Comparison of aluminum nitride thin films prepared by magnetron sputter epitaxy in nitrogen and ammonia atmosphere

Wurtzite-type aluminum nitride (AlN) is used as the piezoelectric layer in surface acoustic wave (SAW)^1,2 and bulk acoustic wave (BAW) resonators due to its exceptional attributes, such as high intrinsic electromechanical coupling coefficient $k t 2$ (6.5%),³ longitudinal piezoelectric coefficient $d 33$ (6 pC/N),⁴ and thermal conductivity (2.85 W cm⁻¹ K⁻¹).⁵ In SAW resonators, increased roughness of the piezoelectric film can cause dispersion of surface acoustic waves,⁶ entailing AlN thin films with low surface roughness. AlN can also be used as a seed layer to promote oriented growth of molybdenum (Mo).^7–9 Mo due to its optimum mass density, resistivity, and acoustic impedance emerges as a compelling electrode for BAW resonators.^10,11 Resistivity of the electrode directly impacts the quality factor (Q) of the resonator.¹² A key factor that affects the resistivity of Mo is its grain size. Large grain size reduces scattering of electrons at the grain boundaries, thereby enhancing the conductivity of the film.^13,14 AlN thin films with large grain size can increase the grain size of Mo top contacts and hence its conductivity.

Apart from being used as a seed layer, AlN also finds application as a “compensating layer” to mitigate the thermal challenges imposed by aluminum scandium nitride (Al_1−xSc_xN), when used as an active layer in BAW resonators. Al_1−xSc_xN has gained a lot of interest in recent years due to its improved $k t 2$ and $d 33$⁠,^15,16 and therefore, is replacing AlN based acoustic resonators. Despite its attractive piezoelectric properties, thermal conductivity of Al_1−xSc_xN has been found to be significantly lower than AlN due to phonon-alloy scattering,^17,18 leading to self-heating issues, frequency shifts, heightened strain in thin films, and long term degradation in resonator performance.^19,20 Especially for BAW resonators which are operated at frequencies above 3 GHz, the thermal challenge becomes critical, as the dimensions of the resonator are expected to decrease with the increase in frequency.²¹ A possible way to reduce the adverse effects of AlScN on the resonator performance is to grow AlScN on AlN thin films.²² Therefore, AlN thin films exhibiting large grain size, low surface roughness, and high thermal conductivity are attractive for both SAW and BAW resonators.

Different growth techniques, such as molecular beam epitaxy (MBE),²³ metal organic vapor-phase epitaxy (MOVPE),²⁴ and magnetron sputter epitaxy (MSE),²⁵ are commonly used for AlN growth. MSE is advantageous over the other growth techniques due to its relatively low cost, high growth rates, and applicability to large substrates.²⁵ Florian et al.²⁶ demonstrated that a crystal quality similar to that obtained by MOVPE and MBE can be achieved through MSE by introducing ammonia (NH₃) during sputtering.²⁶ In addition to its structural and morphological properties, physical properties of AlN grown in NH₃ atmosphere are not explored.

In this work, we deposit AlN in both nitrogen (N₂) and in a mixture of nitrogen and ammonia (NH₃) atmosphere to compare some of their properties by characterizing them using atomic force microscopy (AFM), x-ray diffraction (XRD), time-of-flight secondary ion mass spectrometry (ToF-SIMS), spectroscopic ellipsometry (SE), scanning electron microscopy (SEM), transmission electron microscopy (TEM), and transient thermoreflectance (TTR). In addition, we propose a growth model explaining AlN growth in NH₃ atmosphere.

An Evatec clusterline 200II planar configured pulsed DC magnetron sputtering tool with a leak rate of (⁠ $≈$7.5  $×$ 10  $− 7 )$ mbar l/s was used for depositing AlN thin films. A 304 mm diameter aluminum (Al) target with a purity of 99.9995% was used to deposit both AlN films on 200 mm Si(111) substrates. Growth parameters used for depositing AlN in nitrogen [AlN(N₂)] and in a mixture of nitrogen and ammonia [AlN(N₂ + NH₃)] are listed in Table I. For more information about target treatment, substrate preparation, and the AlN(N₂) layer, readers are directed to our previous work.²⁷ 

The thermal conductivity of the samples was measured by the transient thermoreflectance (TTR) technique.^28,29 In the measurement, the samples were heated by 532 nm Nd:YAG pulsed laser (with a frequency of 50 Hz) through a gold (Au) transducer having a thickness of 200 ± 5 nm. The transient decay of the sample surface temperature after pump laser pulses is monitored by measuring the reflectance of 488 nm Ar-ion probing laser and ultra-fast detector. The heating laser beam size is approximately 200  $μ$m which is more than one order of magnitude smaller for the probing laser, ensuring that only thermal transport along the surface normal direction is probed. The thermal conductivity is obtained by the least-square fit of the measured reflectance transient by the one-dimensional heat transport equation. In the fit, the thermal conductivity of Au and the specific heat of Si, Au, and AlN are used as input parameters and are taken from the literature.^30–33 

