New Synthesis Route for Complex Borides; Rapid Synthesis of Thermoelectric Yttrium Aluminoboride via Liquid-Phase Assisted Reactive Spark Plasma Sintering

YxAlyB14 ceramics are of high interest as high temperature thermoelectric materials with excellent p, n control. In this study, direct synthesis of dense polycrystalline YxAlyB14 (x ~0.64, 0.52 ≤ y ≤ 0.67) ceramics was successfully carried out by spark plasma sintering using commercially available precursors. YB4, AlB2 and B powders were reactively sintered with an additive AlF3 at 1773 K for 5–60 min in reduced Ar atmosphere. The sinterability was remarkably enhanced by liquid phase sintering comparing to conventional synthesis techniques. Phase composition analysis by X-ray diffraction showed that main peaks belong to YxAlyB14 with the MgAlB14 structure type and no peaks of AlF3 were detected. The thermoelectric behavior was changed from p-type to n-type with increasing Al occupancy. Power factor and ZT values measured in this study were found to be in the same range as the best values previously reported. This original synthesis process is found to be less precursor-consuming as compared to previous synthesis processes, and strikingly, less time-consuming, as the synthesis time, is shortened from 8 h to 5 min for p-type and to 1 h for n-type. The total process time is shortened from ≥3 days to ~4–5 h. This discovery opens the door for more accessible synthesis of complex borides.


Results and Discussion
show the XRD patterns measured on the surface of YA0 composition samples sintered with and without AlF 3 at 1773 K, respectively. The XRD pattern of sample sintered for 10 min without AlF 3 exhibits numerous peaks of secondary phases such as unreacted YB 4 , YB 6 , YB 12 , α-alumina, and so on, as shown in Fig. 2. The peaks of secondary phases are still detected on the YA0 composition sample which were even sintered for 1 h. It means that the reaction for synthesis of Y x Al y B 14 is independent on sintering time in this case.  42 , which is orthorhombic structure belonging to Imma space group. The boron atoms B1, B3, B4 and B5 form B 12 icosahedra structures, which are interconnected via external B-B bonds or via B2 atom. Y and Al atoms partially occupy interstitial positions in the rigid three-dimensional B network.
In contrast, the XRD patterns of the same composition samples sintered with AlF 3 at the same sintering condition exhibit relatively much less peaks of secondary phases, and no peak of AlF 3 was detected. It implies that AlF 3 dramatically improved the reactivity among starting materials by liquid phase sintering, and then was completely removed without remaining by evaporation under the reduced Ar atmosphere. The relatively purest Y x Al y B 14 was obtained for the sample sintered for 5 min under the uniaxial pressure of 80 MPa with AlF 3 (Fig. 3). Although some peaks of α-alumina and YB 6 phases are still detected as in the previous conventional synthesis studies 43,44 , polycrystalline Y x Al y B 14 was successfully synthesized via reactive SPS with AlF 3 .
The axial displacement changes of the YA0 composition samples during SPS are shown in Fig. 4. The temperature at the minimum displacement indicates the starting initial stage of sintering. In the case of the sample sintered without AlF 3 under ~30 MPa which is the minimum pressure of SPS, the shrinkage was started at ~1550 K but densification was not finished even at 1773 K. On the other hand, in the case of the sample sintered with AlF 3 under ~30 MPa, the shrinkage was started at ~1450 K and densified well during heating process. This result suggests that the sinterability and densification behavior were remarkably improved by AlF 3 . The melting point of AlF 3 is ~1573 K at 1 atm. This temperature is higher than that for the starting shrinkage (~1450 K). It is therefore  www.nature.com/scientificreports www.nature.com/scientificreports/ considered that the reduced-pressure Ar atmosphere lowered the melting point of AlF 3 , and the initial stage was shifted earlier, namely ~100 K lower by formation of liquid phase of AlF 3 .
