Graphene research has prospered impressively in the past few years, and promising applications such as high-frequency transistors, magnetic field sensors, and flexible optoelectronics are just waiting for a scalable and cost-efficient fabrication technology to produce high-mobility graphene. Although significant progress has been made in chemical vapor deposition (CVD) and epitaxial growth of graphene, the carrier mobility obtained with these techniques is still significantly lower than what is achieved using exfoliated graphene. We show that the quality of CVD-grown graphene depends critically on the used transfer process, and we report on an advanced transfer technique that allows both reusing the copper substrate of the CVD growth and making devices with mobilities as high as 350,000 cm 2 V –1 s –1 , thus rivaling exfoliated graphene.

Keywords

Here, we report on a delamination method that overcomes these problems and results in CVD-grown graphene devices with properties comparable to those of high-quality exfoliated graphene. The key aspect is an advanced transfer process based on van der Waals forces, which we use to pick up CVD-grown graphene directly from the copper and to encapsulate it in hBN while avoiding any contact with intermediary epoxy, polymers, or other chemical agents. This method minimizes all contaminations of the graphene surface due to processing and, at the same time, allows performance of multiple growth cycles on the same copper substrate. The high quality of the hBN/graphene heterostructures obtained using this method is probed by Raman spectroscopy and transport measurements. All investigated devices exhibit an extremely narrow Raman 2D peak and high carrier mobilities (μ ≈ 100,000 cm 2 V –1 s –1 or higher), similar to those observed in hBN/graphene heterostructure based on exfoliated graphene ( 24 , 25 ). Remarkably, we observe no noticeable difference between devices realized in the first or subsequent growth cycles, indicating that the copper substrate is a reusable resource in our fabrication process.

Chemical vapor deposition (CVD) on copper ( 1 – 11 ) is a convenient alternative to epitaxial growth ( 12 – 16 ) for obtaining large graphene crystals. However, to serve as channel material in electronic devices such as high-frequency transistors ( 17 ), Hall sensors ( 18 ), and various other applications ( 19 ), CVD-grown graphene needs to be transferred from the growth substrate (typically copper) onto an insulating substrate, for example, SiO 2 or hexagonal boron nitride (hBN). In state-of-the-art transfer methods ( 20 ), a transfer polymer is deposited directly onto the graphene, and the copper substrate is chemically etched away. This approach has two major drawbacks: first, chemical residues strongly degrade the electronic properties of graphene and reduce its mobility well below the values obtained in exfoliated graphene; second, it leaves no copper to be reused in a new growth step and thereby significantly increases the production costs and creates chemical waste. An alternative, very promising transfer method is delamination. In this case, graphene is peeled from the growth surface by exerting an external force, thus preserving the copper foil for reuse. Both electrochemical ( 21 , 22 ) and dry delamination ( 23 ) of CVD-grown graphene have been recently demonstrated. However, the electronic quality of the graphene obtained in these pioneering works still appears to suffer from chemical- and process-related contaminations.

RESULTS

Graphene flakes are grown by low-pressure CVD (1, 10, 26), using methane as a precursor and copper as a growth substrate (see Materials and Methods). A schematic of the furnace is illustrated in Fig. 1A, whereas Fig. 1G shows a scanning electron microscopy (SEM) image of a typical graphene flake on copper. The diameter of the flakes can reach up to hundreds of micrometers (1, 10) (see also the Supplementary Materials). After a few days of exposure to ambient conditions, the graphene flakes on copper undergo a significant change in optical contrast and become visible with an optical microscope (see Fig. 1C). Energy-dispersive x-ray (EDX) spectroscopy indicates that this change of contrast is associated with an increased concentration of oxygen at the graphene-copper interface (see the Supplementary Materials), which is in agreement with the recent observations by Schriver et al. (27), that CVD graphene actually promotes the oxidation of copper at room temperature. Although this effect might be detrimental for certain applications, it increases the success rate of the dry delamination method illustrated in Fig. 1B to above 90%, without affecting the electronic properties of the transferred graphene. Details on the transfer process are given in Materials and Methods. In brief, we prepare a polymer stack consisting of polyvinylalcohol (PVA) and polymethylmethacrylate (PMMA) on a polydimethylsiloxane (PDMS) stamp. The stack is covered with an exfoliated hBN flake and brought into contact with the graphene on copper. Thanks to the strong van der Waals interactions between the graphene and the hBN, the graphene is picked up from its growth substrate when separating the stack again. It can then be transferred to an arbitrary substrate, for example, another exfoliated hBN flake, as for the samples investigated in this work.

