Atomic layer-by-layer etching of graphene directly grown on SrTiO₃ substrates for high-yield remote epitaxy and lift-off

Remote epitaxy, a new epitaxial growth technique that allows near universal growth and lift-off of freestanding single-crystalline membranes, has recently been gaining much interest as a promising technology to produce multiple single-crystalline semiconductor membranes off a single host wafer. In this method, a mono- or bilayer graphene is transferred onto a compound single-crystalline substrate, such as III-V, III-N, and complex-oxides, followed by epitaxial homo- or heteroepitaxial growth on top of the graphene-coated single-crystalline substate.^1–12 It has been found that due to the extreme thinness of graphene, the surface potential of the underlying single-crystalline substrate can penetrate through the graphene layer(s), resulting in epitaxial growth that is seeded by the substrate rather than graphene. Moreover, due to the existence of graphene between the epitaxial layer and the substrate, no bonding is formed between the epilayer and substrate, allowing facile exfoliation of the epilayer off the substrate. Since the exfoliation process is through the van der Waals (vdW) gap between the graphene and the epilayer, the surface of the host wafer is completely unharmed, unlike other epitaxial lift-off methods, such that the graphene-coated wafer can potentially be reused without additional surface refurbishment.³ This allows unprecedented creation of artificially made mixed-dimensional heterostructures by stacking ultrathin single-crystalline membranes composed of 2D and 3D materials regardless of its crystalline structure and growth compatibility such as thermal budget.^13,14

Despite the advantages of remote epitaxy, there is a notable drawback that leads to a low exfoliation yield using this method. This drawback is caused by the transfer step of the graphene onto the host substrate, where typically epitaxial graphene grown on silicon carbide (SiC) substrates is exfoliated using a thin Ni handling layer.¹⁵ During this semi-dry transfer process, small amounts of organic and metallic residues are left behind on the surface of the graphene-coated substrate, leading to patches of polycrystalline regions where the residues are present. Additionally, due to the macroscopic defects such as cracks and holes that form during transfer, direct atomic bonds are made in those regions, which lead to spalling of the substrate or epitaxial film during exfoliation. One can conclude that unless a method to transfer large area graphene sheets with absolutely zero defects onto other arbitrary substrates is developed, the transfer step will often prevent achieving high-yield remote epitaxy and lift-off of the epitaxial membrane, both in terms of epitaxial area and run-to-run variations. Thus, another method of forming pristine graphene on single-crystalline substrates is needed.

To solve this issue, we demonstrate a method to reliably form mono-/bilayer graphene on any substrate that can allow multilayer graphene to form. In this work, we took single-crystalline strontium titanate (SrTiO₃, abbreviated as STO) as an example. STO is an archetypal perovskite complex-oxide material that exhibits various physical properties such as superconductivity, ferroelectricity, and is used as a substrate for many other interesting complex-oxide materials.¹⁶ Complex-oxides are an important class of materials for next generation electronic, photonic, and quantum information devices, as evidenced by Intel’s recent proposal of the magnetoelectric spin–orbit (MESO) logic device.¹⁷ The ferroelectric, ferromagnetic, and superconductive properties as well as unique interfacial effects and strain-dependent properties make complex-oxides a one of a kind material platform for physicists, material scientists, and electrical engineers to explore novel phenomena and devices.^18–20 

It has been reported previously that graphene can be grown on STO substrates using the chemical vapor deposition (CVD) growth technique.²¹ However, due to random placement of nucleation and growth nature of vdW materials, the thickness cannot be controlled precisely to stop at monolayer growth at wafer-scale and will typically grow up to trilayers or more to achieve full coverage of graphene over the entire substrate surface. Moreover, due to the nature of vdW epitaxial growth, nanoscale islands of graphene in the form of clumped carbon atoms also emerge due to secondary nucleation on top of existing graphene layers. The crystalline quality is severely degraded around these multi-dots during growth, leading to islands of polycrystalline regions. Such thick and rough graphene cannot be used for remote epitaxy, which only allows up to bilayer graphene for complex-oxide materials.² The key to solving this problem is through atomic layer etching (ALE), which allows atomic layer-by-layer removal of graphene as well as etching away multi-dots, allowing precise control of the graphene thickness down to the monolayer/bilayer limit and cleaning up the graphene surface suitable for remote epitaxy. Here, we show that a combination of direct grown graphene on STO substrates with ALE allows high-yield remote epitaxy and exfoliation to produce freestanding thin-film STO membranes.

