Heterogeneous integration of single-crystalline rutile nanomembranes with steep phase transition on silicon substrates

Unrestricted integration of single-crystal oxide films on arbitrary substrates has been of great interest to exploit emerging phenomena from transition metal oxides for practical applications. Here, we demonstrate the release and transfer of a freestanding single-crystalline rutile oxide nanomembranes to serve as an epitaxial template for heterogeneous integration of correlated oxides on dissimilar substrates. By selective oxidation and dissolution of sacrificial VO2 buffer layers from TiO2/VO2/TiO2 by H2O2, millimeter-size TiO2 single-crystalline layers are integrated on silicon without any deterioration. After subsequent VO2 epitaxial growth on the transferred TiO2 nanomembranes, we create artificial single-crystalline oxide/Si heterostructures with excellent sharpness of metal-insulator transition (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\triangle \rho /\rho$$\end{document}△ρ/ρ > 103) even in ultrathin (<10 nm) VO2 films that are not achievable via direct growth on Si. This discovery offers a synthetic strategy to release the new single-crystalline oxide nanomembranes and an integration scheme to exploit emergent functionality from epitaxial oxide heterostructures in mature silicon devices.

The authors have examined etching approaches both similar and slightly different from the above manuscripts to make nanomembrane structures with TiO2 and VO2 on top. Given the extensive prior literature in this field, it will be important for the authors to go beyond simply measuring resistance versus temperature as a metric for VO2 growth. The above papers from nearly a decade back all report sophisticated electrically-wired functional devices (eg ionic liquid FETs; multi-state memory etc).
The present manuscript falls short of this clearly, although the authors have indeed made a systematic effort to structurally characterize their films grown on TiO2 by TEM methods and X-ray diffraction. But this aspect is not particularly new or original, as TiO2 is historically one of the chosen substrates for VO2 growth and hundreds of papers on structured/strain etc already exist on this same structure. There is also no particularly new physical property that is striking, despite observing a slightly more narrower IMT curve in thin films, but that is straightforward to engineer by synthesis conditions as I am sure the authors are aware of (since this behavior is not intrinsic to VO2 but due to VO2 growth on TiO2; and why is 10nm special, not clear to the reader?). There is also no thermal cycling or electrical fatigue data presented on the fabricated films and this is certainly essential to validate the process. Otherwise the resistance curve alone is not useful even if it seems sharp for the first switching measurement, since any drift in it will render the claimed applications irrelevant.
In this referee's opinion, original data to demonstrate some viable device or new concept with VO2-TiO2 that is uniquely enabled by some process feature reported in this study (going beyond the above references) will be desired for a major journal along with cycling data to show that the reported fabrication process is reliable and reproducible for making useful VO2 devices.
Reviewer #3 (Remarks to the Author): Lee et al. report on monocrystalline nanomembranes (NM) of TiO2 and its transfer to silicon substrates for use as template for VO2 heteroepitaxy. The VO2 films show very sharp M-I transitions.
The main novelty of the paper is the use of VO2 sacrificial layers (SL), deposited epitaxially on TiO2 substrates and removed by a H2O2 solution. Compared to the popular Sr3Al2O6 (SAO), VO2 SL presents a lower lattice mismatch with TiO2, expanding the range of epitaxial oxides that can be prepared as NM by etching of oxide SLs. In addition, VO2 is much stable to air exposure than SAO, thus overcoming an important practical limitation of SAO.
The results increase the current capabilities of oxide NMs and thus I consider the results to be relevant. However, before recommending publication in Nature Communications, the authors should review the paper considering several points. 1) In Introduction (p. 4, l. 75) the authors state "...the development of oxide NM has been limited for perovskite structure so far1,2,4,14,20,24; to extend the materials spectrum for freestanding oxide NM, new combination of sacrificial layer and etchant needs to be developed..." However, the recent paper cited in ref. 1 (Kum, H. S. et al. Heterogeneous integration of singlecrystalline complex-oxide membranes. Nature 578, 75-81, (2020)) shows that oxide heteroepitaxy is possible when oxide substrates are covered by a few monolayers of graphene. In this reference, high quality oxide NM of perovkites, spinels and garnets are fabricated by this remote epitaxy method. Please revise the written sentence.
2) XRD scans shown in the manuscript were measured using synchrotron radiation. Similar scans would be useful, at least some of them presented as Supplementary Information, (for example to other authors who may try to replicate the fabrication method in the future).
3) The VO2 SL is deposited at 300 ºC. Please indicate, or confirm, if the deposition conditions for VO2 on the transferred TiO2 NM are the same. 4) The indicated area of the NM is "a few millimiters". Can be obtained and transferred NMs of larger area (at least the largest areas usual in standard PLD, around 10x10 or 10x8 mm2)? How does the etching time depend on the area? 5) Can the authors provide more information/results on the mechanical stability of the transferred TiO2 NMs? The mechanical stability could be limited by stress at the VO2/TiO2 interface (f is around 0.86%) or at the TiO2/Si interface (TEC mismatch). 5A) Are VO2 films tens of nm thick on TiO2/Si stable? 5B) Are films (VO2 or other oxides) deposited at higher temperature (example 700 ºC) stable? 6) Can another oxide, different from TiO2, be deposited on VO2 SL, or can it be deposited on TiO2 / VO2 before etching? 7) Can the authors illustrate the potential of the TiO2 NM by depositing other functional oxide(s)? 8) The authors compare the properties of the VO2 film deposited on the transferred NM to VO2 films grown on SiO2/Si. The films on SiO2/Si are polycrystalline and thus the enhancement of properties is not surprising. Can it be compared to VO2 films grown epitaxially directly on buffered Si(001)?

Detailed list of revisions to the manuscript
The following revisions have been made to the manuscript.

