Influence of Oxygen Partial Pressure during Processing on the Thermoelectric Properties of Aerosol-Deposited CuFeO2

In the field of thermoelectric energy conversion, oxide materials show promising potential due to their good stability in oxidizing environments. Hence, the influence of oxygen partial pressure during synthesis on the thermoelectric properties of Cu-Delafossites at high temperatures was investigated in this study. For these purposes, CuFeO2 powders were synthetized using a conventional mixed-oxide technique. X-ray diffraction (XRD) studies were conducted to determine the crystal structures of the delafossites associated with the oxygen content during the synthesis. Out of these powders, films with a thickness of about 25 µm were prepared by the relatively new aerosol-deposition (AD) coating technique. It is based on a room temperature impact consolidation process (RTIC) to deposit dense solid films of ceramic materials on various substrates without using a high-temperature step during the coating process. On these dense CuFeO2 films deposited on alumina substrates with electrode structures, the Seebeck coefficient and the electrical conductivity were measured as a function of temperature and oxygen partial pressure. We compared the thermoelectric properties of both standard processed and aerosol deposited CuFeO2 up to 900 °C and investigated the influence of oxygen partial pressure on the electrical conductivity, on the Seebeck coefficient and on the high temperature stability of CuFeO2. These studies may not only help to improve the thermoelectric material in the high-temperature case, but may also serve as an initial basis to establish a defect chemical model.


Introduction
With thermoelectric generators, thermal energy can be directly converted into electrical energy. Great efforts have been undertaken in the past few decades to increase the efficiency-characterizing figure of merit (ZT) which depends on the Seebeck coefficient (S), electrical conductivity (σ), and thermal conductivity (κ). If one considers only the electrical parameters, the power factor (PF) is an established parameter of thermoelectric materials: ZT values above 1 were reported for semiconductors like Bi 2´x Sb x Te 3 and filled skudderudites like Ba 0.3 Ni 0.05 Co 3.95 Sb 12 or SnSe [1][2][3][4][5][6][7], and can be further enhanced when optimizing the thermoelectric properties through nanostructuring [8][9][10][11]. However, the commercial application of these materials was removed in a rotary evaporator (Heidolph Instruments, Schwabach, Germany). To elucidate the influence of the oxygen content of the gas atmosphere during the solid state reaction, CuFeO 2 fired in 100% N 2 and CuFeO 2 fired in 1% O 2 were synthesized in a high temperature furnace at 1050˝C for 12 h. The obtained delafossite powders were reground in a planetary ball mill using the above mentioned method, sieved with a 90 µm screen in order to reduce agglomerates and finally dried in a furnace at 200˝C for at least 24 h. A scanning electron microscope (SEM, Zeiss, Oberkochen, Germany) image of the calcined and milled delafossite powder used for the AD process is shown in Figure 1. It can be seen that there is a broad particle size distribution ranging from 0.1 to 30 µm which is uncommon for AD processes. Bulk CuFeO 2 samples were formed into brick shaped pellets, uniaxially cold pressed, and sintered at 1050˝C under the same gas atmosphere as the corresponding starting powder. In order to determine the thermoelectric properties, platinum/gold thermocouples and platinum wires were attached to the sintered samples with platinum conductor paste. Details of the setup are shown in Figure 2.
Materials 2016, 9,227 3 of 16 (FRITSCH, Idar-Oberstein, Germany) with cyclohexane as solvent. After milling the powders for 4 h, the solvent was removed in a rotary evaporator (Heidolph Instruments, Schwabach, Germany). To elucidate the influence of the oxygen content of the gas atmosphere during the solid state reaction, CuFeO2 fired in 100% N2 and CuFeO2 fired in 1% O2 were synthesized in a high temperature furnace at 1050 °C for 12 h. The obtained delafossite powders were reground in a planetary ball mill using the above mentioned method, sieved with a 90 µm screen in order to reduce agglomerates and finally dried in a furnace at 200 °C for at least 24 h. A scanning electron microscope (SEM, Zeiss, Oberkochen, Germany) image of the calcined and milled delafossite powder used for the AD process is shown in Figure 1. It can be seen that there is a broad particle size distribution ranging from 0.1 to 30 µm which is uncommon for AD processes. Bulk CuFeO2 samples were formed into brick shaped pellets, uniaxially cold pressed, and sintered at 1050 °C under the same gas atmosphere as the corresponding starting powder. In order to determine the thermoelectric properties, platinum/gold thermocouples and platinum wires were attached to the sintered samples with platinum conductor paste. Details of the setup are shown in Figure 2. The AD films were processed in a setup similar to previously published works [45][46][47][48]. It generally contains an aerosol generator, a deposition chamber and a vacuum pump (Edwards Germany, Kirchheim, Germany). In the deposition chamber and in the aerosol generator, a vacuum of 8 mbar is induced. Oxygen serves as a carrier gas at a flow rate of 6 L/min in the aerosol generator where an aerosol is created from the ceramic particles. These particles are transported through a slitnozzle with an orifice size of 10 × 0.5 mm² and accelerated up to several hundred m/s due to the pressure drop from the aerosol generator into the deposition chamber. The streaming aerosol is ejected on the target at a distance of 3 mm from the nozzle to the substrate and forms dense ceramic layers of several microns. For electrical measurements, AD films were deposited on alumina substrates (CeramTec, Marktredwitz, Germany) of a thickness of 635 µm, a length of 25 mm, and a width of 12.5 mm, on which screen-printed platinum/gold electrodes have been applied before. To obtain XRD patterns, silicon was used as substrate material (CrysTec, Berlin, Germany). The silicon wafers had an orientation of (911), exhibiting no silicon reflexes in the measured XRD diffraction angle range, avoiding substrate influences to the diffraction pattern.
