II. RESULTS AND DISCUSSION

Initial attempts to deposit h-BN directly on Si, SiO₂, and sapphire using high temperature growth conditions (1000 °C) resulted in thick h-BN films with a high surface roughness not suitable for device development. Therefore, a two-step deposition process, similar to the 2-step growth of h-BN on Ni substrates reported elsewhere, ^{Reference Shi, Hamsen, Jia, Kim, Reina, Holfmann, Hsu, Zhang, Li, Juang, Dresselhaus, Li and Kong11} was developed to reduce surface roughness. The process consists of depositing a film of polyborazylene (from sublimation of ammonia borane) onto a substrate at low temperatures followed by a high temperature anneal at 1000 °C to convert the polyborazylene to h-BN. ^{Reference Shi, Hamsen, Jia, Kim, Reina, Holfmann, Hsu, Zhang, Li, Juang, Dresselhaus, Li and Kong11} The deposition of polyborazylene was performed under 100 Torr of H₂/N₂ (15% H₂) carrier gas at two temperatures, 250 and 400 °C. These two temperatures were selected to represent a temperature within the first hydrogen loss regime of polyborazylene and one slightly above the regime. A postgrowth anneal was then performed at 1 Torr H₂/N₂ (15% H₂), 1000 °C for 1 h to facilitate the conversion to boron nitride.

The surface morphology of direct-grown h-BN depends critically on the deposition and anneal conditions of polyborazylene. High temperature growth (1000 °C, 100 Torr, 10 min) of h-BN on Si and Al₂O₃ without the use of an initial polyborazylene deposition [Figs. 1(a) and 1(b)] results in a 25-nm-thick film with a root-mean-square (RMS) roughness >3 nm. Utilizing an initial polyborazylene deposition at 400 °C (400 °C, 100 Torr, 10 min) followed by a postgrowth anneal at 1000 °C [Figs. 1(c) and 1(d)] yields a reduced film thickness of 8 nm and a significant reduction in surface roughness of >13x (RMS roughness = 0.24 nm) compared to the single-step high temperature growth process. Additionally, a >16x decrease in RMS roughness (4.02–0.24 nm) was found when this sample was compared to an h-BN film of similar thickness grown via CVD on a copper substrate and transferred to Si(111) (see Supplemental Information, Fig. S2d). Further reducing the polyborazylene deposition temperature to 250 °C (250 °C, 100 Torr, 120 min) yields a 2-nm-thick film with an RMS roughness of 0.13 nm on the Al₂O₃ sample. Here, a >25x reduction in RMS roughness (3.28–0.13 nm) was observed compared to the high temperature deposition [Fig. 1(e)] and a >9x reduction (1.18–0.13 nm) compared to transferred h-BN of similar thickness (see Supplemental Information, Fig. S2b). Interestingly, however, when the same conditions are used for h-BN growth on Si(111), partial film voiding occurs [Fig. 1(f)]. This has been observed previously for spin-coated polyborazylene on Si after high temperature annealing and was attributed to degradation of the underlying native oxide. ^{Reference Chan, Rothman, Palladino, Sneddon and Composto12} Therefore, a pregrowth anneal at 1000 °C and 100 Torr in an H₂/N₂ environment was performed to remove any native oxide on the Si surface prior to growth at 250 °C. ^{Reference Yu and Jang13} This resulted in an increased RMS roughness of the film to approximately 1 nm [Figs. 1(g) and 1(h)]. This roughening may result from several possible mechanisms. One possibility is that an initial deposition of BN during the 1000 °C pregrowth anneal may occur due to desorption and redeposition of residual BN from the furnace walls; acting as a seed layer with a rough morphology, as similarly observed for the 1000 °C BN deposition shown in Figs. 1(a) and 1(b). An alternative possibility is that the Si and Al₂O₃ substrate surfaces may reconstruct and roughen during the 1000 °C pregrowth anneal. This roughening of the substrate surface may translate to the morphology of the deposited polyborazylene film. Although further analysis is required to fully understand the surface roughening resulting from the high temperature pregrowth anneal, what can be concluded is that the film voiding on the Si sample was successfully eliminated through use of this high temperature pregrowth treatment [Fig. 1(h)].

FIG. 1.

AFM scans of h-BN directly grown on Al₂O₃(0001) and Si(111) at various deposition conditions. (a, b) Growth of h-BN on Al₂O₃ and Si, respectively, at 1000 °C without the use of an initial polyborazylene deposition. (c, d) Growth of h-BN on Al₂O₃ and Si, respectively, using an initial polyborazylene deposition temperature of 400 °C. (e, f) Growth of h-BN on Al₂O₃ and Si, respectively, using a polyborazylene deposition temperature of 250 °C. (g, h) Growth of h-BN on Al₂O₃ and Si, respectively, using an initial polyborazylene deposition temperature of 250 °C and a pregrowth anneal of 1000 °C. Note that the scale bar for (a, b) is 20 nm while the scale bar for (c–h) is 4 nm.

