Two simple ways of making clean nitrogen-doped graphene for TEM studies

2.1 Sample preparation

Graphene monolayers on TEM grids were purchased from Graphenea, consisting of a gold TEM grid covered with a holey carbon film (Quantifoil) over which a monolayer CVD grown graphene is spanned. Cleaning of the samples was conducted either by downstream argon plasma in the Tergeo EM Plasma Cleaner or by heating to 450 °C at 0.1 mbar for 60 min.

2.2 Method 1: treatment of samples with RF plasma

Samples were treated with a Tergeo EM Plasma Cleaner from Pie Scientific, which is capable of generating a conductively coupled plasma at a frequency of 13.75 MHz. Doping was conducted with N₂ gas. In addition, it can also create a downstream plasma, where ions are filtered out and only reactive species reach the sample. RF power, nitrogen flow and exposure time can be controlled to change the plasma properties and thus the treatment of each sample. Figure 3 shows an image and a schematic for both working modes of the plasma cleaner. RF1 and RF2 were cleaned by downstream argon plasma for 4 min. RF3 was cleaned by heating to 450 °C at 0.1 mbar for 60 min. The plasma at the position of the sample was characterized using a Langmuir probe method, as detailed in the supplementary information.

Figure 3:

(A) Image of the plasma cleaner used in this work, with N₂ plasma running and inserted TEM holder. (B) View through the window of the plasma cleaner, showing sample position. (C) Schematic drawing of immersion plasma mode (left) that is used to dope the sample and downstream plasma mode (right) that is used to clean the sample.

2.3 Method 2: treatment of samples with DC gas glow discharge plasma

DC plasma was realized by constructing a simple vacuum chamber, containing a copper rod as positive electrode and a sample holder as negative electrode (grounded), both being electrically connected to the outside and spaced by a distance of 15 mm from each other. Via two leakvalves, the chamber is connected to a Pfeiffer HiCube 300 turbopump station and the N₂ gas line, allowing to adjust the desired pressure, which is monitored by a Pfeiffer CCR 363 pressure gauge (we note that a turbo pump is not strictly needed to obtain the pressures used in this study). A voltage of 750 V was applied to ignite the plasma inside the chamber, an exemplary photograph is shown in Figure 4A. Both sample holder and positive electrode were electrically connected to the outside and connected to a high voltage power supply (3B Scientific 5 kV power supply, adjustable from 0 to 5 kV, max. current 2 mA) power supply, as the scheme shows in Figure 4B. A relay driven by an Arduino microcontroller is used to switch the connection between the chamber and the power supply on for a selected amount of time, and thereby short pulses of discharges are obtained in a controlled way. The applied voltage was monitored via a multimeter. Most ions that strike the sample are created between the cathode and the negative glow region (see Figure 4B), thus only being accelerated by a part of the applied potential. All samples were cleaned by heating to 450 °C in a separate vacuum chamber at 0.1 mbar for 60 min prior to DC discharge plasma treatment.

Figure 4:

(A) Glow discharge inside DC plasma chamber at 6.5 mbar pressure and 750 V. (B) Sketch of observed glow discharge, showing electrical contacting and sample position.

