•Classical Frank-van der Merwe mode is realized in bilayer graphene growth•We have invented machine-learning-assisted Raman analysis tool to characterize graphene•We developed a support-free graphene transfer approach based on Marangoni flows Bilayer graphene is now a rising star in the discoveries in unconventional physics. While the number of exciting physical phenomena observed in bilayer graphene increases, a big gap persists in transforming these discoveries into useful applications, owing to the small-scale samples obtained via a top-down approach. We realized a layer-by-layer (that is, Frank-van der Merwe) growth mode in large-scale bilayer graphene, with no island impurities, which is unprecedented in any van der Waals-stacked materials. Our study showcases the intriguing layer-by-layer growth, machine-learning-assisted materials characterization, and support-free transfer of bilayer graphene and paves the way for scalable bilayer graphene applications. Bilayer graphene has attracted significant interest because of its unique properties, including fascinating electrical behavior when one layer is slightly rotated relative to the other. However, the quality of large-area bilayer graphene is often limited by the layer-plus-island growth mode in which islands of thicker graphene present as unavoidable impurities. Here, we report the observation of the layer-by-layer, Frank-van der Merwe (FM) growth mode in bilayer graphene where multilayer impurities are suppressed. Instead of the conventional surface adhesive energy, we found it possible to tune interface adhesive energy with an oxidative pretreatment. The FM-grown bilayer graphene is of AB stacking or with small twisting angle (θ = 0°–5°), which is more mechanically robust compared with monolayer graphene, facilitating a free-standing wet transfer technology. Bilayer graphene has attracted significant interest because of its unique properties, including fascinating electrical behavior when one layer is slightly rotated relative to the other. However, the quality of large-area bilayer graphene is often limited by the layer-plus-island growth mode in which islands of thicker graphene present as unavoidable impurities. Here, we report the observation of the layer-by-layer, Frank-van der Merwe (FM) growth mode in bilayer graphene where multilayer impurities are suppressed. Instead of the conventional surface adhesive energy, we found it possible to tune interface adhesive energy with an oxidative pretreatment. The FM-grown bilayer graphene is of AB stacking or with small twisting angle (θ = 0°–5°), which is more mechanically robust compared with monolayer graphene, facilitating a free-standing wet transfer technology. High-quality synthesis, systematic characterization, and clean transfer of large-area bilayer graphene (2LG) are important for transforming the unusual physical properties of 2LG into disruptive technologies.1Ohta T. Bostwick A. Seyller T. Horn K. Rotenberg E. Controlling the electronic structure of bilayer graphene.Science. 2006; 313: 951-954Crossref PubMed Scopus (2770) Google Scholar, 2Mayorov A.S. Elias D.C. Mucha-Kruczynski M. Gorbachev R.V. Tudorovskiy T. Zhukov A. Morozov S.V. Katsnelson M.I. Fal’ko V.I. Geim A.K. et al.Interaction-driven spectrum reconstruction in bilayer graphene.Science. 2011; 333: 860-863Google Scholar, 3Gao Y. Cao T. Cellini F. Berger C. de Heer W.A. Tostatti E. Riedo E. Bongiorno A. Ultrahard carbon film from epitaxial two-layer graphene.Nat. Nanotechnol. 2018; 13: 133Google Scholar Specifically, small twisting angle 2LG (θ = 0°–5°) exhibits fascinating properties, including superconductivity4Yankowitz M. Chen S. Polshyn H. Zhang Y. Watanabe K. Taniguchi T. Graf D. Young A.F. Dean C.R. Tuning superconductivity in twisted bilayer graphene.Science. 