Homogeneous high In content In_xGa_1−x N films by supercycle atomic layer deposition

Nanometer-thin In_xGa_1−x N films are known as very promising optoelectronic materials for efficient blue and green light-emitting diodes (LED). A monolithic full color LED has been demonstrated by stacking In_xGa_1−x N layers with various x within a complete LED structure.¹ The emission energy of In_xGa_1−x N is determined by several factors with the value of x considered as the most important. Due to the miscibility gap between InN and GaN and very high critical temperature for phase separation (1250 °C),² a homogeneous In_xGa_1−x N film is regarded to be metastable. The most common techniques to prepare In_xGa_1−x N and LED structures are metal-organic chemical vapor deposition (MOCVD) and molecular beam epitaxy (MBE) by which the In-content (x) is mainly controlled by the growth temperature and III/V ratio. Phase separation to InN and GaN is typically seen for x > 0.3,³ even in In_xGa_1−x N quantum well structures that are just a few nanometers thick.⁴ 

Recently, atomic-layer-deposition (ALD) has emerged as a promising method for the growth of III-nitrides, such as binary AlN,^5,6 GaN,^7,8 InN,^9,10 and their alloys.^11–13 The ALD approach is characterized by the alternating pulses of metal and nitrogen precursors, purged by inert gas between pulses. This makes the film growth fully dependent on surface chemical reactions. In our previous work, we showed that the gas exchange dynamics combined with plasma activation of the nitrogen precursor (NH₃) at temperatures below 350 °C was essential to achieve high structural quality epitaxial InN layers for just 5 nm thick.^14,15 Further extension of the ALD technology to ternary III-nitrides can open a new path for integration of ALD with existing LED and high-electron mobility transistors.

Unlike ternary oxides,¹⁶ ternary nitrides are much less explored by ALD. Recently, we explored the approach of co-evaporating precursors for In_xGa_1−x N, by mixing In- and Ga-triazenide precursors in the evaporator and introducing them simultaneously into the reactor.¹⁷ More common for ternary materials is to use a supercycle ALD approach based on the combination of growing respective binary compounds as a multilayer. This has been demonstrated to some extent for nitrides.^11–13 The growth of In_xGa_1−x N is accomplished by a repeating cycle that comprises of a number $(m)$ of InN ALD cycles $(UInN)$ followed by a number (n) of GaN cycles $(UGaN)$⁠. The complete growth of In_1−xGa_xN can be formulated as $(m⋅UInN+n⋅UGaN)×k$⁠, where k is the total number of super-cycles. According to the rule of mixtures,¹⁸ x in In_xGa_1−x N is determined not only by the respective number of unit cycles but also the growth per ALD cycle (GPC) as

$m$ and n are the number of unit cycles of InN and GaN, respectively. $GPC∗InN$ and $GPC∗GaN$ are the GPC of InN and GaN in the growth of In_1−xGa_x N, bearing in mind that $GPC∗InN(GPC∗GaN)$ may not be the same as GPC_InN (GPC_GaN) determined from the binary InN (GaN) process. $ρInN$ and $ρGaN$ are the atomic densities of InN and GaN. Although varying the ratio of m/n has been shown effective in tuning In-content,^11,12 the impact of GPC on the In content has not been discussed to our knowledge.

In this work, In_xGa_1−x N films with various x were prepared based on the supercycle ALD approach, in which m and n were chosen to obtain respective binary monolayers. We find that In_xGa_1−x N with 0.2 < x < 0.8 can be deposited without phase separation. We also find a low $GPC∗InN$⁠, which should decrease x (In content) but it is, on the contrary, found notably higher than the predicted values. This is ascribed to the effective intermixing of In into GaN. We believe that our results provide insight into the control of both the thickness and composition of In_xGa_1−x N by ALD and shows the importance of the ALD technique for potential In_xGa_1−x N devices.

