Assessing the quality of large-area monolayer graphene grown on liquid copper for size-selective ionic/molecular membrane separations

Monolayer graphene growth on liquid copper (Cu) has attracted attention due to advantages of a flat/smooth catalytic growth surface, high synthesis temperature (>1080 °C) as well as the possibility of forming graphene domains that are mobile on the liquid Cu with potential to minimize grain boundary defects and self-assemble into a continuous monolayer film. However, the quality of monolayer graphene grown on liquid copper and its suitability for size-selective ionic/molecular membrane separations has not been evaluated/studied. Here, we probe the quality of monolayer graphene grown on liquid Cu (via a metallurgical process, HSMG®) using Scanning Electron Microscope (SEM), High-resolution transmission electron microscope (HR-TEM), Raman spectroscopy and report on a facile approach to assess intrinsic sub-nanometer to nanometer-scale defects over centimeter-scale areas. We demonstrate high transfer yields of monolayer graphene (>93% coverage) from the growth substrate to polyimide track etched membrane (PITEM, pore diameter ∼200 nm) supports to form centimeter-scale atomically thin membranes. Next, we use pressure-driven transport of ethanol to probe defects > 60 nm and diffusion-driven transport of analytes (KCl ∼0.66 nm, L-Tryptophan ∼0.7–0.9 nm, Vitamin B12 ∼1–1.5 nm and Lysozyme ∼3.8–4 nm) to probe nanoscale and sub-nanometer scale defects. Diffusive transport confirms the presence of intrinsic sub-nanometer to nanometer scale defects in monolayer graphene grown on liquid Cu are no less than that in high-quality graphene synthesized via chemical vapor deposition (CVD) on solid Cu. Our work not only benchmarks quality of graphene grown on liquid copper for membrane applications but also provides fundamental insights into the origin of intrinsic defects in large-area graphene synthesized via bottom-up processes for membrane applications.

In this context, the synthesis of graphene on liquid copper via the metallurgical method (High Strength Metallurgical Graphene -HSMG ® ) has attracted attention as a new avenue for the scalable production of highquality graphene [25][26][27][28] due to advantages such as: (i) it allows large graphene domains via low graphene nucleation density and faster growth rates at high temperature >1080 °C [29][30][31][32][33][34][35], (ii) the smooth liquid surface during growth coupled with the high temperature allows for potentially reduced defects in the graphene domains [29], and (iii) the liquid Cu enables the graphene domains to be mobile allowing self-assembly in a continuous film via domain rotation that could potentially minimize grain boundary defects [31,36].However, to the best of our knowledge, the quality of HSMG ® for membrane applications (centimeter-scale areas) has not yet been assessed, evaluated or studied.
Here, we report on a facile approach to characterize quality of large-area HSMG ® by forming atomically thin membranes and probing pressure-driven as well as diffusion-driven transport of analytes through the same membrane area.We probe sub-nanometer and nanometer scale defects over centimeter-scale areas of HSMG ® and compare its quality with high-quality CVD graphene grown on solid Cu.Our results show that although liquid Cu method offers large graphene domains and potentially results in low-angle domain boundaries, intrinsic sub-nanometer and nanometer scale defects still exist and are no less than found in high-quality CVD graphene grown on solid Cu [5-7, 9, 10, 12, 13, 19-22, 24, 37].Our results indicate that the quality of graphene for membrane applications and ionic/molecular transport in particular are susceptible to intrinsic defects incorporating within individual graphene domains during growth of monolayer graphene via bottom-up synthesis methods.

Graphene growth
High Strength Metallurgical Graphene was synthesized on a liquid Cu surface as reported in detail elsewhere [35,38,39] with some minor modifications.At first, 0.2 mm Ni carburized foil was electroplated with Cu using a solution of copper sulphate pentahydrate (CuSO 4 •5H 2 O), at a current density of 0.02 A cm −2 , for 6 h.Next, the plated foil was heated to 1060 °C under 10 Pa pressure in a reactor (SuperCarb, Seco/Warwick SA) and carburized in an atmosphere of acetylene (0.4 l min −1 ), ethylene (0.4 l min −1 ), and hydrogen (0.2 l min −1 ) mixture using four cycles (each including 5 s of dosing followed by 15 min of hold time for diffusion).Finally, the reactor was heated to 1100 °C in pure argon atmosphere at 2 kPa (5 l min −1 ) to melt the Cu and held for 5 min before the sample was cooled down with the reactor to room temperature naturally.

