Advances in Light Emission from Group-IV Alloys via Lattice

Sep 29, 2014 - LeRoy Eyring Center for Solid State Science, Arizona State University, Tempe, Arizona 85287-1704, United States. ABSTRACT: Intrinsic an...
0 downloads 0 Views 5MB Size
Download PDF
Article pubs.acs.org/cm

Advances in Light Emission from Group-IV Alloys via Lattice Engineering and n‑Type Doping Based on Custom-Designed Chemistries C. L. Senaratne,† J. D. Gallagher,‡ T. Aoki,§ J. Kouvetakis,*,† and J. Menéndez‡ †

Department of Chemistry and Biochemistry, Arizona State University, Tempe, Arizona 85287-1604, United States Department of Physics, Arizona State University, Tempe, Arizona 85287-1504, United States § LeRoy Eyring Center for Solid State Science, Arizona State University, Tempe, Arizona 85287-1704, United States ‡

ABSTRACT: Intrinsic and n-type Ge1−ySny alloys with y = 0.003−0.11 have been grown on Ge-buffered Si via reactions of Ge3H8 and SnD4 hydrides using UHVCVD techniques. The films exhibit large thicknesses (t > 600 nm), low dislocation densities (107/cm2), planar surfaces (AFM RMS ≈ 2 for intrinsic films) and mostly relaxed microstructures, making them suitable for subsequent characterization of the emission properties using photoluminescence (PL) spectroscopy. The PL spectra are acquired at room temperature and show tunable and distinct direct and indirect gap emission peaks versus composition. The peak intensity in a given sample is found to increase by exposing the layers to hydrogen plasma, indicating that surface passivation plays an important role in eliminating carrier recombination traps. The PL intensity is further increased by n-type doping with P/As atoms at levels 0.8− 7 × 1019 cm−3using P(GeH3)3, P(SiH3)3, and As(SiH3)3 precursors, indicating that desirable direct gap conditions can be approached even at relatively modest 6−8% Sn contents. The indirect and direct gap energies of the samples are then used to determine the direct gap cross over point at ∼9% Sn. Collectively the results in this paper show that strong light emission can be generated in this class of narrow gap alloys by adjusting the Sn content, subjecting the samples to post growth passivation treatments or doping the system n-type. The influence of precursor chemistry on the activation properties and optical behavior of the materials is explored with the objective to optimize the PL response near the indirect−direct gap threshold. New methods embodying environmentally safe conditions are designed to produce the dopant compounds in high purity for application in future generation working devices requiring enhanced IR optical performance.

■

INTRODUCTION Elemental Ge exhibits significant direct gap light emission in spite of the indirect character of its fundamental band gap. This is due to the sizable population of photoexcited electrons in the Γ valley of the conduction band at the Brillouin zone (BZ) center, resulting from the small energy difference between the minimum of this valley and the lowest-energy valley in the conduction band near the L point in the Brillouin zone. Alloys of Ge and Sn (Ge1−ySny) were recently introduced as a means of extending the optical capabilities of Ge in the group IV photonics arena.1−8 In Ge1−ySny the separation between the Γ and L valleys of the Ge-like band structure is further reduced as Sn is incorporated into the Ge lattice.9−11 This causes an increase in the carrier concentration in the Γ valley, thereby inducing more radiative transitions and thus significantly enhancing light emission relative to pure Ge.7,12 Ge1−ySny alloys are purported to undergo an indirect to direct gap crossover for y ≈ 0.06−0.10, indicating widespread applications in future generations of light emitting devices compatible with existing silicon technologies.9,10,13 In spite of recent advances in mapping the electronic structure of these materials, the unambiguous determination of the cross over composition point is very challenging.11,14−22 This is because it is difficult to separate the individual contributions from two © 2014 American Chemical Society

very close optical transitions if, as is the case in Ge1−ySny, one of them (the direct gap) has an oscillator strength about 2 orders of magnitude larger, and in addition the two transitions are broadened due to the alloy potential. The only technique in which the direct and indirect gaps produce comparable signals is room temperature photoluminescence spectroscopy (PL), and even in this case, measurements of the indirect gap are only possible for y ≤ 0.06. A complete and accurate mapping of the two transitions over a broader range of compositions may require a combination of PL with direct absorption measurements in very thick films and/or measurements of the electrical characteristics of intrinsic and doped samples as a function of temperature. A key requirement to accomplish this goal is the availability of undoped and doped alloy films of comparable quality and thickness over the entire y < 0.1 range of compositions. In this paper, we report on the synthesis and properties of such films grown on Ge-buffered Si substrates. We use X-ray diffraction, Rutherford backscattering, and highresolution electron microscopy to understand their structural properties, particularly the defect structure and elemental Received: August 13, 2014 Revised: September 22, 2014 Published: September 29, 2014 6033

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

lower energy shoulder because of indirect transitions. As the Sn content is increased the primary peak shifts to lower energies and its intensity increases dramatically near the expected cross over composition regime near 9%. At the same time the direct and indirect edges, clearly resolved here, exhibit a systematic decrease in separation with increasing Sn content as expected. This supports our interpretation of the cross over composition occurring in the vicinity of 8−10%. The PL intensity of Ge1−ySny alloys with compositions near the direct gap threshold (y = 0.05−0.09) is further enhanced by populating the Γ-valley in the band structure with electrons through heavy doping with P atoms at levels up to 3 × 1019 atoms cm−3 using the single sources P(GeH3)3 and P(SiH3)3 whose molecular structures are displayed in Figure 1. We show