To verify the phase composition of the starting powders and to elucidate the effect of the AD on the crystallography of CuFeO2, X-Ray diffraction patterns of both the calcined powder and the aerosol deposited films were taken at room temperature using a PANalytical Xpert Pro system (PANalytical, Almelo, Netherlands) operating with CuK radiation (1.541874 Å). The intensities were recorded within 2 = 25° .. 60° at a step size of 0.02°. The morphology of the AD films was examined by scanning electron microscopy images of both the cross section and the fracture pattern of the AD samples.   Figure 2 depicts the setup to determine the thermoelectric properties of bulk (a) and aerosoldeposited CuFeO2 (b). In both cases, the resistance is measured by a four probe technique with offset compensation (digital multimeter Keithley 2700). By knowing the geometry, the electrical conductivity can be calculated: In Equation (3), s is the spacing between the inner Pt electrodes, R is the measured resistance, b the width of the sample, and d the thickness of the pellet or the AD film, respectively. The latter was measured by a stylus profilometer (PGK/S2, Mahr, Göttingen, Germany).
To determine the Seebeck coefficient, S, an additional modulation heater in front of the samples generated an alternating temperature gradient over the specimens. The temperature difference between the thermocouples TC1 and TC2 was determined via the Au and Pt thermocouple tracks and contact pads, while the thermovoltage Umeas of the film was measured between the Pt contacts.
Since the Seebeck coefficients of Pt and Au, SPt and SAu, respectively, are known, the Seebeck The AD films were processed in a setup similar to previously published works [45][46][47][48]. It generally contains an aerosol generator, a deposition chamber and a vacuum pump (Edwards Germany, Kirchheim, Germany). In the deposition chamber and in the aerosol generator, a vacuum of 8 mbar is induced. Oxygen serves as a carrier gas at a flow rate of 6 L/min in the aerosol generator where an aerosol is created from the ceramic particles. These particles are transported through a slit-nozzle with an orifice size of 10ˆ0.5 mm² and accelerated up to several hundred m/s due to the pressure drop from the aerosol generator into the deposition chamber. The streaming aerosol is ejected on the target at a distance of 3 mm from the nozzle to the substrate and forms dense ceramic layers of several microns. For electrical measurements, AD films were deposited on alumina substrates (CeramTec, Marktredwitz, Germany) of a thickness of 635 µm, a length of 25 mm, and a width of 12.5 mm, on which screen-printed platinum/gold electrodes have been applied before. To obtain XRD patterns, silicon was used as substrate material (CrysTec, Berlin, Germany). The silicon wafers had an orientation of (911), exhibiting no silicon reflexes in the measured XRD diffraction angle range, avoiding substrate influences to the diffraction pattern.
To verify the phase composition of the starting powders and to elucidate the effect of the AD on the crystallography of CuFeO 2 , X-Ray diffraction patterns of both the calcined powder and the aerosol deposited films were taken at room temperature using a PANalytical Xpert Pro system (PANalytical, Almelo, Netherlands) operating with CuK α radiation (1.541874 Å). The intensities were recorded within 2θ = 25˝.. 60˝at a step size of 0.02˝. The morphology of the AD films was examined by scanning electron microscopy images of both the cross section and the fracture pattern of the AD samples. Figure 2 depicts the setup to determine the thermoelectric properties of bulk (a) and aerosol-deposited CuFeO 2 (b). In both cases, the resistance is measured by a four probe technique with offset compensation (digital multimeter Keithley 2700). By knowing the geometry, the electrical conductivity can be calculated: In Equation (3), s is the spacing between the inner Pt electrodes, R is the measured resistance, b the width of the sample, and d the thickness of the pellet or the AD film, respectively. The latter was measured by a stylus profilometer (PGK/S2, Mahr, Göttingen, Germany).
To determine the Seebeck coefficient, S, an additional modulation heater in front of the samples generated an alternating temperature gradient over the specimens. The temperature difference between the thermocouples TC1 and TC2 was determined via the Au and Pt thermocouple tracks and contact pads, while the thermovoltage U meas of the film was measured between the Pt contacts.
Since the Seebeck coefficients of Pt and Au, S Pt and S Au , respectively, are known, the Seebeck coefficient S of the delafossite film versus Pt can be determined from U meas . It has to be corrected by the known Seebeck coefficient of platinum, S Pt . Details of the evaluation of S can be found in [49] S " S Pt´U meas ∆T (4) A periodic voltage, U heater = U 0¨c os(2π¨ƒ mod,heater¨t )was applied to the modulation heater. It generated the temperature difference ∆T = T TC2´TTC1 with the frequency f mod : ∆T " ∆T 0¨c os p2π¨f mod¨t q Since heater power and applied modulation heater voltage show a quadratic relation, the temperature difference is modulated with the double frequency as the modulation heater voltage, i.e., f mod = 2f mod,heater [49]. In Equation (5), ∆T 0 is the amplitude of the temperature modulation, f mod stands for the frequency of the temperature modulation, and t is the time. U meas /∆T is determined by a regression analysis of many measured data pairs of the two signals ∆T j and U meas,i . They are plotted according to the following linear equation: The slope, a, represents the quotient U meas /∆T for Equation (6). This method allows elimination of interfering offset voltages. Further details of the data evaluation procedure and accuracies are given in [50]. For our experiments, f mod = 12.5 mHz was used, being low enough for our aerosol-deposited specimen to sustain a frequency-independent temperature gradient over the sample [51]. To circumvent interferences between the measurement of the electrical conductivity and the thermopower measurement, a custom-made switching device was used, enabling the automatic alternate measurement of both and electrically insulating them from each other.
