanocolumns by porous-anodic-alumina (PAA)-assisted anodizing, incorporate them into solid-state microdevices, study their electron transport properties, and reveal that this anodizing approach is suitable for growing TiO2 nanostructures exhibiting resistive switching. The electrical properties and resistive switching behavior are both dependent on the electrolytic formation conditions, influencing the concentration and distribution of oxygen vacancies in the nanocolumn material during the film growth. Therefore, the PAA-assisted TiO2 nanocolumn arrays can be considered as a platform for investigating various phenomena related to resistive switching in valve metal oxides at the nanoscale.


Introduction
Investigation of switching behavior of several metal oxides, which may have their resistance states modulated by voltage applied to or by the current flowing through them, is nowadays a growing research field motivating researchers and engineers to develop high-speed low-power nanodimensional non-volatile Resistive Random Access Memory (ReRAM) devices. In such devices, a layer of metal oxide is usually sandwiched between two metallic electrodes for obtaining metal/semiconductor/metal structures. TiO 2 , among other metal oxides like Ta 2 O 5 , HfO 2 , ZrO 2 , ZnO, WO 3 , or NiO is one of the most intensively investigated resistive switching materials because of many advantages, such as the versatile fabrication techniques and suitable electrical properties resulting from a wide band gap and high dielectric constant [1,2,3].
A common mechanism of resistive switching in metal oxides is the formation of a thin conducting filament within the oxide layer, based on the field-induced movement of mobile oxygen vacancies [4,5]. A further filament rupture and re-creation at one of the interfaces between the metal oxide and a metal electrode, while applying voltage, is often the underlying phenomenon behind a bipolar resistive switching [5]. The filament is thus composed of a substoichiometric crystalline phase of the metal oxide, a Magnéli phase [6], formed also due to the high temperatures achieved during the local heating [7].
Nanostructuring of TiO 2 to the lateral size being comparable with a filament diameter (~15 nm [6]) may spatially confine the filament formation, leading to phenomena different from those occurring in continuous thin films. This motivated our efforts to incorporate arrays of TiO 2 nanocolumns prepared via the porous-anodic-alumina (PAA)-assisted anodizing [8,9] into a 3-D solid-state microdevice and investigate their electrical transport properties. The advantage is that the sizes of nanocolumns can be well controlled through the anodizing variables. In addition, the specific electrolytic conditions applied during the PAA-assisted film growth may enable tuning stoichiometry of the oxides and thus modification of their electrical behavior. In the present work, we fabricate PAA-assisted TiO 2 nanocolumn arrays, embedded in the anodic alumina matrix, by varying the anodizing conditions. We incorporate the arrays into metal/semiconductor/metal microdevices by electrochemically depositing top gold electrodes and measure their room-temperature current-voltage behavior, revealing the right combination of electrolytic conditions appropriate for the occurrence of bipolar resistive switching in the anodic films.

Fabrication of the TiO 2 arrays
A silicon wafer covered with a 500 nm thick thermal oxide layer was used as a starting substrate. A high quality Al/Ti bilayer (aluminum-on-titanium, 500/300 nm thick) was deposited via ion beam sputtering. The wafer with the deposited metal layers was cut into 10 mm ×15 mm pieces, which were individually processed in a closed two-electrode PTFE cell with a stainless steel counter electrode, exposing a surface area of 28 mm 2 . Anodizing and re-anodizing experiments were performed in 0.3 M oxalic acid solution at 10 °C. Anodizing at a constant potential of 40 V led to formation of a PAA matrix having self-ordered nanopores, with nanoprotrusions of TiO 2 within the alumina barrier layer at the bottom of the PAA film having a mean diameter of ~35 nm. Re-anodizing to a higher anodic potential with a sweep rate of 2 V·s −1 resulted in relatively longer TiO 2 nanocolumns, while their length was tuned by setting the value of re-anodizing potential to 70, 100, and 130 V (arrays A) [10,9]. In addition, the nanocolumns grown to 100 V were prepared also by re-anodizing with a sweep rate of 0.2 V·s −1 (array B-100V). The typical current-time and potential-time transients obtained for the two sample types, re-anodized to 100 V, are shown in figure 1, supported by the schemes of nanoarray structure at the different stages of PAA-assisted film growth. The samples were further modified by annealing at 500 °C in vacuum (10 −4 Pa).
To form the top contact to the TiO 2 nanocolumns and thus obtain a memory cell, gold was electrochemically deposited onto the column tops within and over the PAA nanopores [11,12]. The size of the top Au electrode was defined by an etch-back photolithography to ~100 μm × 100 μm.

Characterization
Experimental samples (surfaces and cross-fractures) were observed by scanning electron microscopy (SEM) using a Tescan Mira II at different stages of sample fabrication. The electrical characterization was performed at room temperature using a Keithley 4200 SMU system connected to a CASCADE M150 probe station with magnetic micromanipulators, by recording I(V) curves with a scan rate of 100 mV·s −1 from 0 to +5.0, then to −5.0 and back to 0 V in each cycle. To avoid damage of the microcells, the current compliance limit was set to 100 mA. The top Au pad was biased in all I(V) measurements, while the bottom Ti layer, remaining after the PAA-assisted anodizing and being common for all devices prepared on a single chip, was grounded. The measured current and the calculated resistance were related to the projected area of the microcells.

