Composite GaN-C-Ga ("GaCN") Layers with Tunable Refractive Index.

This article describes novel composite thin films consisting of GaN, C, and Ga (termed "GaCN", as an analogue to BCN and other carbonitrides) as a prospective material for future optical applications. This is due to their tunable refractive index that depends on the carbon content. The composites are prepared by introducing alternating pulses of trimethylgallium (TMG) and ammonia (NH3) on silicon substrates to mimic an atomic layer deposition process. Because the GaCN material is hardly reported to the best of our knowledge, a comprehensive characterization is performed to investigate into its chemical nature, primarily to determine whether or not it exists as a single-phase material. It is revealed that GaCN is a composite, consisting of phase-segregated, nanoscale clusters of wurtzitic GaN polycrystals, in addition to inclusions of carbon, nitrogen, and gallium, which are chemically bonded into several forms, but not belonging to the GaN crystals itself. By varying the deposition temperature between 400 and 600 °C and the NH3 partial pressure between 0.7 × 10-3 and 7.25 mbar, layers with a wide compositional range of Ga, C, and N are prepared. The role of carbon on the GaCN optical properties is significant: an increase of the refractive index from 2.19 at 1500 nm (for carbon-free polycrystalline GaN) to 2.46 (for GaCN) is achieved by merely 10 at. % of carbon addition. The presence of sp2-hybridized C=N clusters and carbon at the interface of the GaN polycrystals are proposed to determine their optical properties. Furthermore, the formation of the GaN polycrystals in the composite occurs through a TMG:NH3 surface-adduct assisted pathway, whereas the inclusions of carbon, nitrogen, and gallium are formed by the thermal decomposition of the chemisorbed TMG species.

Introduction

To keep up with the performance improvement of chips at a time when we are at the limits of the existing silicon (Si)-based integrated circuits (IC) technology, the introduction of novel materials, to supplement the capabilities of the current technology, is a strong requirement. Gallium nitride (GaN), one of the group III nitride semiconductors (the other well-known examples are aluminum nitride (AlN) and boron nitride (BN)), is an important candidate because of its attractive properties. Among them, large breakdown field and piezoelectric coefficient, high carrier mobility, and a wide and direct bandgap[1] can be mentioned. Combined with the mature Si technology, GaN and the other group III nitride semiconductors have the potential to expand the range of superior devices beyond the known high electron mobility transistors (HEMT)[2] and light-emitting diodes (LED).[3,4]

At present, GaN is widely used in LEDs.[5−9] The outstanding material properties of GaN can be optimally exploited in its monocrystalline form, and therefore in LEDs, it is grown epitaxially on specific substrates, usually on sapphire.[10] In parallel, studies on perfecting the epitaxial growth of crystalline GaN films on Si are pursued to enable the use of the low-cost mature Si technology and the higher electronic and thermal conductivity of Si.[8,9] The growth of high-quality monocrystalline GaN on Si substrate suffers from several problems (melt-back etching,[11] large lattice mismatch, and creation of a large tensile stress postgrowth[12]), which deteriorate the quality of GaN and ultimately the device performance.[8,9,13]

In parallel to growing thick (a few micrometers) epitaxial GaN films on Si, the exploration of the properties of thin (submicrometer) polycrystalline GaN (poly-GaN) films also attracts attention in view of several electron device concepts.[14−20] Because the requirement of a high crystal-quality is less crucial for poly-GaN, the growth can be performed at lower temperatures than required to obtain epitaxial GaN and can be accomplished on a wide variety of substrates including Si.[21−24] Poly-GaN can further be deposited by means of a wide variety of techniques,[18,21] and among them, atomic layer deposition (ALD)[22,24−27] is an industrially relevant approach to fabricate modern electronic devices due to enabling the precise thickness control and excellent film conformality on complex device structures.[28]

Poly-GaN potentially allows the addition of other elements in quantities far beyond the doping level, enabling the further exploration of material properties. Such elements can be, for example, located at grain boundaries between the polycrystals. The addition of controlled amounts of carbon to certain group III nitrides has opened up the phase triangle of the metal, carbon, and nitrogen. This has allowed tuning the various material properties by changing the relative composition of the three components. For instance, in the boron carbonitride (BCN) system, the composition has a profound influence on the mechanical and optical properties.[29−32] Importantly, these material properties were composition-dependent, irrespective of the fact whether the film consisted of a single homogeneous chemical phase or multiple phases with different chemical states of boron, carbon, and nitrogen.

A related carbonitride system which is hardly reported is that of Ga. A reason behind the lack of reports on this potential material is perhaps because of the focus on growing stoichiometric, epitaxial GaN layers to satisfy the demands of the optoelectronic industry. For these applications, carbon is occasionally added as a dopant, but higher concentrations are regarded to be detrimental to the device performance.[33] A second reason is the thermodynamically unstable nature of gallium carbide,[34] prohibiting its use as a stand-alone material. However, whether or not gallium carbonitride (GaCN) can exist as a single-phase material or as a mixture of chemical phases containing Ga, C, and N in various proportions, investigating its properties in view of potential applications should not be ignored.

