High-Temperature Carbon Deposition on Oxide Surfaces by CO Disproportionation.

Carbon deposition due to the inverse Boudouard reaction (2CO → CO2 + C) has been studied on yttria-stabilized zirconia (YSZ), Y2O3, and ZrO2 in comparison to CH4 by a variety of different chemical, structural, and spectroscopic characterization techniques, including electrochemical impedance spectroscopy (EIS), Fourier-transform infrared (FT-IR) spectroscopy and imaging, Raman spectroscopy, and electron microscopy. Consentaneously, all experimental methods prove the formation of a more or less conducting carbon layer (depending on the used oxide) of disordered nanocrystalline graphite covering the individual grains of the respective pure oxides after treatment in flowing CO at temperatures above ∼1023 K. All measurements show that during carbon deposition, a more or less substantial surface reduction of the oxides takes place. These results, therefore, reveal that the studied pure oxides can act as efficient nonmetallic substrates for CO-induced growth of highly distorted graphitic carbon with possible important technological implications especially with respect to treatment in pure CO or CO-rich syngas mixtures. Compared to CH4, more carbon is generally deposited in CO under otherwise similar experimental conditions. Although Raman and electron microscopy measurements do not show substantial differences in the structure of the deposited carbon layers, in particular, electrochemical impedance measurements reveal major differences in the dynamic growth process of the carbon layer, eventually leading to less percolated islands and suppressed metallic conductivity in comparison to CH4-induced graphite.

Introduction

Nowadays there is a huge need to develop new and more efficient energy production technologies and at the same time limit the emission of greenhouse gases. Solid oxide fuel cells (SOFCs) are considered as one of the most promising energy converters producing electricity by electrochemical combination of a fuel with an oxidant. These devices operate at high temperatures (1073–1273 K), which expedites electrode reactions and allows the use of a wide variety of fuels. In general, an SOFC consists of YSZ, which functions as an oxide–ion conductor with Ni–YSZ cermet anodes, where the hydrogen oxidation occurs, and Sr-doped lanthanum manganite (LSM) cathodes, where the oxygen reduction takes place.[1−4] The high operating temperatures thereby provide SOFCs with excellent fuel flexibility. Hydrogen and various hydrocarbons (such as CH4, etc.) are typically reformed either internally or in an external reformer prior to the fuel cell. The internal reforming operation results in a higher efficiency, a simplified SOFC system, and hence lower costs of the whole power generation system due to the exclusion of the prereformer.[1−6] However, this study is not focusing on the characterization of an internal reforming gas such as CH4; rather its aim is to shed some light on the effect of pure CO, which is in combination with the H2 part of synthesis gas and used for external reforming.[4−8]

A major setback upon using both reformate gas and hydrocarbon fuels is the formation of carbon deposits on the Ni–YSZ electrode especially at high operating temperatures.[3,7−13] This can deactivate the Ni catalyst and will cause rapid cell degradation. The carbon formation, e.g., specifically by using CH4 or CO, can proceed via two possible pathways: either methane dissociation (1) or following the inverse Boudouard (2)[14−16] reaction

There are several approaches to solve this major problem: lowering the operating temperatures, which results in a decrease in cell efficiency, development of new catalysts to replace Ni in Ni–YSZ, modifying the anode with other oxides, or choosing a pathway via syngas (CO and H2) using an external reformer and admitting this mixture to the anode instead of the hydrocarbons directly. It has also been reported that the quantity and quality of carbon species of deposited carbon during CH4 reforming is intensely affected by the operating temperature and the methane/steam ratio.[2,3,13] Even though there have been many ideas on how to solve this problem, none of them has been fully satisfactory. Currently, the pathway via syngas has been one of the most promising solutions. Thus far, carbon deposition and growth have then been basically ascribed to the metal phase, whereas the contribution of the oxide has been mainly neglected. However, recent studies on a variety of oxides (ZrO2, Y2O3, YSZ, among others) and using different carbon fuels (ethylene or methane) showed that under special conditions—usually at sufficiently high temperatures beyond 1000 K—also oxides can act as effective structural steering mediators for different carbon deposits.[17−19] As on metals, the structure, amount, and properties of the so-formed carbon deposits may depend on the fuel used and thus originate from different formation pathways.

To highlight the effects of usage of different carbon precursor materials, especially on the basis of previous studies using methane as carbon source, the aim of this work is to provide data on the interaction of the reformate component carbon monoxide (CO) with three oxides, namely, Y2O3, YSZ, and ZrO2. These oxides have recently been found to be interesting for forming different carbon architectures following treatment in methane.[17] This will not only directly reveal the already anticipated different formation mechanisms of the carbon deposits but also clarify to which extent the transformation of a carbon fuel into CO-containing syngas mixtures will affect the oxide component of a realistic SOFC material. The discussion is hereby restricted to the carbon growth in pure CO as a benchmark system. To finally achieve this task, an approach using a multitude of complementary structural and spectroscopic characterization tools has been used, including FT-IR, Raman, and EIS as well as dedicated electron microscopy work. We note already at this stage that the combination of FT-IR and impedance spectroscopy is a particularly rewarding combined characterization tool, since it allows one to exactly correlate the changes in carbon-containing adsorbate structures with associated changes in surface conductivity. The latter has in a previous study been determined to be one of the most crucial parameters for characterization of the deposited carbon layers.[17]

Experimental Section

Materials and Sample Pretreatment

Commercial powders of Y2O3, ZrO2, and YSZ were used as starting materials. Cubic (bcc) Y2O3 (yttrium(III) oxide, nanopowder, <50 nm particle size) and tetragonal YSZ (zirconium(IV)oxide Y2O3 stabilized, nanopowder, containing 8 mol % Y2O3 as stabilizer) were supplied by Sigma-Aldrich, and monoclinic ZrO2 (zirconium(IV) oxide, 99.978%) was supplied by Alfa Aesar. All samples were pretreated by calcination at 1273 K in air and subsequently checked by XRD for structural changes upon annealing. To ensure the same starting conditions for all experiments, the samples were heated in pure oxygen up to 1273 K prior to each experiment (using an isothermal period of 1 h at 1273 K and a flow rate of ∼0.8 mL s–1). The surface areas after the pretreatments were determined by nitrogen adsorption at 77 K according to the Brunauer–Emmett–Teller (BET) method as 120 (Y2O3), 32 (YSZ), and 2 m2 g–1 (ZrO2). For BET measurements, a Quantachrome Nova 2000 Surface Area and Pore Size Analyzer was used. The gases were supplied by Messer (CO 4.7, CH4 4.5, O2 5.0, He 5.0). For a typical experiment with the in-situ EIS and FT-IR setup, the samples were heated up to 1273 K, held at 1273 K for 1 h, and subsequently cooled down to 300 K at a rate of 10 K min–1 in the respective gas atmosphere under flowing conditions (between 0.8 and 1 mL s–1). To ensure dry conditions, a liquid N2–ethanol cooling trap at a temperature of approximately 163 K for CO, CH4, and O2 and for He a N2 cooling trap at 77 K was used.

