Synergistic effects of zirconium- and aluminum co-doping on the thermoelectric performance of zinc oxide

Abstract This work aims to explore zirconium as a possible dopant to promote thermoelectric performance in bulk ZnO-based materials, both within the single-doping concept and on simultaneous co-doping with aluminum. At 1100–1223 K mixed-doped samples demonstrated around ∼2.3 times increase in ZT as compared to single-doped materials, reaching ∼0.12. The simultaneous presence of aluminum and zirconium imposes a synergistic effect on electrical properties provided by their mutual effects on the solubility in ZnO crystal lattice, while also allowing a moderate decrease of the thermal conductivity due to phonon scattering effects. At 1173 K the power factor of mixed-doped Zn0.994Al0.003Zr0.003O was 2.2–2.5 times higher than for single-doped materials. Stability tests of the prepared materials under prospective operation conditions indicated that the gradual increase in both resistivity and Seebeck coefficient in mixed-doped compositions with time may partially compensate each other to maintain a relatively high power factor.


Introduction
Thermoelectric (TE) conversion of waste or solar heat into electricity represents a promising solution to meet growing needs in low-carbon and energy-efficient technologies [1][2][3].
The efficiency of thermoelectric generation is limited by the Carnot efficiency and characterized by the figure of merit ZT=α 2 T/ρκ, combining Seebeck coefficient (α), electrical resistivity (ρ), thermal conductivity(κ) of the candidate materials, and working temperature (T). Prospective applications require the thermoelectric materials with high thermal and chemical stability, the absence of toxicity and high natural abundance of the constituent elements. These represent the main advantages of oxide-based TE materials over traditional, Bi2Te3, Bi2Se3, PbTe -based thermoelectrics. Yet, ZT values obtained for the best-known oxide thermoelectrics are much lower than those required by most potential applications [4].
TE oxides arrived at a turning point when good TE properties were reported for NaCo2O4 in 1997 [5]. In the last two decades, more than an order of magnitude enhancement in ZT of oxides was achieved [6,7]. While being rather known for promising optoelectronic, catalytic and photochemical properties [8,9], donor-doped zinc oxide (ZnO) was also considered as a potential high-temperature thermoelectric material [10,11]. Doping with elements capable to possess the oxidation states higher than 2+ is a known straightforward approach to tune TE performance of ZnO. Representative examples include aluminum [10][11][12][13], indium [14][15][16], iron [17], nickel [18], bismuth [15,19], etc. From those, aluminum can be considered as a most used and common dopant.
The co-doping strategy was also found fairly effective [20][21][22][23]. In particular, the ZT values of Al-Ga-, and Al-Ni-co-doped ZnO materials reach up to 0.47-0.65 at 1173-1243 K, being among the highest observed so far in oxide-based thermoelectrics [20,21]. This behavior was attributed to the microstructural evolution in co-doped ceramics, leading to a decrease in the thermal conductivity while maintaining an appropriate electrical performance. Another interesting effect of co-doping, leading to an enhancement of aluminum solubility due to the presence of nickel cations and a X-Ray diffraction (XRD), differential scanning calorimetry (DSC) and diffuse reflectance spectroscopy (DRS) studies were performed on fine powders, prepared by grinding the sintered ceramic samples. Both polished and thermally-etched, and freshly-fractured ceramics were characterized by scanning electron microscopy combined with energy dispersive spectroscopy (SEM/EDS). For the combined total conductivity and Seebeck coefficient measurements the sintered ceramics were cut into rectangular bars ~1.5×2.5×15 mm 3 . Thermal diffusivity studies were performed on ~1.00 mm thick disc-shaped ceramic samples.
XRD patterns were recorded at room temperature using a PANalytical X'Pert PRO diffractometer (Cu Kα) and scanning in the range 2Θ=10º-80º, with a step of 0.02º and an exposition time of 200 s. SEM (Hitachi SU-70 instrument) and EDS (Bruker Quantax 400 detector) equipment were used for the microstructural characterization of the polished and fractured ceramics. The optical band gap (Eg) of the powdered samples, obtained by grinding of the sintered ceramics, was assessed by DRS using a Shimadzu UV 3100 (JP) spectrometer, equipped with an integrating sphere and a white reference material, made of BaSO4 and Spectralon ® , respectively.
The spectra of the samples were acquired in the UV-Vis range (250-825 nm), with 0.2 nm step size.
