This action might not be possible to undo. Are you sure you want to continue?
PUBLISHED ONLINE: 14 APRIL 2013 | DOI: 10.1038/NPHOTON.2013.67
Analysis of enhanced light emission from highly strained germanium microbridges
¨ ess1,2,3†, R. Geiger1†, R. A. Minamisawa1†, G. Schieﬂer1,2, J. Frigerio4, D. Chrastina4, G. Isella4, M. J. Su R. Spolenak2, J. Faist5 and H. Sigg1 *
Tensile strain is a widely discussed means for inducing a direct bandgap in Ge for the realization of a semiconductor laser compatible with Si microelectronics. We present a top-down fabrication approach for creating high uniaxial tensile stress in suspended Ge structures, which enhances—by a factor of more than 20—the strain induced by thermal mismatch of Ge layers grown on silicon or silicon-on-insulator substrates. Strain values up to 3.1% are measured using Raman spectroscopy, in excellent agreement with simulations using a biaxial thermal strain of 0.15%. As expected from the high value of strain, a 210 meV peak energy shift in the emission with respect to bulk Ge and a strong increase (325) in the integrated photoluminescence intensity are observed. Although 3.1% uniaxial strain does not transform Ge into a directgap material, our model calculation predicts an optical gain of 460 cm21 for 1 3 1019 cm23 n-doped structures at an electron–hole injection density of 3 3 1019 cm23.
he ongoing increase in the integration density and complexity of complementary metal–oxide–semiconductor (CMOS) circuits poses new challenges for materials science and physics. To beneﬁt from the increasing switching speed of transistors in modern microprocessors, the signal propagation delay in interconnects needs to decrease. The limited bandwidth of on-chip data transfer via electrical wiring limits the performance of the newest generation of processors, despite the ever-growing operating frequencies and packing densities1. Optical data distribution for onchip communication as well as interconnection is therefore a promising alternative for CMOS-based devices, as it will enhance signal synchronization, improve noise immunity and lower power dissipation2. Nearly all the main components for optical interconnects have been demonstrated on a Si platform, including modulators3, photodetectors4 and nonlinear elements5; the only exception is an efﬁcient light emitter. Although laser devices using III–V direct-bandgap semiconductor emitters on Si have been demonstrated6,7, they have rarely been incorporated into Si-based CMOS technology because of integration challenges and high costs. Recently, Ge has been the subject of considerable attention as a potential candidate in the development of an integrated laser on Si platforms8 because of its compatibility with standard CMOS technology9. Although Ge is an indirect-bandgap semiconductor, with the valence band maximum located at the G point and the conduction band minimum located at the L point, a local minimum is present in the conduction band at G, and manipulation of the band structure through strain and doping may enable efﬁcient radiative recombination. If a high tensile stress is applied to a Ge crystal, the induced deformation (with tensile strain components along the main stress directions) can introduce a crossover of the G and L conduction band minima, thus converting Ge into a direct-bandgap semiconductor10. This transition is predicted to occur somewhere between 1.7% and 2.2% for equibiaxial strain parallel to the (100) crystallographic plane11,12, and at .4% uniaxial strain along the  crystallographic directions13. Researchers have reported results for lasing in Ge
induced by optical14 and electrical15 pumping in low-strained (0.25%) but heavily doped Ge. The applied strain of 0.25% is insufﬁcient, by far, to lower the G conduction band minimum below the L valley. However, it lowers the dopant (and therefore the injection) density required to ﬁll up the L conduction band valleys13. Several works have also proposed mechanical approaches to achieve a direct bandgap in Ge layers, using released and stressed Ge nanomembranes16, or stressor layers deposited around patterned Ge stripes17. Most recently, up to 1% biaxial and 1.4% uniaxial strain has been applied by patterning and releasing Ge on insulator, with stressor layers deposited at the contact pads18. Although all these methods have demonstrated high tensile strain and therefore enhanced the photoluminescence intensity of Ge, none is suitable for integrating highly stressed Ge layers directly on Si. Additionally, the strain level and/or the layer thickness obtained with methods using external stressor layers are limited by the efﬁciency of strain transfer between the ﬁlms18. In this Article, we report a powerful technique to fabricate highly uniaxially stressed Ge structures on Si and on Si on insulator (SOI). Our