Introduction

The advent of chemical vapor deposition (CVD) as a method for synthesizing graphene1,2 and other 2D materials3 has revolutionized their manufacturing. In contrast to the laborious and low-throughput process of exfoliating small flakes from crystalline materials4,5, CVD offers the ability to quickly synthesize large areas of high-quality, monolayer materials2,6,7,8,9, making them highly amenable to scalable processing. However, synthesis by CVD occurs on specific substrates, typically metallic surfaces10,11,12,13,14, which are essential to catalyze the nucleation and growth of the nanomaterial, but are unsuitable for subsequently making use of its optoelectronic properties. Consequently, CVD-grown films must usually be transferred from their growth substrate to a dielectric substrate compatible with electrical or optical measurements15,16,17. Transfer steps are also needed to assemble vertical assemblies of different 2D materials, also known as van der Waals heterostructures, whose combinatorial diversity is the object of great interest to achieve tunable optoelectronic properties18,19, twistronics20 and various applications in field-effect transistors21, photodetectors22 or biosensors9. Whichever the purpose, transfer processes are difficult to scale-up and remain a critical challenge impeding the industrialization of devices based on 2D materials23.

Graphene is among the most prominent 2D materials facing this challenge, as it is typically grown on copper substrates10,13,24. The traditional method developed to transfer graphene is the manual wet transfer technique17,25,26. In this technique, graphene on copper is covered with a thin polymer layer and placed floating at the surface of a copper etching solution. Once the copper is fully etched, the graphene/polymer stack is carefully picked up using a flat substrate and subsequently transferred in multiple cleaning baths, until it is finally picked-up by the destination substrate. A visual representation of this process is illustrated in Supplementary Fig. 1 in the Supporting material. A growing alternative to the chemical etching of copper is delamination27,28, in which the polymer-covered graphene sheet is electrochemically peeled off the copper substrate, followed again by successive rinsing steps until the final pick up with the target substrate. An obvious inconvenience of manual wet transfer techniques is that they require a high degree of manual dexterity, first to avoid tears and folds when manipulating the delicate polymer-graphene layer, but also to properly position the slippery hydrophobic material on the destination wafer. Each such transfer is thus time consuming and, more importantly, difficult to automate. These challenges are amplified exponentially when trying to integrate such techniques with full-sized wafer processing.

Various alternatives to wet transfers have been explored, all facing their own set of challenges. Dry transfer techniques, for instance, use an intermediary substrate to mechanically detach the graphene sheet from its growth substrate and transfer it onto the desired wafer, often assisted by heat or pressure. Such techniques have been successfully scaled up in roll-to-roll systems enabling the transfer of large surfaces of graphene12,16, but such processing has been observed to introduce impurities and defects in the graphene, and is primarily adapted to flexible destination substrates, such as kapton29. Recently, dry transfer automation has been proposed using robotics, albeit with relatively small areas of exfoliated material18. Other approaches have focused on the growth of graphene directly on dielectric substrates such as sapphire30, quartz31,32, germanium33, or PET34. These methods eliminate the need for transfer steps altogether, but they are constrained to specific substrates that may not necessarily align with further application requirements.

All things considered, wet transfer techniques still offer the most favorable combination in terms of versatility for the destination substrate and quality of the transferred material. Nevertheless, their viability for wafer-scale applications needs to be addressed. Over the last decade, several promising improvements to the basic manual technique have been proposed, aiming to reduce reliance on manual dexterity. For instance, Bosca et al.35 described an automated, manipulation-free wet transfer of cm2-scale graphene using peristaltic pumps to exchange fluids under the graphene instead of using manual pick-ups. Gao et al.36 proposed a wafer-scale face-to-face transfer wherein the growth occurs directly on a very thin copper layer deposited on the destination wafer: the graphene was then shown to directly stick to the wafer during the wet etching of the intercalated copper. Bae et al.15 demonstrated a roll-to-roll implementation of graphene growth, wet copper etching, and transfer on flexible or solid substrates, enabling the processing of large surfaces of graphene. In all large-area transfer techniques, it should be noted that the consumption of materials is notably high, which may represent another important hurdle towards industrialization. Specifically, the copper is entirely dissolved in the process, and most of the graphene transferred on the application wafer is subsequently etched away to leave only small graphene motifs embedded within the array of devices. Although electrochemical delamination is an avenue for reuse of the copper, most of the synthesized graphene material always goes to waste. Kireev et al.37 proposed to maximize the utilization of graphene material by parallelizing the wet transfer of an array of small graphene areas. However, their method requires delicate manipulations for transferring each graphene piece through multiple etching and cleaning baths and then into the openings of a floating kapton mask above the destination wafer; the resulting process is again time-consuming and reliant on advanced manual dexterity.