2.4 TEM imaging and analysis

For HRTEM imaging an image-side corrected Jeol ARM200F TEM with CEOS aberration corrector and Gatan OneView camera was used at 80 kV acceleration voltage. Imaging and identification of the graphitic nitrogen substitution follows precisely the approach that was described previously [41], and is summarized here briefly: the contrast of the graphitic nitrogen is influenced by the rearrangement of charge density due to the perturbation of the electronic structure in the lattice by the nitrogen atom. As a result, it has a specific and rather unusual dependence on the defocus value, requiring optimized imaging conditions. Specifically, to enhance the contrast of the nitrogen defects, the defocus was set so that the graphene diffraction peaks were on the second ring of the contrast transfer function, i.e. the first zero crossing is just inside the six innnermost reflections. This corresponds to −18 nm defocus at 80 kV. After finding a suitable area to image, a series of roughly 30 images was taken to reach about 10⁴ counts per pixel. For further analysis following procedure was applied using ImageJ: First the dark current reference was subtracted from the images and then they were divided by the gain reference. Intensity variations by slightly uneven illumination were removed by dividing each image by its strongly Gauss blurred duplicate. After aligning the series to compensate any drift, a summation or averaging of the series was done, resulting in a final image which could be further analyzed. As the contrast difference between a nitrogen defect and the pristine graphene lattice is minute, a Fourier transform of the image was created, the graphene peaks removed and a subsequent inverse Fourier transformation yielded a new image without crystal lattice. This way any defects are more easily visible in the image. The statistical analysis of graphitic nitrogens and other defects concentration was performed as follows: By measuring the size of the contamination-free area in each image, the number of carbon atoms for which individual point defects should be visible was estimated. The number of nitrogen or other defects in the same area was determined by simply counting them. Atomic concentrations were calculated by dividing the number of nitrogen or other defects by the number of expected carbon atoms. Three separate regions were imaged per sample and setting for statistical analysis. In the supplementary information, examples of the contrast analysis are given for the monolayer (as in the earlier study) and in comparison to the doped bilayer, which was not reported before.

3.1 Radio-frequency plasma (method 1)

First attempts for nitrogen doping of graphene were done through the plasma cleaner. As the instrument is capable of a much higher plasma strength than needed, it was necessary to use it at very low power. Figure 5 shows an HRTEM image of one sample, where incorporated graphitic nitrogen as well as other defects are visible. Table 1 lists the parameters used for each sample. Using a higher power or longer exposure time lead to strong damaging or even complete removal of the graphene monolayer, thus these parameters were determined to be inside a suitable range for doping. Table 1 shows the determined graphitic nitrogen concentration as well as the concentration of other defects. As can be seen, nitrogen doping was achieved in all samples. Sample RF3 was subjected to an additional plasma cycle, in anticipation of a higher nitrogen concentration. However, the determined value seems to be even lower than before and strong damage of the sample was observed. Although the concentration of other defects is slightly higher than that of graphitic nitrogens in all samples, the latter can still be reliably found within the clean graphene sections of the sample.

Figure 5:

Monolayer graphene after doping with RF plasma, imaged by HRTEM (RF3). Arrows point at positions of graphitic nitrogen defects. On the left the complete image is shown, the same image with removed crystal lattice is shown on the right.

3.2 Direct current glow discharge plasma (method 2)

The plasma cleaner offers a quick and easy way to implant nitrogen inside graphene, but we achieved only relatively low doping concentrations. In search of a more suitable approach for doping, a DC glow discharge plasma chamber was built. The DC glow discharge is arguably the easiest way to create a plasma, probably even the first artificial one discovered as it has already been described by William Crookes in 1879 [92]. This low-effort and comparably cheap plasma chamber ionizes the incoming nitrogen gas by applying a high voltage between sample and an electrode, as seen in Figure 4. With this setup a series of doping experiments was conducted, using varying nitrogen gas pressures inside the chamber. Best results were achieved inside a pressure range from 3.5 to 8 mbar, higher pressures would either lead to strong contamination/destruction of the graphene or stable graphene with no incorporated nitrogen. Furthermore, it was found that short pulses of approximately 40 ms yield satisfactory and controllable results. Samples were also treated several times, first with 2 pulses, second time with 4 pulses and finally with additional 4 pulses. All parameters are shown in Table 2. Samples were observed via HRTEM and graphitic nitrogen could be found after doping, Figure 6 shows an example of such an image. In Figure 7 the measured concentration of graphitic nitrogen as well as other defects are visualized. As seen in Figure 7A, higher concentrations of graphitic nitrogen were achieved compared to the plasma cleaner method. Furthermore, with repeated treatment the graphitic nitrogen concentration increased, but at some point the graphene took up too much damage or got too contaminated to conduct further analysis. Only with 3.5 mbar we were able to still observe non-contaminated and free areas after 10 pulses. It seems that the graphitic nitrogen concentration reached a threshold at around 0.15 at%, before the destruction of bigger structures set in. Although the nitrogen content increased with further plasma treatment, it was not proportional to the number of applied pulses and the rise in concentration differed between the experiments. Here again, other defects were often in close vicinity to the graphitic nitrogen defects, as can be seen in Figure 6. For the 3.5 and 8 mbar series the number of defects outpaced the one of graphitic nitrogen or was similar, but the 5 and 6.5 mbar series showed higher amounts of graphitic nitrogen than other defects.