Electrochemical Impedance Spectroscopy (EIS)

The operando impedance cell consists of an outer quartz tube with two inner quartz tubes to which the sample and the electrodes are attached. Heating was provided by a tubular Linn furnace and controlled by a thermocouple (K element), located in the reactor about 5 mm downstream of the sample, and a Micromega PID temperature controller. The impedance was measured by a Zahner Messsysteme IM6e impedance spectrometer, which provides data on the impedance and phase angle of the current as a function of voltage. The powder samples were pressed into pellets with a pressure of 250 MPa (5 mm diameter, thickness ≈ 0.2 mm, sample mass about 20 mg) and placed between two circular Pt electrodes which form a plate capacitor in mechanically enforced contact with the sample pellet. For all temperature-programmed conductivity measurements described in this article, an excitation voltage of 20 mV and a frequency of 1 Hz were applied to the Pt electrodes. A schematic drawing of the cell as well as the impedance setup is shown in Figure S1.

A typical Nyquist plot is obtained isothermally at a given temperature in a frequency range between 100 mHz and 0.1 MHz at the same excitation voltage that is also used for the temperature-dependent impedance measurements. The real and imaginary parts of the impedance are first measured from 1 kHz to 0.1 MHz and then from 0.1 MHz to 100 mHz (total measuring time 4 m 38 s). This is done to check the stability of the system at higher frequencies.

On the basis of the assumption that the conductivity and charge carrier concentration are proportional, determination of the activation energies was performed by Arrhenius analysis (ln(charge carrier concentration) vs the reciprocal of the reaction temperature).

Fourier-Transform Infrared (FT-IR) and Imaging Studies

FT-IR spectra were recorded in transmission mode on an Agilent Cary 660 spectrometer with a mid-infrared source and a DTGS detector. The powder samples were pressed into thin pellets using a pressure of 125 MPa (10 mm diameter, thickness ≈ 0.1 mm, sample mass about 20 mg) and subsequently placed inside a home-built operando reactor cell.[20] This cell provides a chemically inert surrounding of the sample in the heated area, and operando measurements up to 1273 K under flowing and static conditions can be performed. Also, measurements in vacuum with a minimum pressure of 3 × 10–5 mbar are possible. The window material BaF2 allows access to wavelengths above 800 cm–1. Experiments in flowing mode can be exactly correlated with associated EIS measurements. In static mode, the gases are preadsorbed on a 5 Å zeolite trap, binding water sufficiently strongly, before the dried gases are desorbed into the evacuated and thoroughly degassed cell. All reported spectra are corrected by the spectrum of the dry preoxidized oxide pellet at room temperature and under vacuum prior to exposure to the gases.

FT-MIR (mid infrared) imaging was performed using a commercially available Spectrum 400 FT-IR/FT-NIR spectrometer (PerkinElmer) coupled to an infrared microscope Spotlight 400 FT-IR Imaging System (PerkinElmer) with a built-in mercury–cadmium–telluride (MCT) 16-element linear array detector. All measurements were carried out at ambient conditions, and data analysis was performed using the SpectrumIMAGE software (R1.8.0.0410). During the preceding preparation the powder samples were pressed into thin pellets using a pressure of 125 MPa (10 mm diameter, thickness ≈ 0.1 mm, sample mass about 20 mg), and gaseous pretreatment was done with the operando FT-IR cell.[20] After oxidizing (up to 1273 K, 10 K min–1, 0.8 mL s–1) the pellets were heated in flowing CH4 or CO (10 K min–1, 0.8 mL s–1) and spectra were collected in situ every 10 K step. Then the experiment was stopped at 1163 K for CH4 and 1233 K for CO, applying active vacuum and cooling back to RT with 10 K min–1. This creates a carbon layer that is not fully percolated. The resulting samples were loaded to the imaging setup fixed on CaF2 carrier provided by KORTH KRISTALLE GmbH. A background spectrum of this CaF2 carrier was recorded before each experiment (resolution 8 cm–1, 30 summed spectra) in order to be subtracted from the data set automatically. The pellets were measured in transmission mode with a lateral resolution of 6.25 μm × 6.25 μm. Transmission spectra were recorded between 4000 and 750 cm–1. Each imaging measurement covers roughly an area of 10 mm2 containing about 270 000 spectra. Before final data analysis a noise reducing procedure via principal component analysis (PCA) was performed, and further image visualization is based upon signal ratios.

Raman Spectroscopy

Confocal Raman spectra of the polycrystalline samples in the range of 50–3800 cm–1 were recorded with a Horiba Jobin Yvon Labram-HR 800 Raman micro spectrometer. The samples were excited using the 532 nm (2.33 eV) emission line of a frequency-doubled 25 mW Nd:YAG laser under an Olympus 100× objective lens with a numerical aperture of 0.9. The size of the laser spot on the surface was approximately 1 μm in diameter. The scattered light was dispersed by an optical grating with 1800 lines mm–1 and collected by a 1024 × 256 open-electrode CCD detector. The spectral resolution, determined by measuring the Rayleigh line, was better than 2 cm–1. The spectra were recorded in unpolarized mode at ambient conditions. The accuracy of the Raman line shifts, calibrated by measuring a silicon standard, was on the order of 0.5 cm–1.

Transmission Electron Microscopy (TEM)

High-resolution imaging (HRTEM), high-angle annular dark-field (HAADF) imaging, as well as electron-energy loss spectroscopy (EELS) and energy-dispersive X-ray spectroscopy (EDXS) were performed using a 200 kV FEI TECNAI F20 S-TWIN analytical (scanning) transmission electron microscope (S)TEM equipped with a Gatan GIF Tridiem filter.

Results and Discussion

Combined EIS and FT-IR Measurements

Alternating current (ac) impedance analysis at 1 Hz excitation frequency and 20 mV excitation voltage in the temperature range from room temperature (∼300 K) to 1273 K was carried out for the Y2O3, YSZ, and ZrO2 samples to detect thermally induced changes in the conductivity, as well as potential charge-carrier-inducing (surface) stoichiometry changes (i.e., vacancy formation) during exposure to CO.

Exposing the Y2O3 sample to CO and restricting the heating up to around 1173 K (Figure ) only results in the reversible formation of thermally excited charge carriers. However, a drastic decrease of the impedance, leading to a final value of around 26 Ω at 1273 K, is observed once the temperature exceeds ∼1173 K. The following impedance course and correspondingly final low value at the highest temperature indicates the presence of a material with almost metallic conductivity, which—as will be shown by Raman spectroscopy and electron microscopy in sections and 3.5—is due to the presence of a highly distorted conducting carbon layer formed via the inverse Boudouard reaction. Even upon recooling to 300 K, metallic conductivity is still preserved (with a corresponding impedance value of 8 Ω at room temperature), implicating that the carbon layer still prevails on the Y2O3 surface.

