Reinvestigation of the tensile strength and fracture property of Ni(111)/α-Al2O3(0001) interfaces by first-principle calculations

Xiancong Guo,Fulin Shang

DOI: https://doi.org/10.1016/j.commatsci.2011.01.001

IF: 3.572

2011-01-01

Computational Materials Science

Abstract:Research highlights ► Novel Ni(1 1 1)/α-Al 2 O 3 (0 0 0 1) interface models without initial lateral stresses are used. ► Lateral stresses remarkably influence the mechanical properties of the interfaces. ► Truly uniaxial tension is a requisite for study mechanical properties of interfaces. ► Some inconsistencies on failure behavior of Ni/Al 2 O 3 interface are clarified. Abstract First-principle calculations are performed to reinvestigate the mechanical tensile property and failure characteristic of Ni/Al 2 O 3 interfaces, in order to clear the inconsistence existed in the literatures. Four types of interface models without initial lateral stresses are used, i.e., Al-terminated O-site, O-terminated Al-site, Al-terminated Al-site and Al-terminated H-site models. Two kinds of tensile methods, viz. , uniaxial extension and uniaxial tension, are adopted to check the mechanical responses of these interface models. It is found that the results under uniaxial extension are generally consistent with those under uniaxial tension, including the overall shapes of stress–strain curves and the values of tensile strengths. Moreover, the initial lateral stresses have an apparent influence on the mechanical properties of the interfaces during the loading process, such as tensile strength, fracture strain and the work of separation. Our simulation results also clarified that, under tensile loading, the most stable O-terminated Al-site interface model tends to fracture in a brittle way along the sublayer between in-plane Ni–Ni atomic bonds, while all of the Al-terminated interface models will fail in a ductile fracture manner with relatively lower stress levels, breaking along the interlayer between the Ni(1) and Al(1) layers. Keywords Thermal barrier coatings Ni/Al 2 O 3 interfaces Tensile strength Fracture First-principle calculations 1 Introduction Thermal barrier coatings (TBCs) spallation and delamination are well-known problems that have been addressed by a number of experimental and theoretical studies [1–8] . Premature failure has been observed to occur often along the interface between the bond coat and the thermally grown oxide (TGO) by microcrack nucleation, propagation and coalescence, leading eventually to spalling and delaminating of the TBC. To prevent premature failure in TBCs, it is important to improve the adhesion strength between the bond coat and the TGO. The exact structure and composition of real interfaces between bond coat metals and TGOs can be very complicated. In this study, focus was put on modeling Ni/α-Al 2 O 3 interfaces. Johnson and Pepper [9] was one of the pioneering works that conducted the first-principle calculations of the Ni/Al 2 O 3 interface, and therein a self-consistent scattered-wave approach was used. To give a quantitative evaluation of interfacial adhesion strength, many researchers resorted to the parameters such as work of separation and ideal tensile strength [10,11] . Zhang et al. [10] indicated that O-terminated Ni/α-Al 2 O 3 interfaces possessed large work of separation, and the interfacial separation processes induced a substantial plastic dissipation in the metal; in contrast, Al-terminated Ni/α-Al 2 O 3 interfaces with work of separation several times smaller than either of the adjoining materials and failed in a brittle manner. According to the suggestion of Shi et al. [12] , however, it may not be reliable to evaluate the strength of interfacial adhesion by using this work of separation parameter. In addition to this, Shi et al. [12] demonstrated that the Al-terminated interfaces would rather fail by ductile fracture, while the O-terminated interfaces tend to undergo brittle fracturing. Noticing these inconsistent results existed in the literatures, it is worthwhile to carefully reinvestigate the adhesion strength and fracture characteristic of the Ni(1 1 1)/α-Al 2 O 3 (0 0 0 1) interface. The second motivation to revisit the mechanical behavior of Ni/Al 2 O 3 interface under tension is related to the simulation methodology used in first-principle calculations. Conventional first-principle tensile