About the manuscript
We have revised the following sentences in the manuscript. Response: We thank reviewer #1 for clarifying comments. As the reviewer #1 is concerned, there are residues (e.g., polymers or carbons) from the supporting layer (polydimethylsiloxane (PDMS) or thermal released tape (TRT)) on the surface of transferred TiO2 NM (See Fig.   R1b). However, we would like to point out that these polymer residues on TiO2 NM was perfectly removed using our modified cleaning method based on previously reported RCA cleaning (see ACS Nano, 5, 9144-9153, (2011)).
Below is our detailed procedure for the removal of residues on the surfaces of TiO2 NM Step 1: Transferred TiO2 NM/Si was immersed in acetone (30 minutes), isopropyl alcohol (10 minutes), and de-ionized water (10 minutes) to remove organic residues.
Step 2: To strengthen the bonding between the TiO2 NM and Si substrate and prevent delamination of the TiO2 NM during subsequent cleaning process, transferred TiO2 NM/Si was annealed at 500 ℃ for 3 hours.
Step 3: After thermal annealing, transferred TiO2 NM/Si was immersed in the solution of 5:1:1 H2O / NH4OH / H2O2 at 80 o C for 10 minutes to removing residual organic contaminants.
To compare the surface quality of TiO2 NM during transfer and subsequent cleaning process, we performed atomic force microscopy (AFM) of "as-grown" TiO2/VO2 epitaxial films on the TiO2 single crystal substrates, "as-transferred" TiO2 NM on silicon substrate, and TiO2 NM on silicon substrate "after the cleaning process" (Fig. R1).
After transfer process by using PDMS or TRT, average roughness (Ra) increased from 0.089 nm (Fig. R1a) to 0.180 nm or 0.359 nm (Fig. R1b), respectively, due to the existence of 4 residues on the surface. Our cleaning methods revert Ra down to ~ 0.09 nm by effectively removing these residues (Fig. R1c); the surface of the TiO2 NM on Si is as good as that of asgrown TiO2 epitaxial films on VO2/TiO2; the surface quality is significantly improved by erasing these residues from rigid supporting layers (PDMS or TRT). We would like to emphasize that this cleaning method is critical for growing high-quality epitaxial VO2 films on TiO2 NM/Si substrates (Fig. 4 in the manuscript).

Changes (Page 8 in the manuscript and SI):
We added some sentences on cleaning procedure in the manuscript and AFM images during the cleaning procedure in the Supplementary Figure 9 to share the critical process for epitaxial growth of VO2 films on transferred TiO2 NM.

Figure R1
Atomic force microscopy image of a. "as-grown" TiO2/VO2 epitaxial films on the TiO2 single crystal substrates, b. "as-transferred" TiO2 NM on silicon substrate, c. transferred TiO2 NM on silicon substrate "after our RCA cleaning process". claimed that our work may be limited to only VO2/TiO2 system. However, we would like to emphasize our work is generally applicable for the fabrication of "single-crystalline" rutile oxide nanomembranes (NM).

Figure R3
Heterogeneous integration of "single-crystalline" RuO2 films on TiO2 NM/Si. a, schematic of an epitaxial TiO2/VO2 heterostructure on TiO2 mother substrate. b, The VO2 layer is dissolved in H2O2 to release the top TiO2 film with the mechanical supporting layer (e.g., PDMS and TRT). c, The freestanding TiO2 NM is transferred onto the desired substrates (e.g., silicon). d, By removing the rigid supporting layer, single-crystalline rutile oxide NM is heterogeneously integrated into a silicon substrate. e, epitaxial RuO2 film is grown on the TiO2-NM-templated Si substrates.