Figure 1

Electrochemical impedance measurements on Y2O3 in flowing dry CO (flow ≈ 0.8 mL s–1). Linear heating and cooling rates of 10 K min–1 between room temperature and 1273 K have been applied.

If one takes a closer look at the impedance course of the heating routine for T < 1173 K it becomes clear that there are two discernible regions with a different course: the first one between 472 and 748 K with an associated activation energy of 27 kJ mol–1 and the second one between 748 and 1162 K (EA ≈ 112 kJ mol–1). The detailed characteristics of the drastic impedance drop in the highest temperature region (1162–1273 K) most likely originate from reaction-induced changes of surface chemistry and/or from ongoing formation and percolation of carbon islands. This will be made visible via FT-IR microscopic imaging experiments in section and results in a more and more percolated conducting graphite layer. This implies that the higher the system is heated, the more connected the formed carbon islands become, eventually forming a fully percolated C layer with only ohmic resistance.

As for a first comparison to the corresponding methane measurements, it is already known that CH4 dissociation on the studied oxides takes place at temperatures above ∼1000 K.[17] Upon comparison of the EIS experiments for Y2O3 in CH4 and CO it can be safely concluded that a conducting carbon layer was formed in both cases, which directly explains the low final impedance value. Even though completely different mechanisms are taking place, e.g., on CH4 oxide-specific H abstraction from methane by active surface entities capable of binding H2 in combination with gas-phase radical reactions at T > 1000 K takes place,[17,19] whereas for CO, the inverse Boudouard reaction, which is likely mediated by surface-reduced centers/oxygen vacancies, is responsible for the low final impedance value.

To link these conductivity changes to eventual changes in surface chemistry, Figure reveals that the FT-IR spectra on Y2O3 in flowing CO indeed perfectly correspond to the correlated EIS data (Figure S2 shows the same impedance graph with insets of FT-IR data). Carbon deposition is visible starting at T > 1023 K (Figure , drop of the total transmittance). Especially at higher temperatures (T > 1173 K)—as in methane[17]—the transmittance drops to zero while the impedance reaches ohmic resistance, caused by a conducting carbon layer. At lower temperatures (T < 1173 K) the surface chemistry is dominated by formate formation (br-formate: νas(OCO) = 1595 cm–1, δ(CH) = 1382 cm–1, νs(OCO) = 1332 cm–1, ν(CH) = 2856 cm–1).[21] This chemisorption process of CO starts around 473 K, reaches a signal maximum at 573 K, and is negligible above a temperature of 773 K. Note that this is exactly the start of the temperature region, 473–758 K, where a very low EA of 27 kJ mol–1 has been determined. This is in close correlation with the determined activation energy via the Arrhenius plot; in this case the next temperature region is the one between 748 and 1162 K with a much higher EA of ∼118 kJ mol–1. The rapid decrease of formate signals goes along with appearance of polydentate carbonates (νas(CO3) = 1444 cm–1 and νs(CO3) = 1352 cm–1). However, we already want to emphasize at this point that this is the total opposite of the behavior of YSZ in flowing CO (cf. Figure B) where the p-CO3 species arise first and their amount decreases to zero when the signals for formates become significant. Note that surface reduction of Y2O3 and related CO2 formation by reaction with surface oxygen atoms starts at 673 K. However, visible carbon deposition (in the case of Y2O3 associated with a decrease of the total transmittance) only starts at 1023 K and ends with zero transmittance above 1173 K, prevailing also upon recooling (cf. Figure S2).

Figure 2

In-situ FT-IR spectra of Y2O3 in flowing CO upon heating up to 1273 K (10 K min–1, flow ≈ 0.8 mL s–1). Relevant signals and temperatures are marked.

Figure 3

Combined EIS ((A) heating and cooling) and FT-IR ((B) heating and cooling) measurements on YSZ in flowing dry CO (flow ≈ 0.8 mL s–1). (Inset in A) Reference EIS experiment in flowing dry He (flow ≈ 1.0 mL s–1). For both methods linear heating and cooling rates of 10 K min–1 between room temperature and 1273 K were applied. FT-IR spectra were collected every 20 K step.

Figure A shows the corresponding impedance experiments in CO on the YSZ sample up to 1273 K (yellow and light blue trace). At the beginning of the heating routine (300–413 K), a plateau where the impedance hardly changes at all is oberved (note that all values above 3 × 109 Ω exceed the measurement range of the impedance spectrometer). At temperatures T > 413 K, basically four temperature regions with a different impedance course are discernible: the first one with a pretty steep impedance drop is observed between 458 and 665 K (with a corresponding EA of ∼84 kJ mol–1), the second one with a much less pronounced impedance drop is between 665 and 875 K (EA ≈ 32 kJ mol–1), the third one between 875 and 1070 K (EA ≈ 61 kJ mol–1), and the fourth one with the steepest impedance course, and the corresponding highest EA of ∼160 kJ mol–1 is between 1070 and 1265 K. At the end of the heating routine, a final value of 824 Ω is obtained. Even during the isothermal period at 1273 K for 1 h this value hardly changes. Recooling in CO to 300 K re-establishes the high impedance as present before admission of CO, in contrast to the Y2O3 sample. During recooling, again several different regions are visible: the first one between 1178 and 1287 K (EA ≈ 125 kJ mol–1), the second one with a very drastic impedance rise between 906 and 1178 K (EA ≈ 161 kJ mol–1), the third one with a much less pronounced impedance rise between 781 and 906 K (EA ≈ 48 kJ mol–1), the fourth one with a plateau-like course and basically no activation energy between 641 and 781 K, and the fifth one between 469 and 641 K (EA ≈ 105 kJ mol–1). This last region appears very similar as compared to the first temperature region in the heating procedure with a slightly higher activation energy. Although the same impedance value is obtained in the end, the impedance courses for the heating and cooling procedure differ somewhat, especially between 650 and 1030 K. If the temperature-dependent impedance traces and the FT-IR spectra are compared in Figure S3, a general trend becomes clear: over the whole temperature region there are four different regions where different conduction mechanisms are taking place. The first one is the low-temperature section between RT and 413 K: here the impedance is in the GΩ range and the sample shows insulating properties. This is the region where the surface chemistry of CO is suppressed—only a very small amount of p-CO32– species (νas(CO3) = 1444 cm–1, νs(CO3) = 1411 cm–1, Figure B green signals) and br-formate species (νas(OCO) = 1583 cm–1, δ(CH) = 1384 cm–1, νs(OCO) = 1361 cm–1, ν(CH) = 2877 cm–1, Figure B red signals) are visible starting from 373 K.[22] The second section is between 413 and 650 K. This is one of the areas where the surface chemistry plays an important role: the bands for p-CO32– and br-fomates (red and green signals in Figure B) increase drastically with increasing temperature, whereas the signals for p-CO32– reach a maximum at 573 K and are rapidly decomposed upon further heating and are no more present above 593 K. This is connected to a first measurable maximum in the CO2 gas-phase signal (labeled in blue in Figure B, 2300–2400 cm–1). Along with the disappearance of the carbonates, the bands for formates do increase strongly and reach a maximum intensity at 653 K. It is also noticeable that there is a correlation of formate adsorbates and a ν(OH) signal at 3675 cm–1 which also emphasizes the fact that surface chemistry and related hydroxylation play a dominant role in this temperature region. As can be clearly seen in Figure S3B the maximum of these adsorbate species is exactly between temperature sections 2 and 3. The third temperature area is between 650 and 1030 K. In this region the signals for OH groups and formates start to decrease and are completely removed at around 933 K. Parallel to the total decrease of surface adsorbate species a distinct and continuous CO2 gas-phase signal is observed, indicating a direct catalytic vacancy-mediated mechanism (Mars-van-Krevelen mechanism, CO + Ooxide → CO2 + #). Note that this is a flowing experiment where any initially formed gas-phase species is removed by the CO gas flow, so a continuous signal implies a continuous CO2 formation. Figure S3A highlights the beginning of the fourth temperature section which exactly coincides with the beginning of carbon deposition at 1030 K as can be also seen in Figure S3B (gray traces). The CO decomposition is visualized in the FT-IR spectra with a prominent fingerprint feature below 1800 cm–1 (marked in light blue in Figure B), and subsequently, in line with the inverse Boudouard reaction the CO2 gas-phase signal increases immensely during decomposition.

