Improving the Through-Thickness Thermal Conductivity of Carbon Fiber/Epoxy Laminates by Direct Growth of SiC/Graphene Heterostructures on Carbon Fibers

Poor thermal conductivity in the through-thickness direction is a critical limitation in the performance of carbon fiber-reinforced polymer (CFRP) composites over a broad range of applications in the aviation industry, where heat dissipation is required (e.g., battery packs, electronic housing, and heat spreaders). In this work, it is demonstrated for the first time that a hierarchical network of vertically oriented graphene nanoflakes (GNFs), with nanoconfined silicon carbide (SiC) nanocrystals, self-assembled on carbon fibers (CFs) can provide significant improvement to the thermal conductivity (TC) of CFRPs in the through-thickness direction. The vertically aligned SiC/GNF heterostructures were grown directly on CFs for the first time by single-step plasma-enhanced chemical vapor deposition (PECVD) employing tetramethylsilane (TMS) and methane (CH4) gases at temperatures of 800 and 950 °C. At the deposition temperature of 950 °C, the controlled introduction of SiC/GNF heterostructures induced a 56% improvement in through-thickness TC over the bare CFRP counterparts while simultaneously preserving the tensile strength. The increase in thermal conductivity is accomplished by SiC nanocrystals, which serve as linkage thermal conducting paths between the vertical graphene layers, further enhancing the smooth transmission of phonons in the vertical direction. The work demonstrates for the first time the unique potential of novel SiC/GNF heterostructures for attaining strong and thermally conductive multifunctional CFRPs.

SiC/GNFs on CF at 800 ºC: gSi800°C and SiC/GNFs on CF at 950 ºC: gSi950ºC. (b) and (c) are Raman spectra of gSi950ºC and gSi800°C respectively, of the dotted region indicated at (a),

S3. XRD measurements
The XRD spectra revealed only a carbon peak around 25.5 degrees with no differences observed between the two samples (same FWHM, position of the carbon band, intensity). The green highlighted area for both samples (increased intensity for gSi950°C) is due to the copper tape used to mount the samples on the XRD sample holder. Error bars represent standard deviation from 5 independent measurements.

S7. Finite element analysis (FEA) of the heat dissipation into the CFRP matrix
Finite element analysis (FEA) software (Autodesk Fusion 360) was used to compare the heat dissipation of GNFs and the SiC/GNFs structures into the CFRP matrix. In a typical simulation ( Figure 12), a simple thermal model under steady state conditions was selected in the software, where a temperature of 60 °C was applied linearly at the bottom face of the simulation matrix.
The parameters used in the simulation are being quoted in the following table:

S8. SiC yield estimation
The SiC yield for our PECVD reactor can be defined by the following expression: We used 10 sccm (0.01 L/min) total gas flow in all fabricated samples (8 sccm TMS + 2 sccm CH4), 30 min total deposition time and the same deposition area. This practically means that by changing the % of SiC (we define as % of SiC, the % of Si-C bonding estimated from the XPS Si2p analysis (reported in Table 4) multiplied by the total amount of Si-(reported in Table  3) in every sample, we can roughly define the yield of the PECVD reactor and investigate the effect of different temperatures: YieldgSi800 C = 21 cm × 2 cm * (0.06/100) 30*0.01 = 0.085 cm 2 /L YieldgSi950 C = 21 cm × 2 cm * (3.91/100) 30*0.01 = 5.5 cm 2 /L Hence, when comparing the two yields we can clearly see that: YieldgSi950 C / YieldgSi800 C = 5.5 / 0.085 ≈ 65 That means that by increasing the temperature from 850 C to 950 C there is an enormous increase in the yield of SiC growth by 65 times.

S9. Void content (%) calculations
We used some cross section SEM micrographs to estimate the void content (%) of the bare CF composite. Because all samples were fabricated together using the VARI method we assume that the void contents are similar in all samples. The calculations were based on optical observations of the voids and more specifically on the %Area of the highlighted (red colour) voids, in comparison with the total %Area of every micrograph (units are pixels). With the help S9 of ImageJ, it was possible to obtain the %Area of the highlighted voids, which however is overestimating the real void content, because there are some other areas on the micrograph which are not actually real voids, but matrix deformation for example. However, we consider them all in our estimation.
According to these calculations the average %Area of the highlighted areas, is about 2.485%.
So, we assume that the void content based on this area recognition procedure is <2.485% if we consider the overestimation discussed previously. Figure S9. Image recognition of the voids using ImageJ.