2019; 363: 1059-1064Crossref PubMed Scopus (931) Google Scholar, 5Cao Y. Fatemi V. Fang S. Watanabe K. Taniguchi T. Kaxiras E. Jarillo-Herrero P. Unconventional superconductivity in magic-angle graphene superlattices.Nature. 2018; 556: 43Crossref PubMed Scopus (3325) Google Scholar, 6Codecido E. Wang Q. Koester R. Che S. Tian H. Lv R. Tran S. Watanabe K. Taniguchi T. Zhang F. Bockrath M. et al.Correlated insulating and superconducting states in twisted bilayer graphene below the magic angle.Sci Adv. 2019; 5: eaaw9770Crossref PubMed Scopus (96) Google Scholar, 7Sharpe A.L. Fox E.J. Barnard A.W. Finney J. Watanabe K. Taniguchi T. Kastner M.A. Goldhaber-Gordon D. Emergent ferromagnetism near three-quarters filling in twisted bilayer graphene.Science. 2019; 365: 605-608Crossref PubMed Scopus (657) Google Scholar and electrically tunable band gaps8Zhang Y. Tang T.T. Girit C. Hao Z. Martin M.C. Zettl A. Crommie M.F. Shen Y.R. Wang F. Direct observation of a widely tunable bandgap in bilayer graphene.Nature. 2009; 459: 820Crossref PubMed Scopus (2827) Google Scholar, 9Tang T.T. Zhang Y. Park C.H. Geng B. Girit C. Hao Z. Martin M.C. Zettl A. Tang T.T. Crommie M.F. Louie S.G. et al.A tunable phonon–exciton Fano system in bilayer graphene.Nat. Nanotechnol. 2010; 5: 32Google Scholar, 10Zhou S.Y. Gweon G.H. Fedorov A.V. First P.N. de Heer W.A. Lee D.H. Guinea F. Castro Neto A.H. Lanzara A. Substrate-induced bandgap opening in epitaxial graphene.Nat. Mater. 2007; 6: 770Google Scholar compared with other 2LG with larger twisting angles.11Zhang J. Zhao J. Mechanical properties of bilayer graphene with twist and grain boundaries.J. Appl. Phys. 2013; 113: 043514Google Scholar To further transfer these physical observations into disruptive innovations, many efforts have been made to target large-scale high-quality 2LG synthesis using chemical vapor deposition (CVD).12Liu L. Zhou H. Chen R. Yu W.J. Liu Y. Chen Y. Shaw J. Zhong X. Huang Y. Duan X. High-yield chemical vapor deposition growth of high-quality large-area AB-stacked bilayer graphene.ACS Nano. 2012; 6: 8241-8249Crossref PubMed Scopus (217) Google Scholar, 13Liao L. Wang H. Peng H. Yin J. Koh A.L. Chen Y. Xie Q. Peng H. Liu Z. van Hove singularity enhanced photochemical reactivity of twisted bilayer graphene.Nano Lett. 2015; 15: 5585-5589Crossref PubMed Scopus (44) Google Scholar, 14Yin J. Wang H. Peng H. Tan Z. Liao L. Lin L. Sun X. Koh A.L. Chen Y. Peng J. et al.Selectively enhanced photocurrent generation in twisted bilayer graphene with van Hove singularity.Nat. Commun. 2016; 7: 1-8Google Scholar Significant progress has been made with CVD 2LG synthesis,15Ma W. Chen M.L. Yin L. Liu Z. Li H. Xu C. Xin X. Sun D.M. Cheng H.M. Ren W. Interlayer epitaxy of wafer-scale high-quality uniform AB-stacked bilayer graphene films on liquid Pt 3 Si/solid Pt.Nat. Commun. 2019; 10: 2809Crossref PubMed Scopus (29) Google Scholar, 16Reina A. Thiele S. Jia X. Bhaviripudi S. Dresselhaus M.S. Schaefer J.A. Kong J. Growth of large-area single- and bi-layer graphene by controlled carbon precipitation on polycrystalline Ni surfaces.Nano Res. 2009; 2: 509-516Google Scholar, 17Wu Y. Chou H. Ji H. Wu Q. Chen S. Jiang W. Hao Y. Kang J. Ren Y. Piner R.D. et al.Growth mechanism and controlled synthesis of AB-stacked bilayer graphene on Cu–Ni alloy foils.ACS Nano. 2012; 6: 7731-7738Crossref PubMed Scopus (141) Google Scholar, 18Huang M. Bakharev P.V. Wang Z.J. Biswal M. Yang Z. Jin S. Wang B. Park H.J. Li Y. Qu D. et al.Large-area single-crystal AB-bilayer and ABA-trilayer graphene grown on a Cu/Ni (111) foil.Nat. Nanotechnol. 