The variations in the heating and cooling curves hence arise due to a different conduction mechanism. The differences in the high-temperature area (1030–1273 K) are most likely due to carbon that has been deposited in this region, which can somewhat enhance the impedance but will also block “active” sites, which could otherwise be used for adsorption of different species. The FT-IR spectra in Figures B as well as S3B also prove that the carbon fingerprint does not change upon recooling, and no rehydroxylation or formation of formates, carbonates, or OH groups takes place in the third temperature region (650–1030 K), which also explains the major variations in the heating and cooling curves of the temperature-dependent impedance plot (Figures A and S3A). The plateau-like feature that appears between 650 and 773 K during recooling might be due to a complex interplay between deposited carbon and other surface-related conduction effects. During recooling absolutely no signal for gas-phase CO2 is detected in the infrared measurement. This also suggests that all “active” vacancy-related reaction sites are already consumed or blocked due to carbon deposition. Interestingly, there is a very similar impedance behavior in temperature section one and two for the heating and cooling procedures, although the IR spectra do not show any indication of surface adsorption activity such as OH groups or formates. Nevertheless, a certain reactivation of the surface areas between the carbon islands (see discussion concerning island-growth mechanism in section ) cannot be excluded upon cooling.

Taking a closer look at the role of the deposited carbon, a second temperature-dependent impedance experiment was conducted (Figure A red and dark blue traces) where the system was heated to a temperature (∼1020 K), which is just before the carbon deposition starts. Even though the heating and cooling curves for the two measurements differ somewhat (note that kinetic effects also play a major role in all those measurements), a very similar trend is observed. This time three temperature regions are apparent, which are a very good match for the first three sections from the measurement that was conducted up to 1273 K. Just like for the area between 650 and 1030 K in Figure A (yellow and light blue trace) a very similar trend can be seen in this experiment as well, with the difference being that the temperature region where the heating and cooling varies is between 650 and 923 K. This is a strong indication that the deposited carbon does play a relevant role in this temperature region, especially during cooling.

To get a better understanding of the temperature region between 650 and 1030 K, a temperature-programmed impedance experiment was conducted in flowing dry He (inset Figure A). A similar feature as for the CO measurements appears between 650 and 747 K, but in this case the heating and cooling curves hardly differ at all. This measurement helps to verify that the first part of the experiment (from RT to 747 K) is strongly influenced by surface chemistry and that at T > 747 K oxide ion conduction is dominant. Note that this fits perfect to the temperatures where YSZ is used for its high oxide ion conduction.

Obviously, in the case of CO disproportionation on YSZ, metallic conductivity (as for Y2O3) could not be obtained, but a roughly two magnitudes higher final value was observed at 1273 K. This indicates that carbon deposition must have taken place but not to the same extent as on the Y2O3 sample; hence, no fully conducting (percolated) carbon layer could be formed. As outlined in a previous publication, treatment of the YSZ sample in CH4 leads to such a conducting carbon layer composed of structurally highly distorted graphite with a very low value for the impedance (15 Ω) at 1273 K.[17] Methane was therefore dissociated predominantly via gas-phase radical reactions/H-abstraction at temperatures above 1000 K. This indicates that the degree of carbon deposition might vary to some extent upon using either CH4 or CO as carbon precursor. The most obvious reason clearly is that the surface of the used oxide must play an important role in the inverse Boudouard reaction, since it is effectively a surface-catalyzed reaction without the influence of gas-phase radical chemistry. Reduced centers could be essential for carbon deposition, which would then coincide with the associated formation of CO2 (for a detailed discussion we refer to section ). It is known from a previous study that all three sampled oxide materials are reducible in surface-near regions in H2.[23] The associated FT-IR experiments show that these oxides are also reducible in CO, as indicated by the decrease of transmission and an increase in the CO2 gas-phase signal.

Figure A shows the corresponding impedance measurements on ZrO2 in CO, which are strikingly different especially with respect to Y2O3 but also to YSZ. During the heating routine in CO three temperature regions with different activation energies are observed (Figure B): the first one starts just when semiconductive behavior is observed at 786 K and ends at 872 K with an EA of ∼111 kJ mol–1, the second region is between 872 and 1065 K with an EA of ∼145 kJ mol–1, and the third one is between 1065 and 1252 K with an EA of ∼222 kJ mol–1.

Figure 4

Combined EIS ((A) heating and cooling and (B) Arrhenius plot) and FT-IR ((C) heating, (D) cooling) measurements on ZrO2 in flowing dry CO (flow ≈ 0.8 mL s–1). Both methods were performed at linear heating and cooling rates of 10 K min–1 between room temperature and 1273 K. FT-IR spectra are shown every 100 K step.