In addition to surface morphology of direct-grown h-BN, the chemical state of the films is also critical to understand because this can lead to charge impurity scattering in graphene overlayers, and ultimately reduce performance of graphene-based devices. XPS was used to investigate the bonding of boron nitride grown at different polyborazylene deposition temperatures. Figure 2 shows the deconvoluted B 1s spectra of h-BN grown on Al₂O₃ substrates using the four growth conditions described previously. The binding energies of all spectra were obtained after charge correction of the C 1s line to adventitious carbon at 284.8 eV. Except where noted, the chemistries of the h-BN samples grown on Si(111) are similar to the h-BN samples on Al₂O₃(0001). Initial studies focused on direct growth at 1000 °C [Fig. 2(a)], which yields a mixture of B–N–O bonding. The primary peak at 190.25 eV, accounting for 69.8% of the main B 1s line, is indicative of the h-BN structure where the core boron atom is bonded to three nitrogen atoms in the planar hexagonal configuration and can be written as [BN₃]. ^{Reference Petravic, Peter, Fan, Yang and Chen14} The smaller shoulder at 191.01 eV indicates the presence of oxygen and occupies 30.2% of the main B 1s line. Pure B₂O₃ has been previously reported to be located between 193.0 and 193.6 eV. ^{Reference Guimon, Gonbeau, Pfister-Guillouzo, Dugne, Guette, Naslain and Lahaye15–Reference Schild, Ulrich, Ye and Stüber17} This intermediate peak between the h-BN [BN₃] peak and the reported B₂O₃ peak was similarly observed by Guimon et al. ^{Reference Guimon, Gonbeau, Pfister-Guillouzo, Dugne, Guette, Naslain and Lahaye15} and was attributed to the core boron atom bonding to nitrogen and oxygen, where oxygen may be a substitutional element in place of nitrogen, forming a BN _x O _y ternary species. Since an ultrahigh vacuum system was not used for this study, oxygen incorporation is likely due to oxygen impurities in the gas phase and has been commonly reported in the literature. ^{Reference Guimon, Gonbeau, Pfister-Guillouzo, Dugne, Guette, Naslain and Lahaye15,Reference Schild, Ulrich, Ye and Stüber17} Additionally, the π–π* plasmon shake-up peak at 198.62 eV is also present, corresponding to sp ² bonded hexagonal boron nitride. ^{Reference Schild, Ulrich, Ye and Stüber17,Reference Trehan, Lifshitz and Rabalais18}

FIG. 2.

XPS B 1s spectra of h-BN directly grown on Al₂O₃(0001) at (a) 1000 °C, (b) 400 °C, (c) 250 °C, and (d) 250 °C with a 1000 °C pregrowth anneal. Component peaks were deconvoluted using a Gaussian fit. Peak positions are given after performing a charge correction of the C 1s core level peak to 284.8 eV.

Polyborazylene deposition at 400 °C, while yielding a smooth surface morphology, leads to loss of stoichiometry [Fig. 2(b)]. XPS reveals that there are four distinct chemical bonding phenomenon, as evident after deconvolution of the B 1s peak. The primary [BN₃] bonding configuration is found at 190.65 eV and accounts for 42.7% of the B 1s spectrum. Additionally, two shoulders at energies lower than the [BN₃] peak are found at 188.86 and 187.60 eV. Elemental boron–boron bonding has been reported previously in the range of 187.3–187.7 eV. ^{Reference Petravic, Peter, Fan, Yang and Chen14,Reference Goto and Hirai19} Here, the peak at 187.60 eV accounts for 17.2% of the B 1s spectrum and can be attributed to elemental boron with no bonds to nitrogen atoms due to boron–boron bonding in either interstitial or interlayer positions, or at defective domain boundaries. Bonding with nitrogen atoms creates a higher binding energy compared to elemental boron–boron bonding due to the core boron atom bonding with the higher electronegative element nitrogen. ^{Reference Petravic, Peter, Fan, Yang and Chen14} Therefore, the peak in between the h-BN [BN₃] and the B–B configuration, located at 188.86 eV (25.7% of the B 1s spectrum), can be attributed to either a [BN₂] configuration, with boron bonded to two nitrogen atoms, or a [BN] configuration with boron bonded to one nitrogen atom. In other words, the intermediate peak corresponds to a single or double nitrogen vacancy (V _N). A similar peak was reported by Guimon et al., ^{Reference Guimon, Gonbeau, Pfister-Guillouzo, Dugne, Guette, Naslain and Lahaye15} where after 4 keV Ar⁺ sputtering, a peak at 189.2 eV appeared. Likewise, Schild et al. ^{Reference Schild, Ulrich, Ye and Stüber17} reported the emergence of a peak at 188.8 eV after sputter cleaning of c-BN samples. In both cases, this peak was attributed to preferential sputtering of nitrogen atoms, leaving behind a boron-rich surface. Additionally, a peak at 191.98 eV again is attributed to substitutional oxygen contamination in the form of a BN _x O _y ternary species and accounts for 14.5% of the B 1s spectrum. Due to the elemental boron bonding, nitrogen vacancies, and substitutional oxygen bonding, a B/N ratio of 1.92 was obtained for this sample, indicating a highly boron-rich film. Interestingly, the presence of B–B bonding in this film may result in a reduction of free electrons in the material by compensating for the difference in valence between nitrogen and oxygen, where the oxygen impurities will induce additional valence electrons. Therefore, despite the fact that this film is highly nonstoichiometric, the B–B bonding may act to create a more charge-neutral dielectric material (compared to direct-grown BN films containing oxygen contamination but no B–B bonding) which may result in reduced electrostatic doping to graphene.