The densification behavior of the YA0 composition sample sintered with AlF 3 under uniaxial pressure of 80 MPa was further enhanced. The shrinkage of the pressure-assisted sample was started from ~1360 K, and the shrinkage rate was changed from ~1450 K. Since the temperature where shrinkage rate was changed is the same as the temperature of the starting initial stage of the sample sintered with AlF 3 under ~30 MPa, the shrinkage observed from 1360 to 1450 K indicates the effect of the applied pressure. All samples exhibited changes of shrinkage rate during the intermediate stage. These phenomena can be mainly attributed to the different shrinkage rate between the main phase and the secondary phases. Figure 5 shows XRD patterns of Y 0.62 Al y B 14 samples (YA0-3; y = 0.71-1.86) sintered at 1773 K with AlF 3 for 1 h under ~30 MPa. Together with the prominent peaks of Y x Al y B 14 , weak peaks of several secondary phases such as α-alumina, σ-alumina, YB 4 , YB 6 , YB 12 and SiO 2 were detected. The Al-poor sample (YA0) exhibited peaks of α-alumina, YB 6 , YB 12 and SiO 2 , whereas the Al-rich samples (YA1, YA2 and YA3) exhibited peaks of σ-alumina in addition. The structure of σ-alumina (Al 2.667 O 4 ) is described as a defect cubic spinel in which Al cations randomly and partially occupy its octahedral and tetrahedral sites 57 . The result thus implies that addition of Al and reductive atmosphere affects formation of σ-alumina during sintering of Al-rich samples. For Al-rich samples, the intensity of the detected YB 6 and YB 12 peaks decreased with increasing Al content. It indicates that the excess Al compensated the Al loss caused by the formation of alumina phase during sintering. Relatively homogeneous yttrium aluminoborides were obtained for the YA3 composition. The XRD pattern for the YA2 sample shows the presence of YB 4 . It is attributed to inhomogeneous mixing of starting materials. To obtain Y x Al y B 14 without secondary phases, such as SiO 2 , YB 4 , YB 6 and YB 12 , optimization of the mixing process is required. We also found out that the SiO 2 in the XRD patterns of the samples came from the agate mortar, which was used to pulverize each  sample before powder XRD measurement. This was indicated because the intensity of the peak at 2θ = 26-27° increased with increasing pulverization time.
The fractional occupancies of x and y, of Y and Al, in the Al-poor (YA0-P) and -rich samples (YA1, YA2 and YA3) prepared by reactive SPS with AlF 3 were estimated from Rietveld refinement within the technique limit ( Fig. 6). At first starting atom position from Korsukova et al. 42 model was used to perform refinements. Ultimately, atom positions were refined, together with partial occupancies for Y and Al positions. The synthesis conditions and the results of Rietveld refinement of each sample are summarized in Table 1 and Table 2, respectively. The refined Y contents (x in Y x Al y B 14 ) of all samples are ~0.64, which is almost the same value as that of the starting materials. This result implies that no Y loss during reactive SPS occurred. The refined Al content (y in Y x Al y B 14 ) of YA0-P is 0.52 though the starting content was 0.71. It suggests that Al was evaporated during sintering due to high sintering temperature and reduced Ar atmosphere. Meanwhile, the refined Al contents of YA1, YA2 and YA3 are ~0.67, which are much less than their initial Al contents (1.24, 1.55 and 1.86). Since these compositions are quite similar to those determined from single crystal x-ray diffractometry Y 0.62 Al 0.71 B 14 42 , it is considered that the Y and Al atoms occupy their interstitial sites as maximumly as possible to satisfy the 4 electron deficient nature 53,54 , and thus excess Al could not occupy Al sites.
The phase fractions and the theoretical and relative density of each sample are given in Table 3. At around 5%, the largest impurity phase is the alumina phase. We have tried to remove it but intrinsically find that it is difficult. As raw materials, amorphous boron is used for many sintering synthesis of complex borides which are difficult to synthesize by simple arc melting, for example 54 . There is some oxidation of the sensitive amorphous boron and in combination with aluminum for the present compound, alumina is formed. We tried to remove the oxygen  www.nature.com/scientificreports www.nature.com/scientificreports/ by heating the amorphous boron and evaporation of BO, but found that handling in air, the amorphous boron is easily oxidized again. Therefore, in this case, the alumina impurity phase is intrinsically difficult to eliminate. However, together with metallic YB 6 , etc. which typically appear since mixing is generally not ideal, it is indicated that these secondary phases do not have a significant effect on the thermoelectric properties, as discussed below.
