Probing nanoscale oxygen ion motion in memristive systems

Ion transport is an essential process for various applications including energy storage, sensing, display, memory and so on, however direct visualization of oxygen ion motion has been a challenging task, which lies in the fact that the normally used electron microscopy imaging mainly focuses on the mass attribute of ions. The lack of appropriate understandings and analytic approaches on oxygen ion motion has caused significant difficulties in disclosing the mechanism of oxides-based memristors. Here we show evidence of oxygen ion migration and accumulation in HfO2 by in situ measurements of electrostatic force gradient between the probe and the sample, as systematically verified by the charge duration, oxygen gas eruption and controlled studies utilizing different electrolytes, field directions and environments. At higher voltages, oxygen-deficient nano-filaments are formed, as directly identified employing a CS-corrected transmission electron microscope. This study could provide a generalized approach for probing ion motions at the nanoscale.

for the sake of clarity.

Supplementary Note 1. Nature of the electrostatic force between the probe and the sample
In order to verify the nature of the accumulated charges observed in 1ω signal, we conducted phase detection electrostatic force microscopy measurement (Supplementary Refs. 1-3). The HfO2/TiN sample was firstly probed by a Pt probe on the surface to perform C-AFM measurements with different DC sweeping voltages, and then phase detection EFM was conducted in lift mode with +2 V tip bias to extract the phase shift. Generally, the cantilever was vibrated by a small piezoelectric element near its resonant frequency in the present EFM measurements. The resonant frequency of the cantilever changes in response to any additional force gradient. In principle, attractive forces make the cantilever effectively "softer", which reduces the resonant frequency, so that the phase shift ΔΦ is negative. Conversely, repulsive forces make the cantilever effectively "stiffer", which increases the resonant frequency and shifts phase positively.
With a positive voltage applied to the tip during EFM, as shown in Supplementary Fig. 1, the phase shift ΔΦ is more negative in the stimulated region than the rest of the sample, which means the electrostatic force between the tip and the stimulated region is attractive instead of repulsive, further confirming that the charges that locally accumulated near the surface after positive voltage sweeps are indeed negatively charged species, e.g. oxygen ions or trapped electrons.

Supplementary Note 2. EFM characterization after negative voltage sweeps
In order to further verify that the migration and accumulation of oxygen ions can indeed be probed by the 1ω component of the electrostatic force gradient between the probe and the sample, we have further performed comparative EFM studies on HfO2/TiN samples where negative voltage sweeps are performed in preceding C-AFM measurements (Fig. 3).

Supplementary Note 3. EFM characterization using Al2O3 as the electrolyte material
In order to verify that the migration of oxygen ions can indeed be probed by the 1ω component of the electrostatic force gradient between the probe and the sample, we have further performed control experiments on Al2O3(~5 nm)/TiN samples (Fig. 4). It is well known that Al2O3 has extremely low diffusion coefficient of oxygen ions even close to its melting point (~2000 °C). In the meantime, Al2O3 has a high activation energy of ~6.5 eV for oxygen diffusion (Supplementary Ref. 4). This decides that the oxygen ion motion in Al2O3 should be significantly retarded. Supplementary Fig. 3 shows the I-V curves of the Pt tip/Al2O3/TiN structure in a series of voltage sweeps with different amplitudes (3,5,7,9,11, 13 V). The samples were then subjected to EFM measurements, which indeed shows the absence of charge accumulation below 11 V (which is a high voltage compared with results in HfO2). This is also in agreement with the I-V characteristics in Supplementary Fig. 3, where the current level stays fairly low below 11 V (<50 nA), due to the difficulty of oxygen ion motions in Al2O3.

Supplementary Note 4. EFM measurements in vacuum
In order to assess the role of the testing environment especially surface adsorptions, we have performed control experiments in an ultrahigh vacuum SPM setup (Scienta Omicron variable temperature SPM, vacuum level of ~8×10 -11 mbar, shown in Supplementary Fig. 7a).
The sample was also baked at ~140 °C prior to C-AFM and EFM characterization in order to further remove possible adsorbates from the surface, and identical C-AFM and EFM measurements with that in air were performed afterwards. Supplementary Fig. 4 shows the 1ω As a result, the switching voltage was increased to ~10 V in vacuum (compared with ~8.4 V in air), and the current level was also much lower due to the current compliance of the equipment (~200 nA, compared with 100 μA in Fig. 1p), as shown 16 in Supplementary Fig. 4j. The thermal effect during switching was thus significantly reduced, which in turn avoids severe structural damages shown in Fig. 1q and hence prevents pronounced features from being observed in the 2ω signal. All the results are highly consistent with that in ambient conditions and precludes the influence of surface adsorptions.