The growth of In_xGa_1−x N was done using a Picosun R-200 plasma ALD system. Trimethyl indium (TMI), TEG, and (NH₃ + Ar + N₂) plasma were used to provide In, Ga, and N, respectively. The In_xGa_1−x N was grown at 320 °C and 6 mbar. Si (100) substrates were chemically cleaned using standard RCA cleaning procedure before loading into the reactor. After temperature stabilization at 320 °C for 60 min and prior to the growth of In_xGa_1−x N, the Si substrate was subjected to (Ar + N₂) plasma treatment for 2 min with the intention to remove residual surface oxide. After the plasma treatment, the system was purged with 100 SCCM N₂ for 10 s followed by the first TMI pulse of the InN ALD cycle. The details of the InN⁸ and GaN¹⁹ ALD processes can be found in our previous reports. The GPC of InN⁸ and GaN¹⁹ was determined to be ∼0.45 Å/cycle when grown as binary materials on silicon (100) substrates. Considering the preferred growth direction along <0001> for III-N wurtzite crystals,¹⁸ $7UInN$ and $5UGaN$ are set for a monolayer of InN and GaN.

In some experiment that was done to show the heteroepitaxy, on-axis Si-face-CMP-polished 4H-SiC substrates was used. SiC substrate was placed next to Si (100) substrate in the same growth run. The SiC substrate was subjected for the same RCA cleaning procedure as the Si substrate prior to loading into the reactor.

The surface morphologies of the InN samples were studied by a high-resolution LEO 1550 Gemini field emission scanning electron microscope. The crystalline quality, thickness, and the macroscale roughness were characterized by using x-ray diffraction (XRD) (PANalytical X'Pert Pro with a Cu-anode x-ray tube). Grazing-incidence XRD (GIXRD) and symmetric 2θ-ω and ω-scans were used to study the crystalline quality. The film thickness was determined by analyzing the x-ray reflectivity (XRR) results. The XRR data were fitted by software PANalytical X'Pert reflectivity.

Elemental composition of the films was obtained using x-ray photoelectron spectroscopy (XPS) (Kratos AXIS ULTRA DLD with Al-anode excitation) and Rutherford backscattering spectrometry (RBS). Gaussian–Laurentius functions and Shirley background were used to fit the experimental XPS data. RBS spectra were fitted by SIMNRA 7.02 code with an 1% statistic uncertainty to determine elemental compositions.

Scanning transmission electron microscopy (STEM) and selective area electron diffraction characterization were performed using the Linköping double Cs corrected FEI Titan³ 60–300 operated at 300 kV.

Spectroscopic ellipsometry was used to determine the absorption coefficient and, thus, to obtain their optical bandgap by using Cody plots (see supplementary material²⁸ for more description about the model).

Figure 1 shows the GIXRD results of the In_xGa_1−x N films grown by varying the number of GaN unit cycles (n) while keeping all other growth parameters unaltered. We observe only one peak between 20° and 40°, in which III-nitrides show their crystal planes with low Miller index such as $(101¯0)$⁠, (0002), and $(101¯1)$⁠. In addition to the dominant peak between 20° and 40°, another relative weak peak ∼60° is seen (see supplementary material²⁸ for the result in Fig. S1). Both GIXRD peaks shift toward lower 2θ values with decreasing n, revealing that the spacing between corresponding planes is enlarged and consequently the In-content (x) is enhanced accordingly. Unlike MOCVD and MBE, ALD is probably the most straight forward technique to tune x in a broad range (between x = 0.2 and 0.8) as evidenced by GIXRD results of In_xGa_1−x N films made with decreasing number of GaN cycles (n) in the growth process.