Graphene transfer
Graphene was transferred onto PITEM substrates using a polymethyl methacrylate (PMMA) carrier layer [40].PMMA (4 wt% in anisole) was spin-coated on graphene at 1000 rpm for 15 s and 1500 rpm for 30 s, followed by drying in air for 2 min.The process was repeated three times to coat four-layers of PMMA on the graphene surface.Subsequently, the PMMA/graphene/metal stack was baked at 130 °C for 20 min.After cooling to room temperature, the metal foil was completely etched in 0.5 M FeCl 3 solution for 24 h and the obtained PMMA/ graphene stack was floated on 3% HCl solution and DI water successively.
PITEM (hydrophobic, ∼200 nm diameter track etched pores, ∼8 μm thick) support (purchased from it4ip) was used as the transfer substrate.Prior to transfer, PITEM substrate was pre-treated with didecylamine (DDA, 0.05 M in ethanol) for 1 h to enhance its hydrophobicity, followed by 3 rinses (10 min each) in separate batches of ethanol to wash off any DDA residue.The PMMA/graphene stack floating on DI water was scooped onto the PITEM substrate, dried in air at room temperature and baked at 120 °C for 30 min to improve adhesion.Finally, acetone was used to remove the PMMA, followed by rinsing the graphene/PITEM in isopropanol and drying in air.
Graphene was also transfer onto 300 nm SiO 2 /Si wafers using the PMMA method as detailed above [5-7, 9, 13, 20, 21, 23].The only difference was once the PMMA/graphene stack was transferred onto a SiO 2 (300 nm)/Si wafer it was dried in air, followed by baking at 45 °C, 60 °C and 90 °C for 30 min each, washing in acetone overnight, and cleaning in IPA for 1 h.
The PMMA method was also used to transfer graphene to TEM grids [5-7, 9, 13, 20, 21, 23].The PMMA/ graphene stack was scooped from DI water using a copper TEM grid.The TEM grid with PMMA/graphene was baked in air at 60 °C for 30 min, and then placed in acetone hot vapours for 20 min, followed by washing in acetone overnight, cleaning in IPA, and drying in air.

Graphene characterization
SEM images of HSMG ® graphene on Cu/Ni foils were acquired using FEI Helios NanoLab G3 CX scanning electron microscope at ∼2 kV.SEM images of HSMG ® graphene on PITEM membranes were collected by using a Zeiss Merlin Scanning Electron Microscope with Gemini II Column operated at ∼2 kV.
The surface morphology of HSMG ® graphene on growth substrate (Cu-Ni) was evaluated by a confocal laser scanning microscope (CLSM, Nikon MA200 microscope with inverted optics, equipped with a C1 confocal system) using an argon laser with a wavelength of λ = 488 nm.Images with resolution of 512 × 512 pixels were recorded using an EZ-C1 Free Viewer software, while detailed analysis of confocal microscopy data was performed using a Mountains Map Premium software (Digital Surf, France).
Raman spectra were obtained by using a Thermo Scientific DXR Confocal Raman spectrometer (∼532 nm laser) with 1 mW laser power.
High-resolution transmission electron microscopy (HR-TEM) images were recorded by using an FEI Talos F200X transmission electron microscope.

