Epitaxial Growth of Cubic Mnsb on Gaas and Ingaas(111)

Copyright and reuse: The Warwick Research Archive Portal (WRAP) makes this work of researchers of the University of Warwick available open access under the following conditions. This article is made available under the Creative Commons Attribution 3.0 (CC BY 3.0) license and may be reused according to the conditions of the license. For more details A note on versions: The version presented in WRAP is the published version, or, version of record, and may be cited as it appears here. The cubic polymorph of the binary transition metal pnictide (TMP) MnSb, c-MnSb, has been predicted to be a robust half-metallic ferromagnetic (HMF) material with minority spin gap & 1 eV. Here, MnSb epilayers are grown by molecular beam epitaxy (MBE) on GaAs and In 0:5 Ga 0:5 As(111) substrates and analyzed using synchrotron radiation X-ray di®raction. We ¯nd polymorphic growth of MnSb on both substrates, where c-MnSb co-exists with the ordinary niccolite n-MnSb polymorph. The grain size of the c-MnSb is of the order of tens of nanometer on both substrates and its appearance during MBE growth is independent of the very di®erent epitaxial strain from the GaAs (3.1%) and In 0:5 Ga 0:5 As (0.31%) substrates.


Introduction
Half-metallic ferromagnetism 1,2 is a highly desirable property for advanced spintronic devices which require very high spin polarization at the Fermi level and has been explored in several classes of materials. 3,4 The cubic polymorphs of many binary transition metal pnictides (TMPs), such as MnAs or CrSb, have been predicted to be half-metallic ferromagnetic (HMF) materials with Curie temperatures ðT C Þ well above room temperature which is essential if they are to be exploited in any functional device. [5][6][7] Many TMPs normally adopt a double hexagonal closepacked B8 1 niccolite structure (abbreviated n-), which has ABAC stacking along the c-axis (A = transition metal, B,C = pnictogen). The structure of the cubic (c-) TMP polymorphs are predicted to be signi¯cantly less energetically favorable than the n-polymorphs under normal growth conditions. However, because the TMPs have excellent engineering compatibility with many mainstream III-V and Group IV semiconductor materials, a their cubic polymorphs in particular have enormous potential as highly spinpolarized contacts for hybrid semiconductor spintronics. Combining TMPs with semiconductors allows the greatest possible°exibility in spintronic device design: arbitrary all-epitaxial ferromagnet/antiferromagnet/semiconductor heterostructures could be grown (e.g., MnSb/CrSb/In 1Àx Ga x As).
While most band structure calculations made using density functional theory (DFT) apply to zero temperature it has been known for some years that the nonzero temperature behavior of HMFs is very important in determining their real-world spin polarization and hence device performance. [8][9][10] The possibility of a critical temperature T Ã ( T C for the onset of a reduction in the spin polarization was highlighted by a recent DFT calculation employing the disordered local moments (DLM) approach to model nonzero temperature band structures. 5 In that work, both NiMnSb (the canonical HMF semi-Heusler alloy, 2 more consistently labeled MnNiSb 11 ) and c-MnSb were studied. NiMnSb has a small minority spin gap 0.5 eV giving a low T Ã in the region of 100 K. This low value of T Ã arises from defect-like states in the minority spin gap, whose spectral weight increases with spin disorder, and easily reaches the Fermi level with an overall magnetization reduction of only 5%. Conversely the large minority spin gap ! 1 eV of c-MnSb, together with a mid-gap Fermi level, means that magnetization reduction must reach around 20% before minority spin magnetic disorder states reach the Fermi level, giving a T Ã probably in excess of 300 K. 5 The prospect for room-temperature HMF behavior in c-MnSb makes it a highly promising spintronic material.