top-down fabrication method consists of patterning and underetching constricted structures from biaxially strained Ge layers directly grown on Si and SOI substrates, resulting in a uniaxial stress state. As the force in the ﬁlm and its thickness are constant and the stress is inversely proportional to the cross-sectional area, the stress increases at the constricted regions, but decreases in the pads. Because this stress enhancement effect is based on pure elastic deformation and depends only on the geometrical dimensions of the suspended structure, the maximum strain is limited only by the mechanical strength of the material19. Previously, we have used this approach to enhance an initial biaxial strain of 0.8% in 13 nm thin Si nanowires by a factor of 5.5 (ref. 20). Here, we extend this method to micrometre-sized Ge structures, thereby demonstrating its potential applicability over a wide range of length scales and substrate materials. We achieved a longitudinal strain of 3.1% (corresponding to an enhancement factor EF of 22)
Laboratory for Micro- and Nanotechnology (LMN), Paul Scherrer Institut, CH-5232 Villigen, Switzerland, 2 Laboratory for Nanometallurgy (LNM), Department of Materials Science, ETH Zurich, CH-8093 Zu ¨rich, Switzerland, 3 Electron Microscopy (EMEZ), ETH Zurich, CH-8093 Zu ¨rich, Switzerland, 4 L-NESS, Dip. di Fisica del Politecnico di Milano, Polo di Como, I-22100 Como, Italy, 5 Institute for Quantum Electronics (QOE), ETH Zurich, CH-8093 Zu ¨rich, Switzerland; † These authors contributed equally to this work. *e-mail: firstname.lastname@example.org
NATURE PHOTONICS | VOL 7 | JUNE 2013 | www.nature.com/naturephotonics
© 2013 Macmillan Publishers Limited. All rights reserved.
HR-XRD reciprocal space maps of the as-grown Ge/SOI material.5 Qx (nm−1) 32. A schematic of the fabrication process ﬂow is presented in Fig. a. NATURE PHOTONICS | VOL 7 | JUNE 2013 | www.NATURE PHOTONICS a DOI: 10. in top-down fabricated Ge structures without external stressors. as the starting substrates. and can be visualized using differential interference contrast (DIC) microscopy (Fig.1038/NPHOTON. as well as the length B and width b of the pads (Fig. 1b). yielding high photoluminescence enhancement. showing the Ge peak offset that indicates initial biaxial strain. the geometrical dimensions of the structure can be deﬁned by the length A and the width a of the constriction.67 ARTICLES (224) 46. We used equibiaxially tensile-strained Ge layers on bulk Si (Ge/Si) and on SOI (Ge/SOI). 1d). A cross-section of a Ge/SOI layer structure is shown in the scanning electron microscope (SEM) image of the focused ion beam (FIB) cross-section in Fig. This parameter is relevant only for the Ge/SOI structures because of the isotropic etch of the buried oxide layer (BOX). 1c (for details see Methods).0 45.com/naturephotonics Figure 1d presents a light microscope image of a constricted structure patterned on Ge/SOI and Fig.0 Ge 45.2013. The constrictions are aligned along k100l. As the thicknesses of the layers do not vary over the area of a pattern.5 Si 46. FIB-SEM cross-sectional image of Ge/SOI substrate.0 45.5 (004) 46. The depth of the etching beneath the frame (L) is highlighted by the outer dotted line shown in white.5 31. is maximal.15%) and are in good agreement with values found with high-resolution X-ray diffraction (HR-XRD) reciprocal space mapping (Fig. . d.0 33. All rights reserved.0 44.5 44. which are in the case of our samples induced by bending of the under-etched layer due to relaxation (see below). c.0 0. B and b.5 Qz (nm−1) SiO2 Si 45. 1a. In the case of Ge/Si structures.0 31. b. Tilted SEM image of a Ge/Si structure showing the geometrical parameters A.5 0. Differential interference contrast light-microscopy image of a Ge/SOI structure.5 b Surface 46. The equibiaxial tensile strain values of the unpatterned layers were determined with Raman spectroscopy (1bi ¼ 0. Finally.0 32. 1e).nature. DIC microscopy allows imaging of topographical gradients. a. respectively.5 44. Parameter L represents the length of the etched region under the Ge/Si layers.0 −0.0 c Electron-beam lithography RIE etching Wet etching d Ge/SOI e Ge/Si L <001> 10> A <0 <100> a b B <010> 50 μm <100> 50 μm Figure 1 | Characterization of prestrained Ge layers and fabrication of suspended constricted structures. the etching length L is negligible 467 © 2013 Macmillan Publishers Limited. beyond the frame of the structures. 1e an SEM image of a constricted structure patterned on Ge/Si. Schematics of the process ﬂow used to fabricate suspended and constricted structures. e.5 1 μm 44. and thus the direct gap recombination efﬁciency. we show that the band offset induced by the strain gradient in the Ge structure causes photogenerated carriers to diffuse efﬁciently to the constriction where the strain.