Figures 1(a)–1(d) show a schematic illustration of remote epitaxy on a direct-grown graphene/STO substrate with an added atomic layer etching (ALE) step to precisely control the thickness of the graphene grown on the STO substrate. Before graphene growth, a TiO₂ terminated STO surface was prepared by dipping the substrate in buffered-HF for 30 s, followed by an anneal process in a tube furnace at 1100 °C for 2 h. The graphene is grown via a conventional CVD method²¹ with Ar as the carrier gas and a mixture of CH₄ (methane) and H₂ as the carbon source. The Ar, CH₄, and H₂ flow was set to 200, 150, and 16 SCCM, respectively. The STO samples (5 × 5 mm²) were placed inside a tube furnace with a set temperature of 1090 °C for a total growth time of 180 min. We visualized the growth morphology of graphene on STO under a scanning electron microscope (SEM) as a function of growth time, as shown in Fig. 2(a). First, small islands of graphene can be seen nucleating at random locations on the surface of the STO during the first few minutes of the growth. As the growth time progresses, the islands grow larger and begin to merge until they cover the entire surface of the substrate. After 180 min of growth, full coverage of graphene is achieved, which is verified by SEM. The graphene grown on STO was then transferred onto an SiO₂/Si substrate for Raman spectroscopy using the semi-dry transfer method, in which Ni is deposited using e-beam evaporation on top of the graphene, followed by peeling of the Ni/graphene stack using thermal release tape (TRT). The TRT/Ni/graphene stack is then placed on top of a SiO₂/Si substrate, followed by removal of the TRT by placing the sample on a hotplate at 120 °C, then etching away the Ni film in ferric chloride (FeCl₃). The graphene is transferred from STO to SiO₂ using this method for Raman spectroscopy since the spectra of STO overlap with the graphene peaks, making it difficult to analyze the resulting Raman spectrum.

Raman spectroscopy is mainly used to characterize the quality and thickness of graphene layers,²² resulting in the growth of graphene on STO as shown in Fig. 2(b). As can be seen, the intensity ratio of the 2D peak at 2780 cm⁻¹ to the G peak at 2430 cm⁻¹ indicates that multiple layers of graphene was grown on the STO substrate. We note that although it is difficult to see multiple layers of graphene in the SEM image shown in Fig. 2(a), the Raman results unambiguously indicate multiple layers of graphene. In contrast, epitaxial graphene grown on SiC substrates transferred onto SiO₂/Si via a semi-dry transfer method results in macro- and microscopic defects such as wrinkles, tears, and cracks. Moreover, due to the use of thermal release tape and the Ni handling layer during the semi-dry transfer step, macroscopic residues are left on the substrate even after thorough cleaning of the substrate in acetone and isopropyl alcohol. An example of graphene defects post transfer is shown in Fig. 2(c). The figure on the left shows epitaxial graphene transferred onto the SiO₂/Si substrate, and the middle and right SEM figures show examples of residues and macroscopic defects of transferred graphene that are made during the transfer process, which is carried out by hand. These defects inevitably lead to direct growth of the epitaxial layer through the microcracks in graphene, forming strong atomic bonds that prohibit exfoliation of the film off of the substrate. In addition, polycrystalline regions are grown on and around areas with residues, leading to reduced crystalline quality of the film.

