Role of dislocations on Sn diffusion during low temperature annealing of GeSn layers

The growth of high Sn content GeSn alloys has become a subject of intense study in recent years.¹ From the point of view of optical properties, GeSn features a tunable direct bandgap for Sn concentrations above 6.5–9 at. % as well as high optical absorption.^2–5 The successful growth of GeSn has prompted the fabrication of GeSn photodetectors^6,7 as well as optically and electrically pumped lasers.^8–11 In addition, recent investigations has shown and improved carrier mobility of GeSn alloys compared to Ge, which is essential for the next generation metal-oxide-semiconductor field-effect transistors.^12,13 As a result, the development of the GeSn semiconductor presents an exciting opportunity for new and novel group IV optoelectronic devices that can be integrated on a Si platform using complementary metal-oxide-semiconductor (CMOS) process flow.

While creating exiting opportunities, GeSn alloys are metastable materials due to less than 1 at. % solid solubility of Sn in Ge at equilibrium at high temperatures.¹⁴ This makes Sn-rich GeSn alloys susceptible to Sn precipitations, segregation, and recrystallization during growth or thermal processing.^15–20 For example, Moto et al.¹² have shown that for annealing temperatures of about 475 °C, the excess Sn was precipitated as a thin film on the surface of polycrystalline GeSn. In addition to segregation, substantial compressive strain develops in epitaxial GeSn layers containing sufficient Sn grown on a Ge/Si(001) substrate. For example, there is an approximate 15% lattice mismatch between pure α-Sn and Ge. As a result, in relatively thick GeSn layers (on the order of 100 nm), strain relaxation produces (1) a network of misfit dislocations (MDs) at the layer/substrate interface and (2) threading dislocations (TDs) in the {111} slip planes propagating toward the surface.²¹ The presence of dislocations is currently thought to enhance material decomposition effects, such as, Sn out-diffusion along the dislocation lines or Sn segregation at dislocations^22,23 or both. In that case, Sn can be expected to be observed locally near dislocation lines. However, the evidence from XRD and PL in this study is that Sn is lost from throughout the sample and not just locally at threading dislocations. Consequently, the contradiction between the expectation of local Sn segregation on the surface and the observed bulk segregation is an indication that the mechanism of large area GeSn alloy decomposition through dislocations is not well understood.

In this research, a series of GeSn grown samples were thermally annealed for different durations of time at a temperature of 300 °C to examine the mechanism of material degradation due to Sn segregation. We investigate the role of dislocations on Sn out-diffusion especially at large distances from dislocation lines.

The GeSn layers with 9.5 at.  % Sn and 300 nm thickness were grown at 300 °C by chemical vapor deposition (CVD) compressively strained on the Ge-buffered Si(001) substrate using an ASM Epsilon reduced pressure chemical vapor deposition reactor. Two samples were annealed in sealed quartz tubes, one for two, and one eight hours at 300 °C to observe Sn segregation at the surface. The temperature of 300 °C was selected because higher temperatures resulted in the total loss of Sn.¹⁸ The surface morphology of our samples was then studied by atomic force microscopy (AFM) using a Bruker 3000 dimension III in a tapping mode equipped with ultrasharp silicon tips. The x-ray diffraction (XRD) measurements were performed using a Philips X’pert MRD system equipped with a standard four-bounce Ge(220) monochromator, 1.6 kW Cu Kα1 x-ray tube, and a Pixel detector. The 10 K photoluminescence (PL) spectroscopy measurements were performed using a pulsed laser emitting at 1064 nm. Raman measurements were performed in an unpolarized backscattering geometry using a 632.8 nm He-Ne laser. The Raman shifted laser light was collected by a spectrometer (Horiba Jobin-Yvon LabRam HR) equipped with a thermoelectrically cooled Si charge-coupled device (CCD) detector.

The AFM measurements were performed to examine the surface evolution of the GeSn layers annealed for different annealing time durations. For the as-grown sample, the AFM image [Fig. 1(a)] is characterized with a smooth surface but with a dense crosshatch pattern. The pattern is typical for epitaxial layers with misfit dislocations (MDs) running along $⟨ 110 ⟩$ crystallographic directions at the layer/substrate interface.²⁴ Additional round shaped microparticles with the density of 3.5 × 10⁸, 1.8 × 10⁸, and 1.4 × 10⁸ cm⁻² can be seen on the surface of samples annealed for two, four, and eight hours, respectively. Moreover, the particle size increased substantially from 137 ± 23, to 236 ± 36, to 314 ± 72 nm [Fig. 1(e)] with increasing annealing time. Together, the results indicate the total Sn on the surface remains about the same from 2h to 4h to 8h, suggesting a hypothesis that (1) Sn migration occurs early during annealing, possibly by threading dislocations to the surface and by diffusion to the surface and (2) Sn coalescence on the surface occurs as Sn appears on the surface.

