Furaday Discuss. Chern. SOC., 1990, 89, 21-30 Glancing Angle XAFS and X-Ray Reflectivity Studies of Transitionmetal/ Aluminium Interfaces Steve M. Heald, Enrique V. Barrera and Huaiyu ChenT Brookhaven National Laboratory, Upton, N Y 11973, USA The combination of glancing angle X-ray absorption fine structure (XAFS) and X-ray reflectivity measurements provide detailed structural information about interfaces. Results from transition-metal/Al interfaces are used to illustrate the capabilities of the techniques. Reflectivity measurements are used to determine the thickness and density profile of the films, along with the amount of interfacial or surface roughness. The XAFS measurements provide information about the local bonding at the interface, and are well suited for studying the interfacial chemistry.Of interest is the role of impurities such as oxygen in modifying interface structure and thermal stability. For Cu/AI, Ni/AI and Cr/AI a small amount of interfacial oxygen is found to affect strongly both the initial interface structure, and the response of the interface t o thermal annealing. For NiiAl it is also found that interfacial oxygen afiects primarily the interface reaction, while oxygen within the A1 layer suppresses the grain boundary reaction channel. The study of surfaces and interfaces by X-rays is greatly facilitated by using glancing incidence angles.” At small angles of incidence, X-rays can undergo total external reflection, resulting in an extremely small penetration depth. This small penetration greatly reduces the bulk contribution to the signal, allowing the small interface or surface signal to be studied in detail.Such an approach, combined with synchrotron radiation, has allowed detailed X-ray studies of a variety of surface reconstructions and ad layer^.^-' In this paper we report the application of glancing angle X-rays and X-ray absorption fine structure measurements (XAFS) to study buried interfaces. Examples are given f o r transition-metal/Al interfaces. X-ray reflectivity measurements are sensitive to the density profile with depth. The method can provide nanometer depth resolution and is sensitive to elemental concentra- tions as low as 1%.’ From the density profile it is straightforward to calculate the penetration of the X-rays to determine the region of the sample being probed by XAFS.The density profile can also be used as a basis for correcting the anomalous dispersion distortions present in the XAFS as a consequence of the glancing angle geometry.’ Such a correction is necessary for detailed quantitative analysis, although, as shown in this paper, qualitative examination of the raw spectra can provide much information. pi-eviously these techniques were used to make a detailed study of the influence of interfacial oxygen on the interface structure and thermal stability of AI/Cu interfaces.”’ Small amounts cjf oxygen were found to have a strong effect on both properties. This work has been extended to Al/Ni and AI/Cr interfaces, where again oxygen is found to have a strong influence on interface structure and stability.In particular, for Ni/AI we can distinguish between the effects of 0 impurity in the layer and 0 impurity at the i n t e r fa c e . +Current address: Physics Dept., Illinois Inst. of Tech., Chicago, IL 60616, USA.22 Glancing Angle XAFS and X-Ray Rejlectivity Studies - 1 g O L : Interface + c t P? PI V l 2 O O O L ; , _1.- 0 ZOO LOO 600 800 depth/ A Fig. 1. Steps in the analysis of reflectivity data as illustrated for a Cu/AI bilayer annealed at 120" C. ( a ) Reflectivity data (points) and a least-square fit (line). V i . PI and P2 refer to angles of the first minimum, first peak and second peak, respectively. ( b ) Cu density profile derived from the fit in ( a ) . ( c ) Calculated contribution to the Cu fluorescence signal from the three angles shown in ( a ) .Background The index of refraction for X-rays is slightly less than one, leading to total external reflection a t small incident angles below the critical angle, 8,. This leads to a very small penetration depth ( a few nm), and it was Parratt' who demonstrated the potential for using glancing angle X-rays to probe surface structure. For bilayer systems the X-rays can be made to reflect from the interface if the overlayer is less dense than the substrate. Interface reflection occurs when the incident angle is greater than 8,- for the surface, but less than 8, for the substrate. Measurements of the substrate absorption can then be used to determine the bonding environment of the atoms within the penetration depth of the interface. The penetration depth can be varied by varying the incident angle to look at the depth dependence of the bonding.These ideas were beautifully verified by extensive studies of the interfaces in AI/Cu bilayers. 