Several groups have investigated the growth of cubic TMP polymorphs by molecular beam epitaxy (MBE) on substrates with square symmetry. 6,12,13 The presence of cubic phases has been inferred from structural measurements such as X-ray di®raction (XRD) and transmission electron microscopy (TEM), while for some ultra-thin¯lms (< 1 nm) normally antiferromagnetic materials such as CrAs can show slightly open hysteresis loops, suggesting the presence of a ferromagnetic cubic polymorph. 14 However, detailed DFT work suggested that these ultra-thin¯lms were not actually cubic in structure and the observed magnetic hysteresis was likely due to uncompensated spins in a highly strained distorted orthorhombic epilayer. Furthermore, for such ultra-thin¯lms, interpretation of TEM and XRD data is far from being unambiguous. More recently it has been claimed that c-MnAs can be grown directly on InP(001) alongside the ordinary hexagonal phase. 15,16 We recently demonstrated the growth of c-MnSb within n-MnSb¯lms on GaAs(111), where the large grain sizes of the c-MnSb (! 10 nm) made structural identi¯cation more straightforward.
In this work we extend our MBE growth of MnSb from InP, 17 GaAs 18 and Ge 19 to relaxed In 0:5 Ga 0:5 As (111) virtual substrates. Furthermore, In 1Àx Ga x As structures are attractive for semiconductor spintronic applications thanks to their high electron mobility, high Land e g-factor and low Schottky barriers. 20 Furthermore, the nominal lattice mismatch to n-MnSb is only 0.31% for In 0:5 Ga 0:5 As, compared to 3.1% for GaAs. Hence, by comparing the growth of MnSb grown directly on GaAs with that on In 1Àx Ga x As, it is possible to determine whether the substrate lattice strain plays a role in polymorph formation. In fact, we will show that the formation of c-MnSb occurs in a very similar fashion on both GaAs and In 0:5 Ga 0:5 As substrates suggesting that the substrate strain is of minimal importance, at least in this family of substrate materials. These results are consistent with our earlier suggestion that the c-MnSb forms epitaxially on top of n-MnSb in a mixed c-MnSb/n-MnSb polymorphic layer, rather than in contact with the substrate. The outlook for exploiting TMP polymorphs in spintronics is brie°y discussed. a By \engineering compatibilty" we mean straightforward epitaxial growth, usually favorable interface chemistry, and ease of device processing and fabrication.

Experimental Details
All MnSb¯lms were grown by MBE in a dedicated home-built chamber, using wafer pieces cut to typically 8 Â 8 mm. After sonication with organic solvents and rinsing with deionized water to remove dust and cutting debris, substrate wafers were cleaned in vacuo by careful degassing, gentle Ar ion sputtering (500 eV, 0.5 A, 10 min) and annealing without incident Group V°ux. For the (111)A oriented substrates this produced the expected (2 Â 2) reconstructions on both GaAs and In 0:5 Ga 0:5 As. Mn and Sb e®usion cells operating close to 857 C and 355 C, respectively produced an Sb 4 :Mn beam equivalent pressure ratio of 6.6:1 as measured by a retractable ionization gauge. This appears to be the optimum°ux ratio for MnSb polymorph formation on GaAs(111) and the°uxes were balanced carefully before each growth. Sample rotation during growth is not possible within our MBE system. The growth rate was 2.7 nm min À1 and the substrate temperature was maintained at ð415 AE 5Þ as measured by a thermocouple on the sample manipulator. Films varying in thickness from 1 nm to 300 nm were grown: here we focus on MnSb¯lms in the thickness range 100 nm to 300 nm where cubic polymorphs can be found.
Both GaAs wafers (moderately n-doped, exactly oriented) and 400 nm thick In 0:5 Ga 0:5 As virtual substrates were used. The latter were grown in a separate Varian MBE system on 50 mm (2 inch) GaAs(111)A wafers. In 0:5 Ga 0:5 As was deposited under standard conditions on to a 100 nm GaAs bu®er layer (InGaAs growth rate 16.7 nm min À1 , substrate temperature 500 C, sample rotation) and a protective As cap was deposited after epilayer completion. Due to the high lattice strain, the surface roughness and crystalline mosaic were signi¯cantly higher for the In 0:5 Ga 0:5 As virtual substrates than for the GaAs wafers. The virtual substrates were transferred through air, cut and mounted in the same way as GaAs wafers and prepared for subsequent MnSb deposition using the same protocols except for a longer pre-anneal to desorb the As cap.