the relaxed pad. Commonly. All rights reserved. 2c) as well as the Raman line scan in Fig.1038/NPHOTON. the equibiaxially strained layer. the Raman shifts in all regions of the structure. showing 1.5 −40 −20 0 Distance (μm) 20 40 60 Figure 2 | Raman spectroscopy and FEM investigation of the strain distribution in the suspended constricted Ge structures. the uniaxially stressed centre. Increasing L results in higher strain at the constriction because it essentially extends the size of the suspended pad.ARTICLES a 288 Raman shift (cm−1) 294 Bulk Centre Layer Pad 300 306 NATURE PHOTONICS b <001> <010> <10 0> DOI: 10.7% longitudinal strain (that is. so as to make a comparison with the measured Raman maps (see Fig. The Raman spectrum of a Ge bulk reference is also plotted for reference.) 50 μm c −5. this relation is purely linear. Raman shift (cm−1) −1 Experiment Simulation −2 Sim: εxx Sim: εyy Sim: εzz 1. The Ge–Ge mode peak measured in the starting biaxially stressed Ge layer (red curve) shifts 20. 2c match closely. d. 1yy and 1zz ). is the highest strain level in Ge reported using top-down fabrication approaches (strain value uncertainties are discussed in Supplementary Section S3b). Sim. c. to date. uniaxial stress. 468 and thus cannot be directly determined from a stress-shift coefﬁcient. for example.0 −2. as well as the transition zones between them. a.5 d 2.5 2 0 Rel. 2c and Supplementary Sections S1 and S2). Figure 3b shows the Raman spectra of Ge/SOI structures with different pad lengths.65 cm21 with respect to the bulk Ge peak. SEM image of a Ge/SOI structure with a strain EF of 22 (see equation (1)). The calculated and measured Raman mappings shown in Fig.8 cm21 with respect to the bulk Ge peak. The accurate match between the Raman data and FEM simulations suggests that the strain in the structures is mainly due to elastic deformation. which. Normalized Raman spectra taken at the positions indicated by coloured triangles in b.15% biaxial tensile strain. We therefore simulated the Raman mapping of the constricted structures using ﬁnite element method (FEM) models. Measured and calculated Raman shift mapping recorded on the Ge/SOI structure shown in b. which indicates that the strain in the constriction is enhanced at the expense of relaxation of the pads. . 2d. are well reproduced.5 0 −2 6 4 2 0 X (μm) −2 −4 −6 −3 0. corresponding to longitudinal strain 1xx ¼ 3. 2b). indicating that the stress/strain tensors obtained by FEM simulations are correct. patterning dimensions and etching parameters. strain in semiconductors is correlated to Raman shift by solving the secular equation under the presupposition of crystal orientation. Figure 2d displays the simulated orthogonal strain tensor components on a line scan along a constricted Ge/SOI structure. EF ¼ 11) at the centre of the structure. due to the strong selectivity of potassium hydroxide (KOH) towards (111) facets of the Si substrate.0 0 1. The Ge–Ge mode measured in the constriction (green curve) moves by up to 24. In contrast. This behaviour is apparent in the Raman map of the Ge/SOI structures (Fig. Structures with higher EFs feature stronger Raman shifts and therefore a higher strain. b. the Ge–Ge mode measured on the pad (blue curve) is very close to the bulk reference. Red lines indicate the position-dependent normal strain components obtained by FEM simulations (1xx .nature.com/naturephotonics © 2013 Macmillan Publishers Limited.0 0.5 −4. that is. Again. Figure 3a displays the Raman spectra of Ge/SOI structures with different under-etching lengths L. The spectrum for bulk Ge is plotted for reference (black dashed curve). involving a single stress-shift coefﬁcient22. stress/strain directions and phonon deformation potentials21. However. The colour bar represents the relative Raman shift (in cm21). 2 0. For simpliﬁed cases.1%. corresponding to 1bi ¼ 0.5 −1. Figure 3c compares the experimentally extracted EF and NATURE PHOTONICS | VOL 7 | JUNE 2013 | www.0 Strain (%) −2 Y (μm) Exp.2013. Figure 2a shows the Ge–Ge mode of Raman spectra measured at different positions on a suspended Ge/SOI structure (Fig. applying known elastic constants and phonon deformation potential parameters. Linescan of the measured (blue circles) and simulated (blue line) Raman shift along the centre part of a Ge/SOI structure (EF ¼ 11). the components of the strain tensor are not trivially related. with the exception of the centre of our constricted structures.0 −4 −60 −0.67 Intensity (norm. The cutoff in the experimental data is due to the limited range (100 mm) of the Raman piezo stage.
