Chemical reactions at CdS heterojunctions with CuInSe2

Chemical reactions at CdS heterojunctions with CuInSe2 Angel Aquino and Angus Rockett Citation: Journal of Vacuum Science & Technology A 31, 021202 (2013); doi: 10.1116/1.4775341 View online: http://dx.doi.org/10.1116/1.4775341 View Table of Contents: http://scitation.aip.org/content/avs/journal/jvsta/31/2?ver=pdfcov Published by the AVS: Science & Technology of Materials, Interfaces, and Processing Articles you may be interested in Chemical bath deposition of Zn(O,S) and CdS buffers: Influence of Cu(In,Ga)Se2 grain orientation Appl. Phys. Lett. 102, 051607 (2013); 10.1063/1.4788717 Electronic effects of Cd on the formation of the CdS/CuInS2 heterojunction J. Vac. Sci. Technol. A 30, 04D114 (2012); 10.1116/1.4721639 CuInS 2 – CdS heterojunction valence band offset measured with near-UV constant final state yield spectroscopy J. Appl. Phys. 106, 073712 (2009); 10.1063/1.3211918 Photoemission study and band alignment of the CuInSe 2 (001)/ CdS heterojunction Appl. Phys. Lett. 84, 3067 (2004); 10.1063/1.1712034 Direct evidence of Cd diffusion into Cu(In,Ga)Se 2 thin films during chemical-bath deposition process of CdS films Appl. Phys. Lett. 74, 2444 (1999); 10.1063/1.123875 Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 130.113.111.210 On: Fri, 19 Dec 2014 16:47:20 Chemical reactions at CdS heterojunctions with CuInSe2 Angel Aquino and Angus Rocketta) Department of Materials Science and Engineering, University of Illinois, 1304 West Green Street, Urbana, Illinois 61801 (Received 27 October 2012; accepted 21 December 2012; published 9 January 2013) The stability of the CdS/CuInSe2 (CIS) heterojunction is critical to understanding the projected lifetime of CIS devices and the effect of processing conditions on the nanoscale chemistry of the heterojunction. This article reports the results of annealing heterojunctions between CdS deposited by chemical bath deposition and single crystal and polycrystalline CIS films between 200 and 500  C for 10 to 150 min. No atomic movement was observed by secondary ion mass spectrometry at temperatures of 300  C and below. At 400  C even for the shortest time studied, Cu and In were found throughout the region initially consisting of CdS only and Cd was found to have moved into the CIS. In the polycrystal, annealing at 500  C resulted in movement of Cd throughout the CIS layer. No time dependence was observed in the 400 and 500  C anneals indicating that a reaction had occurred forming a compound that was in thermodynamic equilibrium with the remaining CIS. Diffusion turns on rapidly between 300 and 400  C, indicating a high activation energy for atomic movement (2.4 eV). The onset of diffusion is consistent with the onset of Cu diffusion in CIS. C 2013 American Vacuum Society. [http://dx.doi.org/10.1116/1.4775341] V I. INTRODUCTION Increasing concerns about global energy resources coupled with significant reductions in the cost of photovoltaics are driving renewed interest in this topic.1 The very high performance of copper chalcopyrite-based devices makes them particularly attractive candidates for detailed study.2,3 One of the issues that arises is the nature of the heterojunction in the devices and in particular those from which the best devices are produced: CdS/CuInSe2 (CIS) heterojunctions and similar junctions with the Ga-containing alloy Cu(In,Ga)Se2. Various groups have studied this interface and have found evidence of reactions and surface doping. The atomic-scale structure of CdS/CIS interfaces has been examined and Cd has been found to dope the surface of CuInSe2.4,5 However, no significant penetration of the chalcopyrite material by Cd has been demonstrated in nanoprobe transmission electron microscopy studies.6,7 Reactions at the heterojunction have been demonstrated by various groups on polycrystalline layers. Primarily this has related to polycrystalline thin films suitable for devices.8–11 In situ, hightemperature, x-ray diffraction experiments by Krishnan et al. show the formation of a CuCd2(In1xGax)Se4 phase at the interface.12 Some of these studies involved only partial electrolyte treatment of the sample surface.13 Studies of heterojunctions with single crystals have also been reported.14 Kazmerski suggested that a device based on a single crystal epitaxial layer was