In order to determine the influence of the oxygen partial pressure on the thermoelectric properties, the transducers were placed in a tube furnace and gas mixtures of oxygen and nitrogen were applied. The oxygen partial pressure was increased stepwise from 10´2 .6 bar, being the lower limit of the employed mass-flow-controller, to 1 bar and both the electrical conductivity and the Seebeck coefficient were measured during each pO 2 -step while each pO 2 measurement cycle was conducted at 700˝C, 800˝C, and 900˝C.

Characterization of the Synthesized CuFeO 2 Powders and AD Films
The crystal structure of CuFeO 2 was determined by XRD from the calcined powders. Figure 3 shows the pattern of CuFeO 2 fired in 0% O 2 (pure N 2 ) and the pattern of the 1% O 2 (1% O 2 , 99% N 2 ) fired powder together with the reference pattern (JPCD 39-0246). The characteristic diffraction peaks of CuFeO 2 can be observed in the pattern, indicating the rhombohedral 3R type with the R3;´m space group symmetry [52]. In order to determine the influence of the oxygen partial pressure on the thermoelectric properties, the transducers were placed in a tube furnace and gas mixtures of oxygen and nitrogen were applied. The oxygen partial pressure was increased stepwise from 10 −2.6 bar, being the lower limit of the employed mass-flow-controller, to 1 bar and both the electrical conductivity and the Seebeck coefficient were measured during each pO 2 -step while each pO 2 measurement cycle was conducted at 700 °C, 800 °C, and 900 °C.

Characterization of the Synthesized CuFeO2 Powders and AD Films
The crystal structure of CuFeO2 was determined by XRD from the calcined powders. Figure 3 shows the pattern of CuFeO2 fired in 0% O2 (pure N2) and the pattern of the 1% O2 (1% O2, 99% N2) fired powder together with the reference pattern (JPCD 39-0246). The characteristic diffraction peaks of CuFeO2 can be observed in the pattern, indicating the rhombohedral 3R type with the R3;¯m space group symmetry [52]. While the XRD pattern of CuFeO2 fired in 1% O2 appears free from secondary phases, a secondary phase can be seen for the 0% O2-fired CuFeO2 at 2θ = 43° (indicated in Figure 3 by #), accounting for elemental copper. We assume this metal impurity is related to the low oxygen content of N2 in the alumina tube furnace, resulting in a reduction of Cu2O to Cu: A similar behavior has already been reported by Zhao et al. [53] for delafossites calcined under Ar atmosphere. For further aerosol deposition of powders and for the measurements of the thermoelectric properties, CuFeO2 calcined in a mixture of 1% O2 in nitrogen was used to avoid traces of the above mentioned copper impurities. The lattice parameters were calculated by Rietveld analyses for Cu-delafossites fired in 1% oxygen to be a = 3.0341 Å and c = 17.169 Å, which corresponds to pure CuFeO2 data reported earlier [54]. While the XRD pattern of CuFeO 2 fired in 1% O 2 appears free from secondary phases, a secondary phase can be seen for the 0% O 2 -fired CuFeO 2 at 2θ = 43˝(indicated in Figure 3 by #), accounting for elemental copper. We assume this metal impurity is related to the low oxygen content of N 2 in the alumina tube furnace, resulting in a reduction of Cu 2 O to Cu: In order to determine the influence of the oxygen partial pressure on the thermoelectric properties, the transducers were placed in a tube furnace and gas mixtures of oxygen and nitrogen were applied. The oxygen partial pressure was increased stepwise from 10 −2.6 bar, being the lower limit of the employed mass-flow-controller, to 1 bar and both the electrical conductivity and the Seebeck coefficient were measured during each pO 2 -step while each pO 2 measurement cycle was conducted at 700 °C, 800 °C, and 900 °C.

Characterization of the Synthesized CuFeO2 Powders and AD Films
The crystal structure of CuFeO2 was determined by XRD from the calcined powders. Figure 3 shows the pattern of CuFeO2 fired in 0% O2 (pure N2) and the pattern of the 1% O2 (1% O2, 99% N2) fired powder together with the reference pattern (JPCD 39-0246). The characteristic diffraction peaks of CuFeO2 can be observed in the pattern, indicating the rhombohedral 3R type with the R3;¯m space group symmetry [52]. While the XRD pattern of CuFeO2 fired in 1% O2 appears free from secondary phases, a secondary phase can be seen for the 0% O2-fired CuFeO2 at 2θ = 43° (indicated in Figure 3 by #), accounting for elemental copper. We assume this metal impurity is related to the low oxygen content of N2 in the alumina tube furnace, resulting in a reduction of Cu2O to Cu: A similar behavior has already been reported by Zhao et al. [53] for delafossites calcined under Ar atmosphere. For further aerosol deposition of powders and for the measurements of the thermoelectric properties, CuFeO2 calcined in a mixture of 1% O2 in nitrogen was used to avoid traces of the above mentioned copper impurities. The lattice parameters were calculated by Rietveld analyses for Cu-delafossites fired in 1% oxygen to be a = 3.0341 Å and c = 17.169 Å, which corresponds to pure CuFeO2 data reported earlier [54].