XRD $2 θ / θ$ scans of AlN(N₂ + NH₃) sputtered on Si(111), Al, AlN(N₂)/Al/Si(111), and c-Al₂O₃(0001) are shown in Fig. 1. Al layer was deposited for 5 s using the parameters listed in Table I. The scans from all the samples showed reflection from AlN 000l (l = 2, 4, 6) indicating that AlN grows with its (0001) plane-parallel to the surface of the substrate. Interestingly, an additional reflection from AlN (⁠ $1 ¯ 101$⁠) plane was observed only when AlN(N₂ + NH₃) was sputtered on Si(111). Therefore, a possible reason for mixed orientation is reaction of species present in the N₂ + NH₃ plasma with the Si substrate. Some of the species present in the NH₃ plasma are N₂, $N 2 +$⁠, H, NH, and N.^34,35 Since H is expected to desorb at temperature greater than 500 °C (Refs. 36–38) and mixed orientation was not observed when AlN was sputtered in pure nitrogen plasma,²⁷ it is most likely that mixed orientation is due to reaction of NH species with the Si substrate. Although mixed orientation was suppressed in both AlN(N₂ + NH₃)/Al/Si(111) and AlN(N₂ + NH₃)/AlN(N₂)/Al/Si(111) stacks, 2D growth of AlN was observed only in a three-step process, i.e., AlN(N₂ + NH₃)/AlN(N₂)/Al/Si(111) stack. An AFM image of AlN(NH₃)/Al/Si(111) indicating 3D growth is shown in Fig. S1 in the supplementary material. Therefore, a three-step process was used for further investigation.

In order to determine the minimum thickness required to achieve 2D growth, stacks of AlN(N₂ + NH₃)/AlN (N₂)/Al/Si(111) was grown with constant Al deposition time of 5 s and AlN(N₂) thickness of $≈$30 nm, while growing AlN(N₂ + NH₃) at three different thicknesses: 60, 90, and 120 nm. The corresponding AFM micrographs are shown in Figs. 2(a)–2(c), respectively. From the images, it can be seen that a minimum thickness of 90 nm is required for the AlN(N₂ + NH₃) layer to achieve 2D growth. In addition, the results indicate that there is a gradual increase in grain size prior to a transition in the growth mode, as has similarly been observed in a previous research.²⁶  Figure 2(d) shows a schematic of the change in the growth mode from 3D to 2D with increasing AlN(N₂ + NH₃) layer thickness. Our TEM analysis shows a flat interface between AlN(N₂ + NH₃) and AlScN (here, AlScN is used as a differential layer), indicating 2D growth. In addition, the analysis also shows a presence of a rough interface between the AlN(N₂) and AlN(N₂ + NH₃) layer (see Fig. S4 in the supplementary material).

To understand the reason why ammonia facilitates 2D growth, three samples were investigated. Stack details and the growth temperature of the samples are listed in Table II.

AFM images of the samples listed in Table II are shown in Figs. 3(a)–3(c). Except for sample 3, none of the samples exhibited 2D growth, indicating that both high temperatures and NH₃ are essential to achieve it. The diffusion length of ad-atoms is given by λ =  $D τ$⁠, and the surface diffusion coefficient is given by D =  $D 0 e ( − E / k B T )$⁠.³⁹ Here, D, τ, E, T, and k_B are the surface diffusion coefficient, the mean residence time of ad-atom at the surface, the energy barrier for surface diffusion coefficient, the growth temperature, and Boltzmann's constant, respectively. According to the definition, D increases exponentially with increasing T and decreasing E.

Ab initio calculations predict improved mobility of gallium (Ga) ad-atoms on GaN when the presence of excess N on the surface is limited, and efficient incorporation of N ad-atoms when there are excess Ga ad-atoms on the surface (Ga rich surface).⁴⁰ Kieu et al. have shown that, on a Ga-rich surface, NH₃ gets adsorbed either on a Ga ad-atom or on a Ga atom, decomposes into an NH unit, which then incorporates into the growing film, and diffuses on the GaN surface,^41,42 providing an additional pathway for diffusion of N ad-atoms. Similar to these observations, introduction of NH₃ probably reduces excess N and creates an Al rich surface, improving the mobility of both Al and N ad-atoms, as also suggested by Dadgar et al.⁴³ In addition, high temperatures increase ad-atom mobility; therefore, a combined contribution of these effects facilitates 2D growth.

