Scripta Materialia

Volume 63, Issue 7, October 2010, Pages 784-787

High-resolution nanostructural investigation of Zn4Sb3 alloys

• Baptiste Gault ^a⁎,
• Emmanuelle A. Marquis ^a,
• David W. Saxey ^a,
• Gareth M. Hughes ^a,
• Dominique Mangelinck ^b,
• Eric S. Toberer ^c,
• G. Jeff Snyder ^c

1. ^⁎ Corresponding author.
2. ^a Department of Materials, University of Oxford, Parks Road, Oxford OX13PH, UK
3. ^b IM2NP, CNRS, Aix Marseille Université, Case 142, Faculté de Saint Jérôme, 13397 Marseille Cedex 20, France
4. ^c California Institute of Technology, Pasadena, CA, USA

Abstract

Zinc antimonides exhibit high thermoelectric figures of merit and bear great potential for power generation. To advance the understanding of the structure–property relationship in these alloys, the microstructure of a nanocrystalline Zn₄Sb₃ alloy, containing 57.29 at.% Zn, was investigated by scanning electron microscopy and atom probe tomography. Chemical inhomogeneities were observed at grain boundaries. Within the grains the distribution of Zn shows large fluctuations inducing a complex microstructure made of Zn-rich and Sb-rich regions at a nanometer scale.

1. Introduction

Thermoelectric materials show great potential for clean and renewable power generation, thanks to their unique ability to harvest energy directly from sunlight or to capture waste heat and convert it into electricity [1]. Ideal thermoelectrics have been described as phonon-glass electron-crystals, having the poor thermal conductivity of a glassy material and the electrical conductivity of a perfect crystal [2]. Increasing the efficiency of thermoelectric materials, via reduction of the lattice heat conductivity, can be achieved by engineering the material microstructure to enhance phonon scattering processes on defects and interfaces [3–5]. This requires a clear establishment of the structure–properties relationship to enable the development of efficient bulk thermoelectric devices [2,6], with control over the long-term stability of the nanostructure.

Zinc antimonides represent one of the most promising groups of candidates for thermoelectric power generation at mid-range temperatures [2]. β-Zn₄Sb₃ exhibits extremely low lattice thermal conductivity [7], making it the prototype of the electron-crystal phonon-glass structure [8]. Extensive structural characterization via X-ray or neutron diffraction [8–10] has revealed a very complex atomic organization within a single phase structure. However, microstructural investigations have shown the presence of multiple phases [11,12]. In the present letter, the microstructure of Zn₃Sb₄ alloys was investigated via high-resolution scanning electron microscopy (HRSEM) and atom probe tomography (APT).

2. Materials

Samples were prepared via melting of pure elements (57.29 at.% Zn), sealed in a quartz ampoule under vacuum, followed by high-energy ball milling. The alloy was subsequently subjected to hot uniaxial pressing at 350 °C for 2 h to induce densification and consolidate the material. Such processes enable the fabrication of a variety of materials with a typical grain size of about 100 nm [13]. This was followed by stress-free annealing for another 2 h at the same temperature, chosen to avoid the phase transition at 400 °C [14].

3. Results

A Zeiss NVision 40 focused ion beam (FIB) HRSEM was first used to clean the sample surface to reveal the internal structure of the material via FIB milling. Subsequently, HRSEM was performed in the cleaned areas using a combination of secondary electron imaging with an in-lens detector and an energy-selective back-scattered electron detector for compositional contrast in images at low electron beam energy [15,16]. The FIB cleaning cut, shown in Figure 1(a), was made on the edge of the sample with a 54° tilt between the ion beam and the surface. Electron imaging was then performed with normal incidence at low energy (1 kV) to enhance the chemical contrast and probe the very surface of the sample.

2. B. Gault et al. / Scripta Materialia Volume 63, Issue 7, October 2010, Pages 784-787

Secondary electron imaging reveals a sub-micron grains structure highlighted by bright boundaries (Fig. 1(b)), with an average grain diameter of ∼200 nm in good agreement with transmission electron microscopy observation [12]. Brighter grains are also observed. Such contrasts could relate to specific topology, orientation via an electron channeling effect or chemical contrast, with a locally higher concentration of heavier elements (Sb). Channeling contrast was ruled out by imaging over a range of tilt angles, where no obvious change was observed. From back-scattered electron imaging, as shown in Figure 1(c), the contrast can be directly related to a local enrichment of Sb at grain boundaries and in some grains. The literature suggests that those grains could have a composition of ZnSb [11]. No obvious nanoscale structure was observed within the grains.