A similar behavior has already been reported by Zhao et al. [53] for delafossites calcined under Ar atmosphere. For further aerosol deposition of powders and for the measurements of the thermoelectric properties, CuFeO 2 calcined in a mixture of 1% O 2 in nitrogen was used to avoid traces of the above mentioned copper impurities. The lattice parameters were calculated by Rietveld analyses for Cu-delafossites fired in 1% oxygen to be a = 3.0341 Å and c = 17.169 Å, which corresponds to pure CuFeO 2 data reported earlier [54].  Figure 4 depicts the XRD patterns of aerosol deposited CuFeO 2 on silicon substrates. No secondary phases or impurities were observed but the peaks got broader compared to the powder measurements arising from the reduction of the grain sizes during deposition. Based on the Rietveld refinement, the calculated mean grain size of AD CuFeO 2 was 90 nm compared to 300 nm for the calcined powder. This is a well-known effect in AD films and has been observed for many aerosol-deposited materials [47,55]. In addition, the relative peak intensities differ from the pattern of bulk and reference CuFeO 2 indicating high lattice strain of aerosol-deposited films, which is also a known phenomenon for aerosol-deposited materials [56,57].
Materials 2016, 9,227 6 of 16 Figure 4 depicts the XRD patterns of aerosol deposited CuFeO2 on silicon substrates. No secondary phases or impurities were observed but the peaks got broader compared to the powder measurements arising from the reduction of the grain sizes during deposition. Based on the Rietveld refinement, the calculated mean grain size of AD CuFeO2 was 90 nm compared to 300 nm for the calcined powder. This is a well-known effect in AD films and has been observed for many aerosoldeposited materials [47,55]. In addition, the relative peak intensities differ from the pattern of bulk and reference CuFeO2 indicating high lattice strain of aerosol-deposited films, which is also a known phenomenon for aerosol-deposited materials [56,57].  SEM cross-sectional images shown in Figure 5a and b indicate crack-free bulk CuFeO2 and dense layers of aerosol deposited CuFeO2 on alumina, respectively. The film thickness is around 25 µm. From the scanning electron microscope images shown in Figure 5c, the nano-sized microstructure of the aerosol deposited films becomes obvious. The primary particle size ranges from 50 nm to 100 nm, being consistent with the XRD analysis, while agglomerates 400 nm in size are embedded in the nanosized matrix. This inhomogeneous distribution of grain sizes is due to the particle size distribution of the starting powder for the ADM. While the CuFeO2 powders exhibit a d50 = 6.5 µm (the medium value of the particle size distribution), the d90 value (90 percent of the distribution lies below this value) of the particles is much larger (d90 = 16.1 µm). The film forming mechanism for AD layers is supposed to favor mid-range particles around 1 µm, so mainly these particles contribute to the layer formation. The larger particles of the aerosol stream may have less energy to form new ceramic layers and are therefore intercalated between the AD-formed ceramic planes. This phenomenon has also been observed for other aerosol-deposited materials [39,58,59]. SEM cross-sectional images shown in Figure 5a and b indicate crack-free bulk CuFeO 2 and dense layers of aerosol deposited CuFeO 2 on alumina, respectively. The film thickness is around 25 µm. From the scanning electron microscope images shown in Figure 5c, the nano-sized microstructure of the aerosol deposited films becomes obvious. The primary particle size ranges from 50 nm to 100 nm, being consistent with the XRD analysis, while agglomerates 400 nm in size are embedded in the nano-sized matrix. This inhomogeneous distribution of grain sizes is due to the particle size distribution of the starting powder for the ADM. While the CuFeO 2 powders exhibit a d 50 = 6.5 µm (the medium value of the particle size distribution), the d 90 value (90 percent of the distribution lies below this value) of the particles is much larger (d 90 = 16.1 µm). The film forming mechanism for AD layers is supposed to favor mid-range particles around 1 µm, so mainly these particles contribute to the layer formation. The larger particles of the aerosol stream may have less energy to form new ceramic layers and are therefore intercalated between the AD-formed ceramic planes. This phenomenon has also been observed for other aerosol-deposited materials [39,58,59].

Electrical Conductivity of Aerosol-Deposited and Bulk CuFeO2
In order to compare the power factors, PF (s. Equation (2)), of AD-processed CuFeO2 and standard ceramic-processed delafossites, both the electrical conductivity and the Seebeck coefficient were determined. Figure 6 shows the temperature dependency of the electrical conductivity of AD-CuFeO2 and bulk CuFeO2 as well as the activation energy of conduction. AD-processed samples show an offset in the electrical conductivity compared to bulk samples of almost one decade at room temperature, getting smaller with increasing temperature. This effect can be attributed to the microstructure of the deposited CuFeO2 films. While sintered bulk samples exhibit almost perfect grain interconnections, AD samples show regions of less densely connected grains. In addition, high strains, as they are common for the room temperature impact consolidation process, impede movements of the charge carriers and diminish the electrical conductivity [47]. With

Electrical Conductivity of Aerosol-Deposited and Bulk CuFeO 2
In order to compare the power factors, PF (s. Equation (2)), of AD-processed CuFeO 2 and standard ceramic-processed delafossites, both the electrical conductivity and the Seebeck coefficient were determined. Figure 6 shows the temperature dependency of the electrical conductivity of AD-CuFeO 2 and bulk CuFeO 2 as well as the activation energy of conduction.

