Local probing of electrochemically induced negative differential resistance in TiO₂ memristive materials

25 and 30 nm thick TiO₂ thin films were prepared on Pt/TiO₂/Si substrates by atomic layer deposition (ALD) at a growth temperature of 150 °C. Titanium (IV) tetraisopropoxide (TTIP; Ti[OCH(CH₃)₂]₄, 99.99%) and deionized water vapor were used as a Ti precursor and an oxidant, respectively. The two precursors were contained in separate stainless steel canisters, which were heated to 40 °C for the TTIP and maintained at room temperature for the water. TTIP vapor was delivered to the process chamber with a pure N₂ gas (99.999%) at a flow rate of 100 sccm. The unit ALD cycle was composed of TTIP exposure (3 s), N₂ purge (10 s), water vapor exposure (0.2 s), and N₂ purge (10 s). N₂ gas at a flow rate of 200 sccm was injected into the process chamber during the N₂ purge. According to x-ray diffraction analysis, the resulting TiO₂ film turned out to be amorphous.

Scanning probe microscopy studies were performed on a commercial system (Asylum Research Cypher) equipped with a current amplifier (Femto DLPCA-200) and home-built Labview/Matlab. A glove box (MBraun) filled with Ar was used for environmental control. During the voltage–current measurements, series resistors of 10 or 30 MΩ were used to limit maximum current levels. ESM measurements were carried out using a band-excitation (BE) waveform of 80 kHz width centered on the contact resonance (315 kHz) and an amplitude of 1.5V_pp applied to a Pt/Cr coated probe (Budget sensors ElectriMulti75-G). The BEPS–ESM measurements were performed in a voltage range of ±6 V and collected the hysteresis loop in the off-field mode over an area of 10 × 10 μm². Smoothing was performed for each topographic difference image. Notably, it has been commonly accepted that the surface adsorbates can affect the absolute value of the measured signal, without significantly affecting the comparative measurements along the surface [32, 33]. In addition, capacitance is present and is in parallel to the resistance/capacitance of the tip–surface junction. Since the signal in electromechanical based AFMs such as ESM is affected by the potential induced by the tip at the tip–surface junction, the signal for sufficiently high input impedance of the current source is unaffected by the cantilever–surface capacitance and other non-local elements of the equivalent circuit [34]. Hence, the cantilever–surface capacitance does not affect the ESM signals.

TEM studies were performed on an FEI Titan aberration-corrected TEM. A TEM sample was prepared by a Hitachi NB5000 dual-beam instrument with one electron beam column and one ion beam column, focused at the region deformed by positive bias sweep up to +7 V in 25 nm TiO₂ thin films.

Here, we explore the variability of memristive formation using multiple I–V measurements over a spatial grid of 10 × 10 points over an area of 16 × 16 μm². In order to avoid hard dielectric breakdown, the measurement circuit included current limiting series resistors of 10 or 30 MΩ. At each grid point, two different positive voltage sweeps were applied to a conductive probe. The first (forming) bias sweep induced an electrochemical process under a high electric field, and the second (probing) bias sweep was used to establish whether its electrical state was changed.

Figure 1 shows I–V curves and their corresponding topographic difference images. For I–V measurements, voltages were applied up to 12 and 20 V, respectively. Interestingly, all the I–V curves show NDR features at around 7–8 V (see the blue arrows in figures 1(a) and (c)). In the inset of figure 1(a), a histogram of maximum voltages of the NDR feature (V_NDR) shows a very uniform distribution at around 7.5 V. After the I–V measurements of figures 1(a) and (c), a uniform surface deformation (disk shaped feature on the topographic image) is observed at each measurement location, as shown in figures 1(b) and (d). The magnitude of surface deformation was about 0.3–0.4 nm. The lateral size of the disks in figure 1(d) was broader than those in figure 1(b). Note that the large size (>600 nm) as compared to the 10–30 nm tip–surface contact area indicates a large column of bias-transformed region and can be related to the water layer on the sample surface. These surface deformations could be observed after >10 h, indicative of thermodynamic stability.

Zoom In Zoom Out Reset image size

Under a positive bias to the top electrode, an oxygen gas bubble can be generated at the interface between the top electrode and the memristive material [35]. However, in this case the top electrode is absent and hence bubble formation in the near-surface layer is unlikely. Furthermore, complete electroforming was not achieved by the first bias sweep (see the second bias sweep in figures 1(a) and (c)). Even though the TEM data indicate partial crystallization of the material (see figure 6(b)), this process decreases the molar volume and hence cannot account for the observed behavior [36]. Thus, a gas bubble and complete electroforming can be excluded as possible reasons for the NDR behavior and corresponding surface deformations. As an alternative mechanism, oxygen species such as O₂,O⁻, or O₂⁻ can be adsorbed on the sample surface under high applied bias voltage. The attractive force by the adsorbed oxygen species can contribute significantly to the topographic imaging by non-contact mode AFM [37]. However, accumulation of adsorbed oxygen is hardly detectable by contact mode AFM. Furthermore, similar topographic change was observed even using a much stiffer cantilever (diamond coated cantilever, >40 times higher than a Pt/Cr coated cantilever) under a higher contact force (not shown here). Since there is a clear long-lasting topographic change, the adsorbed oxygen can be excluded as a dominant phenomenon in the topographic change.