The observation of a single, dominant peak in Fig. 1 indicates that the studied In_xGa_1−x N films is rather homogeneous. Such homogeneity can diminish as evidenced by showing multiple and broader XRD peaks when small m and n were applied (see supplementary material²⁸ for the experimental result in Fig. S2). Considering that the <0002> is the natural preferred growth direction of wurtzite III-nitrides on substrates without epitaxial relationship¹⁸ and corroborated by our own observations,^8,19 we believe the XRD peaks arise from (0002) In_xGa_1−x N. Assuming strainfree In_xGa_1−x N films, the In-content can be tuned from x = 0.87 for n = 2 to x = 0.33 for n = 25 based on Vegard's law. However, accurate determination of the In-content by XRD peak position is very challenging as it is influenced by both In-content and strain condition. The strain originates not only from the lattice mismatch between the substrate and the film, but also the In/Ga ratio and its distribution within the In_xGa_1−x N films. Haider et al. reported substantial deviations of In-content obtained from different characterization techniques such as GIXRD, energy dispersive x-ray spectroscopy (EDX), and XPS.¹¹ To further investigate this issue, RBS and XPS are employed in our study. RBS is a powerful, nondestructive, and standardfree ion-beam-based technique for a fast quantitative analysis with high accuracy that is not sensitive to the strain in the target material. In addition, the high uncertainties in quantification by XPS, mainly coming from sputter-etching artifacts, are not applicable in RBS.²⁰ From our XPS analysis, the carbon level is not detectable, indicating that it is below 1 at. %, and the oxygen level is in the range of 2–7 at. %. The detected oxygen level is found to increase with the In content. We would like to emphasize that the studied InGaN films were very thin (<30 nm) and without any protective/anti-oxidation measures. Postoxidation and surface absorbents are inevitable and will disturb the accuracy of the results significantly. In addition, the N content of our InGaN films cannot be extracted accurately. First, the excitation source of the Al Kα line for our XPS facility renders spectral overlap between the auger peaks (LMM) of Ga and N1s. Second, the detection of light elements in RBS, such as B, C, N, and O, is hindered by the substrate with heavy element.²⁰ (see supplementary material²⁸ for the discussion of metal/nitrogen ratio extracted from RBS in Sec. III). As the main focus of this work is about In tunability and crystalline quality, we will not further discuss about impurities and also metal/nitrogen ratios of our InGaN films. Table I summarizes the In-content and thickness of In_xGa_1−x N thin films obtained by different techniques, together with their optical bandgaps. The In-content determined by all analytic techniques is similar within a range of a few atomic percent, indicating that the In_1−xGa_x N films are rather homogeneous. Constant overestimation of In-content by GIXRD is likely due to residual strain within the In_xGa_1−x N films and in line with previous observations.¹¹ 

According to the dimensionless mixing number (Mx) proposed to differentiate the interfacial sharpness related to the layer thickness in the ALD process,²¹ a distinguishable composition variation can only happen if the diffusion length of one metal in another metal nitride is smaller than the thickness of each bilayer.¹⁶ Although the diffusion length of In (Ga) in GaN (InN) under the growth environment is not known, the finite thickness of GaN in our case (ideally no thicker than 1.2 nm) should prevent the formation of separated phases.

We would like to point out that there is no sign of epitaxy between In_1−xGa_x N and Si. The crystal grains are randomly oriented with respect to the Si substrate. Despite the hexagonal crystal grains as majority, cubic inclusion is also found. (see supplementary material²⁸ for the TEM result of InGaN on Si in Fig. S3). To highlight the heteroepitaxy and the good homogeneity of the films, the SiC substrate is used for STEM study. Figure 2(a) shows a constant In/Ga ratio across entire In_xGa_1−x N (7U_InN + 5U_GaN) film from the energy dispersive x-ray elemental line profiles (STEM-EDX). Pronounced In segregation is not seen according to our STEM-EDX element mapping images. [Figs. 2(c)–2(e)]. In addition, the heteroepitaxy of the In_1−xGa_x N film is evidenced in the corresponding Fast Fourier Transform pattern [Fig. 2(b)]. The AB-type stacking of wurtzite In_xGa_1−x N and the ABAC-type stacking of the 4H–SiC substrate are well-aligned along $[011¯0]$ and [0001] directions. The a and c lattice constants derived from the FFT are ∼3.454 and ∼5.521 Å, respectively. This corresponds to an In content (x) of 0.72, in line with the GIXRD, but can also be affected by the strain in the material. From the lattice-resolved image, the density of structural disorder, such as stacking faults, at the interface is higher than for binary InN.¹⁴ The same constant In/Ga ratio and heteroepitaxy can also be seen from a Ga-rich In_xGa_1−x N (7U_InN + 25U_GaN) film (see supplementary material²⁸ for the TEM result in Fig. S4). However, in the Ga-rich films, there seems to be an In-rich regime close to the surface and the lattice stacking is then less coherent especially after a few layers on top of the Ga-rich In_xGa_1−x N film. The reason why Ga-rich In_xGa_1−x N film is worse than In-rich ones requires further investigation.

