Crystal structure and chemistry of tricadmium digermanium tetraarsenide, Cd3Ge2As4

A cadmium germanium arsenide compound, Cd3Ge2As4, was synthesized using a double-containment fused quartz ampoule method within a rocking furnace and a melt-quench technique. The crystal structure was determined from single-crystal X-ray diffraction, scanning and transmission electron microscopies, and selected area diffraction and confirmed with electron backscatter diffraction. The chemistry was verified with electron energy loss spectroscopy.

In our previous paper (Riley et al., 2012), we described the synthesis process for maximizing the production of the Cd 3 Ge 2 As 4 crystalline phase where the target composition was batched and processed under similar conditions used for making CdGeAs 2 , including the double-containment ampoule method. In this paper, we report the crystal structure of the Cd 3 Ge 2 As 4 crystalline phase; using XRD and analytical electron microscopies, it was shown to belong to the trigonal system in the R3 space group (No. 148). The SC-XRD analysis of small shards of the crystalline phase were utilized to propose the atomic positions of each of the elements, and subsequent use of a double-tilt stage holder in (S)TEM coupled with EELS analysis confirmed these data. Through the combination of these two techniques, the structure of this newly formed phase could be accurately described. Using the crystallographic information file (CIF) created with SC-XRD, EBSD was used to demonstrate successful pattern indexing from the Kikuchi patterns.

Structural commentary
The SC-XRD data yielded a trigonal unit cell of space group R3 (No. 148) with lattice parameters a,b = 7.3748 (13) Å and c = 27.415 (5) Å (see Table 1). The heavy-scattering Cd atom was assigned to a general position ca 1.4 Å above the ab plane with some confidence, but Ge and As have similar scattering factors and were distributed amongst the remaining sites with less certainty. The lowest R factor (3.62% for all data) was found for the final refinement of the atomic poisitions, which also agreed with the expected stoichiometry of Cd 3 Ge 2 As 4 . The structure is shown in Fig. 1 and contains sheets of metal atoms stacked along the c axis. The Cd site and both As sites are tetrahedrally coordinated to neighboring metal atoms. The two Ge sites have tetrahedral (to three Cd and one As) and octahedral (to six Cd) coordination environments, respectively (see Fig. 2). Having Ge in these two environments is consistent with the expected valences of the metal atoms. If Cd and As are assigned their expected oxidation states of Cd 2+ and As 3À , respectively, then one Ge 2+ and one Ge 4+ are required for charge neutrality.