Finally, the surface deformation may be related to the defect states of the present TiO₂ thin films. Since Ti interstitials can exist on TiO₂ surfaces and, furthermore, can be oxidized [38, 39], surface deformations can occur under positive bias sweeps due to the field-enhanced surface oxidation [40, 41]. In other words, non-stoichiometric TiO₂ transforms to the stoichiometric insulating state, resulting in the NDR effect. To confirm the field-enhanced surface oxidation as a possible reason for the surface deformation, the same experiments as shown in figure 1 were performed using the AFM enclosed in a glove box filled with Ar (O₂ and H₂O partial pressures below the 1 ppm level). During the positive bias sweeps, there was no distinguishable current bump in the I–V curves (see figure 2(a)). This indicates that there is no NDR behavior under an Ar atmosphere. Furthermore, surface deformations were not observed after the positive bias sweeps. The I–V curves and corresponding topographic information clearly confirmed that oxygen is a required component in the observed phenomena, and hence the field-induced surface oxidation is responsible for the formation of the surface deformation.

Zoom In Zoom Out Reset image size

We further explored the reversibility of the electrochemical processes that yield static surface deformations though different polarity sequences of bias sweeps. First, positive and subsequent negative bias sweeps (A bias sweep in figure 3(a)) were carried out at four different locations. As a reference point, single positive bias sweeps, which were similar to those in figure 1, were carried out at the center of the scanned region (B bias sweep in figure 3(a)). The I–V curves of both figures 1(b) and (c) show the NDR phenomena in the positive bias sections. However, only the B bias sweep shows surface deformation and there is no surface deformation after A bias sweeps, as seen in figure 3(d). This means that a negative bias sweep just after a positive bias sweep leads to recovery of surface deformation. Hence, the formation of topographic protrusions and NDR are the result of local surface oxidation that can be reversed by local reduction. This scenario is further confirmed by the large fluctuations in current during the negative bias sweep, indicative of an electrochemical process near the surface and tip–surface contact instability. Thus, it was found that the static surface deformations can be reversed by applying different polarity of the bias sweeps.

Zoom In Zoom Out Reset image size

The CAFM results in figures 1 and 3 show the NDR and the corresponding surface deformation. However, CAFM measurements are limited by the fact that the measurements are non-local, i.e. the conductivity of the whole material region between the tip and the bottom electrode is probed. Furthermore, the relationship between oxygen vacancy concentration in TiO₂ and conductivity is highly complex [42], precluding an unambiguous picture of oxygen vacancy dynamics in the near-surface layer from being obtained. To complement CAFM studies, we probed the near-surface oxygen vacancy behavior using ESM. ESM was originally developed for Li-ion conductive materials [29, 43, 44] and recently applied for oxygen ion conductors [31, 45]. ESM probes electrochemical activity including oxygen reduction and evolution (redox) processes using bias-induced electrochemical strain as the measured signal [29, 31]. Here, we explore the electrochemical activity on the pristine and electroformed TiO₂ surfaces through spatially resolved ESM hysteresis loop measurements.

Figure 4(a) shows disk-type surface deformations formed by the same bias sweeps as figure 3(c) at four different locations. As shown in the inset of figure 1(a), surface deformations occur at voltages higher than 6.5 V. Therefore, in order to explore the electrochemical behavior without further surface deformations, ESM measurements were carried out in the bias range of ±6 V, i.e. below V_NDR. In this manner, we both characterized the reversible electrochemical transformations under the tip and ascertained the changes in the reversible electrochemical activity due to the irreversible surface changes induced by prior application of bias pulses. We further monitored current behavior during ESM to correlate the changes in local conductivity and electrochemical activity.

Zoom In Zoom Out Reset image size

The four disks formed by the application of anodic waveforms clearly show lower maximum current compared to the outside of the disks, as is visible in the maximum current map of the first loop of figure 4(b). This is also visible in the average I–V loops of figure 4(d) and the maximum current map of the second loop (not shown here). Interestingly, the second I–V loop on the outside of the disks shows a much lower current level compared to the first one. The insides of the disks also show the same behavior, but the absolute values are much lower. This indicates that, even though there is no surface deformation under ±6 V, there can be electrochemical processes that are relevant to the electrical properties, e.g. crystallization and/or the early stages of redox processes, etc.

