Direct Epitaxial Nanometer-Thin InN of High Structural Quality on 4H-SiC by Atomic Layer Deposition

Direct Nanometer-Thin InN of High Structural Quality on by Atomic Indium nitride (InN) is a highly promising material for high frequency electronics given itslow band gap and high electron mobility. The development of InN-based devices is hamperedby the limitations in depositing very thin InN films of high quality. We demonstrate growth ofhigh-structural-quality nanometer thin InN films on 4H-SiC by atomic layer deposition (ALD).High resolution X-ray diffraction and transmission electron microscopy show epitaxial growthand an atomically sharp interface between InN and 4H-SiC. The InN film is fully relaxed already after a few atomic layers and shows a very smooth morphology where the low surfaceroughness (0.14 nm) is found to reproduced sub-nanometer surface features of the substrate. Raman measurements show an asymmetric broadening caused by grains in the InN film. Our results show the potential of ALD to prepare high quality nanometer-thin InN films for subsequent formation of heterojunctions. ABSTRACT Indium nitride (InN) is a highly promising material for high frequency electronics given its low band gap and high electron mobility. The development of InN-based devices is hampered by the limitations in depositing very thin InN films of high quality. We demonstrate growth of high-structural-quality nanometer thin InN films on 4H-SiC by atomic layer deposition (ALD). High resolution X-ray diffraction and transmission electron microscopy show epitaxial growth and an atomically sharp interface between InN and 4H-SiC. The InN film is fully relaxed already after a few atomic layers and shows a very smooth morphology where the low surface roughness (0.14 nm) is found to reproduced sub-nanometer surface features of the substrate. Raman measurements show an asymmetric broadening caused by grains in the InN film. Our results show the potential of ALD to prepare high quality nanometer-thin InN films for subsequent formation of heterojunctions.

Indium nitride (InN) is a highly promising material for high frequency electronics given itslow band gap and high electron mobility. The development of InN-based devices is hamperedby the limitations in depositing very thin InN films of high quality. We demonstrate growth ofhigh-structural-quality nanometer thin InN films on 4H-SiC by atomic layer deposition (ALD).High resolution X-ray diffraction and transmission electron microscopy show epitaxial growthand an atomically sharp interface between InN and 4H-SiC. The InN film is fully relaxed already after a few atomic layers and shows a very smooth morphology where the low surfaceroughness (0.14 nm) is found to reproduced sub-nanometer surface features of the substrate. Raman measurements show an asymmetric broadening caused by grains in the InN film. Our results show the potential of ALD to prepare high quality nanometer-thin InN films for subsequent formation of heterojunctions.
File list (2) download file view on ChemRxiv manuscript_revised.pdf (620.65 KiB) download file view on ChemRxiv supporting information_revised.pdf (337.38 KiB) Group III-nitrides are highly valuable semiconducting materials for electronic and optoelectronic devices. The low bandgap of InN 1 allows light-emitting-diodes (LEDs), with high In-content InGaN or InN as active layers, to operate to the near infrared regime, and facilitates new photovoltaic and photocatalytic applications. 2 The high electron saturation velocity of InN and high mobility predicted to reach up to reach 12000 cm 2 /Vs 3 raise the interest in this material for high-frequency devices. Modelling has predicted that the two dimensional electron gas (2DEG) density, electron localization, and 2DEG mobility can be greatly enhanced by either inserting a single layer of InN between AlGaN and GaN in a high electron mobility structure, or by using an electron channel of up to 5 nm of InN, if the underlying buffer layer could prevent the InN from relaxation. 4,5 High quality nanometer thin InN layers are therefore of high interest for future LED and high-frequency transistor technologies.