The total electrical conductivity (σ) and Seebeck coefficient (α) were measured using an experimental setup described elsewhere [29]. The measurements were performed simultaneously in flowing air on stepwise cooling from 1223 K to 523 K, followed by up to 0.5 hour thermal equilibration at each temperature, or vs. time in isothermal conditions. Good ohmic contacts for electrical characterization of ZnO-based materials at low temperatures may represent a certain problem [30]; therefore, in several cases, the electrical conductivity data was extrapolated down to 380 K to obtain estimates for the lattice thermal conductivity. The estimated experimental error in measured values did not exceed 3-5% for σ and 5-7% for α. Similar thermal equilibration procedures were implemented for the thermal diffusivity (D) and specific heat capacity (Cp) All indexed reflections belong to a hexagonal wurtzite structure, indexed in accordance with ICDD reference pattern 04-009-7657. Based on the XRD results, all prepared materials are apparently single-phase, except for the Zn0.993Zr0.007O sample containing a detectable amount of the monoclinic ZrO2 phase (ICDD reference pattern 04-008-7682). The corresponding peak at 2Θ~28.2º is hardly visible for Zn0.993Al0.002Zr0.005O composition, however, suggesting that this impurity may also be present albeit in a lower amount. Table 1 lists the unit cell volumes for all prepared materials, calculated from the lattice parameters estimated using a profile matching method in Fullprof software [31]. Although aluminum incorporation into ZnO lattice was studied in many works, unambiguous identification of the relevant defect types and their effects on the lattice parameters is still debatable [32][33] [34]. 3.25 * -the theoretical density was calculated from the XRD data assuming the phase-pure composition Presence of fourfold-coordinated Al 3+ substitution for Zn 2+ is expected to decrease the unit cell volume, in agreement with the ionic radii listed by Shannon [35] and literature data (e.g., [33]). This is the case for both Zn0.997Al0.003O and Zn0.993Al0.007O compositions (Table 1). Nearly equal unit cell volumes of Zn0.997Al0.003O and Zn0.993Al0.007O, however, indicate that for the selected processing conditions the mechanism of aluminum incorporation is altered on increasing its content. The latter may involve the formation of octahedrally and tetrahedrally-coordinated interstitial aluminum cations, core-shell structures with inhomogeneous Al content and a minor amount of side phases like ZnAl2O4 spinel ( [32] and references therein). Furthermore, an increase in the lattice parameters was pointed out in [33], due to the occupation of the interstitial sites by

Al atoms at high dopant contents
In accordance with the literature data, the lattice parameters of wurtzite phase in ZnO:Zr thin films and Zr-containing ZnO nanoparticles are increasing on zirconium content [24][25][26]36].
A similar tendency is also observed for the bulk Zn0.997Zr0.003O and Zn0.993Zr0.007O ceramics in the present work ( Table 1). The ionic radius of fourfold-coordinated Zr 4+ (0.59 Å) is very close to that of Zn 2+ (0.6 Å); hence, the lattice expansion on doping likely originates from the lattice stresses promoted by excessive charge of Zr 4+ cations and/or incorporation of zirconium in the interstitial positions with a higher coordination number [26]. The rather interesting case is represented by the mixed-doped compositions showing the unit cell volume notably below that for undoped ZnO (Table 2). This might be considered as a fingerprint of more complex defect structure, guided by a compromise between presumably different affinities of Al 3+ and Zr 4+ to reside in the lattice and interstitial sites with various coordination environments. Importantly, the tendencies for lattice expansion expected on doping with zirconium contradict to the experimental results (Table 1), especially in the case of mixed-doped Zn0.993Al0.002Zr0.005O and Zn0.993Al0.005Zr0.002O samples. A plausible explanation may be based on the assumption that the presence of zirconium species somehow facilitates aluminum incorporation into the lattice by substituting Zn 2+ cations in tetrahedral coordination. The excessive charge of the ⋅⋅ point defects might introduce a local lattice stress and promote lattice expansion, which can be, to a certain extent, compensated by the presence of smaller Al 3+ cations in tetrahedral coordination, along with the formation of cation vacancies ´´ [32,34]. For given processing conditions, such a compensation mechanism may increase the solubility of aluminum in zinc oxide lattice, resulting in an overall decrease of the unit cell volume. This hypothesis correlates with the studies of the microstructural and transport properties, as discussed below.  Zirconia inclusions are clearly identified for Zn0.993Zr0.007O (Fig. 3D); hence, actual zirconium doping level attained under applied processing conditions is well below 0.7%. A smaller amount of excess zirconia particles was also observed in the Zn0.993Al0.002Zr0.005O sample. On the contrary, no ZrO2 impurities were identified for Zn0.993Al0.005Zr0.002O, while the separation of residual amounts of aluminum-enriched nanophase containing zinc was observed at the grain boundaries ( Fig. 3E). Taking into account the typical reactions occurring in aluminum-doped zinc oxide [11,38,39], one can identify the nanophase as ZnAl2O4 spinel. Similarly, the presence of only vestigial amounts of ZnAl2O4 and ZrO2 (not shown) were found in the Zn0.993Al0.007O and Zn0.994Al0.003Zr0.003O samples by EDS, implying that their actual composition is essentially close to the nominal. It should be noticed that the solubility limits for both aluminum and zirconium in wurtzite structure may vary significantly depending on the processing approaches (bulk ceramics, thin films, nanoparticles) and conditions [10][11][12]24,26,32,33]. However, detailed structural studies rather suggest that it is well below 1% at., provided by steric, electronic and natural coordination preferences of zinc and these substituting cations [25,34].
