Real-time in situ process monitoring and characterization of GaN films grown on Si (100) by low-temperature hollow-cathode plasma-atomic layer deposition using trimethylgallium and N₂/H₂ plasma

Gallium nitride (GaN) is one of the III-nitride binary compounds that has grabbed a lot of attention due to its wide and direct bandgap (3.4 eV) accompanied with a high two-dimensional electron gas mobility at the AlGaN/GaN interface.^1–4 GaN-based III-nitride alloys have found applications in high-electron mobility transistors,^5,6 high power light-emitting diodes (LEDs),⁷ photodetectors,⁸ multijunction solar cells,^9,10 and various sensors.¹¹ Moreover, GaN has also found applications in power electronics mainly owing to its high breakdown voltage and high thermal stability.^12–17 Conventionally, GaN is typically deposited by either metalorganic chemical vapor deposition (MOCVD) and molecular beam epitaxy (MBE), where average deposition temperatures reported are in the range of 800–1100 °C to achieve device quality layers.^18,19 Such high substrate temperatures are needed for efficient ammonia bond breaking and nitrogen incorporation but on the other hand increase the nitrogen partial pressure within the growth reactor. This, in turn, leads to a higher concentration of nitrogen vacancies in the GaN film, which is reported to be one of the main reasons for the higher background-carrier concentration in MOCVD-grown GaN.^20,21 Moreover, the high-temperature deposition methods cannot be used for post-CMOS integration of III-nitride layers due to the relatively low (typically below 450 °C) thermal budget limits of Si-based transistors.²² On the other hand, reducing the substrate temperature in epitaxy reveals reduced surface ad-atom surface mobility resulting in weaker structural and hence poor electrical and optical film properties.²³ Another challenge at a lower temperature is the insufficient thermal decomposition of NH₃.²⁴ 

Plasma-assisted atomic layer deposition (PA-ALD) recently emerged as an alternative low-temperature growth technique with significant potential to resolve these issues for compound thin films including III-nitride alloys.^25,26 Owing to its low-temperature, surface reaction-based self-limiting growth process featuring sequential delivery of coreactants, ALD facilitates submonolayer thickness control along with ultimate 3D conformality.^27–29 So far, promising PA-ALD results have been reported for metal nitrides including conducting nitrides (TiN, TaN, WN),^30–33 insulating nitrides (SiN_x, HfN_x),^34–36 as well as wide bandgap semiconductor III-nitrides (AlN, GaN, InN, BN).^37–41 Using various ALD plasma sources and reactor designs, III-nitride films of promising crystal quality have been grown at substantially lower substrate temperatures.^42,43 Among those, Kim et al.⁴⁴ demonstrated the first GaN thin film deposition using thermal ALD on the Si (100) substrate within a relatively higher temperature window of 500–700 °C, with a halide Ga precursor (GaCl₃) and NH₃ as a coreactant. Ozgit et al.^45,46 reported the first low-temperature self-limiting GaN deposition via PA-ALD utilizing nonhalide organometallic Ga precursors and two different plasma sources (quartz-based inductively coupled and metallic capacitively coupled hollow-cathode plasma source), demonstrating crystalline films at substrate temperatures as low as 200 °C. Motamedi et al.⁴⁷ investigated crystalline structural and electrical properties of GaN films deposited using PA-ALD with triethylgallium and forming gas as the precursor and coreactant within the temperature range of 150–425 °C. In another work,⁴⁸ they reported GaN film growth at 275 °C where the films were epitaxial for the first ∼5 nm and polycrystalline afterward. The initial epitaxial growth was not sustainable and transformed into a three-dimensional island growth regime after the first few nanometers of growth. Nepal et al.⁴⁹ demonstrated plasma-enhanced atomic layer epitaxy of ternary InGaN and AlGaN films on MOCVD-grown GaN/sapphire substrates using trimethylgallium (TMG) and N₂/Ar (75/200 sccm) plasma at 400 °C. In a more recent work by Pedersen and co-workers,⁵⁰ low-temperature ALD of epitaxial GaN on 4H-SiC (0001) was reported using tris(dimethylamido)gallium (III) with NH₃ plasma at 250 °C with a relatively high growth rate of 1.4 Å per cycle.

