Crystallinity degradation and defect development in (Al_xGa_1−x)₂O₃ thin films with increased Al composition

Gallium oxide (Ga₂O₃) has great potential for power electronics due to its ultrawide bandgap of 4.5–5.3 eV.¹ Among the six polymorphs of Ga₂O₃ (⁠ $α$⁠, $β$⁠, $γ$⁠, $δ$⁠, $ε$⁠, and $κ$⁠), monoclinic $β$-phase is thermodynamically the most stable, and bulk single crystals can be grown by conventional melt growth techniques.^2,3 The availability of large-size substrates has driven diverse studies on $β$-Ga₂O₃ for purposes related to thin film growth,¹ device processing,⁴ and power device applications represented by Schottky barrier diodes^5,6 and field-effect transistors (FETs).^7,8

In addition to power device applications, $β$-Ga₂O₃ can also be applied to wireless communication devices in harsh environments by taking advantage of the chemical stability and radiation hardness of its metal-oxide-semiconductor FETs (MOSFETs).⁹ Thus far, promising radio-frequency (RF) characteristics, such as a current-gain cutoff frequency of 10 GHz and a maximum oscillation frequency of 27 GHz, have been obtained for lateral $β$-Ga₂O₃ MOSFETs with gate lengths of 80 and 200 nm, respectively.¹⁰ For an even higher RF performance, the challenge is how to suppress the short-channel effect (SCE). It occurs when the gate length is scaled down to less than 200–300 nm, and the representative phenomenon is a reduction of the threshold voltage. Solutions include thinning the channel layer and inserting a back-barrier layer. Here, (Al $x$Ga $1 − x$⁠)₂O₃ is suitable for the back-barrier layer of Ga₂O₃ MOSFETs because (⁠ $i$⁠) it can be easily grown on a $β$-Ga₂O₃ substrate, after which a $β$-Ga₂O₃ layer can be grown on it,¹¹ and (⁠ $ii$⁠) the bandgap of (Al $x$Ga $1 − x$⁠)₂O₃ is wider than that of Ga₂O₃.¹² In fact, inserting the (Al $0.13$Ga $0.87$⁠)₂O₃ back-barrier has successfully shifted the threshold voltage by +8 V, thus indicating its effectiveness in suppressing the SCE.¹³ 

When forming an epitaxial structure of Ga₂O₃ MOSFETs, an (Al $x$Ga $1 − x$⁠)₂O₃ layer is primarily grown on a $β$-Ga₂O₃ substrate, and an unintentionally doped (UID) $β$-Ga₂O₃ layer is grown thereon. In this upper layer, a channel will be formed in a later device process. It is, therefore, important to check the quality of the underlying (Al $x$Ga $1 − x$⁠)₂O₃ layer because it will influence the final device characteristics. Specifically, the following points must be carefully considered to grow the (Al $x$Ga $1 − x$⁠)₂O₃ layer. First, it is, of course, preferable to have good crystallinity and a flat surface because a Ga₂O₃ channel is formed on it. Second, a larger Al composition $x$ is better to widen the bandgap;¹² however, attention must be paid to the Al solubility limit and the resultant decrease in crystallinity. Third, the (Al $x$Ga $1 − x$⁠)₂O₃ layer should be thick enough, but if the thickness exceeds the critical value and lattice relaxation occurs, the emergent dislocations could affect the device characteristics as a leakage current source. Here, we provide an additional explanation of why the thicker layer is required. In the Ga₂O₃ device processing, Si ion implantation doping can be utilized to form a channel and Ohmic contact regions in the UID Ga₂O₃ layer.¹⁴ While this technique improves the degree of freedom in device design, it also has downsides. For one, the implantation generates defects through which Fe diffusion from the Fe-doped Ga₂O₃ substrate to the channel layer is enhanced during the activation annealing after implantation.¹⁵ Moreover, the Gaussian-like depth profile of the implanted Si ions inevitably has a tail region below the channel layer. Thus, the (Al $x$Ga $1 − x$⁠)₂O₃ layer should be thick enough to counteract the Fe outdiffusion and to accommodate the implantation tail region. Note here that a high-quality (Al $x$Ga $1 − x$⁠)₂O₃ layer must be grown while maintaining the balance between the Al composition and the thickness, as there is also the matter of critical thickness.