Thus, it is clear that direct growth graphene is the best solution for remote epitaxy and lift-off as it fully covers the entire substrate without any defects and leaves behind no residues. The only issue is precisely controlling the thickness of the graphene over the entire wafer surface, as well as the multi-dots. These multi-dots severely reduce the uniformity of the graphene surface and prevent high quality remote epitaxy to occur. To solve this problem, we invoked a method to etch away thick layers of graphene in atomic steps using an atomic layer etching (ALE) technique. Using this technique, thick graphene layers can be directly grown on the substrate to fully cover the surface, and etch the graphene layers atomic layer by atomic layer until the desired number of graphene layers is remaining, as well as clean up any undesirable surface roughness as explained in the following paragraphs.

Figure 3(a) shows the layer-by-layer etching of graphene grown on STO with ALE cyclic steps in which O₂⁺ ion beam irradiation for chemical adsorption and Ar⁺ ion beam irradiation for physical desorption are sequentially performed. In particular, as graphene grows on SrTiO₃, multi-dots that exist due to secondary nucleation are also etched in the ALE process due to the dangling bonds present around the nano-dots and the sidewall etching effect. Figure 3(b) shows a schematic of the ALE process consisting of the sequential reactant adsorption step and the reacted compound desorption step for one-cycle graphene ALE. When a controlled oxygen atom is adsorbed on the bilayer graphene, the electron migrates from the surface carbon to the adsorbed oxygen atom due to the higher electronegativity of oxygen atoms, thereby reducing the binding energy of the graphene top-layer. On the other hand, the binding energy of the graphene bottom-layer to which oxygen is not adsorbed can be maintained. Therefore, only the top-layer can be selectively removed without etching the bottom-layer of the bilayer graphene through controlled Ar⁺-ion bombardment. We observed the energy distribution of O₂⁺ and Ar⁺ ions reaching the substrate through quadrupole mass spectrometer (QMS) analysis to effectively remove one monolayer graphene using the one-cycle ALE process [Figs. 3(b) and 3(c)]. As shown in Fig. 3(b), the threshold energy of O₂⁺ ions decreased as the rf power decreased from 100 to 15 W. When an axial magnetic field of 30 G was applied together with an rf power of 15 W, a further decrease in the threshold energy and an increase in ion flux were observed according to the helical motion of electrons in the plasma source. The increase in O₂⁺ ion energy as rf power increases shows the characteristics of a capacitively coupled plasma (CCP) mode rather than an inductively coupled plasma (ICP) mode due to the low electron density of oxygen plasma. We used an axial magnetic field of 30 G and an rf power of 15 W for chemical adsorption at low O₂⁺ ion energy. On the other hand, the threshold energy of the Ar⁺ ion decreased from ∼55 to ∼23 eV as the rf power increased from 100 to 500 W [Fig. 3(c)]. The reason why the trend is opposite to that of the O₂⁺ ion energy distribution originates from the different operating modes of the plasma ion source. Ar plasma was operated in the ICP mode due to its high electron density. This is also because oxygen has a high electron affinity, whereas Ar has no electron affinity. We performed Ar⁺ ion beam desorption at an rf power of 500 W with a high ion flux and a threshold ion energy of ∼23 eV for the physical desorption of chemisorbed species.

Figure 3(d) shows Raman spectra analysis after one- and two-cycle ALE on trilayer graphene grown on SrTiO₃. As the graphene layer was etched layer-by-layer according to the number of ALE cycles, a gradual increase in the 2D peak intensity and a down shift of the 2D peak position were simultaneously observed.^23,24 In addition, we compared the surface morphology with optical microscopy (OM) and atomic force microscopy (AFM) images to confirm the removal of multi-dots formed on trilayer graphene according to the number of ALE cycles [Figs. 3(e)–3(j)]. Chemically stable graphene has a honeycomb structure without dangling bonds, whereas multi-dots have very effective chemical adsorption and physical desorption due to numerous unstable dangling bonds on the entire surface, as well as the sidewall etching effect of multi-dots. Therefore, it can be seen that most of the multi-dots have been removed after one- and two-cycle ALE in the OM images [Figs. 3(e)–3(g)]. As shown in Figs. 3(h)–3(j), the rms roughness after one- and two-cycle ALE on trilayer graphene was also confirmed to gradually decrease to 5.876, 2.037, and 1.109 nm, respectively. These results show that ALE is an effective method to remove nanoscale graphene defects during growth, such as the multi-dots, as well as precisely control the thickness of graphene directly grown on single-crystalline substrates for remote epitaxy. We note that the ALE process does not guarantee complete monolayer graphene coverage over the entire sample. However, the majority of the substrate area will be monolayer and very small nanoscale patches will be bilayer or trilayer. Nevertheless, due to the strong ionicity of complex-oxide materials, there is no issue performing remote epitaxy.² 

