Investigation on the formation and propagation of

ECS Transactions, 64 (6) 895-901 (2014) 10.1149/06406.0895ecst ©The Electrochemical Society

Investigation on the formation and propagation of defects in GeSn thin films A. Mosleha,b, M. Benamarac, S.A. Ghetmiria,b, B. Conelya,b, M. Alherb,d, W. Dub, G. Sune, R. Sorefe J. Margetisf, J. Tollef, S.-Q. Yub, H. Naseemb a

microElectronics-Photonics Program, University of Arkansas, Fayetteville, AR 72701, USA b Electrical Engineering Department, University of Arkansas, Fayetteville, AR 72701, USA c Institute for Nanoscience and Engineering, University of Arkansas, Fayetteville, AR 72701, USA d Mechanical Engineering Department, University of Karbala, Karbala, Iraq e Department of Physics, University of Massachusetts Boston, Boston, Massachusetts 02125, USA f

ASM, 3440 E University Dr. Phoenix, AZ 85034

Defect formation and propagation in Ge1-xSnx/Ge thin films were investigated by using X-ray diffraction, transmission electron microscopy and scanning electron microscopy techniques. Samples with 0.9 to 7 % Sn mole fraction were deposited on a Si substrate with a Ge buffer layer using a chemical vapor deposition technique. The X-ray diffraction results show that the Ge1Transmission xSnxfilms are 46-100 % compressively strained. electron microscopy images show that symmetrical and asymmetrical Lomer dislocations are formed at GeSn/Ge interface, which stops propagation of defects parallel to the growth direction. Average threading dislocation density (TDD) in the Ge buffer layer is measured to be 9.71×108 cm-2. Adding Sn to Ge made the Ge1-xSnx films to be more flexible, which traps 80 % of TDD propagating in Ge buffer layer. Reduction of TDD to 3×107cm-2 indicates achievement of higher quality of Ge1-xSnx film.

Introduction Silicon photonics is rapidly growing because of the recent introduction of GeSn alloys as a group IV direct bandgap material. Viability of GeSn alloys for complementary metaloxide-semiconductor technology, as demonstrated recently, is another step towards commercializing this material for integrated circuits applications (1). However, ability to grow high quality materials is imperative for device applications. In order to compensate the high lattice mismatch between the Si substrate and the GeSn film different strain relaxed buffer layers are proposed such as such as Ge and SiGe (2). Difference in lattice constant of Si, Ge and Sn introduces a network of misfit dislocations to relax the biaxial strain in the film. Although majority of the misfit dislocations can be trapped, some will thread through the film causing threading dislocation, which decreases the performance of the final device (3). Current methods to increase trapped misfit dislocations at the

895

Downloaded on 2015-10-22 to IP 52.2.249.46 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

ECS Transactions, 64 (6) 895-901 (2014)

surface of Si/Ge are growth of graded Si1-xGex buffer layers with different Ge mole fractions (2) and multi-step growth of Ge (4). Plastic relaxation of higher lattice constant materials (SiGe, Ge, GeSn) on top of Si forces the layers to form 60o misfit dislocations. These dislocations are caused by slip of (110) and (111) planes in the diamond lattice structure, which has a screw components and will thread through the grown film. Growth of low temperature Ge or SiGe film, which is followed by the high temperature growth can be a solution to confine the threading dislocations (4). Formation of Lomer misfit dislocation can also prevent the propagation of threading dislocation into the top layer. The main reason behind this phenomenon is that the Burgers vector in Lomer misfit dislocation, for a (001) substrate, lies parallel to the interface. Therefore, the substrate and film lattices can match while no perpendicular component exists. In this case, the film is grown relaxed without introducing structural imperfection such as threading dislocations, which degrades the material quality. Theory of Lomer dislocation was explained by Mader et al. (5) in 1974, thereafter, different models were proposed including a five-atom ring and seven-atom ring without dangling bonds (6). These structural models are derived from high resolution electron microscope images along with image simulations. Available models show that Lomer dislocations can combine two lattices through symmetric and asymmetric loops (7), (8). In this paper, we report on the defect formation and propagation mechanism in Ge1films and its properties of trapping defects on the interface. Silicon is chosen as the substrate for the growth of Ge1-xSnx with different Sn mole fractions on a Ge buffer layer. The X-ray diffraction (XRD) characterization of the samples show that Ge1-xSnx films are compressively strained over strain relaxed Ge buffer layers. High resolution transmission electron microscopy (TEM) images display the relaxation mechanism of thin films through threading and Lomer dislocation. Germanium tin films exhibit lower density of defects, showing that they have trapped the propagating threading dislocations through the Ge films. Plan view TEM images reveal that relaxation happens through propagation of defects by forming cross-hatch patterns. Scanning electron microscope images verify the threading dislocation density (TDD) measurements by TEM.