distributions, and we use room-temperature PL to monitor their optical quality from the shape and strength of the spectra. For the case of n-type Ge1−ySny alloys, we report new doping strategies leading to full dopant incorporation and substantially enhanced PL signals. The initial observation of tunable PL was obtained from sample prototypes grown directly on Si substrates by CVD reactions of Ge2H6 with SnD4.8 In this case indirect and direct gap PL was only observed from layers with up to 3% Sn, while the corresponding signals from more concentrated analogs degraded significantly with increasing Sn content. This was likely because of the increased defectivity associated with the progressively lower growth temperatures needed to achieve higher Sn concentrations. The introduction of higher reactivity Ge3H8 in place of Ge2H6 enabled the growth of thicker layers on silicon with Sn contents up to 10% Sn, spanning the direct gap composition crossover.23 These samples were found to possess a significant volume fraction of optical quality material away for the defective interface region, leading to sizable PL emission over the entire range of compositions in the samples. The latter optical data were sufficient to allow an initial determination of the compositional dependence of the direct gaps versus Sn content. However, the quality of the PL signal was insufficient to unambiguously resolve both the direct and indirect signals needed to determine the crossover composition and establish the nature of the fundamental band gap. Furthermore, from a device perspective the low intensity and suboptimal quality of the PL signal for high Sn content alloys (4−10% Sn) indicated that the materials, as grown on Si wafers, are not suitable for the fabrication of laser diodes. The above initial studies provided the motivation to pursue development of an improved procedure that could unlock the full potential of GeSn as a bona fide optical semiconductor system. If the crystal can be grown on a suitable platform that precludes larger scale defect densities, the innate properties of the material could be fully exploited. An obvious choice is bulk Ge, but the wafers are brittle and expensive and comparatively difficult to process. An alternative approach to the growth of high optical quality GeSn samples is to use our recently developed virtual Ge substrate technology comprising of low defectively layers grown directly on Si wafers with large thicknesses up to several microns and beyond using the unimolecular chemical precursor Ge4H10 as the Ge source.24 The defect levels in the buffer layers are controllable as a function of thickness in the 106−107 /cm2 regime. In spite of these imperfections, these templates have the advantage of providing a mechanism of absorbing misfit strains in a way that it is not possible with bulk Ge. More importantly in the context of PL, these virtual substrates allow the formation of thick and largely relaxed epilayers with interface microstructures engineered to significantly reduce nonradiative recombination velocities relative to samples grown directly on Si. In this study, we show that intrinsic and n-type Ge1−ySny layers on Ge buffers allow systematic elucidation of the relationship between direct and indirect behavior needed to design direct gap materials for application in emitters and detectors. We present room temperature PL measurements of a large number of samples across the composition range and optimize the response by developing surface passivation procedures. To ensure consistency in the dependence of the PL signals on Sn content we compared samples of similar thickness in the range of 600 nm and above. The spectra show a strong signal corresponding to the direct gap emission and a

Figure 1. Equilibrium structures of P(GeH3)3 and P(SiH3)3 indicating close agreement between calculated and experimental (in parentheses) bonding parameters Reprinted from ref 25. Copyright 2013. American Chemical Society.

that the PL intensities of the films are approximately 10 times those of the intrinsic counterparts with similar thicknesses possessing the same Sn concentration. Experiments using the As(SiH3)3 compound yielded higher carrier densities of 7 × 1019/cm3 allowing additional performance gains in the PL response. The results suggest desirable direct gap conditions for possible laser applications can be achieved in n-type materials at relatively modest Sn contents below the direct gap threshold.

■

EXPERIMENTAL SECTION

The phosphorus doped Ge1−ySny alloys were grown using trigermane (Ge3H8) and deuterated stannane (SnD4) gaseous sources. The P dopant atoms were introduced using the trigermylphosphine (P(GeH3)3) compound which enables full activation of the incorporated donors at the ultralow temperatures 295−325 °C used for these depositions. For a typical experiment the required precursor mixtures were prepared by combining ∼40 Liter-Torr of the Ge3H8 with ∼0.16 Liter-Torr of P(GeH3)3 in a 3 Liter container. Thereafter variable amounts of SnD4 were added as required to attain the target alloy compositions. For example 5 to 7% Sn alloys were produced using 5 to 6.5 Liter-Torr of SnD4 in the mixture, respectively. In the final step each mixture was diluted with research grade H2 such that the Ge3H8 concentration was ∼2%. Similar precursor mixtures were prepared to investigate the possibility of using trisilylphosphine (P(SiH3)3), which would be a more facile source for introducing dopant P atoms, due to its higher thermal stability and superior volatility. Epilayers with Sn concentrations of y = 0.04−0.09 were produced in this manner. The corresponding arsenic compound, trisilylarsine (As(SiH3)3), was also used as a dopant source. In this case, it was found that by increasing the percentage of dopant precursor in the starting mixture, it was possible to achieve much higher doping levels in the epilayers. These results are discussed in detail in subsequent sections below. The synthesis of the above P and As delivery agents were carried out using newly designed solvent-free experiments involving reactions of the lithium phosphide (Li3P) and lithium arsenide (Li3As) with carbon-free SiH3Cl or GeH3Cl inorganic sources. The absence of organic solvents precludes contamination of the final product with 6034