Polyborazylene deposition at 250 °C without the use of a pregrowth anneal results in the presence of two types of bonding, assigned to h-BN [BN₃] (190.64 eV) and BN _x O _y (191.73 eV). Interestingly, for this sample, the BN _x O _y peak accounts for 57.7% of the B 1s spectra, indicating significant oxygen incorporation. Electron energy loss spectroscopy (EELS) (not shown) indicates that for the majority of these samples, interdiffusion between the substrate and the h-BN layer occurs up to 3 nm. Given that this sample is approximately 2 nm, the additional oxygen incorporation is likely induced from the Al₂O₃ substrate and substitutional bonding occurs during the high temperature anneal. The nitrogen vacancy peaks and B–B bonding peak (observed in the film grown at 400 °C) are not observed for this sample, possibly due to oxygen occupying vacancy sites. However, the additional substitutional oxygen contamination results in a B/N ratio of 1.51. Similarly, a polyborazylene deposition temperature of 250 °C “with” the use of a 1000 °C pregrowth anneal [Fig. 2(d)] resulted in two types of bonding, assigned to h-BN [BN₃] (190.61 eV) and BN _x O _y (191.78 eV). However, here the BN _x O _y peak is significantly reduced compared to the sample grown at 250 °C “without” a pregrowth anneal, accounting for 17.7% (compared to 57.7%) of the B 1s spectrum, despite the fact that these films were of similar thickness (∼2 nm). This may be attributed to the 1000 °C pregrowth anneal, where hydrogen from the H₂/N₂ (15% H₂) environment may act to reduce the Al₂O₃ surface, leaving an oxygen-deficient interface for polyborazylene deposition. This would intuitively result in a reduced interdiffusion of oxygen into the BN film. Alternatively, a BN seed layer may be deposited during the 1000 °C pregrowth anneal that acts to reduce interdiffusion between the Al₂O₃ substrate and the h-BN layer. Unlike the sample grown at 400 °C however, the lower energy peaks corresponding to nitrogen vacancies and B–B bonding are absent and the B/N ratio was found to be 1.01, indicating excellent stoichiometry. This may again be a result of a BN seed layer grown during the high temperature anneal, which may act to promote further BN bonding and reduce B–B bonding. Further analysis is required, however, to fully understand the mechanisms responsible for interdiffusion between BN and the substrate.

Table I shows the B, N, O, and C atomic percentages and B/N ratios for h-BN grown on both Al₂O₃(0001) and Si(111) for all four growth conditions. In three of the four cases, the atomic concentrations and B/N ratios are comparable between both substrates. However, the sample grown at 250 °C without the use of a pregrowth anneal shows significant variance between the films grown on Al₂O₃ and Si. Unlike the sample grown on Al₂O₃, the film grown on Si shows significantly reduced oxygen contamination and excellent stoichiometry with a B/N ratio of 1.00. Clearly, oxygen contamination from the substrate is reduced with the use of a Si substrate; additionally, the overall oxygen content is reduced in comparison to the other h-BN films grown on Si. Due to the small thickness of this sample (∼2 nm), it is possible that oxygen from the gas-phase environment diffuses through the BN layer and reacts with the Si substrate to form SiO₂, rather than reacting with the newly formed polyborazylene film. The formation of this SiO₂ layer may also explain the pitting observed with AFM in Fig. 1(f), where during the 1000 °C postgrowth anneal, this SiO₂ layer decomposes and results in surface pitting.

TABLE I.

Atomic concentrations of boron, nitrogen, oxygen, and carbon and B/N ratios as measured with XPS of h-BN films grown at various conditions on Al₂O₃(0001) and Si(111).

The exact mechanisms for the variance in stoichiometry and bonding configurations between different growth conditions are not fully understood, although boron-rich stoichiometries have been reported previously for boron nitride films synthesized from polyborazylene. ^{Reference Chan, Rothman, Palladino, Sneddon and Composto12,Reference Kho, Moon, Nouet, Ruterana and Kim20} To rule out the effect of variance in gas-phase chemistry on film stoichiometry, mass spectrometry was used. Here, the concentrations of various gas-phase BN precursors (produced from sublimation of ammonia borane at 135 °C) were obtained for furnace temperatures of 250, 400, and 1000 °C. The results, shown in the Supplementary Material (Fig. S5), show that the concentrations of borazine, diborane, aminoborane, and BH₂ remain independent of temperature. Therefore, the observed differences in stoichiometry at these growth temperatures are not due to the gas-phase chemistry in the CVD environment. Instead, nitrogen may be lost during the high temperature annealing process required to form BN. Indeed, Chan et al. ^{Reference Chan, Rothman, Palladino, Sneddon and Composto12} demonstrated an increase in the B/N ratio, from 1.18 to 1.37, from spin-coated polyborazylene to BN upon annealing to 900 °C. It is broadly accepted that the formation of BN from polyborazylene at high temperatures results in the loss of hydrogen during cross-linking of unaligned chain branch structures. ^{Reference Fazen, Remsen, Beck, Carroll, McGhie and Sneddon21} It is possible that some B–B bonding may occur during this process, where the bonding energy of B–B bonds has been reported to be 38% less than that of B–N bonds (310 and 500 KJ/mol, respectively), ^{Reference Bartl, Bohr, Haubner and Lux22} resulting in a loss of nitrogen locally and the formation of a boron-rich film.