xSnx

Experimental A single wafer reduced-pressure chemical vapor deposition (RPCVD) epitaxial deposition system (Epsilon®) has been adapted to grow films at ASM Co. The deposition system has a load lock chamber and a cold wall quartz chamber with a silicon carbide coated graphite susceptor heater. Growth temperature of Ge1-xSnx layers was kept below 450°C to be CMOS compatible. A relaxed buffer layer of Ge was grown on Si (001) as virtual substrate prior to the growth of Ge1-xSnx Alloys. Table I shows the composition, film thickness, critical thickness (by Matthew- Blansee and People-Bean methods (1)) and relaxation percentages of the grown samples. Tin concentration is measured using secondary ion mass spectroscopy (SIMS). High Resolution TEM (TITAN®) with an accelerating voltage of 300 kV is used to investigate crystal orientation, defects and thickness of the grown epi-layers. A Phillips X'pert PRO diffractometer is used to confirm the Sn mole fraction and measure the lattice constant and strain in the Ge1-xSnx films.

896

Downloaded on 2015-10-22 to IP 52.2.249.46 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

ECS Transactions, 64 (6) 895-901 (2014)

TABLE I. Composition, thickness, critical thickness and relaxation percentage of the Ge1-xSnx Films. Tin mole fraction is measured through secondary ion mass spectroscopy (SIMS). Critical thickness is calculated by Mathew-Blanskee (M-B) and People-Bean (P-B) methods Sn % (x) SIMS Ge1-xSnx thickness Critical thickness Relaxation (%) (nm) M-B, P-B (nm) 0.9 327 85, 6000 15 1.43 287 40, 3000 54 2.57 257 30, 850 16 2.7 173 28, 800 30 3.2 76 22, 500 16 3.2 128 22, 500 28 6a 49 12, 120 0 7 240 7, 80 52 a Concentration determined from Vegard calculation based on XRD results.

Results and discussion Formation and propagation of defect in the films is a direct consequence of lattice mismatch between the grown layer and the substrate layer. Initially, films could be grown defect free under full compressive strain but as soon as the layers start to relax the strain, defects form and propagate through the film. In order to measure the strain in the grown films, XRD 2θ-ω scan of the samples were performed from symmetric (004) plane (Fig. 1). The peak at 66° corresponds to Ge (004) buffer layer. Due to different Sn mole fractions, Ge1-xSnx layers have different diffracted angles from 66-64°. Obtained XRD data from (004) symmetric plane provides out-of-plane lattice constant of the layers, however, in order to calculate the strain in the films, the in-plane lattice constant is also needed. An asymmetric Reciprocal Space Mapping (RSM) form ( 2 2 4 ) plane provides the in-plane ( a ll ) lattice constant as well as the out-of-plane ( a ⊥ ) lattice constant which was previously measured by 2θ-ω scan. Knowing the elastic Poisson ratio (ν), the total lattice constant of Ge1-xSnx film is calculated by following relation (1): (a + 2allν ) a0GeSn = ⊥ (1 + 2ν ) [1] If the in-plane lattice constant of GeSn film is equal to bulk Ge ( a ⊥GeSn = a0Ge ) then the grown film is 0 % relaxed. The film would be 100% relaxed if the in-plane lattice constant of Ge1-xSnx equals the out-of plane one ( a GeSn = a llGeSn = a0GeSn ). Measured strain ⊥ of the films show that based on the Sn mole fraction and thickness the samples are relaxed from 0 to 54%. Relaxation of the films start at different thicknesses based on the lattice constant of the Ge1-xSnx layers. A higher Sn composition film would have a higher lattice constant and relaxation starts at lower thicknesses. The thickness beyond which the higher lattice constant layer cannot accommodate the strain, is critical thickness. Relaxation process starts by formation of defects such as misfit dislocation on the surface and will propagate through the film by threading dislocations.