Despite homogeneous In-content (x), the predicted In-content is significantly lower than experimentally determined values. Moreover, all In_xGa_1−x N films are thinner (determined by XRR) compared to their predicted values as indicated in Table I. Both observations indicate that the actual $GPC∗InN$ and $GPC∗GaN$ for In_1−xGa_x N are smaller than those found for binary InN and GaN. According to the rule of mixture,¹⁸ the thickness (T) of an In_xGa_1−x N film grown via the supercycle ALD approach can be expressed by using

Considering $GPC∗InN$ and $GPC∗GaN$ as two variables, each In_xGa_1−x N film grown with different combinations of m, n, and k can be plotted as a line. Those lines shown in Fig. 3 be seen as mathematical binary linear simultaneous equations. Assuming $GPC∗InN$ and $GPC∗GaN$ are constant and independent on the number of m, n, and k, a single solution for $GPC∗InN$ and $GPC∗GaN$ is expected. However, such single solution is not available according to our experimental results. Instead, the closest approximation based on the intersections between all lines, the outcome of $GPC∗InN∼0.26Å/cycle$ and $GPC∗GaN∼0.40Å/cycle$ is obtained. According to our analysis, both $GPC∗InN$ and $GPC∗GaN$ are smaller than those estimated from binary cases, with a more pronounced reduction for $GPC∗InN$⁠. In general, the reduction of overall GPC is often ascribed to nucleation delay when switching from one binary ALD process to another, which has been reported for ternary oxides.^16,22,23

Nucleation delay is often associated with the density of reactive surface sites, exchange reaction, and persistent ligands.^16,24 Those impacts are expected to fade upon increasing ALD cycles,^24,25 and thus, the GPC* should increase accordingly until a steady state. This argument is further supported by seeing increasing thickness of In_xGa_1−x N films with increasing m and n (Fig. S4), where all lines possess identical expression of $420⋅GPC∗InN+300⋅GPC∗GaN$ If $GPC∗InN$ and $GPC∗GaN$ were independent of m and n, superposition lines would have been observed. According to our analysis (see supplementary material²⁸ for the analysis in Sec. 6), the increasing thickness of In_1−xGa_x N films with increasing m is mainly due to the enhancement of $GPC∗InN$ (increasing from 0.26 to 0.36 Å/cycle) rather than $GPC∗GaN$ (0.4 Å/cycle). From this perspective, the effect of nucleation delay is more pronounced for InN than GaN. A more severe nucleation delay of InN than of GaN was also found by using tris(1,3-diisopropyltriazenide)indium(III) as precursors for In_xGa_1−x N in our reactor.¹⁷ 

Since $GPC∗InN$ and $GPC∗GaN$ are significantly smaller than we anticipated at the first place, the prediction of In-content shown in Table I should be revised. Although the actual GPC^* is not known for accurate prediction, reduced $GPC∗InN$ than $GPC∗GaN$ will lead to even lower In-content prediction [Eq. (1)]. Considering that all our experimental results suggest a higher In-content than the estimated, a plausible explanation is that the “GaN” grown in the process is in fact In_xGa_1−x N. Unintentional In incorporation into the adjacent GaN layer is often seen in In_1−xGa_xN/GaN multiple quantum well structures grown by MOCVD.²⁶ We speculate that the interdiffusion of In into GaN consumes the In-adatoms on the surface of GaN, resulting in reduced $GPC∗InN$ and excessive In content seen in our experiment.