2020; 15: 289-295Crossref PubMed Scopus (76) Google Scholar particularly, wafer-scale growth of uniform AB-stacked bilayer graphene film reported recently on a liquid Pt3Si/solid Pt substrate15Ma W. Chen M.L. Yin L. Liu Z. Li H. Xu C. Xin X. Sun D.M. Cheng H.M. Ren W. Interlayer epitaxy of wafer-scale high-quality uniform AB-stacked bilayer graphene films on liquid Pt 3 Si/solid Pt.Nat. Commun. 2019; 10: 2809Crossref PubMed Scopus (29) Google Scholar and AB-stacked bilayer or multilayer graphene (MLG) larger than a centimeter using a Cu/Ni(111) substrate.18Huang M. Bakharev P.V. Wang Z.J. Biswal M. Yang Z. Jin S. Wang B. Park H.J. Li Y. Qu D. et al.Large-area single-crystal AB-bilayer and ABA-trilayer graphene grown on a Cu/Ni (111) foil.Nat. Nanotechnol. 2020; 15: 289-295Crossref PubMed Scopus (76) Google Scholar Some of these methods are via controlling the cooling to let the dissolved carbon precipitate out,15Ma W. Chen M.L. Yin L. Liu Z. Li H. Xu C. Xin X. Sun D.M. Cheng H.M. Ren W. Interlayer epitaxy of wafer-scale high-quality uniform AB-stacked bilayer graphene films on liquid Pt 3 Si/solid Pt.Nat. Commun. 2019; 10: 2809Crossref PubMed Scopus (29) Google Scholar, 16Reina A. Thiele S. Jia X. Bhaviripudi S. Dresselhaus M.S. Schaefer J.A. Kong J. Growth of large-area single- and bi-layer graphene by controlled carbon precipitation on polycrystalline Ni surfaces.Nano Res. 2009; 2: 509-516Google Scholar, 17Wu Y. Chou H. Ji H. Wu Q. Chen S. Jiang W. Hao Y. Kang J. Ren Y. Piner R.D. et al.Growth mechanism and controlled synthesis of AB-stacked bilayer graphene on Cu–Ni alloy foils.ACS Nano. 2012; 6: 7731-7738Crossref PubMed Scopus (141) Google Scholar, 18Huang M. Bakharev P.V. Wang Z.J. Biswal M. Yang Z. Jin S. Wang B. Park H.J. Li Y. Qu D. et al.Large-area single-crystal AB-bilayer and ABA-trilayer graphene grown on a Cu/Ni (111) foil.Nat. Nanotechnol. 2020; 15: 289-295Crossref PubMed Scopus (76) Google Scholar or others through the “pocket” approach.19Hao Y. Wang L. Liu Y. Chen H. Wang X. Tan C. Nie S. Suk J.W. Jiang T. Liang T. et al.Oxygen-activated growth and bandgap tunability of large single-crystal bilayer graphene.Nat. Nanotechnol. 2016; 11: 426-431Crossref PubMed Scopus (235) Google Scholar Particularly, in the pocket method the most common observation is that the synthesis follow a Stranski-Krastanov (SK) mode,20Humphreys C. Controlling crystal growth.Nature. 1989; 341: 689Google Scholar,21Gurioli M. Wang Z. Rastelli A. Kuroda T. Sanguinetti S. Droplet epitaxy of semiconductor nanostructures for quantum photonic devices.Nat. Mater. 2019; : 799-810Google Scholar in which growth within a layer and nucleation of additional layers unavoidably occur at the same time. This method makes it unlikely that only 2LG forms uniformly over a large area, as regions of single-layer graphene (1LG) or other multilayers (3LG, 4LG, etc.) always exist. We report uniform growth of 2LG using the Frank-van der Merwe (FM) growth mode in CVD graphene20Humphreys C. Controlling crystal growth.Nature. 1989; 341: 689Google Scholar using the pocket approach. In the FM mode, the existing layer continues to grow and merges into a film prior to the nucleation of an additional layer. This requires the energy to nucleate a new layer (i.e., surface adhesive energy) to be much larger than the energy to grow the existing layer (i.e., atomic cohesive energy). Previous studies found that the growth of additional graphene layers happens at the interface of the existing graphene layer and the Cu substrate instead of growing on top of the existing layer.19Hao Y. Wang L. Liu Y. Chen H. Wang X. Tan C. Nie S. Suk J.W. Jiang T. Liang T. et al.Oxygen-activated growth and bandgap tunability of large single-crystal bilayer graphene.Nat. Nanotechnol. 