tests of interfaces up to now have dealt with only uniaxial extension rather than uniaxial tension and their differences lie in whether a triaxial stress state or lateral stress is involved or not. Shi et al. [12] examined the mechanical properties of Ni(1 1 1)/α-Al 2 O 3 (0 0 0 1) interface through so-called relaxed-type and rigid-type tensile tests. The relaxed-type, or equivalently uniaxial extension, tensile test can induce a triaxial stress state due to a large Poisson’s ratio. Strictly speaking, it is essential to perform truly uniaxial tension simulations on this interface, in order to allow for elasticity through the inclusion of Poisson’s ratio in the tensile tests. From this point of view, it is necessary to re-consider the first-principle study in capturing the tensile property and failure behavior of the Ni/Al 2 O 3 interfaces. In this work, we made use of four types of Ni(1 1 1)/α-Al 2 O 3 (0 0 0 1) interface models without lateral stresses, and performed the first-principle tensile simulations of these models under both the conditions of uniaxial extension and uniaxial tension. And the computational results are examined and compared with those existed results. In particular, the mechanical properties including tensile strength and work of separation are discussed, as well as the failure characteristic of these interfaces. 2 Methodology In the present study, all calculations were performed using the density functional theory (DFT) [13,14] code VASP (i.e., the Vienna ab initio Simulation Package) [15,16] . The single-particle Kohn–Sham equations are solved using a plane-wave basis set and periodic boundary conditions. The ion core potentials are described by projector augmented wave (PAW) potentials [17,18] . Electronic exchange and correlation terms are described using the PW91 version [19] of the generalized gradient approximation (GGA). For the k -point sampling of the Brillouin zone, we used the Monkhorst–Pack (MP) scheme. Ground-state atomic geometries were obtained by minimizing the Hellman–Feynman forces using a conjugate gradient algorithm. To ensure the convergence of results, a plane-wave cutoff energy of 500 eV and an energy convergence criterion of 10 −5 eV for self-consistency are adopted throughout all the calculations. The magnetic contributions are neglected. Test calculations were performed for the bulk and surface properties of Ni and α-Al 2 O 3 at first to assess the accuracy of the computational methods used for interface systems. The equilibrium lattice constants were obtained to be 3.514 Å for Ni and 4.806 Å for α-Al 2 O 3 . The surface energies were obtained to be 1.95 J/m 2 for Ni(1 1 1) and 1.52 J/m 2 for stoichiometric α-Al 2 O 3 (0 0 0 1) by following the fitting method in Refs. [20,21] . The good convergence of surface energies showed that using five layers has been sufficient for Ni while 15 layers for Al 2 O 3 . All our calculated values are in good agreement with available experimental and other first-principle results [22–29] . A popular sandwich model of the Ni(1 1 1) (√3 × √3)/α-Al 2 O 3 (0 0 0 1) (1 × 1) coherent interface without vacuum regions is dealt with in present work [10–12] . Based on the results of the above surface convergence tests, the supercell is constructed by a slab of α-Al 2 O 3 (0 0 0 1) layers consisting of five O atomic layers and 10 Al atomic layers for Al-terminated model while both sides are removed for O-terminated one, sandwiched between two Ni slabs. Convergence tests demonstrated that 1–2 meV/Å per atom degree of convergence with respect to k -point sampling was attained upon using 5 × 5 × 1 points for the interface systems. By fully relaxing the entire slab, equilibrium configurations without initial stress were obtained, i.e., all components of the stress tensor are less than 0.01 GPa, when the Hellman–Feynman forces are less than 0.01 eV/Å. After obtaining the equilibrium configurations, the so-called uniaxial extension and uniaxial tension ideal tensile tests were performed. With the former, the simulation supercell and its contents are expanded uniformly along the direction of tensile loading, allowing atomic relaxation while keeping lateral lattice vectors fixed. With the latter, the system is also expanded along the direction of the applied load, but the lateral lattice vectors are relaxed along with the ionic positions so that the stresses in the direction orthogonal to the loading direction vanish (less than 0.01 GPa). In both tests, the ions relaxations are performed iteratively until the atomic forces are less than 0.05 eV/Å, and the C 3 i symmetry of the supercell is retained during the relaxation. 