The rapid development of monolithic optical-electronic integrated circuits (MOEICs) on silicon opens a door to achieving broadband, high-density and high-speed on-chip signal interconnections strongly desired for explosive data processing and communications.1 So far, the biggest challenge encountered for MOEICs is the on-chip light source compatible with the Si integration technology, which has been continuously pursued for decades.2–4 This is because Si(Ge) cannot generally accomplish efficient radiative recombination due to the indirect bandgap nature. A series of approaches have been proposed to overcome this shortage, e.g. (i) enhancing emission efficiency of Si/Ge nanocrystal5 and quantum dots (QDs)6,7 via quantum confinement effects; (ii) achieving direct-bandgap emissions in highly n-doped/ tensile strain Ge8,9 and GeSn alloy10,11 by band engineering; (iii) integration of III–V materials on Si.12 Recently, the extraordinary emission of hexagonal-GeSi (H-GeSi) demonstrates an alternative solution because of its direct-bandgap nature.13 By changing the crystal structure of GeSi from a general cubic lattice to a hexagonal lattice, the conduction band minimum (CBM) of cubic-GeSi (C-GeSi) at the L point along the [111] direction can be folded back onto the \({\Gamma }\)-point in the hexagonal Brillouin zone (BZ) of H-GeSi, leading to a direct-bandgap semiconductor.14–16 Moreover, the two lowest conduction bands of H-GeSi can be inverted by applying tensile strain17 or via the symmetry reduction due to alloying.18 Accordingly, the strong dipole-allowed transition in H-GeSi can be realized. In comparison with the cubic phase, the hexagonal phase of GeSi is not energetically favorable. Thus most GeSi are in the cubic phase via conventional growth techniques. So far, the H-GeSi is generally fabricated in small volumes via the crystal-phase transfer from the III-V hexagonal-nanowire template,13 ultraviolet laser ablation19 and strain induced crystal transformation.20,21 Nevertheless, those obtained H-GeSi is not fully compatible with Si-based COMS processes, given the existing III–V substrate (epitaxial template) or unconventional nanostructures. The hexagonal local domains have also been formed in C-Si or C-Ge nanowires via generally termed stacking faults.22–24 The inherent mechanism is related to compressive-strain-induced phase transformation from cubic lattice to hexagonal lattice.21,25−27 Although the compressive strain naturally exists in Ge film epitaxially grown on Si substrates,28 the H-Ge local domains originated from the stacking fault in the C-Ge film has essentially not been addressed due to the limited density of stacking faults and the lack of related characterizations.
Herein, we demonstrate H-Ge nanostructures directly realized within C-Ge on Si (001) substrates. These H-Ge nanostructures are obtained via epitaxial growth of thin Ge layers at a rather low temperature of 200°C, namely LTGe. They are demonstrated by the atomic stacking-fault in high-resolution transmission electron microscope (TEM) images, and discussed in terms of strain-induced phase transformation under far-from equilibrium growth conditions. Particularly, the direct bandgap features of H-Ge nanostructures are corroborated by the power- and temperature-dependent PL spectra of the metasurface composed of C-GeSi nanodisks with H-Ge nanostructures. Our results disclose a feasible strategy to realize H-Ge nanostructures in C-GeSi film, which demonstrates promising optoelectronic properties. Considering different band structures and charge densities, such an unprecedented heterostructure with different crystal phases may significantly modulate the electrical, thermal, and optical properties of GeSi. These H-Ge nanostructures have great potentials for amazing multifunction devices, particularly for the innovative light source in Si-based MOEICs.