Synthesis and characterization
The Cd 3 Ge 2 As 4 sample, made using a double-containment quartz ampoule setup (see Fig. 3a), is discussed elsewhere in more detail (Johnson et al., 2009;Riley et al., 2012), but will be briefly described here. For the double-containment ampoule setup, high purity Cd (4.4916 g, 99.9999%), Ge (2.9030 g, 99.9999%), and As (5.9907 g, 99.99999+%) (Alfa Aesar, Ward Hill, MA) were added into a pre-cleaned fused quartz vessel [10Â12 mm; GE214, GM Associates, Inc., Oakland, CA; cleaning details are provided elsewhere (Riley et al., 2012)] while inside a nitrogen glovebox (< 0.1 ppm of O 2 /H 2 O; M-Braun, Inc., Stratham, NH). Then, the vessel was moved out of the glovebox while connected to a sealed gate valve and 12.5-mm compression fitting to make sure that the contents of     Crystal structure of the Cd 3 Ge 2 As 4 showing (a) the distribution and linkages of atoms and (b) the layer structure composed of GeAs 4 tetrahedra and GeCd 6 octahedra; the legend is shown in the bottom right (Cd = red, Ge = blue, As = green). The reader is referred to the online version for reference to color. the ampoule were not exposed to the atmosphere. Then, the ampoule was evacuated, purged with semiconductor grade Ar/ 2.6%H 2 to 1.6Â10 4 Pa (120 Torr), and sealed with an oxypropane torch. This is the inner ampoule shown in Fig. 3b. This inner ampoule was loaded into a 19Â22 mm fused quartz tube (outer ampoule, Fig. 3c,d) that was sealed at the base, and the annulus between the two vessels was filled with 17.8 g of Cu powder ( 75 mm). The ampoule was heat treated in a rocking furnace where it was ramped at 3 C min À1 from room temperature to 400 C, held for 2 h in a fixed position, ramped to 625 C at 3 C min À1 , held for 2 h, ramped to 850 C at 3 C min À1 , held for 25 h, rocking was paused for 10 min to allow the melt to settle to the base, and then the ampoule assembly was rapidly removed from the furnace and quenched in an ice bath filled with NaCl.
Upon cooling, the ingot was cut into discs and polished in oil-based diamond suspensions for further analysis. Optical microscopy was utilized with cross-polarized light to visualize cross-sectioned specimens at various magnifications (12.5-1000Â) using a Leitz Orthoplan optical microscope (Leica Microsystems GmbH, Wetzlar, Germany). Imaging was performed using a Sony DSC-F717 digital camera connected to the microscope through a C-mount adapter.
A crystal suitable for structure determination was identified amongst the small pieces, which broke off one of the cut discs. A series of Omega scans were collected at 150 K from an untwinned crystal 30Â70Â70 mm using a Bruker APEXII diffractometer (Bruker AXS Inc., Madison, WI) and Mo K radiation (more details are provided in the Refinement section). Absorption corrections were applied using the multiscan method and the structure refined with SHELXTL.
The EBSD analysis was performed on a polished cross section of the sample using a Bruker e-Flash HR (Bruker AXS Inc., Madison, WI) coupled to a JSM-7001F field-emission gun SEM (JEOL USA, Inc., Peabody, MA). Mapping was performed under 30 kV acceleration voltage at 200Â magnification at 1024Â768 pixel resolution (0.2789 mm 2 map size), a 12.9 frames per second average collection rate, a 2.9 detector tilt angle, and a 70 sample tilt angle. Data were analyzed using ESPRIT (v1.9) software (Bruker AXS Inc.). Data processing revealed a 92.7% phase match with 7.3% of unidentified regions (zero solutions).
The STEM and TEM analyses were performed on an electron-transparent lamella ($5Â10 mm 2 ) prepared via a cross-sectional lift-out technique using a FEI Quanta 3D focused ion beam (FIB). A JEOL ARM 200CF, a cold field emission, aberration probe corrected TEM (JEOL USA, Inc.; Peabody, MA) operated at 200 kV was utilized to examine these samples. Several techniques were used, including SAD in conventional TEM mode, STEM high-angle annular darkfield (HAADF) and annular brightfield (ABF) as well as EELS. The EELS analysis was performed on a Gatan Quantum 965 ER Gatan Image Filter (Gatan Inc., Pleasanton, CA) equipped with fast shutter and dual EELS modules. The EELS spectra and spectroscopic mapping were performed with a convergence angle of 27.5 mrad and a collection angle of 90 mrad using the Ge L , As L , and Cd M lines. Back-ground subtraction and spectral math were performed with the Gatan Digital Microscopy Suite (v3.0).

Refinement
Crystal data, data collection and structure refinement details are summarized in Table 1. A suitable crystal for SC-XRD was selected and mounted on a Bruker APEXII diffractometer with a microfocus Mo-K X-ray tube and a CCD detector. Data were collected with 0.7 frame widths in ! and 20 s dwell times per frame with the crystal held at 150 K. A total of 2198 frames were collected, and the total exposure time was 12.21 h. The frames were integrated with the SAINT (Bruker, 2012) software package using a narrow-frame algorithm. The integration of the data using a trigonal unit cell yielded a total of 20756 reflections to a maximum angle of 42.35 (0.53 Å resolution), of which 2026 were independent (average redundancy 10.245, completeness = 98.6%, R int = 6.18%, R sig = 3.19%) and 1702 (84.01%) were greater than 2(|F| 2 ). The final cell constants of a = 7.3748 (13) Å , b = 7.3748 (13) Å , c = 27.415 (5) Å , and volume = 1291.3 (5) Å 3 , are based upon the refinement of the XYZ-centroids of 6406 reflections above 20 (I) with 8.723 < 2 < 82.60 . Data were corrected for absorption effects using the multi-scan method with the SADABS software program (Bruker, 2001). The ratio of minimum-to-maximum apparent transmission was 0.476. The calculated minimum and maximum transmission coefficients (based on crystal size) were 0.2330 and 0.4730. The structure was solved and refined using the SHELXTL (Sheldrick, 2008) software package, using the space group R3, with Z = 6 for the formula unit, As 4 Cd 3 Ge 2 (referred to here as Cd 3 Ge 2 As 4 ).  The final anisotropic full-matrix least-squares refinement on F 2 with 29 variables converged at R 1 = 2.53%, for the observed data and wR 2 = 4.91% for all data. The goodness-of-fit was 1.032. The largest peak in the final difference electron density synthesis was 1.996 e À Å À3 and the largest hole was À2.180 e À Å À3 with an r.m.s. deviation of 0.324 e À Å À3 . On the basis of the final model, the calculated density was 6.034 Mg m À3 and F(000), 2040 e À .

