Ga-doped Ca12Al14O33 mayenite oxide ion conductors: synthesis, defects, and electrical properties

Although mayenite Ca12Al14O33 has been known as an oxygen ion conductor for several decades, its relatively low oxide ion conductivity limits its applications in electrochemical devices. Thus, many efforts have been made by researchers, employing a doping strategy, in order to further improve its ionic conductivity, but with little success. In this work, a series of pure phase Ca12Al14−xGaxO33+δ (0 ≤ x ≤ 1.2) materials were synthesized by a traditional solid state reaction method. Scanning electron microscopy (SEM) combined with energy dispersion spectrum (EDS) analyses disclosed well-sintered ceramics with uniform Ga distributions. The defect formation energies for Ga replacing the two distinguishable Al1 and Al2 sites in Ca12Al14O33 calculated by static lattice atomistic simulation are nearly identical, ∼3.03 and ∼3.04 eV, respectively, consistent with the results of Rietveld refinements based on the XRD data, from which no preferred distribution of Ga on Al1 or Al2 site was observed. The electrical properties investigated by alternating current (AC) impedance spectroscopy show increased bulk conductivities for 0 ≤ x ≤ 0.4. Thus, here we present the first work that successfully improves the bulk oxide ion conductivity of Ca12Al14O33 by Ga-doping.