2013.3% deviation (Supplementary Section S3b). but different L. The calculated EFs from the analytical model are depicted in green and blue. where ti are the layer thicknesses and Yi the Young’s moduli in the  direction for Si and Ge. the FEM describes the experimental data much better than the analytical model. Furthermore. c.24. 469 © 2013 Macmillan Publishers Limited. b. strain (%) 2 3 4 4 b Bulk εxx/εbi = 10x 12x 16x 22x Bulk 25 Model enhancement 20 15 Ge/Si: analytical Ge/SOI: analytical Ge/SOI: FEM 3 Model strain (%) Intensity (norm. tSi/tGe .76 for Ge/SOI structures and h ¼ 1 for Ge/Si samples. the constriction exhibits a 25% additional strain enhancement (Supplementary Section S3a).NATURE PHOTONICS a L = 15 μm 20 μm 24 μm Intensity (norm. The dashed line represents coincidence between experiment and model calculation. Compilation of measured versus calculated EFs for both Ge/SOI and Ge/Si structures.1038/NPHOTON. a. Normalized Raman spectra collected on Ge/SOI structures with identical A. We attribute this difference to the inﬂuence of different process ﬂows and/or the different total number of dislocations per structure. The impact of bending on the strain of the Ge/SOI constrictions as a function of the thickness ratio between the Si and the Ge layers.) 2 10 1 5 0 292 296 300 304 292 296 300 304 0 5 10 15 20 25 Raman shift (cm−1) Raman shift (cm−1) Exp. 3d).4 μm) Removed SiO2 (1 μm) Si substrate −2 Ge (1. 20: EF = 1xx 2L + B A a A 1+ / + =h B B−A b B−A 1bi (1) Equation (1) depends only on ratios between the widths and lengths. The difference in stiffness between the Ge and Si layers for the Ge/SOI structures is compensated by the factor h ¼ tGeYGe/ (tSiYSi þ tGeYGe). even though equation (1) is a simpliﬁed model obtained from assuming a uniaxial stress/strain state and isotropic stiffness. In the present case h ¼ 0. so the strain enhancement is independent of the size scale of the structures. d. Fig.4 μm) −4 Si (340 nm) Removed SiO2 (1 μm) 0 1 Strain εxx (%) 2 3 Si substrate Figure 3 | Raman spectra and EFs collected on structures with different dimensions. Figure 3c compares the EF and the respective strains extracted from the Raman measurements versus the analytical and FEM predictions for both Ge/Si and Ge/SOI. L extends the pad length B. processing and material variations contribute a small variability in strain from different constrictions with equivalent design. The surface strain at the upper surface in the centre of the constriction of Ge/SOI structures can be enhanced by a factor of up to 33% if tSi/tGe ≈ 0. amounting to a maximum of 3. B and b.1% for Ge/SOI structures. In this case. the structures will bend upwards after being released. This is because the model does not assume the occurrence of bending induced by out-of-plane strain gradients. The analytical model underestimates the enhancement for the Ge/SOI NATURE PHOTONICS | VOL 7 | JUNE 2013 | www.nature. In our case. where tSi/tGe ¼ 0. As the upper layer is made of the tensile-strained Ge. . As already mentioned. This effect is most pronounced in the centre of the constriction where the curvature is negative (Fig. a. but at 1xx ¼ 3. enhancement 0 d −8 0 −6 −4 −2 X (μm) 0 2 4 6 8 −2 Z (μm) −4 0 Ge (1. was investigated using FEM (Supplementary Section S3a). Normalized Raman spectra collected on Ge/SOI structures with different EFs.) B = 30 μm DOI: 10.com/naturephotonics structures.67 ARTICLES c 0 1 Exp. 3c indicates that the Ge/Si structures fracture at 1xx ¼ 1%. All rights reserved. that calculated analytically for Ge/Si and Ge/SOI. and the EF values as obtained from FEM calculations in red. resulting in an inhomogeneous strain distribution along the z-direction. The overall agreement between the experiment and the analytical equation is excellent for Ge/Si structures. using the equation introduced in ref.5. Also. FEM simulations of the cross-sections of 1xx along the x ¼  and z ¼  directions through the centre part of a structure without (top) and with (bottom) a Si layer under the pre-strained Ge layer.