In this work, we report on the deposition, detailed material characterization, and a plausible explanation behind the high refractive index as well as study the growth kinetics of composite GaCN thin layers. The composite is found to be phase-segregated in the nanoscale, consisting of polycrystalline wurtzitic GaN clusters, along with Ga, C, and N chemically bonded in several fashions. The refractive indices of the GaCN layers are dependent on their carbon content. An index as high as 2.46 at a low-dispersion region of 1500 nm of layers containing 10 at. % carbon has been achieved in this work. We identify that sp2-bonded C=N species present in the composite, as well as carbon residing at the interface of the GaN polycrystals, are responsible for determining the refractive index and the optical bandgap of GaCN. Because of the possibility of tuning these properties by controlling the carbon addition, we show the dependence of the growth rate and the material composition on deposition temperature and NH3 partial pressure.

Experimental Section

The GaCN layers,a with thickness varying between 7 and 110 nm, were deposited on 4 in. Si wafers from alternating pulses of trimethylgallium (TMG) and ammonia (NH3) to eventually mimic an ALD process. An inert (argon) gas purge was applied between consecutive precursor doses. The depositions were performed between 400 and 600 °C at a range of reactor pressures between 0.01 and 10 mbar, corresponding to NH3 partial pressures between 0.7 × 10–3 and 7.25 mbar, respectively. A polycrystalline AlN layer (25 nm thick) was first in situ deposited on Si by ALD as a buffer layer. The pulsed deposition approach was undertaken to minimize gas-phase reactions between the precursors and grow the layer merely through surface reactions. The reactions of TMG and NH3 in the gas phase might produce particles at the deposition temperatures used here[35] and should thus be avoided.

Several thin film analysis techniques were used in this work. The layer thickness and morphology were studied by a Merlin scanning electron microscope (SEM) from Zeiss equipped with an energy selective backscattered (ESB) detector and an Icon atomic force microscope (AFM) from Bruker, respectively. The phase segregation study was performed by a Fourier transform infrared (FTIR) spectrometer from Thermo Scientific, a Quantera SXM X-ray photoelectron spectrometer (XPS) from PHI, a X’pert Powder X-ray diffractometer (XRD) from Malvern Panalytical, and a CM300ST-FEG transmission electron microscope (TEM) from Philips. The layer growth was in situ monitored, determining the thickness and the optical constants (i.e., refractive index and extinction coefficient) by spectroscopic ellipsometry (SE) using a Woollam M-2000 ellipsometer and J.A. Woollam CompleteEASE software. The optical constants were determined by a multisample analysis approach[36] using a Kramers–Kronig consistent B-spline parametrization-based optical model,[37] while also accounting for the surface roughness of the layers (see the Supporting Information for details).

Results and Discussion

Characterization of the GaCN Composite

Layer Cross Section and Morphology

Figure a shows the cross-sectional ESB image of a representative GaCN layer. The AlN and the GaCN layer thicknesses are revealed to be 24 and 114 nm, respectively. The plan-view image, as acquired with SEM (Figure b), reveals a grainy layer with an average grain size of 73 nm (with a standard deviation of 16 nm). The grains were also observed with AFM (inset) and found to be of a similar size.

Figure 1

(a) Cross-sectional ESB image of a representative GaCN layer deposited on an AlN buffer layer. The substrate is Si (111). (b) Top-view SEM image of the same layer which reveals a grainy morphology. Inset: an AFM image of the same surface.

Bonding and Phase-Segregation Analysis

FTIR Spectroscopy

All the GaCN FTIR analyses were done in the transmission mode in the infrared (IR) range from 400 to 4000 cm–1. In addition, a reference stoichiometric (according to XPS) polycrystalline GaN sample was also analyzed and compared with the GaCN material. Similar to the GaCN samples, this reference GaN sample was deposited from TMG and NH3 precursors[38] on a 25 nm thick AlN buffer layer. The spectral transmittance was calculated as the ratio of the sample single beam spectrum to the background single beam spectrum.[39] The background spectrum was acquired with a sample containing an ALD AlN layer on a Si substrate (see section ). In this IR range, only vibrational modes of the GaN and GaCN layers contributed to the spectral transmittance. The FTIR spectra of a representative GaCN layer and the reference GaN layer are shown in Figure . The spectra have no absorption bands to be attributed to the AlN buffer since AlN is reported to absorb around 610 cm–1 (A1 (TO) mode), 655 cm–1 (E22 mode), 675 cm–1 (E1 (TO) mode), and between 884 and 905 cm–1 (A1 (LO) mode).[40−42]

Figure 2

FTIR spectra of the reference GaN and representative GaCN sample, showing their respective absorptions.