Figure 1: (a) SEM secondary electron (SE) images taken at 20 kV accelerating voltage of the FIB cleaned area. Low-energy (1 kV incident beam) imaging of secondary electrons (b) and low loss back-scattered electrons (c).

A FEI Dual Beam Helios and the same Zeiss NVision 40 were used to prepare APT specimens via in situ lift-out [17]. APT provides an analytical three-dimensional image of the distribution of atoms within the analyzed specimen [18] and, thanks to recent breakthroughs in its design, is progressively becoming a mainstream technique for the characterization of semiconductors for nano-electronics applications [19]. APT results of the intra-granular region are presented in Figure 2, where a three-dimensional (3-D) reconstruction and the mass spectrum are displayed. The analysis was performed at 60 K in pulsed laser mode (0.05 nJ/100 kHz), with an average detection flux of 5 × 10⁻³ atoms per pulse on an Imago LEAP 3000X HR. The mass spectrum (Fig. 2(b)) exhibits a set of peaks corresponding to the different elements in different charge states. Interestingly, some molecular ions are observed, a common feature of compound semiconductors analysis [20]. This makes compositional analysis more complex, as, for instance, peaks relative to ¹²¹Sb¹⁺ and ²⁴²Sb₂²⁺ are overlapping. In this case, an additional peak at 122 atomic mass unit (amu) also appears due to the complex (¹²¹Sb + ¹²³Sb)²⁺. Further, the relative amplitude of the mass peaks of a molecular ion can be calculated based on the natural isotopic abundances. Hence, the contribution of each ion type (N(¹²¹Sb¹⁺) and N(²⁴²Sb₂²⁺), where N is the number of atoms of a given ionic species) can be quantified precisely, enabling adjustment of the concentration [21]. This yields an overall concentration of 54.3 at.% Zn, lower than expected but not indicative of ZnSb.

Figure 2: Intragranular atom probe microanalysis: (a) 3-D atom distribution and (b) mass spectrum.

A preferential loss of Zn during APT analysis could originate from its low sublimation energy in such materials [22]. In this case, Zn atoms would leave the sample, preventing their identification by the instrument and contributing to the background. However, the relatively low level of the random background suggests otherwise. The poor thermal conductivity of the material induces a relatively slow decrease in the surface temperature after a laser pulse, which results in relatively large tails after mass peaks (Fig. 2(b)). This affects both Zn and Sb peaks in a similar manner and should not provoke a specific loss of Zn, as a consistent procedure was used to create atomic ranges that define the atomic identity based on their mass-to-charge ratio. Deviations from the expected composition may also find their origin in the materials fabrication, where a loss of zinc is often observed; during the specimen preparation method via FIB, where a temperature increase due to ion irradiation might induce vaporization of Zn; or by the presence of Zn-rich precipitates, as identified by TEM in similar materials, not encompassed by APT analysis [12].

Although deconvolution of the mass peaks enables accurate global concentration measurements, the exact position of individual ions cannot be identified unambiguously on the local scale. Indeed, it is impossible to discriminate between different ions with similar mass-to-charge ratios and belonging to the same mass peak, which induces a loss of spatial information and accuracy. Original methodologies were developed to overcome peak overlaps between molecular ions. For instance, N(²⁴²Sb₂²⁺) positions were selected at random from atoms within a single peak (here at 121 amu), and attributed the chemical nature of a two-atom ion. Subsequently, assuming that molecular ions are formed by atoms that were originally nearest neighbors within the material, molecular ions were dissociated into a sum of single ions occupying the same position. This procedure was repeated for all the different molecular ions encountered in the analysis, in order to perform a species-specific statistical investigation of the spatial distribution of Sb and Zn.