2016; 11: 426-431Crossref PubMed Scopus (235) Google Scholar,22Fang W. Hsu A.L. Song Y. Birdwell A.G. Amani M. Dubey M. Dresselhaus M.S. Palacios T. Kong J. Asymmetric growth of bilayer graphene on copper enclosures using low-pressure chemical vapor deposition.. 2014; 8: 6491-6499Google Scholar By tuning the interface adhesive energy, our calculations and experiments both show that the 3rdLG nucleation can be substantially suppressed to favor the 2ndLG growth. Although a fully continuous 2LG film is not achieved in this study yet, the occurrence of graphene multilayers is substantially suppressed in the FM mode-grown samples compared with that of the SK mode. Through first-principles calculations, the likelihood of an incoming carbon atom for 3rdLG nucleation versus to extend the 2ndLG growth at the interface of first layer of graphene (1stLG) and modified Cu were estimated (Figure 1A ; Notes S1 and S2). In the Cu growth substrate, we chose oxygen atoms as adatoms, because they promote graphene growth with larger crystal sizes23Hao Y. Bharathi M.S. Wang L. Liu Y. Chen H. Nie S. Wang X. Chou H. Tan C. Fallahazad B. et al.The role of surface oxygen in the growth of large single-crystal graphene on copper.Science. 2013; 342: 720-723Crossref PubMed Scopus (871) Google Scholar and facilitate the dehydrogenation of the carbon precursor leading to large single-crystal bilayer graphene growth as well.19Hao Y. Wang L. Liu Y. Chen H. Wang X. Tan C. Nie S. Suk J.W. Jiang T. Liang T. et al.Oxygen-activated growth and bandgap tunability of large single-crystal bilayer graphene.Nat. Nanotechnol. 2016; 11: 426-431Crossref PubMed Scopus (235) Google Scholar For comparison, we constructed a model with the same configuration, except that the Cu substrate is original and contains no adatoms (Figure 1B) and determined the carbon atom formation energies at eight different locations (Figure 1C). The atomic cohesive energies of 2ndLG are typically much smaller compared with the interface adhesive energies of 3rdLG nucleation. This relative difference implies that an incoming carbon atom should extend the 2ndLG growth, rather than nucleating the 3rdLG for both bare and modified Cu substrates. Compared with bare Cu, the interface adhesive energy required for 3rdLG nucleation on the modified Cu substrate (CuxO1−x) is much higher (Figures 1C and 1D). The ΔE of CuxO1−x is twice that of the bare Cu (1.82 versus 3.73 eV/C), indicating that the incorporation of oxygen atoms on Cu surface can substantially suppress the 3rdLG nucleation. We can calculate the Boltzmann factor, which is the ratio of probabilities of two events (2ndLG growth versus 3rdLG nucleation), using:exp(−ΔEF2+ΔEM2kBT)exp(−ΔEF3+ΔEM3kBT)=exp(−ΔEF2−ΔEF3kBT)exp(−ΔEM2−ΔEM3kBT),(Equation 1) where ΔEFi (Figure 1C) and ΔEMi are the formation energy and migration barrier of an incoming carbon atom attaching to the ith layer of graphene, respectively, kB is the Boltzmann constant, and T is the growth temperature (i.e., 1,050°C). Assuming that the 3rdLG nucleation requires no further interface diffusion, ΔEM3∼0, and that the reported activation energy for a C atom to perform both bulk and interface diffusion, ΔEM2, is 1.42 eV,10Zhou S.Y. Gweon G.H. Fedorov A.V. First P.N. de Heer W.A. Lee D.H. Guinea F. Castro Neto A.H. Lanzara A. Substrate-induced bandgap opening in epitaxial graphene.Nat. Mater. 