Both the GaN and GaCN samples show strong absorption at 530 cm–1, corresponding to Ga–N stretching vibration in wurtzitic GaN.[43] Irrespective of the bulk composition of the GaCN layers containing different ratios of Ga, C, and N, all the samples exhibit absorption around the same wavenumber. This absorption indicates the presence of Ga–N bonds in the GaCN layers, which could perhaps be clustered into wurtzitic GaN crystals.

The GaCN sample shows a major absorption around 1660 cm–1, which is not observed in the GaN sample. This absorption can occur due to[44] (i) deformation of the sp3-hybridized N–H bonds in primary amines, (ii) sp2-hybridized C=N stretching vibration in imines, azines, and so on, and (iii) sp2-hybridized C=C stretching vibration in hydrocarbons. On the other hand, whereas the GaN sample has a band at 945 cm–1 (corresponding to the NH2 bending mode), the GaCN layer hardly shows any absorption there. Similarly, the NH and NH2 band at 3250 cm–1 (i.e., the NH2 stretching mode) is much stronger in the GaN layer. These observations rule out the formation of sp3-hybridized primary amines in the GaCN. Instead, the presence of sp2-hybridized C=N bonds is a stronger possibility, and this will be further addressed in the following subsection. The hydrocarbon sp2-hybridized C=C stretching mode is reported to have an overtone absorption between 1800–1850 cm–1.[44] This in fact appears in the GaCN layers at 1840 cm–1 but not in the GaN sample. Interestingly, when the GaCN layers become richer in carbon, the absorption at 1660 cm–1 decreases. This suggests to attribute the 1660 cm–1 absorption to the sp2-hybridized C=N bonds instead of C=C bonds.

Carbon can be additionally present in GaCN through incompletely decomposed TMG residues (“fragments”), such as Ga-(CH3)2 (i.e., dimethylgallium or DMG) and Ga-(CH3) (i.e., monomethylgallium or MMG). For instance, the absorption at 746 cm–1 has been assigned to CH3 rocking vibrations in TMG.[45] The weak band at 758 cm–1 in Figure (which is stronger in GaCN) can therefore be attributed to such TMG decomposition products.

The overshoot in the spectral transmittance at 2365 and 2328 cm–1 occurs due to the presence of small amounts of CO2 in the spectrometer. Because of the high extinction coefficient, even slight variations of the CO2 partial pressure between the background scan and the actual measurements cause a large overshoot.

XPS Analysis

Sputter depth profiled XPS (using 2 kV Ar+ ions) was performed on the layers to obtain bonding information from the bulk. An Al Kα monochromatic X-ray source was used. The surface carbon XPS peak position was calibrated at 284.8 eV to coincide with the ambient hydrocarbon contamination.[46] For obtaining the bonding information in the layers, the Ga 2p3/2, N 1s, and C 1s spectra were analyzed. All spectra were deconvoluted by Gaussian–Lorentzian bands after assuming an iterated Shirley background.

The C 1s Spectrum. A typical C 1s XPS spectrum from a GaCN layer, acquired during the sputter depth profiling, is shown in Figure a. The shape suggests multiple bonding environments of carbon. The spectrum can be best deconvoluted into three bands, with binding energies (BE) of 286.3, 284.7, and 283.5 eV. The highest BE band position matches precisely with that of the sp2-hybridized C=N bonds, for instance, as reported for CN and BCN layers.[32,47] This supports the previous attribution of the FTIR band at 1660 cm–1. The observation of the C=N bond (and its attribution to a separate double-bonded entity), however, does not so far exclude the possibility of carbon being chemically bonded to the GaN crystals, although no evidence for the latter was found from the FTIR. The XPS band at 284.7 eV is near the position of sp2-bonded carbon, reported between 284.2 and 284.6 eV by Barr et al.[46] This correlates with the FTIR observation of the hydrocarbon overtone signal at 1840 cm–1. (To note: the contribution of ambient hydrocarbons to the 284.7 eV peak may be ruled out as the spectrum is obtained after several sputter cycles. The XPS peaks of surface hydrocarbons have been observed to completely disappear in all carbon-free samples.) The band at 283.5 eV has an excellent agreement with C–Ga bonding reported in the TMG molecule and its decomposition products.[48,49] This observation thus again hints toward the presence of such species in the GaCN layers.

Figure 3

GaCN XPS analysis and deconvolution of (a) the carbon C 1s spectra, (b) the nitrogen N 1s and the Auger Ga L2M45M45 spectra, and (c) the gallium Ga 2p3/2 spectra. (d) Depth profiled composition of a GaCN/AlN/Si stack. The GaCN layer has an average stoichiometry of Ga0.66C0.10N0.24.