The results of combined electrical studies of Zn1-x-yAlxZryO ceramics are shown in Fig. 4. correspond to 400-440 K [42,43], while doping is expected to shift it to lower temperatures [41].
Among the studied materials the lowest resistivity is observed for Zn0.997Al0.003O composition.
Although the values of the lattice parameters apparently suggest a similar level of Al 3+ for Zn 2+ substitution in wurtzite structure (Table 1), Zn0.993Al0.007O sample demonstrates ~5-times higher resistivity and a weaker temperature dependence than Zn0.997Al0.003O, an evidence for more complex charge carrier scattering mechanism, likely involving an impurity scattering. Somewhat similar dependence of the electrical transport on composition is also observed for Zn0.997Zr0.003O and Zn0.993Zr0.007O samples, containing ZrO2 traces. These samples demonstrate even higher resistivity values than for Al-containing analogues. Taking into account that two free electrons are expected to be produced by Zr 4+ for Zn 2+ substitution as compared to one electron from Al 3+ substitution, one might conclude that under discussed processing conditions the solubility of zirconium cations in wurtzite lattice is significantly below that for aluminum. The observed trends of the resistivity behaviour with temperature agree well with the changes in the Seebeck coefficient ( Fig. 4B), which is notably higher for Zr-containing samples. cations, having a smaller size compared to Zn 2+ . However, the latter is not sufficient to explain a low unit cell volume of the Zn0.993Al0.002Zr0.005O composition. In such mixed-doped materials, one might expect a complex interplay between substitutions in the lattice and interstitial sites, affected by mutual effects provided by the charge and size differences of Al 3+ and Zr 4+ cations. Presence of even minor phase impurities may decrease the charge carrier mobility; this may account for the slightly higher resistivity of mixed-doped samples as compared to Zn0.997Al0.003O, while the corresponding Seebeck coefficients are essentially comparable. Still, the slightly higher Seebeck coefficient of the Zn0.994Al0.003Zr0.003O is responsible for the highest power factor, observed for this composition (Fig. 5). The mixed-doped Zn0.993Al0.002Zr0.005O sample also demonstrates a relatively high power factor while containing a significant amount of zirconium, reaching 530 µW×m -1 ×K -1 at 1173 K. This behaviour is often observed in the heavily-doped ZnO and is associated with various manybody effects on the conduction and valence bands [44,47]. In other words, the DRS data suggests that the donor-doping level in Zn0.994Al0.003Zr0.003O, containing an equal amount of both dopants, may be the highest among the mixed-doped samples. Still, no additional conclusions on the mutual solubility effects provided by zirconium and aluminium can be taken from the DRS results due to the presence of minor impurity phases affecting the actual chemical composition of the wurtzite phase. As a hypothesis, slightly lower electrical resistivity of Zn0.993Al0.002Zr0.005O (Fig. 4A) may be likely attributed to the higher electron mobility, notwithstanding the evidence of phase impurities in this material (Fig. 2).
