3.1. General Trends Observed in Morphology of 3C-SiC Epilayers
Morphologies and common trends observed on the grown films are, first, reported and discussed.
Figure 1 illustrates the morphology obtained on different 3C-SiC/AlN/Si-grown samples.
Figure 1a,b shows the morphologies of the 3C-SiC layers obtained after 3 min of SiC growth (i.e., including the C/Si “rich” nucleation followed by a 90 sec and C/Si = 1 growth stage), for MBE AlN/Si(111) 4°
off axis and
on axis. Similar trends were obtained using the MOCVD AlN/Si templates. The development of triangularly shaped grains, forming either as a stepped structure (
Figure 1a) or more randomly distributed with some coalescence defects (
Figure 1b), was observed. These differences of morphologies can be attributed to the initial presence of the AlN steps, perpendicular to the Si [1-10] crystalline direction on
off axis substrates (see
Figure S2 in the Supplementary Material). With a similar growth duration, a comparable quantity of SiC was assumed to form for both cases. We noted that islands developing on the
off axis AlN/Si template (
Figure 1a) tended to grow on the surface by forming triangles pointing either perpendicularly to the AlN steps or in the opposite direction with a truncated side, as indicated in the figure. This point is discussed thoroughly hereafter, but this is a typical trend observed during the growth of a three-fold symmetry material growing on a six-fold symmetry substrate (twinning mechanism), as in the case of 3C-SiC growth on hexagonal SiC substrates [
24,
25].
Figure 1c,d shows the morphologies observed on the 3C-SiC/AlN/Si samples with a longer duration of growth (20 min, ~300–320 nm, measured by FTIR). We obtained, for both the
off and
on axes, a SiC surface presenting some large domains, separated from each other by trenches. The AFM measurements (not shown) allowed us to determine that, on a given visible domain, the roughness was quite low (RMS = 1 nm on a 5 × 5 µm
2 scan). On the other hand, the trenches that formed domain boundaries had typical depths in the range of 10–20 nm. According to
Figure 1a,b, the observed grains should correspond to some enlarged grown SiC twinned domains. It can be noticed that, for almost equivalent SiC film thickness, the SiC domains appear qualitatively larger on the
off axis templates, showing a benefit to using initially misoriented substrates. In addition,
Figure 1e shows the morphology obtained on a 320 nm-thick 3C-SiC grown on an
on axis AlN/Si(110) template. It presents the same microstructure but with larger twinned domains compared to
Figure 1c,d. This opens the question of the underlying AlN influence on SiC growth.
Indeed, if some differences between the
on and
off axis templates can be attributed to the stepped-like feature of the
off axis templates, noticeably larger domains on the AlN/Si(110) were obtained even for non-disoriented substrates. As mentioned before, AlN grows with the same [0001] growth direction, both for Si(111) and Si(110). Nevertheless, it was shown that, according to a favorable lattice match, AlN presented some better structural properties on Si(110) than on Si(111) [
22]. This appears interesting to favor the development of large SiC domains.
Furthermore, as mentioned before, some growth processes occurred simultaneously on AlN/Si and 6H-SiC, and the latter was taken as a reference sample for validating the growth process.
Figure 1f shows the morphology obtained on such a 6H-SiC substrate. A smooth surface, featured by the development of steps, is clearly visible. The presence of a domain included within a larger one can be noted. This appears to be typical of a 3C-SiC film formed on a 6H-SiC, with small domains included in larger ones, revealing the development of twinned domains [
26]. The formation of such a typical 3C-SiC film on a 6H-SiC polytype confirms that the adopted growth conditions were suitable for homoepitaxial growth and validates the chosen C/Si ratios. It can also be noticed that the lateral expansion of a given domain was, by far, larger than those observed on the AlN/Si templates, revealing a possible influence of the underlying AlN film.
