Study of Surface Morphology, Impurity Incorporation and Defect Generation during Homoepitaxial Growth of 4H-SiC Using Dichlorosilane

The growth of 4◦ off-axis 4H-SiC epilayers by chemical vapor deposition is studied using dichlorosilane as the Si precursor in a chimney reactor. The unintentional and intentional nitrogen doping at various C/Si ratios and N2 flow rates are investigated. The C/Si ratio has a significant influence on the growth rate, surface morphology, conversion of basal plane dislocations (BPDs), and generation of other defects (e.g., in-growth stacking faults and morphological defects). Addition of N2 has no obvious influence on growth rate and BPD conversion. It is preferable to grow high quality n+ epilayers at a C/Si ratio in the range of 1.3–1.8 along with the addition of a suitable amount of N2 in consideration of high growth rate, good surface morphology, and low defect density. In this case, the conversion ratio of BPDs to threading edge dislocations is greater than 99.8% regardless of N2 addition. Therefore >99.8% substrate BPDs can be buried in an n+ buffer layer, which is beneficial to SiC bipolar power devices. No special treatment prior to, during or after the epigrowth is necessary. The epilayers with doping concentration all the way from p− (∼1e15 cm−3) to semi-insulating, then to n+ (∼1e18 cm−3), can be achieved, giving a great range of flexibility in growth using dichlorosilane precursor. Further optimization of in-situ etching and growth conditions to eliminate step bunching has also been suggested. © The Author(s) 2014. This is an open access article distributed under the terms of the Creative Commons Attribution 4.0 License (CC BY, http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse of the work in any medium, provided the original work is properly cited. [DOI: 10.1149/2.0071503jss] All rights reserved.

4H-silicon carbide (4H-SiC) is a promising wide bandgap material for high power electronic devices, owing to its unique thermal and electrical properties such as high breakdown field, high thermal conductivity, and high electron saturation velocity. 1,2 Chemical vapor deposition (CVD) is a well-developed technique adopted in semiconductor industry to produce high purity and high quality thin films. 3 Homoepitaxial growth of SiC via the CVD technique is one of the key steps in the fabrication of high performance SiC devices. Currently, chloride-based CVD has attracted much interest to potentially replace the traditional silane based CVD for SiC epitaxial growth to reduce the Si droplet formation at high growth rates, because of the higher bonding energy of Si-Cl (400 kJ/mol) than Si-Si (226 kJ/mol). 4 Dichlorosilane (SiH 2 Cl 2 , DCS) is a gas (boiling point: 8.2 • C) at room temperature. It has been demonstrated that using DCS and propane (C 3 H 8 ) as the precursors, high crystalline quality 8 • off-axis SiC epilayers were achieved at a high growth rate (up to 100 μm/h). 5,6 At present, the standard off-axis angle of commercially available SiC substrates is lowered to 4 • to reduce the material loss during substrate preparation from the crystal boules.
Step-bunching is a typical surface feature of the epilayers grown on 4 • off-axis substrates, resulting in increase in surface roughness. 7 Morphological defects, specifically the triangular defects and inverted pyramids, are prone to be generated in epigrowth on low off-axis angle substrates. 8 Hence, determination of optimized growth conditions to produce SiC epilayers of low defect density with good surface morphology requires a systematic study when evaluating a new precursor system.
It is well known that basal plane dislocations (BPDs) are one kind of performance limiting defects in 4H-SiC bipolar devices. Most of the BPDs in the substrate convert into relatively benign threading edge dislocations (TEDs) during the epitaxial growth. 9,10 The rest of the BPDs propagate into the epilayer. Under device current stress, the BPDs in the active layer act as nucleation sites of Shockley staking faults causing increase of forward voltage drift. 11,12 This aspect has led to significant interest in the study of the BPD to TED conversion ratio during epitaxial growth. It has been reported that in the epitaxial growth on 8 • substrates, 70-90% BPDs in the substrate are converted into TEDs due to the image force between the off-axis surface and the BPD. 9,10 It has been reported that the BPD to TED conversion ratio is higher in the case of growth on 4 • substrates, [13][14][15] probably related to the interaction between macro-steps and BPDs 9 . Also the conversion of a BPD to TED is energetically favorable as discussed in detail in a reference. 16 The ratio of elastic energy per unit growth length for a BPD (W BPD ) to that for a TED (W TED ) can be derived from the references 16,17 and written as where E BPD and E TED are the elastic energies per unit length of dislocation line for a BPD and a TED respectively, and β is the substrate off-axis angle (≤8 • ). Since E BPD is very close to E TED , 16 equation 1 leads to W BPD W TED 1, i.e.,W BPD W TED . It is energetically favorable for a BPD to convert to a TED during epitaxy. And the smaller the substrate off-axis angle (β), the larger the W BPD W TED ratio is. Thus the BPD to TED conversion is greatly enhanced by the reduction of substrate off-axis angle.