273.7.8 Some examples drawn from this work are shown in fig. 1 and 2 to demonstrate the basic experimental procedure. First. the reflectivity is measured asS. M. Heald, E. V. Barrera and H. Chen 23 2 4 6 8 10 12 k / A - ' Fig. 2. Extended fine structure data for the three angles shown in fig. l ( a ) compared to CuAI, and C u standards; ( a 1 CuAI,, ( h ) V1, ( c ' ) P1, ( d ) P 2 , ( e l Cu. The standards have been multiplied by 1/2 to approximate the anomalous dispersion distortion of the glancing angle data. shown in fig. l ( a ) , usually at X-ray energies above and below the absorption edge of the substrate (9.2 and 8.6 keVf for Cu ).From this data a density profile can be determined by fitting. An example of the quality of fit is also shown in fig. l ( a ) . Fig. l t h ) shows the Cu density profile derived from fig. l ( a ) . The derived density profile can then be used to calculate the contribution to the XAFS from the Cu atoms at various depths, as shown in fig. l ( c ) , for the three angles indicated in fig. l ( a ) . The fluorescence contribution is calculated by multiplying the Cu concentration by the depth-dependent electric field intensity. The oscillatory behaviour comes from interference in the electric field due to the surface and interface contributions to the reflectivity. The reflectivity oscillations in fig.l ( a ) come from the same source. Different parts of the sample are probed for different angles. This is confirmed by the extended region of the XAFS (EXAFS) spectra shown in fig. 2. At small angles the spectra are similar to those of CuAI,, indicating substantial Cu-AI bonding, while for larger angles more Cu-Cu bonding from the substrate is evident. These measurements allow us to deduce the amount of chemical interaction at the interface and the types of bonds present even for small amounts of reaction. Note, the interface width derived from reflectivity can also be due to a rough interface with little intermixing, and from the EXAFS measurements it is simple to distinguish between the two cases. In the single-scattering approximation the EXAFS spectrum can be written as" .;-I e V = 1.60218~ 10 "'J.24 Glancing Angle X A FS and X-Ray Reflectivity Studies where p is the total absorption and po is the smooth background characteristic of an isolated atom. The sum is over the neighbouring atomic shells, where N is the number of atoms in a shell, R is their average distance, and CT' is the mean-square displacement of the N atoms about R.A ( k ) and @ ( k ) depend on the types of atoms involved, and are usually calibrated using standard compounds such as Cu for Cu-Cu bonding and CuAI, for Cu-AI bonding. Each nearest neighbour distance gives rise to a distinct frequency of oscillation and using Fourier analysis along with fitting techniques it is possible to extract the structural parameters N, R and CT for the first few nearest neighbour shells.Since EXAFS is a local probe, structural information can be gained even if no long-range order exists. The other component of the XAFS is the near-edge structure which is more sensitive to the chemical state of the absorbing atom. In the glancing angle mode, the absorption coefficient, p, is not directly measured. Either the reflectivity or the fluorescence signal can be measured, and they must be corrected to obtain the absorption. In the present work the fluorescence signal was found to be of higher quality. For very thin layers or dilute systems there is a direct proportionality to the absorption, but in other cases anomalous dispersion distortions become important.' However, once a model for the reflectivity has been obtained it is a simple matter to make the corrections.The distortion affects primarily the amplitude of the XAFS, and for qualitative comparisons as in fig. 2 corrections are not necessary. Experiment a1 The samples were prepared in an ion-pumped UHV chamber with a base pressure of 5 x 1OP"' Torr.? Cu and Ni were evaporated using alumina crucibles and Cr was evapor- ated from a W basket. Al was deposited two ways. For the Cu/Al samples, W boats were used while for the Ni/AI and Cr/AI samples an alumina crucible was used. Subsequent Auger analysis showed that the alumina deposited aluminium had a small (2-3%) amount of 0 contamination and second set of Ni/AI samples was prepared using a W boat for the aluminium. This resulted in clean samples and allowed us to make a comparison of the two cases.Float glass substrates (2.5 x 5 cm) were used, and the substrate holder was cooled to keep the samples at or below room temperature during deposition. In each case two samples were made: one with a clean interface and one in