In this paper, we introduce a scalable method for transferring customized arrays of small graphene ribbons to an application wafer. Our approach uniquely combines both automation and parallelization, as well as innovative features enabling multilayer stacking capability and ribbon orientation control. First, we detail the conception and operation of the transfer apparatus, which consists of an array of glass chambers coupled with a dual inflow/outflow pumping system to enable wet transfer without any manual manipulations of the graphene layer. We then demonstrate how this setup can be used to automate the copper etching and transfer of multiple graphene ribbons in parallel, with accurate positioning and controlled orientation on a pre-patterned destination wafer. Using multiple characterization techniques, i.e. optical imaging, scanning electron microscopy (SEM), Raman hyperspectral imaging, and electrical transport, we show that the quality of the graphene transferred with our automated method is on par with that achieved through manual transfer by an expert user. Furthermore, we illustrate the unique capability of the technique for assembling vertical structures, by overlaying graphene ribbons with controlled relative orientations. We conclude by analyzing the practical advantages of the proposed technique, highlighting its suitability for future industrial wafer-based applications.

Results

Automated and parallel transfer apparatus

We conceived here an original instrument, affordable and simple to assemble, developed to enable the automated and parallelized wet transfer of oriented ribbons of CVD-grown 2D materials. Figure 1a illustrates the general architecture of the system, which consists of two key components: (1) a rigid array of transfer columns and (2) a dual inflow/outflow pumping system. The complete apparatus is photographed in Supplementary Fig. 2. The diameter and spacing of the transfer columns is tailored to match the pitch of the device array on the destination application wafer. For example, Fig. 1b, c show top views for two different array layouts: a 37-device layout covering a circular 4-inch wafer (panel b) and a 3 × 3 array on a square wafer (panel c). The shape of the columns will be discussed further along, after explaining the general operation of the system. This structure is placed in a large crystallizer or similar container, to serve as a bath. The rigid array of transfer columns is suspended slightly above the bath floor by external pillars, allowing for the destination wafer to be easily inserted and secured in place under the columns, or removed therefrom. Solutions in the bath are exchanged via tubing connected to two peristaltic pumps, one for inflow and another for outflow. Incoming fluid is injected in the spacing between the transfer columns, and each transfer column is equipped with two symmetrical openings in their walls, approximately at mid-height, allowing for the circulation of solution between columns. Simultaneously, outflowing fluid is pumped from the bottom of the bath, enabling to maintain a consistent filling height in the columns.

Fig. 1: Apparatus for the automated, parallel transfer of an array of graphene ribbons.
figure 1

a Schematics of the parallelized system, here for a 3 × 3 array. b, c Top view of different layouts of the array of transfer columns: (b) a circular array of 37 columns for transferring non-oriented ribbons on a 4-inch wafer, and (c) a 3 × 3 array of H-shaped columns for transferring oriented ribbons on a 3 cm × 3 cm wafer. d–g Process steps of the automated transfer: (d) Ribbons of polymer-graphene-copper are placed floating in the transfer columns filled with a solution of copper etchant. e Once the etching is complete, clean water is pumped in while the etching solution is simultaneously pumped out. f Once the columns are entirely filled with clean water, the destination substrate is secured under the columns, and (g) water is completely pumped out to delicately lower the graphene onto the substrate.