In this work, we investigate the low-temperature self-limiting deposition of GaN thin films for the first time via real-time in situ process monitoring, which enables to resolve the individual ligand-exchange surface reactions. TMG and N₂/H₂ plasma were utilized as the metallic precursor and the nitrogen coreactant in a compact remote-plasma-ALD system equipped with a large-diameter capacitively coupled hollow-cathode plasma source. We report the in situ and ex situ ellipsometric analysis as well as the chemical and structural characterization of GaN thin films grown on Si (100) substrates within a temperature range of 120–240 °C. Our results demonstrate that single-phase crystalline hexagonal GaN thin films are obtained at substrate temperatures higher than 200 °C and rf-plasma power values above 100 W, with stronger preferred (002) orientation when Ar-free N₂/H₂ plasma gas composition is utilized.

An OKYAYTECHALD P100 (Okyay Technologies, Inc., Turkey) remote-plasma-ALD reactor equipped with a stainless-steel capacitively coupled hollow-cathode plasma (HCP) source (Meaglow Ltd., Canada) was used to deposit the GaN films. The plasma-ALD reactor features optical ports that are connected to an FS-1 multiwavelength ellipsometer (MWE) (Film Sense LLC, NE) source and detector units. An Si (100) wafer/sample cleaning process was carried out using isopropanol, acetone, and de-ionized water in an ultrasonic bath followed by nitrogen drying, whereafter the samples were loaded into the growth reactor. The plasma-ALD chamber was then pumped to reach the base pressure of ∼20 mTorr after which N₂ carrier flow was set to 10 sccm and Ar/N₂ plasma gas flows were adjusted to 50/50 sccm, resulting in a process pressure of ∼700 mTorr. Next, a 10 min N₂-only (50 sccm) plasma cleaning process was carried out at 100 W to remove the adsorbed water-vapor from reactor walls and minimize the undesirable impacts of atmospheric exposure during the sample loading process. TMG was used as the Ga precursor at the minimum exposure time of 15 ms and N₂/H₂ plasma as the nitrogen coreactant with and without Ar. To evacuate the excess precursor molecules and reaction byproducts, Ar/N₂ purge intervals were utilized in between the TMG pulse and N₂/H₂ plasma exposure cycles. Deposition of GaN films was performed within the temperature and rf-plasma power ranges of 120–240 °C and 50–250 W, respectively.

An FS-1 MWE unit that utilizes four visible (blue, green, yellow, and red) LEDs as light sources was used to monitor the film growth process in real time. The MWE enabled us to record the evolving film thickness with reasonably low data noise and sufficient time resolution to observe the individual surface reactions. While subsequent 10-cycle short runs were utilized to extract the saturation curves obtained on a single sample, 600-cycle long runs were carried out to obtain thicker GaN films for detailed ex situ materials characterization purposes. To analyze the dynamic in situ ellipsometer data, we used a layer model featuring a “pseudo” substrate representing the Si (100) substrate and the native oxide (SiO₂) layer on top along with a transparent Cauchy-layer for the grown GaN film. The pseudo-optical constants were directly inverted from the ellipsometer data acquired before starting the deposition experiment.