Changes in the electrochemical activity were also observed in the ESM loops in the bias range of ±6 V (see figure 4(d)). The clear loop opening in the ESM loops shows that there are electrochemical processes during the bias sweeps. Since the increase of the ESM signal occurs earlier than that of the current signal (see figures 4(d) and (e)), it indicates that electrochemical reactions occur at lower bias voltages before current flows. Indeed, the operational principle of ESM is based on the local electrochemical reactions induced by a high electric field concentration underneath the AFM tip [29, 31]. Hence, as a first early stage of electroforming, electrochemical reactions without current flow occur near the film surfaces.

We further observed that the changes in ESM signal in the disks could be clearly visualized during multivariate data analysis, as shown in figure 4(c). Since the difference between the ESM loops of the inside and the outside of the disks is relatively small and the loop shape is different from the prototypical square shape, classical analysis strategies to extract the loop parameters fail [46]. Here, to establish spatial variability of the ESM data, we used principal component analysis (PCA) [47, 48]. PCA selects and ranks relevant response components based on variance within the data set. Hence, PCA provides an effective pathway to rapidly process, de-noise, and compress data. Here, PCA allowed us to identify differences in the local electrochemical activity between the inside and the outside of the disks.

Figure 5 shows eigenvectors and principal component maps from the PCA of the local ESM images. In some cases, semi-qualitative interpretation of PCA decomposition data can be achieved based on the shape of the corresponding eigenvectors. For instance, the first component of figure 5(a) can be roughly identified as corresponding to the ESM loop. The first and second components of figure 5(b) do not show any significant response in the disks, whereas the third component of figure 5(b) clearly shows a higher response in the disks.

Zoom In Zoom Out Reset image size

The PCA results clearly confirm the change in the electrochemical activity. Hence, they further allow analysis of the ESM loops. Even though the first positive bias sections of the first ESM loops are different from each other, the second ESM loops in both the inside and the outside of the disks are nearly identical, which shows that the electrochemical activity is similar for the second loops (i.e. the measurement sequence has affected the electrochemical state of the formed region and reset it with the substrate). Since the first positive bias section in the inside of the disks shows lower ESM signal, it means that the insides of the disks are electrochemically less active during the first bias sweeps. As mentioned above, the disk-type surface deformation is achieved by oxidization of the defective surface.

In order to provide more information on the surface deformation, we further explored the surface deformed areas using a TEM. Notably, figure 6(a) shows the phase transition on the surface deformed area from an amorphous phase (right side) to a crystalline one (left side) after a positive bias sweep up to +7 V. The high-resolution TEM images in figures 6(b) and (c) more clearly show distinguishable FFTs from crystalline TiO₂ (c-TiO₂) and amorphous TiO₂ (a-TiO₂). In fact, significant current flow at more than 4 V in the first bias sweep of figures 1 and 3 already shows the feasibility of crystallization by Joule heating. However, since the TiO₂ was primarily crystallized to insulating phases such as anatase and rutile (see the insets of figure 6(b)), the second bias sweep did not show any significant increase in current. In other words, complete electroforming was not achieved in the surface deformed area.

Zoom In Zoom Out Reset image size

Therefore, a plausible explanation can be given as follows. In the first bias sweep, the less defective surface deformation layer may need a higher critical bias to activate electrochemical processes and corresponding measurable strains due to the surface oxide layer. However, once the entire volume along the thickness direction is electrochemically active during the first bias sweep, then the electrochemical behavior becomes nearly identical for both the inside and the outside of the disks. Nevertheless, since there is an additional surface layer on the surface deformed area, the electrical behavior can be different, as seen in figure 4(d).

On the basis of experimental results as well as previous reports, electroforming procedures under positive bias sweep to the conductive probe are expected to be described as follows (see figure 7).

• (1)  
  Local electrochemical reactions, i.e. oxygen vacancy formation, occur underneath the AFM tip (figure 7(a)(ii)). However, there is no significant oxygen vacancy movement and/or electron injection.
• (2)  
  Both electron injection from the bottom electrode and oxygen vacancy movement from the film surface induce current flow, resulting in local Joule heating (figure 7(a)(iii)).
• (3)  
  Joule heating induces crystallization of TiO₂, e.g. brookite, anatase, rutile, and/or Magnéli (figure 7(a)(iv–v)).
• (4)  
  The surface oxidation, which is relevant to the NDR, occurs during the crystallization process of stage (3) (figure 7(a)(vi)).
• (5)  
  As the positive bias voltage keeps increasing, a conductive filament can be formed along the thickness direction, so called complete electroforming (not shown here).

Zoom In Zoom Out Reset image size