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

diffraction experiments. The (004) ω rocking curves show full width at half-maximum (fwhm) values on the order of 0.18° for alloys with 5−7% Sn and slightly higher for the 8−9%Sn materials depending on the final thickness. Measurement of the (224) reciprocal space maps indicated a high degree of lattice relaxation in the as grown films in spite of the mismatch induced by increasing the Sn content (see Figure 2). Films with

deleterious C−H impurities and improves environmental compatibility further increasing the attractiveness of the new procedures. The Li3P and Li3As were obtained via reactions of the elemental materials using standard methods. Li3P was produced by combining 0.86 g Li ribon with 1.27 g of red phosphorus in a stainless steel tube equipped with a high vacuum valve. The mixture was heated at 450 °C for 5 h after which time the sample was cooled to room temperature. The product was ground under an inert atmosphere using a mortar and pestle and then reheated at 450 °C for 5 h to ensure complete equilibration. Li3As was prepared using a similar procedure involving direct reaction of Li ribbon (0.71 g) and arsenic powder (2.3 g) in stoichiometric amounts. The final products in both cases were used without further purification. The synthesis of P(GeH3)3 utilized the ClGeH3 starting material which was produced by standard chlorination of GeH4 with SnCl4. In a typical experiment the Li3P salt was placed in a Schlenk tube and was reacted with a 6% excess of ClGeH3 at −45 °C for 6 h. The yield obtained by this method was close to 20%. The syntheses of As(SiH3)3 and P(SiH3) were carried out by combining commercially available ClSiH3 directly with the corresponding lithium salt. The latter was placed in a stainless steel tube and ∼10% excess of ClSiH3 was condensed on top of it. The tube was then allowed to warm to room temperature, at which ClSiH3 is expected to be present as a liquid with a vapor pressure of 94.7 psi (6.5 atm). In the case of P(SiH3) the reaction was allowed to proceed for 16 h, and the final product was obtained with a yield of ∼50%. For As(SiH3)3 the reaction was allowed to proceed for 5 days producing similar yields to that of P(GeH3)3 above. In all of these syntheses, separation of the product from the starting materials was readily achieved by trap-to-trap distillation, and the purity of the material was confirmed by spectroscopic methods. We note that As(SiH3)3 was also produced in excess of 50% by reacting stoichiometric mixtures of Li3As and BrSiH3 at −78 °C for 20 h using diethyl ether as the solvent. The purity of the product appears to be comparable to that obtained by the solvent free method above. The film depositions were conducted on Ge buffers possessing an average thickness of 800 nm and an AFM RMS roughness of less than 1 nm. The substrates were high resisitivity p-type Si(100) wafers with 4 in. diameters. After growth these were subjected to basic characterizations to ensure optimal quality and then cleaved into quarter pieces for subsequent deposition. In a typical experiment the Ge buffers were first cleaned by dipping them in an aqueous 5% HF solution for 2 min, followed by a distilled water rinse to remove surface oxides formed at ambient conditions. The substrates were then dried under a flow of UHP nitrogen, and immediately loaded into the hot wall UHV CVD reactor where a further surface cleaning step was carried out by flowing 5% digermane (Ge2H6) and H2 for 5 min over the wafers. When the Ge2H6 flow was stopped the precursor mixture was admitted into the chamber through calibrated mass flow controllers, and the pressure inside the reaction zone was set to 200 mTorr using an electronically controlled throttle mechanism. The average deposition time period was 120 min, after which the precursor flow was stopped and the wafers were removed from the reactor and subjected to a thorough characterization in order to determine their materials and optical properties. Finally, we note that the intrinsic samples produced for PL measurements presented in this study were grown following the same protocols as for the n-type analogs with the same composition including deposition temperature and pressure as well as formulation of the various Ge3H8/SnD4 coreactant stock mixtures employed in the experiments. The film structural compositional and morphological characterizations described below are also common to both intrinsic and doped materials.

Figure 2. (Top) 224 reciprocal space maps of Ge0.94Sn0.06/Ge sample. The in plane and vertical lattice parameters are determined to be 5.6984 and 5.7113 Å, respectively indicating compressive strain of 0.1292%. The buffer layer exhibits a slight tensile strain of 0.1291% due to the thermal expansion mismatch with the Si wafer. (bottom) AFM image of the same film exhibits cross hatch patterns arising from dislocation patterns near the interface penetrating through to the free surface.

Sn composition of ∼2−3% were nearly relaxed as grown on Ge buffers. Films with higher concentrations in the 4−10% range exhibited residual compressive strains from 0.12% to 0.28% as grown. The strains in the more concentrated samples (6−10%) were further reduced by rapid thermal annealing down to 0.12% for the 6−7% material corresponding to 80% relaxation. The cubic lattice constants were obtained and the Sn contents were determined from them using the compositional dependence relationship from ref 26. The results are in excellent agreement with the values obtained from RBS analysis. The active carrier concentrations in the alloys were determined using spectroscopic ellipsometry yielding values of 8 × 1018−7 × 1019 cm−3 in all of the samples produced in this study. Nomarski optical microscopy and atomic force microscopy (AFM) were used to determine the surface topology. In common with intrinsic Ge1−ySny/Ge/Si epilayers, the doped samples exhibit cross-hatch patterns on the surface arising from

■

RESULTS AND DISCUSSION Material Properties of Ge1−ySny Samples. Rutherford backscattering (RBS) was used to determine the thickness and Sn content of the doped samples. Channeling experiments were also carried out, the results of which demonstrate a high degree of epitaxial alignment of the epilayers, as well as the complete substitutionality of the Sn atoms within the Ge lattice. The RBS results are further corroborated by high-resolution X-ray 6035

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

the defects generated at the interface (see Figure 2). The rootmean-square (RMS) roughness as determined by AFM was ∼3−4 nm. Further structural characterizations were performed by cross sectional transmission electron microscopy using a JEM −4000 EX microscope operated at 400 kV. The low magnification images revealed single phase monocrystalline epilayers with uniform morphologies and planar surfaces. Figure 3 shows

Figure 4. STEM images of the n-type Ge0.936Sn0.064 film grown on Ge showing defects present in this sample. (a) Enlarged bright field view of the interface showing stacking fault patterns commonly found in this class of materials. (b) An individual defect is highlighted by red box illustrating rotated dimer column along {111} direction. (c) Dark contrast in the center of the image indicates the location of the 60° partial dislocation. (d, e) Inverse FFT images of selected {111} lattice planes in the vicinity of the 60° dislocation is used to identify in this case the defect type denoted by the red circle.