64 eV 2.9 2.6% strain. Compared to bulk Ge. using an excitation power of 4.6 3 Norm. from which we deduce an approximated power density of 0. where G.1% Energy (eV) 0. . Unfortunately. the photoluminescence shift to lower energies.) 1.3 4. as demonstrated by the reﬂection spectrum shown at the bottom of Fig. 0/×1 1.0 0. The oscillations in the spectra are attributed to Fabry–Perot (FP) resonances of the suspended lamella structures. Although the onset remains constant for different measurement positions y.nature. L and Vup refer to the conduction band minima at G and L.5 Photoluminescence intensity (a. d. This upper bound is extracted from and conﬁrmed by power-dependent photoluminescence measurements performed in an all-metal-optics infrared microscope and by heat-ﬂow FEM simulations.64 eV decreases at increasing distances from the centre of the constriction ( y ¼ 0).67 0.0 3.8 0. c.25.ARTICLES a PL Simu.1% results in a redshift of the energy peak by 210 meV.3 2. taken at a comparable NATURE PHOTONICS | VOL 7 | JUNE 2013 | www. mPL spectra taken from structures with increasing longitudinal strain up to 3.7 0. respectively.8 0.u. All rights reserved.u. Position-dependent band-edge energies of the structure measured in b.0 0.7 0. e.5 0.7% 0.4 εxx (%) 2 1 0 0.2% 0.8 Photon energy (eV) 0. As the strain increases. b.6 0.2 4 d L L e 0.85 MW cm22.0 1. and an intensity increase by a factor of .com/naturephotonics © 2013 Macmillan Publishers Limited.9 −4 −2 0 Y (μm) 2 4 6 0 2 Y (μm) Energy (eV) 0.5/×5.5 mW from a laser source operating in continuous wave at a wavelength of 532 nm.4 0.6 Y (μm)/Norm.8 0.8 mm. and the upper valence band maximum. which is a clear indication of the reduced bandgap energy23 as well as of the closing of the offset between the L and G states. The mPL spectrum from Ge bulk is plotted for comparison.2 0.6 0. Figure 4a shows micro-photoluminescence (mPL) spectra recorded at room temperature in the constriction of Ge/SOI structures with different strain. The phase shift of the oscillation observed from sample to sample and also sometimes from bridge to bridge is caused by the thickness variability of the Ge layers. Normalized mPL spectra measured with the excitation laser scanning along the y-axis of a Ge constricted structure with 2.2 Photoluminescence intensity at 0. the photoluminescence intensity measured at 0. FEM strain map (1xx ) of the structure shown in b. At the laser powers used. respectively (Supplementary Section S4).0/×2.7 0.1038/NPHOTON.5 3.0 2 εxx (%) X (μm) 1 Detector cutoff Reﬂection 0 −2 0 0. An infrared resolved photoluminescence spectrum of the bridge with 3.7 Photon energy (eV) 0.8 Bandgap (eV) c Photoluminescence intensity (a. we found that a maximal redshift of 20–25 meV occurs due to heating of the constriction. a longitudinal strain of 3.2013.6% 0.9 2.0 Bulk Vup −6 0.5 0.8 DOI: 10. 4a. is well resolved.5 −6 −4 −2 0 Y (μm) 2 4 6 Figure 4 | mPL spectra measured at room temperature for different suspended constricted Ge structures and excitation positions. the difference between the conduction band energy minima at G and L valleys reduces. increasing the population of electrons at the G point and therefore the emission intensity.5/×1. NATURE PHOTONICS b 0.4 0. The laser spot size was set to a diameter of 0.1% strain.4 1. a.0 1. photoluminescence intensity 1.1% and excitation in the centre of the constriction. The curves are offset for clarity.) 1.6 0. the FP resonances obstruct the identiﬁcation of the contribution from the several possible radiative transitions from the L and G conduction band-edge minima to the strain split valence bands at G. as 470 well as the rise in intensity with increased longitudinal strain.6 0.7% 0. Despite this obstruction.5 0.2 −0.