It is also demonstrated that the BPD to TED conversion can be enhanced by growth interruptions, 18,19 substrate KOH etching, [20][21][22] or epilayer post annealing. 23 For SiC bipolar power devices, it is required that the BPD to TED conversion occurs in an n + buffer layer (i.e., BPDs are buried in the n + buffer layer.) where the electronhole recombination is too low to nucleate Shockley staking faults. 19 Therefore, increasing the BPD to TED conversion ratio in highly n doped SiC epilayers is important, especially for the device die size and yield considerations.
In this paper, the feasibility of applying DCS as the Si precursor in epigrowth on lower off-axis substrates is discussed. The unintentional and intentional nitrogen doping behaviors are investigated at various C/Si ratios and N 2 flow rates. The effect of C/Si ratio and N 2 flow on the growth rate, doping concentration, surface roughness, morphological defects, in-grown stacking faults, and BPDs are studied in detail. The optimal growth condition for achieving high quality, low defect density SiC epilayers is reported.

Experimental
A 3-inch (0001) 4H-SiC wafer with 4 • off-axis toward [1120] was obtained commercially. The Si-face was subjected to chemical mechanical polishing (CMP) and ready for epigrowth. Then the wafer was diced into 8 mm × 8 mm pieces which were then used as substrates for epitaxial growth in this study. The epitaxial growth was performed in a home-build hot-wall chimney CVD reactor which is  illustrated in Fig. 1. The sample holder, hot-wall and gas delivery tube were made of high purity graphite with TaC coating to reduce dusting and impurities. DCS and C 3 H 8 were used as the Si and C precursors respectively. Palladium membrane purified H 2 with a flow of 6 l/min (slm) was used as the carrier gas. N 2 gas was used for intentional doping. The growth temperature was 1600 • C monitored at the back side of the sample holder by a pyrometer. The temperature distribution in the reactor was simulated and calibrated to ensure that the real substrate surface temperature was very close to the monitored temperature. The working pressure was 107 mbar in all runs. A background pressure less than 1 × 10 −6 mbar was achieved by a turbo molecular pump. Since the substrates were polished with the CMP process, intentional H 2 etching prior to epigrowth was not necessary. During the temperature ramp up, 6 slm H 2 and 0.5 sccm C 3 H 8 were introduced in the reactor to avoid defects and degradation induced by excessive H 2 etching of the substrates. Unintentionally doped layers were achieved at C/Si ratios of 0.4-3.5 with a constant DCS flow rate of 4.5 sccm and varying C 3 H 8 flow rate of 0.6-5.25 sccm. Intentionally doped layers were obtained at the above C/Si ratios by adding 2-15 sccm N 2 . In order to evaluate epitaxial defects close to the epilayer/substrate interface, thin epilayers with a similar thickness of 6 μm were obtained by varying the growth time for the different growth conditions. The thickness of the epilayers was measured by Fourier transform infrared (FTIR) reflectance. The net doping concentration was measured by the capacitance-voltage (C-V) mercury probe method. The surface roughness and morphology were studied using atomic force microscopy (AFM) in the tapping mode. The root mean square (RMS) surface roughness was calculated in 20 μm × 20 μm scan areas. Morphological defects including carrots, invertedpyramids, growth pits, and triangular-defects on the as-grown epilayers were studied using a Nomarski optical microscope (NOM). After the above characterizations, the epilayers were etched in molten KOH at 550 • C for 3-8 minutes to delineate the dislocations. Basal plane dislocations (BPDs) and in-grown stacking faults (IGSFs) were specifically studied in terms of the effects of C/Si ratio and N 2 addition.