Ex situ ellipsometric measurements were performed using a spectroscopic ellipsometer (V-VASE, J.A. Woollam Co., Inc., NE) within the 370–1000 nm wavelength interval for three different angles (65°, 70°, 75°). CompleteEase software was used with Cauchy dispersion function to measure optical constants, surface roughness, and thickness values of the GaN thin film samples prepared on Si (100) substrates. A grazing-incidence x-ray diffraction (GIXRD) scan was performed using Cu Kα radiation with a Rigaku SmartLab multipurpose x-ray diffractometer (Rigaku Corporation, Japan) at 45 kV and 40 mA to determine the crystalline structure of the grown GaN films. Samples are scanned for GIXRD spectra in the 2θ range of 25–75° at a step size of 0.02° and 1.0 s counting time. X-ray photoelectron spectroscopy (XPS) experiments were performed with a Kratos AXIS 165 XPS instrument using a monochromatic Al K-α radiation source. Survey spectra were done at a pass energy of 80 eV, and high-resolution spectra were recorded at a pass energy of 20 eV. Binding energies are referenced to C 1 s at 285.0 eV. For some samples, a light sputtering was done using 4 keV Ar ions for 1 min. To study the atomic-resolution crystalline film structure of the grown GaN/Si samples, cross-sectional high-resolution scanning/transmission electron microscopy (HR-S/TEM) using FEI Talos F200X (Thermo Fisher Scientific, US) was carried out. Selected-area electron diffraction (SAED) and fast Fourier transform (FFT) analysis were performed on imaged samples to further confirm the crystalline structure of low-temperature GaN films.

Figure 1 summarizes the GaN saturation experiments using N₂/H₂/Ar (50/50/50 sccm) plasma where we have extracted the average growth per cycle (GPC) values from the optical thickness changes recorded via in situ ellipsometry measurements. Saturation curves are recorded as a function of plasma power, plasma duration, purge time, and substrate temperature. The plasma duration dependence of GPC shows that 20 s of plasma exposure is sufficient to achieve self-limiting growth conditions [Fig. 1(a)]. Figure 1(b) shows the GPC variation as a function of rf-plasma power, which indicates a relatively strong increase up to 150 W. This characteristic can be explained by the more effective plasma-induced ligand removal process at higher rf-power values. When plasma power exceeds 150 W, the GPC becomes relatively constant, reminding of a possible plasma-window (150–250 W) for this process. This behavior is similar to what we had observed in our previous HCP-ALD grown AlN studies, where a GPC saturation was obtained for plasma power values higher than 100 W at 200 °C substrate temperature.⁵¹ Based on our prior GaN experiments using larger plasma-ALD reactors, the high vapor pressure of TMG enabled using the shortest pulsing times, which is 15 ms for the high-speed ALD valves of use. As the current HCP-ALD reactor has a smaller footprint, the TMG saturation curve is shown in Fig. S1 of the supplemental material,⁵² which confirmed our expectations: The minimum available pulse time of 15 ms was sufficient to achieve complete surface saturation.

Figure 1(c) shows the GPC dependence on both precursor and coreactant purge times, which depicts an insignificant change confirming that the shortest 5 s purge duration at 200 °C is sufficient to effectively remove the excess precursor molecules and reaction byproducts. To remain at the safe side, we decided to use 10 s purge cycles to achieve saturation behavior for the entire temperature range of our study (120–240 °C). The dependence of GaN GPC on the deposition temperature depicts a rather linear increase within 120–200 °C (from ∼0.47 Å at 120 °C to ∼0.65 Å at 200 °C), followed by an almost constant GPC behavior up to 240 °C (∼0.67 Å), signaling the existence of a possible constant GPC ALD-window between 200 and 240 °C [Fig. 1(d)]. The stronger temperature-dependent GPC increase within 120–200 °C can be explained by activation-energy limited growth behavior, which leads to more complete surface reactions at higher substrate temperatures. We will comment on some of these noted saturation curve behavior within the following paragraphs where we correlate these results with the averaged single-cycle in situ ellipsometer data.