3. B. Gault et al. / Scripta Materialia Volume 63, Issue 7, October 2010, Pages 784-787

The results, and that of a corresponding randomized volume, are shown in Figure 3(c). The Zn and Sb distributions have a Pearson coefficient μ (reduced χ²) of the order of 0.46. This indicates deviation from a random solid solution and significant fluctuations of the concentration [23]. An isoconcentration surface, calculated from a concentration grid with 1 × 1 × 1 nm³ voxels, highlights regions with a concentration above 59 at.% Zn (d). Zn-rich regions (∼60 at.% Zn) represent about 10% of the volume, while Zn-depleted regions can have concentrations down to ∼40 at.% Zn, the remaining volume exhibiting a concentration close to ∼51 at.% Zn.

Figure 3: (c) Statistical distribution with a block size of 250 atoms; the dashed line corresponds to the binomial distribution expected for a perfect random solid solution. (d) five nanometer slice taken through the tomographic reconstruction shown in figure 2a, parallel to the analysis axis, in which red dots correspond to Sb atomic and molecular ions. A blue isoconcentration.

APT analysis of a region surrounding a grain boundary was also performed, and the resulting tomographic reconstruction is shown in Figure 4(a). Data were also obtained at 60 K in laser pulsed mode (0.02 nJ/50 kHz), at an average detection rate of 2 × 10⁻³ atoms per pulse. The presence of a grain boundary is assessed by a local change – here a drop – in atomic density in the vicinity of the boundary due to trajectory aberrations, a typical feature of APT analysis of grain boundaries [24,25]. Interestingly, the boundary also corresponds to a region depleted in Zn, as highlighted by the isodensity surface, in Figure 2(a). Away from the grain boundary, the measured concentration is 49.9 at.% Zn. Enrichment in Sb is observed at the grain boundary and evidenced by the two-dimensional composition map displayed in Figure 4(b). These results are in very good agreement with the HRSEM images of Figure 1(b and c), where similar contrasts are observed at the grain boundaries and within the ZnSb grains. From the micrographs, the size of the Zn-depleted layer was deduced to be in the range of 30–50 nm, larger than the field of view in the experiment presented in Figure 4.

4. B. Gault et al. / Scripta Materialia Volume 63, Issue 7, October 2010, Pages 784-787

Similar observations in low-Zn alloys have already been reported at a much larger scale [26].

Figure 4: (a) APT reconstruction showing the 3-D distribution of ions. An isodensity surface (15 at.% nm⁻³) was added to highlight the location of the grain boundary. (b) 2-D concentration map of Sb computed within the thin slice highlighted in light blue in (a).

4. Discussion

Two main mechanisms could underpin the formation of an Sb-rich layer at the grain boundaries. A low-Zn content could be a consequence of vaporization of Zn during the processing or sample preparation, yielding composition gradients which are kinetically trapped [22]. Alternatively, such gradients could be thermodynamically stable and arise from surface energy minimization of the grain boundaries.

The large composition fluctuations described in Figure 3 could be associated with distortions in the local structure of the β-Zn₄Sb₃[27], and might correspond to the early stages of phase separation. Phase stability in β-Zn₄Sb₃ is somewhat tenuous – calculations reveal that Zn₄Sb₃ may not be stable at low temperatures, with disproportionation into ZnSb and Zn. Instead, the bulk stability of Zn₄Sb₃ appears to be driven by entropic stabilization at elevated temperatures [27]. Together with the grain boundary-specific chemistry, these fluctuations are likely to impact the thermal conductivity of these alloys.

5. Conclusion

In summary, a combination of HRSEM and APT has brought significant insights into the nanoscale structure of these materials. Further investigations of Zn₄Sb₃ alloys with different compositions or heat treatments will enhance the understanding of their structure–property relationship.

Acknowledgement

  The Oxford FIM Group acknowledges the EPSRC for funding under Grant No. EP/077664/1, with which the UK National Atom Probe Facility was established. Thanks to Drs. M. Descoin and K. Hoummada and Prof. M.C. Record for fruitful discussions, help and advice with specimen preparation. B.G. acknowledges the financial support of the European Commission via an FP7 Marie Curie IEF Action No. 237059. E.A.M. acknowledges financial support from the Royal Society via a Dorothy Hodgkin fellowship.

References

• About ScienceDirect
• What is ScienceDirect
• Content details
• Set up
• How to use
• Subscriptions
• Developers

• Contact and support
• Contact and support

• About Elsevier
• About Elsevier
• About SciVerse
• About SciVal
• Terms and Conditions
• Privacy policy
• Information for advertisers