Structural characteristics of LTGe
The overall layer structure of LTGe sample is schematically shown in Fig. 1a. The details can be found in the Methods section. Figure 1b and 1c shows the surface morphologies of the third and the first layer LTGe, respectively. It can be seen that very small QDs with an average height of 1.18 nm are obtained in the first-layer LTGe. Such a morphology is associated with the low growth temperature of 200 oC, which can significantly reduce the surface diffusion length of adatoms. In general, the smaller QDs, the less strain relaxation.28 Accordingly, the lattice mismatch between Ge and Si facilitates the formation of defects, involving stacking faults.29–32 For multilayer growth of LTGe, the accumulation of misfit strain energy promotes the formation of the big QDs, as demonstrated in Fig. 1b. This is consistent with previously reports.6,33 Meanwhile, it also stimulates the formation of stacking faults in the subsequent LTGe.
To obtain more structural information, Raman spectra of LTGe samples before and after rapid thermal annealing (RTA) are measured, as shown in Fig. 2. Obviously, the Raman spectra of annealed and unannealed samples are nearly the same. This result indicates that the present annealing process essentially does not change the composition and strain of LTGe and SiGe alloy. Based on the Raman spectra, the Ge composition and the strain in LTGe and the SiGe alloy film can be obtained, as shown in Table S1 in the Supplementary Information. The ultrahigh content (~ 97.2%) of Ge in LTGe layers is related to the suppressed intermixing of Ge and Si between LTGe and surrounded SiGe alloy grown at the low temperature. A similar result has been obtained in LTGe embedded in Si.34 The Ge composition of the SiGe alloy film is ~ 11%. It is consistent with the nominal value based on the growth conditions.
The H-Ge nanostructures formed in LTGe
Figure 3a shows the high-angle annular dark field (HAADF) scanning transmission electron microscopy (STEM) image of the SiGe film embedded within LTGe. It can be found that the first layer of LTGe looks very like a quantum well since the QD height is much smaller than the wetting layer thickness. This phenomenon agrees well with the AFM result shown in Fig. 1. The high-resolution STEM image of the first-layer LTGe is shown in Fig. 3b. It can directly reveal the characteristic dumb-bell-like features of Si/Ge dimers and their stacking sequence, which is always employed to characterize the crystal structure of Si and Ge.22 The atomic stacking fault can be distinguished in the colored region in Fig. 3b. The related atom arrangement around the stacking fault in the blue-box in Fig. 3b are clearly demonstrated in the zoomed-in STEM image in Fig. 3c. Figure 3d shows the HAADF-STEM around third layer of LTGe. The rolling boundaries of LTGe and SiGe alloy indicates the formation of big QDs, which agree with the AFM results in Fig. 1b. The high-resolution STEM image of the region in the blue box in Fig. 3d is shown in Fig. 3e. In the regions around stacking faults exhibited in Fig. 3c and 3e, an ABA stacking sequence is embedded in the ABC stacking sequence that is the fingerprint of the C-Ge demonstrated in Fig. 3f. Such an ABA stacking sequence can be regarded to be a segment of the typical ABAB… stacking sequence along the [0001] direction as the hallmark of a hexagonal crystal structure,13,20,22 as shown in Fig. 3g. Accordingly, this stacking fault can be regarded as a H-Ge nanostructure, i.e. partial crystal structure of LTGe in the region is transformed from C-Ge to H-Ge. The H-Ge nanostructure in Fig. 3c is composed of 3 layers of dumb-bell-like dimers, named as 3L H-Ge. It can be considered to be the production from the nearby C-Ge through twin operations or mirror operations with the mirrors denoted by green dashed-lines (with label m) in Fig. 3c.22 With the accumulation of the compressive strain energy in the subsequent LTGe, more mirror operations can occur. They lead to the complex atomic layer structures involving adjacent 2L, 3L and 4L H-Ge, single 2L H-Ge, and C-Ge with different crystal orientations, as demonstrated in Fig. 3e. The similar crystal structure transformation from C-Si to H-Si around stacking faults has also been reported.22
This crystal phase transition from C-Ge to H-Ge is associated with the compressive stress in the epitaxial Ge on the Si (001) substrate and the far-from-equilibrium growth conditions. It has been found that the residual compressive stresses around controlled nanoindentations in the Ge amorphous film plays an important role in the observed phase transition from C-Ge to H-Ge.21 The stress-induced phase transition from C-Ge to H-Ge has also been observed in Ge nanowires.20 In addition, the required stress for the onset of phase transition can be reduced for the small nanostructures.20 The H-Si can also be generated from C-Si owing to the stress occurring under experimental conditions of high pressure and annealing.22,25 The epitaxial growth of Ge on the Si (001) substrate naturally gives rise to the substantially large compressive stress in the Ge, which facilitate the crystal phase transition. Moreover, the LTGe is obtained at a rather low growth temperature of 200 oC. Such a far-from-equilibrium growth condition facilitates the formation of the metastable phases of Ge, e.g. H-Ge. The low temperature growth always leads to a rough surface with nonuniform surface stress, which has been found to induce phase transition in nanostructures.35 It has also been found that a fast growth rate can freeze stacking faults to form H-Si.23 Similarly, a rather low growth temperature can also freeze stacking faults to form H-Ge nanostructures in the present cases. As a result, sufficient H-Ge nanostructures are realized in the LTGe epitaxially grown on Si (001) substrates.