limited by diffusion at the heterojunction.15 Finally, the diffusion coefficient for Cd in CIS was studied by Rutherford backscattering examination of a metallic Cd layer on a CIS polycrystal and fitting of the results with an error function.16 None of these studies, except the latter and more recently Krishnan et al.,12 have examined time dependent reactions as a function of temperature. Time and temperature dependence are critical to the ultimate a) Electronic mail: [email protected] 021202-1 J. Vac. Sci. Technol. A 31(2), Mar/Apr 2013 stability of the heterojunction, particularly if it becomes necessary to heat the junction when forming a second junction, as may be necessary in some tandem device structures. It also has implications for ultimate lifetime and reliability of photovoltaic modules based on CIS. An excellent review of the stability of the interfaces in these devices may be found in Ref. 17. To further address the issue of how CdS interacts with CuInSe2, we have examined the reactions/diffusion of species across CdS/CuInSe2 heterojunctions. In the current study, a single crystal epitaxial layer of CIS grown on a GaAs substrate was used to avoid rapid diffusion along grain boundaries and to minimize surface topography effects that make depth profiling difficult.13 A polycrystalline CIS layer was also used for better comparison with previous results. II. EXPERIMENT A single crystal epitaxial layer of CuInSe2 was grown by a hybrid sputtering and evaporation method on a (100)oriented GaAs substrate. The deposition process has been described in detail elsewhere.18 For consistency in the measurements, a single epitaxial layer with a very smooth surface was chosen for this study. That sample was grown at 730  C for 50 min at a growth rate of 1 lm/h. The film composition as measured by energy dispersive spectroscopy was found to be 19.4% Cu, 24.1% In, 2.8% Ga, and 53.6% Se. For comparison, a polycrystalline layer also grown on (100) GaAs at 720  C for 50 min was also studied. The composition of that sample was 21.5% Cu, 24.7% In, 2.2% Ga, and 51.6% Se. Both the single crystal and the polycrystalline layers were 0.75 lm thick. The samples were coated with 25 nm of CdS by chemical bath deposition using a recipe described in Ref. 19. Following CdS deposition, the samples were overcoated with SiO2 to prevent loss of constituents during annealing. The SiO2 was deposited by sputter deposition for the epitaxial layer and by evaporation for the polycrystalline sample. 0734-2101/2013/31(2)/021202/6/$30.00 C 2013 American Vacuum Society V 021202-1 Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 130.113.111.210 On: Fri, 19 Dec 2014 16:47:20 021202-2 A. Aquino and A. Rockett: Chemical reactions at CdS heterojunctions with CuInSe2 The samples were cleaved into small pieces such that both the CIS and the CdS would be as similar as possible. Each piece was annealed face-up on a quartz boat in flowing N2 in a tube furnace. Annealing temperatures were 200, 300, 400, and 500  C for 10, 30, 60, 90, 120, and 150 min. Times were from insertion to removal from the furnace so there was some heat up time in each anneal that is included in the above times. This is not expected to have significantly affected the results reported here as temperature makes more difference than time at that temperature. The experimental process is shown schematically in Fig. 1. Following annealing, the samples were analyzed by secondary ion mass spectrometry (SIMS) in a Cameca IMS 5f instrument using Csþ primary ions incident on the sample at an energy of 5.5 keV and detecting positive secondary ions ejected from the sample surface. An electron gun was used to neutralize the surface of the sample when sputtering due to the insulating characteristics of SiO2 to avoid charging effects. The depth of each profile was determined by measurement of the crater in which the analysis took place by microprofilometry using a Sloan Dektak-3 instrument. III. RESULTS Typical composition depth profiles for the epitaxial and polycrystalline samples as deposited are shown in Fig. 2. The SiO2 layer is roughly 50 nm thick and appears in the first few data points near the surface of each sample. In this region, the other signals are changing rapidly due to the large chemical change in the matrix and the onset of sputtering. Therefore, the