the L minimum and the G minimum.0 5. We have also shown that gain is expected for doped structures. the onset of photoluminescence emission taken from the infrared microscope occurs at a slightly lower energy because of the improved sensitivity of the infrared setup towards long wavelengths. given a starting substrate with lower defect density. The position of photoluminescence onset is independent of excitation position (marked as dashed circles in Fig. 4e).0 0. loss (x1.7% longitudinal NATURE PHOTONICS | VOL 7 | JUNE 2013 | www. Systematic strain characterizations were performed using Raman spectroscopy and FEM simulations. which is presently 3. respectively.nature. Energy-dependent gain and loss of an undoped (dashed lines) and an n-doped at 1 × 1019 cm23 (full lines) Ge structure with 3. In the case of structures with 3. respectively (Fig. and the strain. Although the maximum tensile strain of 3. Figure 4b presents mPL spectra measured at different excitation positions along the constriction of a Ge/SOI structure with 2. 44 and 100 mV for the upper valence band maximum (Vup).1% strain. b. loss (x1. can further enhance the optical gain. All rights reserved.0 0. However. .000 cm−1) εxx = 3. Photoluminescence measurements on those Ge structures show a peak energy shift of more than 200 meV as well as an increase in integrated emission intensity by a factor of more than 25 with respect to bulk Ge. There is good overall agreement between the two spectra with 3.0 1. increasing the breaking stress of the Ge layer. The inset of Fig. the simulated band edges of the relevant conduction and valence band.64 eV.000 cm−1) DOI: 10.5 mm away from the centre. For excitation 4.5 n-doping 1 × 1019 cm−3: Gain Loss Intrinsic: Gain Loss 0. we deduce that carriers drift efﬁciently to the centre as long as the excitation is in the region with large strain gradients between about y ¼ 1.21 GPa for Ge/SOI and 1.64 eV shown in Fig. which could be useful in constructing efﬁcient carrier injectors for a laser. Instead of doping.56 0.1% strain at an injection density of 3 × 1019 cm23 electrons and holes.1%. This indicates that the region with maximum strain (that is.NATURE PHOTONICS a 2. the constriction) dominates the emission. (Their average spacing is about two and ten times smaller than the length of the constriction for Ge/SOI and Ge/Si structures. and matches the value expected from the strain in the centre of the constriction. 4a. our approach opens the possibility to do so. Comparison between peak gain and valence band losses at the position of peak gain versus injection density for a doped sample with a strain of 3. considerably exceeds the loss for the doped sample (net gain of 460 cm21).1% strain. We propose that bipolar carrier diffusion occurs from the region of lower strain (larger bandgap) to regions of higher strain (smaller bandgap).5 mm and y ¼ 3.1% strain at an injection density of 3 × 1019 cm23. excitation intensity. The spectra taken with the infrared microscope also show a more pronounced FP oscillation due to its smaller numerical aperture and the reduced acceptance angle. To conclude. Figure 5b presents a comparison between the peak gain and the valence band losses at the position of peak gain versus the injection density for the doped sample with a strain of 3. is shown for comparison by the dashed line in Fig.0 0. the collection efﬁciencies are even higher due to the larger band bending. 5a). Each spectrum is normalized to its peak maximum.) Either reducing the dimensions of the structure.5 1.0 3. doped and undoped.67 ARTICLES b 3.68 0. as investigated here.0 2. We have also demonstrated bipolar carrier diffusion from regions with lower strain to those with higher strain. the carrier collection efﬁciency drops to below 15%.0 4.52 1. obtaining longitudinal strain up to 3.0 Injection density (×1019 cm−3) Figure 5 | Gain simulation for 3. Both behaviours suggest that diffusion is driven by the strain gradients.1%. 4b. with optical gain calculations using Fermi–Dirac statistics and Fermi’s golden rule.1038/NPHOTON. reaches transparency but exhibits no overall gain. Transparency is reached at the injection of 1 × 1019 cm23.5 3.72 0.1% n-doping: 1 × 1019 cm−3 2. which assumes the crossover to a direct bandgap in Ge at 4.60 0. our work has considerably simpliﬁed the challenge of making a direct-bandgap semiconductor from Ge.5 Gain Loss 1. In summary. where a higher rate of radiative recombinations occurs. Both will be the subject of further studies.13 GPa for Ge/Si.6% longitudinal strain. Figure 4b–d displays the dependence on the position of the excitation spot on the bridge measured from the centre ( y-coordinate) of the photoluminescence intensity measured at 0.64 Energy (eV) 0. By comparing the photoluminescence intensities at 0. subject to the availability 471 © 2013 Macmillan Publishers Limited.0 Gain.0 6. respectively. we have reported a top-down fabrication approach of highly strained Ge directly on Si and SOI without external stressors. but only at high injection densities.5 0. The differences in the band edges from the highly strained region ( y ¼ 0 mm) to the lower ( y ¼ 6 mm) are 270. The process is based on pure elastic deformation and can be fabricated using the tools in a silicon foundry. The reduction of yield strength compared to the maximal theoretical strength for perfect crystalline material19 is attributed to the relatively high threading dislocations density of 2 × 107 cm22.5 mm.1% for Ge/SOI and 1% for Ge/Si is insufﬁcient to achieve a direct bandgap. To discuss optical gain in our tensile-strained structures.1% strained Ge structures. We ﬁnd that the material gain.0 Gain. Figure 5a shows the energy-dependent absorption of an undoped and an n-doped (1 × 1019 cm23) Ge structure with 3.2013. 4d). or improving the growth process could solve this problem. we combined the model for calculating the band structure24. The shaded area represents the excess gain.0 7. 4a presents a comparison between the bandgap at the G point extracted from the onset of the photoluminescence measurements and those simulated using a band-structure calculation based on literature values of deformation potentials24. plotted with reverse sign (Fig. The absorption cross-sections of the intraband transition in the valence band—found to be the main loss channel—were extracted from our previous pump and probe experiments25 (Supplementary Section S5).0 1. The match between experiment and theory conﬁrms that the peak shifts observed in our measurements are predominantly attributed to strain. but an undoped sample. a.1% Injection: 3 × 1019 cm−3 εxx = 3.1%.5 2.com/naturephotonics strain.
