Nanoscale multistate resistive switching in WO3 through scanning probe induced proton evolution

Multistate resistive switching device emerges as a promising electronic unit for energy-efficient neuromorphic computing. Electric-field induced topotactic phase transition with ionic evolution represents an important pathway for this purpose, which, however, faces significant challenges in device scaling. This work demonstrates a convenient scanning-probe-induced proton evolution within WO3, driving a reversible insulator-to-metal transition (IMT) at nanoscale. Specifically, the Pt-coated scanning probe serves as an efficient hydrogen catalysis probe, leading to a hydrogen spillover across the nano junction between the probe and sample surface. A positively biased voltage drives protons into the sample, while a negative voltage extracts protons out, giving rise to a reversible manipulation on hydrogenation-induced electron doping, accompanied by a dramatic resistive switching. The precise control of the scanning probe offers the opportunity to manipulate the local conductivity at nanoscale, which is further visualized through a printed portrait encoded by local conductivity. Notably, multistate resistive switching is successfully demonstrated via successive set and reset processes. Our work highlights the probe-induced hydrogen evolution as a new direction to engineer memristor at nanoscale.

tips is shown to provide a catalytic behaviour in such process, allowing for localized, controllable conduction states throughout the area of the dielectric. The work is well organized and clearly written, though some grammar issues and typos can be found (some were pointed, but careful proofreading is encouraged). Some figures could benefit from small improvements. The methodology is consistent with that of other works on the topic, though some aspects of the characterization should be clarified for reproducibility and some figures of merit should be more carefully addressed (detailed in comments below).
The analysis and structure of the work largely resemble a recent study by the group (Ref. 30) for VO2 on Al2O3 substrate. The main difference (apart from the change of materials to WO3 on STO) is the data reported in Figure 4 demonstrating resistive switching behaviour. Although this could be interpreted as incremental contribution, the results are of potential interest and could be recommended for publication, but a few aspects may be worthy of revising before that stage. Find below some aspects that should be improved, from this reviewer's perspective (these comments are also attached as a .docx file for the authors' convenience). 1) In the introduction, scanning probe memories are mentioned in general, but with no reference provided about their importance or nature. Since it would be a niche for the reported phenomenon, some general description may aid on describing the potential of the technology.
2) STO is known for its great potential for forming atomically abrupt, conductive interfaces with other semiconductors and dielectrics, hence it has been explored in a wide variety of devices ranging from memristors [a,b] to photocatalysis cells [c]. However, in this paper, its role is not well described. This is a bit shocking since one of the main differences between this work and Ref. [30] is the replacement of Al2O3 as bottom electrode of the switching layer for STO. What is the reason for selecting highly doped STO in this case? In photocells, it aids on the provision of electrons to hydrogen ions. Does this characteristic or band alignment play any role here? A more detailed explanation could be helpful to strengthen the study. 3) In supplementary Fig. 1b, a small correlation seems to be observed in the homogeneousness of the map and the regions that were scanned for potentiation and depression. Can this be related to the additional sweep performed in the de-hydrogenated region, that cleans the surface a bit more? 4) In Supp. Fig.2, the authors say: The writing voltage at the middle region was gradually increased from 0 V to 5 V. Does this mean that the voltage is increased as the scan takes place? Is that why the current profile shows higher currents on one side? If so, providing a voltage vs. position graph may help understand and replicate the experiment conditions. 5) A very important aspect of memristor reliability, specially for storage, is the retention time. The authors provide a discussion on the poor retention of oxygen driven switching in O2 environments in Supp. Fig. 4. However, there is no mention to the retention of the reported phenomena, namely the one evaluated in a forming gas atmosphere. This is a fundamental analysis that should be provided, mainly after mentioning that the oxygen counterpart does not provide sufficiently good results. 6) Hydrogen catalytic capability is mentioned for Pd and Pt coated probes but it is not discussed not referenced. Please, provide some details on the fundamentals behind this and/or some literature references to support this.