In situ ellipsometric analysis of GaN growth process was carried out by dynamically recording the temporal evolution of the film thickness during unit plasma-ALD cycles. Figure 2 shows how the GaN film thickness evolves during a single ALD cycle as a function of substrate temperature. Each of these four curves is obtained by averaging the recorded in situ data over five consecutive ALD cycles to average out the individual measurement noise. Strong and relatively fast TMG chemisorption was observed at all substrate temperatures including the lowest (120 °C), which indicates a rather temperature insensitive surface adsorption process during a TMG pulsing step. Thickness decay during purging is observed at all temperatures, which is mainly attributed to the desorption of physisorbed surface groups. This desorption component is more significant for a 120 °C sample, which indicates a higher portion of physisorbed surface groups when compared to chemisorbed counterparts. This result is somewhat expected as mainly the substrate temperature provides the energy needed for surface chemisorption (ligand exchange) reactions to occur. As the initial TMG chemisorption and subsequent desorption characteristics are nearly identical for 160, 200, and 240 °C samples, we can conclude that 160 °C is sufficient for the self-limiting TMG chemisorption process saturating the available surface reaction sites.

The analysis of the plasma half-cycle, on the other hand, leads to the following observations: (i) Except 120 °C sample, N₂/H₂/Ar plasma exposure featuring energetic neutral hydrogen (H*) and nitrogen (N*) radicals causes a relatively fast thickness decrease (though not as sharp as the TMG chemisorption reaction), which relates to the plasma-induced removal of surface methyl (CH₃) groups. While H* species are responsible for the removal of CH₃ groups through formation of volatile gaseous CH₄, energetic N* species aim to incorporate into the film by chemisorbing to available surface Ga- dangling bonds. (ii) The slow thickness decay in a 120 °C sample depicts a less-efficient ligand removal process, which indicates partly insufficient surface energy to enable methane formation reactions. On the other hand, there is a relatively significant difference in the thickness decay behavior during the plasma exposure half-cycle: A faster drop in film thickness is observed at 240 °C when compared to 200 and 160 °C processes. The total amount of thickness decay is also higher at a higher substrate temperature. We believe that this difference stems from the more efficient ligand-exchange reactions and the effective desorption of the reaction byproducts (CH_x ligands) at an elevated surface temperature. At higher substrate temperatures, the surface energy is comparatively higher, which aids the surface ligand-exchange reactions and leads to more facile desorption of the reaction byproducts. We have observed and reported the similar behavior in our previous AlN study, where we focused on the real-time in situ monitoring of AlN film growth via HCPA-ALD.⁵¹ 

When compared to the higher temperature samples, the film thickness at the end of 20 s plasma exposure is highest for a 120 °C sample, which we attribute to the incomplete removal of CH_x ligands, resulting in a carbon-rich amorphous film. (iii) The film thickness variation during the purge cycle at the end of plasma exposure is also noteworthy: for a 120 °C sample, the film thickness shows a rather dramatic and continuous decrease, which might be due to further desorption of weakly bonded CH_x surface groups. Moreover, it also reminds of a possibly weaker nitrogen incorporation process, as the overall thickness gain at the end of the unit cycle is the lowest at 120 °C. While a 160 °C sample shows almost no change in film thickness during the postplasma purge cycle, both 200 and particularly 240 °C samples exhibit a thickness increase in the initial few seconds of the purge period. We attribute this thickness gain to their higher surface thermal energy, which benefits both the ligand removal reactions and the nitrogen incorporation reactions.

Taking into account that the postplasma thickness increase only takes place at the highest substrate temperature tested, this gain is possibly related to the beneficial nitrogen incorporation. When the plasma power is turned off, there is a certain amount of nitrogen radicals, which will reach the substrate surface within the first few seconds of the purge cycle. A certain portion of those late N* radicals possibly react with the available surface reaction sites under lower H* presence, leading to additional nitrogen incorporation. Considering a possible 2–3 s HCP source-to-substrate transportation and surface reaction time and that we have a 1 s delay in MWE data acquisition, the film thickness stabilizes within the first 3–4 s of the purging cycle. The temperature dependence of this film thickness variation during the postplasma purge cycle also possibly confirms the more effective nitrogen incorporation at higher substrate temperatures.