Optical properties of H-Ge nanostructures
To characterize the optoelectronic properties of H-Ge nanostructures, the three layers of LTGe are embedded in a metasurface composed of C-GeSi nanodisks. The schematic structure and SEM images of nanodisk array are shown in Fig. S1 (Supplementary Information). Ordered SiGe nanodisks in a two-dimensional hexagonal lattice are obtained. The period, the diameter and the thickness of the SiGe nanodisks are about 820, 670 and 195 nm, respectively. Such small sizes of nanodisks are beneficial for the limited cavity modes around communication wavelengths (1310 and 1550 nm).36
Figure 4a shows the PL spectrum of nanodisk array with embedded LTGe at room temperature (293 K). There are three peaks in the PL spectrum from 1200 to 1850 nm with the labels of P1 (1250 nm), P2 (1560 nm) and P3 (1838 nm), respectively. The enhancement factors caused by coupling into cavity modes around P1-3 are 4.6, 11.2 and 2.1, respectively. Figure 4b demonstrates the simulated emission spectrum. The simulated peak wavelengths of cavity modes in the nanodisk array are consistent with those of the experimental peaks P1-3. These modes can be described as the Mie-like multipolar modes according to their field distributions shown in Fig. 4c-f. The peak P1 is related to an electric quadrupole mode in the nanodisk.37 The peak P2 is associated with the anapole mode.38 It is the ideal cavity mode applied in enhancing the emission of QDs due to unique light confinement within the nanodisk.39 The electric field for peak P3 in Fig. 4e indicates a vertically oriented magnetic dipole mode.37
More interestingly, abnormal temperature-dependent intensity of peak P2 is observed, as shown in Fig. 5a. For the temperature in the range of 16–160 K, the integrated intensity increases slightly at first and then decreases remarkably with increasing temperature. These features are consistent with those of PL Intensity from C-Ge/Si SK nanostructures.40 However, the integrated intensity is essentially constant for the temperature above 160 K. This result is completely different from that of the typical C-Ge/Si nanostructures with the indirect bandgap nature, which shows a significant quenching near the room temperature.41 In addition, the peak shape is nearly the same in the temperature range of 160–293 K, as shown in the inset of Fig. 5a. These results demonstrate a nearly temperature-insensitive radiative efficiency in the range of 160–293 K. Such a temperature-insensitive radiative efficiency is a typical feature of direct-bandgap semiconductors.13,41 This means that the emissions in LTGe change from indirect-bandgap to direct-bandgap radiative recombination over 160 K. This transition is further corroborated by the power-dependent PL integrated intensity at 16 K and 293 K, as shown in Fig. 5b. The PL intensities (I) of peak P2 vs excitation power (P) can be fitted by I ~ Px. The indices x at 16 K and 293 K are 0.675 and 1.048, respectively. The index x of 0.675 at 16 K is very close to that of 0.62 for the PL of un-patterned sample, as demonstrated in Fig. S2d in the Supplementary Information. It means that the index x is mainly associated with the emissions of LTGe. This result is rational since the coupling efficiency between the emissions of LTGe and the cavity modes of nanodisk array is essentially power-independent, given the broadband emissions of LTGe and the relatively slight blue-shift with excitation power shown respectively in Figs. S2c and S2d. Accordingly, the behaviors of power- and temperature-dependent PL peak P2 of nanodisk array are essentially similar to those of PL spectra of LTGe. The sublinear power dependency of integrated intensity at 16 K is attributed to the indirect-bandgap of LTGe.42,43 Whereas the index x of nanodisk