III. RESULTS AND DISCUSSION

To evaluate the nucleation process for the growth of GaN on BN, several attempts were made using a LT-GaN nucleation layer (NL). However, due to the lack of dangling bonds on the sp²-bonded BN surface, most of these attempts resulted in a low density of nonuniformly distributed GaN crystalline islands, which often fail to coalesce into a continuous layer with good morphology. Occasionally, continuous layers were obtained but with relatively poor structural quality [an example is given in Fig. 1(b)]. It has been previously reported ^{Reference AlBalushi, Miyagi, Lin, Wang, Calderin, Bhimanapati, Redwing and Robinson21} that GaN tends to nucleate on epitaxial graphene only in highly defective regions, producing small single crystal GaN nuclei and damaging the underlying graphene. On the other hand, AlN has been found to nucleate on graphene in a polycrystalline layer, while preserving the underlying graphene layer. For growth using an AlN NL of 6–20 nm on BN/sapphire, we consistently obtained continuous layers with mirrored surface morphology. Figure 1(c) shows the FWHM values of XRD rocking curve measurements for GaN grown with different AlN NL thickness. The best FWHM values for the (0002) and ( $10\bar 12$ ) diffractions (ω-scan) are 0.13° and 0.17°, respectively [Fig. 1(a)]. These FWHM values are comparable to GaN grown on other 2D materials, as reported in the literature. ^{Reference Nepal, Wheeler, Anderson, Kub, Mastro, Myers-Ward, Qadri, Freitas, Hernandez, Nyakiti, Walton, Gaskill and Eddy11–Reference Kobayashi, Kumakura, Akasaka and Makimoto14} Although a thickness of 15 nm seems optimum for the AlN NL under our current process conditions, the dependence on thickness was relatively weak with high scatter [Fig. 1(c)], possibly due to variations in the BN/sapphire surface roughness. This effect will be elucidated in a future paper; instead this paper will focus on AlGaN/GaN heterostructures grown on atomically smooth BN (RMS <0.2 nm) using 9–15 nm thick AlN as the nucleation layer, since thicker AlN NLs tended to cause self-delaminating after the samples were unloaded at room temperature. The dislocation density estimated using ω-scan FWHM values ^{Reference Dunn and Kogh22,Reference Srikant, Speck and Clarke23} is on the order of 10⁹ cm⁻², which is higher than typical values obtained for GaN layers grown on conventional sapphire substrates. However, studies on the electron transport properties and intrinsic mobility limits of 2DEG in AlGaN/GaN ^{Reference Jena, Smorchkova, Elsass, Gossard and Mishra24,Reference Jena, Gossard and Mishra25} have shown that such dislocation density levels of 10⁹ cm⁻² typically do not interfere with the 2DEG, which typically has a density of 10¹³ cm⁻². This implies that further removal of charged defects (dislocations, background impurities, etc.) would not be efficient for improving the transport mobility of the 2DEG. In fact we observed improved transport mobility in our AlGaN/GaN grown on BN/sapphire compared to the same material grown on sapphire, even though the XRD reveals more structural defects. This behavior indicates that the dislocations at the density present in our AlGaN/GaN are not the dominant scattering mechanisms limiting the transport of 2DEG. More detailed results are presented below.

FIG. 1.

XRD results for (0002) and ( $10\bar 12$ ) rocking curves obtained from GaN grown on BN/sapphire using (a) AlN nucleation layer, and (b) GaN nucleation layer. (c) FWHM of (0002) and ( $10\bar 12$ ) diffractions is given as a function of AlN NL thickness, where dash lines are a guide for the eye only.