Results and Discussion
Growth rate and doping concentration.-The growth rate for the epilayers was investigated at various C/Si ratios (with a constant DCS flow rate of 4.5 sccm). Addition of N 2 does not have an observed influence on the growth rates. As shown in Fig. 2, the growth rate increases with C/Si ratio for C/Si < 1, and then reaches saturation for C/Si > 1. This indicates a conversion of the growth regime from C supply limited (for C/Si < 1) to Si supply limited growth (for C/Si > 1). This trend is quite common as has been observed in SiC growth using silane 7,13 or other chloride-based silicon precursors. 4 The unintentional doping concentration at various C/Si ratios is shown in Fig. 3. N-type doping is achieved for C/Si < 1.3 and P-type for C/Si > 1.8. This doping dependency behavior is in accordance with the site competition mechanism 24,25 which was also reported in silane and other chloride-based epitaxial growths. The N atoms replace C atoms in the SiC lattice; thus N incorporation is suppressed by increasing the C/Si ratio. The N incorporation decreases more efficiently at higher C/Si ratios (1.1-1.3). For C/Si ratio above 1.3, the donor concentration is lowered down to the level of residual acceptor impurities. Epilayers grown at C/Si = 1.5 have a doping concentration < 5 × 10 13 cm −3 which is the limit of our C-V mercury probe measurement. High resistivity or semi-insulating epilayers (i.e., nearly completely compensated doping) or even p-type conduction is achieved. The high purity semi-insulating epilayers are achieved through a compensation scheme that controls the formation of Si-related vacancies through defect competition. 6 Fig. 3 also shows that the net p-type doping concentration begins to saturate when C/Si ratio reaches 2.2, indicating that the doping type is dominated by nitrogen containing species which are released from the sample holder, hot-wall and insulation material.
The n-type doping concentration increases with the N 2 flow as shown in Fig. 4. It is noted that the N ≡ N bond in N 2 is extremely strong. At the growth temperature (1600 • C) N 2 conversion to N-containing species in gas phase is very low. Therefore the kinetic mechanism of surface reaction is important, in which N 2 dissociates and reacts with Si-sites on the surface. 26 Theoretically, the desired highly doped (up to 10 18 cm −3 ) epilayers could be achieved at any C/Si ratio with the addition of sufficient N 2 flow. It is therefore necessary to determine in which condition the epitaxial growth is the most efficient and the epilayers have the highest quality in terms of morphology and defects. This will be discussed in the following sections.  Surface roughness and step-bunching.-The surface morphology of the above epilayers was studied by AFM. The RMS roughness was measured in 20 μm × 20 μm areas and the values are plotted in Fig. 5. As for the unintentionally doped epilayers, the RMS value decreases with increasing C/Si ratio from 0.4 to 1.3, and the value has no significant change (RMS = 1.7-2.0 nm) for C/Si = 1.3-2.2. At C/Si = 2.6, the RMS value reduces to 1.35 nm. Most of the epilayers have obvious step-bunching (Fig. 6b-6e) that is usually observed in low off axis epitaxy. 27,28 The only exception is the epilayer grown at C/Si = 0.4, which shows a dramatically wavy surface (Fig. 6a) instead of a surface with clear macro-steps. This implies a different growth or step-bunching behavior at C/Si = 0.4. For C/Si = 1.1-2.2 ( Fig. 6b-6d), uniform step-bunching with the macro-step height 4-8 nm is observed on the 4 • epilayers grown, and the spaces between macro-steps are quite uniform (0.2-0.3 μm). At C/Si = 2.6 ( Fig. 6e), the macro-step height reduces to 1.0-1.5 nm. It is also found that addition of N 2 only causes slight degradation in surface morphology. Fig. 5 shows that the RMS values slightly increase by ∼0.5 nm while the N 2 flow rate increases from 0 to 15 sccm (the doping concentration increases by 2-4 orders of magnitudes). Considering the above results, it is preferable that the highly n-type doped epilayers are grown at a higher C/Si ratio (>1.3) by simultaneously adding N 2 during the growth. However, defect generation and BPD conversion should be considered when determining the most suitable C/Si ratios for epigrowth, which will be discussed in the following sections.