array at 293 K is close to 1. This linear power dependency is usually related to the direct-bandgap nature.13,41 Therefore, the different behaviors of power-dependent PL intensities at different temperatures reveal the transition of radiative recombination in LTGe from indirect-bandgap to direct-bandgap. Moreover, the indices x of 0.688, 0.866, 1.007 and 1.003 are observed in anther sample at 22, 100, 200 and 293 K, respectively, as shown in Fig. S3 in the Supplementary Information. It strongly indicates the transition from indirect-bandgap to direct-bandgap via increasing the temperature. Given that the sub-nanosecond radiative recombination lifetime of carriers in H-GeSi is comparable to that of direct-bandgap group-III–V semiconductors,13 the behaviors of the linear power-dependent PL intensity and the temperature-insensitive PL spectra near the room temperature are justified since the photon-generated carriers can readily take part in the radiative recombination before nonradiative recombinations via Auger effect or nonradiative centers. They are the superior features of the direct-bandgap transition contrast to the indirect-bandgap transition.
To interpret the unique features of the PL spectra, the band diagram of LTGe (C-Ge) with H-Ge nanostructures embedded in the SiGe alloy film is proposed, as schematically shown in Fig. 5c. The band diagram highlights the type-I band alignment for the indirect-bandgap transition between the \({\text{E}}_{c\_L}^{C-Ge}\) and \({\text{E}}_{v\_hh}^{C-Ge}\) (denoted as the blue arrow), the type-II band alignment for the indirect-bandgap transition between \({\text{E}}_{c\_\varDelta }^{C-SiGe}\) and \({\text{E}}_{v\_hh}^{C-Ge}\) (denoted as the black arrow), and the direct-bandgap transition in H-Ge nanostructures from \({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\)to \({E}_{v\_hh}^{H-Ge}\)(denoted as the red arrow). Taking into account the Ge content and the strain listed in the Table S1, we estimate those transition energies in the LTGe based on self-consistent calculations.34,44,45 The details can be found in the Supplementary Information (S5). The transition energy for C-Ge between \({\text{E}}_{c\_L}^{C-Ge}\) and \({\text{E}}_{v\_hh}^{C-Ge}\) at 16 K is calculated to be about 0.810 eV (1531 nm). It is consistent with the peak B1 of the un-patterned film in Fig. S2b in the Supplementary Information. We argue that the peak P2 of nanodisk array at 16 K mainly arises from the emissions of indirect-bandgap transition between \({\text{E}}_{c\_L}^{C-Ge}\) and \({\text{E}}_{v\_hh}^{C-Ge}\) coupling into the anapole mode of nanodisk array. Thus, the peak P2 exhibits sublinear power-dependence with the index of 0.675. At the room temperature (293 K), the wavelength related to this indirect-bandgap transition increases to 1704 nm. The related emissions can hardly be coupled into the anapole mode of nanodisk array due to the spectra mismatching. On the other hand, the energy of the direct-bandgap transition from \({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\)to \({E}_{v\_hh}^{H-Ge}\) in H-Ge at room temperature is calculated to be 0.7892 eV (1571 nm), which is quite close to the wavelength of peak P2 of the nanodisk array. In this case, we argue that peak P2 at room temperature is related to the direct-bandgap transition in H-Ge nanostructure from \({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\)to \({E}_{v\_hh}^{H-Ge}\). Accordingly, the intensity of peak P2 at room temperature exhibits a liner power-dependence in Fig. 5b. It should be noted that the CBM of H-Ge is \({E}_{c\_{\varGamma }_{8c}^{-}}^{H-Ge}\) rather than \({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\).15,17 However, the transition between \({E}_{c\_{\varGamma }_{8c}^{-}}^{H-Ge}\)and \({E}_{v\_hh}^{H-Ge}\) is three orders of magnitude weaker than the typical dipole-active transitions in semiconductors. Therefore, the transition between the \({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\)and \({E}_{v\_hh}^{H-Ge}\) with the light polarization preferentially perpendicular to the [0001] axis of H-Ge provides the efficient recombination at the \({\Gamma }\) point for a sufficiently high temperature.17