data in the region of the SiO2 should not be 021202-2 considered reliable. Note that the signals for Cu, In, and Se settle down to relatively constant values so the depth profiles are only shown near the surface where changes are occurring. The CdS layer is visible where the Cd and S show a strong rise, while the CIS layer shows Cu and Se. (In is not plotted to simplify the graphs but follows the trend in the Cu signal.) The rise in the Cu signal near the heterojunction is due to an ion yield and sputtering rate change passing from the Se to the S matrix. A similar bump occurs in the In signal for the same reason. The bump in the Cu and In intensities does not correctly reflect the real Cu and In contents. The roughness of the sample surface in the case of the polycrystal reduces the effect of this yield change so the bump is not as obvious. Note that the S, Cd, and Cu signals are all dropping together in this sample below the heterojunction, reflecting the penetration of Cd and S into the polycrystalline material.6 The drop in the Cu signal with the drop in S is the same effect as the bump in Cu in the single crystal sample profile but spread out over the range over which the S signal dies away. The Cu signal eventually settles down to a constant level approximately 0.4 lm into the film. In both samples, there is a region clear of significant Cu, In, and Se further defining the CdS layer. When the samples were annealed, the Cu, In, and Cd were found to move while the S and Se stayed in the same place. The In movement was similar to that of Cu so the remainder of this work focuses on the Cu, but the In profiles were the same. Evidence for the lack of movement of S and Se can be seen by plotting the S and Se depth profiles as a function of annealing temperature and time. The normalized FIG. 1. (Color online) Schematic of the experiments performed. The samples were CuInSe2 layers on GaAs substrates coated with 25 nm of CdS and 50 nm of SiO2. The sample was cleaved into pieces that were furnace annealed and analyzed by secondary ion mass spectrometry, x-ray diffraction, and other techniques. J. Vac. Sci. Technol. A, Vol. 31, No. 2, Mar/Apr 2013 Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 130.113.111.210 On: Fri, 19 Dec 2014 16:47:20 021202-3 A. Aquino and A. Rockett: Chemical reactions at CdS heterojunctions with CuInSe2 021202-3 FIG. 3. (Color online) S and Se signals across the heterojunction after annealing at four temperatures for 90 min each. Further annealing produced no additional movement of atoms. The polycrystalline sample showed similar behavior. FIG. 2. (Color online) Composition depth profiles near the sample surface for (a) the epitaxial and (b) the polycrystalline sample with CdS and SiO2 coating as deposited. S and Se profiles for the epitaxial film as a function of annealing conditions are shown in Fig. 3. Note that no significant movement of either species is observed even for annealing at 500  C for 90 min. The 500  C 150 min anneal also produced no movement of the two elements. Thus, the chalcogen profile at the heterojunction is highly stable. This also provides a good marker for the location of the original heterojunction. To characterize the movement of Cu and Cd, we need to deal with the ion yield change at the heterojunction. Several attempts were made to compensate for this change based on the S to Se ratio measured but no satisfactory result was obtained. We therefore use the depth profiles obtained for the as-deposited films (Fig. 2) as a reference and compare these to the annealed sample profiles to determine if movement occurred. There was no change in the profiles within the noise in the data for the 200 and 300  C annealed samples for any annealing time. Annealing at 400 and 500  C produced significant movement in both the single crystal and polycrystalline samples. Figure 4 shows the depth profiles from the epitaxial single crystal sample annealed at 200  C for 90 min and the sample annealed at 500  C for the same FIG. 4. (Color online) Composition depth profiles of the epitaxial film annealed (a) at 200  C for 90 min and (b) at 500  C for 90 min. The Se and S profiles are unchanged. The peak in the Cu signal has moved outward and Cd has moved inward during annealing at 500  C. No significant difference was observed between the sample