2007; 6: 770Google Scholar the Boltzmann factors of 2ndLG growth versus 3rdLG nucleation on pure and modified Cu substrates are 3.2 × 101 and 7.1 × 108, respectively. This implies that, if we have a 2LG film of 1 cm2 size, the maximum areas that the 3rdLG can grow at the 2ndLG/substrate interface for both Cu and CuxO1−x are 3.1 mm2 and 0.14 μm2, respectively (see Note S1). This suggests that centimeter-scale, high-purity 2LG film can be realized by tuning the interface adhesive energy. We implemented this strategy by using a mild oxidation pretreatment step integrated into the CVD pocket approach (Figure 1E) (see experimental procedures). For comparison, the same growth steps are carried out for both bare and treated Cu foil. For graphene samples synthesized on bare Cu, we observed random growth of 3LG islands and inevitable nucleation of thicker layers, similar to previous reports.22Fang W. Hsu A.L. Song Y. Birdwell A.G. Amani M. Dubey M. Dresselhaus M.S. Palacios T. Kong J. Asymmetric growth of bilayer graphene on copper enclosures using low-pressure chemical vapor deposition.. 2014; 8: 6491-6499Google Scholar,24Ta H.Q. Perello D.J. Duong D.L. Han G.H. Gorantla S. Nguyen V.L. Bachmatiuk A. Rotkin S.V. Lee Y.H. Rümmeli M.H. Stranski–Krastanov and Volmer–Weber CVD growth regimes to control the stacking order in bilayer graphene.Nano Lett. 2016; 16: 6403-6410Google Scholar For graphene samples synthesized on treated Cu, we observed no 3LG impurities (Figure 1E). We saw pure 1LG and 2LG regions without 3LG under the optical microscope over an area of 2 cm2. We performed the same comparison studies on two types of commercial Cu foils (Alfa Aesar 13380 and 46986; Figure 1F) and both observed successful suppression of the 3rdLG nucleation, indicating that this mechanism is not limited to the specific types of Cu foils. In addition, we conducted a series of surface analyses on the treated Cu and found that our pretreatment modifies the surface stoichiometry of a bare Cu to Cu1.7O (see Note S4). This surface-level oxidation enhances as we increase the pretreatment duration (Figure S1). It is notable that XPS spectra only show Cu2O features but no Cu2+ (CuO) peaks, therefore the oxidized phase should be Cu2O. In this study, galaxy-like 2LG morphologies were obtained at the centimeter-scale 1stLG/Cu interface. Over a 1 × 1 mm2 optical imaging area, we observed merging of the 2LG domains into an elliptical film surrounded by numerous 2LG sparse domains and 1LG regions (Figure 1H). Similar graphene adlayer morphologies were observed regardless of whether the Cu was treated or not (Figure S2). We attributed this galaxy-like growth morphology to the sparse locations of bulk diffusion of carbon atoms (i.e., from Cu interior to exterior through the foil thickness), and the limited interface diffusion length of carbon atoms at the 1stLG/Cu interface,25Yoon T. Shin W.C. Kim T.Y. Mun J.H. Kim T.S. Cho B.J. Direct measurement of adhesion energy of monolayer graphene as-grown on copper and its application to renewable transfer process.Nano Lett. 2012; 12: 1448-1452Google Scholar corresponding to steps (ii) and (iii) in Figure 1G (see Note S1). Bulk diffusion may happen via preferred paths, due to the presence of uncontrollable impurities in the commercial Cu foil.26Li S. Li Q. Carpick R.W. Gumbsch P. Liu X.Z. Ding X. Sun J. Li J. The evolving quality of frictional contact with graphene.Nature. 2016; 539: 541-545Google Scholar Carbon atoms that penetrated through bulk Cu will emerge at several preferred points within the Cu surface lattice and start to migrate via interface diffusion. According to diffusion theories, carbon concentration surrounding a penetration point follows Gaussian distributions.27Crank J. The Mathematics of Diffusion. Oxford university press, 1979Google Scholar Engineering of the treated Cu to provide increased bulk diffusion paths should increase the area coverage of 2LG. We statistically analyzed the SK mode- and FM mode-grown samples using optical microscopy (Figures 2A and 2C ). On the bare Cu substrate, the SK growth mode appears to be dominant (Figures 