To realize large-diameter 3C-SiC substrates, chemical vapor deposition (CVD) heteroepitaxy on silicon is currently the most mainstream and mature route. It must be clearly understood that, limited by the silicon melting point (1414 °C), silicon-based CVD cannot directly grow bulk 3C-SiC; its product position is a high-quality 3C-SiC seed template, which must later be combined with sublimation epitaxy (SE) to complete the bulk substrate. This article breaks down the CVD stage in detail.

Fig. 1 Process flow of silicon-based 3C-SiC CVD heteroepitaxy.
1. The Classic Process Framework: The Nishino Multi-Step Method
The classic CVD process for 3C-SiC on silicon originates from the multi-step scheme proposed by Nishino et al. in 1982, and it remains the foundational framework for silicon-based 3C-SiC production:
1. HCl etch pretreatment: high-temperature H₂/HCl etches the silicon surface to remove the damage layer and native oxide, yielding an atomically flat surface;
2. High-temperature short-time carbonization: only the carbon source (propane, C₃H₈) is supplied to grow an ultra-thin 3C-SiC buffer layer in situ on the silicon surface;
3. Thick-film epitaxial growth: silane (SiH₄) and propane (C₃H₈) are supplied simultaneously to grow a thick 3C-SiC film continuously on top of the buffer layer.
2. The Carbonized Buffer Layer: The First Gate for Interface Quality of the Substrate Template
During carbonization, propane is supplied in a hydrogen atmosphere; carbon reacts with silicon atoms on the substrate surface to form an extremely thin 3C-SiC buffer layer that accommodates the huge 19.7% lattice mismatch. Carbonization temperature is the most critical knob for interface quality.
2.1 TEM Evidence for Carbonization Temperature
The paper compares two carbonization temperatures by cross-sectional TEM:
• 1020 °C (low-temperature carbonization): numerous channels and deep cavities appear inside the silicon substrate; the interface is rough, with large unprotected silicon areas that are subsequently etched by H₂; 3C-SiC even intrudes into the silicon bulk, and polycrystalline diffraction rings appear in the electron diffraction pattern.
• 1120 °C (high-temperature carbonization): the interface is continuous and low in roughness; no channels or cavities remain in the silicon bulk; the diffraction pattern shows high-quality single-crystal 3C-SiC features with no polycrystalline rings.
Fig. 2 Cross-sectional TEM comparison after carbonization at 1020 °C vs 1120 °C.
2.2 Effect of Substrate Orientation on Buffer-Layer Growth Kinetics
The squared buffer-layer thickness is observed to be linear with reaction time, proving that carbonization is limited by silicon-atom diffusion. The buffer-layer growth rates of the three common silicon orientations rank as:
(110) > (100) > (111)• (110) slope: 1.85 ± 0.08 nm²/min
• (100) slope: 0.67 ± 0.09 nm²/min
• (111) slope: 0.19 ± 0.01 nm²/min
Fig. 3 Squared buffer-layer thickness vs carbonization time for the three orientations.
2.3 Two Process Paths for Reducing Interface Voids
Interface voids originate from silicon atoms diffusing outward from the substrate to participate in the reaction. After extensive process development, two mature optimization paths have emerged:
1. Raise the C/H₂ ratio: a higher carbon-source fraction increases the 3C-SiC nucleation density on the silicon surface and reduces H₂ etching of the silicon bulk. The paper shows that as C/H₂ rises, the effective void area falls from 11% to 5%, with void density dropping in parallel.
2. Introduce silane (SiH₄) during carbonization: an external silicon source reduces the outward diffusion of substrate silicon atoms, thereby suppressing the size and number of voids.
In addition, the ramp-up between carbonization and high-temperature growth must not be overlooked: optimizing the ramp curve lowers wafer curvature and reduces interface protrusions. For antiphase-domain (APD) suppression, low-pressure LPCVD clearly outperforms atmospheric APCVD: LPCVD can eliminate APD completely within 5 μm, whereas APCVD still shows antiphase domains at 17 μm.
3. Thick-Film Growth: Self-Annihilation of Defects and Process Trade-offs
After the buffer layer, raising the temperature and introducing silane starts the thick 3C-SiC epitaxy stage. The Si/3C-SiC interface has a 5:4 lattice match (five 3C-SiC unit cells match four Si unit cells), forming a Lömer-type array of 90° misfit dislocations with a spacing of about 1.61 nm that relieves most of the lattice-mismatch stress. Residual stress, however, still induces bulk defects such as microtwins, stacking faults, and antiphase boundaries.
3.1 Microtwins (MT): A "Self-Annihilating" Defect
Microtwin density decays rapidly with film thickness: above 3.5 μm, the microtwin fraction is generally below 1%; in some experiments microtwins are fully annihilated at only 2 μm. MT also correlates strongly with growth rate: at a fixed thickness, the lower the growth rate, the lower the MT density; a higher Si/H₂ ratio (faster growth) raises MT density.
3.2 Stacking Faults (SF): The Hardest-to-Eliminate Electroactive Defect
Stacking faults start at the Si/3C-SiC heterointerface and extend upward into the film, carrying strong electrical activity. A key figure: a one-order-of-magnitude rise in linear SF density raises leakage current by three orders of magnitude under 600 V reverse bias—directly destroying device reliability.
SFs self-annihilate: oppositely oriented stacking faults cancel when they meet. In very thin films the SF density can be as high as 10⁹ cm⁻², dropping exponentially with thickness; but SFs never vanish completely and reach a saturation value—even at 250 μm, a linear density on the order of 5×10³ cm⁻¹ remains. The thicker the film, the farther apart the SFs and the lower their collision-and-annihilation probability.
Fig. 4 Double-log plot of microtwin (MT) and stacking-fault (SF) density vs epitaxial thickness (MT drops rapidly toward ~0; SF drops then saturates).
3.3 X-Ray Criterion of Crystal Quality: The FWHM Trade-off
The full width at half maximum (FWHM) of the (200) X-ray rocking curve is a direct measure of crystal quality. The paper gives two key rules:
- The thicker the film, the smaller the FWHM and the better the crystal quality;
- At a fixed thickness, the higher the growth rate, the broader the FWHM and the worse the crystal quality.
This means a high-quality CVD 3C-SiC seed template requires a lower growth rate, at the cost of throughput—a trade-off the substrate product must make between yield and cost.
4. Wafer Warpage: An Engineering Problem the CVD Route Must Address Head-on
The silicon melting point is 1414 °C, and the growth ceiling for 3C-SiC on silicon is locked below 1350 °C, so direct CVD growth of a thick bulk layer hits a temperature ceiling. The more practical issue is thermal-mismatch warpage: during cooling, the differential contraction of 3C-SiC and silicon creates tensile stress that arches the wafer upward. The paper uses a modified Stoney equation to relate film stress to wafer curvature:
where m = E_film/E_sub and n = h_film/h_sub. Warpage depends on epitaxial thickness, substrate thickness, crystal orientation, and elastic modulus. Severe warpage directly causes subsequent lithography, etching, lapping, and polishing to fail.
Fig. 5 Warpage of the silicon-based 3C-SiC epitaxial wafer.
| Summary: Silicon-based CVD can produce 8-inch large-diameter 3C-SiC seed templates and is the starting point of the entire 3C-SiC substrate supply chain. Carbonization temperature, C/H₂ ratio, growth rate, and film thickness are the four process knobs for controlling voids, twins, and stacking faults. However, CVD itself cannot yield a bulk substrate—the high-quality CVD seed template is the prerequisite for the sublimation epitaxy (SE) described in the next article that makes bulk 3C-SiC substrates. |