The nanoscale structure of h-BN is also influenced by growth parameters. Cross-sectional TEM was utilized to characterize the cross-sections of h-BN films grown on Al₂O₃ substrates [structures on Si(111) are similar]. For comparison, an ∼5-nm h-BN film grown on a Cu foil and transferred to Al₂O₃ is shown in the Supplementary Material (Fig. S6). The sample grown directly at 1000 °C is shown in Fig. 3(a), and it suggests that the initial h-BN layers grow nearly parallel to the Al₂O₃ surface. After ∼5 nm however, the c-axis orientation of the h-BN layers becomes highly random and is likely the cause of the rough film surface. The film thickness is ∼25 nm after only 10 min of growth, indicating a comparatively large growth rate of 2.5 nm/min. Additionally, the film is polycrystalline with domain sizes <5 nm. The small domain size is likely a result of stress induced in the h-BN layers due to both the mismatch in coefficient of thermal expansion (CTE) between BN and the substrate (where h-BN has a negative CTE, leading to expansion and compressive stress incorporation in the BN film upon cooling), ^{Reference Bresnehan, Hollander, Wetherington, LaBella, Trumbull, Cavalero, Snyder and Robinson7} and to the large lattice mismatch between Al₂O₃ and h-BN of 46%, where Al₂O₃ and h-BN have c-plane lattice constants of 4.67 and 2.50 Å, respectively. ^{Reference Chubarov, Pederson, Hogberg, Darakchieva, Jensen, Persson and Henry23} When polyborazylene is deposited at 400 °C [Fig. 3(b)], a ∼5-nm film was obtained after 10 min of growth indicating a growth rate of 0.5 nm/min. This film appears largely amorphous, although some layering of nanocrystalline domains can be observed, indicating a turbostratic structure. For the films grown at 250 °C without a pregrowth anneal [Fig. 3(c)] and at 250 °C with a pregrowth anneal [Fig. 3(d)], the crystallinity is decreased even further, and the films appear amorphous with no long-range order. The poor crystallinity obtained in the films grown with the initial polyborazylene deposition can be attributed to the nature of the initial polyborazylene film, which consists primarily of unaligned polyborazylene chain branches that likely bond with only short-range order upon conversion to BN. ^{Reference Frueh, Kellett, Mallery, Molter, Willis, King’ondu and Suib24} In agreement with Chan et al. ^{Reference Chan, Rothman, Palladino, Sneddon and Composto12} and Kho et al., ^{Reference Kho, Moon, Nouet, Ruterana and Kim20} EELS (not shown) indicates that h-BN films grown at 250 °C without a postgrowth anneal [Fig. 3(c)] resulted in significant oxygen incorporation arising from interdiffusion from the substrate, also in agreement with XPS [Fig. 2(c)]. This interdiffusion layer was found to be present in all four samples, where aluminum and oxygen from the Al₂O₃ substrate diffused into the h-BN layers. For the thicker films, grown at higher temperatures [Figs. 3(a) and 3(b)], the interface region was found to be ∼3 nm in width. This indicates that for the ∼2-nm-thick samples [Figs. 3(c) and 3(d)], significant fractions of the h-BN film are likely to be BN _x O _y , corresponding to XPS measurements showing the atomic concentrations of oxygen to be higher in these films (Table I). However, EELS shows somewhat reduced incorporation of oxygen in the sample grown at 250 °C with the use of a pregrowth anneal, indicating that a BN seed layer or hydrogen reduction of Al₂O₃ during the pregrowth anneal may act to partially reduce oxygen interdiffusion, in agreement with the XPS results described previously [Fig. 2(d)]. Similar results were observed for the films grown on Si(111) in this study, where Si was interdiffused in the BN layer.

FIG. 3.

Cross-sectional TEM micrographs of h-BN films grown on Al₂O₃ at (a) 1000 °C without an initial polyborazylene deposition, (b) a polyborazylene deposition temperature of 400 °C, (c) a polyborazylene deposition temperature of 250 °C without the use of a pregrowth anneal, and (d) a polyborazylene deposition temperature of 250 °C and with the use of a 1000 °C pregrowth anneal.

To evaluate these films for long-term prospects for applications in graphene-based or additional electronic systems, it is instructive to understand their influence on electronic and thermal properties of the heterostructure system. Therefore, h-BN films were investigated for their effective thermal interface conductance with the time-domain thermoreflectance (TDTR) method, ^{Reference Gengler, Roy, Jones and Gord25} which is sensitive to the interatomic bonding and ordering in the thin-film structure as well as the h-BN roughness. Table II summarizes results of the TDTR measurements, whereas data fitting examples are provided in Fig. S7. For thicker h-BN films produced by a direct one-step growth at 1000 °C, it was possible to extract the film cross-plane thermal conductivity. It was interesting to observe for these films that the h-BN thermal conductivity was very low. Conductivities of about 600 W/m/K along base planes and 30 W/m/K in cross base plane directions are normally expected for bulk h-BN crystals. For thin ordered h-BN, these are reduced by a factor of two, where recent evaluations of suspended exfoliated few layer h-BN films had shown 250 W/m/K of in-plane thermal conductivity for 5 ML h-BN films. ^{Reference Jo, Pettes, Kim, Watanabe, Taniguchi, Yao and Shi26} The results of our TEM studies show that h-BN films synthesized from 1000 °C direct growth are nanocrystalline and randomly oriented, which critically influence their overall thermal conductivity. Thermal conductivity values reported by Muratore et al. ^{Reference Muratore, Varshney, Gengler, Hu, Bultman, Smith, Shamberger, Qiu, Ruan, Roy and Voevodin27} for nanocrystalline hexagonal thin films of MoS₂, WS₂, and WSe₂ indicate that several orders of magnitude reduction in thermal conductivity can be expected in disordered hexagonal structures with weak cross-plane bonding characteristics.

TABLE II.

Results of TDTR studies of h-BN films directly grown on silicon and sapphire substrates at different growth conditions.

^a Effective conductance, including thermal transport through film thickness and interfaces.