As a result of the enhanced nitrogen incorporation, both 200 and 240 °C samples exhibit higher overall thickness gains, corresponding to GPC values of ∼0.66 and ∼0.68 Å, respectively. (iv) One of the open questions here is related to the nitrogen incorporation reactions: as the N₂/H₂/Ar plasma contains both H* and N* species, we would expect that ligand removal and nitrogen incorporation reactions take place simultaneously throughout the plasma exposure duration. However, the different behavior right after plasma stops hints a difference in N incorporation efficiency and possibly the related crystallization process. The exact reason for this postplasma deposition/incorporation behavior needs to be analyzed in further detail, ideally with in situ/in vacuo surface analysis experiments, which is beyond the scope of the current study.

The in situ recorded growth cycles were averaged over five ALD cycles for the plasma rf-power dependence study and are shown in Fig. 3 as a function of the substrate temperature. At the lowest rf-plasma power, i.e., 50 W, a considerably reduced TMG chemisorption followed by a weak and slow ligand removal behavior is observed within the entire 120–240 °C substrate temperature range. The main reason for this significantly reduced chemisorption might be explained by the plasma-induced ligand removal behavior observed under 50 W exposure. The rather slow thickness decrease during the 20 s plasma exposure signals an ineffective and most probably incomplete ligand removal process, which results in reduced density of available reaction sites on the film surface. This result is in good agreement with our previous AlN observations, where sub-50 W plasma power necessitated prolonged plasma exposure times to complete the ligand removal process.⁵³ These low-power grown AlN films were amorphous, indicating the insufficient surface energy needed for film crystallization.

When rf-power is increased to 100 W, a notable increase is observed in the thickness gain during the TMG chemisorption process along with an improved ligand removal characteristic, which again correlates with a higher density of available surface reaction sites. However, at 100 W, still the ligand removal process does not show the typical sharp thickness decrease, instead displaying a faster but similarly gradual and softer ligand-exchange process. It is only for rf-plasma power values at 150 W and beyond, where we start to see the sharper and faster thickness decrease in film thickness in the initial stages of plasma exposure cycles, reminding a more effective ligand removal process. However, we should also note that this behavior is not independent of the substrate temperature. The ligand removal is stronger at higher temperatures, while the rather slow removal process continues at low substrate temperatures, which is particularly evidenced in the 120 °C data. At this temperature, even at the highest rf-power value tested (250 W), the energetic plasma species are still not capable of efficiently breaking the Ga–C bonds, which can be attributed to the insufficient surface energy. As the temperature is increased to 160 °C and beyond, a more effective ligand removal process takes place for rf-power values higher than 100 W and becomes the strongest at 240 °C. Moreover, when unit ALD cycle thickness gain is analyzed—which corresponds to the growth per cycle (GPC) parameter—the general trend shows that the GPC increases with rf-plasma power, which agrees well with our saturation curve studies. We can attribute this increase mainly to the more effective nitrogen incorporation process, as we observe that the amount of chemisorption saturates within the 150–250 W power range. Therefore, we conclude that 150 W and higher rf-plasma power values result in effective ligand removal, creating higher density of available reaction sites and a more efficient nitrogen incorporation.

Figure 4 shows the recorded in situ film thickness data of the 600-cycle GaN deposition experiments at 120, 160, 200, and 240 °C. These samples were grown not only to observe the linearity behavior of the HCP-ALD process but as well to obtain thicker films for ex situ materials structural, chemical, and optical characterization including XRD, x-ray reflectivity (XRR), XPS, spectroscopic ellipsometry (SE), and high-resolution transmission electron microscopy (HR-TEM). The TMG pulse time, N₂/H₂/Ar flow rates, rf-plasma power, plasma exposure time, and purge times were fixed at 15 ms, 50/50/50 sccm, 150 W, 20 s, and 10 s, respectively, for these long runs. Though the plots look considerably linear except the second half of 120 °C run, we notice that the early growth stage in all samples exhibit a similar growth character and GPC in the first 5–10 cycles, followed by a linear growth curve. The GPC in this first linear portion of the growth curve shifts toward a slightly higher value for growth processes at 200 and 240 °C after the ∼100th cycle and proceeds linearly thereafter at this constant GPC at all substrate temperatures except 120 °C. The only exception to the relatively linear growth character is observed in the 120 °C sample, which displays a decreasing GPC trend in the second half of the growth run. The exact reason for this nonlinearity at a low substrate temperature is not clear; however, we believe that it has a strong correlation with the insufficient ligand removal process, which leads to reduced density of reaction sites on the surface, limiting the nitrogen incorporation further as the growth proceeds and results in amorphous films with high carbon impurity incorporation. The linearity curves obtained for the TMG–N₂/H₂/Ar plasma HCP-ALD process confirms the improved growth characteristics with higher GPC values at substrate temperatures 200 °C and above. The first 40 cycles of the linearity curves are shown in Fig. S2 of the supplemental material and depict no nucleation delays in the growth process on Si substrates, while showing substrate-enhanced growth behavior with higher initial GPC values.⁵² 