The trend exhibited by the surface roughness RMS value versus C/Si ratio indicates that in epitaxial growth using DCS, step-bunching is enhanced under a Si rich condition. This is opposite to most of the results observed in epigrowth using silane as the precursor, 13,29 in which step-bunching increased under a C rich condition. 30 However, Ishida et al. also reported that in growth using silane giant stepbunching occurs during epigrowth at very low or high C/Si ratios and they proposed that this is due to the generation of Si or C clusters on the terraces. 31 The enhancement of step-bunching under the Si rich condition was also found by Yazdanfar el al. in the growth using SiH 4 + C 2 H 4 + HCl precursor system. 27 The step-bunching obtained on samples in this study is most likely due to the non-optimized in-situ H 2 etching and/or low Cl/Si ratio (which is 2 in DCS) during the growth. The optimization of in-situ H 2 etching is suggested by reducing the C 3 H 8 flow rate (or using pure H 2 ), 32 adding HCl, and/or reducing the etching temperature 28 . Addition of HCl to obtain Cl/Si ratio of 3-5 during the growth was found to be an optimum solution for growth in chloride-CVD. 4 Reduction of the growth temperature may also help to moderate/eliminate the step-bunching, 28 though the generation of triangular defects and ingrown stacking faults should be taken into account in low temperature growth conditions.  Morphological defects.-The morphological defects (Fig. 7) are different on epilayers grown at different C/Si ratios. Adding N 2 does not show a clear influence on these morphological defects. The carrot defects (Fig. 7a) and shallow growth pits (5-50 μm dia. Fig. 7b) are mainly observed on the as-grown epilayers grown at C/Si ≤ 0.7 with total densities 5-50 cm −2 . At a C/Si ratio of 1.1, inverted pyramids (Fig. 7c) and triangular defects (Fig. 7d) are frequently observed on the epilayers with total densities <50 cm −2 . It is believed that the Si droplets contribute to the formation of morphological defects in SiC epilayers. 33 Results in this paper imply that Si droplets could still be formed in the Si rich condition using DCS precursor. It is known from a simulation study that in epigrowth using DCS, elemental Si is still generated from decomposition of DCS and its partial pressure is low enough to avoid Si droplet formation only at a very low working pressure (∼40 mbar). 34 The reason for Si droplet formation in growth using DCS is most likely due to the low Cl/Si ratio. Addition of HCl into DCS gas system (e.g., Cl/Si = 3-5) 4 during the growth is suggested to be a solution to completely eliminate Si droplet formation even in a Si rich condition.
For C/Si = 1.3-2.6, the epilayers are almost free of the above morphological defects (Fig. 7), except occasionally have some shallow growth pits (similar to amphitheater depressions observed by Burk et al. 35 ) in the density <20 cm −2 . With KOH etching of the epilayers, these shallow pits are not associated with any dislocations and believed to be relatively benign for device performance 35,36 compared with the other morphological defects (e.g., triangular defects). For C/Si ≥ 3.0, very large triangular defects (Fig. 7e) identified as 3C inclusion by Raman spectra are observed. A 3C inclusion can cover fairly large area (e.g., >30 mm 2 in our study) of the epilayer at C/Si = 3.5.