The temperature dependence of electrical conductivities for Al-poor (YA0-P) and -rich samples (YA1, YA2 and YA3) prepared by reactive SPS is shown in Fig. 7. At the low temperature region, the Al-poor sample exhibits extremely low electrical conductivity compared to Al-rich samples. However, a particularly large increase of electrical conductivity is observed with increasing temperature due to its conducting mechanism which will be described below. On the other hand, all the Al-rich samples exhibit several orders higher electrical conductivities. The highest electrical conductivity values are obtained for YA3 sample which is ~5 times higher than that of the YA1 sample. We consider the possibilities of the secondary phases having any effect on the thermoelectric properties. Amongst the thermoelectric properties, the electrical conductivity can be considered to have the potential to be most sensitive to small amounts of metallic phases or insulating phases, considering that these phases may have orders different resistivity than the main material. As an illustrative example we compare YA1 and YA3. From the Rietveld analysis, the composition ( Table 2) and phase fraction (Table 3) of the main Y x Al y B 14 phase is quite similar, as are the relative densities of the samples which are both close to 100% (Table 3). YA1 has a significantly larger amount of metallic YB 6 impurities and also slightly less amount of insulating alumina impurities compared to YA3. However, as noted above, the conductivity of YA1 is actually around 5 times lower than  Table 3. Phase fraction of constituent phases and theoretical and relative density of each sample. The phase fraction of constituent phases and theoretical densities are estimated using Rietveld refinement. www.nature.com/scientificreports www.nature.com/scientificreports/ YA3, indicating that the secondary phases in this case are not having a significant effect on the thermoelectric properties.
We consider the origin of the difference in properties. The large difference between YA1 and YA3 is only the starting composition (Table 2), where the amount of initial excess Al is around 1.5 times higher for YA3. Since Al itself cannot be seen in the XRD patterns, we speculate that Al atoms may be seeded in the grain boundaries and result in an enhancement of the electrical conductivity. Such an effect has been proposed before 58 and observed in Ba seeding of Ca 3 Co 4 O 9 for example 59 . Figure 8 shows the temperature dependence of Seebeck coefficients for each sintered sample. The Seebeck coefficient of Al-poor sample exhibits positive values, whereas those of Al-rich samples exhibit negative values. This indicates that the thermoelectric behavior of Y x Al y B 14 samples are changed from p-type to n-type with increasing Al occupancy. The Seebeck coefficient values for Al-poor sample increases up to 473 K, and then decreases with increasing temperature. In contrast, the absolute values of negative Seebeck coefficients for Al-rich samples increase with increasing temperature. The maximum positive and negative values, 530 μV/K and −160 μV/K, are obtained for YA0-P at 473 K and for YA1 at 973 K, respectively. Figure 9(a,b) show the temperature dependence of electrical conductivity of each sintered sample in accordance with thermally activated conduction (TAC) mode and variable range hopping (VRH) mode, respectively. According to the TAC model, the conductivity is expressed as the following; Meanwhile, in disordered systems, according to the VRH model, the conductivity for three dimensional systems is given by where σ 0,TAC and σ 0,VRH are constants which do not depend on temperature, E A is the thermal activation energy, T 0 is the characteristic temperature determined by the density of localized states near the Fermi level, k is the Boltzmann's constant and T is the absolute temperature 60 .
We consider the validity of the two different conduction models for our samples. In Fig. 9(a), the temperature dependence of the electrical conductivity of the Al-poor sample appears to generally follow the TAC dependence, except it appears to deviate from a straight line below 573 K. In contrast, in Fig. 9(b), the temperature dependence of the electrical conductivity appears to not be as good a fit to the VRH dependence as it shows a curvature over a wider temperature range. This implies that TAC is the dominant conduction mechanism for Al-poor Y x Al y B 14 above 573 K. In this case, the calculated E A and 0,TAC σ from the fitted line are 0.4 eV and 3.95 × 10 4 S/m, respectively. This activation energy value is similar to the band gap energy in the result of band structure calculation for Y 0.50 Al 0.75 B 14 45 .