Transport measurements
The pressure-driven ethanol transport and diffusion-driven solute transport measurements across the membranes were performed as reported elsewhere [2, 5-7, 9, 13, 20, 21, 23, 24, 41, 42].A side-bi-side glass diffusion cell (figure 3(E), 7 ml volume on each side, 5 mm orifice, PermeGear, Inc.) with a gastight syringe (250 μl) installed on the short opening of the feed-side cell (left cell) was used for probing the transport of analytes through the membranes.The membrane was installed between two diffusion cells with the graphene side facing the feed side.Magnetic Teflon-coated stir bars were stirred vigorously at 1500 rpm in both cells to prevent concentration polarization.All the measurements were repeated for three times to obtain average values and standard deviations.Initially, the system was washed three times (5 min each time) with ethanol (190 proof) to fully wet the membrane.Subsequently, ethanol was filled into both cells and a height difference in ethanol between feed and permeate side was generated to induce a hydrostatic pressure gradient.A digital camera was used to record the drop of ethanol meniscus level along the graduated syringe every 1 min.Ethanol permeance was calculated by permeance , where ΔV is the ethanol volume change (drop), ΔP is the pressure difference across the membrane (∼300-450 Pa), Δt is the time interval, and A effective is the effective membrane area during the transport process.Normalized flux was calculated by computing the ratio of ethanol permeance for PITEM +graphene membrane with respect to bare PITEM membrane.Before performing diffusion-driven transport, the system was washed with DI water for five times to completely replace and remove ethanol.Four model solutes were selected for the transport measurements: KCl (hydrated diameter of K + ion ∼0.662 nm, hydrated diameter of Cl − ion ∼0.664 nm), NaCl (hydrated diameter of Na + ion ∼0.716 nm), Vitamin B12 (B12, molecular diameter ∼1-1.5 nm), and Lysozyme (Lz, molecular diameter ∼3.8-4 nm).For salt transport measurement (KCl or NaCl), the feed-side cell was filled with 7 ml of 0.5 M salt solution, while the permeate side was filled with 7 ml of DI water.A conductivity meter probe (connected to a Mettler Toledo SevenCompact S230 conductivity benchtop meter) immersed in the permeate side was used to measure the conductivity every 15 s for 15 min.For organic molecule transport measurements (B12 or Lz), the feed-side cell was filled with 7 ml of organic molecule solution (1 mM in 0.5 M KCl), while the right cell was filled with 7 ml of KCl solution (0.5 M).A fiber optic dip probe (attached to an Agilent Cary 60 UV-vis Spectrophotometer) immersed in the permeate side was used to record the absorbance spectra in the range of 190 to 1100 nm every 15 s for 40 min.The UV-vis intensity differences between B12 (peak position ∼360 nm), Lz (peak position ∼282 nm) and DI water (reference wavelength ∼710 nm) were used for computing the concentration of permeating species.
The transport rate of each solute was calculated via the slope of concentration change in the permeate side, and the normalized flux was computed by computing the ratio of transport rate for PITEM+graphene membrane with respect to bare PITEM substrate.The effective normalized flux was calculated by ( ) normalized flux of each species normalized flux of ethanol normalized flux of ethanol 1 --

Results and discussion
3.1.Characterization of graphene synthesized on liquid Cu Large-area graphene was synthesized on a liquid Cu via the HSMG ® process [35,38,39] with some modifications (figure 1(A)).Pre-carburized Ni foil was used as a support for electroplating Cu due to the higher melting point of Ni and good wettability of nickel by Cu.The bimetal was heated in a chamber and carburized in hydrocarbon vapors without forming any carbon deposits on the Cu surface.Next, the Cu is molten by increasing the temperature and the lower carbon solubility in liquid Cu than solid Cu pushes carbon to the top of the Cu surface, resulting in the formation of monolayer graphene [31,35,39,43].The main advantage of this method is the potential for self-assembly for the growing graphene domains (high mobility and/or rotation on the liquid Cu surface) minimizing defects along domain boundaries in a continuous monolayer film [31].SEM image in figure 1(B) shows large graphene domains (∼500 μm) due to low graphene nucleation density and faster growth rates at high temperature (>1080 °C) [29][30][31][32][33][34][35]. Figure 1(C) shows a continuous film of graphene covering the catalyst surface identified via wrinkles like features in graphene that arise from differences in thermal expansion between graphene and Cu during cooling process in HSMG ® .Some small areas of multi-layer graphene regions (darker contrast) resulting from excess carbon supply during synthesis are also seen.
The surface morphology of HSMG ® graphene on growth substrate (CLSM image in figure 1(D)) shows visible Cu grains of various sizes with a surface roughness Ra ∼ 0.779 ±0.298 μm.We note that the surface roughness results from the slow heteroepitaxial growth of solid Cu grains on the nickel substrate [44].
To assess the quality of HSMG ® , we acquired HR-TEM images as well as Raman spectra (figure 2).HR-TEM images (figure 2(A)) show a film with some line like features.Magnified regions in the center, left, and right side of the feature are marked with red, green, and purple boxes, respectively.Figure 2 [45,46].The D peak and increased I D /I G ratio indicate the presence of defects associated with graphene wrinkles.We note that the I D /I G ratio of monolayer HSMG ® ∼0.09 is higher than that of high-quality CVD graphene [20] ∼0.027, further confirming higher intrinsic defect density in HSMG ® .