At the low temperature (e.g. 16 K), the photon-generated electrons are preferentially confined in the C-Ge of LTGe and the C-SiGe due to the relatively lower conduction band minimum, as shown in Fig. 5c. Therefore, the PL at a low temperature is mainly related to the indirect-bandgap transitions of \({\text{E}}_{c\_L}^{C-Ge}\)-\({\text{E}}_{v\_hh}^{C-Ge}\) in C-Ge and of \({\text{E}}_{c\_\varDelta }^{C-SiGe}\)-\({\text{E}}_{v\_hh}^{C-Ge}\), denoted by blue and black arrow respectively in Fig. 5c. More details are seen in the Supplementary Information. Given the indirect-bandgap nature of those transitions, a sublinear power-dependent PL intensity at 16 K (Fig. 5b) and the significant decrease of integrated intensity with the increase of temperature from 16 to 160 K (Fig. 5a) are observed. At the temperature over 160 K, the PL intensity associated to the indirect-bandgap transition becomes substantially weak since its long radiative recombination lifetime facilitates the thermal activation of electron out of \({\text{E}}_{c\_L}^{C-Ge}\) and \({\text{E}}_{c\_\varDelta }^{C-SiGe}\) to \({\Gamma }\) valley in H-Ge (\({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\)). In addition, the red-shift of indirect-bandgap transition with temperature suppresses the related emissions coupling into the anapole mode of nanodisk array for peak P2. On the other hand, the thermally activated elections can readily take part in the emissions due to the rather short radiative recombination lifetime associated with the direct-bandgap.13 Thereby, the direct-bandgap transition of \({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\)-\({E}_{v\_hh}^{H-Ge}\) in H-Ge dominates the emissions around 1560 nm, which can couple into the anapole mode (P2) of nanodisk array. As a result, a temperature-independence and a linear power-dependence of peak P2 are exhibited for a sufficiently high temperature.
It is worth mentioning that the direct-bandgap transition of \({E}_{c\_{\varGamma }_{7c}^{-}}^{H-Ge}\)-\({E}_{v\_hh}^{H-Ge}\) in the un-patterned film is not observed at room temperature as shown in the Supplementary Information (S3). One reason is that the density of H-Ge nanostructure in LTGe is still not so high. Moreover, considering the light polarization and the [0001] axis of H-Ge,17 a large fraction of light generated inside H-Ge is guided in the slab by total internal reflection at surface for un-patterned film, which can hardly be detected. Whereas, the emissions of H-Ge can be remarkably enhanced and characterized by coupling into the cavity modes supported in SiGe nanodisk array.39 Given the full compatibility with the Si integration technology and the suitable wavelength around 1550 nm for communication, H-Ge nanostructures embedded in the metasurface can be a promising candidate for the innovative light source in the Si-based MOEICs. In addition, it has been found that the group III dopant can be more easily introduced in the H-Ge due to the local C3v symmetry,46 in comparison with the C-Ge. The H-Ge can even be stabilized by introducing carriers.47 Accordingly, the H-Ge nanostructure within C-Ge readily provides heterostructures with different band structures and charge densities. Such an unprecedented heterostructure of Ge may have unique electronic, thermal, and optical properties, which may be exploited for amazing multifunction nanodevices.