Sun. P. H. L. 10. Z. J. Rieger. Author contributions All authors contributed extensively to the work presented in this Article. a biaxially tensile-strained Ge layer of typically 0. 6. Camacho-Aguilera. & Kelsall. or on bulk Si substrates with (001) surface orientation using low-energy plasma-enhanced chemical vapour deposition26. SiGe. Opt. 35..5 mm for the Ge/Si samples. M. Opt. E. R. V. 15. Phys. Nextnano: general purpose 3-D simulations. Acknowledgements The authors acknowledge hospitality from the IR beamline of the SLS. et al. Strain-induced changes to the electronic structure of germanium. 586–601 (2001). J. Dev.G. 2004).G.S.67 Methods Biaxially tensile-strained Ge was grown directly on SOI with a 340 nm device layer and 1 mm BOX thicknesses. & Kimerling.15% was obtained for both Ge/Si and Ge/SOI substrates. A. H. J. M. 4. Express 15. Electronic Device Architectures for the Nano-CMOS Era 524–559 (Pan Stanford. Express 20. 17. T. Solid State Electron.M. Pollak. M. et al. USA 108. 9203–9210 (2006). and strain. Appl.5%. E. Appl.. Rideau. et al. Optimum strain conﬁgurations for carrier injection in near infrared Ge lasers. 18893–18898 (2011). 1096 (2012). & Lochtefeld.F. High-performance interconnects: an integration overview. Niquet. Minamisawa. J. Condens. Prediction that uniaxial tension along 111 produces a direct band gap in germanium. performed the gain simulation. 527–534 (2010). et al. E... 245201 (2009). 679–681 (2010). The layer thickness was 1. 541–545 (2006). J. Onsite matrix elements of the tight-binding Hamiltonian of a strained crystal: application to silicon. An electrically pumped germanium laser.. Silicon optical modulators. III–V/silicon photonics for on-chip and intra-chip optical interconnects.S. Ruoff. Y. Correspondence and requests for materials should be addressed to H. Pinczuk. Scr. 26.S. Electron. M. & Cardona. germanium. a series of annealing cycles between 600 8C and 780 8C were performed in situ to reduce the density of threading dislocations.1038/NPHOTON.. NATURE PHOTONICS DOI: 10. W. 195802 (2012). Opt. How to convert group-IV semiconductors into light emitters. Phys. F.S. and J. doping. et al. U. 22. R. 12. M. H. 102. and R. Nature Photon.4 mm for the Ge/SOI samples and 2. Phys. 48. D. 5. M. Grimes. Nature Photon.J. 111. and Ge channels for high-mobility metal–oxide– semiconductor ﬁeld-effect transistors. 21. R. 3.. Nature Photon. 011101 (2005). IEEE Trans. Ghrib. Sci. Lett. B 79. Gardes. 156401 (2009). 13. A. R. 518–526 (2010). A. 16. Phys. 201104 (2012). 053106 (2012). R. Leuthold. & Michel. R. J. Jaouen. et al. and their alloys. 751–779 (2010).A.ARTICLES of a sufﬁciently strong Ge on Si substrate capable of resisting forces in the 4–5 GPa range. C. Reprints and permissions information is available online at www. L. wrote the draft manuscript. H. O. Solid State Commun. Owing to the mismatch between the thermal expansion coefﬁcients of Si and Ge and the hindered relaxation of the Ge layer upon cooling. Phys. Lee. 8. Isella. performed the optical measurements. P. 23. Lett. J.M. C. J. After growth at T ¼ 500 8C. Received 3 December 2012. 4. 4. Chips of the substrates were cleaned using acetone and isopropanol. Strain-induced shift of phonon modes in Si1–xGex alloys.J. 54.S. supervised the experiments and coordinated data interpretation. W. Ikonic. Ge-on-Si laser operating at room temperature. Rev. R. G. Havemann. performed the sample growth and XRD analysis. R. Rev. & Thomson. The authors also acknowledge support from the CARIPLO foundation regarding the project NANOGAP. 