Figure 6:

Monolayer graphene after doping inside DC plasma with 10 pulses at 3.5 mbar, imaged by HRTEM. Arrows point at positions of graphitic nitrogen defects. On the left the complete image is shown, the same image with removed crystal lattice is shown on the right.

Figure 7:

Results taken from Table 2. (A) Doping level of graphitic nitrogen as well as (B) defect level were determined from images.

3.3 Bilayer graphene doping

During imaging an area containing monolayer as well as bilayer graphene was found, as shown in Figure 8. Interestingly, graphitic nitrogen defects were detectable in both regions, as is shown in Figure S9.

Figure 8:

Bilayer graphene (most of the area) with some monolayer regions (holes formed in one of the two layers) after doping inside DC plasma with 2 pulses at 5 mbar, imaged by HRTEM. Arrows point at positions of graphitic nitrogen defects. On the left the original image is shown, the same image with removed crystal lattice is shown on the right.

3.4 Adsorption and desorption on a nitrogen substitution

During observation of a free area containing nitrogen defects, a different contrast was noticed temporarily for one of the graphitic nitrogens. Closer inspection revealed that only for a few frames of the image series an enhanced darker contrast was observable, whereas before and after the usual contrast value for graphitic nitrogen could be measured. Figure 9 shows three images from the series, where this change in contrast is visible. We interpret this as the temporary adsorption of heavier atom or small cluster at the nitrogen site, as discussed further below.

Figure 9:

Observation of contrast change at position of a graphitic nitrogen defect, which is marked by the yellow box. Images show the nitrogen before (A), during contrast change (B) and afterwards (C). A slight Gauss blurring was applied to emphasize the change in contrast.

4.1 Doping inside radio-frequency plasma (method 1)

Observing the results, as shown in Table 1, it is apparent that the concentration of graphitic nitrogen tends to stay below the concentration of other defects. This circumstance could be explained by the fact, that nitrogen defects are mostly created in a narrow range of ion energies [74], whereas other types of defects are more probable, e.g. vacancies at higher ion energies. Furthermore, it would be quite conceivable to conclude that some defects could be of pyrrolic or pyridinic nature. Unfortunately, these types can hardly be distinguished from simple vacancy defects inside the graphene lattice using HRTEM. It appears that a gas flow of 25 sccm introduces the highest number of defects (graphitic nitrogen and others), so that this setting could be just inside the right range for efficient doping of graphene. On one hand, a gas flow of 1 sccm might create too little ions due to the low gas pressure. On the other hand, 50 sccm might produce a too high pressure, thus hindering ions to accelerate and reach higher energies. Both, a too low ion concentration and energy, might lead to inefficient doping. However, from the characterization of the RF plasma (see Table S1) 1 sccm shows highest concentration in ions, contradicting this explanation. It should be kept in mind that the sample RF3 (25 sccm) was cleaned by heating in vacuum instead of downstream argon plasma, like RF1 and RF2 (1 and 50 sccm). Cleaning of the sample might have a significant impact on the implantation, as contamination that is already present before doping would shield the graphene from incoming nitrogen ions. Also, after doping contamination might preferably attach to more doped or defective regions, thereby hiding a portion of the dopant atoms in a selective way that affects the statistics.

A related study with freely-suspended graphene was conducted by Iyer et al. [70], where they used N₂ RF plasma and analyzed the samples by X-ray and Raman spectroscopic methods. They were able to confirm different types of nitrogen species inside the sample, but no concentrations were determined. Compared to the results of this study, higher power (100 W) and longer exposure times (up to 10 min) were used. These values differ by orders of magnitude from the range of parameters determined here, outside of the range where we obtain stable structures. However, the power of the plasma source might not be directly comparable.

