Oxide-mediated nitrogen doping of CVD graphene and their subsequent thermal stability

Heteroatom doping of graphene is a promising approach for tailoring its chemical and electronic properties—a prerequisite for many applications such as sensing, catalysis, and energy storage. Doping chemical vapour deposition (CVD) graphene with nitrogen during growth (in situ doping) is a common strategy, but it produces a distribution of inequivalent dopant sites and requires substantial modifications to the CVD growth process. In this study, we demonstrate a novel and simple oxide-mediated approach to introduce nitrogen dopants into pre-existing CVD graphene (ex situ doping) which achieves comparable doping densities to in situ doping methodologies. Furthermore, we demonstrate that thermal annealing of N-doped graphene can selectively remove pyridinic, retaining graphitic and pyrrolic nitrogen dopants, offering an attractive route to further modify graphene functionality. The methodologies we present are simple and scalable to precisely tailor graphene properties without the need to alter CVD growth protocols.


Introduction
Controlled doping of graphene is challenging due to strong sp 2 hybridised carbon atoms bound within its hexagonal lattice. However, imperfection sites within the carbon lattice, such as point defects and edges, can act as access points for these chemical bonds [1]. Substitutional doping (i.e. replacing the carbon atoms with that of another element) is an efficient route to tailoring graphene's structural, physiochemical, electrical, and optical properties. Due to the similarity between the atomic radii of carbon and nitrogen, it is an ideal candidate for doping, expected to introduce n-type carriers into the carbon system [2]. N-doping also induces a bandgap in graphene, a critical requirement for electronic and opto-electronic applications [3]. The addition of N atom within the lattice have also been shown to influence the spin density and charge distribution of neighbouring C atoms, creating 'activation regions' [4]. Hence, the properties of nitrogen-doped graphene (NGr) greatly enhances graphene's applications within the semiconductor, energy storage, ultracapacitor and sensor industries [4,5].
The different doping configurations (pyrrolic, graphitic, and pyridinic) differ electronically and structurally and therefore there is interest in selective doping of graphene for varying applications. Pyridinic doped graphene is being investigated for lithium-ion battery applications as empirical studies suggest that pyridinic dopants yield the greatest lithium storage capacity [6]. Pyridinic nitrogen has also shown to play an important role in the oxygen reduction reaction in N-doped graphene quantum dots [7]. Pyrrolicdoped graphene quantum dots have demonstrated tunable colour emission for applications in bioimaging whilst graphitic nitrogen doping has been shown to significantly enhance blue light emission and electrical conductivity of graphene [8,9].
In order to optimise these properties it is vital that the position of the nitrogen in the graphitic network (i.e. through doping, intercalated gaseous nitrogen, nitrogen located in nonplanar sp 3 sites, chemisorbed) is understood and can be controlled [2].
Several methods have been reported to synthesise NGr, these include; chemical vapour deposition (CVD) [10,11], thermal annealing of graphene oxide (GO) with ammonia/ urea [12,13] and nitrogen plasma treatment [14,15] etc. These can largely be split into two categories, in situ methods, where doping occurs simultaneously to graphene growth and ex situ methods which involve altering the graphene itself.
As previously shown by Zahra et al [16,17], graphene nanobubbles can be formed under CVD graphene. This is achieved by a simple substrate oxidation followed by NH 3 annealing. Consequently, the formation of N-dopants was seen in small atomic concentrations (at. %), ca. 3%. Commercially bought pristine graphene (PrGr) was used and oxidation of samples was achieved by heating on a hot plate in air at 473 K for 20 min. These results are reproduced in this paper with lab-grown CVD graphene, where the NH 3 treatment is then performed in the same furnace. Here an investigation of N-dopant species through the use of x-ray photoelectron spectroscopy (XPS), Raman spectroscopy, and atomic force microscopy (AFM) is undertaken with a focus on maximising and controlling the N-dopant species.