7)
The "apparent suppression of peak A" in Fig. 2d is not too evident. What's the uncertainty of each measurement? Could this be variability? Namely, if various (say, 10) measurements were performed, would this apparent suppression be consistent in all these? The figure in its current form does not give much supporting information to this specific claim.
9) The claim on the enhanced speed of the hydrogenation process when compared to VO2 samples it is interesting. However, in Ref. 30 this metric is inferred from the peaks in the Raman spectra after different experiments that used different scanning rates. In this work, however, the time is inferred from the measured change in conductivity (in-situ). So, are these two metrics comparable? I would expect both measurements extracted with the same technique for a fair comparison. At least, this should be mentioned in the text when performing such claim. 10) Voltage effects on the area at which hydrogenation-related resistance tunning is observed is a very interesting phenomenon. However, the authors do not suggest any origin for this nor provide systematic measurements showing its behaviour. What is this effect attributed to? Spreading of electric field lines? Lateral diffusivity of hydrogen within the WO3? Since this is being showcased, at least a short comment could provide directions for further research. Figure 3d, while impressive, shows peak currents that are much smaller (around 10 times) than in the rest of the experiments shown in the manuscript. Is there a reason for this? What was the scan rate in this experiment? Is there a correlation between scan rate and resolution (meaning by resolution the lateral dimension from panel c and Supp. Fig. 7)? Please clarify.

11)
12) Fig. 4b shows in the inset its normalized data. Normalized against what magnitude? 13) Current in the loops of Fig. 4d is much higher (easily by 2 orders of magnitude) than the current reported on Fig. 1, even at lower applied voltages. What is the origin of such a large difference? What's the variability of the I-V loops in different regions of the material? If discussing reliability, as suggested by the authors in the next to last paragraph, this is important to be addressed. 14) In the same spirit as comment (12), I do not agree with saying that cycle-to-cycle variation is negligible in Fig. 4c. Rather than variability, and this is an impression without having the raw data, the observed effect seems more like a consistent drift with the accumulation of cycles, similar to the one observed in Fig. 4d for successive cycles. Therefore, this shows a cycle (and therefore, time) dependent degradation of the switching capability. I suggest avoiding the mention "negligible" and rather carefully address the observed drift with the accumulation of cycles/higher applied voltages (mimicking accelerated stress conditions).
15) At the end of next to last paragraph, the authors say they "demonstrate its good reliability". For 500 cycles of a single device, this claim is a bit too strong. It is a promising show of endurance, well displayed for the scope of the work but, from this reviewer's perspective, it doesn't demonstrate "good reliability". 16) See typos and grammar, e.g.: "To direct directly visualize the pattern, …" "The switching behaviors retains remains stable after 500 cycles …" "… in this work descripts describes the potential …" Reviewer #3 (Remarks to the Author): In their work, the authors report an demonstration of a convenient scanning-probe-induced proton evolution within WO3. In principle, the topic of the publication is appropriate for the Nature Communications; however, the paper needs to be revised with minor corrections before being accepted and the authors should take into account the following points:

Methods:
In this section, it should be included the information on the equipment used for RS properties measurements, the measurement ranges and the sensitivity of the equipment. The size of the contacts must also be placed to have a clear relation of the dimensionality of the WO3 and the contacts for the I-V measurements.

RESPONSE:
We thank the reviewer for this question. As requested by the reviewer, we have extended our studies with WO 3 thin films at different thickness (e.g., 10 nm, 50 nm, and 100 nm). As shown in Fig. R1, all samples are hydrogenated within a defined region through the scanning probe, in which a pronounced IMT is observed; while thinner samples have higher conductivity after hydrogen intercalation, since the conductance is inversely proportional to the sample thickness.