4.2 Doping inside direct current glow discharge plasma (method 2)

Higher graphitic nitrogen concentrations could be reached within the series of DC plasma doping experiments as compared to the RF plasma. However, there still seems to be a limit to the doping concentration that can be reached without destroying the sample or creating excessive other types of damage to the graphene. Similar to the RF plasma, it seems that doping is most efficiently achieved at medium pressures (5 and 6.5 mbar), at least in regards of the ratio between graphitic nitrogen and other defects. Again, the pressure inside the chamber might play an important role and can have an impact on the ion energy distribution. At too low pressure, the ion’s mean free path is big enough to accumulate more energy, making defects like single or double vacancies more probable. However, too high pressures should lead to lower ion energies and fewer number of defects of any kind, but this we can not observe in these results. It should be noted that only a small part of the possible parameter space could be explored in the scope of this study. For example, variation of the electrode geometry and applied voltage might lead to varying fractions of ions, radicals and neutrals bombarding the sample [93], [94]. Another relevant factor is the amount of hydrocarbons that is present inside the chamber during doping. By collisions with ions and neutrals these hydrocarbons might be deposited on the graphene as contamination and even turn at some point to amorphous carbon due to the energy transfer, similar to what happens under an electron beam. As noted before, contamination might shield the graphene from incoming ions, so that bakeout of the graphene in the same vacuum prior to doping would be desirable. However, examining the HRTEM images (Figures 5 and 6) reveals that many defects are in close vicinity to contamination. This circumstance might hint towards a deposition of contaminates after doping, possibly attracted by various defects.

A related study applying DC nitrogen plasma to multilayer graphene can be found in the literature, reaching doping concentrations of up to 3.12 at% [95]. In this case 350 V, 4.6 mbar and 20–40 s exposure time were used, which exceeds our time of 40 ms by magnitudes. However, the amount of graphitic nitrogen was not determined, and atomic level analysis (which would show the defectiveness of graphene besides the implanted nitrogen) is absent. In this case, 4–5 layer graphene on a substrate was used, which can be expected to withstand much harsher treatment conditions and is not easily comparable to freely-suspended graphene.

As discussed previously, HRTEM has some drawbacks when it comes to give a precise quantitative estimation for the nitrogen concentration of the samples. In particular, contamination accumulated during the transfer from DC chamber to the microscope might hide a portion of incorporated nitrogen atoms. However, in the case of doping graphene by ion implantation, contamination is also an often disregarded trap for other analysis methods. It is known that graphene tends to accumulate hydrocarbons from the air quite quickly [96]. Dedicated investigations have shown that the hydrocarbons themselves incorporate nitrogen atoms in a wide range of configurations, thus distorting the results obtained by commonly used methods as Raman spectroscopy and XPS [97]. Although often plagued by such contamination, HRTEM offers the possibility to truly confirm the integration of the nitrogen atom inside graphene’s structure. The samples obtained in this study display regions that are free from contamination, where nitrogen can be found surrounded by a clean lattice, for at least a significant fraction of the area. In spite of the doping, the overall appearance of clean regions interspersed with contamination networks is comparable to that of pristine graphene.

4.3 Doping and detection of nitrogen in bilayer graphene

As seen in Figure 8, graphitic nitrogen can be detected even in bilayer graphene. The defects show the same contrast in both mono- and bilayer (see Figure S9). However, we can not determine whether the nitrogen is preferentially implanted into one of the layers, as might be expected for the layer that points towards the other electrode in the DC discharge. It should be kept in mind that other doping mechanisms, involving chemical reactions with low energy ions, are theoretically conceivable [98]. A study of this aspect was conducted by Cress et al. on bilayer graphene [99]. They showed how low ion energies (45 eV) only modify the top layer, whereas higher energies (200 eV) modify both layers. As they used Raman spectroscopy to identify which layer is affected, there is no direct proof for the type of defect created.