Full size image

The step-by-step automated wet transfer is illustrated in Fig. 1d–g, and technical details are provided in the Methods section. First, copper foils covered with CVD-grown graphene coated with polymer are cut into small pieces, typically squares or ribbons of a few mm2. A single piece is placed floating in each of the transfer columns of the apparatus, previously filled with a copper etchant solution (Fig. 1d). Once the copper is fully etched, the two peristaltic pumps are activated, allowing for the simultaneous removal of the etching solution and circulation of clean water to remove impurities, while keeping an almost constant filling height in the chambers (Fig. 1e). After completion of the solution exchange, the application wafer is immersed in the bath, slid under the transfer columns and held aligned by the pillars elevating the rigid column array (Fig. 1f). The remainder of the water is then fully and slowly pumped out of the crystallizer, letting the polymer-covered graphene pieces deposit delicately on the surface of the destination wafer (Fig. 1g).

The last important consideration is the design of the transfer columns themselves, specifically their shape and material, in order to control the placement and orientation of graphene ribbons. First, it is critical to prevent the graphene pieces from drifting and sticking to the column walls. It is, therefore, essential for the interior surface of the columns to be hydrophilic: this creates a concave meniscus in the aqueous solution that naturally keeps the graphene stack in the center of the column35. It also renders the column walls unfavorable for graphene adhesion, due to its high hydrophobicity. We tested different materials for the transfer columns, including plasma-treated teflon35 and 3D-printed polymer, but glass was found to give the most durable concave meniscus profile (Supplementary Fig. 3). Glass tubes are also readily available in various formats, and easy to cut and assemble in an arrayed structure using standard glassware tools. Cylindrical columns, such as illustrated in Fig. 1b and Supplementary Fig. 4a, make use of the concave meniscus principle to lower graphene pieces steadily in the center of the columns, allowing to position them precisely on the destination wafer. The center of each graphene piece, therefore, adheres first and keeps it in place while the rest of the water fully drains out from below the column via a small ~1 mm spacing between the cylindrical wall and the destination wafer. However, the graphene pieces can freely rotate in the center of the cylindrical columns because of the circular symmetry of the meniscus. To enable orientation control, we designed H-shaped columns, illustrated in Fig. 1c and Supplementary Fig. 4b, by affixing glass plates on two opposite walls of square-shaped glass columns. The resulting elongated meniscus maintains the alignment of ribbon-shaped graphene pieces, thus enabling the deposition of an array of aligned ribbons on the destination wafer. We found better results using H-shaped columns than simple rectangular columns, in which the graphene ribbons would often twist or buckle, probably due to an excessive curvature of the meniscus at the extremities of the ribbon.

Array positioning

Figure 2 shows photographs of an automated, parallel graphene transfer at each step of the process, using an array of circular columns matching a 3 × 3 motif of electrodes-bearing chips with a 1.2 cm pitch on a square wafer. Panel 2a shows a top-view photograph of the system in its initial state: all columns were filled with the copper etching solution and a small-area polymer-graphene-copper piece (approximately 4 × 4 mm) was placed floating inside each column. The system was left untouched for ~12 h, during which the etching of the copper occurred passively, then the dual pumping system was activated to exchange the bath content with clean water. A Supporting Movie of the solution exchange step is available online. Panel 2b shows the same system after etching and solution exchange: the copper is now dissolved in the solution and the remaining polymer-graphene stacks appear transparent. The 3 × 3 square wafer was inserted underneath the transfer columns and aligned using the side legs of the construct. With the inflow pump stopped, the water was entirely pumped out of the columns, thus slowly lowering the polymer-graphene stack until it was deposited on the wafer. A Supporting Movie of the deposition step is also available online. Panel 2c shows a photograph of the dried wafer: with the polymer layer still on top on the graphene, one can clearly see the nine fragments well-positioned in the center of each chip in the 3 × 3 motif. The polymer was then stripped off (panel 2d), leaving the wafer ready for further processing (e.g., channel patterning, device isolation, and/or passivation).

Fig. 2: Images of an automated, parallel graphene transfer in a 3 × 3 array.
figure 2

a Polymer-covered graphene on copper floating in cylindrical transfer chambers filled with a copper etching solution. b Remaining polymer-graphene after copper etching and solution exchange to clean water, with the patterned substrate inserted underneath the columns. c Polymer-covered graphene transferred in a 3 × 3 array matching the pre-patterned substrate. d Wafer after stripping off the resist. Scale bars in panels a–d is 1 cm. e Optical microscope image (×20, scale bar 80 μm) of the electrodes-patterned substrate covered with graphene, with no visible tears or folds.