We will consider the mechanism for the low temperature deviation below together with the discussion on the Seebeck coefficient. For Al-rich samples, neither the TAC (Fig. 9(a)) or VRH ( Fig. 9(b)) is a good fit, with the rate of increase of the electrical conductivity apparently suppressed at higher temperatures. We speculate that the reason of this suppression may be due to the presence of the Al segregation. This effect only appears to be noticeable for the highest conductive samples (i.e. Al-rich samples at high temperatures) and does not appear to be www.nature.com/scientificreports www.nature.com/scientificreports/ pronounced for the Al-poor sample where the level of electrical conductivity is much lower to begin with. For the Al-rich samples we attempt to obtain insight on the conduction mechanism through the temperature dependence of the Seebeck coefficient discussed below.
The temperature dependences of Seebeck coefficients for TAC and VRH are shown in Fig. 10. When the carrier concentration is increased by thermal activation, the Seebeck coefficient is generally expressed as the following 61,62 ; where e is carrier charge, k is Boltzmann constant, E s is the characteristic carrier-generation energy, T is the absolute temperature, and A is the heat-of-transport constant. However, the plots of Seebeck coefficients of all samples do not appear to have a clear T -1 component, not only for n-type but also for p-type Y x Al y B 14 ( Fig. 10(a)) which generally follows TAC for the electrical conductivity as discussed above. The reason can be elucidated by considering the mixed conduction case in semiconductors. In that case, the total Seebeck coefficient is given by 61 where n σ and σ p are the electrical conductivity due to electrons and holes, and S n and S p are the Seebeck coefficient for n-type and p-type conduction, respectively. Considering the temperature dependence of electrical conductivity discussed above, namely, the deviation to higher electrical conductivity at higher temperatures, the decrease of Seebeck coefficient of the Al-poor YA0-P above 573 K can therefore be attributed to the increase of thermally excited electrons.
In VRH systems, the temperature dependence of the Seebeck coefficient can be approximated by the following 62 ; Figure 9. Temperature dependence of electrical conductivity of each sintered sample. The lnσ is plotted versus (a) T -1 for TAC mode and (b) T -0.25 for VRH mode, respectively. The straight line for Al-poor sample in TAC mode expresses the fit to Eq. 1 for the temperature range from 573 K to 973 K. Also, the straight lines for Al-rich samples in TAC mode and VRH mode express the fit to Eq. 1 and to Eq. 2 for the temperature range from 373 K to 673 K, respectively.
where N(E) is the density of states at the Fermi level. For Al-rich samples, the temperature dependence of the Seebeck coefficients is almost proportional to T 0.5 , as shown in Fig. 10(b). From this temperature dependence of the Seebeck coefficient it is indicated that the Al-rich samples are VRH systems, since insulating or metallic impurities have less effect on the Seebeck coefficient.
To summarize what we have concluded from the data, the conduction mechanism of the Al-poor sample appears to follow TAC dependence, with deviations in the temperature dependences above 573 K indicated from thermally excited electrons. With the large increase of Al inserted atoms, and thereby electrons, for the Al-rich samples, not to mention a speculated effect of excess segregated Al atoms for YA3 in particular, the magnitude of electrical conductivity increases greatly compared to the Al-poor sample. However, despite this, it is indicated that the conduction mechanism follows the VRH mechanism, indicating that the electrical carriers have some localization due to disorder. Figure 11 shows the densities of the sintered bodies measured by Archimedes' method. All samples prepared by reactive SPS exhibit rather high densities, 2.97-3.07 g/cm 3 , compared to that of the reference sample (2.43 g/ cm 3 ) 44 , and are close to the theoretical density of the single crystal Y 0.62 Al 0.71 B 14 , 3.02 g/cm 3 . The measured densities for Al-rich samples are a little bit higher than the reported theoretical density of single crystal due to the presence of secondary phases. However, even taking into consideration the presence of a relatively small amount of secondary phases, highly dense samples were obtained (Table 3). This result suggests that the new synthesis process via reactive SPS with AlF 3 is also much more effective for densification of yttrium aluminoboride than the conventional process.