Another point of interest relies on the similar morphologies observed both for
on axis and
off axis samples (
Figure 1c,d). According to the difference in morphologies observed for thinner films, different morphologies should have been expected for larger thicknesses. A possible assumption for explaining this can be the possibility of two separated SiC domains growing on the same AlN step with a 180° misorientation from one another, with an equivalent probability (twinning). Therefore, the use of a stepped surface becomes an insufficient condition for removing the twinning mechanism because, as long as some twinned domains merge continuously during the film thickening, no preferential domain orientation becomes dominant. More details about the twinning issue are given in the next sections.
Nevertheless, despite the presence of twinned domains, some interesting features must be underlined. First, the typical sizes of SiC-grown grains are quite large relative to the film thickness. Indeed, although they have irregular shapes, grains can expand laterally over 2 µm along a direction, namely about ten times larger than the thickness. For comparison, in the case of growth of 3C-SiC directly on Si(100)-oriented substrates, epitaxial grains can present two different
in-plane epitaxial relations and form antiphase domains differing from each other by a 90° rotation on the (100) growing plane [
27]. Thus, domains are separated by boundaries (anti-phase boundaries, APBs) which constitute a morphological defect very similar, in terms of roughness, to the grain boundaries presently observed. For 3C-SiC/Si(100), the lateral expansion of each domain is comparable to the film thickness. Similar observations were made for 3C-SiC/Si(111) [
28]. It can be mentioned that the presence of prominent APBs on the surface of the as-grown SiC epilayers can be solved using a chemomechanical polishing (CMP) process which allows for reducing the overall RMS, including APBs, to below one nm [
29]. It can be expected that the present observed grain boundary-induced roughness could be reduced in the same way. Other points of interest of the grown 3C-SiC layers rely on their structural properties, as discussed hereafter.
3.2. Structural Investigations
The structural properties of the grown films were investigated using XRD and TEM-HRTEM analysis.
Firstly, the 2θ−ω scans (
Figure 1a) confirm that the 3C-SiC films were (111)-oriented, and no other crystalline direction was observed. It must be highlighted that the same orientation was conserved on the AlN films grown either on the Si(111) or Si(110) substrates.
According to the symmetry difference between wurtzite AlN and cubic SiC, the latter is expected to develop twinned domains, as previously mentioned. This twin formation is similar to that encountered during the growth of cubic SiC on hexagonal SiC polytypes [
24,
25]. The azimuthal scans, recorded along diverse
out-of-plane reflections, demonstrate the presence of twinned domains in SiC, as illustrated in
Figure 2b,c. Indeed, six peaks, with a 60° spacing, were recorded along the SiC(220)
out-of-plane direction, corresponding to the six SiC{220} planes (top spectra in
Figure 2b,c), whereas only three, with 120° spacing, were expected in the case of a single-domain SiC(111)-oriented film. This clearly denotes the presence of SiC twinned domains, differing from each other by a 180° rotation and corresponding to the grains observed by SEM in the previous section. From these measurements, the
in-plane epitaxial relations of the SiC/AlN heterostructures grown on
(i) Si(110) and
(ii) Si(111) could be determined, respectively, as follows:
- (i)
3C-SiC [0-11]//2H-AlN [11-20]//Si [1-10]
and 3C-SiC[-211]//2H-AlN [1-100]//Si [001]
- (ii)
3C-SiC [0-11]//2H-AlN [11-20]//Si [01-1]
and 3C-SiC [-211]//2H-AlN [1-100]//Si [-211]
These epitaxial relationships are similar to those already reported for SiC growth at a low temperature using the PLD technique both for AlN and SiC formation [
4] on Si(110) substrates. According to the almost equivalent diffraction intensities of different domains, we can conclude that the statistical occurrence of both domains is equivalent. In the present case, it must be highlighted that the use of
off axis AlN/Si(111) templates did not reduce the relative contribution of a given domain (see
Figure 2b,c). Our results differ noticeably from those of [
4], where a single-domain SiC film was obtained using the
off axis AlN/Si(110) substrates. Discrepancies between this observation and our results can be related to the very different growth conditions between our work and those of [
4].