Basal plane dislocations (BPDs).-After the above characterizations, the epilayers were etched in molten KOH to delineate the dislocations on the epilayer surfaces. Fig. 8 shows the BPD density at various C/Si ratios for unintentionally doped epilayers. The BPD density reduces rapidly with increasing C/Si ratio from 0.4 to 1.1, and then does not show significant change over C/Si = 1.1-2.6. The observed lowest BPD density of 0-5.6 cm −2 has been achieved in a wide C/Si window. This trend is different from what was reported by Kallinger et al. in the growth using silane precursor. 37 They found that the minimum BPD densities of <15 cm −2 were achieved for C/Si∼1. For lower and higher C/Si ratio, the BPD densities increased. In our study, all of the epilayers have a similar thickness of 6 μm. The change of the BPD density with C/Si ratio could be interpreted by the influence of growth rate. Canino et al. reported that in epigrowth using trichlorosilane precursor the BPD to TED conversion ratio was enhanced by increasing the growth rate. 38 A similar phenomenon may take place in the epigrowth using DCS precursor. For a C/Si ratio of 0.4 the growth rate is low (Fig. 2), resulting in low BPD to TED conversion ratio and thus exhibits a high BPD density (Fig. 8). While the growth rate is higher and almost constant for C/Si ratios of 1.1-2.6 ( Fig. 2), steadily low BPD densities are achieved (Fig. 8) because of efficient BPD conversion.
It is also noted that in a slightly C rich condition (C/Si = 1.1-1.5), while the addition of N 2 will achieve n-type doping concentrations up to ∼2 × 10 17 cm −3 , there is no obvious change in BPD densities (Fig. 9). The BPD density in the substrates was found to be ∼5000 cm −2 . Hence the BPD to TED conversion ratios greater than 99.8% within 6 μm thick epilayers are achieved for C/Si = 1.1-1.5 regardless of N 2 addition. This is different from what was reported in 8 • epitaxial growth where N 2 inhibits the BPD conversion. 39 The  high BPD conversion ratio in our study is among the best reported up to now. 13,14,27,37 It should be noted that the previously reported results on 4 • epitaxy were obtained in low doped (or unintentionally doped) epilayers, while in this study, this high conversion ratio can be directly achieved in higher n doped epilayers. Therefore the BPD faulting can be effectively suppressed, being of great benefit to SiC bipolar power devices.
In-grown stacking faults (IGSFs).-Generation of IGSFs was also studied in the above conditions. After molten KOH etching of the epilayers, the IGSFs exhibit as two shell-like etch pits (like BPDs) at the down-step side, one is along step-flow direction, another is inclined (as shown in Fig. 10 inset). These two shell-like etch pits may be connected with a shallow groove line. 15 These stacking faults are Shockley-type faults with 8H stacking sequence. 40 SiC Schottky barrier diodes containing these IGSFs tend to show higher leakage current and reduced breakdown voltage. 41 Figure 10 shows the IGSF density at various C/Si ratios. No evident influence of N 2 addition on the IGSF generation is found. An IGSF density <20 cm −2 was achieved over C/Si = 1.1-1.8. Significant increase of IGSF density is observed at low (≤0.7) or high (≥2.2) C/Si ratios. Abadier et al. postulated that generation of IGSF is correlated with 2D nucleation of adsorbed species on the growing surface at the Figure 11. Chart summary of 4 • 4H-SiC epitaxial growth using DCS and C 3 H 8 precursor. initial growth condition. 40 Data in Fig. 10 imply that either excessive Si or C containing species can enhance the adsorption of unreacted species on the surface contributing to the nucleation of IGSFs. This process is even more severe in the C excess condition.
To reduce IGSF densities at low or high C/Si ratios, elevation of growth temperature is helpful 42 since this will increase desorption of surface species. However a higher growth temperature may also reduce the growth rate and increase the surface roughness due to severe step bunching. 43 Providing a clean growth environment, 44 reducing the initial growth rate, 40 and/or optimization of surface pretreatment 42 may also help to reduce IGSF density.

Conclusions
The characteristics of 4 • 4H-SiC epilayers grown using dichlorosilane (DCS) as the Si precursor are summarized in Fig. 11. DCS is a promising chlorine based Si-precursor to achieve high quality epilayers with low defect densities. A wide range of doping concentration can be well controlled by the C/Si ratio and simultaneous addition of N 2 gas. The epilayer morphology and defects are primarily determined by the C/Si ratio. High quality epilayers without morphological defects are obtained at C/Si in the range of 1.3-1.8, even with the addition of N 2 for achieving higher n-type doping concentrations. A BPD to TED conversion ratio greater than 99.8% and IGSF density of 0-20 cm −2 can be achieved in this C/Si ratio range even in high n doped epilayers.