While co-doping with aluminum and zirconium is shown to provide significant synergistic boosting effects on the electrical conductivity, as discussed above, the long-term stability of ZnObased thermoelectrics under elevated temperatures still remains one of the main concerns limiting the potential applications. The ageing behaviour of selected prepared materials was evaluated at 973K and 1173 K, the results are shown in Fig. 6. The essentially stable electrical performance was observed at 973 K at least during first 20 hours ( Fig. 6 A,B,C), indicating that this temperature may be still acceptable for the hot side of thermoelectric modules involving the discussed compositions. Higher temperatures result in increase of both electrical resistivity (Fig. 6D) and Seebeck coefficient (Fig. 6E) (Fig. 6F). Additional studies are required to corroborate and explain this behavior, which may represent an interesting pathway to ZnO-based thermoelectrics with stable performance.
Yet another great concern regarding the application of donor-doped zinc oxide as thermoelectric material is only slightly addressed by the described co-doping approach. In general, mixed doping is favorable for decreasing the lattice thermal transport due to local changes in density and elastic constants, associated with the different atoms [48]. Fig. 7 shows the temperature dependence of the total (A) and lattice (B) thermal conductivity, calculated from the Wiedemann-Franz´ law as: where L (2.45⋅10 -8 W×Ω×K -2 ) is the Sommerfeld value of the Lorenz number [49]. As for the case of electrical resistivity, the experimental data on thermal conductivity was normalized to 100% density using the Maxwell correction [12,37]. with the highest doping level. On the other hand, the lower solubility of zirconium in ZnO leads to the higher thermal conductivity of the Zn0.997Zr0.003O and Zn0.993Zr0.007O samples. It should be noticed, however, that the presence of phase impurities as those shown in Fig. 3D,E may contribute to suppressing the thermal transport by additional phonon scattering at the corresponding interfaces. Still, the doping level and mixed doping itself produce rather noticeable effects on the thermal transport, as evidenced by the relatively high thermal conductivity of Zn0.993Zr0.007O, containing the highest fraction of secondary phase.
Finally, Fig. 8 summarizes the relevant effects and shows a cumulative impact of the electrical and thermal properties on the thermoelectric performance of the prepared materials, represented by ZT. The highest ZTs at T>900 K are observed for Zn0.997Al0.003O, mixed-doped Zn0.993Al0.002Zr0.005O and Zn0.993Al0.003Zr0.003O. At 1173 K the average ZT values of these mixed-doped materials are up to ~2.3 times exceeding the corresponding average value for single-doped Zn0.993Al0.007O and Zn0.993Zr0.007O, possessing roughly the same nominal substitution level. The latter clearly underlines the presence of synergistic effects provided by simultaneous aluminum-and zirconium co-doping, which significantly boosts the thermoelectric performance. Although Zn0.993Zr0.007O actually demonstrates the lowest performance (Fig. 8), it appears that higher content of zirconium in mixed-doped compositions is surprisingly favorable for higher ZT, as illustrated by the difference in performances of Zn0.993Al0.002Zr0.005O and Zn0.993Al0.003Zr0.003O compared to Zn0.993Al0.005Zr0.002O. This synergy mostly originates from enhanced electronic transport in Zn(Al,Zr)O, as evidenced by the results shown in Fig. 4A, while the variations of the Seebeck coefficient and thermal conductivity on co-doping present rather minor contribution. It is believed that further significant enhancement of the thermoelectric performance of mixed-doped compositions is possible only by applying nanostructuring approaches similar to those implemented in recent works (e.g., [27,52,53]), having a major objective to suppress extremely high thermal conductivity of zinc oxide.

Conclusions
In order to demonstrate the effects of zirconium doping on the thermoelectric performance of bulk ZnO-based thermoelectrics, a set of single-doped and mixed-doped samples with nominal composition Zn1-x-yAlxZryO (x=0-0.007, y=0-0.007) was prepared via conventional solid state route. Electrical studies revealed significantly lower resistivity of Zn0.993Al0.005Zr0.002O, Zn0.994Al0.003Zr0.003O and Zn0.993Al0.002Zr0.005O samples as compared to single Al-doped and Zr- compositions demonstrated a stable electrical performance at 973 K for 20 hours. At 1173 K a noticeable increase in the electrical resistivity and Seebeck coefficient was observed leading to the degradation of the power factor, which was less pronounced for mixed-doped compositions. The lowest lattice thermal conductivities were observed for Zn0.994Al0.003Zr0.003O, Zn0.993Al0.002Zr0.005O and Zn0.993Al0.007O. Higher content of zirconium in mixed-doped compositions is favorable for higher ZT reaching up to 0.12 at 1173 K.