During the isothermal period (1 h at 1273 K) the impedance again hardly changes. Upon recooling, a very similar behavior as for the heating routine is found with three temperature regions (821–973 K, EA ≈ 140 kJ mol–1; 973–1068 K, EA ≈ 172 kJ mol–1; 1068–1272 K, EA ≈ 213 kJ mol–1).

When comparing the temperature-dependent impedance experiments for the three oxides, it is clear that there is a large difference in the impedance value at the highest temperature. If the Y2O3 sample is treated in CO, metallic conductivity is established and the system remains in this state even during the cooling process. In the case of the YSZ sample, an impedance value of 824 Ω at 1273 K is obtained, which is almost two magnitudes higher than the one for the Y2O3 sample. Note that this value is very similar to the one obtained during treatment of YSZ in dry flowing He at 1273 K (inset Figure A purple and green traces). This strongly indicates that the value obtained at 1273 K in dry CO is most likely solely due to the ionic conductivity of the YSZ pellet, demonstrating that the influence of the C islands is hardly recognizable. At the end of the cooling routine the starting conditions could be re-established since the impedance is again in the GΩ area, which is due to the fact that no percolated carbon layer was formed and only isolated carbon islands are present on the surface of the sample. However, regarding the ZrO2 sample an impedance value of 1.12 × 105 Ω is obtained at 1273 K, and at room temperature after the cooling process the sample shows insulating properties again. This impedance course suggests that there was definitely not as much carbon deposited on this surface as compared to the other samples, and one can exclude the formation (at least during the first heating and cooling cycle) of a conducting carbon layer due to the reversibility toward the initial low conductivity value. From this we can safely conclude that the carbon deposition in CO on the three oxides is most efficient on Y2O3, followed by YSZ and then ZrO2. This will in the following be directly corroborated by Raman measurements and TEM experiments.

As mentioned before, on ZrO2 treated in flowing CO, very limited carbon deposition takes place, which is also reflected in the FT-IR experiment depicted in Figure C and 4D. There is no decrease or fingerprint formation in the transmittance over the whole wavenumber range but only a noticeable “negative” impact while heating (especially above ∼473 K), which matches surface reduction via CO2 formation. The maximum of this effect is reached at 973 K, and upon further heating and cooling the spectra do not change anymore. During heating weak signals for formates can be detected (νas(OCO) = 1563 cm–1, ν(CH) = 2871 cm–1, temperature region between 533 and 873 K), but formates are absent while recooling back to room temperature. This indicates that all reactive surface sites (hydroxyls) of the initial state are consumed in the heating process like on YSZ, but on ZrO2 no bands for p-CO3 species, like on the other oxides, could be detected.

As a tentative explanation for the observed differences of the three oxides, which is also directly deducible from the FT-IR experiments, it is again obvious that the amount of hydroxylation is strongly related to the amount of surface adsorbates,[17,20,23,24] in this case formate signals. There seems to be no direct connection between formate formation and C deposition since the formates are decomposed at much lower temperatures than the carbon deposition takes place, but it appears that the more hydroxyl groups are present, the stronger the subsequent carbon depositing effect is.

Table summarizes the relevant temperature regions alongside its determined activation energies for the three different samples in flowing CO. These activation energies are not only a function of the experimental conditions but of course also of the temperature region where the linear fit was applied to. However, upon comparing the EA’s for CO, a general trend is apparent: there are always at least three or more processes with different activation energies taking place. This can at all times be traced back to the change of the surface of the sample as detected in the FT-IR experiments. The formation or removal of hydroxyl groups, carbonates, or formates strongly affects the electrochemical impedance and hence the Arrhenius plot with its EA’s. Another trend that becomes clear by comparing the determined activation energies is (i) that they are usually higher for the cooling processes than for the heating routines and (ii) that the high-temperature processes show the highest EA’s. This is due to the removal of OH groups during the heating routine, which are partially recovered again upon recooling. Hence, several processes, including OH-mediated surface protonic conductivity,[25] are superimposed on bulk anion transport. This means that multiple processes simultaneously alter the surface of the used sample and therefore result in higher or lower activation energies.

Table 1

Activation Energies for Y2O3, YSZ, and ZrO2 in Dry Flowing CO

sample	treatment	EA CO (kJ mol–1)	temperature range (K)
Y2O3	heating	27 ± 1.6	473–748
		112 ± 6.7	748–1162
YSZ	heating	84 ± 5.0	458–665
		32 ± 1.9	665–875
		61 ± 3.7	875–1070
		160 ± 9.6	1070–1265
	cooling	105 ± 6.3	469–641
		∼0	641–781
		48 ± 3.0	781–906
		161 ± 9.7	906–1178
		125 ± 7.5	1178–1287
ZrO2	heating	111 ± 6.7	786–872
		145 ± 8.7	872–1065
		222 ± 13.3	1065–1252
	cooling	140 ± 8.4	821–973
		172 ± 10.3	973–1068
		213 ± 12.8	1068–1272

To gather more information about the dynamics of the carbon layer growth, Nyquist plots in flowing CO (light blue trace) and, for comparison, in flowing CH4 (dark blue trace) were conducted at various temperatures before and during carbon island growth (Figure , CO at 1123 K and CH4 at 973 K). For CO, in comparison with Figure , it becomes immediately clear that this is exactly the temperature just before the drastic impedance drop, i.e., ongoing slow carbon deposition can be assumed. This can be perfectly correlated with a drastic drop in the transmittance in the FT-IR spectra: at 1073 K a transmittance of 59%T and at 1123 K a transmittance of ≥1%T is left. Due to this complex Nyquist plot one can assume that the carbon deposited on the sample dynamically grows to form islands, as it is the case for CH4.[17] High temperatures or long exposure times at high enough temperatures thus lead to the interconnection of those separated carbon atoms/small islands, which eventually form a fully percolated conducting carbon layer. As the impedance was cycled from 1 kHz to 0.1 MHz (within 14 s) and back from 0.1 MHz to 100 mHz (within 4 m 24 s), this usually verifies the electrochemical stability of the system (because on an unaltered system the real and imaginary contributions of the impedance for the same frequencies should not differ), but in the present case it also helps to demonstrate the dynamics of island growth on the chosen experimental time scale. At frequencies of 0.1 kHz and below (i.e., after 31 s) there is hardly any contribution of the imaginary part left, implying only ohmic resistance. This infers the presence of a more or less complete conducting carbon layer. Consequently, a purely “ohmic” resistance in the respective Nyquist plot is observed if the sample is treated in CO and heated up to temperatures above 1123 K (1173 K; Figure S4). The imaginary part of the impedance is then basically zero, and only a “real” resistance (without phase shift) is found. As shown in Figure , only the Y2O3 sample treated in CO shows metallic conductivity and hence this very complex course of the impedance. Nevertheless, Nyquist plots for YSZ and ZrO2 in CO were also recorded, but no such complex frequency-dependent behavior was observed. This special impedance course can only be observed during the process of the separated islands getting more interconnected forming larger islands, which eventually fully percolate to form a continuous conducting carbon layer. However, this is not the case for YSZ and ZrO2, despite the fact that evidently on these two oxides either an almost comparable or a larger amount of carbon was deposited in comparison to CH4 (see Table ). Nevertheless, no fully percolated and thus no conducting carbon layer could be generated (cf. Figures A and 4A).

