D. Stange, C. Schulte-Braucks, N. von den Driesch, S. Wirths, G. Mussler, S. Lenk, T. Stoica, S. Mantl, D. Grützmacher, D. Buca, R. Geiger, T. Zabel, H. Sigg, J. M. Hartmann and Z. Ikonic
Peter Grünberg Institut-9 and JARA-FIT, Forschungszentrum Jülich GmbH, Jülich, 52425, Germany
Laboratory for Micro- and Nanotechnology, Paul Scherrer Institut, Villigen PSI, 5232, Switzerland
Université Grenoble Alpes and CEA-LETI/MINATEC, Grenoble, 38054, France
Institute of Microwaves and Photonics, University of Leeds, Leeds, LS2 9JT, United Kingdom
The continuous progress of computer technology, with a larger amount of data transfer and higher data processing speed, has strongly increased the energy needed in order to run large data centers. With every new processor generation, the power consumption increases with the square of the clock frequency.1 Given the development of IT networks, energy consumption becomes one of the main bottlenecks. A large fraction of the energy consumption in data centers (38% in 2009) is coming from the cooling systems used to dissipate the heat mainly produced by copper interconnects linking devices between and on chips.2
The partial substitution of copper by optical interconnects would reduce the heat generation enormously. Indeed, attenuation of data transmission at the 10 Gb/s rate can be 1000 times lower optically, requiring much less power.1
The present chip technology is based on silicon with increasing number of other materials integrated into electrical circuits. Due to the constant research in silicon photonics, nearly all the components that are necessary for the photonic part of a silicon-based electronic–photonic integrated circuit (EPIC) have already been developed. The only exception is the light source, which today is based on III–V materials, adding complexity to the fabrication process due to the need for bonding techniques on the Si.3 4 Besides being costly, the bonding process of toxic III–V materials onto Si chips faces various technical challenges. For example, layers can be damaged during annealing steps due to thermal mismatch between III–V and Si materials.5 For industrial manufacturing of photonic systems, a straightforward integration of a light source would be preferable. Devices based only on group IV materials directly grown on Si would be ideal from that point of view.
Unfortunately, group IV semiconductors like Si and Ge exhibit an indirect bandgap. Radiative carrier recombination is then not efficient, making them unsuitable for laser applications. Different concepts like Si nanocrystals6 or Ge under high tensile strain7 are presently under investigation. Another approach consists in growing GeSn layers with high amounts of tin. This reduces the energy separation between conduction band valleys and the valence band as Sn content increases, with a stronger effect for the Γ-valley than for the L-valley,8–10 leading to a transition from an indirect to a direct bandgap for sufficiently high Sn content. The transition point for unstrained GeSn compounds was calculated to occur at a Sn concentration between 5% and 12%.11 12
Recently, the fundamental direct bandgap in partially relaxed GeSn alloys was demonstrated experimentally.13 The transition point for fully relaxed layers occurred for a Sn concentration of ∼9%. Lasing by optical pumping of waveguide structures was demonstrated and a differential gain of 0.4 cm/kW was determined.
The growth of strain-relaxed GeSn is not trivial. Indeed, the lattice mismatch between the Ge-on-Si substrate and the GeSn alloy leads to the presence of a compressive strain inside the GeSn layer, which shifts the transition point to really high Sn concentrations. This chapter will present a systematic photoluminescence (PL) study of compressively strained, direct-bandgap GeSn alloys, followed by the analysis of two different optical source designs. First, a direct bandgap GeSn light emitting diode (LED) will be characterized via power- and temperature-dependent electroluminescence (EL) measurements. Then, lasing will be demonstrated in a microdisk (MD) resonator under optical pumping.
The integration of direct-bandgap GeSn-based devices as a light source for on-chip communications offers the possibility to monolithically integrate the complete photonic circuit within mainstream silicon technology.14 The fabrication of a compact electrically pumped laser source could thus become a reality in the coming years.15
The Ge1−xSnx alloys were grown on 2.5 µm thick Ge virtual substrates (Ge-VS)16 on top of 200 mm Si(100) wafers in an industrial AIXTRON TRICENT reduced-pressure chemical vapor deposition (CVD) reactor17 18 from commercial Ge2H6 and SnCl4 precursors,19 with PH3 and B2H6 used for n-type and p-type in situ doping, respectively.
