We transferred graphene by using PMMA with two AMWs (15k and 550k), which were dissolved in anisole at two weight ratios (2 and 4 wt %). PMMA with 950,000 (950k) AMW (at 4 wt %), commonly used for microfabrication processes as an e-beam resist, was used for further comparison (see Experimental section, “Graphene transfer“). Optical microscopy analysis was carried out to visually evaluate the presence of PMMA residues after the transfer process of graphene single crystals using PMMA with different weight percentages and AMWs (Figure 1a–g). As detailed in the description of the graphene transfer process, after the Cu etching process (Supporting Information File 1, Figure S1b, step II), the PMMA-coated graphene is rinsed in a DI water bath at least three times (Supporting Information File 1, Figure S1b, step III). Each cycle includes two actions, namely (i) scooping up the sample and (ii) releasing it into the water bath. After the rinse process, the sample must be moved to a target substrate, which takes one more transfer cycle. Therefore, the wet transfer process entails at least four cycles. A sufficiently high mechanical strength of the supporting PMMA layer is the key requirement for a successful transfer. Figure 1g shows our experimental observations on the cleanness level and maximum number of transfer cycles afforded by each PMMA solution. For statistical purposes, the number of cycles for each test was extended to eight by transferring the PMMA/graphene between two aqueous solutions. A1 and C1 PMMA solutions (see Table 1 below for the sample denominations) allowed for less than four cycles without breakage, demonstrating a low mechanical strength. The strength appears to reach a proper level for A3 PMMA, which allowed for up to six complete cycles. C3 and C4 PMMA provided the highest mechanical support, allowing for up to ten cycles. To evaluate the process cleanness, the residues were quantified by counting the white spots (larger than 2.5 µm) in the images. Micrographs taken on 650 × 500 µm2 areas were compared (Figure 1a–f). The amount of PMMA residues appears to be directly related to the PMMA concentration. A1 showed very little residues (they can be almost completely removed using acetone) and low mechanical strength, whereas C1 provided proper support while maintaining a low residue level. The length of the PMMA molecular chains (proportional to molecular weight and the attractive intermolecular force) appeared to be the determining factor in the mechanical strength (as demonstrated by C3 PMMA, enabling eight complete transfer cycles). Overall, all 550k samples provided strong mechanical support yet translated into a moderate-to-high residue density level. C4 revealed the densest residue distribution due to extended molecular chains and the highest mass concentration.

Figure 1: Examination of PMMA residues after the transfer of graphene single crystals using PMMA at various weight percentages and AMWs. (a, b) A1, (c) C1, (d) A3, (e) C3, and (f) C4. The white spots in the images are PMMA residues. The cross markers with coordinates are imprinted on the wafer to locate the graphene crystals. (g) Number of residue particles counted as function of the average molecular weight of PMMA.

A PMMA mixture (coded B2, 3% mixture of PMMA-15k and PMMA-550k, see Table 1) was designed to lower the potential residue concentration while maintaining proper mechanical support. The rationale for the design of the optimized mixture is based on two hypotheses: (i) The PMMA-15k component provides short polymer chains, which are expected to diminish the molecular chain entanglement and, hence, the residue level. (ii) The sole presence of short polymer chains should, however, weaken the mechanical strength of the spin-coated layer. Therefore, the addition of PMMA-550k compensates for that and grants support during the transfer. B2 was tested in the transfer of both a graphene film and a single crystal (Figure 2a,b). The area analysis (650 × 500 µm2) revealed less than ten residues, indicating an extremely clean transfer process. B2 PMMA allowed for up to six transfer cycles, representing an intermediate, yet acceptable, mechanical support. This proves that the PMMA mixture features good mechanical strength and cleanness (i.e., the acetone bath can thoroughly remove it). The transferred graphene samples were investigated via Raman spectroscopy to evaluate crystallinity, layer number, and structural defect level [23]. The relative intensities of the G (ca. 1585 cm−1) and 2D (ca. 2700 cm−1) bands are typical of monolayer graphene [23-26]. The defect density appears minimal considering the negligible D band intensity at ca. 1350 cm−1 (Figure 2c) [27]. The Raman mapping in Figure 2d–i examines the whole crystal area [28]. The map and the corresponding statistics in Figure 2d and Figure 2g, respectively, show that the I(D)/I(G) ratio is very low (down to 0.03), meaning that no or few defects could be detected. Figure 2e and Figure 2h show that over 95% of the sample has a ratio of I(2D)/I(G) > 1.6 (average of 2.1 ± 0.3) and a FWHM(2D) of 34.2 ± 3.0 cm−1 (Figure 2f,i), implying high-quality monolayer graphene. Together, the data further support uniformity and crystallinity of the sample.