Full size image

As seen by comparing the series of photographs in Fig. 2a–c, the graphene samples placed in the circular transfer columns are free to rotate but stay centered by the concave meniscus. From 21 samples, the graphene pieces were found to settle 0.4 ± 0.2 mm away from the center of the column, which has an inner diameter of 8.0 mm. In comparison, we performed an identical process using columns made of hydrophobic materials (teflon and polylactic acid (PLA)), which resulted in frequent sliding and sticking of the graphene to the column walls, yielding only a few or no graphene sheets on the wafer (not shown). Using the glass columns, Fig. 2c shows successful deposition of the whole array of graphene pieces, in perfect alignment with the motif of the wafer, without any manual user intervention.

Graphene quality

Next, we assessed whether the automated process preserves the integrity of the graphene material, especially in comparison to a benchmark of manual wet transfers obtained by an expert user. First, Fig. 2e shows a microscope image (×20 magnification) of the electrodes area at the center of one of the chips from the wafer in panel 2d, after the strip of the polymer layer. The whole field of the image is actually covered by graphene, and the electrodes motif is clearly visible through the atomically thin material. An image at the same magnification of the edge of the graphene fragment is provided in Supplementary Fig. 5. At the macroscopic scale observed here, the graphene exhibits no folds or tears over an area of 1500 μm2, as seen from the uniformity of the surface in the picture. Out of seven batches of automated transfer, we found no wrinkles or folds, in comparison to ~20 large wrinkles and folds across five individual manual transfers. For comparison, a 100× image of defective graphene showcasing tears, wrinkles, and folds is provided in Supplementary Fig. 6 from a manual transfer by a moderately new user.

The microscopic integrity of the graphene layer was investigated further, using SEM, hyperspectral Raman mapping (RIMA), and electrical characterization (see Methods section for experimental details). All these measurements were done on graphene transferred with the automated setup and compared with the same measurements on graphene transferred manually by an experienced user. The latter serves as a standard for assessing the quality of transferred CVD graphene. Figure 3a, b present SEM images of graphene transferred via our automated transfer setup (panel a) and using the traditional manual wet transfer technique (panel b). The graphene transferred with the automated technique again shows no visible tear or fold, confirming at this high magnification (×5000) the integrity of the graphene layer, as previously observed with optical microscopy. SEM images of the graphene obtained by both the automatic and manual techniques show similar results in terms of quality, which is verified by the absence of visible defects and comparable density of wrinkles.

Fig. 3: Microscopic and spectroscopic characterizations of transferred graphene.
figure 3

a, b Scanning electron microscope (SEM) images at 20 keV of (a) automatically transferred graphene and (b) manually transferred graphene. c, d Hyperspectral Raman maps of the ID/IG band intensity ratio over a 130 μm × 130 μm surface for (c) automatically transferred graphene and (d) manually transferred graphene.

Full size image

Raman spectroscopy is a frequently used technique to assess the integrity of graphene material38,39. A Raman spectra showing the typical D, G, D′, and 2D peaks of graphene is presented in Supplementary Fig. 7 for reference. Among the main peaks, the G band (~1580 cm−1) corresponds to an intrinsic vibrational mode of the graphene layer, while the D peak (~1350 cm−1) is activated by the presence of defects in the structure40. The intensity ratio of the D and G modes (ID/IG) is therefore used as a semi-quantitative indicator of the presence of defects in a graphene layer. Figure 3c, d present hyperspectral maps of ID/IG over a 130 μm × 130 μm area, for graphene transferred via the automated transfer apparatus (panel c) and for graphene transferred manually (panel d). Corresponding spectra averaged over the whole field of view are shown in Supplementary Fig. 8. Both maps show a homogeneous profile, indicating a consistent defect density over the whole field. Histograms of the ID/IG distributions for both manual and automated transfers can be found in Supplementary Fig. 9, allowing to extract both the mean and standard deviation of the intensity ratios. The average ID/IG obtained for the automated method is ~0.27 ± 0.09, which is generally considered representative of graphene with a small amount of defects41. This value is equivalent to that obtained with manually transferred graphene (~0.25 ± 0.08). This indicates that both graphene are of good quality. Hyperspectral Raman data further indicates that graphene transferred with both methods exhibit similar levels of strain and doping. For both methods, the relative positions of the 2D vs. G peaks follow a linear trend with a slope of ~1 (Supplementary Fig. 10), indicating similarly small amounts of strain42,43. Supplementary Figure 11 presents maps of the position of the G band over a 130 μm × 130 μm area for the graphene transferred manually (panel a) and with the automated transfer apparatus (panel b). The resulting distributions (panel c), although not overlapping exactly due to a shift also observed in the D band (panel d), correspond to similar levels of doping44. This suggests that the automated transfer method provides graphene quality that is comparable to that of manually transferred graphene.