2A and 2B). We found 2LG domains using the SK growth mode with an average size of ∼10 μm, but multilayer impurities appear even when the growth is limited to 1 h. As the growth duration increased from 1 to 4 h, bilayer graphene domains become larger in size and eventually merge into a film. However, these are accompanied by not only the concurrent growth of 3LG and 4LG impurities but also the nucleation of subsequent layers up to 8LG. In contrast, when we used treated Cu for 1 h, most of the 2LG domains synthesized on the treated Cu substrate of regular hexagonal shape (Figure 2C). The domain size of the 2LG increased with growth duration, and we observed no 3rdLG nucleation for up to 3 h (Figures 2C and 2D). Our typical 3.33-h grown sample had connection of 2LG domains into a continuous area. We observed the appearance of a μm-sized 3LG domain in a continuous 2LG region for our typical 4-h grown samples, but without any 4LG impurities (Figure 2C). To quantitatively evaluate the suppression of multilayer impurities in our FM mode-grown bilayer graphene film, we developed a statistical method with the help of microscopy analysis. Specifically, we first converted the original microscopy images of graphene films transferred onto Si/SiO2 substrates into grayscale and plotted the corresponding pixel counts (over 5,000,000 pixels for each image). A histogram for a typical FM mode-grown graphene grown for 1 h contains only two peaks (1LG and 2LG), while that of SK mode contains four peaks, which can be attributed to 1LG, 2LG, 3LG, and 4LG (Figure 2E). To systematically label the graphene region with different layer numbers, we assigned a range of gray levels to each histogram peak and converted the corresponding pixels into red (Figure 2E). Upon layer labeling in each microscopy image, we estimated the occurrence probability of 2LG and multilayer impurities by extracting the ratio of pixel counts. For this analysis (Figure 2F), all samples were grown for 3.33 h. We randomly acquired 10 microscopy images, with sizes of at most 40 × 40 μm2, from both the SK mode- and the FM mode-grown graphene samples (Figure 2F). The occurrence probability of 2LG is less than half (46.20% ± 14.22%) for the SK mode-grown samples, while that of FM mode is close to 100% (99.32% ± 0.34%). We found, using Raman spectroscopy, that our FM mode-grown bilayer graphenes are quasi-AB stacked (i.e., either AB stacked or having a small twisting angle [θ = 0°–5°]) (Figure 3A ).28Zhou H. Yu W.J. Liu L. Cheng R. Chen Y. Huang X. Liu Y. Wang Y. Huang Y. Duan X. Chemical vapour deposition growth of large single crystals of monolayer and bilayer graphene.Nat. Commun. 2013; 4: 2096Crossref PubMed Scopus (461) Google Scholar,29Kim K. Coh S. Tan L.Z. Regan W. Yuk J.M. Chatterjee E. Crommie M.F. Cohen M.L. Louie S.G. Zettl A. Raman spectroscopy study of rotated double-layer graphene: misorientation-angle dependence of electronic structure.Phys. Rev. Lett. 2012; 108: 246103Google Scholar From our transmittance electron microscopy (TEM) selected area electron diffraction (SAED) characterization of multiple locations in our FM-2LG sample, AB stacking has been observed (Figure S3); nevertheless, since SAED only examines limited sample areas, one cannot conclude that all 2LG regions are AB stacked. Rather, Raman spectroscopy mapping gives us characterization of large-area regions and, with 2D band's full width at half maximum, one can narrow down the twisting angle θ to <5°.29Kim K. Coh S. Tan L.Z. Regan W. Yuk J.M. Chatterjee E. Crommie M.F. Cohen M.L. Louie S.G. Zettl A. Raman spectroscopy study of rotated double-layer graphene: misorientation-angle dependence of electronic structure.Phys. Rev. Lett. 2012; 108: 246103Google