The N 1s Spectrum. The N 1s spectrum provides a strong indication of the occurrence of stoichiometric GaN clusters in the GaCN layers. Because the XPS was performed with an Al-based X-ray source, the Auger Ga L2M45M45 triplet partly overlaps with the N 1s photoelectron spectrum.[50] Deconvolution of the spectrum into the corresponding bands is shown in Figure b. For the stoichiometric GaN reference sample, the best result was achieved by using two (instead of three) Auger bands for Ga and one XPS band for N 1s—the peak position of the latter corresponding with the N–Ga bonding in GaN (see Figure S1). The spectral deconvolution of the N 1s–Ga L2M45M45 complex in all the GaCN layers studied in this work reveals the existence of three Auger bands for Ga, with the midband always being noticeably stronger than the N 1s XPS band. This cannot be explained by the preferential sputtering of N over Ga,[51] as similar degrees of preferential sputtering can be expected in both GaN and GaCN. The strong midband Auger signal in the GaCN layers therefore suggests an excess amount of Ga in the GaCN. This extra Ga can occur in the form of the TMG-decomposed fragments and/or clusters of metallic Ga.

The very close peak positions of the N 1s band in GaCN (397.0 eV) and in the reference GaN sample (396.8 eV) are in agreement with the literature-reported N–Ga binding energy in GaN (397.1 eV[51] or 396.9 eV[25]). This reveals the presence of possibly chemically noninteracting GaN inclusions in the GaCN layer and is in agreement with the FTIR analysis. The weak shoulder at 399.5 eV could be related to the sp2-hybridized C=N clusters, as reported to occur at 400 eV.[52]

The Ga 2p. The excess Ga in the GaCN layers is additionally confirmed by the Ga 2p3/2 XPS spectra. For instance, the spectrum of the GaN sample (see Figure S2) reveals only a slight asymmetry of the band and is best deconvoluted using two Gaussian–Lorentzian bands. The band at 1117.6 eV occupies up to 95% of the entire area, while the band at 1116.6 eV occupies just 5%. On the basis of the electronegativity difference between Ga and N, we may assign the higher-energy band to Ga–N bonding and the lower band to Ga–Ga bonding.[53] The peak positions also show good agreement with literature-reported values for Ga–N and Ga–Ga bonds at 1117.8 and 1116.5 eV, respectively.[51,54] The small amount of Ga–Ga bonds observed in the stoichiometric GaN reference sample can possibly be explained by preferential sputtering taking place during the depth profiling.

On the contrary, the spectrum of the GaCN layer reveals a significant degree of asymmetry (Figure c). Spectral deconvolution reveals a strong presence of the Ga–Ga band at 1116.5 eV compared to the band at 1117.3 eV, with the former now occupying 66% of the peak area. Unlike in the GaN sample, carbon is also present in the GaCN sample, partially bonded to Ga (recall Figure a). However, distinguishing between the Ga–C and the Ga–N binding energies is not straightforward due to the proximity of the peak positions (for example, in TMG-decomposed fragments, the Ga–C bond is reported at 1117.2 eV,[48] which is indeed quite close to the Ga–N bond position). The 1117.3 eV band is therefore attributed to contributions from both Ga–N and Ga–C bonds.

Depth Profiled Composition. Figure d shows the sputter depth profiled composition of the representative GaCN layer deposited on an AlN buffer layer. Besides the higher Ga levels, we observe that the C content is rather uniform (∼10 at. %) throughout the layer thickness. The exact cause of the minor and opposite gradient in the Ga and N contents is not understood at the moment.

XRD Analysis

The XRD analysis was performed to further examine the structure of the GaCN layers. In Figure S3, the θ–2θ XRD scan of the GaCN layer is compared with a grazing incidence XRD scan of the reference GaN sample. Both materials show a close match with hexagonal wurtzitic GaN when compared to the peak positions reported in the literature.[55] The d-spacing for the first three peaks of the GaCN layer diffractogram (located at 32.3°, 34.3°, and 36.6° 2θ angle), as calculated using the Malvern Panalytical HighScore Plus software, reveals 2.77, 2.61, and 2.45 Å, respectively. The geometrical d-spacing values of a wurtzitic GaN crystal (assuming the lattice constants of a = 3.186 Å and c = 5.186 Å[56]) yield 2.76, 2.59, and 2.44 Å for the (100), (002), and (101) crystal planes, respectively, whereas the reference GaN sample yielded values of 2.78, 2.63, and 2.50 Å, respectively. From these three observations, we conclude that the GaCN material consists of a significant amount of wurtzitic GaN clusters. Furthermore, no additional peak, which could have been attributed to other inclusions in the GaCN, except for those of GaN in the diffractogram is observed. This points to the amorphous or nanoscale crystalline nature of such foreign inclusions—in the latter case, with grain sizes smaller than the XRD detection limit.

TEM, Energy-Filtered TEM (EFTEM), and Energy Dispersive X-ray Spectroscopy (EDX) Analysis

Figure a shows a TEM image of a GaCN/AlN/Si layer stack. Columnar growth can be clearly observed in certain regions in the GaCN layer. The columns seem to extend from the AlN layer.

Figure 4

TEM images of (a) GaCN/AlN/Si stack, (b) zoomed-in image showing various structural features observed in the GaCN layer, and (c) coexisting polycrystalline and amorphous regions in GaCN.