Cross-sectional TEM was performed on the AlGaN/GaN heterostructure grown on BN/Sapphire to evaluate the interfaces between the AlN NL, BN, and GaN. Figure 2(a) shows a high-angle annular dark-field (HAADF) image of sample A, exhibiting the final AlGaN/GaN heterostructure grown on BN/sapphire. (Note that the interface region of BN/sapphire interface is not shown in this figure.) The corresponding high resolution (HR) TEM image of the HEMT region is shown in Fig. 2(b). The AlGaN/AlN/GaN stack shows a very uniform, smooth, and high quality interface between these layers, which is barely distinguishable without reference to the Z-contrast in the HAADF images. The interface region of the AlN/BN/sapphire is shown in Fig. 2(c) as a HAADF image, where the presence of a BN spacer (the dark region) can be observed between AlN and sapphire substrate. The corresponding HR-TEM image of the few-layer BN and AlN nucleation layer is shown in Fig. 2(d), where a layered structure parallel to the c-plane sapphire can be identified, this corresponds to BN with a thickness of ∼1.6 nm. The thickness of AlN nucleation layer is estimated to be ∼14 nm. The HR-TEM results confirm that (i) AlN nucleates from the surface of BN but not from the sapphire substrate; (ii) the AlN nucleation layer is oriented along [0002] direction with lattice plane spacing of ∼0.5 nm, and (iii) the few-layer BN film is well preserved after GaN growth. Compared to the results of using 2D graphene templates in previous III-N MOCVD growth, ^{Reference Nepal, Wheeler, Anderson, Kub, Mastro, Myers-Ward, Qadri, Freitas, Hernandez, Nyakiti, Walton, Gaskill and Eddy11,Reference AlBalushi, Miyagi, Lin, Wang, Calderin, Bhimanapati, Redwing and Robinson21} it is not surprising to see well-preserved 2D BN, since BN is very stable under MOCVD growth conditions.

FIG. 2.

Cross section TEM (a) image taken under HAADF STEM mode near top surface of AlGaN/AlN/GaN structure; (b) HR-TEM image showing GaN/AlN insertion layer and AlGaN barrier; (c) HAADF image of GaN/AlN/BN/sapphire interface, where a BN spacer is clearly shown; (d) HR-TEM showing layered BN with thickness ∼1.6 nm and well-oriented AlN nucleation layer ∼14 nm.

AFM measurements were carried out to examine the surface morphology of the AlGaN/GaN heterostructures. Figure 3 compares the AFM image of a typical surface of AlGaN/GaN grown on BN/sapphire and the same structure grown on sapphire. The root mean square of roughness (RMS) was 0.283 nm for AlGaN/GaN grown on BN, compared to 0.257 nm on sapphire. The undulation exhibited in the AFM images is very similar to the GaN surface morphology commonly observed on sapphire substrates. We also observed some increase in RMS roughness when we use relatively rough BN/sapphire templates, consistent with the known sensitivity of BN surface morphology to various process conditions. ^{Reference Paduano, Snure and Shoaf26,Reference Snure, Paduano and Kiefer27} For BN surface RMS roughness up to 1 nm, relatively smooth AlGaN/GaN surface (RMS ∼ 0.314 nm) can still be obtained. However, increased roughness of the BN/sapphire template did show significant effect on the crystalline quality of the GaN layer as revealed by XRD. When Fe doping was introduced in the GaN buffer, we did not observe a notable change in surface morphology.

FIG. 3.

AFM images of AlGaN/GaN surface morphology (a) grown on BN/sapphire, and (b) grown on sapphire, with a root mean square roughness indicated on each image.