Results and discussion
Through the progression of samples fabricated to optimize production of the Cd 3 Ge 2 As 4 crystals (Riley et al., 2012), optical microscopy was used to assess macroscopic features and crystal yield. While the crystals and the glassy matrix have similar compositions, the crystals could be separately visualized using cross-polarized light illumination; this provided a technique for broadly assessing the crystallite size. This can be seen in Fig. 4 where the crystallites populate the core of the disc and the amorphous outer layer can be separately visualized.
The EBSD mapping shown in Fig. 5 confirms that the CIF generated from SC-XRD analysis of the Cd 3 Ge 2 As 4 crystal indexed well to the Kikuchi patterns. This data was used to corroborate the SC-XRD data. The orientation maps presented in Fig. 5c demonstrate that the grain sizes can span several hundred micrometers.   Cross-polarized light optical micrograph of a cross-sectioned sample disc after polishing showing the crystals within the core and the amorphous rim around the perimeter.  Utilizing both the diffraction patterns in TEM mode and the Kikuchi map in STEM Ronchigram mode, the sample was tilted to obtain low index zone axes in order to better understand the crystallography of the material. Fig. 6 shows HAADF, ABF, and diffraction patterns collected from the {11-20} and {21-30} type planes of the trigonal crystal. Ball-andstick models of these orientations (derived from previously mentioned XRD studies) matched extremely well with the atomic column imaging collected in STEM mode. While slight drift in the imaging of the sample during data collection is apparent in the atomic column images, the local ordering could be related to the XRD data. The contrast of the HAADF images was used to corroborate the symmetry and the location of the atoms, but since the predicted structure did suggest overlap of atomic species, using the contrast in the HAADF and ABF images was insufficient to explicitly prove the structure.
Atomic column dual EELS mapping shown in Fig. 7 provided a clearer picture of the chemical location of each of the atoms within the crystal. The Cd-M edge, starting at 403.7 eV (see Fig. 8), was utilized to map the location of the Cd elements in both orientations shown in Fig. 6. In the {12-30} orientation, the distinct Cd positions could be easily mapped without overlap from either the Ge or As. In the {11-20} orientation, the Cd overlapped with the Ge, and hence the signal was slightly obscured, but individual atomic positions could still be observed. The Ge and the As signals were more complicated because of their L edge onsets at 1217 and 1323 eV, respectively. The higher energy edges were more difficult to map because of shorter collection times and low signal-to-noise ratios. Additionally, since the edges are so close in energy, background subtraction and signal integration were difficult. Theoretically, the Ge-L edge could be decoupled from the As-L edge, but because of the minimal amount of integration before the As-L edge, the amount of intensity does not provide adequate information for successful deconvolution.
In EELS, the As-L signal will always be convoluted with the Ge-L signal, so exact identification of the As versus the Ge proved difficult. Mapping of the Ge-L edge before the As-L edge ($1217-1320 eV) provided a map that suggests there are Ge and As substitutions in both orientations. When the As-L + Ge-L edges were utilized for mapping, the signal validated the XRD data in that there were individual layers of As in both orientations. Mapping of the Ge-L edge showed little Ge amongst the Cd layers, but when the Ge and As signals were used, the atomic positions suggested by XRD and HAADF were expressed. Again, since it is difficult to deconvolve the two signals, atomic column mapping could not be used to definitively identify the positions of the As and Ge atoms within the crystal structure. Difference maps are also provided in Fig. 7 (i.e., RGB mix), which give a good indication of the positions of the elements in relation to one another.

Funding information
Funding for this research was provided by: Pacific Northwest National Laboratory (PNNL), which is operated for the US Department of Energy by Battelle under Contract DE-AC05-76RL01830.