In this work, to obtain high-quality (Al $x$Ga $1 − x$⁠)₂O₃ thin films with an adequate thickness ensuring sufficient back-barrier effects, we evaluated the crystal structure and surface morphology of the (Al $x$Ga $1 − x$⁠)₂O₃ thin films when the Al composition $x$ was systematically changed. Results showed that the crystallinity began to degrade and defects appeared on the surface when the Al composition $x$ exceeded about 0.16. As $x$ increased, the defects developed mainly along the $[201]$ direction and slightly along the $[001]$ direction. These experimental findings are briefly discussed alongside a comparison with the critical thickness.

(Al $x$Ga $1 − x$⁠)₂O₃ thin films were grown on Fe-doped semi-insulating $β$-Ga₂O₃ (010) substrates by plasma-assisted molecular beam epitaxy (MBE). Ga and Al were evaporated from effusion cells. The beam equivalent pressure of Ga (BEP $Ga$⁠) was fixed at $1.9× 10 − 7$ Torr, while that of Al (BEP $Al$⁠) was varied from $5.3× 10 − 9$ to $2.4× 10 − 8$ Torr to change the Al composition $x$⁠. Oxygen plasma was generated by an RF plasma cell. The oxygen flow rate and the plasma power were set to 2.0 sccm and 250 W, respectively. The growth temperature of 630  $°$C was monitored by a pyrometer. The temperature was used as optimized for the growth of unintentionally doped (UID) $β$-Ga₂O₃ thin films on $β$-Ga₂O₃ (010) substrates.¹⁶ The UID $β$-Ga₂O₃ thin films grown below 500  $°$C and above 700  $°$C had rough surfaces due to three-dimensional growth and step bunching, respectively. In contrast, the films grown at 550–650  $°$C had atomically smooth surfaces. In addition, the growth rate decreased above 700  $°$C because of the decomposition of Ga₂O₃ to volatile suboxides.¹⁷ For the above reasons, the growth temperature of 630  $°$C was determined. The film thickness was estimated to be about 100 nm by a surface profiler, and the growth rate was determined to be 0.33  $μ$m/h. The crystal structures of the samples were characterized by x-ray diffraction (XRD) using Cu K $α$ radiation. The surface morphology was observed by reflection high-energy electron diffraction (RHEED) and atomic force microscopy (AFM). The RHEED patterns were recorded at 300  $°$C during the cooling process after the growth, while the AFM observation was carried out at room temperature in air.

Figure 1(a) shows $θ$-2 $θ$ XRD patterns of the (Al $x$Ga $1 − x$⁠)₂O₃/ $β$-Ga₂O₃ (010) samples. The (020) diffraction peaks of the $β$-(Al $x$Ga $1 − x$⁠)₂O₃ thin films monotonically shift to the higher angle side as the Al composition $x$ increases. Here, $x$ was estimated from the (020) diffraction peak separation between the substrate and the coherently grown (shown later) thin film in the $θ$-2 $θ$ pattern.^18, Figure 1(b) shows the relationship between $x$ and the Al BEP ratio. Since $x$ is linearly in proportion to the Al BEP ratio, Al atoms were efficiently incorporated into the thin films. In monoclinic $β$-Ga₂O₃, Ga ions occupy the octahedral and tetrahedral sites. The results of a theoretical calculation to assess the phase stability of (Al $x$Ga $1 − x$⁠)₂O₃ alloy suggested that the octahedral site was preferentially occupied by Al up to $x=0.5$⁠.¹² Ionic radii of sixfold coordinated Ga $3 +$ and Al $3 +$ are 0.62 and 0.535 Å, respectively.¹⁹ This means that the lattice constant becomes smaller with increasing $x$⁠. Further, the thin film is subjected to in-plane tensile strain on the substrate, leading to a contraction along the out-of-plane direction. Therefore, it is reasonable that the diffraction peaks of the $β$-(Al $x$Ga $1 − x$⁠)₂O₃ thin films monotonically shift to the higher angle side with increasing $x$⁠. Clear thickness fringes observed in the $θ$-2 $θ$ XRD patterns indicate the formation of sharp interfaces, from which the film thicknesses were estimated to be 100–110 nm. These values are in good agreement with the ones expected from the growth rate. In Fig. 1(c), for the $β$-(Al $x$Ga $1 − x$⁠)₂O₃ thin films, the (020) diffraction peak intensities and the full widths at half maximum (FWHMs) of the (020) rocking curve peaks are plotted as a function of $x$⁠. Above $x≈0.16$⁠, the peak intensity decreases and the rocking curve is broadened, indicating the lowering of the crystallinity due to approaching the solubility limit.