2) A typical I-V curve would be good for audience to understand the electric transport behaviour.
RESPONSE: Thanks for the suggestion. Fig. R2 shows I-V curves measured at the as-grown, the hydrogenated and the dehydrogenated regions. From the measurements, the hydrogenated region turns into low-resistance metallic state with characteristic linear I-V curve, while the asgown and dehydrogenated areas remain at high-resistance insulating state. We included this figure in the revised supplementary information. Figure R2. Characteristic I-V curves measured at different states. The black, red and blue lines are the I-V curves measured at the as-grown, the hydrogenated and the dehydrogenated regions, respectively.

3) Authors use the SIMS to compare the hydrogen concentration between the as-grown
sample and hydrogenated sample. It is known second ion ionization rate may depend on the electric states of the sample (insulating and conducting). Comments may be required in the manuscript.

RESPONSE:
We thank the reviewer for this suggestion. We note that the SIMS results provide qualitative evidence that the hydrogen ions were indeed intercalated into the WO3 films using the scanning. Therefore, the distinct hydrogen profiles between the as-grown and hydrogenated samples (shown at Fig. 2a in the main text) directly confirm the existence of hydrogen in the hydrogenated sample with dramatically enhanced SIMS intensity. To reduce the electrostatic charge accumulation during the measurements, the samples are prepared on conducting substrate (i.e., Nb dopped SrTiO3), which is grounded during the measurement. This information was provided in the method section.

RESPONSE:
We thank the reviewer again for this constructive comment. The multiple resistance states are attributed to the change of carrier concentration, which is related to the hydrogen content within WO3. We have previously studied the ionic liquid gating induced hydrogenation in WO3 with the sheet resistance measured as different gating voltages 1 , where the results reveal that the hydrogen concentration forms a direct/efficient tunning knob in modulating the sheet resistance of WO3 thin films.
For conductive filaments induced memristive switching, the size of conducting path underneath the metal electrode is the most crucial factor determining the resistive state of the entire device.
In this case, due to the relatively dilute distribution of the filaments, a high-resolution CAFM measurement would result in an inhomogeneous current map, as shown in a recent study of filament induced IMT 2 . However, our current studies clearly reveal that the multi-resistance patterns manipulated by either hydrogenation voltage, time, or writing number (shown in Supp. The reviewer is correct that the proton gradient along the thickness direction would be an important factor to consider. Indeed, our SIMS result indicates that the hydrogen concentration is higher near the surface of the film, which might be related to the intrinsic ionic diffusion process, in which the top layer would have a higher hydrogen concentration.
We added the related discussion in the revised manuscript.

Reviewer #2 (Remarks to the Author):
This manuscript reports on resistive switching properties of WO3 enabled through scanning probe hydrogenation that triggers an insulator-metal-transition. The use of Pt or Pd scanning microscopy tips is shown to provide a catalytic behaviour in such process, allowing for localized, controllable conduction states throughout the area of the dielectric. We note that the hydrogenation-induced IMT in WO3 has been studied by ionic liquid gating and solid-state proton electrolyte 1, 6-8 . Among those studies, either LaAlO3, YAlO3 or SiO2 is involved as the substrate to fabricate the WO3 films, and similar resistance switching behaviors are observed. Therefore, the hydrogenation and the associated electron doping effect play a dominant role in the resistance switching. However, we agree with the reviewer that band alignment engineering would be an effective path to further optimize the performance of resistance switching, which forms an important and interesting project for future studies. To address the roles of NSTO, the following text was added in the revised manuscript "In this study, we employed the highly doped Nb:STO substrates 9-13 as a bottom electrode to facilitate the tip induced hydrogenation as well as subsequent cAFM measurements." Fig. 1b, a small correlation seems to be observed in the homogeneousness of the map and the regions that were scanned for potentiation and depression. Can this be related to the additional sweep performed in the dehydrogenated region, that cleans the surface a bit more?

RESPONSE:
We note that the slight change in topography (Fig. R3) should be attributed to the hydrogenation-induced lattice expansion. This result is consistent with the results obtained from the TEM analysis, where notable lattice expansion (2.3%) along the c-axis is observed at the hydrogenated region. We included this discussion in the revised supplementary information.