High-resolution XRD experiments were performed both in-house and at three synchrotron radiation facilities b all equipped with multi-circle di®ractometers. In this work, we show synchrotron XRD data only, all of which were obtained with samples at room temperature under°owing dry nitrogen using 10 keV photons monochromated using Si(111) crystals. Angular data, which were recorded in a triple-axis geometry using suitable analyzer crystals, have been reduced to reciprocal lattice units in which the Q z direction is de¯ned as normal to the substrate surface and hence parallel to a [111] direction. Scans were recorded as a function of both Q z and the orthogonal, in-plane, Q x , directions. Reciprocal space maps (RSMs) were obtained in symmetric (out-of-plane) and asymmetric (in-plane) di®raction geometries. The TEM experiments were performed using Jeol ARM-200F and 2100 microscopes operating at 200 keV.

Results and Discussion
We present TEM images from typical MnSb/ In 0:5 Ga 0:5 As/GaAs(111)A heterostructures and then compare XRD data from MnSb/In 0:5 Ga 0:5 As and MnSb/GaAs. In Fig. 1, are shown two brighteld TEM images from a 300 nm thick MnSb¯lm on In 0:5 Ga 0:5 As(111)A. The overview of the heterostructure is shown in (a) and the arrows highlight the GaAs-In 0:5 Ga 0:5 As interface (bottom) and In 0:5 Ga 0:5 As-MnSb interface (top). The In 0:5 Ga 0:5 As layer is highly defective with many threading dislocations, as expected from the large lattice mismatch (3.7%) with the substrate. However, its surface is reasonably smooth and has a sharp interface with the MnSb overlayer. Since the MnSb is closely matched to the relaxed In 0:5 Ga 0:5 As virtual substrate, the density of dislocations is much lower in the MnSb¯lm. However, non-niccolite structures can be discerned in the MnSb¯lms. In particular, the surface region of this¯lm shows a granular morphology with di®erent crystal structures. This layer is visible in Fig. 1(b) where the arrow highlights the interface between a granular surface layer incorporating non-B8 1 structures and the underlying \pure" n-MnSb.
This surface layer is shown in more detail in Fig. 2(a). The crystallite size within this layer is typically around 25 nm both laterally and in terms of overall layer thickness. A high-resolution TEM image obtained on a single crystallite from within this surface layer is shown in Fig. 2(b). The atomic columns are readily resolved and their symmetry is b Beamline X22C at NSLS (Brookhaven National Laboratory, USA), the XMaS facility at the ESRF (Grenoble, France) and beamline I16 at Diamond Light Source, UK. consistent with a cubic structure oriented with (111) planes parallel to the n-MnSb(0001) interface. A Fourier transform of the image is shown in the inset of Fig. 2(b). This part of the MnSb¯lm comprises both c-MnSb grains and a continuation of the n-MnSb structure, i.e., it is not a simple epitaxial layer of c-MnSb on top of n-MnSb. The appearance of this granular structure is strikingly similar to c-MnSb on n-MnSb grown directly on GaAs(111). 5 In Fig. 3, are shown selected synchrotron XRD data for a typical MnSb/In 0:5 Ga 0:5 As/GaAs heterostructure. Sharp line features on the symmetric di®ractograms in (a) and (b) are due to multiple scattering from the substrate. The expected substrate and virtual substrate peaks are present along with n-MnSb with its c-axis out-of-plane. Panel (b) shows a smaller angular range around the lowest order peaks. The sharpest and most intense peak is due to the GaAs(111) substrate (the peak is within 10 À3 Å À1 of the position expected for GaAs at room temperature). The virtual substrate appears at lower Q z due to its larger lattice parameter, and the peak is slightly broadened due to residual strain near the GaAs interface. The n-MnSb(0002) peak is intense and quite symmetric re°ecting the high crystalline quality of the epilayer. Lattice parameters derived from¯tting these peaks are a ¼ ð5:853 AE 0:001Þ Å and c ¼ ð5:768 AE 0:001Þ Å for In 0:5 Ga 0:5 As and n-MnSb, respectively. The value for the c-lattice parameter of n-MnSb is consistent with that found in¯lms grown on other substrates 5,19 as well as the bulk material. 