Figure 3. XTEM micrograph of n-type Ge0.936Sn0.064 film grown upon a Ge buffered Si substrate at 315 °C. Inset is a high resolution STEM bright field image of the film/buffer interface region showing the position of the interface marked by arrow and the location of a pair of stacking faults separated by 15 nm. The doping concentration in this sample is 3 × 1019/cm3 as determined by spectroscopic ellipsometry.

the residual stress induced by the lattice mismatch. In addition to stacking faults, the STEM images also reveal 60° dislocations randomly arranged throughout the heterojunction. The dark contrast in panel (c) indicates the location of such a defect which was identified by first calculating a Fourier transform (FFT) of the STEM image, and second, applying masks to only two specific {111} reflection, for example, 11̅ 1 and 111̅ ,̅ filtering out everything else, and finally calculating reverse Fourier transform of a masked FFT. The process was repeated to another set of {111} reflections (11̅1 and 1̅11̅) and then their reverse FFT was calculated. The two reverse FFT images are displayed in panels d and e and show that only one of them (e) contains an extra 111 plane terminated at a point marked by the red circle indicating the presence of a 60 degree perfect dislocation. Finally the Ge0.936Sn0.064 sample was characterized by “element-selective” mapping using STEM and electron energy loss spectroscopy (EELS) to investigate the distribution of the constituent atoms in the lattice at the subnanometer scale. The atomic arrangements in these materials are of particular relevance to the PL studies in this paper, since deviations from alloy randomness may have significant implications on the electronic structure of the crystals. The EELS spectra were collected with spot size 0.13 nm in aberration corrected STEM HAADF (high angle annular dark field) mode using a GATAN Enfinium spectrometer. Figure 5a is a high resolution STEMHAADF image showing projections of atomic columns displayed as pairs of bright spots. These correspond to dimers or dumbbells comprising Ge and Sn aligned along the growth direction. The region of the sample analyzed by EELS is identified by the square box with dimensions of 3 × 3 nm2 in the lateral direction. The thickness of the specimen is 40 nm as determined by the low loss spectra. The EELS spectra in all cases shows peaks corresponding to Ge (L) and Sn (M)

representative micrographs of a Ge0.936Sn0.064 film (t ≈ 600 nm) grown upon a Ge buffer layer (t = 780 nm) at 315 °C. The STEM (scanning transmission electron microscopy) bright field image shows signs of defects and strain fields confined to the interface region. The bulk film exhibits a uniform phase contrast with no threading defects visible within the field of view of several microns in the lateral direction. This observation indicates that the dislocation densities propagating through the film are relatively low and that the misfit strain is mostly compensated at the interface. Hence, we conducted a detailed high resolution analysis to identify the type and distribution of dislocations generated under the low temperature process conditions used in this study. The experiments were performed on JEOL ARM 200F microscope equipped with probe aberration corrector. STEM bright field images were acquired using large collections angles up to 22 mrad and representative data are shown in Figures 3 and 4. The top panel in the latter illustrates an extended view of the interface region indicating the location of stacking faults shown as dark contrast lines along the growth direction. The features are separated from one another by distances ranging from 42 to 15 nm as illustrated in the high resolution image in the inset of Figure 3. These defects are randomly distributed at the GeSn/Ge interface and they penetrate down a short distance into the buffer layers. An enlarged view of an individual stacking fault (panel b) shows that the disruption of the stacking sequence of the 111 planes is caused by rotation of the dimer columns in [110] projection. This imperfection is likely spawned by localized strain fields found in the vicinity of these defects as a means of minimizing 6036

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

Figure 5. STEM and element selective EELS mapping show random Sn substitution in diamond lattice. (a) STEM image of n-type Ge0.936Sn0.064 film identifying the region analyzed by EELS; (b) Ge map created from the L edge showing dimer columns in 110 projection; (c) Sn map generated from the M edge; (d) hybrid Ge and Sn map.

ionization edges at 1217 and 483 eV, respectively. These were then used to generate atomic maps for the Ge (green) and Sn (red) columns as shown in Figure 5b and c, respectively. The Ge map shows distinct Ge−Ge dimer rows of the host lattice. The Sn maps exhibit similar features corresponding to projected bonding sites. The color overlay in Figure 5d of the Ge and Sn maps illustrates a uniform distribution of the green and red spots down each column indicating that both atoms occupy the same lattice and that they are randomly located along [110] projected columns. Collectively the XTEM and EELS analysis results support the notion that the Ge buffers provide a structurally compatible low energy platform that serve as compliant templates to reduce the initial lattice mismatch with Si wafers making it possible to integrate high Sn content alloys with large thickness, low concentrations of threading defects and random alloy structures devoid of interstitials and precipitates as required for meaningful investigation of the optical properties. Photoluminescence Measurements and Band Gap Determination of Intrinsic Alloys. The PL spectra of all samples were collected at room temperature using 200 mW power generated from a 980 nm laser, focused to a spot size of ∼20 μm. The emitted light was collected with a Horiba 140 mm f/3.9 Czerny-Turner micro-HR spectrometer. The system response was calibrated using a 10 W tungsten-halogen lamp. Long-pass filters were used to block the PL signal from the Si wafer at 1960 nm corresponding to second-order diffraction from the grating used in the experiment. The spectra of most samples were recorded using a liquid-nitrogen (LN)-cooled InGaAs detector with a wavelength range of 1300−2300 nm (0.95−0.54 eV). However, the spectra of n-type analogs with Sn concentrations greater than 6% could not be fully resolved using this detector. In this case we employed a thermoelectrically (TE) cooled PbS photodetector with extended detection range down to 2800 nm, allowing unambiguous collection of spectra for samples with composition near the 9−10% Sn regime. Figure 6 shows PL spectra from representative alloys with a common thickness of ∼600 nm and varying Sn contents from 0.3% to 9.1% Sn. These spectra were acquired mostly with the LN-cooled InGaAs detector as evidenced by the sharp cutoff