To verify the effectiveness of ALE on creating graphene-coated substrates ready for remote epitaxy, we remote epitaxially grew STO (100) on top of graphene-coated STO (100) substrates with and without ALE steps using the pulsed laser deposition (PLD) technique. To deposit STO, a KrF laser with an energy of ∼400 mJ with pulse rate of 10 Hz was used to ablate a commercially bought STO target. The substrate was set at a temperature of 850 °C with an oxygen flow of 20 mTorr. Figure 4(a) shows the top-view SEM and electron backscatter diffraction (EBSD) measurement results of STO (100 nm) grown on graphene without ALE. As expected, due to the thick and non-uniform graphene layer that was grown on the STO substrate, the epitaxially grown STO films showed completely polycrystalline growth, evident from the rough multifaceted surface morphology shown in the SEM image and the multi-colored EBSD result. In contrast, the sample with graphene thickness controlled by ALE showed completely single-crystalline growth, as indicated by the smooth top-view SEM image and EBSD result showing only the (100) orientation [Fig. 4(b)]. The inset in the right shows an exfoliated 100 nm STO membrane off of the substrate, supported by a Ni stressor layer and thermal release tape.

Finally, to verify the exfoliation area yield of transferred graphene on STO vs direct-graphene grown on STO substrates, STO membranes from both samples were exfoliated and inspected under the microscope. As can be seen in Fig. 4(c), macroscopic holes ranging from several micrometers to tens of micrometer holes can be observed randomly located throughout the membrane exfoliated off of transferred graphene, which is caused by growth in regions with cracks and holes in the graphene, preventing exfoliation. In contrast, the STO membrane exfoliated off of direct-grown graphene shows absolutely no holes and is a smooth continuous film. This indicates that the STO membrane has, indeed, been grown on graphene and not through any macroscopic defects on the graphene such as holes and tears, which would have prevented us from exfoliating the STO membrane so cleanly. We also carried out cross-sectional TEM of the STO film grown on graphene-coated STO substrates to verify remote epitaxy through graphene, as shown in Fig. 5. Moreover, although the density of holes for STO membranes exfoliated off of transferred graphene varies from sample to sample, it was difficult to produce a perfect film without any holes. Thus, our results demonstrate that it is best to use substrates with graphene grown directly on the substrate for remote epitaxy for obtaining the high-yield remote epitaxy and exfoliation process for freestanding single-crystalline membrane production. We note that Ni was used as a stressor and handling layer, which allows robust handling of the thin and brittle complex-oxide membrane while allowing high yield transfer of the membrane onto other arbitrary substrates without generating cracks and wrinkles.

In summary, we show that by carefully optimizing direct growth of graphene and adding an ALE step, pristine mono- to bilayer graphene on single-crystalline substrates can be made for high-yield remote epitaxy or other applications needing graphene sheets. This opens up a scalable pathway for producing large-area substrates with controllable graphene layers suitable for high-yield remote epitaxy and epitaxial lift-off processes. This method is not only limited to complex-oxide materials demonstrated in this work and should be compatible with other material systems such as multilayer graphene grown on GaAs or sapphire substrates for producing III-V or III-N membranes for advanced (opto)electronic and power device applications.

This research study was supported by the Yonsei University Research Fund, Grant No. 2021-22-0338, and by the Basic Science Research Program through NRF funded by the Ministry of Education (NRF Award No. NRF-2020R1A6A1A03043435).

The authors have no conflicts to disclose.

K.S.K., J.E.K., and P.C. contributed equally to this work.

The data that support the findings of this study are available from the corresponding authors upon reasonable request.