To gain insight into the crystalline film structure of the grown samples, x-ray diffraction scans are carried out. Figure 5 depicts the XRD measurements of various sets of GaN samples grown via HCP-ALD under different temperatures and plasma conditions. The influence of the substrate temperature on the film crystallinity is shown in Fig. 5(a), which clearly confirms the enhanced structural quality at a higher substrate temperature with more effective ligand removal and nitrogen incorporation processes observed during in situ measurements. While the GaN films grown at lower substrate temperatures (120 and 160 °C) show amorphous character with no crystal peaks at all, the 200 °C sample shows early signs of crystallization with two relatively weak and broader diffraction peaks at two angles (32.4° and 36.7°) corresponding to (100) and (101) planes of h-GaN. Increasing the substrate temperature further to 240 °C leads to a significant enhancement in crystallinity with a stronger (002) peak, while the other two neighbor crystal orientations are only slightly enhanced. To check whether these films exhibit any preferred crystal orientation, θ–2θ measurements are carried out. Both 200 and 240 °C samples exhibited a notable diffraction peak around 34.5°, which confirms the preferred orientation of these GaN films along the (002) plane. Hence, the GaN films grown at T > 200 °C can be classified as single-phase crystalline h-GaN exhibiting a preferred (002) orientation. The temperature dependence of film crystallinity implies that 240 °C, which is the highest substrate temperature in our study, corresponds to the optimal growth temperature to obtain GaN films with higher structural quality. The remaining three sets of samples where different plasma conditions are tested are, therefore, all grown at 240 °C.

Figure 5(b) compares the measured GIXRD peaks to understand the effect of removing Ar-flow from the plasma gas mixture on the crystallinity of GaN films grown at 240 °C, 150 W, and 20 s plasma exposure. It is evident from Fig. 5(b) that Ar-free N₂/H₂ plasma results in suppressed (100) and (101) shoulder peaks, with a similar but slightly broader dominant (002) diffraction peak. Finally, Figs. 5(c) and 5(d) display the rf-plasma exposure time and plasma duration dependence of the XRD spectra, respectively. While the dominant (002) peak width and intensity remains fairly unchanged within 150–200 W plasma power, the plasma exposure time dependence at 150 W shows that 10 s N₂/H₂ plasma exposure falls short with reduced (002) peak intensity, while 40 s results in a similar XRD spectrum as a 20 s sample with a very close (002) peak width and intensity. Based on these results, we decided that 20 s rf-plasma exposure at 150 W is the optimal rf-plasma operation point for N₂/H₂-based HCP-ALD grown GaN film samples. Crystallite size calculation has been done to analyze the effect of plasma power variation. The average crystallite size of the polycrystalline wurtzite GaN film is extracted from the (002) reflection at 150, 175, and 200 W and is found to be 10.5, 9.5, and 9.0 nm, respectively.