Finally, the wafer obtained in Fig. 2d was further processed to form graphene field-effect transistors (GFETs), which are increasingly being used as on-chip sensors for chemical and biological detection45. Photolithography and reactive ion etching were used to pattern equivalent graphene channels between each electrode pairs (source and drain), and to remove the remaining graphene to isolate each electrical device from each other. The electrodes motif was designed to create 16 GFETs per chip, each with a 6-μm-wide and 4-μm-long graphene channel (see Supplementary Fig. 12 and Methods section for more details). Figure 4a shows a typical transfer curve for the resulting devices, using the doped Si layer of the substrate as a backgate electrode. The source-drain current (Isd) shows the expected V-shaped modulation by the applied gate voltage (Vg), with the typical asymmetry between hole and electron branches for devices assembled on Si/SiO245. The amplitude of the current and the smooth appearance of the curve are consistent with the presence of graphene of good quality between the electrodes. The distribution of charge neutrality points (i.e. gate voltage of the curve minimum) for multiple devices made from graphene transferred either manually or automatically is shown in Supplementary Fig. 13. Both distributions are equivalent, indicating a comparable doping of graphene after both transfer processes, as previously concluded from Raman data (Supplementary Fig. 11). An effective carrier mobility (μ) was computed for each device from the derivative of their transfer curve, using the following relation

$$\mu =\frac{L}{W{C}_{g}{V}_{sd}}\frac{d{I}_{sd}}{d{V}_{g}}$$
(1)

in which L and W are the length and width of the channel, Cg is the gate capacitance and Vsd is the applied source-drain voltage45. Figure 4b shows the resulting curve numerically differentiated from panel a. Effective electron and hole mobilities were extracted as the amplitudes of the maximum and minimum of the curve, respectively.

Fig. 4: Electrical transport characterization and effective carrier mobility.
figure 4

a Typical transfer curve of a GFET device produced using graphene transferred in the automated apparatus. b Effective mobility computed from the point-by-point derivative of the transfer curve. c Distribution of effective hole mobility in ensembles of GFETs made from automatically transferred graphene (color) and manually transferred graphene (black), both fitted with a lognormal distribution (lines).

Full size image

Figure 4c shows histograms of effective hole mobilities extracted, as just described, for an ensemble of 97 devices made from 4 batches of automated transferred graphene (red) and 81 devices made from 5 chips of manually transferred graphene (black). No significant statistical difference is detected between the two ensembles (p-value = 0.75 for a two-tailed two-sample Student’s t-test assuming non-equal variances), as well as no significant variation between chips and/or batches. By fitting each histogram with lognormal curves, shown as lines in Fig. 4c, we also find equivalent values for the geometrical mean of the effective mobility (μauto = μmanu = 102.4±0.3 cm2/(V ⋅ s) for both automated and manual transfers). One may note that the values of effective mobility extracted here (μauto~μmanu~ 125–500 cm2/(V ⋅ s)) are much smaller than the reported carrier mobilities for exfoliated graphene (~10,000 cm2/(V ⋅ s))46 or CVD graphene (~7000 cm2/(V ⋅ s))47; this is expected because net mobilities are measured using 4-probe measurements, while the effective mobility extracted here includes contact resistance. The effective mobility is used here solely as a base of comparison between devices either obtained from automated- or manually transferred graphene, in which electrical contacts should be comparable. The equivalence between both effective mobility distributions indicates again that the automated transfer techniques results in graphene of similar micron-scale quality as that obtained by manual transfer by an expert user.

