Computational Materials Science 23 (2002) 85–94 www.elsevier.com/locate/commatsci Diffusion of clusters down ð1 1 1Þ aluminum islands M. Bockstedte a a,* , S.J. Liu b, Oleg Pankratov a, C.H. Woo b, Hanchen Huang b Lehrstuhl fur Theoretische Festk€orperphysik, Universit€at Erlangen-N€urnberg, Staudtstrasse 7/B2, D-91058 Erlangen, Germany b Department of Mechanical Engineering, The Hong Kong Polytechnic University, Hong Kong Accepted 1 June 2001 Abstract The key factor determining nucleation processes and faceting in homoepitaxial growth as well as texture competition is the mobility of adatoms and small clusters across step edges and facets. Using a combination of molecular dynamics and ab initio calculations, we investigate the mechanisms of small clusters (dimer and trimer) diffusion down the aluminum (1 1 1) surface. In this paper we report results of molecular dynamics studies. Our study shows that the clusters dissociate at the step-edge of compact islands. As a result, the clusters diffuse down the step by an exchange mechanism with a small or medium Schwoebel barrier. The mechanism of this down-diffusion/dissociation is discussed and the corresponding energetics are calculated using the molecular statics method. We find a large anisotropy between the barriers at the two types of h1 1 0i oriented steps. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Cluster diffusion; Schwoebel barrier; Island; Facet; Aluminum 1. Introduction The microstructure evolution of thin films has been a focus of research for years, driven by both scientific interest and technological importance [1– 17]. For example, aluminum thin films, grown with preferred texture and desirable density, are used in the form of metal lines as interconnect in integrated circuits. It is well known that aluminum films of h1 1 1i texture are resistant to electromigration, a process of atomic mass transport due to electron wind [18,19]. During the growth of alu- * Corresponding author. E-mail addresses: [email protected], and [email protected] (H. Huang) after summer 2002. minum films, the texture competition is affected by many factors; one of them being the faceting of (1 1 1) surfaces [2]. The h1 1 1i texture dominates when large f1 1 1g facets form at the early stage of deposition. The direct result of the large facets is a large fraction of substrate area covered by the h1 1 1i grains at the nucleation stage. The large f1 1 1g facets form when adatoms and clusters (e.g. dimers and trimers) diffuse fast and have low Schwoebel barriers for diffusion hops over the facet edges. Both ab initio calculations [13–15] and molecular dynamics simulations [2] have shown that: (1) an adatom has a small migration barrier on the (1 1 1) surface (<0.1 eV); and (2) the adatom has a small Schwoebel barrier (<0.1 eV). If the small clusters, in particular dimers and trimers, have similar 0927-0256/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 7 - 0 2 5 6 ( 0 1 ) 0 0 2 3 2 - 4 86 M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 migration and Schwoebel barriers, large {1 1 1} facets are expected. On the other hand, confinement of the small clusters on the (1 1 1) surfaces by large Schwoebel barriers or by other mechanisms may substantially suppress the faceting and therefore the dominance of the h1 1 1i texture. Apart from the texture competition, the {1 1 1} faceting also affects the density or uniformity of aluminum thin films. This effect is of crucial importance in the filling of trenches and vias, as part of the aluminum interconnections in integrated circuits. Large {1 1 1} facets formed near shoulders of the trenches and vias aggravate the geometrical shadowing – an unavoidable effect in most physical vapor deposition processes. As a result, voids form in the trenches and vias; the {1 1 1} faceting is detrimental in this regard. With the presence of voids, the effective cross-sectional area of the aluminum metal lines for electrical conduction is smaller. Due to the resulting increased current density – keeping the same current at the reduced cross-sectional area, such metal lines will have a shorter lifetime. Despite the importance of the diffusion of small clusters – on and down the {1 1 1} facets – the details of their dynamics remained largely unknown, in particular for aluminum. Experimental studies on Ag systems [20,21] indicate that small clusters are mobile. The experiments clearly showed initial and final configurations of the clusters at various time intervals. However, the time spent in the vicinity of the transition states is too short for experimental observation of the migration process; consequently, the mechanisms of diffusion remain unclear. Our recent molecular dynamics studies [9] indicate that small clusters, such as dimers and trimers, rapidly diffuse on the aluminum (1 1 1) surfaces via jumps of constituent atoms; the involved diffusion barriers