The thermal conductivities of n-type Y x Al y B 14 samples are shown in Fig. 12. Thermal conductivities generally decrease with increasing temperature indicating Umklapp scattering. YA2 and YA3 show some difference in the www.nature.com/scientificreports www.nature.com/scientificreports/ temperature dependence, with relatively less temperature dependence compared to YA1 and YA0-P. This trend is also reflected to a lesser extent in the intrinsic thermal diffusivities (Fig. S1). The behavior may be due to the Al segregation speculated before, since both YA2 and YA3 have higher Al starting compositions, but it is not clear at present. In comparison to reported samples, all samples exhibit slightly higher thermal conductivities due to their relatively higher densities than what has previously been obtained.
To investigate the microstructure, SEM observation for fracture surface of the samples was carried out (Fig. 13). The samples synthesized by reactive SPS exhibit highly dense microstructure compared to that of previous samples 46 though a minor amount of small pores are observed. All samples show coarse microstructure due to the high sintering temperature and liquid phase introduced during sintering. The microstructures generally consist of two phases, which are observed as a dark region and a bright region. The components in both regions are revealed by EDX analysis, as shown in Fig. 14. The result shows the presence of main elements, Y, Al, B and O. The dark region contains Y, Al and B but not O, whereas the bright region contains Al and O but not Y and B. It thus can be considered that the matrix and the secondary phase are Y x Al y B 14 and alumina, respectively. In the matrix phase, some differences in the contrast are observed. It implies an inhomogeneity of the metal atoms which occupy the interstitial sites in Y x Al y B 14 . Standard molar enthalpy of formation (∆H f 0 ) and standard molar entropy (S 0 ) of each material are listed in Table 4 63 . The calculated free energies of formations (∆G 0 ) for AlB 2 , . The theoretical and relative densities of the samples prepared by reactive SPS in the present study are given in Table 3. www.nature.com/scientificreports www.nature.com/scientificreports/ Al 2 O 3 and B 2 O 3 were listed and plotted, as shown in Table 5 and Fig. 15 64 . The results of thermodynamic calculations show that ∆G 0 of α-alumina is much lower than AlB 2 and B 2 O 3 . As discussed previously, it is considered that the presence of boron oxide layer coated on the surface of particles of the starting material of amorphous boron reacted with Al during sintering, and thus the alumina phase was formed. We have found that even after removing the boron oxide by the pre-treatment of heating, the amorphous boron powder can have oxidation in air again. To suppress the formation of alumina phase, using a closed system to enable heating of the raw material amorphous boron and then continuously, the mixing process and SPS sintering is desirable. www.nature.com/scientificreports www.nature.com/scientificreports/ Figure 16(a,b) show the power factors and ZT values of Y x Al y B 14 samples with the reference sample 44 , respectively. The power factor is dramatically improved by increasing Al occupancy. The maximum power factor, 3.7 × 10 −4 W/mK 2 , was obtained for YA3 at 973 K, which is significantly higher than that of the reference Y 0.56 Al 0.57 B 14 sample( Fig. 16(a)). The large improvement in power factor can be attributed to the high density achieved by the new synthesis process. The maximum ZT value, 0.08 was also obtained for YA3, as shown in Fig. 16(b). This value is comparable with those reported by previous study, because of the higher thermal conductivity. However, looking forward, it can be said that in general it is more straightforward to find ways to selectively lower the thermal conductivity 65,66 , rather than enhance the power factor 67,68 , Although the ZT is not high, there are very few materials which potentially can be used for very high temperature thermoelectric applications 10 , and we succeeded to prepare Y x Al y B 14 by developing a new sintering process, which is much shorter, simpler and more cost-effective.

conclusions
In this study, a new synthesis process with SPS and sintering additive for synthesis of dense yttrium aluminoboride was developed. The addition of 1 wt.% AlF 3 effectively and dramatically enhanced the sinterability and densification behavior of yttrium aluminoboride by the formation of liquid phase during SPS. Consequently, the process was remarkably simplified and total process time was shortened from ≥3 days to only ~4-5 h.
