Role of grain boundaries in Ge–Sb–Te based chalcogenide superlattices

GeTe and Sb₂Te₃ CSL with 20 repetitions has been sputter-deposited from stoichiometric targets of 4N purity on a Si(1 1 1) substrate with ρ  >  5 mΩ cm at a deposition temperature of 200 °C. The DC power was 35 W with an Argon gas flux set to 35 sccm. The growth rate of GeTe and Sb₂Te₃ is 0.1 nm s⁻¹. Superficial oxide on the substrate was removed beforehand by a wet-etch in aqueous hydrofluoric acid. The deposition was carried out at a base pressure of ${{p}_{0}}=~$ 10⁻⁸ mbar in a HV sputtering system with sputtering sources in confocal geometry and individually controlled shutters that allow precise control of the serial deposition.

The electron backscatter diffraction (EBSD) technique was used to investigate the individual Sb₂Te₃ and GeTe layers. The EBSD system (AMETEK, NJ, USA) was integrated into the NanoLab system. The SEM was typically operated with an acceleration voltage of $20\,{\rm keV}$ and a beam current of $1.6\,{\rm nA}$, with a chamber pressure of about 10⁻⁶ mbar. The sample tilt and working distance were set to 70° and 13 mm, respectively. With these parameters, the EBSD lateral resolution was approximately 30 nm, and the typical exposure frame rates for the EBSD camera was around $20$ to 30 fps.

For the overall structural analysis, a Bruker D8 Discover setup equipped with a Goebel mirror, a Ge(2 2 0) asymmetric channel-cut monochromator and a Eulerian cradle was used.

The structural analysis at the atomic level was performed with an FEI Titan ChemiSTEM 80-300 Cs-STEM operating at 200 kV and equipped with a super-X detector. TEM lamellae were prepared with a focused ion beam (FIB) employing a FEI Dual Beam Helios NanoLab system and finally polished using a Fischione NanoMill at low energy (500 eV) to remove residual FIB-induced amorphization. Conventional TEM investigations for the correlative TEM-APT analysis were performed using a FEI Tecnai F20 microscope operated at 200kV.

APT measurements were carried out by CAMECA LEAP 4000X Si instrument with laser pulses of 355 nm wavelength (UV), 10 ps pulse length, and 20–25 pJ pulse energy. The specimen base temperature was set to 40 K. APT samples were prepared using FIB milling as described in [11]. To minimize beam damage, a low energy (5 keV) Ga beam was used at the final ion-milling stage.

The texture of individual 200 nm thick Sb₂Te₃ (R-3m crystal structure) and GeTe (R3m crystal structure) films sputter-deposited on Si(1 1 1) substrate was analyzed by EBSD. This technique provides the orientation of individual grains within the first 30-50 nm of the film. Hence, the character of the GBs can be directly determined.

Figure 1 shows that both films, Sb₂Te₃ and GeTe, contain predominantly red-colored grains, which correspond to the [0 0 0 1] (or [0 0 1]) growth direction as shown in (0 0 0 1) pole figures. This implies that Sb₂Te₃ and GeTe films have a common [0 0 0 1] (or [0 0 1]) texture. However, ~15% of grains in GeTe possess a second, less pronounced orientation, i.e. [0   −1 1 0] texture, colored on the inverse pole figure image in green (see figure 1(b)). For Sb₂Te₃ no other texture besides the [0 0 0 1] texture is detected. Yet, a considerable fraction of Sb₂Te₃ grains of approximatively 35% are randomly oriented in the film. Moreover, the averaged grain diameter in Sb₂Te₃ and GeTe thin-films has been determined to $0.174\pm 0.106$ µm and $0.144\pm 0.106$ µm, respectively; in good agreement with the work of Saito et al [12] where the lateral grain size varied between 50 nm and a maximum of 300 nm.