RESPONSE:
The reviewer is correct that the voltage gradually increases during the scanning, as illustrated in Fig. R4d, in which the voltage increases linearly from the left to the right.
Similarly, a voltage template (with negatively biased voltages) was employed for the dehydrogenation process, after which a current map was obtained, as shown in Fig. R4f. We have updated the Supp. Fig. 2 with Fig. R4.  atmosphere. This is a fundamental analysis that should be provided, mainly after mentioning that the oxygen counterpart does not provide sufficiently good results.

RESPONSE:
As suggested by the reviewer, the retention test was performed in the forming gas environment as well, in which a conducting region was continuously measured with the results shown in Fig. R5. It shows that the hydrogenated area sustains a high conductivity state up to 300 minutes after the hydrogen intercalated. At the same time, a notable change in the conducting region was also observed in the horizontal axis, which should be attributed to the tip-induced protonic out-diffusion from the central region. Nevertheless, the hydrogenated sample still shows notably longer retention as compared with the case for the sample with oxygen vacancies, which has a retention time of a few minutes with strongly reduced current.
This information is also included in the revised supplementary information. Figure R5. Retention test of the tip-induced hydrogenation within forming gas.

6) Hydrogen catalytic capability is mentioned for Pd and Pt coated probes but it is not
discussed not referenced. Please, provide some details on the fundamentals behind this and/or some literature references to support this.

RESPONSE:
As suggested, we added the following discussion in the revised manuscript "We note that noble metals of Pt and Pd are the most widely employed catalysts used for hydrogen spillover. 14 However, compared with Pd, Pt has smaller activation energy for the catalytic reaction, meaning it can facilitate hydrogen spillover in a more efficient manner 15 , leading to a more pronounced conducting state with identical writing conditions." Fig. 2d is not  This also applies to the data from the histograms of Fig. 2c. TEM images with high resolution focus on a very small cross section of the device, especially compared to the scanning tip area of influence (not smaller than 75 nm2, as indicated further along the text). Therefore, I'd like to ask how large is the probed area that generates those histograms? Are these results consistent through various images? Also, can the authors provide a reference on the phenomenon driving this observed effect?

7) The "apparent suppression of peak A" in
Perhaps this information could replace Fig. 2d which, from this reviewer's perspective, does not provide much valuable information.

RESPONSE:
We thank the reviewer for this critical comment regarding the STEM measurements. We note that the suppression of peak A is attributed to the electron doping effect associated with hydrogenation. Specifically, the EELS measurement was carried out using the line scan across the length of 1.2 μm (Fig. R6 left), and the presented EELS data in Fig. 2d were the integrated signals through the scans at different regions. To further verify the trend, we also extracted signals from a smaller window of 0.3 μm integrated width (Fig. R6 right), from which almost identical features were obtained.
Regarding the quantitative measurements of lattice constants shown in Fig. 2c, we showed the analyzed area in the Supp. Fig. 6a, where the lattice parameters of pristine and hydrogenated regions are the averaged results of 225 (15 by 15) unit cells. It is important to note that, through this analysis, a notable lattice expansion along the out-of-plane direction was observed, which is consistent with the scenario of hydrogenation induced lattice expansion as observed in previous studies 6, [16][17][18] ; for instance, our previous study 1 of ionic liquid gating induced hydrogenation into WO3 with emergent IMT shows a chemical expansion up to 3.5%, We have added the related discussion in the revised manuscript.