8
The cross sectional HAADF-STEM images confirm that the epitaxial growth of 9-nm-thick RuO2 on TiO2 NM can realize heterogeneous integration of single-crystal RuO2 films on a Si substrates. Unlike VO2 films, the RuO2 film appears much brighter than TiO2 NM because the atomic number of Ru (ZRu = 44) is much larger than that of Ti (ZTi = 22). The low magnification image reveals that the RuO2 thin films were uniformly grown on TiO2 NM with ~ 9 nm thickness (Fig. R5 a). The atomic-scale resolution image near the interface between RuO2/TiO2, indicated by yellow square in Fig. R5 a, is shown in Fig. R5 b (zone axis : [100] in both TiO2 and RuO2). The atomic columns of RuO2 and TiO2 are coherently matched at the interface with perfect rutile crystal without any noticeable defects; it means TiO2 NM on Si allows the epitaxial growth of RuO2 films; high-magnification HAADF-STEM image shows perfect registry of Ru atoms in the rutile structure and sharp interface from the contrast difference between RuO2 and TiO2 ( Fig. R5 b). Selective area electron diffraction (SADP) spots on RuO2/TiO2 NM regions show that both RuO2 film and TiO2 NM have high quality of rutile crystal structures (Fig. R5 c). The out-of-plane direction diffraction spots (i.e., 002, indicated by blue square) are slightly separated due to lattice mismatch between RuO2 and TiO2, whereas the in-plane direction diffraction spots (i.e., 040, indicated by red square) are completely overlapped as a single spot, also confirming that epitaxy growth of RuO2 thin films on TiO2 NM. When comparing the lattice parameter c/a(=c/b) ratio calculated from the SADP images with the c/a ratio in the bulk states, there were differences by -0.98% and - The out-of-plane direction diffraction spots (i.e., 002, indicated by blue square) are slightly separated due to lattice mismatch between RuO2 and TiO2, whereas the in-plane direction diffraction spots (i.e., 040, indicated by red square) are completely overlapped as a single spot, also confirming that epitaxy growth of RuO2 thin films on TiO2 NM. When comparing the lattice parameter c/a(=c/b) ratio calculated from the SADP images with the c/a ratio in the bulk states, there were differences by -0.98% and -4.48% in TiO2 and RuO2, respectively. The large decrease of c/a in RuO2 indicates that the RuO2 films are fully strained by biaxial tensile strain along the inplane direction.
Our further experiments during the revision clearly demonstrates the "general" aspect of our approach for the production of single-crystalline NM with rutile oxide crystal system. TiO2 NM may facilitate synthesis of other single-crystalline rutile oxide films such as RuO2, SnO2, NbO2 and IrO2 without any extended defects on the Si substrates and devices As reviewer #1 mentioned, SAO sacrificial layer and remote epitaxy has been suggested for general synthetic methods for single-crystalline NM. First of all, a perovskite oxide membrane was gently released by dissolving water-soluble Sr3Al2O6 sacrificial layers, but the moisturesensitive nature of these layers prevents long-time exposure of the sacrificial layers, which restricts practical application for heterogeneous integration of oxide NM, as reviewer #3 commented. Moreover, the development of oxide NM has been limited for perovskite structure (e.g., SrTiO3, BaTiO3 etc.). Therefore, our study enables to "extend" the materials spectrum for freestanding single-crystalline rutile oxide NM by developing new combination of VO2 sacrificial layer and H2O2 etchant.
In addition, we would like to emphasize that remote epitaxy approach for oxide NM may not be suitable for synthesizing ultrathin oxide NM. To prevent oxidation of interfacial graphene to production of oxides membranes, epitaxial oxide films should be grown under vacuum (< 5 × 10 torr) at initial growth (about 5 -10 nm) (Nature 578, 75-81 (2020)). Oxide film growth under vacuum may generate oxygen vacancies in the oxide films and/or interrupt accurate control of stoichiometric ratio (e.g., the formation of oxygen deficient sub-oxide); the generation of defects in ultrathin regime near the interface could suppress the intrinsic properties of multi-functional oxide films. Moreover, the coalescence of localized nuclei in oxide layers on the graphene, along with restriction of oxygen environment during growth, prevents the layer-by-layer growth of VO2 NM with atomic precision and high quality using remote epitaxy.
Recently, remote epitaxy of VO2 NM was demonstrated by using graphene interfacial layer (Nano Lett. 20, 33-42 (2020)), but VO2 NM exhibited domain boundaries by rotational symmetry mismatch between monoclinic (100)R-VO2 films with a 2/m point group and rhombohedral (0001) Al2O3 substrates with a 3 m point group. The existence of these extended defects prohibits steep metal-insulator transition of VO2 NM by remote epitaxy. In these respects, our approach for single-crystalline rutile oxide NM are "unique" and "generally applicable" for heterogeneous integration of oxide NM with any rutile crystal structures on Si substrates.

Changes (Page 4 and Page 14 in the manuscript and SI):
We hope we fully addressed reviewer #1's constructive comments and we also added a sentence on originality and generality of our fabrication method for "single-crystalline" rutile NM and the figures to Si substrates. In terms of processing, this TiO2 NM is likely to prevent chemical reaction between VO2 and the Si substrates, and allows excellent control over the V-oxidation states without any defects; we would admit that TiO2 NM is indispensable ingredient for the heterogeneous integration of single-crystalline VO2 layers on Si substrates, but the its thickness could be minimized as thin as possible.
More generally, we can absolutely growth other "epitaxial" oxides with rutile structures, and this oxide layers can be released from TiO2 mother substrates and transferred on Si substrates, as long as the oxide are insoluble under dilute H2O2 solution; Among many candidates of rutile oxides, RuO2 was selected to prove the generality of our approach for "single-crystalline" rutile NM; other single-crystalline rutile oxide NMs (e.g., RuO2 or IrO2 NM) as a conducting electrodes, instead of TiO2 NM, could be transferred on Si substrates; VO2/electrode (i.e., RuO2)/Si heterostructures could be realized using the same technique.
For RuO2 electrode NM, we firstly grew 9 nm-thick RuO2 films on the VO2/TiO2 heterostructure under pO2 = 40 mTorr and Tg = 400 o C (Fig. R6 a). Then, we performed symmetric XRD 2θ-ω scans of RuO2/VO2 on TiO2 substrates (Fig. R6 e). As expected, strong (002) RuO2 peak appeared at ~ 47.8 o in the sample. Then, rigid supporting layer was coated on the RuO2/VO2/TiO2 heterostructure and immersed it into H2O2 solution at room temperature. RuO2 NMs were released from the TiO2 substrates by using selective etching of 12 VO2 sacrificial layer ( Fig. R6 b). After fully dissolving VO2 sacrificial template, RuO2 NMs was released, and then transferred on Si substrate ( Fig. R6 c), followed by the removal of rigid supporting layer (Fig. R6 d). Symmetric 2θ-ω scan by using synchrotron XRD showed that (002) RuO2 peak appeared with (400) Si substrate ( Fig. R6 f), which confirms freestanding epitaxial RuO2 film was successfully transferred on the Si substrate.

Figure R6
Heterogeneous integration of RuO2 NM on Si a. schematic of an epitaxial RuO2/VO2 heterostructure on TiO2 mother substrate. b, The VO2 layer is dissolved in H2O2 to release the top RuO2 film with the mechanical supporting layer (e.g., PDMS and TRT). c, The freestanding RuO2 NM is transferred onto the Si substrate d, By removing the rigid supporting layer, single-crystalline RuO2 is heterogeneously integrated into a silicon substrate. Corresponding symmetric 2θ-ω XRD scans of the e. Thus, our strategy to release and transfer a freestanding epitaxial rutile oxide is quite versatile techniques and can be applicable to fabricate a number of combination of heterostructures; this will provide an unprecedented platform for emergent phenomena in various oxide heterostructures with various conductivity (i.e., conductors (e.g., RuO2), semiconductors (e.g., SnO2), insulators (e.g., TiO2) and even emerging superconductivity (e.g., strained RuO2)) to be integrated with mature Si-based heterostructures and devices.