Electrical Conductivity of Aerosol-Deposited and Bulk CuFeO2
In order to compare the power factors, PF (s. Equation (2)), of AD-processed CuFeO2 and standard ceramic-processed delafossites, both the electrical conductivity and the Seebeck coefficient were determined. Figure 6 shows the temperature dependency of the electrical conductivity of AD-CuFeO2 and bulk CuFeO2 as well as the activation energy of conduction. AD-processed samples show an offset in the electrical conductivity compared to bulk samples of almost one decade at room temperature, getting smaller with increasing temperature. This effect can be attributed to the microstructure of the deposited CuFeO2 films. While sintered bulk samples exhibit almost perfect grain interconnections, AD samples show regions of less densely connected grains. In addition, high strains, as they are common for the room temperature impact consolidation process, impede movements of the charge carriers and diminish the electrical conductivity [47]. With process, impede movements of the charge carriers and diminish the electrical conductivity [47]. With increasing temperature, the grains sinter as well as the microstrain releases, thus enhancing the electrical conductivity, a mechanism observed, e.g., for aerosol deposited MgB 2 [60].
Since both aerosol-deposited CuFeO 2 and bulk CuFeO 2 behave as though thermally activated, the electrical conductivity increases exponentially and can be described by Equation (8); hence, E a can be derived from the slope of the Arrhenius-like plot of the electrical conductivity as a function of the inverse temperature.
Both aerosol-deposited CuFeO 2 and bulk CuFeO 2 indicate a change in the activation energy. While the aerosol-processed sample exhibits a change from E a = 0.28 eV to E a = 0.38 eV at 200˝C, the bulk sample shows this transition behavior from E a = 0.24 eV to E a = 0.35 eV at 400˝C. The different transition temperature may be attributed to the microstructure of AD films mentioned above. The values as well as the change of activation energy are consistent with previously published work from Dordor et al. [61], where both single-crystals and polycrystalline samples of CuFeO 2 were investigated.
At temperatures above 800˝C, the electrical conductivity of both samples decreases abruptly, supposedly induced by a certain oxygen loss [37]. To study the origin of this conductivity decrease, the dependency of the electrical transport parameters conductivity (σ) and Seebeck coefficient (S) on the oxygen partial pressure (pO 2 ) was investigated at 900˝C for both aerosol deposited and bulk CuFeO 2 . Figure 7 shows a characteristic measurement cycle. Starting with a pure nitrogen gas atmosphere, the oxygen partial pressure, pO 2 , was increased stepwise. Compared to bulk CuFeO 2 , aerosol deposited samples respond much faster to pO 2 steps, promptly reaching an equilibrium state. Below an oxygen partial pressure of 31 mbar (3.1% oxygen), CuFeO 2 shows a p-type conduction behavior, as can be seen by the increasing conductivity with pO 2 . With increasing pO 2 , more oxygen is incorporated into the material, resulting in an increased hole concentration, resulting in an increasing electrical conductivity. Thus the σ (pO 2 ) measurement supports the assumption that the abrupt decrease of the electrical conductivity that occurs at 900˝C (displayed in the inset in Figure 6) may be attributed to a loss in oxygen. increasing temperature, the grains sinter as well as the microstrain releases, thus enhancing the electrical conductivity, a mechanism observed, e.g., for aerosol deposited MgB2 [60].
Since both aerosol-deposited CuFeO2 and bulk CuFeO2 behave as though thermally activated, the electrical conductivity increases exponentially and can be described by Equation (8); hence, Ea can be derived from the slope of the Arrhenius-like plot of the electrical conductivity as a function of the inverse temperature.
Both aerosol-deposited CuFeO2 and bulk CuFeO2 indicate a change in the activation energy. While the aerosol-processed sample exhibits a change from Ea = 0.28 eV to Ea = 0.38 eV at 200 °C, the bulk sample shows this transition behavior from Ea = 0.24 eV to Ea = 0.35 eV at 400 °C. The different transition temperature may be attributed to the microstructure of AD films mentioned above. The values as well as the change of activation energy are consistent with previously published work from Dordor et al. [61], where both single-crystals and polycrystalline samples of CuFeO2 were investigated.