Introduction
Solid oxide fuel cells (SOFCs) have attracted great attention over the last few decades, due to the clean energy conversion technology associated with high efficiency and fuel exibility. [1][2][3] The oxide ion conducting electrolyte plays a key role in the working temperature of a SOFC. [4][5][6] The yttrium-stabilized ZrO 2 (YSZ), as the traditional and most widely used electrolyte in industry, can be applied only at a temperature higher than 750 C. [7][8][9] This high-working temperature raises issues concerning undesired reactions between the electrolyte and electrode materials, and also creates thermal stresses during thermal cycling. Thus, the development of new oxide ion conductors with considerably high conductivity for use in intermediate temperature (500-750 C) SOFCs is an urgent and pressing need. [9][10][11][12] The oxide ion conductivity of aluminate Ca 12 Al 14 O 33 , a stable ceramic material, was rst identied by M. Lacerda et al. in 1988. 13 M Lacerda et al. reported that the oxide ion conduction ($ 1.5 Â 10 À3 S cm À1 ) of Ca 12 Al 14 O 33 materials was only slightly lower than that of YSZ, making Ca 12 Al 14 O 33 a competitive candidate for SOFC electrolyte, if the oxide ion conductivity could be further improved through metal cationic doping strategy. The parent Ca 12 Al 14 O 33 adopts a cubic structure with the I 43d space group and a lattice constant of $11.99Å. There are two Ca 12 Al 14 O 33 molecules in the unit cell which is composed of a positively charged framework built from 12 cages and two free O ions randomly occupying 2 different cages. Thus, it can also be represented by the chemical formula [Ca 24 Al 28 O 64 ] 4+ $2O 2À . The cages, composed of framework Ca, Al (6-coordinated Al1 and 4-coordinated Al2) and O atoms, are approximately 6Å wide and connected to 8 other cages via $3.7Å wide windows (Fig. 1). The free oxygen ions possess high mobility, which makes the un-doped material somewhat of an oxide-ion conductor. The oxygen diffusion in un-doped mayenite had been examined by neutron diffraction, 14,15 DFT and MD calculations 16,17 and oxygen isotope exchange experiments. 18,19 The hightemperature neutron diffraction study on the un-doped Ca 12 Al 14 O 33 suggested that the 'free' oxygen ions are most likely transported via a jump-like process involving exchange of the 'free' oxygen with framework oxygen. 14 This was consistent with the theoretical prediction reported by Sushko et al. 17 Irvine and West et al. made the rst attempts at improving the high-temperature conductivity by zinc doping, as well as with zinc and phosphorus co-doping on Al atom sites. 20 They found that for the zinc-only series compositions, replacement of aluminium by zinc would cause a decrease in conductivity ($6.5 Â 10 À4 S cm À1 at 700 C) and an increasing in activation energy. Similar electrical behaviors were also found for the zinc and phosphorus co-doped compositions. Later, G. Ebbinghaus et al. studied iron-doped single crystal mayenite in which they observed slight decreases in conductivity ($1.26 Â 10 À3 S cm À1 at 700 C, while the conductivity of the undoped single crystal Ca 12 Al 14 O 33 is $1.78 Â 10 À3 S cm À1 ) for the doped samples. 21 Other works have included copper, 22,23 nickel, 24 gallium, 25 manganese, 26 and iridium 27 incorporation into mayenite, but did not examine the effect of these dopants on the conductivity. Some other cations doped materials, such as Bi 3+ and Ln 3+ (Tb 3+ /Sm 3+ /Er 3+ /Nd 3+ /Yb 3+ /Ho 3+ /Pr 3+ ) substituting for Ca ions, were also reported and focused mainly on the luminescent properties, 28,29 without oxide ion conductivities being investigated.
Besides metal cations doping, substituting the free oxygen ions with other anions such as F À , OH À , O 2 À , H À , O À , and so on, [30][31][32][33][34][35][36] had also been reported with some interesting properties. In addition, electrons can occupy the empty space inside each cage in a similar manner, forming electride materials. However, although these anions or just electrons substituted Ca 12 Al 14 O 33based materials may have much higher conductivity, such as the H À introduced and photo-activated material had an electrical conductivity as high as 0.3 S cm À1 (mainly n-type electronic conduction) at room temperature, the oxide ion conduction was not improved, and therefore did not benet its application in SOFCs as electrolyte. These anions or electrons substituted Ca 12 Al 14 O 33 -based materials are thus out of the scope of our study interesting. As mentioned above, the gallium doped Ca 12 Al 14 O 33 materials have been previously reported by Luis Palacios et al. with the structures and reduction behaviors being studied. 25 Through the Rietveld renements based on the combined neutron and X-ray powder diffraction data, unit cell expansions were observed for these doped materials and the Ga ions were reported to mainly occupy the 4-coordinated Al2 sites. In order to get Ga-doped Ca 12 Al 14 O 33 electrides, they red these doped materials under a strong reducing condition, but resulted nally in decompositions and forming a mixture of Ca 12 Al 14 O 33 , Ca 3 Al 2 O 6 , and amorphous Ga metal. In this study we focused on investigating the effects of Ga doping on the phase, defect formation energy, structure and electrical properties of Ca 12 Al 14 O 33 by X-ray diffraction, SEM/EDS, static lattice atomistic simulation, and AC impedance spectroscopy techniques. The results show that Ga ions can be substituted for up to about ten percent of Al ions in the crystal structure, similar with that reported in Luis Palacios et al.'s work. However, we found that both the Rietveld renements based on the high-resolution XRD data and the defect formation energy calculations, did not suggest a preferred occupation for Ga ions replacing the two distinguishable Al1 and Al2 sites. This is different from that reported by Luis Palacios et al. The bulk electrical conduction of Ca 12 Al 14Àx Ga x O 33+d was increased for Ga content in the range of 0 # x # 0.4, aer which a decrease in the conductivity was observed for x > 0.4.