3 Results and discussion In our study, four types of Ni/Al 2 O 3 interface models are considered, i.e., Al-terminated O-site model (in short, Al–O model), O–Al, Al–Al and Al–H models. Their lateral lattice parameter a and the lattice misfits are shown in Table 1 . The equilibrated supercells obtained in our study are similar to the type II (T-II) interface model mentioned in Ref. [10] , which in principle applied an expansion to Ni(1 1 1) layer and a contraction to α-Al 2 O 3 (0 0 0 1) layer, respectively. Zhang et al. [10] pointed out that the interface configuration is relatively insensitive to the pre-existing misfit strain. It is reasonable that lattice misfits are about 11% for all these candidate interface models, although the experimental misfit is 9.488% [23] . Without a fully relaxation of the cell lengths parallel to the interface, the above T-II interface model induced non-zero initial lateral stresses, which is not the case in our models. However, from our calculations, it is found that the initial interfacial structures after relaxation are not very different with those in Refs. [10,11] , including the interlayer distances ( D layer ), bond lengths ( D bond ) and bulk differential (Δ d ij ). It is apparent that whether the lateral stresses exist or not in the supercell model, the initial interface atomic configurations hardly change. As expected, the volume of the each interface model, see Fig. 1 , is expanded during the tensile process, regardless of tensile method. For each interface model, the plots are constructed by combining two sets of calculated results under uniaxial extension and uniaxial tension tests. For all of the four models, their total energies in uniaxial extension tests are higher than those in uniaxial tension tests, due to relaxation of the lateral lattice vectors parallel to the interface. When strains increase to critical values, the stresses reach their maximum values while the total energies do not increase, at these points failure has taken place. Fig. 2 shows the stress–strain curves of the four models with Ni(1 1 1)/α-Al 2 O 3 (0 0 0 1) interface under both tensile tests. Under uniaxial extension, at strains of 6% and 5%, stresses reach maximum values of 7.362 GPa and 6.771 GPa for Al–Al and Al–H interfaces, respectively, compared to maximum stresses of 7.372 GPa and 6.876 GPa at strains 8% and 7% respectively under uniaxial tension. For the Al–O model, the stress reaches a maximum value of 11.098 GPa at a strain of 8% under uniaxial extension, which is lower than the maximum stress of 15.384 GPa (at critical strain of 9.181%) reported by Ref. [12] . For the O–Al interface, which is found to be the most stable configuration among the four interface models, the uniaxial extension calculations reveal a maximum stress of 22.907 GPa at critical strain of 11%. According to the results obtained in Ref. [12] , the maximum stress could be high as 28.301 GPa when strain reaches a critical value of 14%. In other words, our prediction of fracture strength is relatively 19% lower and that of critical strain is also 21% lower than those in Ref. [12] for the O–Al interface model. In our study, the extensile simulations were carried out for the fully-relaxed interface configurations without initial lateral stress. Different from this, Shi et al. [12] started with an interface model corresponding to the type I (T-I) interface model suggested in Ref. [10] and certain non-zero pre-stresses exist in their model. These pre-applied lateral stresses thus prevent the interface model from contracting/shrinking along the lateral directions. Moreover, the lateral stresses in type I (T-I) interface models are larger than that used in the present work during the tensile process. These computational conditions might induce the above differences in the calculated results. Moreover, it can be seen from Fig. 2 that the critical strain for uniaxial extension is smaller than that for uniaxial tension, while their ideal tensile strengths show little discrepancy. This is due to that lateral tension is involved in the