Zoom In Zoom Out Reset image size

Figure 1 also shows the identified GBs overlaid on a plot of image quality (IQ) for the Sb₂Te₃ and GeTe thin-films, respectively. Based on the misorientation (disorientation angle and plane orientation) several types of GBs are present, such as random HAGBs with disorientation angle larger than 15°, low-angle GBs (LAGBs) with disorientation angle smaller than 15°, and highly symmetric HAGBs called twin boundaries (TBs) with specific disorientation angles (for example the Σ3 TBs are characterized by a 60° disorientation angle). We note here that 60° angle does not correspond to a rotational symmetry, but to a defect, since in the trigonal and rock-salt structures there is only a 3-fold rotation symmetry (i.e. they can be rotated 120° each block of the stacking without producing a defect). In figure 1, the TBs are highlighted by yellow and blue lines, whereas the random HAGBs are highlighted by black lines. A high proportion of HAGBs ~94% and a low proportion of LAGBs ~6% were identified for both Sb₂Te₃ and GeTe thin-films. 24.7% and 16.3% of the identified HAGBs corresponds to Σ3 TBs in Sb₂Te₃ and GeTe, respectively. Additionally, the GeTe thin-film exhibits a high amount of Σ16 and Σ31 TBs with a fraction of 43.9%.

The structure of a ~330 nm thick [Sb₂Te₃–GeTe]₂₀ CSL film (20 double-layers of Sb₂Te₃ and GeTe with trigonal space group symmetry P3m1) sputtered-deposited on a Si(1 1 1) substrate was investigated using XRD. The biaxial texture, observed with EBSD for the constituent materials was confirmed for the SL by means of a rotational $\phi$-scan of the $\{0\,1\,\overline{1}\,2\}$ peak family (see S1 in supplemental material available online at stacks.iop.org/JPhysCM/31/204002/mmedia). To analyze the stacking in growth-direction, figure 2(a) shows the results of an XRD $\Theta 2\Theta$ symmetrical scan of the CSL sample.

Zoom In Zoom Out Reset image size

The two most intense diffraction features (at around $2\,{{\mathring{\rm A}}^{-1}}$ and $4\,{{\mathring{\rm A}}^{-1}}$) stem from the Si substrate. The three peak-groups, highlighted in red, are the so called SL peaks flanked by a number of SL satellites. The distance of the series of satellite peaks is known to be $\Delta {{Q}_{z}}=\frac{2\pi }{\Lambda }$, where $\Lambda$ is the SL unit cell (supercell), i.e. one double-layer of Sb₂Te₃ and GeTe [13]. Hence, the supercell size is determined by fitting the satellite peak positions against a peak index (see figure 2(c)). To do that the peak groups are fitted with a set of superposed pseudo-Voigt functions of equal widths and distances (see figure 2(b)). With this the mean supercell size is determined to be $\Lambda =(16.6~\pm 0.1)$ nm.

The distance of the SL peaks (red lines in figure 2(a)) is related to the mean Te-sublattice throughout the CSL which encodes the average chemical composition [14] and yields ${{x}_{{\rm S}{{{\rm b}}_{{\rm 2}}}{\rm T}{{{\rm e}}_{{\rm 3}}}}}=0.50\pm 0.02$. Furthermore, we simulated the perfect superlattice structure using a kinematical scattering approach obtaining a best match in good agreement with the prior result with the XRD data for a CSL supercell of (8.16 nm Sb₂Te₃/8.40 nm GeTe). More details about XRD-spectra of CSL samples, fitting and stoichiometry determination can be found elsewhere [14].

Additional intensity is found at multiples of ${{Q}_{z}}=$$\left( 0.62\pm 0.04 \right)\,{{\mathring{\rm A}}^{-1}}$ (green highlight in figure 2(a)), which correspond to the strong scattering of the 8 quintuple blocks of Sb₂Te₃ within each supercell. There are broad peaks instead of individual satellite peaks observed, since supercells of slightly different sizes coexist in the CSL. The latter has been previously shown for samples grown by molecular beam epitaxy [5]. This understanding is further supported for the sputtered samples of this study by the TEM/APT analysis which shows defects, such as inserted GeTe bilayers and GeSbTe stackings, within the superlattice periods.

To determine the degree of preferential orientation (texture) of the grains in the film a rocking curve scan was performed (see figure S2 in supplementary information). The FWHM $\beta =(3.41\pm 0.04){}^\circ$ indicates that [0 0 0 1] (or [0 0 1]) is the main growth direction (see figure S2) in agreement with the EBSD results obtained on individual Sb₂Te₃ and GeTe layers. However, the obtained value is larger than that obtained by other authors, but we note here that the deposited CSL layer is also thicker than the ones usually investigated (~340 nm, instead of  ~50 nm). XRD probes the whole film, such that formation of defects predominantly in the upper parts of the layer will lead to an overall larger FWHM. This effect will be illuminated in the following local analysis by TEM and APT.