21 Figure 3(c) shows a RSM about the GaAs(422) re°ection (asymmetric di®raction geometry). Several additional di®raction features appear due to the virtual substrate and the two polymorphs of MnSb. we would expect to see only one of these peaks in a given RSM. However, two di®erent epitaxial orientations of c-MnSb are present corresponding to growth on AB versus AC terminated n-MnSb regions. These are mutually misoriented in-plane by 60 . The presence of two epitaxial orientations enhances the grain contrast observed in TEM (Fig. 2). Having both in-plane (4 peaks) and out-ofplane (3 peaks) XRD data for the c-MnSb enables its strain state to be investigated. The lattice parameter is a ¼ ð6:435 AE 0:007Þ Å and we cannot detect any signi¯cant distortion (¯tting three separate lattice parameters always converges to identical values within the experimental error). This value is approximately 1% smaller than previously reported for c-MnSb in¯lms grown directly on GaAs. 5 The epitaxial strain for c-MnSb, at this lattice parameter, on n-MnSb is 10.2% and with small grain sizes of tens of nanometer it is clear that substantial epitaxial strains are to be expected. The lack of observed distortion probably re°ects the complex structure of the polymorphic layer where the c-MnSb grains embedded within a n-MnSb matrix are not subject to simple biaxial stress.
The mosaic spread of the¯lm and substrates were obtained from rocking curves, Q x scans, at the principal XRD peak positions (not shown). The full width at half maximum (FWHM) of the lowest di®raction order rocking curves is as follows: In 0:5 Ga 0:5 As 0.543 , n-MnSb 0.537 and c-MnSb 1.210 . Clearly, the mosaic spread of the n-MnSb lm is dominated by that of the virtual substrate, which in turn is due to the formation of the mis¯t dislocations at the GaAs/InGaAs interface shown in Fig. 1(a). The c-MnSb gives rise to much broader rocking curves due to mismatch-induced mosaic and the smaller grain size, consistent with the TEM and XRD data.
We show in Fig. 4 typical XRD data for MnSb lms grown directly on GaAs(111). The standard di®ractogram contains similar peaks to that of Fig. 3(a) with only the virtual substrate peaks missing. In particular, strong substrate and n-MnSb peaks appear and, despite the very di®erent substrate lattice parameter, c-MnSb is also present. The lowest order family of peaks is shown in Fig. 3(b). Some additional features appear compared to the virtual substrate samples. The peak labeled 1, lying between the c-MnSb and GaAs(111) peaks, is due to GaSb(111). This sometimes occurs at the GaAs/ n-MnSb interface due to Ga droplets resulting from the surface preparation of GaAs. The presence of such alternative III-V compounds at the substratelm interface which can also include InAs forming during MnAs growth on InP, can make identi¯cation of cubic TMP phases ambiguous without care. The peak labeled 2, between GaAs(111) and n-MnSb(0002), is due to n-MnSbð1 101Þ. Small crystallites of this alternative epitaxial orientation can appear in MnSb¯lms on GaAs and have also been observed for low-strain NiSb¯lms on GaAs (111). 22 However, the key point is the presence of c-MnSb on both GaAs and In 0:5 Ga 0:5 As substrates.
A RSM obtained in symmetric di®raction conditions is shown in Fig. 4(c). This data reinforces the interpretation of the principal features of Fig. 4(b). The n-MnSb(0002) and GaAs(111) features are intense and symmetric. The peak assigned to GaSb is also symmetric and sharp, re°ecting good epitaxy directly on the GaAs substrate; the lattice parameter of ð6:104 AE 0:003Þ Å corresponds very well to bulk GaSb (0.13% expanded). The peak assigned to n-MnSbð1101Þ is much weaker and less distinctly symmetrical as expected. The c-MnSb(111) feature is quite broad in Q x as expected for a granular phase with higher mosaic spread; the corresponding rocking curve has FWHM of around 1.1 , similar to its counterpart on In 0:5 Ga 0:5 As. The other interesting feature of the c-MnSb peak is a broad shoulder extending to lower Q z , also very clear in the diffractogram. This can be explained by the presence of very small crystallites in varying high strain states throughout the mixed n-MnSb/c-MnSb polymorphic layer.