An Al-rich surface can be expected in Al-rich growth conditions, and since AlN is expected to be Al polar in Al rich conditions,^44,45 sample 3 is expected to be Al polar. In order to determine the polarity, samples 1 and 3 were etched in phosphoric acid (H₃PO₄) for 7 min at 55 °C, and the resulting surface morphology is shown in Figs. 3(d) and 3(e), respectively. The images clearly demonstrate that the etched area in sample 1 is larger than that of sample 3, indicating that sample 1 is predominantly N-polar while sample 3 is predominantly Al-polar,⁴⁶ suggesting AlN growth in Al rich condition. Presence of an Al interlayer could have also promoted AlN to grow in Al-polar direction,⁴⁷ but when a stack of Si/Al/AlN was etched in similar conditions, the results indicated that AlN is mixed polar (see Fig. S2 in the supplementary material). An explanation to this outcome is diffusion of Al into Si.^26,48 The polarity analysis reinforces the proposed growth model, depicted in Fig. 4.

The average grain size $d grain$ of sample 1 and 3 was determined using a procedure similar to a Williamson–Hall (W–H) analysis.⁴⁹ The outcome of the grain size calculation revealed that with introduction of ammonia, the average grain size increased from 14 to 25 nm, as a result of change in growth mode from 3D to 2D [see Figs. 3(a) and 3(c)]. XRD rocking curves were recorded from 0002 and $1 ¯$102 reflections in order to assess the crystalline quality of sample 1 and 3. The ω-FWHM values extracted from AlN 0002 reflection increased from 0.85° to 1.18°, while the ω-FWHM value of AlN $1 ¯$102 reflection increased from 1.36° to 1.6° when ammonia was introduced, indicating that crystalline quality slightly worsens with NH₃ assisted AlN sputtering [see Figs. S3(b) and S3(c) in the supplementary material]. Oxygen and carbon concentrations in samples 1 and 3 determined using ToF-SIMS measurements are shown in Figs. 5(a) and 5(b), respectively. In sample 1, both carbon and oxygen concentrations remained constant, whereas in sample 3, they increased after the introduction of ammonia. The increase in the oxygen impurities in sample 3 could be the reason for the worsening in crystal quality⁵⁰ and has been previously observed⁵¹ that the purity of the NH₃ gas used is most likely the reason for this outcome.

The thermal conductivity of three samples having different grain sizes $d grain$ (7, 14, and 25 nm) was measured at two different temperatures and shown in Fig. 6. Here, the samples with grain sizes 14 and 25 nm correspond to samples 1 and 3, respectively, while the sample with a grain size of 7 nm corresponds to AlN grown at 500 °C in nitrogen atmosphere. This layer exceptionally exhibits a fiber texture with crystal grains rotating around the c-axis, causing a phonon mismatch in the lateral direction between the grains. However, this effect can be disregarded since we only probe the thermal transport along the surface normal. Among the three samples, the one with the largest grain size (⁠ $d grain$ = 25 nm) grown in N₂ + NH₃ atmosphere with the 2D growth mode has the highest thermal conductivity, 13.6 W/m K at 300 K. The other two samples grown in N₂ atmosphere with the 3D growth mode have smaller grain sizes and their thermal conductivity becomes significantly smaller. Such an increase in thermal conductivity with increasing grain size has been reported for different nanocrystalline materials.^52–54 AlN is a semiconductor, notable for its high thermal conductivity ranging from 270 to 374 W/m K.²⁹ The low thermal conductivity observed in the investigated AlN layers is attributed to the small layer thickness of 140 nm, which intensifies the layer boundary effect.⁵⁵ Moreover, the presence of grain boundary contributes to further reduction in the thermal conductivity of these layers. It is also worth noting that sample 1 contains about twice higher oxygen concentration compared with sample 1. Nevertheless, we anticipate the negligible impact of the oxygen on the measured thermal conductivity as seen Fig. 6. In all three samples, the thermal conductivity of the layers measured at 77 K is found to be consistently 2–3 times lower compared to that at 300 K.

The phonon scattering rate at layer boundary and grain boundary can be expressed in a similar way by $τ B − 1 = v s / a B L$⁠, where L is the layer thickness, and $τ grain − 1 = v s / a grain d grain$⁠.¹⁸ The prefactors a_B and a_grain are used to reflect the ratio between the phonon mean free path (MFP) and the layer thickness/grain size for the two separate cases. The value of a_grain is inherently relative and depends on how the grain size is determined. For instance, the grain size of sample 1 determined by AFM was found to be $≈$46 nm, which is 70% smaller when it is measured by XRD. Therefore, for simplicity, we assumed $a grain = a B = a$ and set it as an adjustable parameter in a least-square minimization fit of the measured data with calculations using Eq. (1). We have found a = 2.5 at 300 K and a = 1.5 at 77 K. The value of a_B obtained at 300 K is comparable with the one obtained for epitaxial GaN layers.⁵⁸ The smaller value of a obtained at 77 K reflects a significantly stronger boundary scattering effect compared with that at room temperature. The thermal conductivity dependence on the grain size calculated by Eq. (1) is shown in Fig. 6.