Results and discussion
Arrays of TiO 2 nanocolumns are prepared via PAA-assisted anodizing to investigate their electrical properties with the focus on resistive switching application. The anodizing procedure is varied to obtain nanocolumns of variable length (increasing with the re-anodizing potential) and of different stoichiometry (possibly tuned by the potential sweep rate during the re-anodizing). To enable a good electrical contact to each nanocolumn in the array, gold is electrochemically deposited onto the TiO 2 columns, filling the alumina pores and growing over the alumina surface, which is constrained to square-shaped pads of a micrometer size. Typical examples of the devices and their cross fractures are displayed in the SEM images of figure 2, where the different array components can be seen and are schematically shown. A homogeneous filling of the alumina pores, with most of the TiO 2 nanocolumns contacted by gold, is desirable. The pore filling achieved in this study was not worse than 75%. Typical initial I(V) characteristics of the arrays A (i.e. having the higher potential sweep rate during re-anodizing) processed to different final potentials are shown in figure 3a in a logarithmic representation to emphasize the differences between the samples. Each of these arrays, comprising TiO 2 nanocolumns of various lengths, reveals non-linear shape of the corresponding I(V) curve, also showing substantial current drops and jumps for the 70V and 130V arrays and an asymmetric diodelike behavior for the A-100V and A-130V samples. The initial resistance at 0 V (R initial,0V ) increases with increasing re-anodizing potential, from ~0.25 Ω·cm 2 for sample A-70V to ~7.5 Ω·cm 2 for both A-100V and A-130V arrays. Further I(V) cycling (see figure 3b) shows again non-linear characteristics with an increased current density as compared with the initial state, being asymmetric only for the A-100V and A-130V arrays. The cycling resistance at 0 V (R cycles,0V ) is thus lower than that in the initial measurement by about ten and two times in case of sample A-70V and samples A-100V and A-130V respectively. The electrical characterization of the arrays A therefore does not reveal any resistive switching behavior, for which a substantial decrease of resistance after the initial I(V) cycle accompanied by a presence of high-and low-resistance states (HRS and LRS, respectively) in the following cycles is expected [12]. The observed current jumps and drops in the initial I(V) curves of arrays A (figure 3a) are rather caused by imperfect contacting, and there are no signs of distinguishable HRS and LRS in the following cycles. On the other hand, the resistance increases with the re-anodizing potential, which is in line with the increasing length of the nanocolumns and with their semiconductive nature. In addition, as samples A-100V and A-130V have their initial and consecutive I(V)'s similar to each other but different from those of sample A-70V (see the curve shape and symmetry), different conduction mechanisms are expected in these array groups. The oxygen-deficient root structure [13] of A-70V columns in the PAA barrier may prevail in the shorter nanocolumns, whereas a more stoichiometric, i.e. less conducting column body situated over the roots may lead to the diode-like behavior of A-100V and A-130V arrays, possibly with a Schottky barrier created at the Au/TiO 2 interface [12].
To alter the electrical behavior of the PAA-assisted TiO 2 nanocolumns, an array of type B (i.e. having the lower potential sweep rate during re-anodizing) was also prepared and electrically characterized. An initial I(V) curve of sample B-100V (figure 3c) shows a non-linear asymmetric behavior with R initial,0V of ~6·10 5 Ω·cm 2 and a current jump at 5.0 V, leading to a drop in the resistance of about one order of magnitude. The consecutive cycling (figure 3d) reveals also a non-linear The array B-100V shows therefore signs of bipolar resistive switching. During the initial, forming I(V) cycle, a conducting filament is probably formed within the columns or at one of the interfaces [12]. The switching direction, i.e. the transformation of HRS into LRS, happening at the positive polarization of the top Au electrode, followed by a transition from LRS into HRS at the negative polarization, indicates that the switching interface is the bottom one (TiO 2 /Ti) [12]. The absence of abrupt current or resistance change during one of these transitions may indicate synaptic behavior of the array [14], executed in this case by inhomogeneous switching events in the multiple channels present in the array.
Comparing the anodizing conditions during the nanocolumn growth with the electrical behavior of arrays A and B, we see the following relations, summarized schematically in figure 4. The higher potential sweep rate during the re-anodizing (arrays A) leads to the formation of nanocolumns having a relatively low initial resistance (between 0.25 and 8 Ω·cm 2 ), whereas the lower potential sweep rate (array B) seems to increase the resistance (6·10 5 Ω·cm 2 ). This may be explained by a formation of more oxygen-deficient and thus more electrically conducting nanocolumns in the case of the faster sweep rate during the column formation (arrays A) as compared to more stoichiometric oxide formed during the slower sweep rate (arrays B) [15]. Such higher-resistance, more-stoichiometric TiO 2 nanocolumns would be expected to lead to bipolar resistive switching behavior [5], as it is indeed observed in the present study. In addition, the length of the nanocolumns seems to have also an influence on their electrical resistance, possibly due to an inhomogeneous distribution of oxygen vacancies along the nanocolumn material, as manifested in arrays A by the increase in the resistance with increasing nanocolumn length, accompanied by the diode-like I(V) characteristics of the longer A-100V and A-130V nanocolumns. To confirm this mechanism, as well as to elucidate the role of the annealing in vacuum on the concentration and distribution of oxygen vacancies, further analytical and electrical characterization of the arrays is being performed. Additionally, TiO 2 nanocolumn arrays grown at modified electrochemical conditions are currently under investigation; the results to be reported in due course. stoichiometry, and electrical properties. The faster re-anodizing rate (arrays A, upper part) results in a lower R initial,0V and reveals no sign of resistive switching, whereas the slower re-anodizing rate (arrays B, lower part) leads to the formation of more stoichiometric columns having a high R initial,0V and showing the resistive switching.