Interstitial Zn atoms do the trick in thermoelectric zinc antimonide, Zn4Sb3: A combined maximum entropy method X-ray electron density and ab initio electronic structure study

The experimental electron density of the high-performance thermoelectric material Zn4Sb3 has been determined by maximum entropy (MEM) analysis of short-wavelength synchrotron powder diffraction data. These data are found to be more accurate than conventional single-Crystal data due to the reduction of common systematic errors, such as absorption, extinction and anomalous scattering. Analysis of the MEM electron density directly reveals interstitial Zn atoms and a partially occupied main Zn site. Two types of Sb atoms are observed: a free spherical ion(Sb +3) and Sb2 4+ dimers. Analysis of the MEM electron density also reveals possible Sb disorder along the c axis. The disorder, defects and vacancies are all features that contribute to the drastic reduction of the thermal conductivity of the material. Topological analysis of the thermally smeared MEM density has been carried out. Starting with the X-ray structure ab initio computational methods have been used to deconvolute structural information from the space-time data averaging inherent to the XRD experiment. The analysis reveals how interstitial Zn atoms and vacancies affect the electronic structure and transport properties of beta-Zn4Sb3. The structure consists of an ideal A12Sb10 framework in which point defects are distributed. We propose that the material is a 0.184:0.420:0.396 mixture of A12Sb10, A11BCSb10 and A10BCDSb10 cells, in which A, B, C and D are the four Zn sites in the X-ray structure. Given the similar density of states (DOS) of the A12Sb10, A11BCSb10 and A10BCDSb10 cells, one may electronically model the defective stoichiometry of the real system either by n-doping the 12-Zn atom cell or by p-doping the two 13-Zn atom cells. This leads to similar calculated Seebeck coefficients for the A12Sb10, A11BCSb10 and A10BCDSb10 cells (115.0, 123.0 and 110.3 mVK-1 at T=670 K). The model system is therefore a p-doped semiconductor as found experimentally. The effect is dramatic if these cells are doped differently with respect to the experimental electron count. Thus, 0.33 extra electrons supplied to either kind of cell would increase the Seebeck coefficient to about 260 mVK1. Additional electrons would also lower S, so the resulting effect on the thermoelectric figure of merit of Zn4Sb3 challenges further experimental work.