Controlled orientation

An interesting aspect of the proposed technique is the design of H-shaped transfer columns (Fig. 5a) to control the orientation of graphene ribbons during the transfer. Figure 5b shows a photograph of such columns, in a 2 × 2 array configuration, loaded with polymer-graphene-copper ribbons on top of the copper etchant solution. The spontaneous alignment of the ribbons is clearly visible in the picture, and this parallel alignment was maintained until the deposition of the ribbons on a wafer, as shown in panel 5c. The resulting array of aligned ribbons on a wafer was achieved directly from the transfer, without requiring any further lithography and etching processes. Measurements of the angle of the ribbons after their transfer on the wafer show an average variation of ~5∘. This could likely be further optimized by fine-tuning the proportions of the H-shape columns.

Fig. 5: Automated transfer with controlled orientation of the graphene ribbons.
figure 5

a Schematics of H-shaped columns designed to maintain the alignment of graphene ribbons during the transfer process. b Photograph of ribbon-shaped polymer-graphene-copper stacks locked in parallel alignment in H-shaped columns. c Resulting transfer of aligned graphene ribbons on a wafer. d Rotation of the transfer columns in a second transfer, in order to overlay two ribbons with different orientations. e Photograph of the superposition of two graphene ribbons with a perpendicular orientation, showing the possibility to electrically connect each ribbon independently.

Full size image

Furthermore, successive transfers can be performed on the same wafer using this apparatus, including with different orientations of the columns (Fig. 5d), in order to create vertical assemblies of ribbons with controlled relative orientations. To illustrate this principle, Fig. 5e shows the overlay of two graphene ribbons with perpendicular orientations, obtained by rotating the apparatus by 90∘ between two successive transfers. A 5∘ angle is measured between the two ribbons, which is within the range of the combined tolerance for the orientation of the two ribbons. In addition, these perpendicular ribbons were deposited onto a wafer pre-patterned with two pairs of electrodes, each connecting one of the ribbons. This illustrates the capability of the technique to independently connect overlaid ribbons, which opens interesting possibilities for the realization of specific intra- and inter-layer measurements in van der Waals heterostructures.

Discussion

The automated parallel wet transfer technique developed here presents a number of advantages compared with either the traditional manual wet transfer or other wafer-scale transfer techniques, as summarized in Table 1.

Table 1 Comparison of the proposed automated parallel wet transfer technique (center) with manual wet transfer technique (left) and wafer-scale transfer techniques (right)
Full size table

First, while the proposed automated parallel technique provides transferred graphene of similar quality as that from traditional manual wet transfer, it is easier and faster in execution. Indeed, the automated technique does not rely on manual dexterity: the user does not need to manipulate graphene ribbons which are maintained in place by the meniscus inside the transfer columns. In our experience, we find that new users trained on the automated system obtain good transfers more quickly, whereas manual transfers require a lot of practice. In particular, the alignment of multiple graphene fragments onto a patterned wafer, a challenging task with manual transfer, is here completely automated by the positioning of transfer columns. This parallelization in multiple columns is also key in increasing the speed and throughput of the transfer step. With manual transfer, each fragment requires separate manipulations, especially in consecutive rinsing baths. Even with a well-organized sequence of operations, transferring a 3 × 3 array of graphene fragments would take ~3 h to an experienced user and the error rate due to missed alignments would be high. In comparison, transfer with a high success rate of the same 3 × 3 array automated with the pumps takes ~1 h. We note that this gain of time actually increases with the size of the array: while the time for a manual transfer scales approximately linearly with the number of transferred fragments, the time for the automated process is similar per wafer regardless of the number of transferred sites. The automated transfer setup could also be made compatible with larger area graphene transfers, but the current design was actually thought out so as to minimize graphene loss during etching processes, making it a more cost-effective system.