are not much higher than 0.1 eV. We further showed that, if dimers are the critical nuclei [2,5], the faceting is limited by dimers diffusion rather than by their dissociation. Therefore, whether dimers or trimers are effective nuclei for three-dimensional growth – a mechanism that limits the faceting – is dictated by their diffusion down {1 1 1 } facets or islands. In this paper we investigate these aspects using compact islands on a (1 1 1) aluminum surface. The formation and time evolution of such islands during the initial stages of homoepitaxial growth have been a focus of investigation recently (cf. [1,12–17] and references therein). It is a model system to study the relevant microscopic processes involved in the kinetics of growth. The main attention, however, has been directed to the diffusion and kinetics of adatoms on the flat surface and at the step edges of islands. Here we address the diffusion mechanism of small clusters (dimers and trimers) down step edges. In particular we focus on the details of the mechanism and its energetics. To answer these questions, we use a combination of (classical) molecular dynamics and ab initio methods to investigate the diffusion of the dimer and the trimer. The molecular dynamics and molecular statics methods are used to elucidate the mechanisms of diffusing down a (1 1 1) island and to estimate the corresponding energetics. As a supplementary tool, ab initio calculations using a plane-wave pseudopotential method [22] provide confirmation/correction of the estimated energetics. As the first part of this investigation, we present the results from molecular dynamics simulations and molecular statics calculations. The ab initio calculations will be presented in another paper. In Section 2, we will describe the simulation method, and present the simulation results in Section 3. The conclusions are summarized and discussed in Section 4. 2. Simulation method As in our previous studies [2], the interatomic interactions of aluminum are described by the EAM potential of Ercolessi and Adams [23–25]. This interatomic potential provides a fairly reliable description of bulk properties, and more importantly the surface energetics. The range of the potential extends up to third neighbors in our ). The simulation calculations (the cutoff is 6.0 A cell consists of three regions, as shown in Fig. 1. Periodic boundary conditions are applied along the two horizontal directions, that is h1 1 0i and h1 1 2i. Within three layers from the bottom of the simulation cell, atoms are fixed to their perfect lattice positions to mimic a large bulk underneath M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 87 Fig. 1. Schematic setup of the simulation cell with a flat surface. the top free surface. Atoms in the three layers right above the fixed region are subject to a Langevin force, to maintain a constant temperature in this region; further they serve as a thermal bath for atoms above. Starting from the simulation cell with a flat (1 1 1) surface, a compact hexagonal island is formed by an extra monolayer of atoms limited by h1 1 0i=f1 0 0g- and h1 1 0i=f1 1 1g-faceted steps right above the flat surface. A dimer or trimer is then introduced on the island to study its diffusion down the island. The island is chosen to be large enough so that the diffusion of a dimer or trimer down one faceted step is not dominated by interaction with other steps; barrier calculation will be carried out at the middle of each step. In the MD simulations, both h1 1 0i=f1 1 1g- and h1 1 0i=f1 0 0g-faceted steps are included so that diffusion down both of them can be sampled in a simulation; all the facets are along h1 1 0i to minimize the potential energy. For simplicity, these two faceted steps will be referred to as {1 1 1}- and {1 0 0}-faceted steps from now on. We choose a step length of 17.0 times of the interaction cutoff length. For our studies of diffusion mechanisms we expect that size effects will be unimportant with this choice. Once the diffusion mechanisms are identified, triangular shaped islands including either {1 1 1}or {1 0 0}-faceted steps are used for the investigation of the migration barriers within static calculations; thereby the simulation cell is effectively reduced maintaining the same separation of islands. To validate the use of the smaller com- putational cell, we also calculate formation energies of a trimer using computational cells of different sizes, and make sure that the variation is less than 0.01 eV. The formation energy of a defect is defined with respective to a reservoir of kinks; that is, the energy of each reference atom is )3.36 eV in aluminum. The simulation cells with all the three types of islands are shown in Fig. 2. Both dynamics and statics simulations of the aforementioned systems are carried out at constant volume. To account for the thermal expansion at finite temperatures, the