Changes (Page 14 in the manuscript and SI):
We hope we fully addressed reviewer #1's constructive comments and we also added a sentence on generality of our fabrication method The authors have examined etching approaches both similar and slightly different from the above manuscripts to make nanomembrane structures with TiO2 and VO2 on top. Given the extensive prior literature in this field, it will be important for the authors to go beyond simply measuring resistance versus temperature as a metric for VO2 growth. The above papers from nearly a decade back all report sophisticated electrically-wired functional devices (e.g. ionic liquid FETs; multistate memory etc).
Response: First of all, we thank the reviewer #2 for his/her interest on our manuscript. We appreciate the reviewer #2 for his/her valuable time and effort in reviewing our manuscript.
Despite the reviewer #2's interest on our study, he/she seems to doubt the originality of our work based on prior literatures on rutile nanomembranes. However, I would like to emphasize that our VO2/TiO2 nanomembranes are perfect "single-crystalline" without any extended defects. Our work clearly shows the first demonstration of single-crystalline VO2/TiO2 14 heterostructures on Si substrates; exceptional MIT characteristics in terms of transition sharpness (∆ ~ 2.85 K, ∆ ~ 3.3 K) and resistivity ratio (∆ρ ⁄ > 10 3 ) were realized in ultrathin VO2 films integrated on Si substrates, benefitting from the superior quality of epitaxial oxide films.
We address reviewer #2's misunderstanding on our work and, more importantly, emphasize the scientific and technical significance of our study in the following responses, which we failed to deliver to reviewer #2 in our manuscript before revision.
First of all, in this comment, the reviewer #2 claimed that our work followed several literatures with identical goals to make membranes. However, we would like to emphasize that previous literatures that the reviewer #2 referred to did not demonstrate perfect "singlecrystalline" VO2/TiO2 hetero-membranes. These membranes showed relatively poor crystal quality and properties compared to our nanomembranes.  (2005)).
As a result, poor structural quality with a number of defects in TiO2 templates could degrade the sharpness and resistivity modulation of metal-insulator transition in VO2 films on TiO2/MgO (e.g., ∆ρ ⁄ ~ 10 2 , ∆ ~ 12.5 K with VO2 thickness of 70 nm in Fig. 2 from Adv. Mater. 24, 2929-2934 (2012)) compared to our fully epitaxial VO2 films on transferred TiO2 NM/Si (i.e., ∆ρ ⁄ > 10 3 , ∆ ~ 3.3 K with VO2 thickness of 10 nm). Likewise, VO2 membranes grown on the Si3N4 substrate, which is referred from reviewer #2, showed polycrystalline nature of structural quality, which significantly suppressed metal-insulator transition (Nanoscale 4, 7056-7062 (2012), Appl. Phys. Lett. 87, 051910 (2005)); in terms of structural quality, VO2 films on the Si3N4 are similar to directly grown VO2 films on Si substrates (control sample in black line of Fig. 5a), which is completely different from fully epitaxial VO2 films on TiO2 NM/Si substrates. Again, our strategy to release and transfer TiO2 single-crystalline NM enables the integration of nearly perfect single-crystalline VO2 films with controlled thickness on Si substrates, which is never reported before. VO2 film grown on the transferred TiO2 NM had no distinct defects. And low-magnification ABF-STEM image shows that the VO2/TiO2 NM is free of defects even after transfer process from the wide field of view (Fig. R8 a). As a further evidence for exceptional crystalline quality, we scrutinized VO2/TiO2 NM integrated on Si using scanning transmission electron microscope (STEM). The high magnification ABF-STEM image sufficiently verifies the perfect registry of Ti (blue) and oxygen (red) atoms in the rutile structure (Fig. R8 b) in VO2/TiO2 NM. Furthermore, high magnification HAADF-STEM images with the [100] zone axis visualized the structural coherency of the freestanding oxide NM and VO2 film (Fig. R8 c); the atomic resolution image implicates that the VO2 layer is tightly constrained from the underlying TiO2 NM, and thus TiO2 NM templates between VO2 and Si enable growth of epitaxial VO2 films with crystallographic perfection free from defects such as stacking faults and dislocation, which enables steep phase transition at the ultra-thin VO2 film.(i.e., under 10 nm).
Based on these multiple supporting data and arguments on the uniqueness and structural integrity of our "single-crystalline" NMs compared to previously reported membranes, we strongly opposed to reviewer #2's under-appreciation of our work. The present manuscript falls short of this clearly, although the authors have indeed made a systematic effort to structurally characterize their films grown on TiO2 by TEM methods and Xray diffraction. But this aspect is not particularly new or original, as TiO2 is historically one of the chosen substrates for VO2 growth and hundreds of papers on structured/strain etc already exist on this same structure. There is also no particularly new physical property that is striking, despite observing a slightly more narrower IMT curve in thin films, but that is straightforward to engineer by synthesis conditions as I am sure the authors are aware of (since this behavior is not intrinsic to VO2 but due to VO2 growth on TiO2; and why is 10 nm special, not clear to the reader?). There is also no thermal cycling or electrical fatigue data presented on the fabricated films and this is certainly essential to validate the process. Otherwise the resistance curve alone is not useful even if it seems sharp for the first switching measurement, since any drift in it will render the claimed applications irrelevant.