According to a previous work on the growth of (Al $x$Ga $1 − x$⁠)₂O₃ thin films, $γ$-phase may be stabilized with relatively high Al compositions.²⁰ However, all the films in this study were grown without phase segregation.²¹ The solubility limit of Al in $β$-Ga₂O₃ increases for higher growth temperature, however, it appears around $x=0.20$ in the case of the MBE growth.²² Since MBE operates in ultrahigh vacuum, the growth temperature is limited to $∼700 °$C to reduce the decomposition of Ga₂O₃ into volatile suboxides.¹⁷ As a result, the value of $x$ is limited. On the other hand, since metal organic chemical vapor deposition (MOCVD) can operate at low to medium vacuum, the growth temperature can be increased to higher temperatures, leading to the higher solubility of Al. For instance, MOCVD growth at 880  $°$C has demonstrated that $β$-(Al $x$Ga $1 − x$⁠)₂O₃ thin films were grown on $β$-Ga₂O₃ (010) substrates up to $x=0.40$⁠.²⁰ 

Figures 1(d)–1(h) show reciprocal space mappings (RSMs) of the (Al $x$Ga $1 − x$⁠)₂O₃/ $β$-Ga₂O₃ (010) samples measured around the asymmetric (420) diffraction. The peak of the (Al $x$Ga $1 − x$⁠)₂O₃ thin film is located at the same $q x$ as that of the substrate, which indicates that these films were coherently grown on the substrates. In the $x=0.19$ sample, as the peak of the (Al $x$Ga $1 − x$⁠)₂O₃ thin film is broadened in the $q x$ direction, the crystallinity is certainly lowered, as shown in Fig. 1(c). In addition, the peak of the $β$-Ga₂O₃ substrate is also broadened in the $q x$ direction for the $x=0.19$ sample, while the peaks are sharp for the $x≤0.16$ samples. This implies a degradation of the substrate crystallinity (discussed later).

Next, we examine the surface morphologies of the as-grown (Al $x$Ga $1 − x$⁠)₂O₃ thin films. Figures 2(a)–2(e) show RHEED patterns for $x=0.05$– $0.19$⁠. The sharp streaky patterns indicate atomically smooth surfaces. However, a closer look at the surface using AFM revealed an interesting difference. Figures 2(f)–2(j) show surface AFM images of the series of (Al $x$Ga $1 − x$⁠)₂O₃ thin films. For $x=0.05$⁠, $0.11$⁠, and $0.12$⁠, the surfaces are flat, and the root mean square roughness (R $q$⁠) is comparable to that of the as-supplied $β$-Ga₂O₃ (010) substrate (data not shown). For $x=0.16$⁠, although the R $q$ value is similar to those for $x=0.05$– $0.12$⁠, small defects are sparsely observable, as indicated by the white arrows in Fig. 2(i), which are aligned along the $[201]$ direction. When $x$ is further increased to $0.19$⁠, more and larger defects appear across the surface [Fig. 2(j)]. According to the AFM images, the defects developed mainly along the $[201]$ direction and slightly along the $[001]$ direction as $x$ increases and approaches the solubility limit.