Figure 2: Raman analysis of graphene samples transferred with the optimized B2 PMMA solution. Optical images of (a) a large-area film and (b) a single crystal. (c) Raman spectra taken at the positions indicated in (b). Raman mapping of (d) I(D)/I(G), (e) I(2D)/I(G), and (f) FWHM(2D), and (g–i) corresponding statistics.

The G phonon band arises from double degeneracy of iTO and iLO phonon modes (E2g symmetry) at the Brillouin zone center, which is an in-plane vibration of sp2 carbon atoms [23], and its position displays a blueshift as the charge carrier concentration rises. That is, the frequency shift of the G band is proportional to |EF|, which sets the carrier concentration. Due to the method and materials employed for the graphene transfer being the same except for the PMMA mixture, we consider that the differences in Raman spectra between the different samples can be attributed to adsorbed PMMA residues. Such residues could absorb water and oxygen molecules, conferring p-type doping to graphene [29]. Conversely, by tracking the G phonon band features in Raman spectra, we can correlate the degree of doping in graphene [30,31] with PMMA residues. A comparison of the statistical data of the G band position and FWHM(G) of graphene transferred with C4 and B2 PMMA is presented in Figure 3. Representative Raman spectra for the two cases are shown in Figure 3a and Figure 3b, respectively. The comparison of Figure 3c and Figure 3d shows that the G peak position is blueshifted in C4 samples compared to B2 samples (from an average of 1587 to 1593 cm−1), indicating that the charge carrier concentration did not rise as much when using B2 PMMA. This shows that B2 leaves behind a much lower (if any) density of residues. This redshift observed upon reducing the PMMA residue concentration is consistent with studies on advanced methods for cleaning PMMA from graphene [32]. The average FWHM(G) for the C4 and B2 samples is 14 and 23 cm−1, respectively (Figure 3e and Figure 3f). The broader G phonon band observed for the B2 samples reveals that a significantly higher number of inter-band decay pathways are available due to a lower Pauli blocking threshold [23] (equivalent to twice |EF|), further indicating that the p-type doping caused by adsorbed PMMA is less intense for the B2 samples. This result again supports that employing B2 PMMA yields fewer residues and may help in avoiding post-transfer treatments for advanced PMMA residue cleaning of graphene, such as annealing and ion beam irradiation [32]. The graph of the G band shift (Supporting Information File 1, Figure S2a) confirms that C4 PMMA leaves the highest level of contamination. This case also shows the highest standard deviation of the G band shift and FWHM(G) (Supporting Information File 1, Figure S2a,b) due to heterogeneous doping levels in the sample. We explain these results by the higher variance in the proportions of PMMA residue aggregates, resulting in alternating regions of intensively local p-type doping (large PMMA aggregates) and regions of less intense p-doping (small PMMA aggregates). In contrast, PMMA mixtures with lighter AMWs showed lighter and more uniform p-type doping over the crystallite area (smaller error bars). Few-layer crystals transferred with B2 are analyzed in Supporting Information File 1, Figure S3. The graphene crystal in Supporting Information File 1, Figure S3a,b is composed of four layers having a thickness of 0.4–0.5 nm (Supporting Information File 1, Figure S3c) [28,33]. The crystal morphology appears very smooth and free of identifiable impurities, with an average surface roughness (Ra) of ca. 0.2 nm (Supporting Information File 1, Figure S3d). This value is one order of magnitude lower than that of graphene crystals transferred with C4 PMMA, which showed a surface roughness of ca. 2.8 nm. The low roughness obtained with B2 PMMA can be also related to a minimal occurrence of nanometer-scale PMMA residues.

Figure 3: Statistical analysis of the Raman spectra of transistors prepared with transfers using PMMA mixtures with different molecular weights. (a, b) Representative Raman spectra of graphene transistor channels prepared using B2 and C4 mixtures, respectively. (c, d) Distributions of the G band peak position for B2 and C4, respectively. (e, f) Distributions of the FWHM of the G band.