Conclusions and Outlook
In this work, MnSb thin¯lms have been grown by MBE on (111)-oriented GaAs and In 0:5 Ga 0:5 As substrates with lattice mismatches of 3.1% and 0.31%, respectively. In both cases, high-quality n-MnSb¯lms can be grown using a suitable°ux ratio [6.6:1 excess Sb] and a substrate temperature of ð415 AE 5Þ C. Under these conditions, all¯lms showed the presence of the technologically important c-MnSb polymorph.
Structural characterization by TEM and XRD is consistent with a model of small c-MnSb grains within a mixed polymorphic layer (c-plus n-with sometimes some wurtzite 5 ). The c-MnSb is epitaxial on and within the n-MnSb matrix, and is under high epitaxial stress. However, this does not appear to be simple biaxial stress due to the mixed nature of the polymorphic layers. There is evidence for a wide range of strain states within the c-MnSb (both among grains in a single¯lm and from¯lm to¯lm) which most likely depends principally on crystallite size. The in-plane epitaxy of the c-MnSb corresponds to (111) planes lying on the n-MnSb(0001) but with two possible in-plane orientations separated by 60 , most likely re°ecting matching to ABlayer or AC-layer terminated n-MnSb. Further TEM work is under way to understand in more detail the strain states and epitaxial relationship of the c-MnSb in n-MnSb.
The key point of this paper is that the presence of the c-MnSb polymorphic layers does not depend on whether the n-MnSb is sitting on In 0:5 Ga 0:5 As(111) or GaAs(111) substrates. Thus, we can rule out the in°uence of epitaxial mismatch at the substrate-¯lm interface as the precursor for polymorphic growth, at least in this pair of epitaxial systems. We have never observed c-MnSb grown directly on any III-V or Ge substrate, even for (100) substrates (not discussed here). Hence, we do not believe that c-MnSb is readily stabilized directly on semiconductor substrates. At present, the precise mechanism for polymorph nucleation within the n-MnSb¯lm remains unknown. It is plausible that the polymorphic transition is related to the growth kinetics on particular surface reconstructions of the n-MnSb (0001) 23 ; layer growth kinetics and nanostructure self-assembly are long known to be a®ected by surface reconstructions in ordinary III-V semiconductor MBE. 24 With a better understanding of the TMP polymorph nucleation mechanism, improved precision in the MBE process should allow c-MnSb to be formed more reliably and with tighter control over crystallite sizes. Vicinal substrates may be useful insuppressing the twin epitaxial orientation of the c-MnSb.
The half-metallic cubic TMP polymorphs share key features with the most promising spintronic Heusler alloys, principally high Curie temperature and large minority spin gap. Considering also their good epitaxial compatibility with mainstream semiconductors, these qualities bode well for applications in room-temperature spintronics. Because the c-MnSb crystal structure only contains two facecentered cubic sublattices rather than the four of a full Heusler alloy, there is probably less scope for atomic-scale disorder to disrupt half-metallicity. However, the growth of cubic TMPs is less advanced than that of Heusler alloys and it is not clear what key defect properties may be important in large crystallites. Further DFT calculations of defective cubic TMPs would be valuable. Presently, our experiments on cubic TMPs focus on the crystallite sizes and strain states, which are also issues shared with polycrystalline Heusler alloy¯lms. Experimental demonstration of true half-metallicity in c-MnSb remains challenging due to the co-existence of n-MnSb but surface-speci¯c techniques such as spinresolved photoemission may initially be most useful, since surface layers dominated by the cubic polymorph can already be grown. Polymorphic layers thick enough to dominate transport in a point contact Andreev re°ection (PCAR) experiment would allow more direct comparison to Heusler alloys for which a signi¯cant body of PCAR data exists. The ability to control the strain states of highly spinpolarized cubic TMP layers formed on or within ordinary niccolite-structured layers o®ers exciting new possibilities in the¯eld of hybrid semiconductor spintronic devices.