Supplementary Note 5. Electrostatic force microscopy on lateral devices
In addition to the vertical devices shown in Figs. 1-4 and 6, we have also prepared devices with a lateral configuration using e-beam lithography, as shown in Supplementary Fig. 5a. The device has a nominal electrode distance of ~150 nm. By applying a positive voltage sweep on the right electrode with respect to the left one (see Supplementary Fig. 5b)

Supplementary Note 6. Oxygen gas eruption and filament formation in HfO2 and TaOx
We have performed further C-AFM measurements on Ti/HfO2/TiN samples, as schematically shown in Supplementary Fig. 6a. As shown in Supplementary Fig. 6b, I-V curves from this structure using C-AFM measurements clearly display hysteretic resistance switching that is characteristic for a Ti/HfO2/TiN memristive structure. Supplementary Figure 6c Fig. 1 and Supplementary Fig. 6a-c therefore point to a consistent picture during the resistance switching of HfO2, where oxygen ions drift as driven by the electric field and the redox reactions at the interface lead to oxygen-deficient filament formation, structural deformation as well as oxygen gas eruption. In the previous measurements on the HfO2/TiN sample without the top electrode ( Figs. 1 and 2), the oxygen gas formed by electrochemical oxidation will directly emit to the ambient and thus was not detected. The introduction of the top electrode in Ti/HfO2/TiN has therefore played a role in signifying such gas formation.
It is worthwhile pointing out that such oxygen ion motion and accompanying effects are universal phenomena for different oxides. We have performed similar C-AFM experiments on 19 TaOx/TiN samples, as shown in Supplementary Fig. 6d, and a sudden resistance switching was observed at ~10 V ( Supplementary Fig. 6e). Topographic analysis once again suggests distortions to the TaOx film as well as formation of oxygen-deficient conduction channels in TaOx, as shown in the inset of Supplementary Fig. 6e and Supplementary Fig. 6f. These results are fully consistent with that observed in HfO2 and once again verify that the ion migrations and redox reactions should be responsible for the resistance switching in oxides based memristive devices. 20

Supplementary Note 7. Effect of moisture on resistance switching
In order to examine the role of ambient condition, e.g. moisture level, in resistance switching of oxides, we have performed additional experiments on the HfO2/TiN samples using C-AFM in two different environments: i) in air (with moisture, the moisture level was controlled to be ~35%) and ii) in vacuum (with no or limited moisture). The vacuum level of the SPM system (Scienta Omicron VT-SPM) used in the new experiments was ~8×10 -11 mbar, and the sample was also baked at ~140 °C before measurements in order to further remove the absorbed moisture. The results are shown in Supplementary Fig. 7b. One can see that when the same bias voltage of 6 V was applied on the C-AFM tip, the sample was always switched to on state in less than 100 s in air, as confirmed by >10 repetitive tests. In stark contrast, repetitive measurements have shown that successful switching was not achieved on the same sample in vacuum condition in much longer time, i.e. up to 800 s. These results clearly demonstrate the role of moisture in assisting the redox reactions and charge transfer that are indispensable for successful resistance switching, in agreement with previous reports on the effect of moisture for both VCM and ECM devices .
In addition to the above current-time (I-t) measurements employing fixed voltage biases, I-V measurements have suggested similar role of moisture in resistance switching. The HfO2/TiN sample was switched to on state at around ~8.4 V in air (Fig. 1p), however the switching voltage has increased to ~10 V in vacuum, as shown in Supplementary Fig. 4j, once again consistent with the role of moisture in facilitating resistance switching processes.

Supplementary Note 8. Controlled programming of resistance states for systematic STEM and EDS characterization
We have adopted a number of varied switching conditions during C-AFM measurements to locally create a collection of stimulated regions existing at different stages of resistance switching to understand the dynamic switching process as well as the physical mechanism corresponding to different resistance states, as shown in the SEM image in Fig. 5a.
Corresponding I-V characteristics in the programming processes are shown in Supplementary   Fig. 8, where the panels are arranged following the same positions of the simulated regions in