which the metal surface was exposed to 0, gas prior to Al deposition. For C u the exposure was 1000 L$ and for Ni and Cr the exposures were 600 L. The X-ray measurements were made at the Cornell High Energy Synchrotron Source (CHESS) beamline C-2, and the National Synchrotron Light Source (NSLS) beamline X-1 1A. The samples were mounted on a precision goniometer and gas ionization chambers were used to measure the incoming, reflected and fluorescence X-rays. The entrance slit was 50-100 p m which resulted in a flux on the sample of ca.10' photons s I . This resulted in a fluorescence signal of ca. lo7 which allowed high quality XAFS data to be obtained in ca. 15 min. Measurements were made at room temperature on as prepared and annealed samples. The samples were annealed in air for the Cu/Al samples at temperatures ranging from 65-2OO0C, and in a getter-purified Ar atmosphere for temperatures ranging from 230- 340 "C for the Ni/AI samples and 240-390 "C for Cr/Al. The same sample was annealed sequentially at successively higher temperatures. The annealing time was 5 min at temperature with ca. 10 min needed to reach temperature. Above ca. 350 "C the float glass substrates had a tendency to break into two pieces, but measurements could be made on one of the pieces and there seemed to be no degradation of the substrate flatness.; 1 Torr i 1 L - 1 0 "Torrs '. 101 325/760 Pa.S. M. Heald, E. V. Barrera and H. Chen 25 2.5 2.0 x Y .- > .- ti 1.0 2 5 0.5 0.2 0.3 0.4 0.5 0.2 0.3 0.4 0.5 angle/" Fig. 3. X-Ray reflectivity data obtained at 8.633 keV for: (a) a Ni/Al bilayer with a clean interface and ( b ) a Ni/AI bilayer with an interface exposed t o 600 L of 02. The A1 layer for these samples had a small amount of 0 contamination as described in the text. 0.2 0.3 0.4 0.5 0.2 0.3 0.4 0.5 angle/" Fig. 4. X-Ray reflectivity data obtained at 6.189 keV for: ( a ) a Cr/Al bilayer with a clean interface and ( b ) a Cr/AI bilayer with an interface exposed to 600 L of 0:. The Al layer for these samples had a small amount of 0 contamination as described in the text.26 Glancing Angle XAFS and X - Ray Reflectivity Studies 2 0 1 5 1 0 0 5 0 0 0 200 400 600 800 0 200 400 600 800 depth/A Fig.5. Ni density profiles derived from fitting the data in fig. 3 aiong with the corresponding reflectivity data obtained at 8.033 keV. Reflectivity Data Fig. 3 and 4 present some of the reflectivity data obtained from the Ni/Al and Cr/AI interfaces. The Ni/Al data in this case was taken on the sample which showed a small amount of 0 contamination in the Al layer. Detailed results for Cu/AI can be found in a previous publication.3 The figures show the data obtained for energies above the Ni and Cr K-edges, respectively, which is the most sensitive to changes in the interfacial mixing. It is clear that interfacial 0 has a strong effect on the annealing behaviour in both systems.For Ni/AI the reaction for a clean interface begins at ca. 250 "C, while for the oxygen exposed interface the reaction begins near 310°C. For Cr/Al the corresponding temperatutes are found to be 330 and 420°C. Fig. 5 shows the Ni concentration profiles obtained from fitting the Ni/Al data in fig. 3. For the clean interface sample the reaction proceeds from the interface, generating a well defined intermediate layer with a density consistent with NiAI3 formation. Other workers''-" have also found that NiAI, is the first compound to form. Only at 310 "C does significant Ni reach the near-surface region (the fall off at the surface is due to an air-formed A1203 layer). This is likely to be due to compound growth along the Al grain boundaries.For the oxygen-exposed interface the intermediate layer is not as well defined, although the Ni concentration near the surface begins increasing at temperatures similar to the clean sample. It appears that the interfacial 0 acts to block the interface reaction, but the reaction in the grain boundaries takes place at similar temperatures for both samples. As mentioned both of these samples had small amounts of 0 in the A1 layer. To check whether this was affecting the annealing behaviour, a second sample was measured which had a clean Al layer and a clean interface. The concentration profiles for this sample are shown in fig. 6. Again the reaction begins at 25OoC, and a well defined NiAI, layer is formed. However, there is also more Ni reaching the surface. This indicates that the grain boundary reaction was inhibited for the first sample.Thus, theS. M. Heald, E. V. Barrera and H. Chen 27 2 C 0 m .C 4-8 L != e, C 4-8 8 .