Changes
In this referee's opinion, original data to demonstrate some viable device or new concept with VO2-TiO2 that is uniquely enabled by some process feature reported in this study (going beyond the above references) will be desired for a major journal along with cycling data to show that the reported fabrication process is reliable and reproducible for making useful VO2 devices

Responses:
We strongly disagree with the reviewer #2's remark. Our work focus on the heterogeneous integration of "single-crystalline" functional oxide with steep metalinsulator transition on the Si substrates. Our work is different from previous VO2/TiO2 membranes with numerous defects in the literatures that reviewer #2 mentioned. Although thick TiO2 substrates has been a well-known substrates for VO2 growth, there has been no report on the "epitaxial" growth of VO2 films on freestanding "nanometer-thick" "singlecrystal" TiO2 membranes, which our work demonstrated in the manuscript; our new strategy to release TiO2 NM enables the first demonstration of perfectly "epitaxial" VO2 films integrated on Si substrates. This steeper and more enhanced MIT properties cannot be engineered simply by direct deposition of thin VO2 films on Si or MgO substrates due to the existence of interfacial layers and the substantial density of defects as we already 19 mentioned in our previous response.
Despite numerous effort to integrate "single-crystalline" oxide nanomembranes on silicon, heteroepitaxial growth of oxide films drastically limits the possible materials combinations due to the requirement of lattice matching between epilayer and substrates. In particular, epitaxial growth of VO2 on Si is fundamentally impossible because direct growth of VO2 on Si forms defective and polycrystalline layers. Despite significant efforts to improve structural and electrical characteristics of VO2 thin films using the above buffer layers on Si, we would like to emphasize that high resistivity modulation (∆ ⁄ ) is difficult to be achieved in ultrathin regime of VO2 thickness due to a number of interfacial defects and domain boundaries between buffer layer and VO2 film (see Comment 10 from reviewer #3). Our work clearly demonstrates the heterogeneous integration of "single-crystalline" oxide films with steep metal-insulator transition on Si; this work will offer unique way to integrate the emergent phenomena of epitaxial oxide films with mature electronic and photonic devices.
Previously, for "single-crystalline" nanomembranes, SAO sacrificial layer and remote epitaxy has been suggested for general synthetic methods for "single-crystalline" NM (Nat. Mater. 15, (2016)). First of all, a perovskite oxide membrane was gently released by dissolving water-soluble Sr3Al2O6 sacrificial layers, but the moisture-sensitive nature of these layers prevents long-time exposure of the sacrificial layers, which restricts practical application for heterogeneous integration of oxide NM. Moreover, the development of oxide NM has been limited for perovskite structure (e.g., SrTiO3, BaTiO3 etc.). Therefore, our study enables to extend the materials spectrum for freestanding single-crystalline rutile oxide NM by developing new combination of VO2 sacrificial layer and H2O2 etchant.

1255-1260
In addition, remote epitaxy of VO2 NM was recently demonstrated by using graphene interfacial layer (Nano Lett. 20, 33-42 (2020)), but VO2 NM exhibited domain boundaries by rotational symmetry mismatch between monoclinic (100)R-VO2 films with a 2/m point group and rhombohedral (0001) Al2O3 substrates with a 3 m point group. The existence of these extended defects prohibits steep metal-insulator transition of VO2 NM by remote epitaxy.
Furthermore, the coalescence of localized nuclei in oxide layers on the graphene and restriction of oxygen environment during growth prevents the layer-by-layer growth of VO2 20 NM with atomic precision and high quality using remote epitaxy (Nature 578, 75-81 (2020)).
In these respects, our approach for single-crystalline rutile oxide NM are unique and generally applicable for heterogeneous integration of oxide NM with any rutile crystal structures on Si substrates.
Despite our strong disagreement of reviewer #2's claim on the originality of our work, we agree with the reviewer #2's suggestion on thermal and electrical cycling of "singlecrystalline" VO2 films on TiO2 NM/Si, which is certainly essential to validate the process for the possible application of integrated VO2 films on Si. First of all, we performed the repeated thermal cycling by measuring temperature-dependent sheet resistance at several times (270 K ~ 320 K) as shown in Fig. R9 a. Obviously, steep sheet resistance modulation with metalinsulator transition across TMI ~ 296 K (∆ρ ⁄ ~ 10 3 ) were observed in 10 nm-thick VO2 on TiO2 NM/Si more than 100 cycles (Fig. R9 b). No drift was observed during the multiple cycle of thermal switching of metal-insulator transition in VO2 on TiO2 NM/Si Figure R9 a. Repeated measurement of temperature-dependent resistance measurement from 270 K to 320 K during 5 cycles. b. thermal cycling at 278 K and 308 K until 100 cycles in the 10 nmthick VO2 / 70 nm-thick TiO2 NM.
In addition to thermal cycling, we also characterized threshold switching (i.e., voltagetriggered insulator-to-metal transition) after the fabrication of two-terminal devices with 5 μm 21 of lateral dimension between two Au/Ti electrodes (Fig. R10 a). Uniform and abrupt voltagetriggered threshold switching (i.e., sudden increase of current at threshold voltage) was observed in epitaxial VO2 films integrated on Si substrates (Fig. R10 b); this observation indicates that reversible electric-field-induced switching from insulating phase to metallic phase occurs at consistent threshold voltage (Vth ~ 5.7 V) in the forward direction and subsequent metal-to-insulator transition at hold voltage (Vh ~ 0.5 V) in the reverse direction of sweep. This two terminal devices with "single-crystalline" showed high reliable selectivity of on and off states more than 100 cycles, which shows excellent the endurance of Ion/Ioff ratio, Vth and Vh during DC bias sweep (Fig. R10 c, d), which can be potentially applicable for selectors and steep-slope switches in the future. Again, No drift was observed during the multiple cycle of electrical switching of metal-insulator transition in VO2 on TiO2 NM/Si Although we did not focus on the device performance as an originality of our manuscript, we sufficiently proves that our single crystal VO2 film integrated on the Si substrate was high endurance during thermal and electrical cycling as the reviewer #2 suggested. Figure R10 a. Plane view of two-terminal threshold switch with single crystalline VO2/TiO2 NM on Si obtained by optical microscopy. b. the I-V characteristic of two-terminal single crystal VO2 device consisted of Au/Ti electrodes on the VO2/TiO2 NM with 5 um channel length. c. the DC endurance of the singe crystal VO2 device, which showed stable resistivity change. d. homogeneous Vth and Vh changes during DC bias endurance more than 100 cycles.