XPS was employed to analyze the graphene samples transferred using C4 and B2 PMMA (Figure 4). Figure 4a shows the chemical structure of the PMMA molecule. The C1s spectra can be decomposed into two prominent peaks originating from sp2-hybridized C–C and sp3-hybridized C–C/C–H bonds. Four PMMA-related peaks can be assigned to C–H, C–C, O–CH3 (methoxy functional group), and O–C=O (carboxy functional group) bonds, respectively [34-36]. PMMA residues on the graphene surface mainly feature three peaks resulting from C–C bonds and carbon–oxygen-related bonds (i.e., methoxy and carboxy functional groups). After PMMA has been removed in the acetone bath, the peak of the C–H bond can rarely be observed because of a broader merger with the peak of sp3-hybridized C–C/C–H bonds in graphene. In Figure 4b, the peak intensities of C–C, O–CH3, and O–C=O bonds are, respectively, 17.6%, 14.6%, and 15.0% of the main peak intensity (sp2 C–C). In Figure 4c, the same ratios decrease to, respectively, 2.9%, 2.6%, and 4.8%. Therefore, regarding O–C=O bonds, the residue caused by B2 was reduced by three times and decreased about six times for the removal of C–C and O–CH3 bonds, implying the crucial role of PMMA-15k in the mixture.

Figure 4: (a) Representation of the molecular structure of PMMA. XPS C1s spectra of graphene samples transferred using (b) C4 and (c) B2. The normalized spectra are fitted by Gaussian–Lorentzian curves. The solid blue and grey fills identify, respectively, sp2- and sp3-hybridized carbon bonds in graphene, located at ca. 284.4 and 285.0 eV, respectively. C–H (blue line), C–C (purple line), O–CH3 (pink line), and O–C=O (orange line) bonds are located at ca. 285.7, 286.3, 287.0 and 289.0 eV, respectively.

The optimized graphene transfer process was statistically validated in a batch fabrication context by comparing the channel resistance of electrolyte-gated graphene field-effect transistors (GFETs, Figure 5) designed to operate as DNA biosensors. Two batches of GFETs having a topmost graphene channel (75 µm width × 25 µm length) were fabricated (see details in Supporting Information File 1). In the first batch [37,38] (including 1755 GFETs), graphene was transferred with the C4 PMMA mixture (see Section 2.2). The data acquired from the first batch were used to benchmark a successive test on a more extensive second batch [39,40] (4200 GFETs) that used B2 PMMA for the graphene transfer. Figure 5 shows the resistance distribution in the two cases. The distributions were fitted with multiple Gaussian curves to identify the predominant resistance values in each transistor batch. The first Gaussian curves of C4 and B2 (which envelop the most common resistance bins, see Figure 5) peak at 1860 and 690 Ω, respectively. This is a difference of 1170 Ω. The much lower channel resistance for the B2 mixture gives further evidence of a more robust, cleaner, and more effective transfer process. The B2 transfer appears to minimize the PMMA residues, known to act as centers of carrier scattering in graphene and to increase its resistance [19,20]. The vast majority (ca. 90%) of B2 data populate the first Gaussian curve with a narrow distribution (FWHM of ca. 300 Ω). In stark contrast, the C4 data are scattered over a much broader range. Almost 50% populate a broader Gaussian curve with two peaks (FWHM of the first peak: ca. 567 Ω) in the range up to 15 kΩ, while more than 20% have values up to 1 MΩ. Differently from B2, a consistent sample subset (above 30%) is above 1 MΩ, which means that a third of the fabricated devices are open circuits due to an imperfect graphene transfer (leading to highly damaged or lacking graphene). Overall, the B2 mixture yields more consistent electric properties of the graphene channel, thanks to a homogenous and reproducible process. Such characteristics ultimately translate into a consistent sensor performance, which is pivotal for industrial fabrication.

Figure 5: Distributions of the graphene channel resistance values in devices made using C4 (blue) and B2 (red) PMMA. The inset shows optical micrographs of representative B2 (left) and C4 (right) transistors. The graphs show the resistance distributions in (a) the 0–15 kΩ range (with 125 Ω bins) and (b) the range from 15 kΩ to 1 MΩ (10 kΩ bins). The multiple Gaussian fits identify the dominant resistance values in both batches. When considering all sub-15000 Ω transistors, the difference between the average resistances is 2139 Ω.