A high-resolution TEM (HRTEM) image in Figure b reveals the structural nonuniformity of GaCN. The layer is composed of (i) long crystalline columns, (ii) spherical inclusions, and (iii) an interstitial matrix. The interstitial matrix revealed a range of d-spacing values (derived by performing fast Fourier transform (FFT) analysis of these regions using the Gatan DigitalMicrograph software), indicating the presence of crystalline domains of various orientations as well as amorphous regions (Figure c and the inset). Because of the predominantly polycrystalline nature of the matrix, the FFT images showed not one but several diffraction spots corresponding to the different crystallites. For instance, three regions having different crystalline nature were identified in the matrix, as shown in Figure S4. Whereas region 3 is almost amorphous, in regions 1 and 2, the peaks observed at 2.47, 2.58, and 2.78 Å have a good match with the set of d-spacing values previously obtained from the XRD analysis. This reconfirms the presence of the (101), (002), and (100) crystal planes of wurtzitic GaN clusters in these regions. Similarly, the FFT performed at the crystalline column of Figure b reveals a d-spacing of 2.74 Å, assigned to the (100) plane of wurtzitic GaN. Finally, the peaks observed at the abnormally high d-spacing (i.e., 5.70 and 6.10 Å in Figure S4) may be related to Moiré fringes visible in the TEM images. In conclusion, the agreement of the d-spacing values from both TEM and XRD analyses, and their match with the corresponding values of wurtzitic GaN, strongly suggests the presence of wurtzitic GaN domains in the GaCN layers.

The normalized EDX spectra (see Figure S5) of the matrix, the inclusions, and the column regions of Figure reveal an almost equal intensity of the Ga signal. A distinct C signal appears from the column. As EDX has a rather low sensitivity to nitrogen, the N signal was hardly observed in the three regions. However, since the d-spacing of the column did not show a deviation from the (100) wurtzitic GaN plane, and also considering the prior arguments from the XPS analysis, the carbon signal of the column cannot be directly attributed to carbon being incorporated into the GaN crystalline network. Instead, we assume the presence of carbon between the grain boundaries of the GaN crystals.

EFTEM analysis,[57] performed for Ga, C, and N elements, further provided evidence of phase separation in the GaCN layers. The zero-energy-loss TEM image is shown in Figure S6a. As shown in Figure S6b–d, the spatial distribution of the elements is clearly not homogeneous. For example, the sizes of the Ga-containing inclusions observed in the lower part of Figure S6b and the inclusions observed in the TEM image of Figure b are comparable (8–11 nm in diameter). The overlay of the individual elemental maps is presented in Figure S6e. On the basis of the relative variations of the elemental intensities, we speculate the existence of three separate chemical phases (Ga, GaN, and CN clusters) in the layer (Table S1).

GaCN Optical Constants Determined by Spectroscopic Ellipsometry

Figure a shows the consistently higher refractive indices of GaCN layers in comparison with the polycrystalline GaN reference sample. The refractive index (n) of the reference sample (2.19, at a low-dispersion region of 1500 nm) reasonably agrees with the index of 2.16 for polycrystalline GaN films reported elsewhere.[25] From the composition of the GaCN layers, it becomes apparent that carbon plays a critical role in determining the refractive index, and interestingly, even small changes in the GaCN bulk composition (a few at. %) result in a significant change in the refractive index (Figure a and Figure S7a). The layers also have a larger extinction coefficient (k) compared to polycrystalline GaN, as shown in Figure b (and Figure S7b), which is again related directly to their carbon content. Figure S8 shows the optical bandgap (Eg) variation of GaCN with C content, as obtained from Tauc plots.[58] Unlike GaN, GaCN is a composite material. Therefore, the obtained Eg values do not reflect true optical bandgaps of any single component but merely describe the onset of absorption in the material.[59] For constructing the Tauc plots, a power factor of 2, which corresponds to a direct bandgap semiconductor,[60] was used for the GaN sample. The so-obtained Eg of 3.59 eV for the carbon-free GaN compares fairly with the previously reported bandgap values of polycrystalline GaN layers (3.62 eV[61] or 3.90[62]). The same power factor was used for the GaCN layers to make a fair comparison. The increasing n and k values (Figure and Figure S7) and the decreasing Eg (Figure S8) with the carbon content in the GaCN layers suggest that the matrix (see section for the matrix definition) and/or defect states in the GaN crystals (e.g., carbon at the grain boundaries) might be contributing to the optical property changes. This will be further discussed at the end of this section.

Figure 5

Variation of (a) refractive index and (b) extinction coefficient of GaN and GaCN layers as a function of wavelength and photon energy.

Because literature on GaCN layers is hardly existent, we refer to some other carbonitride systems to compare and discuss their optical properties. The significant increase in the refractive index and the decrease in the optical bandgap with changes in the layer composition has been widely reported for thin film carbonitrides. For example, in BCN films, these optical variations have been attributed to their increasing C content.[31,32,52,63−69] Remarkably, the optical constants had been affected in both single-phase and multiple-phase instances of these layers. However, contradictory reports on the role of carbon also exist. For example, in the work of Shi et al.[70] on CN layers, the authors report a decreasing Eg with an increasing N/C content. This demands an investigation into the possible causes of the optical property variation of the GaCN layers.