Changes (Page 12, SI) : We added some sentences in the manuscript and figures in the
Supplementary Figures 14 and 15  We also recognized that high quality oxide NM (perovskites, spinels and garnets, not rutile) are fabricated by this remote epitaxy method as reported in the referred literature from reviewer #3 (Nature 578, 75-81 (2020)). However, we would like to emphasize that remote epitaxy approach for oxide NM may not be suitable for synthesizing ultrathin oxide NM. To prevent oxidation of interfacial graphene to production of oxides membranes, epitaxial oxide films should be grown under vacuum (< 5 × 10 torr) at initial growth (about 5 -10 nm) (Nature 578, 75-81 (2020)). Oxide film growth under vacuum may generate oxygen vacancies in the oxide films and/or interrupt accurate control of stoichiometric ratio (e.g., the formation of oxygen deficient sub-oxide); the generation of defects in ultrathin regime near the interface could suppress the intrinsic properties of multi-functional oxide films. Moreover, the coalescence of localized nuclei in oxide layers on the graphene, along with restriction of oxygen environment during growth, prevents the layer-by-layer growth of VO2 NM with atomic precision and high quality using remote epitaxy.
Recently, remote epitaxy of rutile VO2 NM was demonstrated by using graphene interfacial layer in another literature (Nano Lett. 20, 33-42 (2020)), but VO2 NM exhibited domain boundaries by rotational symmetry mismatch between monoclinic (100) In these respects, our approach for single-crystalline rutile oxide NM are unique and shows some advantage in some aspect compared to remote epitaxy; our approach would be generally applicable for heterogeneous integration of oxide NM with any rutile crystal structures on Si substrates.

Changes (Page 4 in the manuscript):
Following the reviewer #3's suggestion, we have revised the sentence in the Introduction of the manuscript as "the development of oxide NM has been limited for perovskite structure among chemical lift-off methods so far".

 Comment 3
2) XRD scans shown in the manuscript were measured using synchrotron radiation. Similar scans would be useful, at least some of them presented as Supplementary Information, (for example to other authors who may try to replicate the fabrication method in the future).

Response:
We thank reviewer #3 for very constructive suggestion to improve our manuscript.
Before synchrotron high-resolution x-ray scattering measurement, we also measured heterogeneous integration of single-crystalline TiO2 NM on silicon substrate using in-house HRXRD with Cu Kα1 radiation.

Changes (Page 7 in the manuscript, SI):
We revised the sentence in the manuscript and added a figure (Fig. R11)  3) The VO2 SL is deposited at 300 ºC. Please indicate, or confirm, if the deposition conditions for VO2 on the transferred TiO2 NM are the same.

Response:
We apologize the lack of our explanation and thank the reviewer #3 for missing information of experimental details. Obviously, this information should be included in the manuscript. The epitaxial VO2 film was successfully grown on the TiO2 NM/Si with an identical growth condition of that of VO2 sacrificial layers on the (001) TiO2 substrate.

Changes (Page 15 in the manuscript):
Based on the reviewer #3's suggestions, we added the sentence in the Method Section that the growth conditions for VO2 films on TiO2 NM/Si is identical with that for VO2 sacrificial layer on TiO2 substrates before the release of NM.

28
 Comment 5 4) The indicated area of the NM is "a few millimiters". Can be obtained and transferred NMs of larger area (at least the largest areas usual in standard PLD, around 10x10 or 10x8 mm 2 )? How does the etching time depend on the area?

Response:
The reviewer #3 asked the achievable area of NMs and the area dependence of etching time for the removal of VO2 sacrificial layers.
As the reviewer #3 commented, the lateral size of NMs is determined by the size of TiO2 substrates, as long as laterally uniform growth is allowed at optimized condition of PLD.
Since the laser-induced plasma plume could be uniformly distributed for up to 5 mm × 5 mm TiO2 substrates in the optimized growth condition, the largest lateral area of NMs was 5 mm × 5 mm in our case. However, the size could be further increased in case of the increased substrate size and larger area deposition technique (e.g., sputtering or MOCVD).
On the area dependence of etching time for the removal of VO2 sacrificial layers, we estimated the release time of TiO2 NMs attached on supporting layer from as-grown TiO2 (70 nm)/VO2 (15 nm)/TiO2 heterostructures with various lateral size (1 mm × 1 mm, 2.5 mm × 2.5 mm, 5 mm × 5 mm); as-grown TiO2/VO2/TiO2 was immersed in dilute H2O2 solution (50 ml H2O2 + 150 ml H2O). Since un-reacted TiO2 epitaxial layers cover whole surfaces of VO2 sacrificial layers during release process, release of epitaxial TiO2 NM by oxidation and dissolution of VO2 sacrificial layers begins from the edge of the substrate (i.e., side etching). Therefore, the etching time increases in proportion to the size of the substrate as shown in

Figure R15
a. repeated temperature-dependent resistance measurement from 270 K to 320 K during 5 cycles. b. thermal cycling at 278 K and 308 K until 100 cycles in the 10 nm-thick VO2 / 70 nm-thick TiO2 NM.
Despite the interesting phenomena on the bendable NM by misfit strain between 70-nm-thick VO2 and TiO2, we would like to emphasize that 10 nm-thick VO2 film grown on the 70 nmthick TiO2 NM is mechanically stable even after repetitive thermal cycling. We performed the repeated thermal cycling by measuring temperature-dependent sheet resistance at several times (270 K ~ 320 K). Obviously, steep sheet resistance modulation with metal-insulator transition across TMI ~ 296 K (∆ρ ⁄ ~ 10 3 ) were observed in 10 nm-thick VO2 on TiO2 NM /Si more than 100 cycles (Fig. R15); this result confirms that 10 nm-thick VO2 on TiO2 on the 70 nm-thick TiO2 NM mechanically robust without degradation.