heated to 200  C for 90 min and the asdeposited profile. JVST A - Vacuum, Surfaces, and Films Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 130.113.111.210 On: Fri, 19 Dec 2014 16:47:20 021202-4 A. Aquino and A. Rockett: Chemical reactions at CdS heterojunctions with CuInSe2 time. The profiles show movement of Cu and In from near the heterojunction in the CIS into the CdS and Cd movement in the opposite direction when annealed at 500  C but not at 200  C. Note that the signals are normalized so the background signal varies from one element to another in different profiles. The Cd, Cu, In, and Se signals had approached the values they would have throughout the remainder of the depth so the profiles are terminated at this depth to show the details of the front heterojunction. The movement of Cd, In, and Cu was found to be time independent for most of the annealing conditions studied. For temperatures of 300  C or below, no movement of any element was found above uncertainty in the data. The epitaxial sample behaved somewhat differently from the polycrystalline sample under higher temperature anneals. At 400  C for times longer than 10 min, Cu and In were found to diffuse completely through the CdS leaving a Cu-deficient layer near the surface of the CIS for both samples. In the 10 min anneal, this diffusion process was not complete for the epitaxial layer and further movement was found for longer anneals. The data are not sufficiently clear to extract a FIG. 5. (Color online) Composition depth profile for the polycrystalline sample annealed at (a) 300  C for 60 min and (b) 500  C for 60 min. The profile in (a) is the same, to within noise in the data, to the profile shown in Fig. 1(b). Note the diffusion of Cd through the entire CIS layer in the higher temperature anneal while no movement of S or Se is observed. 021202-4 reliable diffusivity and only this single data point (400  C, 10 min) showed any time dependence. Cd was found to move into the Cu/In deficient region of the CIS, replacing the lost Cu/In which had moved into the CdS layer. In the polycrystalline sample, outdiffusion of Cu/In was observed at 400  C for 10 min or longer, which did not change with time. At 500  C, more extensive outdiffusion of Cu and In was observed, again without time dependence. Plots of composition profiles for the polycrystalline sample are given in Fig. 5. Significant diffusion of Cd was not observed at 400  C in the polycrystalline sample. At 500  C, the Cd was found to have diffused rapidly into the bulk of the CIS layer and was distributed uniformly throughout the depth of the layer. The behavior of In matches that of Cu. The differences in behavior in the polycrystal relative to the single crystal are probably mostly associated with grain boundary diffusion coupled with the roughness of the surface, which has the effect of averaging the signal observed by SIMS over various depths. The absence of further atomic movement at 400 and 500  C for increasing anneal times indicates that the resulting Cu and Cd distributions [Fig. 4(b)] are in equilibrium for that temperature. The interpretation is that the observed distribution should represent a specific stable (CuIn)xCd1xSe phase. X-ray diffraction analysis of the annealed samples showed evidence of a new Cu0.75Cd0.5In0.75Se2 phase (PDF# 04-013-0597) for the sample annealed at 500  C for 90 min. The unannealed sample showed only the CuInSe2 and GaAs phases (the CdS peaks overlap with the CIS peaks so they cannot be distinguished separately). The XRD spectra are shown in Fig. 6. Energy dispersive x-ray spectroscopy (EDS) measurements were also performed on the annealed film. The composition after annealing was 16.8% Cu, 19.5% In, 2.2% Ga, 36.7% Se, 11.3% Cd, and 13.4% S. Assuming 80% of the Cd is now in the original CuInSe2 layer (which is FIG. 6. (Color online) X-ray diffraction data for two films, after a 200  C, 30 min anneal and after a 500  C, 120 min anneal. The peak positions for the major diffraction peaks for chalcopyrite CuInSe2 and Cu.75Cd.5In.75Se2 are marked. The major peaks at 27 and 66 are due to the GaAs substrate. The other peaks fit very well with those of CuInSe2 for the 200  C, 30 min annealed sample and with Cu0.75Cd0.5In0.75Se2 for the high temperature annealed sample. J. Vac. Sci. Technol. A, Vol. 31, No. 2, Mar/Apr 2013 Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 130.113.111.210 On: Fri, 19 Dec 2014 16:47:20 021202-5 A. Aquino and A. Rockett: Chemical reactions at CdS heterojunctions with CuInSe2 reasonable based on the Cd depth profile), that a respective amount of Cu and In are now in the original CdS layer, and normalizing these compositions, a compound with Cu0.65Cd0.49In0.79Ga0.12(SþSe)1.95 is obtained. These values obtained from EDS are in close agreement with the Cu0.75Cd0.5In0.75Se2 phase observed by x-ray diffraction. IV. DISCUSSION Many authors have referred to “diffusion” of Cd in CIS. From the current work we can estimate a diffusivity range for Cu in CdS containing sufficient cation vacancies (as it replaces moving Cu or In). At 300  C for 150 min, Cu or In would have had to have moved less than 10% of the thickness of the CdS (2.5 nm) to be undetectable by SIMS. At 400  C to have moved completely through the CdS, the diffusion distance would have had to have been significantly more than 25 nm in 10 min. Taking the diffusion distance, x, as x ¼ (Dt)1//2, where D is the diffusivity and t is the time, and solving for D, D ¼ x2/t, the maximum diffusivity at 300  C would have to be less than 1  1017 cm2/s. The minimum diffusivity at 400  C would have to be greater than 1.5  1014 cm2/s. To obtain this difference in diffusivity over 100  C would imply an activation energy for diffusion of Cu and In in CdS of greater than 2.4 eV. This magnitude of activation energy is common in diffusion of impurities in typical semiconductors by a substitutional mechanism.20 Because Cd diffusion turns on in nearly the same temperature range in both samples, the diffusion activation energy for Cd in CIS would have to be approximately the same in single and polycrystalline materials. The estimated values for diffusivities are much smaller and for diffusion activation energy much larger than the previous reported diffusivity for Cd in CIS.16 If the prior work were correct, we should have seen Cd diffusion into the CIS even at 200  C when annealed for 150 min with a detectable variation in observable diffusion of Cd into the CIS at 200  C for the various anneal times. We conclude from our results that diffusion is much slower than reported by Kumar. This is not necessarily inconsistent with the prior work as that experiment was very different. In the current work, we diffused Cd from CdS into single crystal and polycrystalline CIS and measured the results by SIMS. In the earlier work, the Cd was metallic, which would result in a strong reaction with the chalcopyrite CIS resulting in a large number of Se vacancies in the material and a very different point defect density. Furthermore, the Rutherford backscattering analysis has a variety of complications involved in the data interpretation. The 0.47 eV activation energy for diffusion in the earlier work is also more consistent with an interstitial rather than a substitutional diffusion mechanism. Therefore, we consider our results distinct from the prior work. In the current work, we find that the diffusion of Cd is coupled to the diffusion of Cu and In. All of the chalcogen compounds have a very strong tendency to valence compensate. Counterdiffusion of Cu and In and Cd would tend to result in a net exchange of one In and one Cu for two Cd atoms. This is precisely what we observe in the XRD data 021202-5 showing the final compound to be Cu0.75Cd0.5In0.75Se2. Our observation that the diffusion of Cd, Cu and In resulting in the formation of a (CuIn)xCd1xSe compound is consistent with the previous work by Heske et al. who found compound formation by photoelectron spectroscopy but did not investigate the time or temperature dependence of the reaction.8 The phase identified in this work is different from that identified by Krishnan et al.12 in that they observe a more Cd-rich phase. Based on the annealing conditions studied, here we conclude that no true diffusion [i.e., x ¼ (Dt)1/2 driven by a concentration gradient] of any significance of Cd, Cu, or In can occur under conditions of chemical bath deposition (maximum 75  C) or under conditions where air annealing of the heterojunction is performed by some groups (200  C). Any long-range transport of these atoms begins at much higher temperatures, around 400  C, and may result in the rapid formation of