The growth temperature was kept in the 340–350 °C range, with high growth rates in order to avoid Sn precipitation. A TEM micrograph of a typical 400 nm thick epitaxially grown Ge0.875Sn0.125 layer is shown in Fig. 1. The crystalline quality of that layer, which is partly relaxed, is very high. Misfit dislocations are confined at the interface between GeSn and the Ge-VS underneath.19 The GeSn layer is thus free from threading dislocations (at least on the TEM scale).
Material analysis regarding crystallinity, Sn concentration, and strain was performed via Rutherford-backscattering-spectroscopy/channeling (RBS/C) and X-ray diffraction using reciprocal space mapping (XRD-RSM).
Here, we will succinctly describe material properties using Ge0.875Sn0.125 epilayers of various thicknesses. Up to a certain thickness, Ge1−xSnx grows pseudomorphically on Ge-VS (e.g., its in-plane lattice parameter is equal to that of the Ge-VS, while its out-of-plane lattice parameter is higher than its bulk value). For coherently grown Ge0.875Sn0.125, the compressive strain reaches −1.7%. Above a certain critical layer thickness, the alloy starts to plastically relax through the formation of misfit dislocations at the interface.
The amount of strain inside the GeSn lattice strongly affects its band structure. The strain-dependent band structure for Ge0.875Sn0.125 in Fig. 2(a) is calculated via the 8-band k·p method,21 22 using conventional deformation potentials for indirect valleys. With increasing relaxation of the highly compressively strained layers, the conduction band is shifted toward lower energies. However, the Γ-valley decreases faster in energy than the L-valley. The transition from an indirect to a direct bandgap semiconductor is predicted to occur at a compressive strain of about 1%. In addition, the valence band splitting is reduced with decreasing compressive strain, with the valence bands becoming degenerate at zero strain.
Temperature-dependent integrated PL intensity is a suitable method to determine whether a semiconductor has a direct or indirect fundamental bandgap.13 The integrated PL intensity of four layers with thickness increasing from 46 to 400 nm (and hence decreasing compressive strain) is shown in Fig. 2(b). The corresponding strain is given in the figure inset.
The PL intensity of the pseudomorphically grown Ge0.875Sn0.125 alloy (e.g., under a compressive strain of −1.7%) decreases with temperature T. This is due to the decreasing Γ valley electron population as T decreases (reduction of thermal transfer of carriers). This indicates an indirect band structure of this alloy, confirming the calculations of Fig. 2(a).
For a thicker 170 nm layer, we find a residual strain of −1.05% and the integrated PL intensity increases continuously. According to calculations of Fig. 2(a), the material might be just at the transition from an indirect to a direct bandgap semiconductor as indicated by the constant value of integrated intensity for 300 K and 250 K. The trend of continuously increasing signal with decreasing T is observed for all GeSn layers with less than −1.05% compressive strain (see Fig. 2(b)). Given this trend, these materials can be classified as being direct bandgap semiconductors. With decreasing compressive strain, the energy difference between the Γ- and L-valleys increases, which results in increased intensities at a constant temperature. Full temperature-dependent PL spectra of a 400 nm thick GeSn layer are shown in Fig. 2(c). The strong luminescence reveals the high optical quality of the material, which is mandatory for the fabrication of light sources.
LEDs made out of GeSn alloys have already been reported by various groups.23–25 Room-temperature EL was associated with Γ–valence band recombination; however, the directness of these materials remained questionable. Tin concentrations of up to 10% were reached,26 27 but temperature dependence was not reported. The Sn content and the strain in the previously fabricated diodes suggested that the GeSn alloys had an indirect bandgap. As a consequence, the current density needed to inject a sufficient density of carriers remained very high, for example, in the kiloampere per square centimeter range.24 28, 29 Here we will give a short overview of light emission in a direct-bandgap Ge0.89Sn0.11 LED that could fulfill a major requirement for future optoelectronic devices with reduced power consumption.