At temperatures above 800 °C, the electrical conductivity of both samples decreases abruptly, supposedly induced by a certain oxygen loss [37]. To study the origin of this conductivity decrease, the dependency of the electrical transport parameters conductivity () and Seebeck coefficient (S) on the oxygen partial pressure (pO2) was investigated at 900 °C for both aerosol deposited and bulk CuFeO2.  Figure 7 shows a characteristic measurement cycle. Starting with a pure nitrogen gas atmosphere, the oxygen partial pressure, pO2, was increased stepwise. Compared to bulk CuFeO2, aerosol deposited samples respond much faster to pO2 steps, promptly reaching an equilibrium state. Below an oxygen partial pressure of 31 mbar (3.1% oxygen), CuFeO2 shows a p-type conduction Astonishingly, the conduction mechanism changes from p-type to n-type behavior at an oxygen partial pressure of 31 mbar, i.e., with increasing pO 2 the electrical conductivity decreases first sharply with a huge conductivity decrease by more than a half decade and then slightly at higher pO 2 . This effect is more distinctive for aerosol deposited samples, since the response time for the change in pO 2 is larger compared to bulk samples, not reaching a state of equilibrium. The double-logarithmic representation of the final values in Figure 8 accentuates this. Astonishingly, the conduction mechanism changes from p-type to n-type behavior at an oxygen partial pressure of 31 mbar, i.e., with increasing pO2 the electrical conductivity decreases first sharply with a huge conductivity decrease by more than a half decade and then slightly at higher pO2. This effect is more distinctive for aerosol deposited samples, since the response time for the change in pO2 is larger compared to bulk samples, not reaching a state of equilibrium. The double-logarithmic representation of the final values in Figure 8 accentuates this. For typical semiconducting oxides, the electrical conductivity depends on the oxygen partial pressure acc. to Equation (9): In a double-logarithmic plot, the prevalent defect mechanism may be deduced from the slope m. While typically slopes of m = +1/4 or m = −1/6, as they appear for the aerosol-deposited sample, can be explained by classical defect chemical means, see for instance [62][63][64], the slope for the bulk CuFeO2 samples can only be explained if one assumes that no equilibration has been settled, i.e., the final values are not equilibrium values.
The abrupt change of the conductivity at around 31 mbar cannot be explained by classical defect chemistry. Instead, we suggest a decomposition of delafossite-type CuFeO2 to the corresponding spinel phase CuFe2O4 and CuO, following Equation (10) 2 CuFeO2 + 1/2 O2  CuFe2O4 + CuO (10) According to the Ellingham diagram of CuFeO2, this phase change occurs at a pO2 = 30 mbar at 900 °C [65]. While CuFeO2 is a p-type semiconductor, CuFe2O4 is n-type, being in agreement with our conductivity vs. pO2 data [66]. Such a decomposition reaction could also explain the different distinct conductivity changes between bulk and aerosol deposited films. Since the bulk samples are considerably thicker, oxygen diffusion is by far slower, and a mixed phase consisting of CuFeO2 and CuFe2O4 as well as CuO may be present simultaneously. XRD measurements on samples that have been processed under 5% oxygen also support these assumptions since the XRD pattern clearly showed a mixed phase consisting of both CuFe2O4 and CuO (Figure 9). No CuFeO2 was found since the sample was exposed to the 5% O2 atmosphere for a long time (over several hours), so no evidence For typical semiconducting oxides, the electrical conductivity depends on the oxygen partial pressure acc. to Equation (9): In a double-logarithmic plot, the prevalent defect mechanism may be deduced from the slope m. While typically slopes of m = +1/4 or m =´1/6, as they appear for the aerosol-deposited sample, can be explained by classical defect chemical means, see for instance [62][63][64], the slope for the bulk CuFeO 2 samples can only be explained if one assumes that no equilibration has been settled, i.e., the final values are not equilibrium values.
The abrupt change of the conductivity at around 31 mbar cannot be explained by classical defect chemistry. Instead, we suggest a decomposition of delafossite-type CuFeO 2 to the corresponding spinel phase CuFe 2 O 4 and CuO, following Equation (10) been processed under 5% oxygen also support these assumptions since the XRD pattern clearly showed a mixed phase consisting of both CuFe 2 O 4 and CuO (Figure 9). No CuFeO 2 was found since the sample was exposed to the 5% O 2 atmosphere for a long time (over several hours), so no evidence on the transition phase could be obtained. In order to elucidate this mechanism in particular, measurements of the Seebeck coefficient were conducted.
Materials 2016, 9,227 10 of 16 on the transition phase could be obtained. In order to elucidate this mechanism in particular, measurements of the Seebeck coefficient were conducted.

Thermoelectric Properties of Aerosol Deposited and Bulk CuFeO2
Astonishingly, the Seebeck coefficient of bulk CuFeO2 is inferior compared to aerosol deposited CuFeO2 at low pO2. This discrepancy cannot be explained in the manner described for the electrical conductivity, since the thermopower is independent of the geometry (here the interconnection of grains and ceramic layers) and the reduced mobility caused by the high microstrains. Since this behavior is not fully understood, and to elucidate the change of the conduction mechanism from ptype to n-type at pO2 > 31.6 mbar, detailed measurements of the oxygen dependency of the thermopower were conducted. Figure 10 shows the Seebeck coefficient at 900 °C of both aerosol deposited CuFeO2 and bulk CuFeO2 as a function of oxygen partial pressure.