Oxidation dependence
An illustration of the sample preparation can be seen in figure 1. CVD of a pristine graphene sample (no N precursor), was grown on a 6 × 2 cm Cu substrate in an in-house CVD chamber. The pristine graphene was then cut into 6 equal pieces and oxidised in air. Each piece was oxidised for a different duration ranging from 0 to 60 min. Roman numerals have been used to dictate the period of oxidation where i, ii, iii, iv, v, and vi represent 0 min, 1 min, 5 min, 10 min, 30 min and 60 min, respectively. Each piece was further cut in half, with one complete set of the oxidation range being further annealed in NH 3 to form NGr. Throughout this paper this will be referred to as the ex situ method. An image of the NGr, pristine graphene, and a control set of Cu can be seen in figure 1. As the oxidation duration increases the Cu changes in colour from orange to red to yellow. Qualitative evidence of the graphene layer can already be seen by observing the minor colour change in the pristine graphene sample in comparison, as the oxidation duration increases. This is due to the protective nature of graphene [18][19][20][21]. However by 60 min oxidation, the sample is seen to have become more red, as the oxygen has begun to etch away the graphene allowing more direct access to the Cu substrate [22,23]. The variation in Cu oxidation can also be seen in the XPS spectra of the Cu LMM (electronic supporting information, ESI figure S1). Upon annealing in NH 3 , the samples all return to their original colour. XPS spectra of the N 1s for the NGr samples are shown in figure 2. The signals from all 6 samples can be deconvoluted into multiple components. These have been split into three regions. The first region (green) consists of peaks associated with N 2 components at ca. 404.7 eV (N 2 graphene nanobubbles) [16] and ca. 403.4 eV (N 2 intercalation) [16,[24][25][26][27], a consequence of the Cu 2 O reduction reaction upon annealing in NH 3 . The second region (purple) is attributed to N-doping species. Recent research has shown that, though often fitted separately within the literature [4,5], N 1s peaks associated with graphitic and pyrrolic dopants lie too close to one another to be easily distinguishable [28]. Hence, in this article they have been fitted with a single peak ca. 400.2 eV [28], which has a wider full width half maxima (FWHM) than that seen for the pyridinic peak at ca. 398.4 eV [4]. Finally, the last region (orange) also consists of two peaks. These are due to N bonding to Cu, (surface amide, ca. 397.4 eV) [29] and (low levels of atomic N bonded with Cu, ca. 396.6 eV) [29][30][31]. There are no N 2 peaks observed in the samples which underwent 0 or 1 min of oxidation (figures 2(i) and (ii)). This is consistent with previous studies where it was suggested that the underlying Cu substrate was required to be oxidised to initiate the reduction of NH 3 to N 2 [12,16]. We see the greatest contribution from the N 2 components after 10 min of oxidation. At oxidation times greater than 10 min, there is less relative contribution from the N 2 peaks and greater relative contribution from the N-doping and N-Cu components. The highest concentration of N-doping species ca. 3.4 at. %) is seen after 60 min of oxidation. Figure 3 shows the Raman spectra of the graphene film before and after substrate oxidation. Measurements were taken across the surface of each sample, averaged and then background subtracted to remove the photoluminescent background caused by the Cu substrate.
In both spectra, peaks at ∼1590 cm −1 and ∼2710 cm −1 , assigned as the G band and the 2D band respectively, can be observed and are characteristic of graphene. The intensity ratio between the G and 2D ( / I I 2D G ) bands is indicative of the number of graphene layers formed [32]. Analysis of the spectra for pristine graphene with no oxidation (PrGr)(i)), indicates the sample is monolayer, as it has an average 2D:G intensity ratio, / I I 2D G = 1.72 and FWHM 2D = 33.2 cm −1 . A third peak assigned as the D peak can be observed in both spectra at ∼1350 cm −1 and is associated with disordered/defective carbon. After 60 min of oxidation (PrGr(vi)), although the graphene layer remains present, a  significant increase in the D peak is observed, whilst the / I I 2D G decreases to 1.02, and the FWHM 2D increases to 46.6 cm −1 . This indicates that the prolonged oxidation causes an increase in defect concentration and is consistent with what is observed in figure 1 [33].