Graphene transfer onto PITEM substrate and fabrication of atomically thin membranes
The synthesized HSMG ® graphene was transferred onto PITEM substrate using a sacrificial polymer (figure 3(A), see Methods section) to form an atomically thin membrane.We specifically choose PITEM with parallel cylindrical pores ∼200 nm diameter as model supports due to (i) its straight, well-defined channels that effectively prevent inter-connected pores and allow for unambiguous interpretation of transport measurements and (ii) PITEM's compatibility with acetone, which has to be used to dissolve polymer carrier layer from transfer.
SEM images (figures 3(B), (C) confirm successful transfer of HSMG ® onto PITEM support, where most PITEM pores are covered with graphene and appear darker due to graphene's electrical conductivity (figures 3(B), (C)).Uncovered PITEM pores appear brighter due to polymer charging [5,20].Wrinkles in graphene are also observed in the SEM images as well as some tears in graphene that appear as bright open regions co-located along wrinkles in some instances (figure 3(B)).To further confirm the transfer of HSMG ® onto PITEM support, we collected the Raman spectra on bare PITEM and HSMG ® transferred onto PITEM (PITEM+HSMG ® ).As shown in the figure 3(D), compared with bare PITEM, there is a clear 2D peak which belongs to HSMG ® in the PITEM+HSMG ® sample (G peak is overlapped with the Raman signal of PITEM), indicating the successful transfer of HSMG ® onto PITEM.

Assessing the quality of graphene via ionic and molecular transport
The synthesized atomically thin monolayer graphene membrane was mounted in a side-bi-side diffusion cell (figure 3(E)) to evaluate the nanometer and sub-nanometer scale defects over large area by measuring pressuredriven ethanol transport and diffusion-driven transport of salts/organic molecules.The pressure-driven transport of ethanol was performed to assess the tears and large nanopores (>60 nm), since the pressure-driven flow through a PITEM pore and a graphene pore scales as D L respectively, i.e. ∼4 nm defect in graphene will have similar resistance to transport as an open PITEM ∼200 nm pore.Hence, diffusion is sensitive to nanoscale defects in graphene <4 nm and diffusion-driven transports of KCl, NaCl, B12 and Lz were performed to evaluate the sub-nanometer defects as well as small nanopores (4 nm) [2, 5-7, 9, 13, 20, 21, 23, 24, 41, 42].
Since HSMG ® shows higher transport of hydrated ions and organic molecules than high-quality CVD graphene (transferred via IHL), this implies more sub-nanometer and nanometer scale defects existing in HSMG ® .However, the transfer yields are also different.The effective normalized flux computed using the equation ( ) normalized flux of each species normalized flux of ethanol normalized flux of ethanol 1 --facilitates a direct comparison and allows for accounting for any un-transferred areas that would be akin to bare PITEM (figure 3(G)) and can be effectively subtracted out.Compared with high-quality CVD graphene transferred via IHL method (KCl∼26%, NaCl∼25.7%,B12 ∼13.3%, and Lz ∼3.6%) [20] (figure 3(G)), HSMG ® shows similar/comparable effective normalized flux of KCl ∼28.6%, NaCl ∼23.4%, and B12 ∼15.4%, but much higher effective normalized flux of Lz ∼9.6%, indicating that HSMG ® lattice has more defects >4 nm.
The high quality of the HSMG ® graphene is attributed to the high-temperature of synthesis and large graphene domains formed which are highly mobile on the smooth/flat liquid Cu surface, [31,50], while defects >4 nm could potentially also result from re-crystallization of liquid Cu and associated mechanical stress on the graphene.The presence of intrinsic defects in HSMG ® graphene at similar levels as on solid Cu (optimized for membrane applications) indicates the formation of intrinsic defects within the graphene domains during bottom-up synthesis.The synthesis of graphene on liquid Cu also does not eliminate the problem of the wrinkles formation, which is mainly due to the different thermal expansion coefficients of graphene and Cu [51].Notably, solidification of the graphene growth substrate changes the morphology of the Cu surface from perfectly smooth in the liquid state (during graphene growth) to grainy in the solid state (before graphene transfer to PITEM) and the graphene adapts to these changes via elastic stretching.Taken together, the quality of HSMG ® is comparable with high-quality CVD graphene grown on solid Cu and optimized for membrane applications.