Films grown from polyborazylene are much thinner than the single-step, high temperature growths, making direct measurements of their thermal conductivity impossible as the TDTR penetration depth was two orders of magnitude deeper. Instead, for such thin h-BN films thermal effects were interpreted as changes in an overall interface conductance between the top aluminum transducer layer applied for TDTR analysis and the underlying Al₂O₃ or Si substrates. Such effective thermal interface conductance incorporates thermal transport in the h-BN film as well as its interface resistances with the substrate and the aluminum layer. A similar procedure was recently used with graphene layers. ^{Reference Hopkins, Baraket, Barnat, Beechem, Kearney, Duda, Robinson and Walton28} From this perspective, there is a clear trend in Table II that the thinner h-BN films grown at 400 and 250 °C provide better effective thermal interface conductance. In all cases, it was several times higher than that of the conductance measured for thicker films produced at 1000 °C. The observation of higher effective thermal interface conductances for smaller thickness h-BN films can be understood as a decreased resistance of phonon transport through the film thickness. In addition, polyborazylene growth yields better ordered h-BN films near the substrate surface region as well as lower surface roughness values, which would be expected to reduce phonon scattering at the h-BN interfaces. How crystallinity within the h-BN film and its structural alignment at the interfaces affect thermal conductance have to be determined. As one indicator, a well ordered 5-nm-thick h-BN film produced by direct growth on a copper foil (see Fig. S7) and transferred to sapphire shows an effective thermal interface conductance of 101 ± 5 MW/m²/K. This is higher than the values in Table II, indicating that ordering within the h-BN film may contribute to the observed improvement of the effective thermal conductance with the polyborazylene growth. Despite the improvement in thermal conductivity observed for the amorphous BN films, the values are still below bulk h-BN; indicating that films synthesized via the polyborazylene to BN conversion process presented here are likely more suitable for electronic and optoelectronic applications that do not require thermal management considerations (i.e., low power electronics, sensors).

To understand the impact of h-BN as a substrate on the transport properties of graphene, it is instructive to understand the strain and Fermi velocity reduction (FVR) which h-BN induces in graphene overlayers. Figure 4 shows the Raman 2D versus G peak positions for CVD graphene transferred to various dielectric substrates. Vector analysis correlating graphene's 2D (ω_2D) and G (ω_G) peak positions to tensile and compressive strain (e _T and e _C, respectively), FVR (e _FVR), and hole doping (e _H), is also shown and is based on the work of Lee et al. ^{Reference Lee, Ahn, Shim, Lee and Ryu29} and Ahn et al. ^{Reference Ahn, Kim, Ko, Choi, Watanabe, Taniguchi, Hong and Ryu30} Nominally, strain-free and charge-neutral graphene Raman peaks are located at ω_G and ω_2D peak positions of 1581.6 and 2676.9 cm⁻¹, respectively, as indicated by the light blue circle in Fig. 4. ^{Reference Lee, Ahn, Shim, Lee and Ryu29} Giovannetti et al. ^{Reference Giovannetti, Khomyakov, Brocks, Kelly and van den Brink31} have shown that even weak van der Waals forces between graphene and its dielectric environment can lead to perturbation of graphene's electronic band structure through breaking of symmetry between neighboring C atoms in graphene. The modulation of the electronic structure can manifest as a reduction in the Fermi velocity (v _F) of charge carriers in graphene. ^{Reference Ahn, Kim, Ko, Choi, Watanabe, Taniguchi, Hong and Ryu30} This FVR was found to only affect the 2D peak of graphene, whereas the G peak of graphene, originating from the E_2G phonon, is not affected to a first-order approximation. ^{Reference Ahn, Kim, Ko, Choi, Watanabe, Taniguchi, Hong and Ryu30,Reference Ni, Wang, Yu, You and Shen32} Therefore, any reduction in v _F of graphene will manifest as an upward shift in the 2D peak along the e _FVR vector shown in Fig. 4, where the values shown indicate Δv _F/v _F (%) and are negative to indicate a decrease from the intrinsic v _F of graphene (≈10⁶ m/s). ^{Reference Ahn, Kim, Ko, Choi, Watanabe, Taniguchi, Hong and Ryu30,Reference Ni, Wang, Yu, You and Shen32} Raman spectroscopy is also useful in characterizing strain (ε) in a wide variety of materials, since changes in lattice constant due to strain will result in alterations to the respective phonon frequencies. ^{Reference Schadler and Galiotis33} Lee et al. ^{Reference Lee, Ahn, Shim, Lee and Ryu29} then extracted the contribution of strain (ε) for given points in the ω_G − ω_2D space, forming a simple vector model for e _T and e _C, as explained elsewhere. ^{Reference Lee, Ahn, Shim, Lee and Ryu29} Finally, the vector associated with hole doping (e _H) was formed from in situ Raman analysis data obtained by Das et al., ^{Reference Das, Chakraborty, Piscanec, Pisana, Sood and Ferrari34} where hole doping was controlled through electrical gating. Since both electron and hole doping will lead to an increase in ω_G, ^{Reference Lee, Ahn, Shim, Lee and Ryu29} the results presented in Fig. 4 show that the contribution of charge doping to the Raman spectra of the samples investigated here is negligible. Therefore, the ω_G and ω_2D positions of the transferred CVD graphene investigated here can be vector decomposed exclusively into contributions from strain (e _T, e _C) and FVR (e _FVR) as a function of the supporting dielectric substrate.

FIG. 4.

Raman 2D frequency versus G frequency of CVD graphene transferred to various substrates. The data are vector decomposed to correlate peak shifting to tensile and compressive strain (e _T and e _C, respectively), FVR (e _FVR), and hole doping (e _H), using methods reported by Ahn et al. ^{Reference Ahn, Kim, Ko, Choi, Watanabe, Taniguchi, Hong and Ryu30} Inset: expected trajectories of 2D and G peak positions affected by e _T, e _C, e _FVR, and e _H.