4.4 Adsorption to graphitic nitrogen

As shown in Figure 9, a contrast change appears in the middle of the image series at the site of a graphitic nitrogen. Before as well as after this event the nitrogen defect was measurable at the exactly same position, thus it can be assumed that it is also present during the contrast change. Our explanation for this temporary increase in contrast is the presence of additional atoms, either a single adatom or several ones, possibly in the form of a molecule. From the image series it can be confirmed that the adsorption event goes on for roughly 60 s.

There are several theoretical studies concerned with the interaction and adsorption directly or close-by to graphitic nitrogen. Some state that the adsorption of various gases [100] and small hydrocarbons [101] is not enhanced by graphitic nitrogen, whereas many transition metals [101] and water [102] are supposed to bind more likely. Another theoretical study found that a carbon adatom, a likely species to encounter on graphene, would rather bind to the carbon neighbors of graphitic nitrogen than to itself [103]. This is explained by the charge redistribution introduced by the nitrogen, pushing more electron density towards the neighboring carbons and thus giving them a stronger electron donor character. Such change in electron density was confirmed experimentally by EELS [90], [91]. Scanning tunneling microscopy and spectroscopy experiments indicate that the interaction of certain organic molecules with graphene are manipulated by the presence of graphitic nitrogens [43], [44]. However, it is assumed that this results from the interaction with the changed electronic structure, rather than a chemical interaction with the nitrogen. Another study using STEM and EELS found no binding of various metal elements to graphitic nitrogen, but preferably to pyridinic nitrogen [42]. From these findings it could be assumed that graphitic nitrogen displays a weak binding character that is of electronic rather than of chemical nature.

The showcased images allow to observe a direct adsorption event mediated by graphitic nitrogen. Contrast change due to the adsorbate is visible directly at the nitrogen atom and its closer surroundings, making it hard to pin down which atoms participate during adsorption. Moreover, it is not clear if the interaction is of electronic or chemical character, but considering previous works, former might be more probable. This is also in line with the observation that no bonds are broken after the adsorbate has desorbed, leaving the graphitic nitrogen in the same configuration as before. However, our results indicate that graphitic nitrogen could have a higher affinity to adsorbates than previously expected.

4.5 N-doped graphene as substrate for TEM

As showcased before, graphitic nitrogen is able to adsorb atomic or molecular species for a time period which is sufficient for imaging. Of course the long-term stability of graphitic nitrogen under the electron beam is an important question. Indeed, the stability of the graphitic nitrogen substitution in graphene is remarkable, it is much more stable than most other defects under the 80 kV electron beam. Susi et al. showed that electron impact can displace a neighboring carbon atom, leading to the formation of pyridinic nitrogen [104]. However, this event is still quite rare at 80 kV and becomes more important at higher acceleration voltages. Another study by Warner et al. conducted at 60 kV concluded that graphitic nitrogen remains stable over a long time, even at temperatures of 500 °C [91]. This corresponds roughly to our findings, as many graphitic nitrogen defects would stay stable during recording of an image series, which could take more than 15 min. This time span should be enough for many applications, thus making N-doped graphene a suitable substrate for TEM. Another excellent characteristic of graphitic nitrogen is its very low contrast change compared to the pristine lattice, keeping the disturbance inside the image to a minimum. Other options for adsorption sites would be vacancy defects, but these bring stronger contrast change with them. Oxidative doping of graphene is another possibility to enhance adsorption [105], but oxygen defects often include some form of vacancy. There are records of triple bond oxygen in graphene [57], but it brings in more mass and charge compared with graphitic nitrogen, and this defect is also much less stable under the beam. Furthermore, oxidative doping usually demands high temperatures, which are often incompatible with most TEM holders. The plasma doping methods presented here can be applied directly at the TEM holder, keeping sample handling to a minimum. In general, both methods offer a simple, quick and controllable way to change the adsorption properties of graphene without greater disruption of its periodic structure inside the TEM image.