The growth of homogeneous nanometer thin InN layers suffers from large lattice mismatch to common templates and substrates for group III-nitride epitaxy. Furthermore, In-adatoms have high surface diffusivity and high self-cohesive force, meaning that metallic In droplets are prone to form. Deposition of InN is also hampered by the low decomposition temperature of InN. Therefore, InN deposition typically results in separated crystalline islands or films with high density of structural defects. Chemical vapor deposition 6,7 , hydride vapor phase epitaxy 8 , sputtering 9 and molecular beam epitaxy 10,11 (MBE) have been employed for growth of InN to study the bulk properties. MBE has also been used to prepare InN monolayers embedded in a GaN matrix for short-period superlattice. 12 However, the knowledge and deposition techniques for just-a-few-nanometer InN films is still at its infant stage.
Recently, atomic-layer-deposition (ALD) has emerged as a promising method for the growth AlN, 13 GaN, 14,15 and InN. 16,17 The ALD approach, characterized by the alternating introduction of the group III and the nitrogen precursors separated by inert purge gas between pulses, is fully depending on surface chemical reactions, eliminating gas phase reactions between the different precursors. The hallmark of ALD is also a perfectly conformal film on topographically complex substrates, potentially opening up for more complex device structures.
Here, we present homogeneous, wurtzite InN layers epitaxially-grown on on-axis 4H-SiC (0001) by ALD. 4H-SiC possess type-I band alignment with respect to InN which could be a promising substrate to initiate the InN heterojunctions for device applications. 18 The growth of InN is done by ALD at 320 °C and 6 mbar in a Picosun ALD system as previously described in detail. 19 Trimethyl indium (TMI) and (NH3 + Ar + N2) plasma were used to provide In and N respectively for complete reaction of InN. Double-side polished with Si-face chemicalmechanical-polished, on-axis 4H-SiC (0001) substrate used in this work was commercially acquired in test grade from Norstel AB. The 4H-SiC (0001) substrate was chemically cleaned using standard RCA cleaning procedure before loading into the reactor. After stabilization at 320°C for 60 min and prior to the growth of InN, the 4H-SiC (0001) substrate was subjected to (Ar + N2) plasma treatment for 2 minutes to remove residual surface oxide. After the plasma treatment, the system was purged with 100 sccm N2 for 10 sec followed by the first TMI pulse.
An ALD cycle comprises 4 s TMI pulse, 10 s N2 purge, 12 s (NH3+ Ar + N2) plasma exposure and 6 s N2 purge. More information related to the material growth process is given in supporting information. The diffraction peaks at 31.3° and 35.586° can be indexed to the relaxed wurtzite InN (0002) planes and to the (0004) planes of the 4H-SiC substrate, respectively. Given that the critical thickness of InN on 4H-SiC is less than one monolayer, 12 a relaxed InN film is expected.
However, the 2.3 nm InN is seemingly compressively-strained as the profile shifts toward lower 2θ angle. Metallic In, which is a common inclusion in InN thin films or on the surface, seen as a peak at 33 o , is not detected in our films. 11 On the other hand, no peak is observed in a grazing incidence XRD (GI-XRD) 2θ scan in the range of 20-90° using a fixed 0.5° incident angle under current geometry. By tilting the sample 90 o (χ = 90°), the diffraction peaks associated with   Fig. 2c). This edge-type misfit dislocations are known to be the most common and efficient defect for strain relaxation at interfaces with large lattice mismatch as have been observed in InN/AlN 21 and other material systems. 22,23 The only other structural defects observed are some regions where certain atomic columns do not show well-distinguished projections as indicated by the blue arrows in Fig. 2d. We believe these are in-plane irregularities related to point defects caused by the coalescence between separate grains of monolayers in order to minimize the energy. We believe that this is a consequence of extremely low growth rate (0.37Å per ALD cycle), so the imperfections are introduced at boundaries such as vacancies and impurities without any distinguishable Burger vectors. It should be mentioned that the dark contrast propagating along growth direction in the TEM image is not associated with threading dislocations. Instead, they reflect the In-deficient nature due to the high-angle annular dark-field imaging. In vacancies and impurities of light elements are speculated to be the point defects. This argument is supported by our Raman results which will be discussed later.  