Figure 6. Room temperature PL plots vs Sn fraction for intrinsic asgrown Ge1−ySny (y = 0.003−0.09). Main peak is due to direct recombination (E0) and the weak shoulder is attributed to indirect emission (Eind). Peak energies redshift and intensities increase as a function of Sn content. Inset is a schematic of Ge1−ySny band structure showing the two transitions measured in the PL spectra and the band gap narrowing in relation to elemental Ge.

below 0.55 eV in the figure. The PL plots for y = 0.003−0.05 show a strong main peak corresponding to direct gap emission and a weak lower energy shoulder which is assigned to indirect transitions from the L minimum of the conduction band. The two peaks appear to merge into a single broad peak in the spectra of the more concentrated samples (y > 0.06) because of the decreasing separation of the direct and indirect edges with increasing Sn content. The trends here illustrate a shift of the PL maximum to lower energies as a function of Sn content and a sharp increase in the signal intensity as the indirect−direct crossover point is approached. The direct and indirect gap energies are extracted by fitting the PL peaks using methods described for prior work in refs 27−29 for SiGeSn analogs. The fitting procedure corrects for strain effects (which are rather small due to the large strain relaxation) and yields the band gap values corresponding to unstrained films. Figure 7 presents an example of the spectral fits for a sample with composition Ge0.97Sn0.03. The Ge buffer layer peak is also visible in this case at ∼0.8 eV. The inset in the Figure illustrates the compositional dependence of the band gap energies for samples with Sn content up to 11% which exhibited indirect and direct band gap peaks. The lines in the figure represent quadratic fits (see equations below) of the E0 and Eind for nearly 50 Ge1−ySny samples with compositions evenly spread across the entire range y = 0−0.11). E0(y) = E0Ge(1 − y) + E0Sny − b0y(1 − y) E ind(y) =

Ge E ind (1

− y) +

Sn E ind y

− bindy(1 − y)

(1) (2)

In the above expressions, the E0 and Eind for y = 0 and y = 1 are taken as equal to those of elemental Ge and α-Sn, corrected for 6037

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

Figure 8 compares the spectra of an intrinsic Ge0.96Sn0.06 alloy (gray curve) and a plasma-treated film of the same

Figure 7. PL spectrum of a Ge0.97Sn0.03 alloy showing three features including the direct gap the indirect gap of the epilayer, as well as direct gap of the Ge buffer. The dotted/solid/dashed lines represent data fits of the alloy epilayer and Ge buffer band gaps. The fits are performed using a combination of Gaussian for the indirect gaps and EMG (exponentially modified Gaussian) procedures for the direct gaps. Inset is a plot of the compositional dependence of the direct (solid line) and indirect (dotted line) band gap energies indicating a cross over point at ∼9% Sn.

Figure 8. PL plots of intrinsic, surface passivated, and phosphorus doped (as grown) 6% Sn alloy showing the trends in optimizing the peak intensity in the Ge0.94Sn0.06 alloy. The intensity of the n-type material is ∼10 times higher than that for the as grown intrinsic counterpart representing a significant improvement in sample performance.

temperature dependence. The fitting function then contains the bowing coefficients b0 and bind as its only adjustable parameters. Sn Ge Using EGe 0 = 0.796 eV, E0 = −0.413 eV, Eind = 0.655 eV, and Sn Eind = −0.035 eV (ref 27), we obtain bowing parameters b0 = 2.26 ± 0.03 eV and bind = 1.06 ± 0.09 eV. The compositional dependence lines from the intrinsic samples intersect in the vicinity of 8−9% Sn representing the cross over point composition yc from indirect to direct gap semiconductor for bulk like Ge1−ySny. This energy range is significantly lower than predicted using virtual crystal approximation theory (yc = 0.2) but much closer to recent calculations using supercells to simulate the alloy (4.5−6%). Finally we note that the error to determine the peak position for the direct gap is ∼0.1 meV across the entire composition range. The error in the position of the indirect gap for alloy concentrations up to 3% Sn is 0.1 meV, however this value becomes greater as the Sn concentration is increased since the peak features begin to overlap near the indirect to direct gap crossover. The error in the determination of the indirect gap peak position then increases up to 1 meV at 6% Sn. However, the separation between the direct and indirect gap energies at 6% Sn is on the order of 20−30 meV and so the error in the peak determination is small compared to the peak separation across the alloy range for which both direct and indirect gaps can be unambiguously determined. As indicated in the introduction section the PL intensities of the above Ge1−ySny alloys can be enhanced by passivating the surface of the as-grown intrinsic films with H atoms. In this case the samples were subjected to inductively coupled plasma generated at H2 pressure of 90 mTorr by applying RF radiation of 300 W in the coil region above the sample and 50 W in the platen region. To avoid extensive etching of the bulk crystal under these conditions, the exposure time of the wafer to the plasma was limited to 5 min which appeared sufficient in most cases to produce the desired outcome of increasing the PL signal intensity relative to the as grown counterparts.