To evaluate the quality of the AlGaN/GaN heterostuctures, the mobility of the 2DEG was assessed. Hall-effect measurements were carried out as a function of temperature in the 10–300 K range at a fixed magnetic field. Figure 4 depicts the temperature dependence of electron mobility (µ) and sheet electron density (n _sh) for AlGaN/GaN grown on BN/sapphire and sapphire substrates using (a) an undoped GaN buffer and (b) an Fe doped GaN buffer. Measured values of µ for different samples at 10 and 300 K are summarized in Table I, along with the corresponding n _sh, average sheet resistance (R _sh at 300 K), and the XRD results discussed above. The electrical properties from similar structures grown on SiC and sapphire are included for comparison. Figure 4 clearly exhibits near-constant carrier density over the entire temperature range for samples C and D (Fe doped buffer grown on BN/sapphire and sapphire), and for sample A (undoped GaN buffer grown on BN/sapphire). When the undoped GaN was grown directly on the sapphire without the BN (sample E), the apparent sheet carrier density increased with increasing temperature [Fig. 4(a)]. This behavior is commonly seen on sapphire substrates, ^{Reference Smorchkova, Keller, Heikman, Elsass, Heying, Fini, Speck and Mishra28} because undoped GaN grown on sapphire is typically n-type with a background carrier concentration of ∼2 × 10¹⁶ cm⁻³, possibly due to residual oxygen impurity from the substrate. In this scenario, the bulk carrier density in the GaN buffer increases with temperature due to thermal activation. Since the carrier density of the 2DEG (n _2d) is independent of temperature, we can differentiate the contribution of the buffer from that of the 2DEG by comparing the temperature dependences of n _sh in the various samples (Fig. 4). The constant n _sh in samples C and D confirms that parallel conduction is negligible in the GaN buffer (n _sh = n _2d) due to Fe-related compensation, whereas the increasing sheet carrier density with temperature in sample E is clearly due to the conductive buffer. However, sample A grown on BN/sapphire showed no change in sheet carrier density with increasing temperature, indicating negligible parallel conduction in the buffer even without Fe doping, which is similar to the behavior of high resistivity GaN grown on SiC substrates. The sheet resistance across the 2-in wafer (shown in Table I) also showed much better uniformity when the nonuniform buffer conduction is reduced.

FIG. 4.

Sheet carrier density and Hall mobility as a function of temperature for AlGaN/GaN heterostructures (a) grown on undoped GaN buffer, and (b) grown on Fe doped GaN buffer, where red symbols are for sapphire substrates, and blue symbols are for BN/sapphire templates.

TABLE I.

Summary of XRD and Hall-effect measurement results for samples in this study.

^a BN was deposited at low N/B ratio, see discussion.

For AlGaN/GaN heterostructures grown on a conductive GaN buffer (samples E and F), the overall mobility is affected by the 3D transport in the parallel conducting buffer, especially at room temperature; therefore, the measured mobility is lower than the pure 2DEG mobility. ^{Reference Jena, Smorchkova, Elsass, Gossard and Mishra24,Reference Lisesivdin, Yildiz, Balkan, Kasap, Ozcelik and Ozbay29} In this regard, AlGaN/GaN grown on BN/sapphire, benefiting from the highly resistive GaN buffer, exhibited improved µ_{300 K}. However, this improvement due to the template can be significantly affected by the quality of the BN. We observed different magnitudes of sheet carrier density increase depending on the quality of the BN. As an example given in Table I, sample B is grown in the same run as sample A. The difference between sample A and B is that the BN layers were deposited under different V/III ratios—2700 and 900, respectively. As shown in our previous paper, ^{Reference Paduano, Snure and Shoaf26,Reference Snure, Paduano and Kiefer27} the morphology of these BN layers is sensitive to the growth conditions. At a high V/III ratio, growth of BN exhibits a self-terminating mode of growth, resulting in a BN layer with an atomically smooth surface. At a V/III ratio of 900; however, the BN layer is highly defective with a high density of holes. ^{Reference Snure, Paduano and Kiefer27} The lower value of µ_{300 K} in sample B is consistent with the existence of transport contributions from both 2DEG and a parallel channel. Therefore, good quality BN is critical to effectively reduce parallel conduction in the GaN buffer.