present uniaxial extension test that prevents lateral contraction, as depicted in Fig. 3 . Poisson’s ratio generally decreases and drops rapidly at the critical strain. Most importantly, regardless of what tensile method is used, the shapes of the stress–strain curves obtained agree well with each other for the same interface model, a feature which was first noticed by Shi and co-workers [12] . Another important parameter to characterize the property of interface fracture is the area under the stress–strain curve, A . This parameter gives an approximate evaluation of the work required to separate the interface, which is often referred as the ideal work of adhesion (i.e., the energy required to rigidly cleave the interface) in literatures, e.g. [30,31] . According to our study, the calculated area under the stress–strain curve A for the Al–O interface model, (see Fig. 2 a) is 1.44 J/m 2 and 1.48 J/m 2 for uniaxial extension and uniaxial tension, respectively, while those for the O–Al model (see Fig. 2 b) are 3.82 J/m 2 and 4.17 J/m 2 , respectively. It is apparent that the areas under the stress–strain curves for the same interface model are consistent with each other, regardless of tensile method. In other words, A is not much sensitive to the tensile method. By comparing the work of adhesion to the work of separation for an ideal clean Ni/Al 2 O 3 interface given in Ref. [10] , the above calculated A appears to be reasonable. Moreover, it is worthwhile to mention here that the area A is an essential parameter in constructing the widely used cohesive zone models (CZMs) in continuum mechanics framework [32–34] . The parameter A usually represents the work of fracture in CZMs. There exists many different ways to develop a material-dependent CZM and its core task is to extract the material parameters such as the area A and the maximum cohesive strength σ max in the cohesive laws. By using the results from first-principle calculations, however, the CZM could be constructed as well and this kind of CZM will naturally possess a more reliable physical foundation. To better understand the failure process of the four Ni(1 1 1)/α-Al 2 O 3 (0 0 0 1) interface models, atomic bonds near the interface, as shown in Fig. 4 , are checked for uniaxial tensile test. By observing both the behavior of interfacial bonds in Fig. 4 and the stress–strain relationships shown in Fig. 2 , we can clearly determine the fracture mode of each interface model considered. For the Al–O model, Fig. 4 a, the interfacial bond length of Al(1A)–Ni(1A) and the interfacial stress change gradually when strain goes beyond the critical strain, demonstrating that the Al-terminated O-site interface breaks in an analogous manner to ductile fracturing. Similar fracture process has been found for the other two Al-terminated models, viz. , the Al-site and H-site models and both of them have a clear failure between the Ni(1) and Al(1) layers, Fig. 4c and d. As shown in Fig. 5 a, the valence charge density contours also reveal that the charge depletion region propagates to the Ni(1A)–Al(1A) interfacial bond as the strain increases up to the critical value, until the interfacial region is fully separated into two surfaces. This gradual breakage of the atomic bonds is quite distinguishing from the feature of catastrophic failure for the O-terminated Al-site interface. For the O–Al model, Fig. 4 b, the interfacial bond lengths of Ni(2C)–Ni(1B), Ni(2A)–Ni(1C), O(1C)–Ni(1A) and Al(2A)–Ni(1A) increase suddenly at the critical strain point and the stresses drop abruptly. And the remarkable decrease in valence charge density is noticeable in these bonds before and after critical strain point (see Fig. 5 b), which signify that brittle fracture has taken place along the sublayer between the in-plane Ni–Ni bonds. 