The LEDs presented here are p–i–n structures based on GeSn. They contain a Sn concentration of about 11.5% with a residual strain of −0.8%. In accordance with calculations regarding this configuration, the material should have a direct bandgap with a small energy offset between Γ- and L valley ΔEΓL = 2.4 meV. The carrier concentration in the p+ boron-doped region reaches values of 2 × 1018 cm−3, while the n+ phosphorus doping was 3 × 1019 cm−3. The “intrinsic” region is unintentionally p-doped at 9 × 1017 cm−3, possibly due to crystal defects.30
The mesa of the diode was fabricated by reactive ion etching using Cl2 and Ar plasma and passivated by Al2O3/SiO2 10:150 nm stacks deposited via atomic layer deposition (ALD)/plasma-enhanced chemical vapor deposition (PECVD). Contact windows were defined by optical lithography and opened by CHF3 dry etch (in order to contact the p and n regions afterward). A schematic of the LED structure and a 3D SEM can be seen in Fig. 3(a) and (b).31 The contacts were made by sputtering 10 nm of Ni and annealing the structure at 325 °C in a forming gas atmosphere to form NiGeSn.32 As a last step, Al was deposited to facilitate wire bonding.
Electroluminescence was measured under a pulsed injection at a frequency of 1988 Hz. The temperature-dependent EL of a Ge0.89Sn0.11 p–i–n diode is shown in Fig. 4(a). At current densities of about 120 A/cm2, a clear EL signal is obtained. The emission energy is blueshifted when the temperature decreases, in agreement with the expected temperature dependence of the bandgap. At low temperatures, both thermal activation of charge carriers from Γ- into the L-valley and defect-related nonradiative recombination processes are suppressed, resulting in an EL increase. The continuously increasing integrated signal in Fig. 4(b) can then attribute to the direct bandgap of the alloy.
Power-dependent measurements at room temperature show the suitability of these group IV alloys as light emitters at low current densities of 55 A/cm2 (see Fig. 4(c)). Reducing the temperature allows observation of a signal at even lower current densities.
It is, however, obvious that GeSn p–i–n homojunction LEDs do not provide a good confinement of carriers inside the active region (when targeting optimized LED structures and designing electrically pumped lasers). Also, heterostructures like Ge/GeSn/Ge do not offer high band offsets. The Ge layer sitting on top of the thick, nearly fully relaxed GeSn layer, is necessarily under tensile strain, making type I band alignment difficult to reach. Therefore, it seems mandatory to use SiGeSn as a cladding material. In addition, multiquantum well (MQW) structures may offer low threshold powers given their 2D density of states.33
Both the integration of an on-chip light source and dense packaging require a compact resonator geometry. One possible design is the MD geometry, which is already used in III–V technology as a suitable approach for integrated light sources on a silicon chip.34 35
Group IV microdisk experiments were also reported including SiC and pseudomorphic Ge/GeSn/Ge heterostructures. Whispering gallery modes (WGM) were observed in both cases; however, no evidence of lasing was reported.36 37 The same was true for Ge membranes under high tensile strain38 or MDs that used SiN or other stressing methods.39 40
The microdisks presented here are underetched, with an undercut of 3.6 µm using selective etching between Ge and GeSn. This leads to an increase of the effective refractive index and, therefore, improves the optical confinement of WGM modes inside the disk resonator. Another beneficial impact of this undercut is the strain relaxation that occurs at the edges of the disk. By partially etching the Ge buffer beneath the Ge1−xSnx layer, the latter is indeed able to fully relax. With a higher degree of relaxation of the GeSn lattice, the energy separation between Γ- and L-valleys becomes larger, yielding a direct bandgap alloy. The same effect was discussed in the previous section regarding relaxation in Ge0.875Sn0.125 layers. The alteration of the band structure due to the disk geometry (i.e., due to strain gradients) leads to a diffusion of the charge carriers toward the edges of the disk. This results in increased Γ-population at the disk edges, where the WGM are formed, which may help in reducing the lasing threshold. For a strain change from −0.4% to 0% (e.g., fully relaxed Ge0.875Sn0.125 layers), the Γ-population changes from 10% to 30% of the total electron density in the conduction band, as calculated by the 8-band k·p and deformation potential method.