SiC. The contrast of the image is weak, indicating that our InN film is smooth and is free from the common surface non-homogeneities such as large crystal grains and In-rich droplets. 6,11 It should be mentioned that non-optimal ALD growth condition will lead to rougher surface and In-rich clusters as observed in the literature. 24 A smooth surface is further shown by the topographical atomic force micrograph of the InN surface (Fig. 3b). The root-mean square (RMS) roughness of the 20 nm thick InN film is 0.16 nm. For comparison, the root-mean square roughness of the bare 4H-SiC substrate is 0.13 nm. We find that the RMS roughness of InN is independent on the thickness of InN in this study. The surface stepping feature of 4H-SiC caused by the CMP process finished at the substrate supplier (insert in Fig. 3b), is duplicated by the InN film, showing the conformity of the InN growth at the nanometer-scale. Raman spectroscopy was conducted to study the vibrational properties and the strain status of the films. The spectra are shown in Fig. 4, where the spectrum of a bare 4H-SiC substrate is also shown for reference (the bottom curve). Two of the Raman modes allowed in the backscattering geometry, E2(high) and A1(LO), are observed in all samples, but instead of sharp peaks we observe significantly broadened bands peaking approximately at 585 cm -1 and 489 cm -1 , close to the positions of these modes in a relaxed, single-crystalline wurtzite InN grown along the c-direction. 25 The A1(TO) Raman mode is forbidden in the backscattering geometry, but appears in our spectra as a band positioned around 450 cm -1 . Both the broadening of the Raman modes and the appearance of the forbidden A1(TO) can be understood by the grained structure of the films. According to our TEM analysis, the grains are in nanometer-scale defined by the thickness and the "in-plane irregularities". The existence of grains aligned along the c- axis but coalesced laterally with respect to each other implies defects associated with the positioning boundaries between the individual grains. Hence, inhomogeneous strain distribution within the film is expected, which leads to inhomogeneous broadening of the Raman peaks. In addition, the selection rules could be relaxed by the deformation potential originating from small displacement of atoms from their equilibrium positions. Such phenomenon has been observed in nano-crystals 26 or GaN films with anisotropic strain. 27 Finally, a band centered around 380 cm -1 is also observed. In a previous study the appearance of this band has been associated with localized gap mode due to In vacancies. 28 29 It should be mentioned that all Raman signals we observed are much broader than that typical line width from bulk counterparts. 25 The pronounced asymmetric broadening can be due to the combined effects of disorder-activated scattering caused by various grain sizes in ultra-thin InN layer, 30 surface optical modes of InN 31 and Fano interference caused by discrete phonon modes and background continuum electron transitions in the system. 32,33 Despite the fact that our XRD results support the presence of compressive strain, the contribution of LPP+ cannot be ruled out due to the surface accumulation effect especially for InN nanostructures. 26 Further studies are currently underway to unveil the thickness-dependent optical and electrical properties of InN.

DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author upon reasonable request.

Supporting Information
(1) Growth details Plasma treatment The in-situ plasma treatment prior to the growth initiated when the temperature of the substrate was stabilized for 60 min at 6mbar and 320 o C (indicated by the pressure gauge and thermal couple connected to the bottom of the substrate susceptor). The power of 2800 W plasma with 100 sccm Ar and 50 sccm N2 was ignited and was last for 2 min.