Scholar A typical 1LG Raman signature is shown in Figure 3A, with the positions of G and 2D bands (ωG and ω2D) located at 1,582 and 2,670 cm−1, respectively. The intensity ratio of 2D to G bands (I2D/IG) is >1, and the 2D full width at half maximum (Γ2D) is ∼34 cm−1. The other spots exhibited an almost identical Raman spectra with ωG = 1,580 cm−1, ω2D = 2,680 cm−1, I2D/IG = ∼1, and Γ2D = ∼60 cm−1. These features all can be identified as 2LG with either AB stacked or small twisting angles (θ = 0°–5°).28Zhou H. Yu W.J. Liu L. Cheng R. Chen Y. Huang X. Liu Y. Wang Y. Huang Y. Duan X. Chemical vapour deposition growth of large single crystals of monolayer and bilayer graphene.Nat. Commun. 2013; 4: 2096Crossref PubMed Scopus (461) Google Scholar,29Kim K. Coh S. Tan L.Z. Regan W. Yuk J.M. Chatterjee E. Crommie M.F. Cohen M.L. Louie S.G. Zettl A. Raman spectroscopy study of rotated double-layer graphene: misorientation-angle dependence of electronic structure.Phys. Rev. Lett. 2012; 108: 246103Google Scholar We obtained consistent results from 60 different Raman spectra that we randomly collected from 6 different samples, confirming that our growth method produces quasi-AB-stacked 2LG. In contrast, for the SK mode-grown graphene sample (Figure 3B), the Raman spectra taken at four typical spots all had different ωG, ω2D, I2D/IG, and Γ2D values. We identified the Raman spectra of spots 1 and 4 in Figure 3B 1LG and 3LG, respectively. Under the optical microscope, we recognized spots 2 and 3 as 2LG because they shared the same optical contrast, which is at the interval of 1LG and 3LG. One spot was 2LG with a small twisting angle (I2D/IG = 0.77 and Γ2D = 62 cm−1) and the other had a larger twisting angle 2LG (I2D/IG = 2.3 and Γ2D = 37 cm−1).29Kim K. Coh S. Tan L.Z. Regan W. Yuk J.M. Chatterjee E. Crommie M.F. Cohen M.L. Louie S.G. Zettl A. Raman spectroscopy study of rotated double-layer graphene: misorientation-angle dependence of electronic structure.Phys. Rev. Lett. 2012; 108: 246103Google Scholar The variation of twisting angles in the SK mode-grown 2LG is large, unlike those in the FM mode. We also performed Raman mappings on both FM mode- and SK mode-grown graphene samples that we transferred onto Si/SiO2. For the FM mode-grown sample, we found Γ2D remains constant throughout the single domain or two merging domains of 2LG (Figure 3C). We collected Raman spectra from typical SK mode-grown samples, over an area that consists of 1LG, 2LG, 3LG, and MLG (Figure 3D). Surprisingly, we classified the 2LG region, which looks uniform under optical microscopy, into at least two divisions in the Raman map of Γ2D (Figure 3D). We further classified these Raman mapping data using a simple and fast unsupervised machine-learning approach, the k-means algorithm, which measures dissimilarity between two data points using square Euclidean distance (see experimental procedures). We considered all five parameters (ωG, ω2D, I2D, IG, and Γ2D) in our artificial intelligence-assisted Raman data classification. To visualize the classification results, we selected three parameters (ωG, ω2D, and Γ2D) and made three-dimensional plots (Figures 4A and 4C ). The algorithm clustered all 18,000 Raman spectra into two classes in several seconds. By reading the mean value (ωG¯, ω2D¯, Γ2D¯) of each cluster, we identified class I and II as the Raman data of 1LG and quasi-AB-stacked 2LG, respectively (Figure 4B). The error rate of our artificial intelligence-assisted Raman analysis is 0.027, as cross-verified by hierarchical clustering algorithm. We also verified the quasi-AB-stacked 2LG (class II) using TEM (Figure S3). We executed the same algorithm for comparison on the Raman data of SK mode-grown samples (Figure 4C), where