Methods
The samples of Ca 12 Al 14Àx Ga x O 33+d were prepared by traditional solid-state reaction method using stoichiometric amounts of CaCO 3 (Alfa Aesar, > 99.8% purity), Al 2 O 3 (Alfa Aesar, > 99.997% purity) and Ga 2 O 3 (Alfa Aesar, > 99.99% purity) as the starting raw materials. First, the well-mixed and ground raw materials were red at 1000 C for 12 h. Aer regrinding, the calcined powders were then uniaxially pressed into pellets and sintered at 1300 C for 24 hours in an air atmosphere. The densities of these prepared ceramics were estimated from the samples' weight and geometry.
The XRD data were collected on a Panalytical X'pert Pro X-ray diffractometer with Cu Ka radiation over a 2q range of 5À120 . The variable temperature XRD measurements were performed over a temperature range 25-900 C with the 2q range of 10À80 C. The Rietveld renements of the XRD data were carried out using Topas-Academic soware. 37 The microstructure and EDS analyses were performed on a Hitachi (Tokyo, Japan) S4800 scanning electron microscope (SEM). Before the SEM/EDS measurements, all these as-made ceramic pellets were well polished, followed with thermal etched. AC impedance spectroscopy (IS) measurements were performed with a Solartron 1260 frequency response analyzer over a 10 7 to 10 À1 Hz frequency range. Prior to the IS measurements, electrodes were formed by coating platinum paste on opposite faces of the pellets and red at 750 C for 3 h to remove any organic components. Before the impedance measurements, the temperature was equilibrated at each set point for 1 hour.
The energies of Ga 3+ ions substituting for Al 3+ ions were investigated through atomistic-static-lattice simulation, using the General Utility Lattice Program (GULP) 38,39 based on interatomic potential approach. In this work, the Buckingham potential function 40 was used to model the interaction between ions with the shell model 41 to describe the electronic polarizability for the structural modeling. The interatomic potential parameters used for the atomistic simulation are listed in Table 1. The linear increase for x # 1.2 can be explained by the larger effective ionic radius of Ga 3+ ions (0.47Å or 0.62Å in a 4-coordinated or 6-coordinated environment, respectively) than that of Al 3+ ions (0.39Å or 0.53Å in a 4-coordinated or 6-coordinated environment, respectively). 42 Rietveld structure renements were subsequently carried out on these doped materials. For these Rietveld renements, a parent cubic Ca 12 Al 14 O 33 structure (space group I 43d) containing two Ca sites (24d), one 3-linked and one 4linked Al site (16c and 12a, respectively, with the numbers 3 and 4 denoting the oxygen number of a AlO 4 tetrahedron that cornershared with other AlO 4 tetrahedra) and three oxygen sites (one 16c and two 48e) was used as a model. Here, a typical Rietveld renement was described for the x ¼ 1.2 composition which has the highest nominal Ga content, while renements for other single phase compositions were provided in the ESI of this paper (Fig. S1-S6 and Tables S1-S6). † The renement was conduct with the background, lattice parameters, peak shape parameters rened in sequence. This was followed by the positional parameters and then isotropic thermal vibration parameters being rened, which resulted in an acceptable agreements between the measured and calculated diffraction patterns. Finally, the ratios of Ga and Al on the 6-coordinated and 4coordinated positions were varied, subject to the constraint that both sites remained fully occupied. The nal tted plot and structural parameters are given in Fig. 2c and Table 2, respectively. It can be seen that the Ga ions show essentially a random distribution among the Al1 and Al2 sites, which is different from the case reported by Luis Palacios et al. that the Ga ions shown a preferred occupation on the 4-coordinated position. The rened chemical formula of Ca 12 Al 12.784(2) Ga 1.216(2) O 33+d agrees well with the nominal composition of Ca 12 Al 12.8 Ga 1.2 O 33 . Fig. 3a demonstrates the typical SEM micrograph image of the composition x ¼ 1.2, conrming the dense structure of this prepared ceramic. The grain size ranges from 2 mm to 5.0 mm, with irregular morphology. Fig. 3b-d display the EDS element distribution maps of elements Ca, Al, and Ga, respectively. We can see that all these elements show homogeneous distributions, indicating the pure phase nature of the ceramic. This is consistent with the results from XRD data analysis. Fig. 3e shows the element concentrations and reveals a relative ratios for Ca : Al : Ga to be 12 : 13.91 : 1.4, close to the nominal ratios. The SEM and EDS results for other compositions are provided in Fig. S7-S12. †

Static lattice simulation
As stated previously, no more than 10 percent of the Al ions can be replaced by Ga ions. This narrow solid solution usually corresponds with the relatively high defect formation energy for substitution. To conrm this, static lattice simulation technology was used to calculate defect formation energies based on  the appropriate combination of dopant and vacancy defect energies and lattice energies of the binary oxides, and can be determined using the following defect equation: i.e., where E(X) is the calculated total lattice energy or point defect energy of the species of interest aer geometry optimization.
The starting point of the study was to reproduce the experimentally observed crystal structures of Ca 12 Al 14 O 33 , Ga 2 O 3 , and Al 2 O 3 . Using the interatomic potentials presented in Table 1 for simulation, the differences between the calculated and experimental unit cell edges and volumes for all three oxides were found to be less than 4%, as demonstrated in Table 3, validating the rationality of these interatomic potentials used for simulations.
Next, point defect energies for Ga ions replacing both Al1 and Al2 site ions were calculated. Combining the total lattice energies of the interested binary oxides Ga 2 O 3 and Al 2 O 3 , the defect formation energies for Ga ions replacing Al ions can be deduced from eqn (2). These values are summarized in Table  4. It can be clearly seen that the defect formation energies for Ga ions replacing both the Al1 site and Al2 site ions are higher than 3.0 eV, consistent with the relatively narrow solid solution. In addition, the almost same defect formation energies on Al1 and Al2 sites agrees well with the randomly occupying and comparable probability of Ga atoms on the Al1 and Al2 sites, as derived from the Rietveld renements based on the XRD data.