While advantages in speed and technical ease may also be found in some large-scale techniques based on transferring the entire wafer area with graphene15,36, the parallel wet technique proposed here also adds advantages in terms of materials economy, unique capability for controlled vertical assemblies, and potential furtherance towards resist-free transfers, as described in the following. After wafer-scale transfers, the micro-manufacturing of devices typically involves etching away all the area outside of small graphene channels - thus wasting most of the synthesized nanomaterial. This issue also applies to a lesser extent to manual transfers, which are difficult to execute with very small areas of graphene. In comparison, the automated parallel technique allows for transferring multiple graphene fragments, precisely positioned on the parts of the wafer where they are needed. This advantage is scaled when applied to the vertical stacking of multiple layers of nanomaterials. Small-area fragments, as small as a few mm2, can thus be superposed in an array of precisely-positioned stacks. We note that the transfer setup is compatible with exchanging solutions with removal solvents (e.g. acetone for PMMA), which can facilitate in situ removals of polymer supports between successive transfers without moving the substrate. The system also appears compatible with resist-free transfer approaches such as proposed by Zhang et al.48, in which the graphene layer is maintained at the interface between two non-miscible solutions. Our preliminary investigation of this implementation using non-polar n,n-hexane on top of the aqueous copper etchant yielded several transfers (see Supplementary Fig. 14), but more optimization is still needed at the moment to maintain the integrity of the atomically thin layer in a reproducible manner.

In addition to the maximization of graphene usage, parallelization, and automation also reduce the amount of ultrapure water and copper etching reagent used in the transfer process, compared to manual transfers. Manual transfers require consecutive rinsing baths, each with enough volume to dilute copper etchant residues and to accomodate the immersion and pick up motion of the substrate. For a 3 × 3 wafer as described earlier, we typically use a standard petri dish for the etching and then 2–3 water baths of 500 mL for a total of 1.5 L of ultrapure water. A few fragments can occupy the same petri or use the same baths, but these eventually need to be replaced after several individual transfers. In comparison, the automated system uses parallel columns to isolate the whole array of graphene fragments in the same bath during etching and washing. The pumps provide volume exchange to ensure washing of the copper etchant. For a 3 × 3 array, we typically use a 50 mL bath of copper etching solution and a 13:1 volume exchange with ultrapure water (Supplementary Fig. 14), for a total usage of 700 mL for the whole array. Again, this economy would increase for larger arrays. The reduced comsumption of graphene, reagents, and ultrapure water makes the proposed technique more economical and sustainable for environmental considerations.

Finally, we showed that H-shaped columns enable the arrayed transfer of oriented ribbons of graphene onto a wafer. This is exceedingly difficult to achieve with successive manual transfers, but accessible with wafer-scale techniques simply by patterning graphene ribbons with lithography and etching after the transfer process. However, such post-transfer patterning of specific layers in vertical assemblies of 2D materials is still very challenging. Layered 2D materials are most often obtained by successive dry transfers, for which some parallelization was only recently demonstrated with exfoliated small fragments using robotization18. We showed that parallel wet transfers can be used to simultaneously assemble arrays of vertically stacked ribbons with controlled relative orientation between them. The current precision of ~5∘ is of course still too coarse to take advantage of lattice superposition effects, such as long-range Moiré patterns and twistronics; whether it could be improved to that precision with further engineering is to be determined. One should keep in mind though, that although efforts have been made in the growth of large monocrystalline graphene domains49, lattice orientation is usually an uncontrolled variable in standard large-scale CVD growths, which generate polycrystalline materials with grains of various orientations. In that context, the main advantage in controlling respective ribbons orientations is to facilitate their independent addressing by electrodes; for this purpose the current precision of the proposed wet transfer approach is more than sufficient.

Conclusion

In conclusion, the apparatus consisting of an array of strategically shaped and spaced glass transfer columns combined with a dual pump system offers an effective solution for the automated and simultaneous transfer of aligned graphene nanoribbons on pre-patterned wafers. The technique holds significant potential for the industrialized production of large number of wafer-based devices: The method is scalable, reduces reliance on manual dexterity, saves time and materials, and is performed with an apparatus that is affordable and easy to assemble. Moreover, variations of the apparatus with different column orientations can be used sequentially on the same wafer to assemble vertical assemblies of 2D ribbons with relative orientations, facilitating their independent wiring to electrical connections. Finally, this technique is likely to be applicable to various types of large-area CVD-grown 2D materials, such as hBN and transition metal dichalcogenide layers. This broadened use opens promising avenues for studying vertical junctions involving diverse 2D materials, such as van der Waals heterostructures and layered structures incorporating encapsulated particles or molecules.