classes I to V in Figure 4D represent 1LG, quasi-AB-stacked 2LG, and larger twisting angle 2LG, 3LG, and MLG, respectively. Using the artificial intelligence-assigned class number (I to V), we constructed a spatial map (Figure S4) in which the boundaries/edges that separate 1LG, 2LG, and 3LG, etc., can be recognized. For graphene samples on substrates without a good optical contrast, such as glass, quartz, or sapphire, Raman mapping or this facile classification method will be very useful to identify regions of different layer numbers and stacking orders. Utilizing the mechanical advantage of 2LG in comparison with 1LG,30Papageorgiou D.G. Kinloch I.A. Young R. Mechanical properties of graphene and graphene-based nanocomposites.Prog. Mater. Sci. 2017; 90: 75-127Crossref Scopus (1143) Google Scholar we devised a support-free wet transfer approach driven by the Marangoni effect. Owing to graphene's thinness, a transfer support layer is generally required to prevent crack generation and propagation. In conventional polymer-supported CVD graphene transfers,31Leong W.S. Wang H. Yeo J. Martin-Martinez F.J. Zubair A. Shen P.C. Mao Y. Palacios T. Buehler M.J. Hong J.Y. et al.Paraffin-enabled graphene transfer.Nat. Commun. 2019; 10: 867Google Scholar,32Wang H. Leong W.S. Hu F. Ju L. Su C. Guo Y. Li J. Li M. Hu A. Kong J. Low-temperature copper bonding strategy with graphene interlayer.ACS Nano. 2018; 12: 2395-2402Crossref PubMed Scopus (35) Google Scholar a polymer sacrificial support layer is first coated on a piece of graphene/Cu. The coated sample is then floated on an etchant solution to completely remove the Cu. The sample is then rinsed with deionized water at least three times. For these rinsing steps, a flat scoop (e.g., Si wafer or glass slide) is used to move the sample from one beaker of deionized water to another, and an external force (i.e., tweezers or glass slide) is required to drive the sample from the liquid surface to the flat scoop, until the sample is free of Cu etchant residues. If these rinsing steps were carried out without the polymer support layer, the external force usually generates cracks and tears in the graphene layer. To eliminate the need for a sacrificial transfer support layer, we substitute the external force with Marangoni-driven internal forces (Figure 5A), which are the hydrodynamic forces generated whenever liquids with different surface tensions mix.33MacKay G. Mason S. The Marangoni effect and liquid/liquid coalescence.Nature. 1961; 191: 488Google Scholar To achieve these prior to sample loading, we fine-tuned the liquids' surface tension with the help of isopropyl alcohol (IPA). We first wetted a flat scoop with a liquid of higher surface tension (high σ), and quickly dipped it into the lower surface tension liquid with a floating graphene sample. Forced by Marangoni flows, the sample will automatically climb from the liquid surface to the flat scoop in a few seconds (Video S1). When we gently dipped the flat scoop back into the high σ liquid, the graphene sample gradually glides to the liquid surface (see Figure 5A and experimental procedures). Figure 5D shows a typical image of a centimeter-scale 2LG sample transferred onto a Si/SiO2 substrate using our free-standing Marangoni-driven graphene transfer process. We found the yield (in terms of area percentage for unbroken regions) of this transfer approach for our 2LG is >95%, whereas for 1LG it is ∼66.7% (Figure S5; Video S2). This observation was consistent with our molecular dynamics (MD) simulations; AB-stacked 2LG was found to be robust under the Marangoni flow, even in the presence of atomic defects (Figures 5B and 5C; Video S3; Note S5). A frame made of Scotch tape or silver paste was