Cu0.75Cd0.5In0.75Se2 rather than extended transport of atoms into the bulk of the CIS. It is possible that a few atomic layers of atomic exchange may occur at 200  C but if that is the case, it is not detectable by any of the methods employed here. V. CONCLUSIONS We conclude that CdS/CIS interfaces are chemically stable at temperatures below 400  C, although short range transport of atoms over distances too small (monolayers) to measure by the techniques used here may be responsible for degradation of devices.17 The S and Se profiles across the heterojunction are highly stable even at 500  C for 150 min. However, above 400  C, Cu and In begin to diffuse out into the CdS layer. To valence compensate for the loss of these elements from the CIS, Cd moves into the vacated locations but does not diffuse effectively in the bulk of the CIS. However, Cd can diffuse relatively rapidly along grain boundaries in the polycrystalline material. The diffusion results in formation of a stable phase, Cu0.75Cd0.5In0.75Se2. ACKNOWLEDGMENTS The authors gratefully acknowledge the support of the Air Force Research Laboratory contract FA9453-08-C-0172, Space Vehicles Directorate, Kirtland AFB and was carried out in part in the Center for Microanalysis of Materials at the Frederick Seitz Materials Research Laboratory at the University of Illinois. The CIS films used in this study were grown by D. Hebert and C. Mueller at the University of Illinois, whose help is gratefully acknowledged. Assistance with XRD data analysis by Dr. Mauro Sardela of the Materials Research Laboratory at the University of Illinois is also acknowledged. 1 R. Gillette, “Q2 2010 performance summary, business update, and market outlook,” First Solar Quarterly Report, First Solar, Phoenix, AX, 2010. 2 P. Jackson, D. Hariskos, E. Lotter, S. Paetel, R. Wuerz, R. Menner, W. Wischmann, and M. Powalla, Prog. Photovoltaics 19, 894 (2011). 3 I. Repins, M. A. Contreras, B. Egaas, C. DeHart, J. Scharf, C. L. Perkins, B. To, and R. Noufi, Prog. Photovoltaics 16, 235 (2008). JVST A - Vacuum, Surfaces, and Films Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 130.113.111.210 On: Fri, 19 Dec 2014 16:47:20 021202-6 A. Aquino and A. Rockett: Chemical reactions at CdS heterojunctions with CuInSe2 4 Y. L. Soo, S. Huang, Y. H. Kao, S. K. Deb, K. Ramanathan, and T. Takizawa, J. Appl. Phys. 86, 6052 (1999). 5 D. X. Liao and A. Rockett, J. Appl. Phys. 93, 9380 (2003). 6 C. Lei, M. Duch, I. M. Robertson, and A. Rockett, J. Appl. Phys. 108, 114908 (2010). 7 T. Nakada, Thin Solid Films 361, 346 (2000). 8 C. Heske et al., Appl. Phys. Lett. 74, 1451 (1999). 9 H. Wiesner, J. Keane, K. Ramanathan, M. Contreras, and R. Noufi, “Junction formation in CuInSe2 based thin film devices,” National Center for Photovoltaics Program Review Meeting, Denver, CO, 811 September 1998, U.S. Department of Commerce. 10 B. M. Basol and V. K. Kapur, Sol. Cells 30, 143 (1991). 11 P. Hailin, X. Chong, D. T. Schoen, K. McLlwrath, Z. X. Feng, and C. Yi, Nano Lett. 7, 3734 (2007). 12 R. Krishnan, G. Tong, W. K. Kim, R. Kaczynski, U. Schoop, E. A. Payzant, and T. J. Anderson, 38th IEEE Photovoltaic Specialists Conference (IEEE, New York, 2012), 001970. 021202-6 13 S. E. Asher, K. Ramanathan, D. W. Niles, H. Wiesner, and H. Moutinho, AIP Conf. Proc. 462, 126 (1999). F. A. Abou-Elfotouh, L. L. Kazmerski, T. J. Coutts, R. J. Matson, S. E. Asher, A. J. Nelson, and A. B. Swartzlander-Franz, J. Vac. Sci. Technol. A 7, 837 (1989). 15 L. L. Kazmerski, “Ternary compound thin film solar cells,” Project Report, U.S. Department of Commerce document PB2625366, 1976. 16 A. P. Kumar and K. V. Reddy, Semicond. Sci. Technol. 12, 966 (1997). 17 J. F. Guillemoles, L. Kronik, D. Cahen, U. Rau, A. Jasenek, and H. W. Schock, J. Phys. Chem. B 104, 4849 (2000). 18 A. Rockett, T. C. Lommasson, L. C. Yang, H. Talieh, P. Campos, and J. A. Thornton, 20th IEEE Photovoltaic Specialists Conference, Las Vegas, NV, (IEEE, New York, 1988), p. 1505, IEEE Cat. No.88CH2527-0. 19 M. A. Contreras, M. J. Romero, B. To, F. Hasoon, R. Noufi, S. Ward, and K. Ramanathan, Thin Solid Films 403–404, 204 (2002). 20 M. Newberger, III-V Ternary Semiconductors (Plenum, New York, 1971). 14 J. Vac. Sci. Technol. A, Vol. 31, No. 2, Mar/Apr 2013 Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 130.113.111.210 On: Fri, 19 Dec 2014 16:47:20