Not just the overall carbon content, but specifically the sp2-hybridized C=N and C=C bonds are likely candidates that can increase the refractive index (and decrease the optical bandgap) of carbonitrides, and relatively small amounts may be sufficient to cause a strong impact on their optical properties. For instance, in the study of Lei et al.,[31] the BCN layers consisted of a mixture of sp3-bonded B–N and C–N as well as sp2-bonded C=N microdomains. With an increase in the deposition temperature, the ratio of the sp3 to sp2 CN domains in the layers increased, as revealed from XPS spectra. This resulted in low refractive indices and wide optical bandgaps of the high-temperature-deposited layers.

In their BCN layers, Kimura et al.[52] reported a strong decrease of the optical bandgap with increasing carbon content—by 0.9 eV from only 3 at. % increase of C. The authors attributed this change to an increased share of amorphous regions in the layers that were rich in sp2-bonded C=C and C=N species, whereas the sp3-bonded B–N and C–N species appeared to decrease with increasing carbon. The same authors in another publication[69] demonstrated two BCN samples with very similar C content (both ∼13 at. %) but with up to 1 eV difference in the Eg values. The lower Eg sample was revealed to contain a higher amount of sp2 C=N over sp3 C–N bonded species. If indeed the share of carbonaceous amorphous clusters (i.e., containing the sp2-bonded species) increased with the carbon content, then it coincides with similar observations made by Yuki et al.[30] and Sulyaeva et al.[64] Moreover, not only in BCN but also in CN films,[70] the authors attributed the decrease of Eg to a higher fraction of sp2 over sp3 bonded amorphous carbonaceous clusters in the material.

Recalling Figure , the refractive indices of the GaCN layers increase with their C content. The XPS analysis demonstrates a clear decrease of the sp2 to sp3 bond ratio with an increasing C content (Figure S9). As discussed previously, the 286.3 eV BE peak can be attributed to sp2 C=N bonds whereas the 283.5 eV peak to sp3 C–Ga bonds. In the GaCN sample with the lowest carbon content (3 at. %), the 286.3 to 283.5 eV band area ratio is the largest, suggesting the presence of sp2 C=N bonds for the majority of carbon, when the latter’s overall content is low. This may explain why changes in the GaCN optical constants already take effect at low carbon levels, e.g., from 3 at. % (Figures S7 and S8). The sp2 C=N share then decreases gradually with increasing the C content. From Figure S9, beyond 10 at. %, the carbon is almost entirely in the sp3 C–Ga form. However, because of the extremely small (<10 nm) thickness of all our C-rich GaCN layers (with C content >20 at. %), it was hardly possible to determine their optical constants by SE. On the basis of the above discussion, we tentatively assign the high refractive indices and the low optical bandgap values of GaCN composites to the presence of sp2-hybridized chemical bonds.

The Ga0.66C0.10N0.24 layer not only has the highest refractive index and lowest optical bandgap, it also shows a strong sub-bandgap absorption (Figure b). As Figure S7b reveals, such absorption is present in all the GaCN layers and increases with the C content. Its possible causes may be (i) electronic transitions in the matrix and the inclusions surrounding the GaN crystals and (ii) electronic transitions in the GaN crystals involving states that arise from defects in the lattice,[71] possibly due to carbon at their interfaces. The second cause predicts that the absorption coefficient (α) below the bandgap as a function of photon energy (E) should follow the Urbach relation: α ∝ exp(E/EU), where EU is the Urbach energy.[59] The α values generated by the B-spline optical model are fully consistent with the Urbach relation (Figure S13), so that the sub-bandgap absorption in the GaCN layers can be ascribed to C-induced defect states in GaN. However, we also cannot exclude the first cause for the sub-bandgap absorption, since the composite may contain phases that produce spectral absorption of a similar shape. An alternative possibility such as the presence of a single-phase GaCN material is ignored since it was not detected by the previous analyses.

The ability to tune the carbon content gives an opportunity to prepare GaCN layers with a fairly wide range of optical constants. The next section, focusing on the growth kinetics, discusses the exact role of deposition parameters, i.e., temperature and NH3 partial pressure, on the layer composition.