4 Conclusions We have re-considered first-principle calculations of the structural properties, tensile strengths and failure processes for Ni(1 1 1)/α-Al 2 O 3 (0 0 0 1) interfaces. Four types of novel interface models without pre-applied lateral stress were adopted and two kinds of tensile simulation methods, i.e., uniaxial extension and uniaxial tension, were used to check the mechanical responses of these interface models. From the careful examinations of our computational results and also comparisons with available results in the literatures, the following conclusions can be reached: (1) The lateral stresses have negligible effects on the initial interface atomic configurations, but they have remarkable influence on the mechanical properties of the interfaces during the loading process, such as tensile strength, fracture strain and the work of separation. (2) The computational results from both tensile tests for the same interface model are consistent with each other in general tendencies, and the overall shapes in the stress–strain curves are similar. Due to the effect of Poisson’s ratio, truly uniaxial tension rather than uniaxial extension is a requisite for capturing some subtleties behind mechanical properties of Ni/Al 2 O 3 interfaces, including the calculated tensile strengths, fracture strains, and the ideal work of adhesion (i.e., the area under the stress–strain curve). Specifically, the commonly used uniaxial extension could give predictions of fracture strength (peak stress) and fracture strain that are 20% higher than those from truly uniaxial tension simulations. And the exact values of these parameters could be critical in certain cases, for example, in developing a continuum-based cohesive zone model for Ni/Al 2 O 3 interfaces, or in facilitating large-scale molecular dynamics simulations of the same Ni/Al 2 O 3 interfaces. (3) Some inconsistencies or discrepancies existed in the literatures have been clarified. For the Al-terminated O-site, Al-site and H-site Ni/Al 2 O 3 models, fracture would occur along the interlayer between the Ni(1) and Al(1) layers, among them, the O-site interface model fails with the highest ideal strength. Furthermore, the Al-terminated O-site model would break along the Al(1A)–Ni(1A) atomic bond at a relatively lower level of tensile stress and underwent ductile fracturing. Different to the above three interface models, the O-terminated Al-site Ni/Al 2 O 3 model tend to fracture in a brittle way, and its failure occurs along the sublayer between the in-plane Ni–Ni bonds. Acknowledgment This work was supported by New Century Excellent Talents Program from Ministry of Education of China (No. NCET-08-0445). References [1] D.R. Mumm A.G. Evans Acta Mater. 48 2000 1815 1827 [2] A.G. Evans D.R. Mumm J.W. Hutchinson G.H. Meier F.S. Pettit Prog. Mater. Sci. 46 2001 505 553 [3] N.P. Padture M. Gell E.H. Jordan Science 296 2002 280 284 [4] W.G. Mao C.Y. Dai Y.C. Zhou Q.X. Liu Surf. Coat. Tech. 201 2007 6217 6227 [5] X. Chen J.W. Hutchinson M.Y. He A.G. Evans Acta Mater. 51 2003 2017 2030 [6] D.R. Mumm M. Watanabe A.G. Evans J.A. Pfaendtner Acta Mater. 52 2004 1123 1131 [7] A.G. Evans J.W. Hutchinson Surf. Coat. Tech. 201 2007 7905 7916 [8] Y. Wei J.W. Hutchinson Philos. Mag. 88 2008 3841 3859 [9] K.H. Johnson S.V. Pepper J. Appl. Phys. 53 1982 6634 6637 [10] W. Zhang J.R. Smith A.G. Evans Acta Mater. 50 2002 3803 3816 [11] S. Shi S. Tanaka M. Kohyama J. Am. Ceram. Soc. 90 2007 2429 2440 [12] S. Shi S. Tanaka M. Kohyama Phys. Rev. B 76 2007 075431 [13] P. Hohenberg W. Kohn Phys. Rev. B 136 1964 B864 B871 [14] W. Kohn L.J. Sham Phys. Rev. 140 1965 1133 1138 [15] G. Kresse J. Hafner Phys. Rev. B 47 1993 558 561 [16] G. Kresse J. Furthmuller Phys. Rev. B 54 1996 11169 11186 [17] P.E. Blöchl Phys. Rev. B 50 1994 17953 17979 [18] G. Kresse D. Joubert Phys. Rev. B 59 1999 1758 1775 [19] J.P. Perdew Y. Wang Phys. Rev. B 45 1992 13244 13249 [20] J.C. Boettger Phys. Rev. B 49 1994 16798 16800 [21] V. Fiorentini M. Methfessel J. Phys. Condens. Matter 8 1996 6525 6529 [22] B. Hinnemann E.A. Carter J. Phys. Chem. C 111 2007 7105 7126 [23] C. Kittel Introduction to Solid State Physics eighth ed. 2005 John Wiley & Sons New Caledonia [24] W.G. Wyckoff Crystal Structures second ed. 1964 Wiley-Interscience New York [25] F. Mittendorfer A. Eichler J. Hafner Surf. Sci. 423 1999 1 11 [26] L. Vitos J. Kollár A.V. Ruban H.L. Skriver Surf. Sci. 411 1998 186 202 [27] E.A.A. Jarvis A. Christensen E.A. Carter Surf. Sci. 487 2001 55 76 [28] W.R. Tyson W.A. Miller Surf. Sci. 62 1977 267 276 [29] J.M. McHale A. Navrotsky A.J. Perrotta J. Phys. Chem. B 101 1997 603 613 [30] M.W. Finnis C. Kruse U. Schonberger Nanostruct. Mater. 6 1995 145 155 [31] M.W. Finnis J. Phys. Condens. Matter 8 1996 5811 5836 [32] G.I. Barenblatt Adv. Appl. Mech. 7 1962 55 129 [33] D.S. Dugdale J. Mech. Phys. Solids 8 1960 100 104 [34] A. Needleman J. Appl. Mech. 54 1987 525 531