897

Downloaded on 2015-10-22 to IP 52.2.249.46 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

ECS Transactions, 64 (6) 895-901 (2014)

Fig. 1. 2θ-ω scan of Ge1-xSnx/Ge films from (004) plane. These films are grown on Si substrate. Two most common methods to calculate critical thickness is by Matthews-Blakeslee (M-B) and People-Bean (P-B) (9). Critical thickness of Ge1-xSnx films on relaxed Ge are calculated for different Sn mole fractions in Ref (1). Comparison of P-B model with MB model shows that the latter is in better agreement with the experimental data achieved from RSMs. Large lattice mismatch between Si and Ge (4.2%) causes the Ge buffer layer to release the strain by formation of misfit dislocation at Ge/Si interface. Mathew-Blakeslee calculation for critical thickness of Ge growth on Si shows that this process should start after few nanometers of growth (10). Figure 2(a) shows the schematic diagram of the samples with Ge1-xSnx layer grown on Ge buffer layer on Si substrate. Formation of 60o misfit dislocations as a result of plastic relaxation of Ge on top of Si forces the (110) and (111) planes to slip in the diamond lattice structure. The screw component of misfit dislocation will thread through the grown film and form threading dislocations. A crosshatch pattern is observable as a result of formation of misfit dislocations on the interface of Si/Ge and Ge/GeSn, however not all the dislocations might propagate through the second layer. Figure 2(b) shows the cross-hatch pattern observed by plan view TEM of the Ge0.93Sn0.07 sample. The cross-hatch patterns formed at GeSn/Ge interface have overlapped with those of Ge/Si in the image. The denser pattern is attributed to the Ge/Si interface due to higher lattice mismatch. Figure 2 (c) delineates scanning electron microscopy image of decorated threading dislocations at the surface by using cold Schimmel solution (CrO3+HF) for 10 minutes. Density of pits (1.1×107 cm-2) is in good agreement with TDD measured from TEM micrographs (3×107 cm2) [1]. It also confirms that the quality of GeSn material is higher than Ge buffer layer (TDD: 9.71×108 cm-2) by an order of magnitude.

898

Downloaded on 2015-10-22 to IP 52.2.249.46 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

ECS Transactions, 64 (6) 895-901 (2014)

Fig. 2. (a) Schematic diagram of threading dislocation propagation (b) Plan view TEM image of Ge0.93Sn0.07/Ge/Si(c) SEM image of Ge0.93Sn0.07 film exposed to etch pit solution. Propagation of threading dislocations is studied in Ge buffer and GeSn layer. Higher density of treading dislocation is seen in all the Ge buffer layers. However, localization of threading dislocations is observed on the interface of Ge/GeSn interfaces. Figure 3 shows that the defects have not passed through the GeSn film at several locations. The TEM images in Fig. 3 shows the cross-section of Ge buffer layer as well as the Ge0.968Sn0.032 layer for two thicknesses of 76 nm [Fig. 3(a)] and 128 nm [Fig. 3(b)]. Thinner Ge0.968Sn0.032 film (76 nm) in Fig. 3(a) shows less relaxation than the thicker Ge0.968Sn0.032 film (128 nm) in Fig. 3(b), nonetheless, no cross-hatch pattern was observed on the surface of these two films. This indicates misfit dislocations are not propagated through the GeSn film.

Fig. 3. Transmission electron microscopy Ge.968Sn.032/Ge films that are grown onSi substrate. GeSn film has 76 nm thickness in (a) and 128 nm in (b). It can be seen that threading dislocations are locked on the at GeSn/Ge interface. Higher density of threading dislocations in the Ge buffer layer shows lower quality of the film, however, it has provided the opportunity to investigate defect propagation in the Ge1-xSnx film. The TEM image of Fig. 3(b) shows that the Ge1-xSnx layer can lock the threading dislocations that are being propagated through the film at the GeSn/Ge interface. Also the no threading dislocations show being formed at the Ge1-xSnx /Ge interface, nonetheless, some of the threading dislocations were able to pass through the Ge1-xSnx film. As a result of this phenomenon, the average density of threading dislocation in the Ge1-xSnx layer is far less than the Ge buffer layer. The measurements

899

Downloaded on 2015-10-22 to IP 52.2.249.46 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

ECS Transactions, 64 (6) 895-901 (2014)

show that the density of threading dislocations in the Ge buffer layer is 7.1×108 cm-2 while that of the Ge1-xSnx layer is 5×107 cm-2. This phenomenon is also seen in other TEM images of different Sn mole fraction films.