Changes (Page 12 in the manuscript and SI):
We hope we fully addressed reviewer #3's comment on the mechanical stability of VO2 on TiO2 NM, and we also added these sentence in the revised manuscript and figures in the Supplementary Figure 14.
The peak position of (002) RuO2 on TiO2 NM/Si is identical with that of (002) RuO2 on TiO2 substrates; this result indicates that RuO2 film can be epitaxially grown at 600 o C since TiO2 NM is stable without degradation at higher temperature even under oxidizing atmosphere, where graphene interlayer is not stable for remote epitaxy.

Changes (Page 14 in the manuscript):
We hope we fully addressed reviewer #3's comment and we also added the sentences on the thermal stability of single-crystalline TiO2 NM.

Figure R16
Schematic of a. RuO2 film grown on the TiO2 NM / Si substrate and c. RuO2 film grown on the (001) TiO2 substrate at substrate temperature of 600 o C and corresponding symmetric 2θ-ω scans of b. RuO2 / TiO2 NM / Si and d. RuO2 / (001) TiO2 substrate. The location of (002) peak for RuO2 films grown at 600 o C on TiO2 NM/Si was almost identical that of epitaxial RuO2 films grown at 600 o C on (001)  candidates of rutile oxides, RuO2 NM was selected to prove the generality of our approach for "single-crystalline" rutile oxide NM.
Following the reviewer #3's suggestion, we firstly grew 9 nm-thick RuO2 films on the VO2/TiO2 heterostructure under pO2 = 40 mTorr and Tg = 400 o C (Fig. R17 a). Then, we performed symmetric XRD 2θ-ω scans of RuO2/VO2 on TiO2 substrates (Fig. R17 e). As expected, strong (002) RuO2 peak appeared at ~ 47.8 o . Then, rigid supporting layer was coated on the RuO2/VO2/TiO2 heterostructure and immersed it into H2O2 solution at room temperature; RuO2 nanomembranes were released from the TiO2 substrates by using selective etching of VO2 sacrificial layer (Fig. R17 b). After fully dissolving VO2 sacrificial template, RuO2 nanomembranes was released, and then transferred on Si substrate (Fig. R17 c), followed by the removal of rigid supporting layer (Fig. R17 d). Symmetric 2θ-ω scan by using synchrotron XRD showed that (002) RuO2 peak appeared with (400) Si substrate (Fig. R17 e), which confirms freestanding epitaxial RuO2 film was successfully transferred on the Si substrate. Figure R17 a. schematic of an epitaxial RuO2/VO2 heterostructure on TiO2 mother substrate. b, The VO2 layer is dissolved in H2O2 to release the top RuO2 film with the mechanical supporting layer (e.g., PDMS and TRT). c, The freestanding RuO2 NM is transferred onto the Si substrate d, By removing the rigid supporting layer, single-crystalline RuO2 is heterogeneously integrated into a silicon substrate. Corresponded symmetric 2θ-ω XRD scans of the e. Response: We thanks the reviewer #3 for very constructive suggestion. The reviewer #3 suggested that the "general" potential of our approach for the production of TiO2 NM needs to be illustrated by depositing other functional oxides, as well as VO2, which is the essentially same suggestion with reviewer #1. We would like to emphasize that our work is generally applicable for the fabrication of single-crystalline rutile oxide nanomembranes. Figure R18 a, schematic of an epitaxial TiO2/VO2 heterostructure on TiO2 mother substrate. b, The VO2 layer is dissolved in H2O2 to release the top TiO2 film with the mechanical supporting layer (e.g., PDMS and TRT). c, The freestanding TiO2 NM is transferred onto the Si substrate. d, By removing the rigid supporting layer, single-crystalline TiO2 NM is heterogeneously integrated into a silicon substrate. e, epitaxial RuO2 film is grown on the TiO2-NM-templated Si substrates.
To generally demonstrate the heterogeneous integration of other rutile oxide single-crystalline on Si, we grew epitaxial RuO2 films on TiO2 NM/Si (Fig. R18). After the cleaning process to remove the residue on transferred TiO2 NM, 8.5-nm-thick RuO2 thin films, instead of VO2 films, were grown on TiO2 NM/Si by pulsed laser deposition (Fig. R18e). As observed in symmetrical XRD 2θ-ω scans, the intense (002) RuO2 peak appeared at ~ 2θ = 47.8° ( Fig.   39 R19 b), along with peaks related to the TiO2 template (~ 2θ = 49.5°) and Si substrates (~ 2θ = 54.3°). The location of (002) RuO2 peak on TiO2 NM/Si was almost identical that of epitaxial RuO2 films grown on (001) TiO2 substrates (Fig. R19 d); this result implicates that TiO2 NM templates facilitate the formation of epitaxial RuO2 films on Si substrates as well.
The cross sectional HAADF-STEM images confirm that the epitaxial growth of 9-nm-thick RuO2 on TiO2 NM can realize heterogeneous integration of single-crystal RuO2 films on a Si substrates. The RuO2 film appears much brighter than TiO2 NM because the atomic number of Ru (ZRu = 44) is much larger than that of Ti (ZTi = 22). The low magnification image reveals 40 that the RuO2 thin films were uniformly grown on TiO2 NM (Fig. R20 a). The atomic-scale resolution image near the interface between RuO2/TiO2, indicated by yellow square in a, is shown in Fig. RX 20  The sharp diffraction spots shows both RuO2 film and TiO2 NM have high quality of rutile crystal structures. The out-of-plane direction diffraction spots (i.e., 002, indicated by blue square) are slightly separated due to lattice mismatch between RuO2 and TiO2, whereas the in-plane direction diffraction spots (i.e., 040, indicated by red square) are completely overlapped as a single spot, also confirming that epitaxy growth of RuO2 thin films on TiO2 NM. When comparing the lattice parameter c/a(=c/b) ratio calculated from the SADP images with the c/a ratio in the bulk states, there were differences by -0.98% and -4.48% in TiO2 and RuO2, respectively. The large decrease of c/a in RuO2 indicates that the RuO2 films are fully strained by biaxial tensile strain along the inplane direction.
Besides, to experimentally verify the local strain analysis of RuO2 / TiO2 NM / Si substrates, we performed the geometric phase analysis (GPA) strain quantification using STEM techniques (Fig. R21 a). In-plane (εIP) and out-of-plane (εOOP) lattice strain mapping were obtained from geometric phase analysis (GPA) of HAADF-STEM image. Line profile of lattice strain, εIP and εOOP, on the RuO2 films were extracted from GPA strain mapping (blue box in Fig. R21 b). The lattice strains were calculated based on the lattice parameter of reference region in TiO2 NM (yellow box in Fig. R21 a). In-plane lattice parameters were almost invariable across the RuO2/TiO2 NM (Fig. R21 b, c), indicating that the RuO2 thin films were completely constrained from the TiO2 NM-template. On the other hands, out-ofplane lattice parameters of RuO2 films were about 4.5% larger than that of TiO2 NM (Fig.   R21 b, c), and these are also uniform over the entire area of the films.
From our results, we clearly shows that "single crystalline" TiO2 NM had promising template to integrate other "single-crystalline" functional oxide with rutile structure on Si substrates.