Thermoelectric Properties of Aerosol Deposited and Bulk CuFeO 2
Astonishingly, the Seebeck coefficient of bulk CuFeO 2 is inferior compared to aerosol deposited CuFeO 2 at low pO 2 . This discrepancy cannot be explained in the manner described for the electrical conductivity, since the thermopower is independent of the geometry (here the interconnection of grains and ceramic layers) and the reduced mobility caused by the high microstrains. Since this behavior is not fully understood, and to elucidate the change of the conduction mechanism from p-type to n-type at pO 2 > 31.6 mbar, detailed measurements of the oxygen dependency of the thermopower were conducted. Figure 10 shows the Seebeck coefficient at 900˝C of both aerosol deposited CuFeO 2 and bulk CuFeO 2 as a function of oxygen partial pressure. With increasing oxygen partial pressure, the Seebeck coefficient of bulk CuFeO2 declines slightly up to a pO2 of 31.6 mbar, thereafter dropping faster with a slope of −115 µV/K per decade pO2. It becomes even negative at pO2 = 1 bar (100% O2 in the gas), indicating n-type conductivity. Aerosol deposited CuFeO2 shows an always constant Seebeck coefficient of +425 µV/K up to an oxygen partial pressure of 31.6 mbar. However, in contrast to bulk CuFeO2, the transition from the p-type to n-type conductivity mechanism occurs sharper for aerosol-deposited CuFeO2, resulting in a negative Seebeck coefficient of S = −100 µV/K at pO2 = 0.1 bar, which persists at this value up to an oxygen partial pressure of 1 bar.
The changing sign of the Seebeck coefficient supports our assumption of a phase transition of CuFeO2 to CuFe2O4 and CuO with increasing oxygen partial pressure. Since oxygen equilibration kinetics of aerosol deposited CuFeO2 samples is much faster compared to bulk CuFeO2, the transition appears more pronounced, being completed within one measurement cycle, whereas bulk CuFeO2 supposedly exhibits a phase mixture of both CuFeO2 (p-type), CuFe2O4 (n-type) and CuO (p-type), resulting in ambiguous, bipolar thermoelectric effects. With two types of charge carriers present, the Seebeck coefficient of the material is the weighted average of the Seebeck coefficients associated to the different charge carriers as described by Equation (11): S = σ n S n +σ p S p σ n +σ p (11) with the Seebeck coefficients of the materials with different charge carrier types, Sn,p, and their electrical partial conductivities, σn,p, respectively [31]. Keeping in mind that the Seebeck coefficients of the n-type and p-type phases have opposite signs, the weighted Seebeck coefficient of a bipolar thermoelectric can be small compared to the purely n-type or p-type conducting materials. The measurements of the Seebeck coefficient of aerosol deposited CuFeO2 indicate that at a pO2 < 31.6 mbar the prevailing phase is CuFeO2 with a high thermopower of +425 µV/K. When increasing the pO2, bipolar effects occur in the transition region, due to the mixture of the decomposing CuFeO2 and the emerging CuFe2O4 and CuO phases. At high pO2, the transformation ends and the thermoelectric With increasing oxygen partial pressure, the Seebeck coefficient of bulk CuFeO 2 declines slightly up to a pO 2 of 31.6 mbar, thereafter dropping faster with a slope of´115 µV/K per decade pO 2 . It becomes even negative at pO 2 = 1 bar (100% O 2 in the gas), indicating n-type conductivity. Aerosol deposited CuFeO 2 shows an always constant Seebeck coefficient of +425 µV/K up to an oxygen partial pressure of 31.6 mbar. However, in contrast to bulk CuFeO 2 , the transition from the p-type to n-type conductivity mechanism occurs sharper for aerosol-deposited CuFeO 2 , resulting in a negative Seebeck coefficient of S =´100 µV/K at pO 2 = 0.1 bar, which persists at this value up to an oxygen partial pressure of 1 bar.
The changing sign of the Seebeck coefficient supports our assumption of a phase transition of CuFeO 2 to CuFe 2 O 4 and CuO with increasing oxygen partial pressure. Since oxygen equilibration kinetics of aerosol deposited CuFeO 2 samples is much faster compared to bulk CuFeO 2 , the transition appears more pronounced, being completed within one measurement cycle, whereas bulk CuFeO 2 supposedly exhibits a phase mixture of both CuFeO 2 (p-type), CuFe 2 O 4 (n-type) and CuO (p-type), resulting in ambiguous, bipolar thermoelectric effects. With two types of charge carriers present, the Seebeck coefficient of the material is the weighted average of the Seebeck coefficients associated to the different charge carriers as described by Equation (11): S " σ n S n`σp S p σ n`σp (11) with the Seebeck coefficients of the materials with different charge carrier types, S n,p , and their electrical partial conductivities, σ n,p , respectively [31]. Keeping in mind that the Seebeck coefficients of the n-type and p-type phases have opposite signs, the weighted Seebeck coefficient of a bipolar thermoelectric can be small compared to the purely n-type or p-type conducting materials. The measurements of the Seebeck coefficient of aerosol deposited CuFeO 2 indicate that at a pO 2 < 31.6 mbar the prevailing phase is CuFeO 2 with a high thermopower of +425 µV/K. When increasing the pO 2 , bipolar effects occur in the transition region, due to the mixture of the decomposing CuFeO 2 and the emerging CuFe 2 O 4 and CuO phases. At high pO 2 , the transformation ends and the thermoelectric measurements indicate the prevailing n-type CuFe 2 O 4 phase. For bulk CuFeO 2 , this effect arises much more slowly, resulting in a broader bipolar transition region, and the transformation is not finished at high pO 2 within the measurement cycle, resulting in a bipolar thermopower and a Seebeck coefficient of´15 µV/K compared to´120 µV/K for aerosol-deposited CuFeO 2 at pO 2 = 1 bar. In fact, it is believed that the bulk sample with a thickness of 500 µm does not reach an equilibrium within half an hour. If one assumes an oxygen kinetic that is diffusion-controlled, one finds equilibration kinetics to be proportional to the square of the thickness of the smallest geometry. In other words, the equilibration kinetics of the AD sample should be faster by a factor of (d bulk,sample /d AD sample ) 2 « 20 2 « 400. Hence, both the thermopower and the conductivity values of the bulk samples appear to be nonequilibrium values and therefore always lie "between" the AD curves. Nevertheless, since the detailed process of aerosol deposition has not yet been fully understood, the consequences of the room temperature impact consolidation on the thermoelectric properties, especially the diverging Seebeck coefficient of bulk and AD processed samples at a pO 2 < 31.6 mbar, remains an open-ended question for further investigations.