In the case of 3C-SiC (three-fold symmetry) growth on hexagonal (4H- or 6H-) SiC substrates (six-fold symmetry), the formation of twinned domains is an issue that sparks an ongoing debate. Some authors have mentioned that twin suppression can be driven by adjusting the growth parameters [
25,
26]; others reported the need for adequate substrate preparation [
30]. For the growth of 3C-SiC on 2H-AlN, many parameters can also influence the initial nucleation of SiC on AlN. As in the case of growth on hexagonal SiC, a step-flow growth mode [
31] should favor the occurrence of single-domain SiC. However, our data show that the use of stepped AlN, formed on
off axis Si substrates, is insufficient to suppress twinned SiC domains and, in the absence of a step-flow growth mode, these domains can form at two different locations on a given AlN step, as underlined before. Our morphological observations (see
Figure 1a) clearly indicate that this is the case with the appearance of a 3D growth mode, favored by our experimental growth conditions.
Nevertheless, the grown SiC films present an attractive feature. Indeed,
Figure 3 shows the full width at half-maximum (FWHM) of the SiC(111) symmetric diffraction line as a function of the film thickness. In addition to these values, FWHMs of SiC(111) measured on SiC films grown directly on Si(111) substrates in a lab using the optimized carbonization–epitaxy growth process described in [
32] are reported. We noted the very reduced values of the FWHMs for SiC films on AlN were thinner than—300 nm compared to SiC films on Si grown with the carbonization–epitaxy process. For SiC grown on AlN, the FWHMs remained below 2500 arcsec, even for the thinnest films, whereas a very strong broadening of the SiC(111) FWHMs for SiC on Si was observed for thicknesses below 200 nm. This confirms that, according to the very reduced lattice mismatch between SiC and AlN, SiC can grow with a lower density of crystalline defects than on silicon, where the carbonization stage is required and forms, in the early stage of growth, a very defective layer for this range of thickness [
9]. This point is particularly important as it shows the ability to obtain a 3C-SiC film of low thickness with crystalline quality is, by far, better than using only Si as the substrate. This should be very beneficial for the design of microelectromechanical structures where the structural properties influence the mechanical properties and, thus, can potentially impact the physical parameter of devices [
5,
6].
TEM observations allowed us to go a step further in the analysis.
Figure 4 shows two TEM cross-section images of the interface between SiC and AlN along the <1-10> SiC zone axis. The large-scale image of
Figure 4a allows for observing the presence of numerous stacking faults. The formation of stacking faults is a very likely mechanism in 3C-SiC, according to the very low energy of formation, and it is largely reported during growth on silicon. The presence of twins is also confirmed by the good match between the simulated selected-area electron diffraction (SAED) patterns, obtained by assuming the existence of twinned domains, and experimental ones (see the inset in
Figure 4a). Zooming in on the SiC/AlN interface (
Figure 4b) reveals that many SiC zones close to the interface presented a hexagonal arrangement, which appeared as a 6H-SiC polytype, as evidenced by their diffraction patterns (inset in
Figure 4b). These zones can expand some tens of nm away from the interface.
Furthermore,
Figure 5 is an HRTEM image of the AlN/SiC interface along a <11-2>Si/<10-10>AlN zone axis, revealing a structural modification occurring after either the SiC growth or simple annealing of the AlN/Si at 1330 °C. A thin (~3–5 nm) layer, adopting a crystalline arrangement that differs from that of wurtzite AlN and Si, was observed. The Fourier transform (FT) of the image (
Figure 5b) is the superposition of the Si<11-2> and AlN<1-100> diffraction patterns, plus extra diffraction spots (blue arrows) demonstrating triple
in-plane and double
out-of-plane periodicities. This is characteristic of the Al
5+αSi
5+δN
12 compound obtained on AlN layers on sapphire by the diffusion of Si at temperatures between 1350 °C and 1550 °C and first observed by some of the present authors [
33]. The Si/Al ratio is close to 1 in this alloy, which demonstrates a large AlN/Si interdiffusion process during the SiC growth. To the best of our knowledge, such an AlSiN alloy formation on an AlN/Si heterostructure has never been reported.