RESPONSE:
We note that both VO2 and WO3 employed in the previous and current studies demonstrate an IMT through hydrogenation induced electron doping. Importantly, the IMT in VO2 is also accompanied with a structural transition, where the V-V dimerization is suppressed, and this feature is employed as an indicator to trace the phase transition through Raman spectra 19 . We note that the lack of bottom electron hinders and complicates the electrical measurement of VO2 across the IMT, while the Raman spectra do not require electrical contact. For the WO3 sample, the electron doping into empty 5d orbitals results in a metallic state, through which the perovskite structure remains. Therefore, the Raman is not an appropriate method to trace this transition. While, luckily the WO3 was grown successfully on a conducting electrode (Nb:STO), in which a direct electrical measurement is feasible. Although different methods have been employed to trace the IMT phase transition of these two systems, the phase transition itself holds a characteristic feature of the materials, which is highly correlated with the intrinsic bulk proton diffusion coefficients, surface facets, etc. Through these studies, although the hydrogenation was carried out at elevated temperature (50 ℃) and higher voltage (10 V) for VO2, its phase transition speed (~2 ms) is still about one order of magnitude longer than that for the WO3, which was carried out at room temperature with a smaller voltage of 3 V. This comparison clearly highlights the promising potential of WO3 for IMT switching devices through proton evolution.

10) Voltage effects on the area at which hydrogenation-related resistance tunning is
observed is a very interesting phenomenon. However, the authors do not suggest any origin for this nor provide systematic measurements showing its behaviour. What is this effect attributed to? Spreading of electric field lines? Lateral diffusivity of hydrogen within the WO3? Since this is being showcased, at least a short comment could provide directions for further research.

RESPONSE:
The reviewer brought out a very nice point here. Indeed, this should be attributed to the stray field induced lateral diffusion. From the geometric feature of the nano-sized scanning probe used, an inhomogeneous electric field (stray field) would form around the hemisphere-shape tip/sample contact regions 20,21 , leading to lateral diffusion of the proton ions.
An intuitive picture is that, with a higher voltage applied between the tip and the bottom electrode, the spreading electric filed would be larger, and stronger lateral diffusion would be induced. However, a microscopic picture behind this would involve several interesting (though complicated) interactions, such as tip induced charge injections 22 , Joule heating 23 , oxygen vacancy 24 and strain field 25,26 , and these factors would couple with the electric field to manipulate the observed phenomena of tip-induced hydrogenation. So, as suggested by the reviewer, we added the following comment about the voltage effect: "The lateral diffusion of hydrogen would be an inevitable effect during the tip-induced hydrogenation process, which can be further manipulated through the stray field around the hemisphere-shape tip/sample contact. 20,21 To better control the hydrogenation resolution, which would benefit high-density data storage, further investigations need to be conducted to investigate electric field induced ionic (protonic) exchange at the tip/sample interface, where the charge injections 22 , Joule heating 23 , oxygen vacancy 24 and even strain field 25,26 could all play an essential role." 11) Figure 3d, while impressive, shows peak currents that are much smaller (around 10 times) than in the rest of the experiments shown in the manuscript. Is there a reason for this? What was the scan rate in this experiment? Is there a correlation between scan rate and resolution (meaning by resolution the lateral dimension from panel c and Supp. Fig. 7)? Please clarify.

RESPONSE:
We thank the reviewer for the question. The writing of Fig. 3d was performed at the scanning rate of 1 Hz and with the voltage template of 0 to 3 V. We note that both the scanning rate and tip voltage during writing play an important role in the observed electric current. In the voltage lithographic template used for Fig. 3d, most of the pixels have a voltage smaller than 3 V (the maximum value). For instance, the voltage used to sketch the hair and face is about 0.8-2.7 V (The writing voltage at the red and blue circle in the Fig. R7 is about 0.87 V and 2.73 V, respectively). We note that this voltage difference is the main reason resulting in the different currents, as there is a close correlation between them, as shown in Fig.   1c. While the resolution is mainly determined by the external voltage, and for instance, the voltage used for Fig. 3c is 3 V, which results in the lateral resolution of 230 nm, while a slightly reduced biased voltage of 2 V (Supp. Fig. 7) leads to the characteristic width of 75 nm for the conducting wires. We note that this trend can be attributed to the stray field induced proton diffusion along the lateral direction during the writing, in which the larger voltage means a stronger stray field, leading to the transport of protons at a longer distance. We have revised the manuscript to clarify these issues.