Significant variations in strain and FVR can occur when incorporating graphene with dielectric substrates. Graphene transferred to the bare SiO₂ (blue circle) and Al₂O₃ (black circle) substrates both experienced compressive stress (−0.17% and −0.24%, respectively) and large FVR (−7.1% and −5.8%, respectively). The compressive stress likely arises from strong adhesion of the graphene to these dielectrics. It has been shown previously that exfoliated graphene on SiO₂ substrates experiences strong compressive stress and deformation due to ultrastrong adhesion forces with the underlying substrate. ^{Reference Koenig, Boddeti, Dunn and Bunch35} It is possible that graphene may have strong adhesion to Al₂O₃ as well, based on the shift of ω_G − ω_2D along e _C. These strong adhesion forces also perturb the electronic band structure of the transferred graphene, ^{Reference Koenig, Boddeti, Dunn and Bunch35} which again manifests as FVR and results in an upward shift in ω_2D. h-BN films grown directly on Al₂O₃ (square symbols) and Si (triangular symbols) at 1000 °C without a polyborazylene deposition (green symbols) and at 250 °C with a pregrowth anneal (pink symbols) resulted in extensive tensile strain in the transferred graphene, with an average of 0.34% and 0.24% tensile strain, respectively. Here, the tensile strain likely results from the transferred graphene layer being stretched across the rough morphology. Interestingly, the FVR for these samples was low compared to the other growth conditions, where the samples grown at 1000 °C without a polyborazylene deposition and at 250 °C with a pregrowth anneal resulted in a FVR of −2.0% and −4.0%, respectively. This is particularly unexpected since the h-BN samples grown at 1000 °C showed the highest degree of crystallinity, as observed with TEM in Fig. 4(a). Ahn et al. ^{Reference Ahn, Kim, Ko, Choi, Watanabe, Taniguchi, Hong and Ryu30} showed that exfoliated h-BN supporting dielectrics resulted in an increase in FVR compared to amorphous SiO₂ due to the crystalline nature of h-BN, which would induce periodic perturbations to the graphene band structure. However, here the 1000 °C h-BN sample shows lower FVR than the amorphous BN samples (grown with the polyborazylene deposition). This again may be attributed to the high surface roughness of these samples. Since the graphene film would be partially suspended over the rough h-BN surface, a large percentage of the graphene film would not be in contact with the h-BN surface, reducing the van der Waals interactions between graphene and the substrate. In this case, the tensile strain resulting from the partial suspension over the substrate would primarily contribute to the shift in Raman peak position. The samples grown at 250 °C with a pregrowth anneal show higher FVR despite the fact that they are amorphous and would be expected to contribute to less van der Waals interactions. However, this sample was found to be 72% less rough than the 1000 °C sample (see Fig. 1) and thus would be in greater contact with the h-BN surface, thereby interacting with the transferred graphene to a higher degree. Similarly, the ∼5-nm-thick h-BN film grown on a Cu foil and transferred to Al₂O₃ (purple circle) also induces tensile stress (0.14%) to the graphene layer, due to a RMS surface roughness of 1.5 nm (see Fig. S2c), which is similar to that of the sample grown at 250 °C with an anneal. However, the transferred h-BN film shows significant FVR compared to the direct-grown h-BN samples. This may be a result of the improved crystallinity, where these films were found to have much higher crystallographic texturing and domain sizes compared to the direct-grown h-BN films (see Fig. S6), and thus may not only induce strain from surface roughness but also induce periodic perturbations to the graphene band structure over several nanometer ranges.

Boron nitride films grown with a polyborazylene deposition of 250 °C without a pregrowth anneal (orange symbols) were found to result in the most variation in ω_G and ω_2D peak positions between the Al₂O₃ (square) and Si (triangle) substrates. The sample grown on Si resulted in a negligible compressive strain of −0.02% and a low FVR of −2.2%. The amorphous structure of this film is expected to induce negligible van der Waals interactions to the graphene layer, thus resulting in minimal FVR. However, it should be noted that this film experienced surface pitting (see Fig. 1); therefore, the Raman line scan for this sample was taken in a region where surface pitting was not observed locally. Although this sample resulted in ω_G and ω_2D positions comparable with strain-free charge-neutral graphene, this sample is unfortunately unsuitable for graphene device development due to the surface pitting observed across the film surface. In contrast, the BN sample grown on Al₂O₃ resulted in a compressive strain of −0.14% and a FVR of −5.4%. It is not fully understood why this sample shows such contrast to the sample grown on Si under the same conditions, as the relatively high FVR is not expected for an amorphous material. However, based on XPS results shown in Table I, it is known that this film is much different chemically than the film grown on Si. Therefore, the increase in FVR may be due to the fact that this sample was found to have significant oxygen contamination. Substitutional oxygen bonding would induce a high density of dangling bonds due to an extra valence electron compared to nitrogen; therefore, the increased density of dangling bonds likely induces electrostatic perturbations to graphene's band structure and exacerbates the FVR.

The samples grown at a polyborazylene deposition of 400 °C (red symbols) show negligible strain and low FVR for both BN films grown on Al₂O₃ (squares) and Si (triangles), where the strain was 0.01% and −0.08% and the FVR was −3.7% and −3.1% for these films, respectively. The amorphous nature of these films results in minimal van der Waals interactions with the graphene layer and minimally degrades the Fermi velocity. Additionally, the presence of both B–B bonding and BN _x O _y bonding [as observed with XPS in Fig. 2(b)] likely results in an overall compensation of valence, resulting in negligible electrostatic interactions from free electrons, further leading to a low FVR. Even more interesting is that the strain induced into graphene is nearly negligible and comparable to strain-free graphene, likely due to low adhesion between graphene and BN as well as the low RMS surface roughness of these films [as shown in Figs. 1(c) and 1(d)]. These results suggest that the amorphous BN grown with a polyborazylene deposition of 400 °C may be an ideal substrate for CVD graphene integration.