Supplementary Note 9. Characterization of tip-sample interactions in forming process
In order to figure out if there exists any tip-oxide interaction when the volcanos/craters are created during the electroforming process, we have performed extensive EDS mapping and spectral analysis as well as Electron Probe Microanalyzer (EPMA) characterization and Wavelength Dispersive X-ray Spectroscopy (WDS) analysis. Supplementary Figure 9a shows a typical crater that is formed on the HfO2/TiN sample, whose stacking sequence in vertical direction is illustrated in Supplementary Fig. 9b. All the elements included in this stacking structure, i.e. Si, Ti, N, Hf and O, were detected in the EDS measurement, however, with no signal indicating existence of Pt ( Supplementary Fig. 9c). We have also performed EDS mapping on the Pt L edge, and once again only background noise was collected, as shown in Supplementary Fig. 9d. These results exclude the possibility of alloy formation during the forming process, since the Pt element should get incorporated into the sample in that case, at least in the crater region, which is in contrast to the above experimental results.
In addition to the above characterization on the crater, a complete examination on the alloying possibility should also include analysis on the Pt tip, as shown in Supplementary Figs. 9e-f. It is expected that alloy formation should also lead to cross contamination to the tip. It is interesting to find out that the end of the Pt tip seems to have gone through a melting process and a sphere-like shape was formed, probably due to the accompanying significant Joule heating effects during electroforming. In order to further check on the possibility of alloy formation, two areas labeled as "A" and "B" were subjected to EDS analysis. While region A is the pristine Pt tip, region B was in direct contact with the HfO2 surface during the C-AFM measurements, which has been identified by the existence of a small flat surface in region B 23 due to the tip-sample contact. Supplementary Fig. 9g shows that both regions are very similar in composition and only Pt can be detected (besides C and O that are well-known contamination elements from the ambient), while characteristic peaks from Hf or Ti cannot be observed, once again ruling out the possibility of alloy formation.
These conclusions were further testified by WDS analysis in EPMA, whose sensitivity in compositional analysis is about one order of magnitude higher than that of EDS, however at the cost of spatial resolution (Supplementary Ref. 11). Supplementary Fig. 9h shows the analyzed region by EPMA-WDS, where region B in Supplementary Fig. 9f has been included in the analyzed region, and the results from different types of crystals and channels once again exclude the existence of Hf or Ti on the Pt probe ( Supplementary Fig. 9i). Unfortunately, such analysis could not be conducted on the nanoscale craters (~500 nm in diameter), due to the spatial resolution limitation of EPMA as mentioned above.
We believe based on the extensive analysis shown above, it can be safely concluded that due to the extreme programming conditions during the forming process, physical interactions occur between the Pt tip and the HfO2/TiN sample, mainly manifesting as Joule heating effects as suggested by the melted end of the Pt tip, which is also consistent with our TEM observations as discussed in the main text. However, chemical interactions between the probe and the sample, such as alloy formation, seem to be absent. 24

Supplementary Note 10. STEM and EDS characterization on pristine HfO2
Supplementary Fig. 10a shows EDS mapping of Pt M, O K, Hf M, Ti K and N K edges from the pristine state of the HfO2/TiN sample, and Supplementary Fig. 10b further shows the overlaid mapping results. Supplementary Fig. 10c further exhibits the histogram of O K edge signal intensity in the HfO2 film. One can see that in the pristine state of the sample, the concentration distribution of the oxygen element is highly uniform. This is in contrast to the non-uniform distribution of oxygen concentration in the locally switched HfO2 region, as shown in Figs. 5h-o, hence suggesting formation of a conducting filament in the latter case.

Supplementary Note 11. Reversible ion motion in forming and reset processes of HfO2 memristors
The mechanism of reset process in HfO2 memristors was also studied by C-AFM and EFM characterizations, as shown in Fig. 6, and Supplementary Fig. 11 shows the corresponding I-V characteristics. The panels are arranged following the same positions of the simulated regions in Fig. 6a. When a positive voltage of 5 V was applied, Supplementary Figure 11b shows that a relatively high current level of ~200 nA has been reached, therefore indicating the onset of conducting filament formation under 5 V. Figures 6a and 6c also show that when the applied voltage further increased structural distortions will take place. As a result, we have adopted 5 V as the forming threshold. In the second run of experiments (bottom part of Figs. 6a-c), the HfO2/TiN sample was firstly stimulated by a positive voltage sweep to 5 V, which attracts oxygen ions to the top interface, followed by negative voltage sweeps in the same locations with gradually increased negative voltages (0, -1, -3, and -5 V). Backward ion migration was clearly observed in Fig. 6b, where the oxygen ions accumulated at the top interface gradually disappeared. This illustrates the reversible ion motion in forming and reset processes, accounting for the bipolar resistance switching process in HfO2 memristors.