material (black curve) indicating a 4−5 fold intensity increase. Since the PL intensity is approximately quadratic in the photoexcited charge density nex, whereas nex is approximately proportional to the recombination lifetime τ, the results in Figure 8 imply that upon passivation τ increased by a factor of 2. This in turn implies that the recombination lifetime has increased by a factor of 2. The recombination lifetime has an interface, a bulk, and a surface contribution. If we tentatively neglect the bulk contribution (which is in the millisecond range in pure Ge) we can write the effective recombination lifetime, following Mathiessen’s rule, as τ−1 = (sint + ssur)/L, where sint and ssur is the interface (surface) recombination velocity and L the film thickness. Surface passivation will reduce ssur but leave sint intact, so that if sint ≫ ssur, the effect of surface passivation should be negligible. Ge films grown on Si can be close to this limit, because the interface recombination velocity can be as high as sint =4000 cm/s, whereas the surface recombination velocity is approximately ssur = 140 cm/s.30,31 On the other hand, the observation of a significant PL intensity increase in our samples implies that for Ge1−ySny on Ge-buffered Si sint < s0sur, where s0sur is the surface recombination velocity prior to passivation. This means sint ∼ 100 cm/s. An alternative way to estimate sint is by comparing the PL intensity from unpassivated samples grown on Ge-buffered Si with samples grown directly on Si substrates. Here we observe a 10-fold increase in intensity for the former, which would imply sint ∼4000/√10 cm/s ∼ 1300 cm/s for the Ge1−ySny/Ge interface. The discrepancy between the two estimates of sint for the Ge1−ySny/Ge interface might be expected given their crude nature, but both point to a significantly less defected Ge1−ySny/Ge interface compared with the standard Ge1−ySny/Si interface. Photoluminescence Measurements and Band Gap Determination of n-Type Alloys. The PL intensity of the 6038

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

above intrinsic samples can be further enhanced by heavy doping the material with donor atoms. Also compared in Figure 8 is the spectrum of an n-type 6% Sn alloy containing 3 × 1019/ cm3 carriers grown and characterized as described in the previous sections. This sample exhibits an order of magnitude higher PL intensity (dotted curve) relative to that of the corresponding intrinsic specimen (gray curve). In addition there is a slight redshift of the doped sample peak maximum due to the renormalization of the band gap induced by the significant doping level. Figure 9 shows PL peaks of several n-

Figure 9. Normalized PL spectra of n-type Ge1−ySny (n ≈ 1−3 × 1019 cm−3) as-grown samples with y = 0.04, 0.07, and 0.09 recorded at room temperature using a PbS detector. The latter enables full spectral resolution of the PL peaks at this range of Sn compositions and P doping levels allowing an unambiguous determination of the direct band gap energies.

Figure 10. Direct and indirect band gaps extracted from fits of the PL spectra of phosphorus doped alloys. The color code indicates the carrier density in these materials. The solid lines are fits of the direct and indirect gaps for intrinsic samples described in Figure 6. The trends reveal a systematic redshift of the emission energies between doped and intrinsic materials with same compositions. This outcome is attributed to band gap renormalization effects due to the phosphorus doping. The band gap renormalization energies is similar (in Ge based materials) for both direct and indirect transitions in the doped samples therefore direct crossover point should not be strongly affected by doping.

type alloys containing 4.5%, 7%, and 9% Sn and doped with P atoms at ∼3 × 1019 cm−3 active carrier concentrations. The maxima of all curves have been normalized to facilitate comparison between films with different thicknesses. We note that the signal-to-noise ratio in these spectra is inferior to that of Figure 6 because the data were collected using the PbS detector. In spite of this limitation, the peaks generated under these conditions are distinct, well-resolved, and by far more intense relative to intrinsic analogs (not shown) measured using the same protocols. These observations further validate the point that n-type GeSn alloys possess enhanced emission capabilities and these materials are uniquely suited to be employed as requisite active components for the manufacturing of light emitters based on GeSn. The band gap energies from n-type Ge1−ySny samples are extracted from PL spectra measured using both the InGaAs and PbS detectors following the fitting procedure applied previously for the intrinsic analogs. The direct and indirect gaps (circles and squares, respectively) are plotted in Figure 10 and compared with the corresponding gaps of the intrinsic samples shown in Figure 7 (inset). The compositions of all films described in the plots straddle the range for practical band gap engineering beyond Ge into the mid-IR from 0.8 eV down to 0.45 eV at 11% Sn. The active carrier concentrations are found to be in range of 0.8−3 × 1019/cm3 as indicated by the color coded bar in the inset of the figure. It is apparent from the plots that both the direct and indirect band gap energies are lower than those of intrinsic alloys with the same Sn content. This renormalization effect is due to the incorporation of phosphorus donor atoms in the lattice. The redshift trend is clearly apparent for the direct gaps in the plots and the renormalization energy seems to be independent of composi-

tion over the entire range up to 9.5% Sn. In the case of the indirect gaps there is significant noise in the data and it is thus difficult to determine whether the redshift energy value is the same for all compositions. We note that the samples described in Figure 10 are exclusively doped using the P(GeH3)3 compound as the source of P atoms to obtain the desired carries densities up to 3 × 1019/cm3, well within the range needed to enhance the PL signal relative to intrinsic samples. In this study we have expanded the single source doping strategy to study the use of the analogous -SiH3 precursors P(SiH3)3 and As(SiH3)3 as lowtemperature high-efficiency delivery agents of donor atoms into the group IV lattice. As indicated in previous sections the P(SiH3)3 compound is potentially a more practical CVD source than P(GeH3)3 for scalable semiconductor processing due to its superior volatility (22 Torr vs 2 Torr at 22 °C) and enhanced thermal stability. We see no indication of decomposition or degradation of the molecule when it is stored under inert conditions for several months at room temperature. In the case of P(SiH3)3, we produced Ge1−ySny samples with y ≈ 0.04−0.08 exhibiting maximum doping concentration of 3 × 1019/cm3 which is similar to that found in films produced by P(GeH3)3 within the uncertainty of the measurement. Furthermore, the PL energies and emission intensities of the P(SiH3)3 samples were also found to be similar to the P(GeH3)3 counterparts indicating the presence of SiH3 or GeH3 functionalities in the dopant compound does not 6039