11316–11320 (2012).A. 535–544 (2010). 24. Vogl. R.nature. Semicond. All rights reserved. Majewski. 100. G.. Carroll. H. Control of tensile strain in germanium waveguides through silicon nitride layers. G. J.. Kimerling. High-performance Ge-on-Si photodetectors. Liu. 133–138 (1970).. F. H. F. 197–200 (1978). E.. The ﬁnal structures were released from the substrate by selective etching using 20 wt% KOH solution at 73 8C (controlled by a temperature controller) in the case of Ge/Si. R. Top-down fabricated silicon nanowires under tensile elastic strain up to 4. Phys.. Lett.. A. & Zhang. Strained Si. published online 14 April 2013 References 1. performed Raman measurements and carried out ﬁnite element simulations. 9. On the ultimate yield strength of solids.com/reprints. 109. & Blase. J. A. 25. Matter 24. J. A.. 130181). L.com/naturephotonics © 2013 Macmillan Publishers Limited. T. L. L. Fitzgerald. T49. J. & Freude. Aldaghri.. where some of the photoluminescence experiments were performed. X. 14. Zhang. and G. Phys. J. Fang. 057402 (2012).Fr. 19. Tahini. G. Liu. M. A micromachining-based technology for enhancing germanium light emission via tensile strain.. & Abstreiter.2013. A. 11. et al. Nature Photon. Michel. 398–405 (2012). 7.S. Part of this work is supported by the Swiss National Science Foundation (SNF project no. fabricated the Ge/Si samples. Rev. and J.S. Liu. et al. R. typically to 2 × 107 cm22. 11272–11277 (2007). 2137–2142 (2007). Competing ﬁnancial interests The authors declare no competing ﬁnancial interests. A.. the Ge layers were etched to the Si substrate (in the case of Ge/Si) or to the BOX (in the case of Ge/SOI). Bulsara. Deleonibus.. Mashanovich. 8. Effect of static uniaxial stress on the Raman spectrum of silicon. Jain. Sanchez-Perez. Opt. Laser Photon. G. 18. et al. Rev. Nature Commun.nature. 9.. Anastassakis. After dry etching of the Cr hard mask. 6. Currie. Pezzoli.I. Lett. Reed.G. J. Roelkens. 4. Schwingenschlo ¨ gl. Appl. 3. accepted 26 February 2013.S. Appl. W. Camacho-Aguilera. M. Tensile-strained. S. Tavernier. et al. Phys. Low-energy plasma-enhanced chemical vapor deposition for strained Si and Ge heterostructures and devices. Direct-bandgap light-emitting germanium in tensilely strained nanomembranes. H. 2. J... Additional information Supplementary information is available in the online version of the paper. Sci. Birner. n-type Ge as a gain medium for monolithic laser integration on Si. A. M. R.C. J. Phys. fabricated the Ge/SOI samples. S.. Koos. Mater. Constricted structures were patterned using polymethyl methacrylate (PMMA) resist and standard electron-beam lithography. W. A. 4.. Proc.. & Hutchby. Direct-gap gain and optical absorption in germanium correlated to the density of photoexcited carriers.F. T. 476–482 (1993). R.. A. F. et al. Nonlinear silicon photonics. and 50% hydroﬂuoric acid in the case of Ge/SOI.. Crespi. 49. 472 NATURE PHOTONICS | VOL 7 | JUNE 2013 | www. Y. . 20. and then covered with a thin Cr layer ( 24–60 nm). 97. Burstein. R. et al. & Dimoulas. and H. and G. IEEE 89. Proc. D. C. Electrically pumped hybrid AlGaInAs–silicon evanescent laser. Express 14. D. using SF6-based reactive ion etching (RIE). 1317–1323 (2004).. collaborated on the design of the experiment. Chroneos. Natl Acad. X.G. C. Proc.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.