The lower thermal conductivity measured for the layers having smaller grain size is related to the increased phonon-grain-boundary scattering rate $τ grain − 1$⁠. Note that all layers investigated have comparable thicknesses which guarantee that the $τ B − 1$ remained unchanged. The significantly lower thermal conductivity obtained at 77 K compared with that at 300 K can be explained by (i) the dominance of the grain size effect in the thermal transport and (ii) the less contribution of high-frequency phonons (i.e., the low MFP) as they are less activated once the temperature drops. The effect of the point defect (impurities) on the thermal conductivity is also considered (Fig. 6). For the impurity concentrations measured in sample 3, the thermal conductivity is almost unaffected, which can be attributed to the dominance of the grain-size effect in thermal transport. When the grain size approaches the layer thickness (i.e., L = 120 nm), the thermal conductivity drops by only 4.4% at 300 K and 2.1% at 77 K. Both the measurements and calculations indicate that there is ample room for achieving higher thermal conductivity of the layers by further improving their crystal quality.

In summary, this research provides valuable insight into the growth behavior, grain size, and thermal conductivity of AlN thin films sputtered in N₂ and N₂ + NH₃ atmospheres. Introduction of NH₃ in process gas facilitates 2D growth, increasing the grain size and improving the roughness of the AlN films. The thermal conductivity measured at different temperatures demonstrate that the larger grain size contribute to higher thermal conductivity. Impurity studies reveal that introduction of NH₃ leads to significant O and C contamination, which does not impact the thermal conductivity significantly but worsens the crystal quality. Table III compares and quantifies the properties discussed before for AlN sputtered in N₂ with N₂ + NH₃ atmospheres studied in this work. Although there is a slight degradation in crystal quality, the significant increase in grain size and thermal conductivity, and reduction in roughness attainable by NH₃ assisted AlN growth, makes AlN sputtered in NH₃ a superior choice for BAW and SAW resonators.

See the supplementary material for AFM image of AlN (NH3)/Al/Si(111) stack (Fig. S1); polarity analysis of AlN/(Si(111) and AlN/Al/(Si(111) (Fig. S2); XRD analysis (2θ/θ scans, rocking curves) of samples 1 and 3 (Fig. S3); and cross-sectional TEM image of AlScN/AlN(N2 + NH3)/AlN(N2)/Si(111) stack (Fig. S4).

This project was performed within the framework of COMET—Competence Centers for Excellent Technologies and ASSIC Austrian Smart Systems Integration Research Center, which is funded by BMVIT, BMDW, and the Austrian provinces of Carinthia and Styria. The COMET programme is run by FFG. The authors would also like to thank Dr. Maximilian Kessel, Dr. Jürgen Weippert, and Silas Pokorny for their help in integrating ammonia gas into the sputter module. The work at Linköping University is performed within the Competence centers for III-Nitride Technology (C3NiT-Janzén) supported by the Swedish Governmental Agency for innovation systems (VINOVA) under the Competence Center Program Grant No. 2022-03139. We further acknowledge the support from the Swedish Research Council VR under Grant No. 2023-04993.

The authors have no conflicts to disclose.

Balasubramanian Sundarapandian: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Dat Q. Tran: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Lutz Kirste: Data curation (equal); Formal analysis (equal); Writing – review & editing (supporting). Patrik Straňák: Data curation (equal); Formal analysis (equal); Writing – review & editing (supporting). Andreas Graff: Data curation (equal); Formal analysis (equal). Mario Prescher: Data curation (supporting); Formal analysis (supporting). Akash Nair: Writing – review & editing (supporting). Mohit Raghuwanshi: Supervision (supporting); Writing – review & editing (supporting). Vanya Darakchieva: Funding acquisition (equal); Project administration (equal); Resources (equal); Supervision (supporting); Writing – review & editing (supporting). Plamen P. Paskov: Formal analysis (supporting); Funding acquisition (equal); Project administration (equal); Resources (equal); Writing – review & editing (supporting). Oliver Ambacher: Conceptualization (supporting); Formal analysis (supporting); Funding acquisition (lead); Supervision (lead); Writing – review & editing (supporting).

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