The hypothesis is also consistent with the outcome of a comparison between the 10 K PL spectra for GeSn layers measured before and after annealing (Fig. 3). For example, two pronounced peaks (E₁ and E₂) can be seen on each spectrum, whose origin, however, is not as straightforward for a quasi-direct bandgap GeSn alloy. In particular, the magnitude of the separation in energy between the L and Г valley minimums (ΔE_Г−L) for a GeSn alloy is very close to the splitting of the light hole (LH) and heavy hole (HH) bands (ΔE_HH−LH) caused by the expected biaxial strain.³³ As a result, the observed two peaks can be associated with the L-HH transition and Г-HH transition in compressively strained GeSn, or with the bottommost CB valley to HH and LH transitions (inset in Fig. 3). Meanwhile, the experimental observations are (1) an increase in PL intensity, (2) a reduction in the separation (⁠ $ΔE$⁠) between the two peaks, and (3) a reduction in strain are observed after the annealing treatment. The decrease of compressive strain with annealing decreases the HH-LH splitting (ΔE_HH−LH) and, in the case of an indirect bandgap GeSn alloy, the Г-L separation (ΔE_Г−L) also decreases.^33,34 This makes it difficult to tell which transitions are responsible for the two peaks. Moreover, independent of the explanation for the two peaks, (4) the position of the low-energy peak (E₁) is observed to be at its lowest strain and energy after the annealing for two and four hours. Both the Г-valley and L-valley decrease in energy with compressive strain relaxation.³ However, the fact that the E₁ peak for the sample annealed for eight hours is at a higher energy serves as an evidence of a lower Sn content in this sample. In this case, the effect of a slight decrease in Sn acts opposite to the effect of compressive strain relaxation, neutralizes the expected change in E₁, and is consistent with all observations with annealing. These observations, therefore, point to only a slight decrease in Sn content, and a sufficient reduction in strain to produce a net decrease in ΔE_Г−L.

The Raman spectra were also measured for the as grown and annealed samples [Fig. 3(b)]. The Ge-Ge mode of the annealed samples is seen red shifted relative to the Ge-Ge peak of the as-grown sample, which can be understood given its relationship with the in-plane strain (⁠ $ε ∥$⁠) and Sn content (⁠ $x$⁠), $Δω=ax+b ε ∥$⁠.^35–37 Considering the Raman-Sn (a = 84 ± 8 cm⁻¹) and Raman-strain (b = −491 ± 52 cm⁻¹) coefficients,³⁶ the observed red shift of the annealed sample indicates a decrease of compressive strain. In fact, a decrease in strain is often reported after the temperature treatment^17,18,27 and is likely the cause of the red shift. At the same time, the Ge-Ge peak for the sample annealed for eight hours is seen at larger wavenumbers compared to that annealed for two and four hours, which is consistent with a slight loss of Sn atoms.

According to this condition of steady-state diffusion, Sn segregation is expected in the closest vicinity of a dislocation. Therefore, the process of GeSn alloy decomposition is enhanced in the presence of dislocations. While the transport of Sn atoms toward the surface looks feasible through the TDs, Sn diffusion to the surface is expected, as well as ripening of 3D Sn structures at the surface.⁴¹ Together, these results support the hypothesis that (1) Sn migration occurs early during annealing, by both threading dislocations to the surface and by diffusion to the surface and (2) Sn coalescence on the surface occurs as Sn appears on the surface.

In conclusion, we have demonstrated the effect of dislocations dynamics on the enhanced Sn out-diffusion in GeSn layers. The decrease of Sn composition after the annealing was detected by studying changes in the energy bandgap and lattice spacing using photoluminescence spectroscopy and x-ray diffraction technique. According to x-ray diffraction measurements, the strain in the GeSn layer decreased from −6.3 × 10⁻³ and stabilized at −2.1 × 10⁻³ during annealing at 300 °C. This change in strain state was attributed to migration of TDs. Surface particles were observed for annealed GeSn layers. The particles density was in good agreement with the evolution of strain, and thus, showed dependence on the TDs relocation. It turns out that TDs migration strongly enhances the process of Sn out-diffusion by harvesting Sn across the GeSn layer. These results suggest that suppression of Sn diffusion from the GeSn layer can be achieved by reducing the density of dislocations or reducing dislocations migration. Both the density of dislocation and their propagation after the growth can be achieved by utilizing lower growth rates or by introducing growth interruptions. Annealing at 300 °C is low enough to be sensitive to observe low Sn out-diffusion. While the appearance of Sn particles on surface indicates a local diffusion of Sn, we now understand that this diffusion is a result of accumulation of Sn across the sample due to extending misfit dislocations and corresponding threading dislocations.

The work was supported by the Air Force Office of Scientific Research (AFOSR) (Grant No. FA9550-19-1-0341).

The authors have no conflicts to disclose.

Hryhorii Stanchu: Conceptualization (equal); Data curation (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Abdulla Said: Formal analysis (equal); Methodology (equal). Oluwatobi Olorunsola: Data curation (equal); Formal analysis (equal); Investigation (equal); Validation (equal). Sudip Acharya: Data curation (equal); Formal analysis (equal); Validation (equal). Sylvester Amoah: Formal analysis (equal); Methodology (equal); Validation (equal). Mohammad Zamani-Alavijeh: Data curation (equal); Formal analysis (equal); Methodology (equal). Fernando M. de Oliveira: Formal analysis (equal); Methodology (equal); Validation (equal). Santosh Karki Chhetri: Formal analysis (equal); Methodology (equal); Validation (equal). Jin Hu: Formal analysis (equal); Investigation (equal); Validation (equal). Yuriy I. Mazur: Formal analysis (equal); Methodology (equal); Validation (equal). Shui-Qing Yu: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Project administration (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal). Gregory Salamo: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Project administration (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal).

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