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

significantly influence the course of the growth process or the properties of the final product. For example samples with composition Ge0.96Sn0.04 produced using P(GeH3)3 and P(SiH3)3 exhibited nearly identical direct band gap energies at ∼0.62 eV and similar PL intensities in the range of ∼1400− 1600 μV. The film morphology, crystal quality and final thicknesses were also comparable in all films obtained using these approaches. The absolute P and Si concentration in films grown by P(SiH3)3 were measured by SIMS using reference standards and were found to be essentially equal at ∼2−3 × 1019 cm3. This value is consistent with the above-mentioned donor carrier densities measured by elliposmetry (3 × 1019/ cm3) indicating that the P atoms in the material are fully activated. On the other hand Ge0.96Sn0.04 samples grown using As(SiH3)3 exhibited significantly higher active carrier densities at 7 × 1019 cm3 than the P analogs with similar thickness. These densities were determined by the Hall method and corroborated using spectroscopic ellipsometry by modeling of the dielectric function which shows Drude like behavior in the infrared. Fits of the data as described in ref 32 yield the resistivity and the relaxation time from which carrier concentrations and mobility are derived by approximating the effective mass to be that of pure Ge (0.14). The SIMS measurements of these samples indicated that the absolute As content is fully activated within the error of the analysis and it is nearly equal to that of silicon as in the case of the P(SiH3)3 doped materials above. To account for the fact that the P/Si and As/Si ratios are nearly unity in all doped alloys we propose that the M(SiH3)3 (M = As,P) precursors decompose under the deposition conditions to eliminate volatile and relatively stable SiH4 or Si2H6 molecules, which are then pumped away and therefore do not participate in the growth, and highly reactive M−Si−H intermediates that remain adsorbed on the growth surface and eventually incorporate intact M−Si units into the films as shown by the eqs 3 and 4 below. M(SiH3)3 → Si 2H6 + MSi + 3/2H 2

(3)

M(SiH3)3 → 2SiH4 + MSi + 1/2H 2

(4)

Figure 11. Room temperature PL spectra of n-type as-grown Ge0.96Si0.04 produced by P(GeH3)3 and As(SiH3)3. The plots show a significant enhancement in the emission intensity with increasing carrier concentration.

comparable volatility to P(GeH3)3 but significantly enhanced thermal stability making it more viable for widespread CVD use. Again in this case liquid bulk samples do not decompose when kept under inert conditions at room temperature for extended time periods. These results indicate that both P(SiH3)3 and As(SiH3)3 may offer significant advantages over P(GeH3)3 for activation of Ge based semiconductors given that the codoping with Si atoms furnished by the highly reactive −SiH3 ligands remains low, so that the basic properties of the parent lattice do not change in any significant fashion.

■

SUMMARY The work describes fabrication of intrinsic and P and As-doped Ge1−ySny layers (y = 0.003−0.11) grown on Ge buffered Si wafers via UHV CVD reactions of Ge3H8, SnD4, P(GeH3)3, P(SiH3)3, and As(SiH3)3. The quality of these materials has been evaluated using several structural probes and roomtemperature photoluminescence. For applications in lightemitting diodes, we developed surface passivation methods to improve the emission intensity by subjecting the films to a hydrogen plasma source under optimal conditions. Further intensity enhancements were obtained through fabrication of ntype materials. Samples with n = 3 × 1019/cm3 produced using P(GeH3)3 and P(SiH3)3 exhibited a 10-fold increase in PL intensity relative to intrinsic analogs, while samples with higher donor levels n=7 × 1019/cm3 grown with As(SiH3)3 showed additional improvements as expected. Finally, the compositional dependence of the emission properties in the doped samples was studied in detail indicating that both the direct and indirect gap energies are lower than those of intrinsic alloys with the same Sn content. This is due to band gap renormalization effects induced by the dopant atoms.

Finally, we note the SIMS results for all samples show that total amount of the codopant Si atoms in the lattice is too small (in the range of 1019 cm−3) to affect in any meaningful manner the bulk structural properties (lattice contants) and optical behavior (band gap energies) of the Ge1−ySny alloys. In the case of the As(SiH3)3-doped films the PL spectra showed significantly higher peak intensities than those of the P(SiH3)3/P(GeH3)3 doped counterparts as expected because of the 2-fold increase of the free carriers. This is illustrated in Figure 11 which compares the spectra of two representative samples with the same average Ge0.96Si0.04 composition containing 3 × 1019/cm3 P and 7 × 1019/cm3 As donor atoms. The peak intensity of the latter is nearly double to that of the former while the corresponding peak emission energies are very similar. Since band filling effects push this maximum to higher energies, the result imply a larger band gap renormalization for the n = 7 × 1019/cm3 sample, as expected. The ability of the As(SiH3)3 to introduce larger amounts of dopant atoms in the Ge1−ySny lattice may be attributed to the higher reactivity of the delivery compound. Another factor may be the larger size of the As atom which is similar to that of the bulk Ge constituents, allowing a more facile incorporation into substitutional positions relative to the smaller P counterpart.33 Lastly, it is worth noting that the As(SiH3)3 molecule possesses

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This work was supported by the Air Force Office of Scientific Research under contracts DOD AFOSR FA9550-12-1-0208. We would like to thank Liying Jiang for her assistance with the 6040

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041

Chemistry of Materials

Article

(31) Swain, B. P.; Takato, H.; Liu, Z.; Sakata, I. Sol. Energy Mater. Sol. Cells 2011, 95 (1), 84. (32) Tolle, J.; D’Costa, V. R.; Xie, J.; Chizmeshya, A. V. G.; Menendez, J.; Kouvetakis, J. Solid State Electron. 2009, 95 (8), 816. (33) Kim, J.; Bedell, S. W.; Sadana, D. K. Appl. Phys. Lett. 2011, 98 (8), No. 082112.

electron microscopy. We acknowledge the use of John M. Cowley Center for High Resolution Electron Microscopy at Arizona State University.