- z 1 .o 0.5 0.0 0 2 00 400 600 800 depth/ A Fig. 6. Ni density profles obtained for a Ni/AI bilayer with a clean interface and clean A1 layer. effect of 0 in the A1 layer is localized to the grain boundaries, while the 0 placed at the interface affects only the interfacial reaction. Note that for the contaminated sample, the 0-induced reaction barrier is overcome at approximately the same temperature (310 "C) for both the interface and grain boundaries. Detailed analysis has not yet been carried out on the Cr/Al reflectivity data. The analysis is complicated by the high degree of roughness of the surface and interface.This can be seen especially in the data for the clean sample where it is obvious that the reflectivity oscillatio!s are washed out. Model calculations indicated that the r.m.s. roughness is ca. 150 A. This is quite smooth in an optical sense (i.e. the samples still had a mirror-like appearance), but is large when compared to the Al layer thickness of ca. 500 A. By comparison the Ni/AI and Cu/Al samples had r.m.s. roughnesses in the range of 20-30 A before annealing, and the float glass substrates have a roughness in the range of 5-10 8,. The roughness for C r seems to be due to the deposition process, since as shown in the next section, the XAFS results indicate that little mixing has occurred.It is interesting that the 0-exposed sample seems to have a somewhat smoother interface, although comparison with calculations reveals that it is still substantially rougher than for Cu/AI or Ni/AI. XAFS Results Some examples of the extended fine structure are shown in fig. 7 and 8 for Ni/AI and Cr/Al, respectively. The Ni/AI results demonstrate the sensitivity of the EXAFS measurements. Even though the reflectivity showed a similar interface width for the two samples, it requires the EXAFS to observe the room temperature reaction occurring at the interface. The sample with impure Ai shows no reaction to within our sensitivity of 5-10 8, for the reacted layer thickness. Further analysis is required to quantify the amount of reaction in the clean sample, but from comparison with previous Cu/Al28 I I I I 1 1 1 1 I l l ! Glancing Angle XAFS and X - Ray ReJZectivity Studies I I I I I l l 1 1 I l l 1 I I I I 1 1 1 1 I I 1 I I I 1 1 0 - 0 -1 10 12 2 4 6 k 1 A - I 8 Fig.7. EXAFS data for the Ni/AI bilayers with a clean interface, ( a ) Ni; ( b ) C1; ( c ) C2; ( d ) C2, 250 "C; (e) NiAI,. C1 has an 0-contaminated A1 layer and C2 has a clean Al layer (see text). Ni and NiAI, are standard materials. 1 Fig. 8. EXAFS data for the Cr/Al bilayer with a clean interface unannealed ( h ) and after a 330 "C anneal ( c . ) compared with Cr ( a ) and Cr,AI,, ( d ) .S. M. Heald, E. V. Barrera and H. Chen 29 analysis we estimate the reacted region to be 20-30 A thick. Similar amounts of reaction have been observed by photoemission in UHV-prepared interfaces." For Cr/ Al no reaction is observed. However, for this case a second sample with clean Al was not prepared, and the same suppression of the spontaneous interface reaction may be occurring.The reacted samples give spectra similar to the expected reaction products NiAI, and Cr2AIl3 . 1 4 1 For Cu/Al the corresponding compound is CuAI,, and detailed quantita- tive analysis for the reacted spectra gave results which were similar, but not identical to the CuAI, spectrum. From the angle dependence of the spectra there were indications of additional Cu-rich phases (most likely CuAI) near the Cu layer. Detailed analysis is continuing to see if the same effect is occurring for the Ni and Cr cases. Discussion and Conclusions The influence of impurities on the Ni/AI reaction has been studied by Zhao et al." for large-grained Al substrates, and by Ma e f al." for thin-film couples.For clean material they reported planar NiAI3 growth. For both a contaminated interface (6 x 10' L exposure on the Al surface) and a contaminated Al layer (deposited at 8 x lo-' Torr), the reaction rate was reduced and a rough reaction front was created. These results seem to contradict our findings somewhat. In particular, we found planar growth only for a contaminated Al film. However, the previous measurements were based on Rutherford backscattering (RBS) analysis, which does not have the same depth resolution as th? glancing angle technique. Typically, the reaction layers were much thicker (> 1000 A). Examination of the spectra indicate an interface width of several hundred Angstroms which is not inconsistent with the concentration tails observed in our data.Also, the rough interfaces reported in contaminated samples were found for an annealing temperature of 324 "C, a temperature at which we also observe growth along the grain boundaries. Thus, a detailed comparison of our results with past work does not reveal any inconsistencies. Previous work on clean samples found an initial reaction at 250 "C for the thin film couples," and 330°C