HRSTEM studies were also performed to investigate the structure of CSL layers at the atomic level. HAADF mode in STEM allows for direct interpretation of the micrograph intensity being roughly proportional to Z^1.7 [15]. Along the zone axis investigated in the present work, CSL atomic columns are formed either by formal anions or cations. The formal anion columns are composed only by Te atoms while the formal cation ones are composed by Ge (in GeTe regions), by Sb (in Sb₂Te₃ regions) or by a mixture of Ge and Sb (in GeSb₂Te₄ or in Ge₂Sb₂Te₅ regions). As Ge has a lower Z-value, i.e. intensity, than Sb and Te, darker spots in the micrograph correspond to Ge or to a mixture of the Ge/Sb while brighter spots correspond to Te, Sb, or a mixture of Sb and Te.

The STEM HAADF micrograph in figure S3 clearly shows the Sb₂Te₃–GeTe superlattice structure along the [1 1 –2 0] Sb₂Te₃ (or [1 –1 0]) zone axis. Sb₂Te₃ regions can be clearly distinguished by the higher Z-contrast and by the presence of 'vdW-like' gaps. The red box indicates the region in which a Z-contrast line profile was calculated. By means of the Z-contrast line profile we determined the thickness of the Sb₂Te₃ and GeTe regions in the micrographs. This provides only a rough estimation of the thickness because (i) the GeTe/Sb₂Te₃ interface is not always well defined and (ii) the procedure is possible only in the region in which the sample is exactly on axis, i.e. in a quite limited area. Nevertheless, we found the average thickness of Sb₂Te₃ ${{t}_{1}}=(8.33\pm 0.14)\,{\rm nm}$ and the average thickness of GeTe ${{t}_{2}}=(8.24\pm 0.24)\,{\rm nm}$. The supercell size, estimated by HAADF, is therefore $\Lambda ={{t}_{1}}+{{t}_{2}}=(16.6~\pm 0.5)\,{\rm nm}$, which is in very good agreement with the XRD data. The larger error is due to the rough GeTe/Sb₂Te₃ interface that introduces an uncertainty of about 1 nm, i.e. one Sb₂Te₃ block.

In figure 3 we show HAADF STEM micrographs of the sample in cross-view (a) at low magnification, (b) medium magnification and (c) at high-resolution. The sample was tilted within the Si 〈1 1 0〉 zone axis so that the CSL layers are clearly visible. Figure 3(a) proves the typical polycrystalline structure of the sputter deposited CSL film. Diffraction patterns acquired at the bottom of the CSL layer (i.e. first 100 nm, see figure S4(b)) showed the presence of GBs with very low disorientation angles of maximum 5°, i.e. LAGBs, having almost no influence on the periodicity of the CSL. However, in the upper part of the CSL the GBs possess much larger disorientation angles  >15° disturbing the periodicity of the superstructure strongly.

Zoom In Zoom Out Reset image size

In figures 3(b) and (c), Sb₂Te₃ regions can be clearly distinguished by their higher Z-contrast and by the presence of gaps of lower intensity marked with orange dashed lines in figure 3(c). These are electron-depleted vacancy layers with a typically very low Z-contrast [16]. We note here that the distance between two Te atoms in these 'vdW-like' gaps (0.3 nm), is typically smaller than the real vdW-gaps, which is the distance between two graphene layers (0.335 nm [17]). Moreover, in figure 3(c) blocks of five layers about 1 nm thick are also clearly visible, typical of the Sb₂Te₃ stacking on the c-axis. Along the c-axis, the Sb₂Te₃ unit cell is given by the stacking of three blocks, i.e. 15 layers [6]. The red box in figure 3(c) indicates a region in which the Z-contrast line profile was calculated. In this line profile Sb₂Te₃ blocks appear to be well-divided by vdW-like gaps. However, we can also distinguish a block of seven layers, that is typical of GeSb₂Te₄. The presence of a low-Z element (Ge) was confirmed by the presence of a layer with a quite low intensity. Therefore, this observation clearly proves Ge diffusion into the Sb₂Te₃ layer. Moreover, in figure 3(b) (in the middle of the image where the sample thickness is constant) brighter regions are also present in the GeTe layer which suggests that Sb might have also diffused inside GeTe layer during the CSL deposition. Hence, the interdiffusion between the Sb₂Te₃ and GeTe layers explains why the Sb₂Te₃–GeTe interfaces are rather abrupt as clearly highlighted in figure 3(b).