Table I shows the comparative summary of ex situ ellipsometry (multiwavelength and spectroscopic) and XRR measurements, with acquired film thickness, average GPC, index of refraction, and film density values. While the film thickness values extracted via multiwavelength and spectroscopic ellipsometry agree well, we notice a rather considerable discrepancy in the refractive index values, particularly for the 150 W film that exhibits the highest index values. We attribute this difference partly to the enhanced accuracy in the measuring and modeling capacity of the spectroscopic ellipsometer. Another possible reason, particularly for the 150 W sample where the discrepancy is the highest (2.04 vs 2.20), might be partially attributed to the increased surface roughness (4.1 vs 1.9 nm for 150 and 200 W, as measured from an SE analysis) of this sample, which leads to a more complex fitting analysis prone to a higher degree of uncertainty (MSE >5). Film density measurements were carried out using an XRR analysis and has been compared with a bulk GaN density of 6.15 g/cm³. Films grown at 150 W plasma power show density values in agreement with previously published data for polycrystalline h-GaN films with (002) preferred orientation.⁴⁶ The XRR data, which are shown in Fig. S6 of the supplemental material, in general agrees well with the trend in film thickness with a certain offset of 3–4 nm lower thickness values under all plasma conditions.⁵² The plasma power dependence of the refractive index can be due to the surface morphology, variation in crystalline orientation or near-random orientation of crystallites, and/or improved crystallinity. The value of the refractive index at 632 nm is 2.20 for the films deposited at 240 °C. The spectral refractive index and extinction coefficient data are shown in Fig. S3 of the supplemental material, extracted from the spectroscopic ellipsometry measurements.⁵² The refractive index values obtained here are in reasonable agreement with previously published reports on low-temperature plasma-ALD grown GaN samples.^54,55 At this point, we believe that the increase in the refractive index for the 150 W sample stems from the more favorable surface conditions for crystallization, which possibly leads to an enhanced 3D growth mode, revealing an increased surface roughness.^48,56 However, the experimental verification of such a claim needs systematic studies, which can analyze the evolution of the surface morphology as a function of ALD cycle numbers and plasma power.

XPS measurements were used to analyze the composition of the ALD samples along with an MOCVD-grown epitaxial sample as a reference for composition. XPS measurements are complicated by interference from Ga AES peaks that overlap with the N 1 s region for an Al x-ray source and the C 1 s region for an Mg x-ray source. Moreover, sputtering the surfaces to remove adventitious carbon from air exposure also preferentially removes light elements leaving a Ga rich surface composition. Survey spectra for both x-ray anodes comparing the ALD and epitaxial samples are presented (Fig. S4 in the supplemental material), along with high-resolution scans for the Ga, N, O, and C regions (Fig. S5 in the supplemental material).⁵² Combining data from both anodes allows for an estimate of the surface composition of air-exposed samples as 30/20/30/20 at. % for Ga/N/O/C, respectively. These can be compared with 40/30/10/20 at. % (in the same order) for the epitaxial sample. After sputtering with Ar ions for 3 min at 3 keV, the carbon is completely removed, and the composition of the ALD grown films becomes Ga rich with 56/22/22 at. % for Ga/N/O, respectively, which can be compared with 60/37/3 at. % (in the same order) for a similarly sputtered reference epitaxial sample.

HR-TEM, SAED, and high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) measurements were carried out to analyze the layer structure, atomic arrangement, and elemental/compositional mapping within the ALD-grown films. Figure 6(a) shows the HR-TEM micrograph of GaN thin films grown on an Si (100) substrate at 240 °C and 150 W. The GaN layer reveals a polycrystalline film with crystal domains in different crystallographic orientations. The cross-sectional HR-TEM image also confirms the relatively thick amorphous native oxide (SiO₂) layer between the Si (100) substrate and the GaN film. Figure 6(b) shows the FFT analysis obtained from the GaN layer only. FFT over the broader GaN region clearly shows scattered arclike diffraction rings corresponding to multiple reflections, confirming the polycrystalline structure. The pattern obtained in Fig. 6(b) consists of polycrystalline diffraction rings corresponding to the wurtzite (hexagonal) GaN (h-GaN) crystal structure.