for the large-grained Al substrates. However, as pointed out in ref." the cleaning procedures used for the large-grained material may have left a small amount of contamination. In the present work we have shown that very small amounts of interface impurities can inhibit reaction.For the case of Cu, the 1000 L exposure has been shown to give 1/2 monolayer coverage on single-crystal surfaces." On polycrys- talline surfaces the coverage could be greater, but probably not substantially so. Auger analysis is planned to estimate the amount of 0 deposited for exposures employed. These will be made on the metal surface prior to Al deposition. Detection of such small amounts of impurities is difficult for buried interfaces, and could have been missed in some previous studies. Indeed, Auger sputter profiling of the 0-exposed Cu/Al did not reveal any excess 0 at the interface. For Ni/AI the addition of 0 impurities to the Al film did not affect the temperature dependence of the interface reaction, but only the grain boundary diffusion path.This allowed the formation of a well defined NiAI, layer, without extensive Ni diffusion ahead of the reaction front. The impurity 0 is apparently strongly bound up and cannot diffuse back to the interface to inhibit the reaction. This is not unexpected since from the phase diagram 0 in Al should be in the form of Al2O3. This has been observed in contaminated Al thin films,lh and perhaps it is the formation of A1,03 in the grain boundaries which is acting as a diffusion barrier. The EXAFS measurements did reveal that the 0 impurities in the Al inhibit the initial interface reaction. Since the annealing behaviour is essentially unchanged, this implies that the amount of 0 necessary to ; In some references this compound i s rel'erred to as CrAl,30 Glancing Angle XAFS and X - Ray Reflectivity Studies quench the initial reaction is much smaller than that necessary to affect subsequent annealing reactions. This paper has demonstrated some of the features of the glancing angle X-ray techniques for studying buried interfaces.Both chemical and compositional information can be obtained with nanometer resolution. The methods seem especially well suited for studying the initial stage of interface reaction, where the reacted layer is thin and may not have a well defined crystalline structure. In many respects the methods complement the traditional techniques of RBS and TEM. RBS is excellent for following the later stages of reaction in which the thick reacted layers would be more difficult to probe by the glancing techniques. TEM sample preparation is difficult, but for cases such as epitaxial films on semiconductors, atomic-scale resolution can be obtained.However, for systems with more disorder, the results from TEM may not be so clear cut. Also, for cases such as Cu/Al we have found that the sample preparation necessary for TEM can cause enhanced reaction at the interface, obscuring the original structure. In contrast, the glancing angle techniques require little sample preparation aside from the requirement of a flat substrate. In many cases this can be achieved by deposition onto flat substrates, although as seen for the Cr/Al case, work is sometimes needed to establish the proper deposition conditions. We would like to thank J. K. D. Jayanetti and M. W. Ruckman for their help in making the measurements. This work is supported by the US Dept. of Energy, Basic Energy Sciences under contract nos. DE-AS05-ER10742 and DE-AC02-76CH00016. References 1 L. G. Parratt, Phjx Rev., 1954, 95, 359. 2 S. M. Heald, H. Chen and J. M. Tranquada, Phjx Reti. B, 1988, 38, 1016. 3 H. Chen and S. M. Heald, J. Appl. Phj,.~., 1989 66, 1793. 4 W. C. Marra, P. Eisenberger and A. Y. Cho, J. Appl. Phjx, 1979, 50, 6927. 5 G. H. Vineyard, Phys. Rev. B, 1982, 26, 4146. 6 I . K. Robinson, Phys. Rev. Lett., 1983, 50, 1145. 7 H. Chen and S. M. Heald, Solid State lonics, 1989, 32/33, 994. 8 H. Chen, Ph.D. Thesis, (City University of New York, 1989). 9 E. A. Stern and S. M. Heald, in Handbook c!f'Sj.nchrotron Radiation, ed. E. E. Koch (North-Holland, Amsterdam, 1982), vol. 1 b, p. 955. 10 E. C. Colgan, M. Natasi and J. W. Mayer, J . Appl. Phj-s., 1985, 58, 4125. 1 1 X. A. Zhao, H-Y. Yang, E. M a and M-A. Nicolet, J . Appl. Ph-vs., 1987, 62, 1821. 12 E. Ma and M-A. Nicolet, J. Appl. PIij.s., 1989, 65, 2703. 13 M. W. Ruckman, L. Jiang and M. Strongin, J. Vac. Sci. Technol., to be published. 14 A. E. Gershinskii, G . V. Timofeeva and N. A. Shalygina, Thin Sol. Films, 1988, 162, 171. 15 ,4. P. Raddorf and J. F. Wendelken, J. Vac. Sci. Techriol., to be published. 16 M. J. Verkerk and G . J. vander Kolk, J . Vat. Sci. Twhnol. A, 1986. 4, 3101. Paper 9/05381 F; Rc>ceiwtl 14th I>ecember, 1989