Fig. 1 Dry transfer of CVD-grown graphene. (A) Illustration of the CVD furnace with a copper enclosure inside. (B) Process schematic of the contamination-free transfer of CVD graphene from copper onto hBN. (C) Optical microscopy image of grown graphene crystals on copper foil. (D and E) Optical false-color image of graphene on copper after a first (D) and a second (E) growth cycle. The insets show close-ups of areas marked by the dashed rectangles (see also fig. S1). (F) False-color optical image of CVD graphene encapsulated in hBN. The graphene is recognizable by its fractal shape. Turquoise and yellow areas correspond to the top and bottom hBN flakes, respectively (see also fig. S2). (G) SEM image of a graphene crystal on copper.

This transfer process allows reuse of the catalyst foil in multiple growth cycles. This is exemplified in Fig. 1, D and E, where we show graphene grown on the very same copper substrate in two subsequent growth cycles, that is, the flakes in Fig. 1E were grown after those in Fig. 1D were picked up with the transfer technique described above. The arrow indicates a marker cut that was manually placed on copper after the first growth process, to unambiguously identify the specific location on the substrate. The flakes obtained in both growth cycles are similar in size and shape and, as we show below, in structural and electronic properties.

The properties of the CVD graphene/hBN heterostructures are probed by Raman and electrical transport measurements. A typical Raman spectrum is shown in Fig. 2A. It is comparable to the spectra obtained for exfoliated graphene sandwiched in hBN (25, 28–31). The graphene-related G and 2D peaks are visible at positions ω G = 1584.2 cm–1 and ω 2D = 2687.2 cm–1. The absence of the D peak around ω D ≈ 1345 cm–1 indicates the lack of lattice defects (32), giving a first evidence of the high structural quality of the fabricated heterostructure. A second and equally important clue comes from the exceptionally small full width at half maximum (FWHM) of the 2D peak, Γ 2D ≈ 16.5 cm–1. It has recently been shown that random strain variations on the nanometer scale are the main source of broadening of the Raman 2D peak and that low values of Γ 2D are a direct indication of a uniform strain landscape within the laser spot size (25).

Fig. 2 Raman characterization of CVD-grown graphene. (A) Typical Raman spectrum and microscopy image (inset) of a CVD graphene/hBN heterostructure. a.u., arbitrary units. (B) Spatially resolved map of the width of the 2D peak, Γ 2D , measured in the region displayed in (A). Scale bars, 20 μm. Γ 2D shows remarkably low values in the area where the graphene flake is encapsulated in hBN [marked by the dashed line, see also inset in (A)]. (C) Histogram of Γ 2D measured on the encapsulated part of the graphene flake shown in (B). It exhibits a maximum at 16.8 cm–1 and a cutoff at 16 cm–1. (D) Sample-averaged against average strain for 29 samples. Squared and circle data points correspond to CVD-grown graphene wet-transferred to SiO 2 and between hBNs, respectively. Diamond-shaped data points correspond to samples realized with the dry transfer technique starting from a new or a reused copper substrate. Different colors correspond to different growth cycles, according to the color scheme of (E). For each data point, the error bars indicate the 20th and 80th percentile of the corresponding distributions of Γ 2D and strain. (E) Typical Raman spectra of samples made with the dry transfer technique performing multiple growth cycles on the same copper substrate.

For all samples, we record spatially resolved Raman maps as shown in Fig. 2B, and plot the values of Γ 2D in histograms, such as in Fig. 2C. From these, we extract the sample-averaged width of the 2D peak, . For comparison, we also repeat the same procedure on a number of samples with CVD-grown graphene wet-transferred to SiO 2 or sandwiched between hBN flakes using the state-of-the-art wet etching technique (20). The obtained Γ 2D values for the wet-transferred samples are in agreement with literature (33). Remarkably, all heterostructures produced with the dry delamination method show a value of about 20 cm–1, significantly smaller than what is observed in wet-transferred samples (see Fig. 2D). This also holds true for the average strain in the samples, which we estimate following (34) by analyzing ω G and ω 2D (see the Supplementary Materials). The clustering of the diamond-shaped data points in Fig. 2D further shows that our fabrication technique consistently results in samples of high structural quality, with low strain values (about 0.1%) and small strain fluctuations on the nanometer scale over the entire sample, manifested by low values of (25). Moreover, structures fabricated by reusing the copper substrate multiple times show the same properties as those realized in the first growth cycle (see Fig. 2, D and E).