Methods

Graphene growth

Graphene was grown by chemical vapor deposition (CVD)13 on thick copper foils (Alfa Aesar 13382 copper, purity 99.8%, 0.025 mm thick). The copper foils were first annealed at 1000∘C for 30 min under 50 mTorr of H2 (Praxair, UHP, grade 5, O2< 1 ppm). After annealing, 100 mTorr of non-purified CO2 (Praxair, Anaerobic, grade 4, 99.99% purity) was introduced to etch carbon contaminants emanated from the copper during annealing. Finally, the foils were exposed 10 min to 450 mTorr of CH4, the latter acting as the precursor for carbon germination at the surface of the copper. Gas inlets were then cut and samples left to cool down before being removed from the growth chamber. The top of the Cu foil was then spincoated with 495K PMMA A6 polymer, left to cure for 2 h, following which the backside of the Cu foil was etched using an O2 plasma (10 sccm, 150 W, 45 s).

Automated transfer

PMMA/graphene/Cu samples were cut to the desired size and positioned at the surface of the transfer columns previously filled with a copper etchant solution of 0.1 M of ammonium persulfate solution (APS, ACS 98+%, Acros Organics). Samples were left in the columns for ~12 h, or until the copper was etched completely. Using a Control Company 3386 Variable-Speed Peristaltic Tubing dual-pump system with a flow rate varying between 0.4 mL/min and 85 mL/min, the APS was then exchanged with milli-Q water (MQ), using approximately 13:1 MQ:APS volume exchange (Supplementary Fig. 15). The target wafer was then inserted underneath the columns, and the MQ water was slowly pumped out to allow PMMA-covered graphene samples to land on the wafer. See Fig. 1 and Supplementary Fig. 2 for illustrations of the apparatus and its step-by-step operation.

Manual transfer

PMMA/graphene/Cu samples were cut into the desired size and placed at the surface of small petri dishes filled with 0.1 M APS to etch overnight. The resulting PMMA-covered graphene samples were then picked from the APS bath using a clean substrate, and placed floating in a crystalizer filled with 300 mL of MQ water for 10 min. This process was repeated twice using fresh MQ, following which the PMMA-covered graphene was picked with the final target substrate and left to dry. See Supplementary Fig. 1 for a step-by-step illustration of manual transfers.

Device arrays

For transport measurements, pre-patterned Si++/285 nm-SiO2 wafers were diced either in individual chips (manual) or in larger arrays of chips (automated). Each chip had a motif of 16 electrode pairs with a 4 μm separation (see Fig. 2). To form the electrodes, empty substrates were first dehydrated, then spincoated with positive photoresist (OIR-674, 4000 rpm, 30s) followed by annealing at 90∘ for 60 s. Electrodes were defined by photolithography, developed in AZ-726 (45 s, followed by 45 s quenching in water), and metallized by evaporation of 10 nm Ti and 30 nm Au. Lift-off was performed in Remover PG for 1 h, followed by acetone/IPA cleaning and nitrogen drying. Graphene fragments were then transferred on the electrodes motifs using either the automated or manual method. Following transfer, the PMMA protective layer was removed via successive 30-min baths of acetone and IPA, followed by drying under N2 flow. The graphene layer was further etched into 6 μm-wide ribbons connecting each of the electrode pairs (see Supplementary Fig. 12), using photolithography (OIR-674, 4000 rpm, 30s) and oxygen plasma etching (10 sccm, 150 W, 1 min).

Microscopy and spectroscopy

Optical imaging was performed at different magnifications (from 20x to 100x) with a Motic trinocular microscope (BA310MET-T). Scanning electron microscopy (SEM) was performed on a Raith equipement, at 20 kV, with a magnification of 5K. Raman imaging was performed with a Raman hyperspectral imager (RIMA) from Photon Etc., with a 532 nm excitation laser at 3 W, over an area of 130 μm × 130 μm.

Electrical measurements

Electrical measurements were performed using an automated probe station (FormFactor Summit) coupled to a Keysight semiconductor parameter analyzer. A constant bias of Vsd = 50 mV was applied between the source and drain electrodes, while the backgate voltage Vg was modulated between -100 V and 100 V. Source-drain current Isd going through the graphene channel was then recorded for each device.