3.3. Electrical Behavior and AlN Structural Modifications Induced by Process
According to the interest in using AlN as an intermediate insulating layer in the SiC/AlN/Si heterostructure, we investigated some electrical-related trends induced by the SiC growth process. At first, an assessment of the electrical behavior was conducted by vertical I(V) measures.
Figure 6 reports the measurements made contacting the upper face with a Hg probe and the back of the silicon substrate with a metal chuck: (i) on a 100 nm thick AlN/Si(111)—(p type—1–10 ohm.cm) template before the SiC growth, (ii) on the same AlN/Si template annealed at a temperature similar to that used for growing SiC and (iii) on a 270 nm thick SiC layer on AlN/Si. Similar features were obtained using thicker AlN films or using Si(110) substrates. A very low vertical current was recorded for any voltages for the non-annealed AlN/Si heterostructure, attesting to a very high insulating characteristic of the AlN film despite the presence of structural defects.
After annealing, a significantly higher current was recorded through the AlN/Si heterostructure. This trend was further enhanced when considering the complete 3C-SiC/AlN/Si heterostructure, with an ohmic character of the I(V) characteristics and current values seven orders of magnitude higher than in the case of the initial AlN/Si template, evidencing a vertical current flowing through both SiC and AlN.
The electrical flow through SiC can be expected, according to the electrical activity of inclined stacking faults, clearly visible on the TEM images (
Figure 4) [
34]. However, the fact that the current could flow vertically across the AlN film is more intriguing, especially regarding its initial insulating character before SiC growth or annealing. This lets us assume that the high-temperature annealing and/or the CVD SiC growth conditions can affect the AlN electrical properties.
The XRD measurements carried out on these heterostructures are helpful for understanding these trends. Indeed,
Figure 7a reports the azimuthal dependence of the FWHMs recorded on an as-grown AlN(50 nm)/Si(111) (4°
off axis) heterostructure, as well as the FWHMs recorded after either annealing or the regrowth of SiC on this heterostructure. To be sure of performing relevant comparisons, the initial AlN/Si wafer was sliced into two pieces, each of them being dedicated to a specific treatment—annealing or SiC growth.
Figure 7b shows the data recorded on the AlN(100 nm)/Si(111) (
on axis), excluding annealing, which was not performed.
For the AlN grown on
off axis substrates (
Figure 7a), a strong anisotropy of the recorded values, dependent on the azimuthal angle, with two FWHMs maxima separated by 180° and corresponding to the [
11,
12,
13,
14,
15,
16,
17,
18,
19,
20] crystalline direction (i.e., the off-cut direction), was observed. The realization of either thermal annealing at 1330 °C or a CVD SiC growth process on such AlN films led to a significant evolution of the FWHMs. This was especially true for the
off axis heterostructures (
Figure 7a) for which FWHMs were reduced along the crystalline directions previously exhibiting the largest values. We observed that this effect was less pronounced after the simple annealing compared to after the SiC growth process. We also noted that only a slight evolution was observed for the crystalline directions initially presenting the lowest FWHMs ([1-100] and [-1100]). On the other hand, the decrease in the FWHMs was less pronounced (within 10%) in the case of the
on axis heterostructures, for both the Si(111) (see
Figure 7b) and Si(110) substrates.