Figure 5

Nyquist plots of Y2O3 treated in dry flowing CO (circles, flow ≈ 0.8 mL s–1) at 1123 K in comparison to in dry flowing CH4 (squares, flow ≈ 0.8 mL s–1) at 973 K.

Table 2

Quantification of the Inverse Boudouard Reaction as Well as of Methane Decomposition and the Resulting Carbon Layers on Y2O3, YSZ, and ZrO2

sample	gas	total deposited amount of C (μmol g–1)	amount of C (% of value on Y2O3 in CO)
Y2O3	CO	16 500	100
	CH4	8400	51
YSZ	CO	9610	58
	CH4	10 800	65
ZrO2	CO	3130	19
	CH4	1630	10

For comparison, in Figure (dark blue trace) a Y2O3 pellet treated in CH4 at 973 K is also shown, exhibiting a very similar trend as observed in CO. Again, a very complex frequency-dependent impedance course can be observed with a few notable differences: the temperature for this process in CH4 is 150 K lower than for the process in CO, and the contributions of the real and imaginary part of the impedance are generally lower as compared those for Y2O3 treated in CO.

All three samples were analyzed with FT-IR spectroscopy, but only YSZ is shown as an example in Figure , as this is the oxide showing the most pronounced spectral information due to the formation of a distinct fingerprint of the carbon layer in both gases (CO and CH4). The resulting carbon fingerprint signals of the carbon layers are very similar, which is in good correlation with the Raman spectra discussed in section . On Y2O3 carbon deposition leads to a total overall decrease of the transmittance, and in the case of ZrO2 no characteristic impact on the spectra is visible (lowest amount of deposited C in comparison to Y2O3 and YSZ).

Figure 6

Comparison of the FT-IR fingerprint of the carbon layer deposited on YSZ in either dry flowing CO (dark blue, flow ≈ 0.8 mL s–1) or CH4 (light blue, flow ≈ 0.8 mL s–1).

Quantification of the Deposited Carbon

From the above-discussed combined impedance and FT-IR results a conversion of CO into C and CO2 on the oxide surface can be deduced. Consequently, information on the degree of CO conversion as well as on the amount of deposited carbon material is highly desirable. To provide these data, complementary quantitative FT-IR measurements have been performed in close correlation to the already discussed EIS and FT-IR experiments. These studies were performed for CO as well as for CH4 to be able to identify which gas treatment causes more carbon deposition and hence harms the system more with respect to eventual carbon-induced structural breakdown.

Figure S5 shows this procedure for YSZ as an instructive example. Although the data are only shown in the Supporting Information, we provide a brief outline of calibration at this point. First, the carbon was deposited by a heating and cooling routine (10 K min–1) and a subsequent isothermal period at 1273 K for 1 h in flowing CO and CH4. Second, the samples were reoxidized in ∼1 bar static O2, the resulting CO2 peak was then recorded at room temperature, and the peak area was calculated (cf. green peak Figure S5A). To avoid temperature effects on the spectra, the same background as for the deposition was used. Third, vacuum was applied, and various pressures of CO2 were measured with the same background as before to calibrate the CO2 sensitivity (cf. Figure S5B). A calibration line thus results (cf. Figures S5C and S5D). Finally, the resulting CO2 pressure caused by oxidation of the carbon layer was converted into μmol g–1 deposited carbon. The quantification details are summarized in Table .

Table clearly reveals that at least on Y2O3 and ZrO2 treated in CO more carbon was deposited as compared to in CH4. The highest amount of carbon was deposited on Y2O3 in CO; hence, this was set as 100%, and all other carbon amounts are normalized to this value. This finally leads to a sequence of the amount of deposited carbon in CO of Y2O3 > YSZ > ZrO2 as compared to in CH4 of YSZ > Y2O3 > ZrO2.

At this point, a short comparison of the distinct carbon growth mechanisms in CO and CH4 appears useful: it is already known from the literature[17,19] that the mechanism for the C deposition in CH4 proceeds via H abstraction from methane by active surface entities capable of bonding hydrogen to form gas-phase methyl radicals, which not only are highly reactive but also exhibit accordingly fast diffusion properties (i.e., diffusion into all pores of the oxide powder takes place). This gas-phase/surface hybrid mechanism[19,26] suggests that the generated C layer might be much more homogeneous (specifically if H abstraction can occur homogeneously on the growing carbon layer itself), which is also reflected in the impedance spectra of a previous study, which show metallic conductivity upon CH4 treatment on all three oxides.[17] In contrast, the mechanism in CO most likely proceeds via a surface-catalyzed inverse Boudouard reaction (2CO ↔ C + CO2) starting at local defect centers. As there are no gas-phase radicals involved in this reaction, the mechanism probably follows a vacancy-mediated Mars-van-Krevelen-type pathway with temperature-dependent reaction steps. Step 4 is then the crucial step for carbon deposition, occurring only at temperatures T > 1000 K.

The formation of single C atoms is expectedly followed by local density fluctuations of surface carbon atoms to eventually build a sufficiently large and thus stable graphite nucleus. In such a model, highly anisotropic nucleation and growth processes are unavoidable, especially if the individual (“isolated”) C atoms can also react back toward CO. Also, for energetic reasons we may suggest that carbon is preferentially formed at/near and then incorporated into already present graphite islands, specifically because bonding in an already existing graphitic lattice is energetically more favorable and a back-reaction with lattice oxygen toward CO thus less likely. A related defect-dependent pathway, at least for initial CO adsorption, has already been verified for substoichiometric TiO2.[27−29]

This scenario may explain, according to Table , not only why more carbon was generally (except on YSZ) deposited in CO as in CH4 but also its less homogeneous distribution. However, even though metallic conductivity was observed on all three oxides in CH4, in CO it was only obtained on Y2O3 (cf. Figures , 3A, and 4A). As it is known from temperature-programmed reduction experiments in CO, reduction of the surface in CO starts already at ∼680 K, but FT-IR measurements also prove enhanced reducibility at higher temperatures (∼1113 K), as deduced from the strong increase of the CO2 gas-phase signal, coinciding with the occurrence of the carbon fingerprint shown in Figure . On Y2O3 we therefore may suggest a higher initial density of graphitic nuclei at the Y2O3 surface which then could allow for further growth of graphite toward island coalescence/full percolation. In contrast, on the less reducible YSZ and ZrO2 samples, much more localized carbon nucleation and growth and, in turn, less or not at all percolated carbon deposits are to be expected. As shown in Figure for carbon deposition in CO on Y2O3, island growth and the formation of a percolated conducting carbon layer takes place at higher temperatures (shifted by 150 K) and with a higher time offset (shown in Figure C and 8D for CH4 and CO) in comparison to CH4. This suggests that the growth of the graphite islands is energetically more demanding and their coalescence strongly time dependent.