ALD process After the plasma surface treatment, the system is purged by N2 (100 sccm) for 10 sec followed by ALD cycles starting with TMI pulse. An ALD cycle comprises 4 s TMI pulse, 10 s N2 purge, 12 s (NH3+ Ar + N2) plasma exposure and 6 s N2 purge. The TMI was commercially available from SAFC Hitech in epi-grade and was kept in a stainless-steel bubbler. The TMI bubbler was mounted in a Peltier element with the temperature set at 23 °C with an estimated vapor pressure of 1.46 mbar. The carrier gas for TMI is N2. The 4sec TMI pulse was chosen based on our experimental outcomes in which the so called "ALD window" with constant growth rate was found between 3 and 5 sec (ref. 19). 320°C was chosen the growth temperature as it gave the best surface morphology. It should be mentioned that a fill-empty step for TMI is employed to facilitate the TMI flow. Technically, TMI pulse can be seen as a combination of two parts; fill part and empty part. During the first part, the filling-part, the ALD valve is opened and a high flow of 300 sccm nitrogen carrier gas is passing through the bubbler. The high flow increases the total pressure in the bubbler. The partial pressure of the TMI in the bubbler is not expected to be affected by this. Then in the second part, the empty-part, the ALD valve is opened to the bubbler and a low flow of 100 sccm nitrogen carrier gas is flown through the bubbler. The high pressure in the bubbler, created by the high nitrogen flow in the fill-part, serves to empty the bubbler into the reaction chamber. This can be noted by carefully monitoring the pressure in the growth chamber; a short increase in pressure can be noted during the empty-pulse. The two pulses are separated by 10 s to allow gas mixing in the bubbler. For the designated pulse used for the supply of "N" to complete the second half reaction is done under 2800 W plasma with 100 sccm NH3 + 100 sccm Ar + 50 sccm N2.
(2) Grazing Incidence XRD and various scanning geometries to verify hetroepitaxy between InN and 4H-SiC Grazing Incidence-XRD (GIXRD) is a highly surface-sensitive technique used here to detect any crystal grains grown not orthogonal to the substrate surface. The GI-XRD 2θ scans are done in the range of 20-90° using a fixed 0.5° incident angle as shown in Fig. S1. As no peaks can be observed in the GIXRD diffractograms, all the InN grains have grown perpendicular to the substrate surface are indicated.  In addition, six-fold symmetric peaks are expected by fixing 2θ − ω to corresponding angles of either m (101 ̅ 0) and a (112 ̅ 0) planes and performing φ scans (φ = −180 to 180°). All configurations used to record the signals are summarized in Table S1. As can be seen clearly in Fig. S3, two sets of six-fold symmetric peaks are off-set by 30 o . It should be mentioned that the six-fold symmetric peaks recorded from SiC (101 ̅ 2) is the same as those recorded from InN (101 ̅ 0) . 1 Based on all the XRD results, the heteroepitaxial between InN and SiC is unambiguously verified by using XRD technique.  (2) InN thickness determined by the thickness-fringes in XRD 2θ-ω scans.
The thickness of the InN film could be determined by fitting/simulating the curves in in-plane XRD 2θ-ω scans. The software of X'Pert Epitaxy is employed for the simulation. The lattice constants of a=3.538 Å and c=5.703Å are set for wurtzite InN. In our case of InN on on-axis 4H-SiC substrate, the thickness of InN is the main variable in the simulation. The degree of relaxation has to be adjusted to reach best match as indicated in respective simulation curves. As can be seen in Fig. S4, by shifting the curves vertically, our experimental results match well with the simulated curves; for relatively thick to very thin InN samples. Four samples with the thickness between 53.5 and 2.3 nm are shown here. (3) X-ray ω-scan curves of the ALD grown InN on SiC substrates To facilitate the comparison of crystalline quality between InN and SiC substrate, the socalled X-ray rocking curves measurement was performed against InN (0002) and SiC (0004). The ω angles of the peak positions are shifted to zero and their respective peak intensities are normalized to 1. Results recorded from InN films with different ALD cycles (thickness) are shifted vertically for visual clarity. The instrumental resolution is 20 arcsec.
Perhaps owing to the test grade quality of SiC substrate, there exists curvature, defects and crystal grains which can cause small but detectable "mis-orientation". Such "mis-orientations" are reflected by the asymmetric profile as shown by solid lines in Fig. S5. The recorded profiles and FWHM of InN (0002) are nearly identical to those of SiC (0004). Such observation is found independent on the thickness of InN; from the thickest (53.5 nm) to the thinnest samples (2.3 nm) in our study. Despite the presence of non-perfection in the substrate, it strongly suggests near perfect hetero-epitaxy by our process. download file view on ChemRxiv supporting information_revised.pdf (337.38 KiB)