Conclusions
In this work, the quality of large-area graphene synthesized on liquid copper for membrane applications was investigated.Graphene (HSMG ® ) was synthesized using the metallurgical method and assessed via a simple and cost-effective technique based on pressure-driven and diffusion-driven transport of analytes.The as-synthesized HSMG ® showed a high-yield transfer onto PITEM with >93% coverage, but intrinsic sub-nanometer and nanometer scale defects still exist in HSMG ® and are no less than high quality CVD graphene synthesized on solid Cu and optimized for membrane applications.Even though HSMG ® graphene is synthesized at hightemperature and has large hexagonal domains, with potential for self-assembly via high mobility on the smooth/ flat liquid surface, intrinsic sub-nanometer and nanometer-scale defects are inevitably present as reflected in the evidenced transport of analytes (0.66-4 nm).Our study effectively benchmarks quality of graphene grown on liquid copper for membrane applications and also provides fundamental insights into the origin of intrinsic defects in large-area graphene synthesized via bottom-up processes for membrane applications.
(B) shows a high degree of defects along the line like feature, consistent with prior observations of defect clusters along graphene domain boundaries in HSMG ® [31].Figures 2(C) and (D) present atomic resolution images of the graphene lattice from two different regions.Notably, figure 2(C) shows several atoms missing (indicated by green arrows), while figure 2(D) appears to be pristine with no apparent vacancies being readily visible.Figure 2(E) also presents an HR-TEM image of graphene with visible multilayers and wrinkles, which is consistent with SEM images in figure 1(C).Upon transferring HSMG ® graphene onto SiO 2 /Si substrate (figure 2(F)), Raman spectra also confirm the presence of monolayer graphene (I 2D /I G ∼2, full width at half-maximum (FWHM) of 2D peak ∼33.5,I D /I G ∼0.09), multilayer graphene patches (I 2D /I G <1, FWHM of 2D peak ∼54.4,I D /I G ∼0.04) and wrinkles (I D /I G ∼0.36)

3 respectively [ 20 ,
24], i.e. ∼60 nm defect in graphene will have similar resistance to transport as an open PITEM ∼200 nm pore.Diffusiondriven flow on the other hand through a PITEM pore and a graphene pore scales as scales a D

Figure 2 .
Figure 2. (A) Transmission electron microscopy (HRTEM) image of HSMG ® , with magnified regions in the center, left, and right side of the line like feature marked with red, green, and purple boxes, respectively.(B) HR-TEM image of line like feature with defects (red arrows along the feature).(C) Atomic resolution image of the graphene lattice with defects (several atoms missing indicated by green arrows).(D) Atomic resolution image of pristine graphene lattice region with no readily apparent vacancies.(E) TEM image of monolayer HSMG ® graphene film with some multilayer islands/domains and wrinkles.(F) Raman spectra of HSMG ® graphene transferred onto SiO 2 /Si probed over three distinct regions i.e. monolayer, multilayer, and wrinkles (insets show corresponding optical images, scale bars are 10 μm).

Figure 3 .
Figure 3. (A) Sacrificial polymer scaffold method used to transfer HSMG ® graphene onto polyimide track etched (PITEM) support.(B) and (C) SEM images of HSMG ® transferred onto PITEM supports.Red arrows indicate tears.Wrinkles in the graphene are also visible.(D) Raman spectra of PITEM substrate (black curve) and HSMG ® transferred onto PITEM (PITEM+HSMG ® , red curve).(E) Schematic of customized side-bi-side diffusion cell.(F) Normalized flux of HSMG ® and high-quality CVD graphene grown on solid Cu [20] for pressure-driven ethanol flow and diffusion-driven flow of KCl, NaCl, B12, and Lz.The normalized flux was computed by the flow rate of PITEM+HSMG ® membrane divided by that of the bare PITEM substrate.(G) Effective normalized flux of HSMG ® graphene and high-quality CVD graphene [20] after accounting for coverage.