Yet, these results are qualitative and not quantitative, i.e. we are unable to give an exact content of Ge in the Sb₂Te₃ layer. For this purpose, APT investigations have been performed inside CSL grains, but also at the GBs.

Figure 4 summarizes a typical APT investigation of the entire CSL film with GeTe and Sb₂Te₃ layers repeated 20 times. The 3D reconstruction in figure 4(a) clearly shows an average layer thickness of ~8.5 nm for Sb₂Te₃ and ~7.5 nm for GeTe, in good agreement with the XRD and STEM experiments described before. We note here that the overall thickness of the CSL film (~340 nm) in the 3D reconstruction was calibrated based on the measured film thickness directly on the APT specimens by HRTEM. Figure 4(b) shows the 1D concentration profile constructed using a 30  ×  30  ×  0.2 nm³ sampling box placed perpendicular to the Sb₂Te₃/GeTe interfaces. Distinctive layers are clearly visible in the concentration profile. However, their composition deviates from the stoichiometric Sb₂Te₃ and GeTe composition.

Zoom In Zoom Out Reset image size

The APT experiments clearly show a chemical intermixing between Sb₂Te₃ and GeTe. First, a non-negligible amount of Ge is present in the Sb₂Te₃ layer, as expected from the STEM experiments. The content of Ge in the Sb₂Te₃ layers varies from one layer to another between 4 and 7 at.%. Second, surprisingly an even higher content of Sb is present in GeTe layer (5 to 13 at.%), information which was not accessible by STEM HAADF. For a broader overview, figure S5 shows the content of Sb in GeTe and of Ge in Sb₂Te₃ for three other APT measurements. We can clearly see that the maximum Ge (Sb) content in Sb₂Te₃ (GeTe) is 10 at.% (15 at.%). This suggests that the Sb is more soluble in GeTe than Ge in Sb₂Te₃ at temperatures of about 200 °C used during magnetron sputtering deposition.

Correlative TEM-APT investigations were achieved directly on the APT specimens to understand the interplay between structure and composition at the nanoscale. Figures 5 and 6 show an example of such TEM-APT correlative investigations.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The needle-shaped specimen was first analyzed by TEM and the results obtained are summarized in figure 5. The bright-field TEM micrograph of the entire APT specimen acquired in low-magnification conditions (41 kX, figure 5(a)) exhibits a slight change in contrast between Sb₂Te₃ (darker contrast) and GeTe (brighter contrast) layers. By investigating the same APT specimen at higher magnification conditions of 290 kX, a HAGB and a LAGB are clearly visible in the upper and lower parts of the specimen, as highlighted in figure 5(b). We note here, that the disorientation angles of ~33° and ~4° were determined based on the Sb₂Te₃/GeTe interface orientation offsets registered in the APT specimen as highlighted by dotted-white lines. The presence of a HAGB and a LAGB inside the APT tip are further confirmed by the electron diffraction patterns acquired in the CSL layer using an aperture for selecting a sample area of 200 nm in diameter (diffraction pattern in figure 5(c)). The angles 34° and 4.65° obtained by this method match the previous values (there is only one crystallographic rotation of 180° for obtaining equivalent domains, and hence 0°, 4.65° and 34° correspond to 'rotational' defects). The diffraction pattern acquired at the CSL/Si interface (diffraction pattern in figure S4(a)) proves the ${{\left\langle 1\,\overline{1}\,0 \right\rangle }_{{\rm CSL}}}\|{{\left\langle 1\,1\,2 \right\rangle }_{{\rm Si}}}$ orientation relationship in line with XRD $\phi$-scans (see figure S1 in supplementary material). Moreover, the first two CSL layers are slightly tilted by 1.5° with respect to the Si surface.