Here, we discuss the experimental results for the case of $x=0.19$⁠. It is seemingly inconsistent that the RHEED pattern was streaky [Fig. 2(e)] while the AFM image showed a defective surface [Fig. 2(j)]. However, since the RHEED patterns and the AFM images were observed at 300  $°$C and room temperature, respectively, it is likely that the defects were generated during the cooling process after the growth due to the difference in thermal expansion coefficients between the film and substrate, which is a well-known phenomenon. For instance, the defect densities were increased during the cooling process after the growth from $∼ 10 4$ to $∼ 10 7$ cm $− 2$ in the GaAs/Si and GaP/Si systems.²³ Moreover, the AFM images indicate that the defects developed in specific crystal orientations [Figs. 2(i) and 2(j)]. Regarding thin film defects, it is known that defects on substrates are one of the origins. When defects, which are sources of dislocations, are present on substrates, the dislocations are inherited by epitaxial thin films grown thereon. For $β$-Ga₂O₃ substrates, defects along various directions are known to occur,²⁴ and defects along the $[201]$ and $[001]$ directions have also been reported.²⁵ Therefore, the defects observed on the (Al $x$Ga $1 − x$⁠)₂O₃ surfaces in the current study might have been partly triggered by defects that were originally present on the substrates. Note here that not all the defects on the film surface originated from the pre-existing defects in the substrate because the defect density observed in the AFM image was $∼ 10 9$ cm $− 2$⁠, while that in the substrate is $∼ 10 4$ cm $− 2$⁠.²⁶ Returning to the RSM [Fig. 1(h)], not only the thin film peak but also the substrate peak was broadened in the $q x$ direction. This implies a degradation in the substrate crystallinity. As the RSM indicates that the thin film is not relaxed, we presume that the strain energy accommodated in the (Al $x$Ga $1 − x$⁠)₂O₃ thin film is locally released by the generation of the defects, which helps the defect-free regions to remain coherently grown. In this case, the sample with $x=0.19$ exceeds the critical thickness even though the lattice relaxation was not detected by the RSM. Regarding the broadening of the substrate peak in the RSM, one possible scenario is that the defects generated in the (Al $x$Ga $1 − x$⁠)₂O₃ thin film propagated back to the Ga₂O₃ substrate, resulting in the degradation of the crystallinity. In terms of this assumption, a similar situation has been reported in tensile-strained SiGe thin films grown on Ge-on-Si (111).^27,28

The parameters necessary to calculate $h c$ are now all present. Figure 3 shows the $h c$ curves as a function of $x$ and compares them with the experimental results for the (Al $x$Ga $1 − x$⁠)₂O₃ thin films. The boundary, where the crystallinity lowering was confirmed by XRD and the appearance of defects was observed by AFM (⁠ $x∼0.16$⁠), is close to the curve obtained using the M–B model for the slip system of $⟨201⟩{10 2 ¯}$⁠. Regarding the deviation from the calculated $h c$⁠, it is generally known that the $h c$ value is overestimated by the P–B model and is larger than the estimation by the M–B model when the lattice mismatch is below 1 $%$⁠, as in this study.³⁵ In designing the $β$-Ga₂O₃/(Al $x$Ga $1 − x$⁠)₂O₃ epitaxial structure for device prototyping, it is, thus, useful to adopt the Al composition and the (Al $x$Ga $1 − x$⁠)₂O₃ layer thickness estimated by the M–B model in anticipation of a small safety margin.

In summary, we aimed to improve the RF characteristics of lateral Ga₂O₃ MOSFETs by inserting the (Al $x$Ga $1 − x$⁠)₂O₃ back barrier. As a preliminary step, the crystal structure and surface morphology of 100-nm-thick (Al $x$Ga $1 − x$⁠)₂O₃ thin films were evaluated when the Al composition $x$ was changed. For $x≥0.16$⁠, the crystallinity was lowered and the defects appeared across the surface. The defects developed mainly along the $[201]$ direction and slightly along the $[001]$ direction as $x$ increased. The boundary where the crystallinity and the surface morphology changed in the experiments was close to the critical thickness curve calculated using the M–B model assuming the slip system of $⟨201⟩{10 2 ¯}$⁠. This finding will be useful to design $β$-Ga₂O₃/(Al $x$Ga $1 − x$⁠)₂O₃ epitaxial structures for future device prototyping.

This work was supported in part by the Development Program, “R&D of ICT Priority Technology” of the Ministry of Internal Affairs and Communications, Japan (JPMI00316).

The authors have no conflicts to disclose.

Takumi Ohtsuki: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Validation (equal); Writing – original draft (lead); Writing – review & editing (equal). Masataka Higashiwaki: Conceptualization (equal); Funding acquisition (lead); Project administration (lead); Resources (lead); Supervision (lead); Validation (equal); Writing – review & editing (equal).

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