To further confirm the effect of BN on the residual donor impurities in the GaN buffer, SIMS analyses were performed on samples A, E, and F for [O], [C], and [Si]. (There was no intentional Fe doping in theses samples.) Results are summarized in Table II, where the background impurity levels are derived from SIMS profile near the center of the GaN buffer (∼1 µm below surface) using a 50-point average. We see that [Si] is the lowest in all three samples (close to the SIMS detection limit); [C] has a very similar level of ∼1 × 10¹⁶ cm⁻³; and [O] exhibits some variation between the samples. The lowest [O] level was detected in sample A, consistent with its lowest residual donor level in the GaN buffer (as obtained from Hall measurements, Table I). Slightly reduced buffer conductivity in sample F (compared to sample E) is also consistent with a reduced [O] background.

TABLE II.

Summary of SIMS results on background impurity concentrations (average over 50 points at ∼1 μm below surface).

We should note that the residual donor concentration obtained from Hall measurements may not always be in good agreement with SIMS results. This is because SIMS measurements are not always accurate at concentrations near the detection limit, and also some of the impurities may be electrically inactive. ^{Reference Look, Fang and Claflin30} Nevertheless, the differences in measured oxygen concentration are noticeable, especially near the substrate interface.

Figure 5 shows the SIMS profiles from samples A, E, and F [where the background [C] and [Si] profiles are shown for only one of the samples on sapphire substrates in Fig. 5(a)]. [O] exhibits an abnormal increase in all three samples at ∼250–400 nm before the substrate interface, as indicated by marker lines 1 and 2 in the figure. This increase in [O] is consistent with our previous SIMS measurements on films grown on sapphire, and is accompanied by a similar increase in [Al]. It was suggested that during SIMS data acquisition, preferential sputtering in the local area around core of screw dislocations can be responsible for causing long “tunnels” reaching the sapphire substrate. ^{Reference Kim and Andersson31} A related observation is that the distance from the substrate interface at which the abnormal increase occurs corresponds well with the thickness at which the initial growth of GaN becomes fully coalescent. Regardless, the abnormal [O] profile renders the effect of oxygen diffusion near the substrate interface nearly unobservable. We can however see that [O] exhibits a longer “tail” (out to ∼400 nm) in samples E and F (marker line 3 in Fig. 5(a) indicates where [O] approaches the background level in both of these samples), compared to the arguably much sharper change in sample A [in Fig. 5(b), [O] reaches the background level near marker line 1]. This indicates that the out-diffusion of oxygen from the substrate is reduced due to BN, which is consistent with the lowest [O] background and the highest resistivity of the GaN buffer in sample A. These results clearly suggest that a good quality BN plays an important role in effectively reducing residual donor impurities in GaN.

FIG. 5.

SIMS profiles for background impurities [O], [C], and [Si] in (a) GaN grown on sapphire substrates with GaN and AlN as nucleation layers, where only one [C] and [Si] profiles were shown; (b) GaN grown on BN/sapphire, where [B] is also shown as a marker for substrate interface. Marker lines: 1 indicates the point at which the abnormal increase of [O] occurs for AlN NL; 2 indicates the same for GaN NL; and 3 indicates where the [O] approaches the background level in both samples E and F.