Growth Kinetics of the GaCN Composite

The material characterization suggested the occurrence of a mixture of phases in the GaCN, i.e., the coexistence of GaN clusters, excess Ga, and C in different chemical forms. The formation of such a layer can occur due to an interplay between two separate chemical pathways taking place during the layer growth. The first pathway can likely contribute to the formation of the polycrystalline stoichiometric GaN clusters, while the second pathway is expected to lead to the Ga and C inclusions. Identification of the first pathway is a novelty of our work: the initial results have recently been presented elsewhere.[38,72]

Formation of Stoichiometric GaN Clusters

Because of the chemical stability of NH3 at the temperatures used in this study (400–600 °C),[73] the low temperature deposition of polycrystalline GaN films from TMG and NH3 precursors is generally considered to be not facile without an extra activation of NH3, e.g. by plasma. The majority of reports on ALD of poly-GaN films from TMG and NH3[24−26] have therefore employed plasma to produce NH (x = 0–2) radicals to further react with TMG and thereby enable GaN growth at low temperatures. The previous works might have ignored the existence of alternative surface reactions between TMG and NH,which can occur at temperatures as low as 400 °C, without the assistance of the radicals. The formation of the GaN polycrystals in the GaCN layers is speculated to occur by such a chemical route, which is discussed below.

The route relies on the strong adduct-formation tendency of the Lewis acid TMG and the Lewis base NH3,[74,75] expressed as (CH3)3–Ga + NH3 ↔ (CH3)3–Ga:NH3.[76] In several reports exploring the growth chemistry of metal–organic chemical vapor deposited (MOCVD) GaN films, the formation of the gas-phase TMG:NH3 adduct species with its derivatives has been experimentally observed[75−82] and theoretically explored.[83−87] The reaction is reported to be reversible in nature, and therefore its rate is strongly dependent on the NH3 partial pressure.[76,82] Instead of premixing TMG and NH3 in the gas phase (as in MOCVD), ALD introduces the precursors by subsequent pulses. In this case, we speculate the occurrence of an equivalent surface adduct species as a result of the association (physisorption) of NH3 molecules on TMG-chemisorbed surface sites. Similar to the gas-phase counterpart, the rate of the surface adduct formation is speculated to be dependent on the NH3 partial pressure.

Although not experimentally detected to the best of our knowledge, the existence of the TMG:NH3 surface adduct (or a similar adduct species) has been supported by Mazzaresse et al.,[88] who observed that TMG and NH3 combine “on or near the surface” of the growing GaN film during MOCVD. From this study, the exact chemical structure of the adduct was, however, not clear. Furthermore, in a theoretical study of MOCVD GaN growth, Sengupta et al.[89] suggested that one of the pathways could involve surface reactions between TMG and NH3, resulting in the same surface adduct species (the so-called “TCOM1(s)” species in their work). Importantly, the role of the surface adduct in ALD of GaN or GaCN layers has not been emphasized prior to this work.

The experimental evidence of the formation of a similar surface adduct from trimethylaluminum (TMA) and NH3 has been reported for the ALD of AlN films. The works of Bartram et al.[90−92] and Mayer et al.[93] have experimentally confirmed the formation of the TMA:NH3 surface adduct (resulting from the physisorption of NH3 on a chemisorbed-TMA surface site) and, thereafter, the conversion of the surface adduct into an AlN unit. The latter involves a reaction between the NH3 of the adduct and a neighboring −CH3 surface group, leading to the elimination of a CH4 molecule and the formation of a “–NH2–” linkage between the adjacent Al atoms. Such “Al–NH2–Al” linkages are expected to laterally propagate across the surface after each saturating NH3 exposure, ideally forming a monolayer of AlN. Assuming similar surface reactions for the GaCN growth, we hypothesize the formation of the GaN clusters in the GaCN composite through the two-step mechanism: (i) the formation of the TMG:NH3 surface adduct species and (ii) its conversion into the Ga–NH2–Ga linkage. A detailed analysis of the kinetics of the surface-adduct-assisted ALD of GaN films will be published in a future work.

The growth per cycle (GPC) of the GaN clusters in GaCN is expected to show a dependence on the NH3 partial pressure. For instance, if the surface adduct formation is indeed reversible, the surface density of the adduct can be enhanced by maintaining a higher partial pressure of NH3 during its pulse, since this would shift the equilibrium toward the adduct formation. On the contrary, if the adduct formation is irreversible in nature, then a high partial pressure will supply more NH3 molecules for occupying the free TMG-chemisorbed surface sites. In either case, the GPC of the GaN clusters (and ultimately the GaCN layer) would increase.

To test this hypothesis, we first performed a deposition at 400 °C with alternating TMG and NH3 pulses, with inert Ar purges in between, to produce a stoichiometric, carbon-free (from XPS) GaN film[38,72] (which, in fact, was used in this work as the reference GaN sample). The GPC of this film indeed showed a strong NH3 partial pressure dependence (Figure , bottom). This is atypical of an ALD process but can be explained in terms of the proposed surface-adduct pathway. During the GaCN deposition at 550 °C, the GPC showed a similar pressure dependence (Figure , top), indicating that the GaN clusters inside the composite may have been formed by the same route.

Figure 6

Dependence of the GPC of GaN (bottom) and GaCN (top) layers on the NH3 partial pressure.