Finally, the same specimen was investigated by APT and the results obtained are summarized in figure 6. The Sb₂Te₃/GeTe interfaces in figure 6(a) are highlighted in the APT 3D reconstruction by an iso-concentration surface constructed using a Ge iso-concentration value of 30 at.%. These iso-surfaces confirm the presence of two GBs in the upper and lower part of the tip as highlighted in figure 6(a). Furthermore, the disorientation angle $\Theta$ could be determined directly on the 3D APT map by identifying the angle between the normal to Sb₂Te₃ (0 0 0 1) planes (i.e. [0 0 0 1] or [0 0 1] directions) of the neighboring grains as described in the supplementary material in figure S6(b). In contrast to TEM, where only the angle projected along the zone-axis can be determined, the angle obtained by APT can be precisely determined by taking into account the full 3D information on the detector. The disorientation angles for the HAGB and LAGB are $\Theta \approx 34{}^\circ$ and $\sim 4{}^\circ$, respectively, values in very good agreement with the TEM investigations presented in figure 5.

In the HAGB region a distinctive phase which corresponds to Ge₂Sb₂Te₅ is identified. The 1D concentration profile from figure 6(b), constructed at the HAGB position using a sampling box of 13  ×  7.5  ×  6 nm³, clearly shows that the Ge, Sb, and Te average concentrations correspond to 22.3  ±  2.2 at.%, 21.5  ±  2.2 at.%, and 56  ±  2.6 at.%, respectively correspond to Ge₂Sb₂Te₅ phase. Such a heterogeneous nucleation [18] of Ge₂Sb₂Te₅ phase at the CSL GBs has not been reported before. This is mainly because of the very limited region where this Ge₂Sb₂Te₅ phase is present (~8 nm wide and 60 nm long). Such information could not be obtained by STEM because the GBs, which are found in the ~50 nm thick TEM lamellae, are projected in 2D and, hence, such Ge₂Sb₂Te₅ nanoprecipitates are impossible to detect. We mention here that we have observed the heterogeneous nucleation of the Ge₂Sb₂Te₅ phase for other HAGBs investigated in the same sample. Thus we are describing a generic feature observed at HAGBs.

The question which may rise now is whether such a Ge₂Sb₂Te₅ phase is also formed at LAGBs. Figure 7(a) depicts the 3D map of a small portion of a LAGB where (QLs are easily visible in Sb₂Te₃. These QLs are in general 0.8 ÷ 1 nm thick and contain approximately 4 at.% of Ge. However, there are particular regions close to the LAGB where layers thicker than 1 nm are present (see small sampling box situated in Sb₂Te₃ layer in figure 7(a) and the atomic density profile, i.e. the overall number of atoms detected per each bin of 0.1 nm, in figure 7(b)). Such layers are rather called blocks and are highlighted in figure 7(b) by a grey-colored region. The blocks B1 (~1.3 nm), B2 (~1.3 nm), and B3 (~1.5 nm) have a thickness larger than that of a typical 1 nm thick QL. Momand et al [6] determined, by direct HAADF STEM observations, the thickness of a Ge₁Sb₂Te₄ block as typically 1.4 nm. This suggests indeed that the blocks B1, B2 and B3 correspond to Ge₁Sb₂Te₄ phase (see table S1).

Zoom In Zoom Out Reset image size

The detailed 3D composition measurements confirm the presence of both phases, Ge₁Sb₂Te₄ and Ge₂Sb₂Te₅, at the LAGB. Figure 7(b) summarizes the composition detected in each block and layer. The detected concentration values for Ge, Sb, and Te per each block are given in table 1. The compositions of B1, B2 and B3 (13.1  ±  2.6 at.% Ge, 27.8  ±  3.4 at.% Sb,) are indeed close to the composition of GeSb₂Te₄. We note here that we have identified the same GeSb₂Te₄ phase for other LAGBs investigated. However, the Ge concentration at the interfaces between these blocks which are only 0.5 ÷ 0.8 nm thick (such as B1/B2, B2/B3, and B3/B4; see table 1) was determined to be much higher approaching thus the composition of Ge₂Sb₂Te₅. Although the width of these interfaces does not match with the c lattice constant of stable trigonal Ge₂Sb₂Te₅ (1.7 nm) [6], the elemental composition corresponds to the Ge₂Sb₂Te₅ compound.