Being of interest as high temperature thermoelectric material, the electrical conductivity and Seebeck coefficients were investigated at temperatures up to 900˝C. Figure 11 shows the power factor (PF) of both aerosol deposited CuFeO 2 and standard processed bulk CuFeO 2 , exhibiting a maximum of PF = 59 µW/(K²¨m) at T = 800˝C for aerosol deposited CuFeO 2 and PF = 130 µW/(K²¨m) for bulk CuFeO 2 , featuring the same magnitude like other oxide thermoelectrics, e.g., Ca 3 Co 4 O 9 with PF = 225 µW/(K²¨m) or PF = 810 µW/(K²¨m) for doped Na x CoO 2 [20]. measurements indicate the prevailing n-type CuFe2O4 phase. For bulk CuFeO2, this effect arises much more slowly, resulting in a broader bipolar transition region, and the transformation is not finished at high pO2 within the measurement cycle, resulting in a bipolar thermopower and a Seebeck coefficient of −15 µV/K compared to −120 µV/K for aerosol-deposited CuFeO2 at pO2 = 1 bar. In fact, it is believed that the bulk sample with a thickness of 500 µm does not reach an equilibrium within half an hour. If one assumes an oxygen kinetic that is diffusion-controlled, one finds equilibration kinetics to be proportional to the square of the thickness of the smallest geometry. In other words, the equilibration kinetics of the AD sample should be faster by a factor of (dbulk,sample/dAD sample)² ≈ 20² ≈ 400. Hence, both the thermopower and the conductivity values of the bulk samples appear to be nonequilibrium values and therefore always lie "between" the AD curves. Nevertheless, since the detailed process of aerosol deposition has not yet been fully understood, the consequences of the room temperature impact consolidation on the thermoelectric properties, especially the diverging Seebeck coefficient of bulk and AD processed samples at a pO2 < 31.6 mbar, remains an open-ended question for further investigations.

Conclusions
In the present study, the novel aerosol deposition method (ADM) was successfully employed to fabricate dense and crack-free ceramic layers of several microns from the undoped p-type thermoelectric CuFeO2 at room temperature with no further heat treatment, thus avoiding interactions with the substrate or the influence of sinter additives. By employing the aerosol deposition method, measurements could be performed on very thin films enabling very fast

Conclusions
In the present study, the novel aerosol deposition method (ADM) was successfully employed to fabricate dense and crack-free ceramic layers of several microns from the undoped p-type thermoelectric CuFeO 2 at room temperature with no further heat treatment, thus avoiding interactions with the substrate or the influence of sinter additives. By employing the aerosol deposition method, measurements could be performed on very thin films enabling very fast responses. Since the oxygen partial pressure plays a decisive role during the synthesis and application of Delafossites, XRD studies confirmed that a lowly oxidizing calcination atmosphere is essential for the preparation of single phase CuFeO 2 . The process window, however, is small since at higher oxygen partial pressures, pO 2 > 30 mbar at 900˝C, a phase transition from CuFeO 2 to the spinel-type CuFe 2 O 4 and CuO occurs.
Astonishingly, we observed a sudden change of conduction from p-type to n-type at an oxygen partial pressure of pO 2 = 30 mbar. While the electronic structure of CuFeO 2 can be calculated by an enhanced local spin density approximation [67], this change in the conduction mechanism at a defined oxygen partial pressure has not been observed yet. Investigations on changing valance states of the copper and iron sites in CuFeO 2 were also conducted in order to establish a defect chemical model [68]. However, we propose that the change is based, for instance, (at least partly) upon a phase transition from p-type semiconducting CuFeO 2 to n-type CuFe 2 O 4 and CuO, resulting in a bipolar thermoelectric material. While the thermoelectric properties of the n-type phase are inferior to the p-type CuFeO 2 , this material system can be of interest for use in thermoelectric generators, since both p-type and n-type materials can be precisely tailored only by defined process conditions based on the identical starting thermoelectric material. Nevertheless, detailed defect chemical investigations, particularly more measurements of electric transport parameters combined with other non-electrical analytical means, need to be conducted at defined and especially low-oxygen partial pressures in order to develop a comprehensive defect model of CuFeO 2 . The measurements shown in this study may serve as an initial basis. Furthermore, the influence of dopants needs to be studied to tailor the thermoelectric properties, and detailed measurements on the thermal conductivity of thin aerosol-deposited films deserve further investigation since the reduction in grain size, resulting from the room temperature impact consolidation effect, could lead to a reduction of the thermal conductivity of CuFeO 2 , probably due to increasing phonon scattering at grain boundaries thereby increasing the thermoelectric performance of delafossites.