Hall transport provides evidence that substrate choice and h-BN properties have significant impact on graphene electronic properties. Figure 5 summarizes carrier mobility as a function of charge carrier density for CVD graphene transferred to various substrates. CVD graphene was transferred to direct-grown h-BN on Al₂O₃, a 5-nm h-BN film grown on Cu and transferred to Al₂O₃, a bare Al₂O₃(0001) substrate, and a bare 300-nm SiO₂ substrate. Graphene transferred to bare Al₂O₃(0001) (orange circles) shows an average mobility of 4218 (±849) cm²/V-s at a hole concentration of 1.9 × 10¹² (±4.5 × 10¹¹) cm⁻², whereas the graphene sample transferred to the SiO₂ substrate (brown hexagons) shows an average mobility of 2559 (±124) cm²/V-s at a hole concentration of 3.5 × 10¹² (±3.5 × 10¹¹) cm⁻². Both Al₂O₃ and SiO₂ have low surface optical phonon energies, reported at 55 and 60 meV, respectively. ^{Reference Konar, Fang and Jena36} Therefore, both substrates are expected to experience significant phonon–electron coupling, resulting in surface optical phonon scattering. In a separate work, however, it was found that impurity scattering is the dominant scattering mechanism for h-BN gated epitaxial graphene devices, accounting for >90% of the total scattering processes for h-BN coated epitaxial graphene. Though the samples detailed here show reduced carrier densities compared to epitaxial graphene, these samples are likely similarly dominated primarily by impurity scattering. ^{Reference Hollander, Agrawal, Bresnehan, LaBella, Trumbull, Cavalero, Snyder, Datta and Robinson8} Given that the dielectric constants of Al₂O₃ and SiO₂ are 12.5 and 3.9, ^{Reference Konar, Fang and Jena36} respectively, the higher dielectric constant of Al₂O₃ will act to screen charged impurities that would induce Coulombic scattering to graphene, as described by Konar et al. ^{Reference Konar, Fang and Jena36} Therefore, the effective impurity density of the graphene/Al₂O₃ system is reduced compared to graphene/SiO₂, leading to higher mobilities for the graphene/Al₂O₃ system. Based on temperature-dependent Hall effect measurements and subsequent modeling described by Hollander et al., ^{Reference Hollander, Agrawal, Bresnehan, LaBella, Trumbull, Cavalero, Snyder, Datta and Robinson8} it was found that the effective impurity concentration is approximately 5.5 × 10¹¹ cm⁻² for a graphene/SiO₂ system with the mobilities and sheet densities shown in Fig. 5(a). Unfortunately, the impurity density of the graphene/Al₂O₃ system cannot be extracted without temperature-dependent Hall mobility analysis. The variation in mobility for these two samples can be effectively described by impurity scattering, since surface roughness scattering can be ruled out as both substrates display a similar surface roughness (see Fig. S3). Additionally, the compressive strain and high FVR observed with Raman spectroscopy were similar for these two samples (Fig. 4) and likely further limit the mobility for both substrates.

FIG. 5.

(a) Hall mobility versus hole concentration of CVD graphene transferred to h-BN dielectrics synthesized at various growth conditions, as well as bare Al₂O₃ and SiO₂ substrates. Dotted curves indicate the simulated dependency of mobility on carrier density at various impurity densities based on modeling for the h-BN/graphene system described by Hollander et al. ^{Reference Hollander, Agrawal, Bresnehan, LaBella, Trumbull, Cavalero, Snyder, Datta and Robinson8} (b) Log–log plot of graphene Hall mobility versus RMS surface roughness of the underlying h-BN dielectric, as measured by AFM. (c) Log–log plot of graphene Hall mobility versus oxygen content of the underlying h-BN dielectric, as measured by XPS. The symbols shown in (b) and (c) correspond to the legend given in (a).

The dotted lines in Fig. 5(a) represent the simulated dependency of mobility on carrier density for given impurity densities of the h-BN/graphene system. ^{Reference Hollander, Agrawal, Bresnehan, LaBella, Trumbull, Cavalero, Snyder, Datta and Robinson8} This model assumes an acoustic deformation potential of 4.8 eV ^{Reference Fang, Konar, Xing and Jena37} and an optical deformation potential of 25.6 eV/A ^{Reference Fang, Konar, Xing and Jena37} for the CVD graphene and that the surface optical phonon modes (taken as 160 meV) of all h-BN films are equal. ^{Reference Hollander, Agrawal, Bresnehan, LaBella, Trumbull, Cavalero, Snyder, Datta and Robinson8} These simulated trends indicate that the observed variation in mobility between graphene transferred to the various h-BN samples cannot be described simply by changes in carrier density. Instead, based on this modeling, the change in mobility as a function of h-BN growth conditions can be effectively described by impurity concentration. The graphene devices on h-BN films grown at an initial polyborazylene deposition temperature of 400 °C (red square) resulted in an average mobility of 6425 (±975) cm²/V-s at a hole concentration of 1.1 × 10¹² (±3.0 × 10¹¹) cm⁻² and reached a maximum mobility of nearly 8000 cm²/V-s. These mobilities correspond to impurity densities ranging from 2.0 × 10¹¹ to 3.0 × 10¹¹ cm⁻². In contrast, the h-BN films grown at the various other growth conditions result in reduced mobility and increased hole density, leading to increased impurity densities ranging from 4.5 × 10¹¹ to 6.5 × 10¹¹ cm⁻². In other works, ^{Reference Bresnehan, Hollander, Wetherington, LaBella, Trumbull, Cavalero, Snyder and Robinson7,Reference Hollander, Agrawal, Bresnehan, LaBella, Trumbull, Cavalero, Snyder, Datta and Robinson8} we found that the benefits of h-BN dielectrics on epitaxial graphene were heavily dependent on the charged impurity density, where impurity concentrations >5 × 10¹¹ cm⁻² resulted in a charged impurity dominated scattering regime, where high-k dielectrics outperformed h-BN due to their ability to screen Coulombic scatterers. On the other hand, for impurity concentrations <5 × 10¹¹ cm⁻², surface optical phonon scattering begins to contribute more significantly to the total scattering rate. In this scattering regime, the increased phonon–electron coupling induced by high-k dielectrics (having low surface optical phonon modes) limits the mobility of charge carriers in graphene, while h-BN gated devices benefit from h-BN's high surface optical phonon modes. Therefore, the work presented here, as shown in Fig. 5(a), agree with previous reports, where the high-k dielectric Al₂O₃ outperforms h-BN samples with high impurity densities, whereas the h-BN samples with low impurity densities (grown at 400 °C) outperform Al₂O₃.