AFM measurements show the topographical features of the sample (figure 4). Bright features are seen in both samples due to contamination in the Cu substrate prior to growth. In the pristine graphene sample, PrGr(i), wrinkles can be seen propagating around these features as the graphene contracts during cooling. In regions void of these contaminates the surface is much smoother and indicates large graphene coverage, this is most prominent in the adhesion image. The NGr(iv) sample is shown as it contained the largest at.% of N 2 components, seen in the XPS ( figure 2). This can be seen in the form of graphene nanobubbles, where the negative contrast in the stiffness indicates they are less stiff than the surrounding Cu. There also appear to be regions of bare Cu, further suggesting that that oxidation causes some etching of the graphene layer.
We hypothesise that two processes occur here. The first is that the oxidation process causes partial etching of the graphene layer, producing more defect sites, as evidenced by the larger D peak in the Raman spectrum.
Subsequently, during the NH 3 treatment, the reduction of the underlying Cu 2 O then produces atomic and molecular nitrogen radicals such as N, NH, and NH 2 in close proximity to the graphene layer which bond more easily and readily to the defective graphene sites and edge sites, forming NGr [3,34]. Substitution of C atoms in graphene by NH radicals has also previously been demonstrated with ammonia exposure [15]. This also explains why the N 2 component peaks begin to decrease after 30 min of oxidation, as there is less graphene coverage and the defects act as egress sites for the N 2 to escape. A high level of defects in the graphene may ehance transport of NH 3 to the reactive Cu 2 O/Cu surface (as the graphene is a less effective barrier), explaining the higher levels of doping for longer oxidation times.

Thermal stability of nitrogen dopants
To investigate the effects of temperature on the stability of nitrogen species in NGr, two new samples were prepared using the ex situ (S ex situ ) and in situ (S in situ ) methods. The synthesis of both required CVD and used Cu foil as a substrate. The S ex situ was prepared as described in section 2.1, CVD pristine graphene was oxidised in air for 60 min followed by NH 3 annealing [16]. For the S in situ , NGr was prepared by CVD with both carbon (CH 4 ) and nitrogen (NH 3 ) precursors.
These precursors dissociate on the surface of the catalyst (Cu) at high temperatures (∼900°C), forming atoms and radicals. Nitrogen-doped graphene then forms through the precipitation of the carbon and nitrogen atoms [35,36]. Further details on the growth procedures for both samples are provided in section 3.
High resolution XPS spectra of the N 1s are shown in figure 5. Samples were initially annealed to 500 K in order to remove adventitious C. The signals from both samples can be deconvoluted into multiple components, 6 seen in the S ex situ and 3 in the S in situ , where both N 2 components, and the surface nitrogen bonded to Cu are absent in the in situ sample. The mechanism behind the formation of the N 2 components observed in S ex situ has previously been described in literature and shows that NH 3 intercalates between an already formed graphene layer and the copper oxide substrate, and then decomposes [12,16].