RESPONSE:
We thank the reviewer for this critical reading and comparison. The data shown in Fig. 4d were measured through the single point electrical measurement with a static tip, in which the current is ~500 nA at the low resistance state set by 4 V and about 20 nA at the intermedia state set by 2 V. While the data in Fig. 1c were patterned and measured by a scanning tip, in which the current is about 220 nA for 4V and about 15 nA for 2 V. We note this difference should be attributed to the different time duration during the writing and possibly the contact area between the sample and tip as well. [27][28][29][30] As suggested by the reviewers, we have also performed multiple I-V loops at several randomly selected regions on the same sample (Fig.   R8), in which the resistive switching phenomena were consistently observed. We have included this information in the revised supplementary information. Figure R8. I-V measurement at 5 randomly chosen spots A-E on the same sample. The last figure is an overlay of these loops.

14)
In the same spirit as comment (12), I do not agree with saying that cycle-to-cycle variation is negligible in Fig. 4c. Rather than variability, and this is an impression without having the raw data, the observed effect seems more like a consistent drift with the accumulation of cycles, similar to the one observed in Fig. 4d for successive cycles. Therefore, this shows a cycle (and therefore, time) dependent degradation of the switching capability. I suggest avoiding the mention "negligible" and rather carefully address the observed drift with the accumulation of cycles/higher applied voltages (mimicking accelerated stress conditions).

RESPONSE:
We thank the reviewer for this critical comment. As suggested, we described the cycle-to-cycle variation as "the I-V curves show a small fluctuation, but with persistent IMT characteristic features." We also mentioned the small accumulated drift observed in Fig. 4d in the revised manuscript: "The characteristic resistive switching remains after 500 cycles with a small accumulated drift, which further verifies the reversibility of the hydrogenation induced IMT.". 15) At the end of next to last paragraph, the authors say they "demonstrate its good reliability". For 500 cycles of a single device, this claim is a bit too strong. It is a promising show of endurance, well displayed for the scope of the work but, from this reviewer's perspective, it doesn't demonstrate "good reliability".

RESPONSE:
We thank the reviewer for this comment. In this revised manuscript, we changed this statement as discussed above.
16) See typos and grammar, e.g.: "To direct directly visualize the pattern, …" "The switching behaviors retains remains stable after 500 cycles …" "… in this work descripts describes the potential …" RESPONSE: We thank the reviewer for the careful review. In the revised submission, we corrected those mistakes and further repolished the writing.

Reviewer #3 (Remarks to the Author):
In their work, the authors report a demonstration of a convenient scanning-probeinduced proton evolution within WO3. In principle, the topic of the publication is appropriate for the Nature Communications; however, the paper needs to be revised with minor corrections before being accepted and the authors should take into account the following points:

Methods:
In this section, it should be included the information on the equipment used for RS properties measurements, the measurement ranges and the sensitivity of the equipment.
The size of the contacts must also be placed to have a clear relation of the dimensionality of the WO3 and the contacts for the I-V measurements.

RESPONSE:
We thank the reviewer for these comments. The electrical measurements were carried out with an SPM setup (model Cypher ES) from Oxford Instruments, which is equipped with an ORCA module for the current measurement. With this setup, the current range is ±10 µA and the sensitivity is ~1 pA. To perform the measurements, commercial Pt-coated conductive tips (model HQ:NSC18/Pt) from MikroMasch were employed. Figure R9 is an SEM image of a typical HQ:NSC18/Pt tip obtained from the vender's website, from which the radius of the tip is estimated to be about 20~30 nm. We note that this value is notably smaller than the characteristic length scale (~75 nm) of the conducting wires obtained in this study, suggesting that the proton diffusion along the lateral direction forms an essential constraint for the experimentally obtained feature size in the current study. We added this information in the revised manuscript.