Changes (Page 14 in the manuscript and SI):
We hope that we fully addressed reviewer #3's comment about "general" potential of TiO2 NM on our manuscript and we also added a sentence in the revised manuscript and related figures in Supplementary Figures 17-19.

Figure R21
Local strain analysis of RuO2/TiO2 NM on SiO2/Si substrate. a, HAADF-STEM image of RuO2 thin films on the TiO2 NM-template. b, In-plane (εIP) and out-of-plane (εOOP) lattice strain mapping obtained from geometric phase analysis (GPA) of HAADF-STEM image a. c, Line profile of lattice strain, εIP and εOOP, on the RuO2 films extracted from GPA strain mapping b (blue box). These lattice strains were calculated based on the lattice parameter of reference region in TiO2 NM (yellow box in a). The interface between RuO2/TiO2 NM is identified by the orange dotted lines in a, b, and c. In-plane lattice parameters almost invariable across the RuO2/TiO2 NM, indicating that the RuO2 thin films were completely constrained from the TiO2 NM-template. On the other hands, out-of-plane lattice parameters of RuO2 films are about 4.5% larger than that of TiO2 NM, and these are also uniform over the entire area of the films. It is slightly different with bulk state, where the gap of out-of-plane lattice parameter between RuO2 and TiO2 is about 5.4%, because the RuO2 films were tensile-strained along the in-plane direction.

 Comment 10
8) The authors compare the properties of the VO2 film deposited on the transferred NM to VO2 films grown on SiO2/Si. The films on SiO2/Si are polycrystalline and thus the enhancement of properties is not surprising. Can it be compared to VO2 films grown epitaxially directly on buffered Si(001)?
Despite significant efforts to improve structural and electrical characteristics of VO2 thin films using the above buffer layers on silicon, we would like to emphasize that high resistivity modulation (∆ ⁄ ) is difficult to be achieved in ultrathin regime of VO2 thickness due to fundamental limitation of this approach. As shown in previously reported literatures (APL Mater. 4, 026101 (2016)), the degree of high resistivity modulation (∆ ⁄ ) across the metal-44 insulator transition was strongly suppressed as VO2 thickness decreased from 130 nm to 50 nm. (No temperature-dependent MIT result was reported in thinner VO2 films (< 50 nm) on AlN-buffered Si, probably due to inferior ∆ ⁄ modulation). Indeed, we grew VO2 films on the Al2O3 buffered Si substrate with various thickness (10 nm ~ 150 nm) for a direct comparison. The ∆ ⁄ in 10-nm-thick VO2 films (∆ ⁄ ~ 1.50) were considerably reduced compared with that of 150-nm-thick VO2 film on Al2O3-buffered/Si substrate (∆ ⁄ ~ 4.80 × 10 ) (Fig. R22a); The ∆ ⁄ in 10-nm-thick "single-crystalline" VO2 films on TiO2 NM/Si (∆ρ ⁄ ~ 3.3 × 10 3 ) shows more than 3 orders of magnitude higher than that in 10-nm-thick VO2 films on Al2O3-buffered/Si substrate (∆ ⁄ ~ 1.50). In the case of VO2 films on buffered Si, a further reduction in ∆ ⁄ was found to occur with decreasing film thickness, likely due to a number of interfacial defects and domain boundaries between buffer layer and VO2 film; By fair comparison with buffered Si based on reviewer #3's suggestion, our work clearly demonstrates that exceptional MIT characteristics in terms of transition sharpness (∆ ~ 2.85 K, ∆ ~ 3.3 K) and resistivity ratio (∆ρ ⁄ > 10 3 ) were realized in ultrathin VO2 films integrated on Si substrates, benefitting from the superior quality of epitaxial oxide films.

Changes (Page 12 in the manuscript and SI):
We hope that we fully addressed reviewer #3's comment on exceptional properties of the VO2 epitaxial layers grown on the TiO2 NM/Si, even compared to buffered Si substrates, on the revised manuscript. To demonstrate more clearly, we also added a sentence in the revised manuscript and a figure in Supplementary   Figure 16.