At 500 K, the highest combined at. % of the pyridinic, graphitic and pyrrolic peaks (N-dopants), relative to C 1s sp 2 (figure 6) is seen in S ex situ at ca 6.9 at. %; whereas in S in situ  only ca. 3.5 at. % is attributed to N-dopants (detailed at. % of all N species at each temperature is provided in the ESI, tables S1 and S2). Though different initial at. % of each N species were seen in both samples, a similar trend develops upon vacuum heating. In S ex situ both N 2 peaks are absent above 800 K. Similarly in both S ex situ and S in situ both the N-Cu peaks are absent above 700 K. This indicates that these are weakly bonded species. In this paper we shall be focusing on the N-dopant peaks as seen in the second (purple) region, where the most prominent change in the dopants is in the pyridinic nitrogen. Figure 7 reveals that as temperature increases the pyridinic component begins to diminish, and at 1000 K it has decreased by 79% and 85% from its initial value in S ex-situ and S in-situ , respectively. Although the graphitic + pyrrolic peak also decreases with temperature, this is at a much slower rate with a 44% and 32% drop from initial at. % upon 1000 K annealing in S ex situ and S in situ , respectively. This result indicates a difference in thermal stability of the N-doping species, with graphitic + pyrrolic having a significantly higher stability than the pyridinic. This relationship has previously been observed in the literature in carbon nitride films [37][38][39], GO [40,41], and most recently in amorphous C [42]. However, research into thermal treatments of N-doped graphene is very limited. To the best of our knowledge only two articles exist which discuss the relationship of N-dopant species to temperature annealing. Kumar et al [43] conducted a study on the thermal stability of N-species in graphene nanoflakes doped by ion bombardment. Similarly, they found a large decrease in pyridinic above 673 K and only a minor decrease in graphitic. This is in agreement with the results reported here. The other article, by Tian et al [44] is a theoretical study which uses 69 model N-graphene fragments. They find that pyridinic is significantly more thermally stable than graphitic nitrogen, which contradicts both the literature and results shown here. Our results suggest that on CVD graphene on copper, graphitic/pyrrolic dopants have significantly higher stability than pyridinic, however we point out that the relative stability may be significantly affected by other factors, such as interactions with the substrate and the nature of the graphene (the literature examples cited use flakes of graphene rather than a extended CVD monolayer).

Experimental methodology
Cu foil, thickness 125 μm and 99.995+ % purity (Advent), were used for both samples. An in-house built low pressure CVD chamber consisting of a Carbolite hot walled tube furnace, an Edwards rotary pump and three feedthroughs (H 2 , CH 4 and NH 3 ) controlled by Omega mass flow controller was used. The tube furnace has a base pressure of ca. 6´10 −5 mbar. The Cu substrate was first rinsed in acetone and dried. It was then annealed in the tube furnace under a H 2 (CK gas hydrogen (c.99.999% pure)) atmosphere 20 sccm (where sccm is a standard cubic centimetre per minute) at a 303 K min −1 ramp rate. This was maintained for 20 min to improve surface characteristics and remove adventitious contaminants. For the N-doped S in situ a temperature of 1173 K was used. After the annealing period, CH 4 (CK gas methane N5.0 (c.99.999% pure)) and NH 3 (CK gas ammonia N3.8 (c.99.98%)), at 5 sccm and 0.5 sccm, respectively were introduced for a growth period of 40 min, before removal.  The pristine S ex situ was annealed to 1273 K and only CH 4 (5 sccm) was introduced for the growth period (40 min), before removal. H 2 (20 sccm) was maintained in both samples throughout and removed at 773 K. The samples were then allowed to cool to <323 K before being extracted. Ex situ doping of S ex situ was achieved by oxidising the pristine graphene sample in air on a hot plate (473 K) for 0 min, 1 min, 5 min, 10 min, 30 min or 60 min, as indicated by the roman numerals i, ii, iii, iv, v, and vi respectively. The oxidised samples were then annealed in the tube furnace in an NH 3 atmosphere (1 sccm, with the exhaust valve throttled to maintain 1 mbar pressure) at 1023 K for 60 min. NH 3 was removed at 673 K and S ex situ was allowed to cool to <323 K before being extracted. The workflows for both the in-situ and ex-situ nitrogen doping protocols are represented diagrammatically in figure 8.
In section 2.1 XPS measurements were performed using a SPECS NAP-XPS instrument. The system consists of two ultra-high vacuum (UHV) chambers (preparation and analysis), and a PHOIBOS 150 NAP Analyser. All spectra were taken in the UHV analysis chamber. Photoelectrons were excited using monochromatic Al K α radiation (1486.6 eV). A large spot size of ca. 1 mm was used when scanning in standard UHV conditions. For survey/detailed scans E pass of 60/20 eV were used.