When the 3D bulk conduction in the GaN buffer ‘freezes out’ at low temperature, the measured Hall mobility at 10 K (µ_{10 K}) represents the transport of the 2DEG alone. For a simple relation µ _i = eτ _i /m*, where τ _i is a relaxation time (or 1/τ _i is scattering rate) for a process i, Matheissen's rule provides a way to combine the effect of the different scattering processes, assuming they act independently. We found that the typical µ_{10 K} values of our samples (from 15,000 to 33,000 cm²/V·s with n _2d ∼1 × 10¹³ cm⁻²) are in the range where the dominant scattering mechanism limiting µ_{10 K} is interfacial roughness between the GaN and the AlN insertion layer. The scattering rate associated with such interfacial roughness (1/τ_ir) has been treated by Jena et al. ^{Reference Jena, Smorchkova, Elsass, Gossard and Mishra24} for AlGaN/GaN, which correlates µ with n _2d for a given interface roughness (characterized by root mean square roughness height and correlation length). In Fig. 6, a dashed-line depicts this characteristic dependence of µ on n _2d described in Ref. Reference Jena, Smorchkova, Elsass, Gossard and Mishra24, and µ_{10 K} values obtained in this work are plotted as a function of n _2d. Good agreement is seen between experimental µ_{10 K} values and the predicted characteristic of mobility under the limit of interface roughness scattering. Scatter can be attributed to variations in interface roughness. The highest values of µ_{10 K} are associated with undoped GaN buffer, as shown in solid symbols in Fig. 6. Additionally, the quality of the BN can affect the AlGaN/GaN in many ways. As discussed before, we observed little dependence of crystalline quality on the AlN NL thickness [Fig. 1(c)], whereas the BN quality has shown a significant impact. The effect of the BN layer quality was also observed in the 2DEG mobility at low temperature, showing a correlation with BN roughness. In this regard, optimization of the BN layer is likely as important as optimization of the AlN NL. A systematic study is underway to understand the effect of the BN layer on the quality of the subsequent III-nitride.

FIG. 6.

Low temperature mobility (10 K) is plotted as a function of 2D sheet carrier density, where open circles are for samples grown on Fe-doped GaN buffer and solid symbols are for samples grown on undoped GaN buffer, and dashed-line depicts calculated dependence ^{Reference Jena, Smorchkova, Elsass, Gossard and Mishra24} for a given interface roughness with RMS roughness of 0.2 nm and correlation length of 1 nm.

To further assess the quality of the GaN/BN layer, PL measurements were performed at room temperature, as well as at 10 K. Figure 7(a) shows the PL spectrum obtained at room temperature. A strong near-band-edge (NBE) emission was observed at 3.43 eV, while defect-related yellow band (at ∼2.5 eV) and blue band (at ∼2.9 eV) emissions ^{Reference Reshchilov and Morkoc32} were barely noticeable. In Fig. 7(b), the PL spectrum recorded at 10 K exhibits two strong emissions at 3.485 and 3.491 eV, corresponding to neutral-donor-bound-exciton (D^oX_A) and free-exciton (X_A) transitions, respectively. ^{Reference Viswanath, Lee, Yu, Kim, Choi and Hong33} On the high energy side of the spectrum, emissions due to free-excitons X_B and X_C were also visible on the expanded scale. The free-exciton transition energies are close to the reported values in undoped GaN grown on sapphire by MOCVD. ^{Reference Kim and Andersson31} On the low energy side of the spectrum, a few LO phonon replicas at energy separations that are multiples of ∼91 meV are visible, with decreasing intensity as the order increases. An emission corresponding to donor acceptor pair (DAP) transition, typically observed at ∼3.27 eV, ^{Reference Reshchilov and Morkoc32} is not detectable in our GaN grown on BN/sapphire. A very weak and broad emission peak is observed in the range of 2–2.5 eV, corresponding to the defect-related yellow luminescence, ^{Reference Reshchilov and Morkoc32} shown in the insert of the Fig. 7(b) with an expanded scale. Overall, the PL indicates the GaN grown on BN/sapphire is of good optical quality.

FIG. 7.

(a) PL spectrum recorded at room temperature, where YL and BL indicate the positions of yellow band and blue band emissions commonly observed near 2.25 and 2.9 eV respectively. ^{Reference Reshchilov and Morkoc32} (b) PL spectrum measured at 10 K. Inset shows yellow band with an expanded scale.