Fig. 4. (a) Transmission electon microscopy image of Ge.93Sn.07/Ge/Si film. (b) High resolution TEM image of misfit dislocations at the interface of Ge.93Sn.07/Ge film is depicted (c) Symmetrical Lomer dislocation is observed at the GeSn/Ge interface denoted by a white box in Fig. 4(b) (rotated 45o CCW). (d) Asymmetrical Lomer dislocation is obsereved at the GeSn/Ge interface denoted by a white box in Fig. 4(b) (rotated 45o CCW). (e) Is the model of Fig. 4(c). (f) Is the model of Fig. 4(d). Figure 4(a) shows the structure of Ge0.93Sn0.07 layer on Ge buffer. The thickness of this layer is 240 nm and Ge buffer is 760 nm. Critical thickness of such GeSn structure is calculated to be 7 nm by M-B model and 80 nm by P-B model. Therefore a 240 nm Ge0.93Sn0.07 layer has undergone relaxation process through formation of a network of misfit dislocation on the surface and propagation of threading dislocation throughout the film. Figure 4(b) shows misfit dislocation at two points. This is in agreement with the XRD analysis that shows 52 % relaxation and network of cross-hatch pattern that is seen in the plan view TEM image [Fig. 2(b)]. Measurement of the TDD in Ge buffer layer and GeSn film show that 80% of threading dislocations have been trapped at the interface of Ge/GeSn [Fig. 4(a)]. Density of threading dislocations in Ge1-xSnx films is measured to be as low as 3×107 cm-2, which indicates higher quality of the grown film with respect to the Ge buffer layer. Misfit dislocation can also form Lomer dislocations (11) on the surface by forming closed loops at the GeSn/Ge interface. Formation of such dislocation stops propagation of defects along the growth direction. Lomer misfit dislocation is a 90o misfit dislocation that does not have a screw component in contrast to the regular 60o misfit dislocation. Formation of his kind of misfit dislocation at the GeSn/Ge interface leads to less threading dislocations in the Ge1-xSnx films. Fig. 4(c) (rotated 45o CCW) shows the higher magnification of Lomer dislocations that has formed a symmetrical Lomer dislocation. Other kinds of Lomer misfit dislocations are also observable at the interface as it is shown in Fig. 4(d) (rotated 45o CCW) which is an asymmetrical Lomer dislocation. Different models are fit to the Lomer dislocations to show how the atoms are

900

Downloaded on 2015-10-22 to IP 52.2.249.46 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

ECS Transactions, 64 (6) 895-901 (2014)

placed to relieve the stress and keep the top surface defect free. The most recurrent Lomer dislocations are presented in Fig.4 (e, f). Figure 4 (e) shows a seven-atom-ring symmetrical and Fig. 4 (f) delineates an eleven-atom-ring asymmetrical Lomer dislocation.

Conclusion In conclusion, defect formation and propagation of Ge1-xSnx layers is studied through different material characterizations such as XRD, TEM, SEM and etch pit density measurement. Germanium tin films with different Sn mole fractions ranging from 0.9 % to 7 % were grown on strain relaxed Ge buffer layer on Si by using RPCVD technique. Investigation of defect formation and propagation in Ge1-xSnx films showed that Ge1-xSnx films confine propagation of misfit and threading dislocations that are produced in the Ge buffer layer. In addition, formation of symmetrical and asymmetrical Lomer misfit dislocations at the GeSn/Ge interface plays a big role in the reduction of formation of threading dislocations. The current results present GeSn as a unique candidate for universal compliant layer. Using such compliant layer can localize and trap dislocations the interface. Therefore, other lattice mismatched materials can be grown on top of Ge1xSnx films which lead to high quality film growth.

Acknowledgment This material is supported by the National Science Foundation under EPS-1003970. Drs. Soref and Sun appreciate support from AFOSR.

References

1. A. Mosleh, S. A. Ghetmiri, B. R. Conley, M. Hawkridge, M. Benamara, A. Nazzal, J. Tolle, S. Yu and H. A. Naseem, J Electron Mater., 43, 4 (2014). 2. E. Fitzgerald, Y. Xie, M. Green, D. Brasen, A. Kortan, J. Michel, Y. Mii and B. Weir, Appl.Phys.Lett., 59, 7 (1991). 3. L. M. Giovane, H. Luan, A. M. Agarwal and L. C. Kimerling, Appl.Phys.Lett., 78, 4 (2001). 4. J. Hartmann, J. Damlencourt, Y. Bogumilowicz, P. Holliger, G. Rolland and T. Billon, J.Cryst.Growth., 274, 1 (2005). 5. S. Mader, A. Blakeslee and J. Angilello, J.Appl.Phys., 45, 11 (1974). 6. J. Hornstra, Journal of Physics and Chemistry of Solids., 5, 1 (1958). 7. A. Bourret, J. Desseaux and A. Renault, Philos.Mag.A., 45, 1 (1982). 8. A. Nandedkar and J. Narayan, Philos.Mag.A., 61, 6 (1990). 9. Y. H. Jo, I. Jung, C. S. Choi, I. Kim and H. M. Lee, Nanotechnology., 22, 22 (2011). 10. R. Beeler, R. Roucka, A. Chizmeshya, J. Kouvetakis and J. Menéndez, Physical Review B., 84, 3 (2011). 11. J. Kouvetakis and A. V. G. Chizmeshya, Journal of Materials Chemistry., 17, 17 (2007).

901

Downloaded on 2015-10-22 to IP 52.2.249.46 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).