■

REFERENCES

(1) Soref, R. Nat. Photonics 2012, 4, 495. (2) Temkin, R. J.; Connell, G. A. N.; Paul, W. Solid State Commun. 1972, 11 (11), 1591. (3) Kouvetakis, J.; Menendez, J.; Chizmeshya, A. V. G. Annu. Rev. Mater. Res. 2006, 36, 497. (4) Roucka, R.; Mathews, J.; Weng, C.; Beeler, R.; Tolle, J.; Menendez, J.; Kouvetakis, J. IEEE J. Quantum Electron. 2011, 47 (2), 213. (5) Oehme, M.; Kasper, E.; Schulze, J. ECS J. Solid State Sci. Technol. 2013, 2 (4), R76−R78. (6) Sun, G.; Soref, R. A.; Cheng, H. H. J. Appl. Phys. 2010, 108, 033107. (7) Mathews, J.; Roucka, R.; Xie, J. Q.; Yu, S. Q.; Menendez, J.; Kouvetakis, J. Appl. Phys. Lett. 2009, 95 (13), No. 133506. (8) Vincent, B.; Gencarelli, F.; Bender, H.; Merckling, C.; Douhard, B.; Petersen, D. H.; Hansen, O.; Henrichsen, H. H.; Meersschaut, J.; Vandervorst, W.; Heyns, M.; Loo, R.; Caymax, M. Appl. Phys. Lett. 2011, 99 (15), No. 152103. (9) D’Costa, V. R.; Cook, C. S.; Birdwell, A. G.; Littler, C. L.; Canonico, M.; Zollner, S.; Kouvetakis, J.; Menendez, J. Phys. Rev. B 2006, 73 (12), No. 125207. (10) Yin, W.-J.; Gong, X.-G.; Wei, S.-H. Phys. Rev. B 2008, 78 (16), No. 161203. (11) Tonkikh, A. A.; Eisenschmidt, C.; Talalaev, V. G.; Zakharov, N. D.; Schilling, J.; Schmidt, G.; Werner, P. Appl. Phys. Lett. 2013, 103 (3), No. 032106. (12) Chen, R.; Lin, H.; Huo, Y.; Hitzman, C.; Kamins, T. I.; Harris, J. S. Appl. Phys. Lett. 2011, 99 (18), No. 181125. (13) Wirths, S.; Ikonic, Z.; Tiedemann, A. T.; Hollander, B.; Stoica, T.; Mussler, G.; Breuer, U.; Hartmann, J. M.; Benedetti, A.; Chiussi, S.; Grutzmacher, D.; Mantl, S. Appl. Phys. Lett. 2013, 103, No. 192110. (14) He, G.; Atwater, H. Phys. Rev. Lett. 1997, 79, 1937. (15) Moontragoon, P.; Ikonić, Z.; Harrison, P. Semicond. Sci. Technol. 2007, 22, 742. (16) Jenkins, D. W.; Dow, J. D. Phys. Rev. B 1987, 36, 7994. (17) Perez Ladron de Guevara, H.; Rodrıguez, A. G.; NavarroContreras, H.; Vidal, M. A. Appl. Phys. Lett. 2004, 84, 4532. (18) Lu Low, K.; Yang, Y.; Han, G.; Fan, W.; Yeo, Y. J. Appl. Phys. 2012, 112, No. 103715. (19) Chibane, Y.; Ferhat, M. J. Appl. Phys. 2010, 107, No. 053512. (20) Gupta, S.; Magyari-Kope, B.; Nishi, Y.; Saraswat, K. C. J. Appl. Phys. 2013, 113, No. 073707. (21) Perez Ladron de Guevara, H.; Rodrıguez, A. G.; NavarroContreras, H.; Vidal, M. A. Appl. Phys. Lett. 2007, 91, No. 161909. (22) Eckhardt, C.; Hummer, K.; Kresse, G. Phys. Rev. B 2014, 89, No. 165201. (23) Grzybowski, G.; Jiang, L.; Beeler, R. T.; Smith, D. J.; Menéndez, J.; Kouvetakis, J. Appl. Phys. Lett. 2012, 101 (7), No. 072105. (24) Xu, C.; Beeler, R. T.; Jiang, L.; Chizmeshya, A. V. G.; Menendez, J.; Kouvetakis, J. Semicond. Sci. Technol. 2013, 28, No. 105001. (25) Sims, P. E.; Chizmeshya, A. V. G.; Jiang, L.; Beeler, R. T.; Poweleit, C. D.; Gallagher, J.; Smith, D. J.; Menéndez, J.; Kouvetakis, J. J. Am. Chem. Soc. 2013, 135 (33), 12388. (26) Beeler, R.; Roucka, R.; Chizmeshya, A. V. G.; Kouvetakis, J.; Menéndez, J. Phys. Rev. B 2011, 84, No. 035204. (27) Jiang, L.; Gallagher, J. D.; Senaratne, C. L.; Aoki, T.; Mathews, J.; Kouvetakis, J.; Menendez, J. arXiv: 1406.0448v1. (28) Gallagher, J. D.; Xu, C.; Jiang, L.; Kouvetakis, J.; Menéndez, J. Appl. Phys. Lett. 2013, 103, No. 202104. (29) Jiang, L.; Xu, C.; Gallagher, J. D.; Favaro, R.; Aoki, T.; Menéndez, J.; Kouvetakis, J. Chem. Mater. 2014, 26 (8), 2522. (30) Trita, A.; Cristiani, I.; Degiorgio, V.; Chrastina, D.; von Känel, H. Appl. Phys. Lett. 2007, 91 (4), No. 041112. 6041

dx.doi.org/10.1021/cm502988y | Chem. Mater. 2014, 26, 6033−6041