Residual strain in the heterostructure is another important issue affecting device performance, which can be addressed by growing on a 2D BN template. To estimate the residual strain in GaN grown on BN/sapphire, Raman spectroscopy was performed at room temperature on samples grown on BN/sapphire and compared with layers grown on SiC and sapphire substrates. It is known that biaxial strain in c-plane can be well-represented by the frequency shift of the E₂ high ( ${\rm{E}}_2^{\rm{H}}$ ) mode in GaN. ^{Reference Zhao, Xu, Xie and Tong34–Reference Kitamura, Nakashima, Nakamura, Furuta and Okumura36} Figure 8 shows the ${\rm{E}}_2^{\rm{H}}$ mode measured on four samples. For comparison, we adopted the value of 567.6 cm⁻¹ for the ${\rm{E}}_2^{\rm{H}}$ mode phonon frequency in “stress-free” GaN from the literature. ^{Reference Harima37} A blue shift of the ${\rm{E}}_2^{\rm{H}}$ peak by 2 cm⁻¹ can be seen in GaN on sapphire, while a red shift of less than 1 cm⁻¹ is seen for GaN grown on SiC. Consistent with the findings reported in the literature, the GaN layer grown on a sapphire substrate with a LT-GaN NL is under compressive stress, while the GaN grown on SiC with an AlN NL is under tensile stress. For samples grown on BN/sapphire, a shift similar to that of GaN grown on sapphire was observed in Fig. 8, indicting a similar degree of compressive stress in the GaN. However, the spectrum collected on the same GaN layer after release from the substrate showed an ${\rm{E}}_2^{\rm{H}}$ peak close to 568 cm⁻¹, indicating stress relaxation. Using the ${\rm{E}}_2^{\rm{H}}$ frequency change with respect to “strain free” value, Δω (cm⁻¹) = Cσ _xx , where C is the linear stress coefficient, we can estimate biaxial stress (σ _xx ) in each sample (results given in Table III). Since experimental values for stress coefficients in the literature are scattered due to the different methods by which biaxial stress is evaluated, ^{Reference Harima37} we adopt an experimental value of 2.9 cm⁻¹/GPa, which is in better agreement with a theoretical value (2.4 cm⁻¹/GPa) calculated using first-principles. ^{Reference Wagner and Bechstedt38} Estimated stress values given in Table III are an average over multiple (25) spectra collected at the same location for each sample, where the ${\rm{E}}_2^{\rm{H}}$ peak frequency in each spectrum was obtained through curve fitting rather than from the measured raw data. The standard deviation (STD) represents the variation in stress among the 25 spectra analyzed. It is seen that GaN layer grown on BN/sapphire remains compressively stressed until the layer is released from the substrate, consistent with a previous report ^{Reference Kobayashi, Kumakura, Akasaka and Makimoto14} for GaN grown on BN. Comparing the residual stress in GaN before and after it is released from the substrate, we also notice that the stress represented by the shift of the ${\rm{E}}_2^{\rm{H}}$ peak frequency is well correlated with a shift of the NBE peak (∼8 meV) observed in the room-temperature PL spectrum. On a released GaN film, Raman measurements were performed to check the sp²-bonded BN on the backside of the GaN film and on the substrate surface. The signature Raman shift for sp²-bonded BN was observed on both sides at ∼1370 cm⁻¹, indicating that the bonding between the 3D and the 2D surfaces (i.e., substrate/BN or GaN/BN) is stronger than the pure van de Waals bonding between adjacent BN layers.

FIG. 8.

Raman spectra collected at 300 K showing the ${\rm{E}}_2^{\rm{H}}$ mode from GaN layers grown on sapphire, SiC, BN/sapphire, and released from BN/sapphire, as indicated. The black vertical dashed line at ω = 567.6 cm⁻¹ represents ${\rm{E}}_2^{\rm{H}}$ mode phonon frequency in “stress free” GaN. ^{Reference Harima37}

TABLE III.

Biaxial stress in GaN grown on different substrates calculated from ${\rm{E}}_2^{\rm{H}}$ peak shift with respect to ${\rm{E}}_2^{\rm{H}}$ peak frequency of 567.6 cm⁻¹ in stress-free GaN. ^{Reference Zang and Chua35} (STD represents the variation in stress; NBE is from room temperature PL).