Formation of C- and Ga-Containing Clusters

During the deposition of GaCN, besides leading to the formation of the GaN clusters, the chemisorbed TMG also has the possibility of decomposing at the studied growth temperatures (400–600 °C). Being an organometallic compound, it is unstable beyond ∼400 °C,[94] which would likely result in DMG, MMG, free Ga, hydrocarbons, and carbon (e.g., as reported in temperature-programmed desorption studies by Lee at al.[48]). These species are further not able to participate in the adduct mechanism to form the GaN clusters. Instead, they accumulate in the matrix as inclusions; see for example the Ga clusters in Figure b.

Because of the presumed strong dependency of the GPC of GaN clusters in the GaCN composite with the NH3 partial pressure (Figure ), the GaN formation rapidly reduces at low partial pressures. Still, the formation of the Ga- and C-containing clusters may continue due to the ongoing TMG thermal decomposition, which might not follow the same pressure dependence. To investigate this, we performed a deposition at a very low NH3 partial pressure of 0.3 × 10–3 mbar at 550 °C. Indeed, even after 900 deposition cycles, only ∼0.5 nm thick layer (as monitored by in situ SE) was formed. Inspection of the wafer postdeposition revealed a surface densely covered with unusually shaped particles, each consisting of a spherical lobe with several conical protruding “ears” (Figure and the inset). EDX and ESB analysis of the lobe and the ears showed that the former is composed mostly of Ga and the latter mostly of C. Nitrogen was not detected in these particles. Formation of such particles has also been reported during the MOCVD of GaN,[95] where their origin had been attributed to the gas-phase TMG pyrolysis at high temperatures and the agglomeration of the products.

Figure 7

Particles (with a mean size of 5 μm and a standard deviation of 1.2 μm) containing Ga- and C-rich regions, deposited at a low NH3 partial pressure of 0.3 × 10–3 mbar. Inset: ESB image of one particle.

The competing mechanisms of GaN formation and TMG decomposition during the GaCN growth can be visualized by the range of GPC values observed for depositions performed at various temperatures and NH3 partial pressures, as shown in Figure . At a partial pressure of 7.25 mbar, deposition at 450 °C has a higher GPC than at 400 °C. A similar trend is also seen for depositions at 0.7 × 10–3 mbar partial pressure. However, for both pressure regimes, increasing the temperature beyond 450 °C slows down the increase of the GPC, causing even a decrease of the GPC at 0.7 × 10–3 mbar. This coincides with the temperature at which TMG begins to decompose.[94] The decomposition rate, enhanced by the temperature, depletes the chemisorbed TMG species available for participating in the adduct reactions to form the GaN clusters. The resultant GPC accordingly drops with temperature, suggesting a dominant contribution of the adduct pathway to the overall GaCN growth.

Figure 8

Variation of GPC of GaCN layers with a wide range of compositions, deposited under different temperatures and NH3 partial pressures. The film composition (obtained from XPS) is indicated next to the data points: e.g., (49, 6, 44) corresponds to 49 at. % Ga, 6 at. % C, and 44 at. % N. Inset: variation of the carbon content with temperature and NH3 partial pressure.

The inset of Figure shows the effect of the NH3 partial pressure on the C content of GaCN layers deposited at 400 and 550 °C. The compositional trend further supports the existence of the two competing mechanisms in GaCN growth. While increasing the temperature enhances the TMG decomposition, decreasing the NH3 partial pressure suppresses the formation of GaN, but presumably not the TMG decomposition. To obtain C-rich GaCN layers, lowering the NH3 partial pressure and keeping the temperature above the TMG dissociation threshold (>400 °C) are thus required.

Conclusions

In this study, we have shown that carbon addition beyond doping levels to polycrystalline GaN resulted in the so-called GaCN composite, with the material being phase-segregated at the nanoscale. As demonstrated, the carbon content played a significant role in determining the optical properties of the GaCN layers. A refractive index as high as 2.46 (at 1500 nm) has been achieved by adding 10 at. % of carbon to the layers. Accordingly, the optical bandgap decreased from 3.59 eV (stoichiometric GaN) to 2.95 eV (GaCN). Using a wide variety of material characterization techniques (FTIR, XPS, XRD, and TEM), we concluded that the GaCN was not a homogeneous single-phase material, but rather a composite of nanoscale wurtzitic GaN clusters intermixed with Ga-, C-, and N-containing inclusions within the so-called interstitial matrix. The matrix revealed the presence of crystalline domains of various orientations as well as amorphous regions chemically bonded in several forms. The sp2-hybridized  C=N clusters in the matrix and carbon residing at the interface of the GaN polycrystals presumably caused the high refractive index and the strong suboptical bandgap absorption, respectively. The ability to tune the carbon content gave an opportunity to prepare GaCN layers with a wide range of optical constants. To control the C content in the GaCN composite, we studied the influence of the deposition temperature and the NH3 partial pressure on the GaCN growth kinetics and the layer composition. It was proposed that the GaN polycrystals in the composite were formed via a TMG:NH3 surface-adduct-assisted pathway, whereas the other inclusions in the matrix were formed by thermal decomposition of the chemisorbed TMG species.