In section 2.2, XPS measurements were performed at MatLine beamline of the ASTRID2 Synchrotron in Aarhus, Denmark. An excitation energy of 500 eV was used for all data presented. Heating was applied using electron bombardment by an e-beam heater and thermal radiation from the filament at low temperatures, to a maximum temperature of 1000 K. Each sample was heated in steps of 100 K where the sample was heated to the desired temperature, maintained at temperature for 20 min and then allowed to cool before scanning. The sample temperate was measured by a thermocouple attached to the sample plate.
Data processing for all XPS spectra were carried out using CasaXPS processing software. All binding energies, E , B were charge calibrated to Cu. Data from the SPECS system were calibrated to the Cu 2p 3/2 peak at E B = 932.6 eV [45], whereas data from MatLine were calibrated to the Cu 3p 3/2 , E B = 75.1 eV [46,47]. Regions and components were fitted to each peak with the appropriate relative sensitivity factors. Generally, a Shirley background type was used. In cases where the intensity was flat or higher at lower E B (such as the N 1s spectra from MatLine) a linear background type was used. A symmetric Guassian-Lorentzian product (pseudo-Voigt) line shape was used for the N 1s. The Cu 3p, Cu 2p and C 1s sp 2 , where fitted with an asymmetric Lorentzian and hybrid Doniach Sunjic/Gaussian-Lorentzian (product) [48,49] line shape.
Raman spectra were taken using a Renishaw inVia Raman microscope with a 532 nm laser. The beam was focused using a 100´objective lens, laser power of 3.32 mW and a grating of 1800 l mm −1 . The peak fitting and photoluminescent background subtraction were done through a Python programme (further details in ESI).
AFM measurements were carried out at room temperature using a Bruker Multimode8 SPM, operated in Quantitative NanoMechanical mode. A silicon tip on nitride Bruker lever was used (model: SCANASYST-AIR). A variety of parameters were captured using NanoScope 8.2, which were then analysed using Gwyddion.

Conclusions
We have demonstrated that pre-oxidation is a powerful tool to control dopant concentration and species in CVD graphene. The results show that N species present in ex situ methods are dependent on the duration of the oxidation of the PrGr sample, and that this method is successful with lab-grown CVD graphene. In section 2.1 it is shown that the greatest contribution from the N 2 components (ca. 7.7 at. %) is seen after 10 min of oxidation and highest N-dopants (ca. 3.4 at. %) occurs after 60 min oxidation. Raman and AFM indicate that as the oxidation duration increases, the pristine graphene is etched and becomes more defective, creating reactive sites for the dopants to bond to. Though outside of the scope of this paper, further fine-tuning of the oxidation parameters (temperature, duration, pressure) could be used to vary the NGr species and at. %. This offers a simple and effective alternative to the in situ CVD method commonly used to produce NGr and allows users to maintain the use of pre-established pristine graphene recipes and setups.
In section 2.2, a new set of samples were created; an ex situ sample (N-dopants, ca. 6.9 at. %) which followed the same procedure as NGr(vi) and an in situ NGr sample (Ndopants, ca. 3.5 at. %) that was grown using CVD with a N precursor. We have shown that N species in both samples follow the same trend upon annealing and that their at. % can be altered using this method. As temperature increases the pyridinic component begins to diminish, and decreases by 79% and 85% from its initial value in S ex situ and S in situ , respectively, at 1000 K. While the graphitic + pyrrolic component decreases at a much slower rate with both samples having less than a 44% drop from initial at. % after 1000 K. These results indicate that graphitic + pyrrolic nitrogen has a higher thermal stability than that of pyridinic nitrogen. This shows that thermal annealing of NGr can provide a simple and scalable methodology to tailor the surface chemistry and electronic structure of NGr to optimise it for a particular application.