Random strain variations not only affect the linewidth of the Raman 2D peak but also are the dominant cause of scattering for electrons in graphene on hBN (31). Being characterized by remarkably low values of , our samples hold promise to show high carrier mobility. We test this by performing transport measurements in a four-probe configuration, after structuring the CVD graphene/hBN heterostructures into Hall bar devices and contacting them following the scheme of Wang et al. (24). An optical image of a typical device is shown in the inset of Fig. 3A, together with the four-terminal resistance measured as a function of back-gate voltage V BG at low (T = 1.6 K, blue) and at room temperature (T = 300 K, black curve). The small shift of the peaks with respect to V BG = 0 V represents a doping level of 3 × 1011 cm–2. The mobility of each sample is extracted by plotting the conductivity as a function of the charge carrier density n (Fig. 3B), performing a linear regression up to n = 5 × 1011 cm–2, and calculating the mobility in terms of the Drude formula (see also the Supplementary Materials). For the sample shown in Fig. 3A at T = 1.6 K, we obtain μ = 110,000 and 145,000 cm2 V–1 s–1 for electrons and holes, respectively. The green curve in Fig. 3B represents the low temperature conductivity of another sample with carrier mobilities of μ = 320,000 and 350,000 cm2 V–1 s–1 for electrons and holes, respectively. These values exceed the highest ones reported for CVD graphene (6, 9) by roughly one order of magnitude and, when performing magnetotransport measurements, are reflected in a very well resolved Landau fan and broken Landau level degeneracy already at moderately low magnetic fields (see Fig. 3C and fig. S3). Similar features have previously been observed in high-quality devices based on exfoliated graphene (35). At higher temperatures, the mobility of our samples shows the expected tendency to decrease, but remains well above 50,000 cm2 V–1 s–1 up to room temperature (see Fig. 3F), making our fabrication method very appealing for technological applications.

Fig. 3 Transport characterization of dry-assembled devices. (A) Optical image of a Hall bar device and four-terminal resistance as a function of back-gate voltage at 1.6 K (blue) and 300 K (black). (B) Conductance σ as a function of charge carrier density n for the device shown in (A). The green curve shows data measured on another sample with carrier mobility of μ = 350,000 cm2 V–1 s–1. (C) Derivative of the Hall conductivity σ xy with respect to V BG as a function of magnetic field B and V BG for the device corresponding to the green curve in (B). Landau levels (LL) at filling factors ν = …, −10, −6, −2, 0, 2, 6, 10,… are clearly visible. Several LL exhibit full degeneracy lifting already at B = 6 T, see, for example, the appearance of LL at ν = −1, 1. (D) Plot of μ D = σ/ne as a function of the charge carrier density for the device shown in (A) at 1.6 K. The dashed lines represent the mobilities extracted from (B). (E) Procedure to extract n*: two lines are fitted to the double logarithmic plot of σ as a function of n. (F) Mobility as a function of temperature for the device shown in (A).

Another indication of the sample quality is given by the disorder-induced charge carrier density fluctuations n*. This quantity gives a measure of the potential fluctuations experienced by the electrons (31). It can be extracted by plotting the low-temperature conductivity σ against the carrier density n in a double logarithmic scale (see Fig. 3E). For all our devices, we extract very low values of n*, around a few 1010 cm–2, which indicate a rather homogeneous potential background. Moreover, we again observe no difference between samples fabricated in a first or second growth cycle on the same copper substrate (see blue and red squares in Fig. 4). Comparing these values with those given in literature for devices based on wet-transferred CVD graphene (see circles in Fig. 4) (6), we can conclude that our fabrication technique leads consistently to devices with higher mobility and smaller potential fluctuations than those that can be achieved with the wet-transfer method.