Figure 8

(A) FT-IR imaging of YSZ after treatment in CH4 up to 1163 K. (B) FT-IR imaging of YSZ after treatment in CO up to 1233 K. Images are based on the ratio between 1850 and 1750 cm–1 and 1250–1150 cm–1. (C) Image of YSZ pellet after treatment in CH4 up to 1163 K; the imaged region is marked. (D) Image of YSZ pellet after treatment in CO up to 1233 K; imaged region is marked. (E) Single-beam spectra of YSZ with varying amounts of carbon deposited depending on the maximum temperature treatment in CH4.

On the contrary, upon exposure to CH4 and heating up to high enough temperatures on all three oxides metallic conductivity is observed in the temperature-dependent impedance spectra. Even though only 10% carbon (in comparison to the highest value of deposited carbon, which was defined as “100 % carbon” for Y2O3 in CO) was deposited on ZrO2 in CH4, which is 9% less than the amount formed in CO (see Table ), this is obviously enough to create a thinner but nevertheless conducting carbon layer. This again supports the picture that the carbon that is deposited during CH4 treatment is much more homogeneously distributed than in CO. If, beyond the “homogeneous” decomposition of the intermediate methyl radicals on the oxide surface, further growth of the already existing graphite layer is self-limiting in the sense that H abstraction by carbon becomes less favorable and recombination of methyl radicals with hydrogen more favorable then deposition of very thick layers will be rather inhibited. In such a scenario rather thin layers could already be fully percolated.

Raman Spectroscopy

In addition to EIS, FT-IR, and FT-IR quantification studies, Raman measurements were performed to verify the chemical state and structure of the deposited carbon material. This is particularly worthwhile since Raman spectroscopy is a very valuable tool to supply qualitative information on carbon materials and physicochemical properties, such as their phonon and electronic structure or the presence of structural defects.[30] To follow the kinetics of the formation of the carbon layers, Raman spectra of the samples treated in CO up to 1273 K and held at the highest temperature for a selected times have been collected.

Nanocrystalline Y2O3 was used as starting material and hence exposed to flowing CO from room temperature to 1273 K, held at this temperature for 1 h, and cooled to room temperature (Figure yellow trace). Cubic Y2O3 exhibits a dominant vibrational mode at 377 cm–1 and additional bands at 138, 161, 197, 317, 377, 415, 452, 469, and 585 cm–1.[31] However, the Raman spectra of the sample annealed at 1273 K show additional bands due to the formation of carbon layers on the surface of the oxide. The prominent features are the nondispersive G- band which corresponds to double-degenerate in-plane stretching vibrations of sp2-bonded carbon atoms and are present in most graphite-like materials around 1550–1600 cm–1. The dispersive D band originates from structural disorder and is present around 1340–1360 cm–1, whereas the dispersive G′ or 2D band is an overtone or second-order harmonic of the D peak, which appears around 2700 cm–1, and last, the combination mode of the G and D band, also due to structural disorder, peaks around 2950 cm–1.[32−35] In the Raman spectrum of the Y2O3 specimen heated up to 1273 K in CO the G peak is located at 1592 cm–1 and the D peak at 1338 cm–1; the G′ peak and the combination mode of the G and D bands are not visible due to their small intensity compared to the other peaks. The intensity ratio of the G and D peaks indicates that the species formed on the Y2O3 surface are most likely highly distorted nanocrystalline graphite layers.[36−39]

Figure 7

Selected Raman spectra collected on Y2O3 (yellow trace), YSZ (light and dark blue traces), and ZrO2 (green trace) after different annealing treatments (as indicated for the individual graphs) after exposure to flowing dry CO (flow ≈ 0.8 mL s–1) at 1273 K.

A very similar preparation and analysis procedure was performed for YSZ (light and dark blue spectra in Figure ). The Raman spectrum of cubic YSZ is usually characterized by the fundamental optical modes at 149, 254, 316, 469, and 635 cm–1,[40] but these are suppressed by the carbon layer in the spectrum associated with the treatment for 1 h at 1273 K (light blue trace). Carbon-related vibrational modes bands occur at 1344, 1594, 2679, and 2933 cm–1. Just like for Y2O3, the ratio between the G and the D peaks indicates nanocrystalline graphite with a high degree of disorder. A second sample was prepared, where a closer look was taken at the kinetic effects of carbon growth during the isothermal period. This sample was only held at the highest temperature (1273 K) for 10 min, resulting in a G to D peak ratio of almost 1:1. This indicates that the carbon that is deposited up to this point is not as highly distorted as the carbon on the other sample that was temperature treated for a longer time.

Similar carbon-related modes have been observed for pure ZrO2, but the difference between the Raman modes of pure ZrO2 (external modes at 99, 179, 191, 223 cm–1 and internal modes at 306, 334, 348, 382, 476, 503, 537, 559, 616, 639, 758 cm–1)[41] and modes due to deposited carbon is even more pronounced (Figure green trace). In line with that, carbon-related frequencies for the G, D, and G′ bands and the combination modes of the G and D bands compared to Y2O3 and YSZ are observed. Even the intensities of the G and D bands are very similar to the ones for the YSZ sample. It nevertheless also reveals that the amount of deposited carbon is smallest on ZrO2.