For p-type Y x Al y B 14 , the thermally activated conduction mechanism was proposed rather than the usually assumed variable range hopping (VRH) mechanism, to explain the experimental results of the temperature dependence of electrical conductivity and Seebeck coefficient. Conversely, the VRH mechanism appears to be a good match for the observed properties of the n-type Y x Al y B 14 . Because of the increasing Al occupancy and   www.nature.com/scientificreports www.nature.com/scientificreports/  . thereby electron carrier content, the n-type samples have much higher conductivity, however, despite this it is indicated that the electrical carriers have some localization due to disorder. ZT values of yttrium aluminoboride prepared by reactive SPS exhibits overall similar thermoelectric properties as the previously reported best samples prepared by conventional processes. The power factor is significantly enhanced because of the high density achieved by the present discovered method. The higher thermal conductivity balances out the ZT, however, in thermoelectric research there are many methods to selectively lower the thermal conductivity. In any case, the present synthesis method that we discovered radically shortens the time necessary for synthesis, e.g. from 8 h to 5 min for the p-type sample. This breakthrough technique is expected to be able to be applied to synthesize other borides also.

Methods
Commercially available YB 4 (99.9%, Japan New Metals Co., Ltd.), AlB 2 (Sigma-Aldrich Co., Ltd.), amorphous B (99%, New Metals and Chemicals Co., Ltd.) and AlF 3 (Wako Co., Ltd.) powders were used as the starting materials. YB 4 and AlB 2 were selected as the precursor for reactive sintering, and AlF 3 was selected as the sintering additive. The powders of YB 4 , AlB 2 , B and 0-1 wt.% AlF 3 were simply mixed under ethanol for 30 min using SiO 2 mortar with nominal compositions of Y 0.62 Al y B 14 , where y = 0.71, 1.24, 1.55 and 1.86. For y = 0.71 (YA0 samples), five different sintering conditions were tested;10 min and 1 h under the uniaxial pressure of ~30 MPa, each with and without AlF 3 , and 5 min under the uniaxial pressure of 80 MPa with AlF 3 . Subsequent samples YA1, YA2 and YA3 were prepared with AlF 3 additive and 1 h sintering time. The excess amount of Al (y = 1.24-1.86) was used to prepare Al-rich n-type yittrium aluminoborides.
The powder mixtures were dried at 373 K for 3 h in an oven, and then poured into a 10 mm diameter graphite die. Here, graphite paper was used as a release agent. The mixtures were heated up to 1773 K in a SPS apparatus (Dr. Sinter, Fuji Denpa Koki Co., Ltd.) with heating rate of 100 K/min, and then reactively sintered at 1773 K for 5-60 min under uniaxial pressure of 28-80 MPa in a reduced-pressure Ar atmosphere (−0.03 MPa). After sintering, the samples were cooled down with cooling rate of 200 K/min, and finally, the sintered discs, 10 mm in diameter and ~2 mm in thickness, were obtained.
After sintering, as-sintered samples were slightly grinded by a diamond plate (#400) to remove the graphite paper on their surface. The densities of grinded samples were measured using Archimedes' method. The constituent phases of sintered samples were analyzed by X-ray diffraction (XRD, Smart lab 3, Rigaku) with Cu Kα radiation (λ=1.5418 Å). The XRD patterns were collected at 2θ = 10-90° with in a step of 0.01°. The compositions of the prepared Y x Al y B 14 samples by reactive SPS were estimated from Rietveld refinement using FullProf software. The thermal diffusivity of each sample was measured from 373 to 973 K by using a laser flash analysis apparatus (LFA, TC-7000, ADVANCE RIKO, Inc.). Prior to the LFA, the samples were coated with carbon. After LFA, samples were cleaned to remove the carbon on their surfaces, and then machined into bar-shaped specimens by cutting machine. The electrical resistivity and Seebeck coefficient of the specimens were evaluated from 373 to 973 K using a thermoelectric tester (ZEM-2, ADVANCE RIKO, Inc.). The distance between probes used for the measurements was 5.6 mm. The specific heat capacities of the samples were measured from 313-1073 K using a high temperature differential scanning calorimetry (STA449F3-NT24, NETZSCH Japan Co., Ltd.). As crucibles, Pt pans, Pt lids and Alumina liner were used. Microstructure of fracture surfaces of each sample was observed by scanning electron microscopy (SEM, S-4800, Hitachi High-Technologies Co.) at an acceleration voltage of 10.0 kV. To investigate the components of matrix and secondary phases on the fracture surface, energy-dispersive X-ray spectroscopy (EDX, Hitachi High-Technologies Co.) was used.