Unfortunately, the model used for impurity density extraction in Fig. 5(a) does not take into account surface roughness scattering, where significant surface roughness variation between h-BN growth conditions was observed in Fig. 1. Increased interfacial surface roughness has not only been shown in literature to directly induce scattering in graphene ^{Reference Katsnelson and Geim38} but also induces strain in graphene (as validated with Raman spectroscopy in Fig. 4), which has been shown to further reduce mobility. ^{Reference Miao, Tongay and Hebard39,Reference Shah, Mohiuddin and Singh40} Therefore, the average mobilities of the transferred graphene on the various h-BN samples are plotted against RMS surface roughness, as measured with AFM, in Fig. 5(b). A trend is observed for four of the five samples; however, the sample grown at 250 °C without a pregrowth anneal results in a low average mobility [2445 (±535) cm²/V-s at a hole concentration of 5.8 × 10¹² (±1.1 × 10¹²) cm⁻²] despite the fact that this sample displayed the lowest RMS roughness (130 pm) of all the h-BN samples. To understand this, the average mobilities were also plotted against the oxygen concentration of the h-BN films, as measured with XPS, in Fig. 5(c). It is likely that substitutional oxygen bonding contributes significantly to the overall impurity density of the graphene/dielectric system since oxygen possesses an extra valence electron than nitrogen, resulting in an increased density of dangling bonds that would induce Coulombic scattering to graphene. Therefore, the oxygen concentration of the h-BN films can be correlated directly with impurity density. From Fig. 5(c), graphene's mobility is found to decrease with increasing oxygen content, where the sample grown at 250 °C without a pregrowth anneal (light blue cross) is dominated by impurity scattering due to its high concentration of substitutional oxygen impurities. Again however, outliers in the observed trend are found. The sample grown at 1000 °C without a polyborazylene deposition (green triangle) was found to have an average mobility of 2407 (±478) cm²/V-s at a hole concentration of 4.9 × 10¹² (±8.3 × 10¹¹) cm⁻², whereas the transferred h-BN sample (pink “X”) was found to have an average mobility of 2123 (±243) cm²/V-s at a hole concentration of 4.6 × 10¹² (±8.3 × 10¹¹) cm⁻², despite the fact that these samples had relatively low oxygen impurity concentrations. The results of Figs. 5(b) and 5(c) suggest that two scattering mechanisms contribute to graphene transport limitations when transferred to the h-BN films. Scattering can be dominated by surface roughness (and strain) as evident in the rough h-BN sample grown at 1000 °C (RMS roughness > 3 nm) and the transferred h-BN sample (RMS roughness > 1.5 nm), or by impurity density as evident by the ultrasmooth but highly oxygen-rich (>50% oxygen content) sample grown at 250 °C without a pregrowth anneal. Another option is a combination of the two scattering mechanisms, as evident by the h-BN film grown at 250 °C with a pregrowth anneal (blue diamond), where the mobility [3098 (±281) cm²/V-s at a hole concentration of 3.1 × 10¹² (±5.6 × 10¹¹) cm⁻²] was found to similarly lie at intermediate points in the trend lines of both Figs. 5(b) and 5(c). In this case, the moderate roughness and moderate oxygen contamination lead to contributions from both scattering mechanisms. Further work is required to model the exact impurity concentrations of these samples while taking into account the effects of surface roughness scattering. Based on these results, however, it is apparent that when h-BN is synthesized under conditions that result in both low surface roughness and low oxygen contamination, the beneficial effects of h-BN's high energy surface optical phonon modes can be realized. This was achieved in the present work for h-BN synthesis at an initial polyborazylene deposition temperature of 400 °C, where an RMS surface roughness of 240 pm resulted in nearly strain-free transferred graphene (as observed with Raman spectroscopy in Fig. 4), while a low oxygen content of <6 at.% resulted in low impurity densities (<3 × 10¹¹ cm⁻²) which allow h-BN's high energy surface optical phonon modes to benefit graphene devices. With the use of the 400 °C polyborazylene to BN synthesis conditions, transferred CVD graphene devices resulted in an increase in average mobility of 52% and 151% and a reduction in average hole doping of 75% and 224% on the Al₂O₃ and SiO₂ substrates, respectively, with the maximum mobility reaching nearly 8000 cm²/V-s. These results also suggest that even the use of nearly amorphous BN, as shown with cross-sectional TEM in Fig. 3, can benefit graphene electronics.