Spectroscopic ellipsometry is employed to investigate the optical properties of In_xGa_1−x N films with different x. The bandgap of the In_xGa_1−x N film can be determined by extrapolating the linear region of respective Cody-plot as shown in Fig. 4 (see supplementary material²⁸ for the plot of refractive index against wavelength and more information of fitting model in Sec. 7). The optical bandgaps of In_xGa_1−x N films obtained from our spectroscopic ellipsometry increase progressively from 1.99 to 2.43 eV with decreasing x as indicated in Table I. We would like to highlight that the bandgap estimated by our spectroscopic ellipsometry study is significantly higher than expected by considering the bandgap of InN as 0.7 eV. We believe it is due to their Fermi levels that are well above the conduction minimum, which is a known phenomenon for InN and In-rich InGaN.²⁷ 

In summary, In_xGa_1−x N films grown by ALD with a supercycle approach show a broad tunability in the In/Ga ratio by varying the number of GaN ALD cycles in the supercycle. The experimentally determined In-content is higher than the values predicted by the supercycle model. We ascribe this enhanced In incorporation to pronounced intermixing between InN and GaN bilayers. In addition, the actual growth of In_xGa_1−x N is not an ideal linear combination of GPC_InN and GPC_GaN found in the respective binary process. Notable reduction of GPC_InN and GPC_GaN is ascribed to nucleation delay when switching from one binary ALD process to another. No phase segregation is observed from an In_0.6Ga_0.4N film by TEM or XRD. Our results highlight the advantage of the ALD technique to prepare homogeneous, high In-content In_xGa_1−x N films, which could be used for extending the emission wavelength to red and near infrared in light-emitting-diodes.

This work was financially supported by the Swedish Foundation for Strategic Research through the project “Time-resolved low temperature CVD for III-nitrides” (No. SSF-RMA 15-0018) and by the Knut and Alice Wallenberg Foundation through the project “Bridging the THz gap” (No. KAW 2013.0049). The authors acknowledge the Knut and Alice Wallenberg Foundation for support of the electron microscopy laboratory in Linköping and for project funding (No. KAW 2020.0033). The Swedish Research Council and the Swedish Foundation for Strategic Research are acknowledged for access to ARTEMI, the Swedish National Infrastructure in Advanced Electron Microscopy (Grants Nos. 2021-00171 and RIF21-0026). Babak Bakhit gratefully acknowledges financial support from Swedish Research Council (VR), Grant No. 2019-00191 (for accelerator-based ion-technological centre in tandem accelerator laboratory in Uppsala University) and Grant No. 2021-00357.

The authors have no conflicts to disclose.

Chih-Wei Hsu: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Investigation (lead); Methodology (lead); Validation (lead); Visualization (lead); Writing – original draft (lead). Ivan Martinovic: Data curation (equal); Formal analysis (equal); Software (equal); Writing – review and editing (supporting). Roger Magnusson: Data curation (equal); Formal analysis (equal); Validation (supporting); Writing – original draft (supporting); Writing – review and editing (supporting). Babak Bakhit: Data curation (equal); Formal analysis (equal); Writing – review and editing (supporting). Justinas Palisaitis: Data curation (equal); Formal analysis (equal); Resources (supporting); Writing – review and editing (supporting). Per. O. Å. Persson: Conceptualization (supporting); Investigation (supporting); Resources (supporting); Supervision (supporting); Writing – review and editing (supporting). Polla Rouf: Data curation (equal); Formal analysis (equal); Writing – review and editing (supporting). Henrik Pedersen: Funding acquisition (lead); Project administration (lead); Resources (equal); Writing – original draft (supporting); Writing – review and editing (lead).

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