The MD laser mesa structure was defined by e-beam lithography followed by ∼1 µm dry etching with Cl2/Ar. Then, Ge was selectively dry-etched isotropically by a CF4 plasma.41 For the reduction of nonradiative surface recombination, a 10 nm thick Al2O3 layer is deposited at 300 °C by ALD.
The previously described strain relaxation process was investigated via Raman spectroscopy. A WiTec setup with a 532 nm laser diode was used to measure the signal of Ge–Ge vibrations in a GeSn lattice. The Raman modes in Ge bulk are seen at ∼300 cm−1. With an increasing amount of Sn in the alloy, the signal is shifted to lower wavenumbers, whereas strain shifts the Raman peaks toward higher wavenumbers.41 42 As a result, for a given Sn content, plastic relaxation in GeSn should lead to a wavenumber decrease. This is exactly what is seen in Fig. 5(a): a shift toward lower wavenumbers when moving from the center toward the edges of the disk. The complete distribution of the Raman peak's wavenumber inside the disk is presented in Fig. 5(b). A further proof of strain relaxation is given by locally resolved PL measurements. The PL peak from the edge of the disk is redshifted compared to the signal from the center, due to the lower bandgap initiated by strain relaxation at the edges of the disk (see Fig. 5(c)).
Strain variation along the radius of disk induces local band structure changes. The lower energy of the bands at the edges compared to the center leads to a built-in electric fields that results in a diffusion of electrons toward the edges of the disk. Because of the location of WGM at the disk edges, carrier diffusion is a boon to lasing action and reduces the lasing threshold.
Incident light coming from the top from a pulsed 1064 nm Nd:YAG laser diode with a pulse duration of 6 ns was used to optically pump the microdisk laser. The generated laser light is scattered out by imperfections at the microdisk sidewalls and guided to an InSb detector.
An 8 µm diameter disk with a Sn concentration of 12.5% was investigated. The compressive strain in the initial Ge0.875Sn0.125 layer was −0.4% according to X-ray diffraction (XRD). Band structure calculations give the Γ–L energy separation ΔEΓL = 73 meV for Ge0.875Sn0.125 at 100 K. Spontaneous emission becomes visible for optical pumping of the microdisk at low power. With increasing pumping power, an exponential enhancement of emission is observed (see Fig. 6(a)). The lasing threshold at 20 K is 128 kW/cm2. For pumping powers above 500 kW/cm2, light emission slowly starts to saturate.
Figure 6(b) shows PL microdisk spectra below and above threshold. The broad spectrum coming from spontaneous emission reveals a more than 10 times larger full width half maximum (FWHM) compared to the narrow laser linewidth. These measurements demonstrate the typical laser characteristics of this novel group IV semiconductor laser based on GeSn. Details of the microdisk laser can be found in Ref. 43.
In conclusion, we presented growth and optical characterization of high-quality GeSn alloys with very high Sn content. The change in band structure coming from plastic strain relaxation was discussed via PL analysis of different layers with various degrees of relaxation.
We have fabricated direct bandgap LEDs from our GeSn alloys, measured their luminescence, and discussed avenues to improve carrier confinement and obtain even more efficient LEDs.
Special attention was allocated to fabrication and characterization of underetched microdisk GeSn/Ge structures. They were investigated by Raman mapping and band structure calculations. Because of the undercut, nearly full relaxation was obtained at the edges. For a Sn concentration of 12.5%, lasing in those whispering gallery mode devices was achieved at a record-breaking 128 kW/cm2 threshold.
Combining these two results, electrical injection into a resonator geometry coupled with the suggested improvements for LED structures should pave the way toward an electrically pumped GeSn laser grown directly on Si(100). This unique feature of GeSn alloys can be advantageous over bonded III–V light sources and could lead in the future to an affordable use of such heterostructures in complex EPICs.
3.143.239.44