Table II summarizes the measured ring diameters, theoretical values for h-GaN, and the corresponding crystallographic planes. The first and third rings from the center are relatively weaker pronounced. The diameter of these rings was measured and found to be 7.14 and 10.59 nm⁻¹, corresponding to the (100) and (102) planes with calculated interplanar spacing (d_hkl) values of 2.631 and 1.889 Å, respectively. The diffraction ring corresponding to the (002) plane of h-GaN (d_hkl = 2.593 Å), on the other hand, displays brighter spots on its diffraction ring, confirming the dominant orientation within the polycrystal structure. Diffraction rings corresponding to (110) and (112) were also observed in the measured SAED diffraction pattern.

The cross-sectional high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image of the highest crystal-quality GaN/Si sample is shown in Fig. 7(a). The corresponding elemental maps of Ga, N, C, O, and Si for the HCP-ALD GaN/Si sample grown at 240 °C are obtained via electron diffraction spectroscopy (EDX) and is shown in Figs. 7(b)–7(f). Si, Ga, N, and C show strong contrasts over the scanned region on their respective color-mapped micrographs. As expected, Ga and N appear to be homogenous within the bulk of the GaN layer [Figs. 7(b) and 7(c)]. The presence of O almost over the whole scanned area of the sample [Fig. 7(e)] is partly due to the atmospheric exposure of the sample. The relatively strong O signal at the Si substrate interface corresponds to the native oxide layer. On the other hand, the presence of significant carbon, oxygen, and nitrogen on the protective capping layer stems from the organic ligands of the Pt precursor used in the FIB system, while C and N intensity fades out on the scanned region over the bulk of the GaN film [Figs. 7(c)–7(e)]. However, we also notice that the oxygen content within the GaN layer is noticeably higher when compared to the O signal obtained from the Si substrate. This confirms the oxygen incorporation component within the low-temperature HCP-ALD grown samples. Therefore, we conclude that the STEM-EDX elemental mapping agrees well with the XPS extracted elemental composition for the GaN sample grown at 240 °C–150 W.

Self-limiting growth of GaN thin films on Si (100) substrates at different growth parameters was demonstrated for the first time via real-time in situ process monitoring using hollow-cathode plasma-ALD. Single-phase hexagonal polycrystalline GaN films at substrate temperatures higher than 200 °C were obtained. At lower growth temperatures, real-time in situ data confirmed a rather inefficient ligand removal process, which resulted in lower values of precursor chemisorption, reduced GPC, higher impurity incorporation, and amorphous film structure. Increasing the temperature has increased the crystalline quality of the film, which is attributed to optimal condition for sufficient chemisorption and ligand removal. Saturation experiments confirmed that the plasma power has a significant impact not only on GPC but a crystalline structure as well. Even if the temperature is set at the highest value, i.e., 240 °C, low plasma power hinders the chemisorption due to insufficient ligand removal. 150 W rf-power was necessary to obtain high chemisorption, sharper ligand removal, high-GPC, and low-impurity crystalline films. The structure of the GaN film layer was examined using a high-resolution x-ray and electron microscopy analysis, both confirming the deposition of a single-phase poly-GaN layer with a wurtzite crystal structure. GaN samples grown at 240 °C and 150 W plasma power exhibit (002) dominant crystal orientation with near-ideal slightly Ga-rich stoichiometry and below-XPS detection-limit carbon incorporation. With further room for improvement, these low-temperature grown crystalline GaN layers might be used in various device applications such as gas sensors, UV photodetectors, and thin film transistors.

The authors acknowledge the University of Connecticut, School of Engineering Startup Research Funding and the Research Excellence Program (REP) funded by the Office of the Vice President for Research (OVPR) for financial support. For helping with the XRD and XRR measurements, the authors thank Daniela Morales. The authors would like to thank Roger Ristau and Lichun Zhang for their support in FIB sample preparation and HRTEM measurements. The authors also acknowledge the financial support of the National Science Foundation grant (NSF Award No. 1511138).

The data that support the findings of this study are available from the corresponding author upon reasonable request.