FT-IR Microscopic Imaging

To visualize the stepwise percolation of locally growing carbon islands during the deposition process, FT-IR imaging was conducted on YSZ pellets with a deliberately partially C-covered surface. The pellets shown in Figure were heated in CH4 up to 1163 K (Figure C) and CO up to 1233 K (Figure D). The C growth was monitored in situ, and the heating process was stopped by applying vacuum and recooling the samples within the cell right when the changes in the FT-IR spectrum due to carbon deposition were clearly visible but were not yet strong enough to represent a complete carbon layer. Already the optical images of the resulting pellets indicate that the carbon deposition is generally not homogeneous, which is also reflected in the corresponding FT-IR images (Figure A for CH4 and Figure B for CO, respective areas are marked on the pellets). These images result from a ratio evaluation of each spectrum between 1850 and 1750 cm–1 and 1250 and 1150 cm–1. The reason for the chosen wavenumber areas is illustrated in Figure E: the deposition of carbon leads to a successive weakening of the transmittance of the single-beam spectra in the region below 1500 cm–1, but especially in the 1250–1150 cm–1 region this effect is pronounced in comparison to the rest of the spectra. Comparing the carbon growth on YSZ in CH4 and CO supports the suggested differences in the deposition mechanisms and correlated differences in the island-like growth. Both the FT-IR imaging in Figure A and the microscopic image in Figure C show a more homogeneous distribution of areas with a “thick” carbon layer (dark blue/purple pixels indicate a higher amount of C) than in CO in Figure B and 8D. Although carbon deposition starts at a similar temperature in CH4 and CO (∼1030 K) and the total amount of carbon is comparable on YSZ in both gases after heating up to 1273 K (see Table ), the images in Figure B and 8D (after CO treatment) look more spotted than the images in Figure A and 8C after CH4 treatment. Note again that the experiment in CO had to be conducted up to higher temperatures than in CH4 to create a comparable carbon-covered surface, which again highlights the pronounced island growth in CO, since the treatment in CH4 leads to a more homogeneous layer already 70 K earlier (at higher temperatures the pellet treated in CH4 is completely black and for FT-IR imaging is not suitable anymore). A closer look at the image in Figure B in CO unravels a clear purple spot (which is equivalent for a quite “thick” island) at an approximate position of x ≈ −2000 μm and y ≈ −700 μm, which is not visible in the corresponding image in Figure D. This is explained by the fact that the FT-IR imaging process is conducted in transmission mode, and this spot represents a carbon island on the back-side of the pellet, which is obviously not percolated through the pellet but only a surface spot. Sufficient percolation through the pellet volume is therefore not observed, which matches the lack of metallic conductivity in CO despite the high total amount of carbon. The better diffusion properties of the mechanism in CH4 lead to metallic conductivity already at lower temperatures since the percolation on the surface and probably also between the grains of the pellet is favored and reflected by the imaging experiment.

Transmission Electron Microscopy

Figures and 10 in turn highlight the structure of the deposited carbon layers in a joint comparative discussion on Y2O3 and YSZ (and ZrO2 in the SI). For Y2O3, the detected carbon layers are usually thickest and comprise up to 10 graphitic layers almost entirely covering the partially agglomerated individual grains (Figure A; the lower Y2O3 grain shows (222) lattice fringes with a distance of about 3.1 Å). Corroborating the Raman measurements discussed before, these layers are usually structurally severely distorted and appear almost amorphous. The chemical nature of the graphitic covering layer is also reflected in dedicated EELS measurements. Figure B therefore comprises the edge of a Y2O3 grain in high resolution with a comparatively thin shell of carbon on the grain contours. The shell itself is not thicker than five structurally heavily distorted graphitic layers. EEL spectra taken on the carbon layer (marked by the green arrow) directly also reveal inhomogeneous graphite-like features. This follows from the comparison to literature-reported crystalline graphitic and amorphous carbon fingerprint spectra.

Figure 9

High-resolution image of two agglomerated Y2O3 grains (with (222) lattice spacings) after exposure to flowing CO (flow ≈ 0.8 mL s–1) for 1 h at 1273 K covered by a thick shell of 10–15 individual, structurally heavily distorted graphene layers (A). High-resolution image of a Y2O3 grain after exposure to flowing CO (flow ≈ 0.8 mL s–1) for 1 h at 1273 K covered by a thinner shell of up to 5 individual, structurally heavily distorted graphene layers. (Inset) Single C–K EEL spectrum (green) taken at the particle edge at the position where the green arrow is pointing to alongside reference spectra of crystalline graphite (red) and amorphous carbon (blue) (B).

Figure 10

HAADF image of YSZ after exposure to flowing CO (flow ≈ 0.8 mL s–1) for 1 h at 1273 K. The red arrow in the main panel indicates where the C K-edge and total HAADF intensity have been measured (inset).

To further highlight the structure of the carbon layers, Figure shows the corresponding experiments on YSZ. In this particular case and to show that carbon deposition is mainly found on the particle outlines, HAADF imaging was combined with line profiling of the carbon K-edge EELS intensity and the total HAADF intensity. As it is immediately clear, the HAADF intensity in fact follows the particle contours, whereas the carbon intensity clearly peaks only at the edges. The same is, in principle, true for the corresponding experiments on ZrO2. Also, in this case (Supporting Information, Figure S6), the EELS line profile shows carbon enrichment at the particle edges. EELS experiments performed at the particle edges (indicated by the line profile; Figure S6B) again reveal the similar nature of the graphitic carbon layers.

In summary, the TEM images show structurally and electronically more or less similar carbon layers, i.e., are graphite-type. The images also reveal that the overall amount of deposited carbon is highest on Y2O3, and significantly less carbon is found on YSZ and ZrO2 (which follows from several representative sites on Y2O3 that have been checked to substantiate this observation). This is in very good agreement with the carbon quantification studies discussed in the previous section. Of particular importance is also a direct comparison of the obtained carbon layers in CH4[17] and CO, since the carbon precursors are different and hence also the formation mechanism of the deposited carbon. However, in line with the results discussed above, in particular, the FT-IR fingerprints of the carbon layers and the Raman measurements, the analytical TEM results rather support the picture that in fact the structure of the obtained carbon islands is at least locally similar.

Conclusion

From the outlined results it is safe to conclude that the studied oxides, potentially among many others, may act as efficient substrates for carbon deposition via the inverse Boudouard reaction given a high enough temperature to induce C–O bond scission at surface vacancies and/or defects. Carbon deposition and, depending on the oxide, the formation of a highly distorted graphitic carbon layer takes places on all studied oxides at temperatures above ∼1023 K. Carbon deposition was most efficient on Y2O3, which could be quantitatively verified by FT-IR experiments. Temperature-dependent impedance spectra (metallic conductivity) as well as FT-IR studies (decrease in the transmittance to zero in analogy to CH4) also corroborate these results. As for the comparison between CO and CH4, FT-IR experiments reveal that the start of carbon deposition is at exactly the same temperature for CO and CH4. However, as follows from temperature-programmed impedance measurements, the beginning of the steep impedance drop in CH4 is exactly at the starting temperature of carbon deposition (∼1000 K), whereas for CO graphite island coalescence is shifted 100–150 K to higher temperatures (∼1100 K). This is also visualized in the comparative Nyquist plots of Y2O3 in CO and CH4. These differences arise from the different reaction mechanisms, which are a hybrid methyl gas-phase radical/H-abstraction process in the case of CH4 and the surface-vacancy-mediated inverse Boudouard reaction for CO. Nevertheless, the CH4- and CO-deposited graphitic layers are not substantially different with respect to their local structural properties. Thus, the lacking graphite coalescence/percolation in CO on YSZ and ZrO2 might be caused by a disadvantageous combination of gradually more localized growth and the lower total amount of carbon, causing a shortfall below the island percolation limit.