Such observations indicate that both thermal annealing and the CVD SiC growth process can influence the structural properties of the underlying AlN. In the case of thermal annealing, only a physical mechanism can be involved (a thermally induced mechanism), whereas, in the case of SiC growth, both physical and chemical mechanisms could be involved, the latter being related to the reactive gas environment used for the growth. Nevertheless, according to the slight differences in behavior observed between thermal annealing and the SiC growth, it seems probable that the chemical effect would be low compared to the thermally induced effect. This is supported by previous works. Indeed, it was shown that the structural properties of AlN films can be greatly modified by performing specific high-temperature annealing [
35]. In this work, the authors used sputtered AlN deposited on sapphire templates. The annealing temperatures were in the range of 1600–1700 °C. A reduction of the dislocation density was evidenced as a consequence of the thermal annealing.
In our work, we dealt with a quite different system (AlN with better initial crystalline quality, grown by MOCVD on a silicon substrate) and with lower process temperatures (typically 1100 °C). However, the results in
Figure 7 tend to demonstrate that a dislocation healing should have occurred according to the FHWM reduction after annealing and the CVD growth process, with a bit more of a pronounced effect after the CVD SiC growth. Nevertheless, it is not possible to definitely conclude about the specific contribution of the SiC growth stage on dislocation healing. This point remains open for discussion.
It is clear from our work that the temperature conditions were sufficient to modify the AlN microstructure and that the SiC growth conditions induced further modifications, according to the significant increase in the vertical current flow through the entire SiC/AlN/Si heterostructure (see
Figure 6). This could be related to the significant cross-doping of AlN with Si or C species during the growth and the reverse case, namely the incorporation of Al or N in SiC during its growth [
10,
17].
A strong indication for the presence of a cross-doping effect in our experiment is given by the infrared reflectance measured on some 3C-SiC/AlN/Si samples.
Figure 8 shows a reflectance spectrum acquired on the SiC(270 nm)/AlN(100 nm)/Si heterostructure. This spectrum shows the typical SiC and AlN TO-phonon-related structures close to 660 and 800 cm
−1 [
36,
37]. These structures were superimposed onto a high continuous background that slowly declines after—1000 cm
−1.
FTIR experimental reflectance spectra of materials or stacked heterostructures can be modeled by a classical dielectric model, where reflectance can be calculated from both the definition of the dielectric functions of each material present in the stack and the Fresnel coefficients at each interface [
36,
38]. The presence of free carriers (and, thus, doping) can significantly affect the reflectance spectrum and can be modeled by introducing a plasmon term (Drude term) to the dielectric function of the material. This Drude term is well separated from the lattice contribution (vibrations), which are described with a Lorentzian term. Thus, a qualitative assessment of a possible cross-doping effect could be carried out by comparing the experimental data with a simple dielectric calculation where we assumed different carrier concentrations. In the present work, a home-made code under a SCILAB environment [
39] was used for calculating the dielectric–based reflectance of a 3C-SiC/AlN/Si heterostructure, taking as phonon parameters those reported in the literature [
37,
38] and considering a plasmon term, both in the SiC and in the AlN layers, as variable parameters, ranging from 200 to 2200 cm
−1 for targeting carriers concentrations of 10
18, 5 × 10
18, 10
19, 5 × 10
19, and 10
20 cm
−3. In addition to the experimental data,
Figure 8 shows the calculated reflectance spectra obtained with these different doping levels. The background, against which the SiC and AlN phonon-related peaks are observed, was soundly enhanced by the introduction of plasmon terms. For obtaining a qualitatively good agreement between the experiment and calculation, a carrier concentration roughly in the mid-10
19 cm
−3 was required. Obviously, our approach remains a rough estimation, as our model cannot take into account the presence of carrier concentration gradients at the interfaces or inside layers. From an experimental point of view, these gradients are very likely to occur, as cross-doping should generate diffusion mechanisms. At first, the calculated reflectance spectra show trends that support the presence of a noticeable cross-doping effect, at quite a high level, between the SiC and AlN. Further experimental data are required to go further into detail on this point.