Electrical properties
The electrical properties of the Ca 12 Al 14Àx Ga x O 33+d ceramic samples were investigated by AC impedance spectroscopy. Compared to four electrode technique that commonly used in the direct current (DC) conductivity measurements, the alternating current (AC) impedance spectroscopy measurements have the advantage to separate the resistance related to grains (bulk) and grain boundaries due to their different relaxation times, leading to separate semicircles in the complex impedance spectrum. Therefore, the bulk conductivity (s b ), grain boundary conductivity (s b ), and the total conductivity (s t ) can be well calculated from the bulk resistance (R b ), grain boundary resistance (R g ), and total resistance (R t ¼ R b + R g ). The densities of these prepared ceramics, estimated from the samples' weight and geometry, were all higher than 92% of the theoretical prediction. The AC impedance measurements were performed within a temperature range of 400À900 C. The results showed that the conductivity of Ca 12 Al 14Àx Ga x O 33+d ceramic samples did not vary linearly with the doped Ga content. In fact, the conductivity increased within the range 0 # x # 0.4, as shown in Fig. 4a, and then began to decrease when x $ 0.6. For x $ 0.6 the values are plotted separately in Fig. 4b for clarity. Thus, the highest conductivity was observed for the composition x ¼ 0.4, with a value of $ 2.76 Â 10 À3 S cm À1 at 800 C. For the composition x ¼ 0.6, although the conductivity was lower than that of x ¼ 0.4 but still higher than that of the parent material Ca 12 Al 14 O 33 . For x ¼ 0.8, the measured conductivity was almost the same as that of the un-doped material. Conductivity lower than the un-doped Ca 12 Al 14 O 33 was observed for samples with x ¼ 1.0, and the lowest measured conductivity of $6.96 Â 10 À4 S cm À1 occurred for the x ¼ 1.2 sample. This value, measured at 800 C, is about four times lower than the sample with composition x ¼ 0.4. A more direct visual comparison of the conductivities of the Ca 12 Al 14Àx Ga x O 33+d ceramic samples is seen in Fig. 4c where the conductivity values are plotted as a function of x at a temperature of 800 C. In Fig. 4c the trend of conductivities with x values can be clearly seen.
For all these Arrhenius plots displayed in Fig. 4a and b, one can see that there are different slops for each plot at low (#700 C) and high (>700 C) temperature, respectively. The change in slop at different temperature range, however, is not caused by a phase transition, which can be validated by the in situ VT-XRD measurements (Fig. S13 †) performed on the sample x ¼ 0.4. The VT-XRD patterns show a single mayenite phase over the measured temperature range of 25-900 C. Therefore, the slope change may originate from the variation of local defect structure that surrounding the charge carriers, in different temperature range.

Conclusion
In this work, a series of Ga-doped Ca 12 Al 14Àx Ga x O 33+d (0 # x # 1.4) materials were prepared by a conventional solid state method. Pure mayenite phases were obtained for 0 # x # 1.2, as was conrmed by Rietveld renements based on the XRD data, and also by combined SEM and EDS element distribution map and element concentrations analyses. Static lattice atomistic simulation was used to calculate the defect formation energies of Ga ions replacing Al ions. The simulation values of 3.03 eV and 3.04 eV for the defects on Al1 and Al2 sites, respectively, were consistent with the non-preferred occupation of Ga on Al1 or Al2 sites, as derived from Rietveld renements. The bulk conductivities increased with Ga content for 0 # x # 0.4. In summary, we believe that this paper presents the rst example of successful improvement of the bulk oxide ion conductivity of Ca 12 Al 14 O 33 through Ga-doping.

Conflicts of interest
There are no conicts of interest to declare.