dimension of computational cell is fixed according to values of the lattice constant at corresponding temperatures. These values were derived by using the Parrinello– Rahman algorithm [26], and they are given in [9]. A constant temperature is kept during dynamics simulation through the application of the Langevin force in the thermostat, as shown in Fig. 1. In the dynamic simulations, coordinates of atoms in a dimer or trimer are tracked as a function of time. Analysis of the atomic trajectories reveals the mechanisms of diffusions. Following the identified mechanisms, molecular statics calculations are carried out to estimate the corresponding energy barriers. In these calculations, along a path of dissociation and/or diffusion, we draw a straight line between the initial and final configurations. At each point in between, the coordinate of one atom parallel to the line is fixed (this coordinate is our reaction coordinate) and the other two coordinates perpendicular to the line are free to relax. Other atoms in the cluster and substrate are free to relax 88 M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 Fig. 2. Top view of the simulation cell with a cluster (dimer): (a) both kinds of steps h1 1 0i=f1 0 0g – the top step, and h1 1 0i=f1 1 1g – the bottom step are present; (b) only h1 1 0i=f1 1 1g-faceted steps are present, and (c) only h1 1 0i=f1 0 0g-faceted steps are present. The dark circles represent substrate atoms, the open circles represent atoms in the island, and the grey circles represent atoms in a cluster. in all directions to minimize the total potential energy. A continuous path of dissociation and/or diffusion is constructed based on the relaxed configurations; in cases where a discontinuous path appeared, another path is drawn between the two discontinuous points. In this way the energy barrier is obtained as the maximum potential energy along the path. If the migration mechanism possesses more than one barrier, further reaction coordinates are constructed from straight lines connecting adjacent minima on the path. All the M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 energy states are given in terms of the formation energy of the relevant cluster. For dimer (n ¼ 2) or trimer (n ¼ 3), the formation energy Ef is defined as Ef ¼ EðN þ nÞ EðN Þ n ð3:36 eVÞ; where EðN þ nÞ is the total potential energy of the simulation cell containing a cluster and EðN Þ is the potential energy of the simulation cell without the cluster. 3. Simulation results We start with the dynamic simulation results to demonstrate the mechanisms of cluster diffusing down the island. They are supplemented by the static calculations of energy barriers. The trajectories of atoms in a dimer from a simulation of 350 ps at 600 K are shown in Fig. 3(a); their time dependence is shown in Fig. 3(b). The dimer was initially placed at the center of an island, and finally diffuses down a step. The simulation is repeated at 300 K for a much longer time. In these two simulations, we have observed two representative mechanisms of dimer diffusion on the island. One is a diffusive concerted sliding and the other is a non-diffusive (confined) atom-by-atom rotation. The energy barrier of atom-by-atom rotation is very small [12]. As indicated by the dark regions in Fig. 3(a) and the plateaus in Fig. 3(b), relative rotations are much more frequent than diffusive jumps. In both simulations, we have never observed the dissociation of a dimer on the island before it reached step edge. The edge evaporation, corner break and surface melting are also absent even in the high temperature (600 K) simulation. These dynamic simulations indicate that: (1) one atom of the dimer diffuses down the {1 1 1}-faceted step through the exchange mechanism (the mechanism is also valid for adatoms) while the dimer bond remains intact, and (2) the dimer dissociates after this atom has been incorporated into the step edge. According to the identified mechanism of dimer diffusing down a step, we carried out molecular statics calculations of the corresponding energy 89 barrier. In Fig. 4, the calculated potential energy of the dimer is plotted versus the reaction coordinate along the path. The initial configuration (coordinate ¼ 0.0) corresponds to the dimer at the edge of the island, and the final configuration (coordinate ¼ 0.5) corresponds to the dimer with one atom incorporated into the step and the other on the island. The potential energy at the initial configuration corresponds to the formation energy of the dimer in this configuration. The corresponding relaxed configurations are shown below the coordinate axis. At first one of the dimer atoms approaches the step, displaces a step-atom from its FCC-site into the adjacent HCP-site. In order to fully replace the step-atom this dimer-atom has to push the step-atom further away. The latter moves to the nearest FCC-site at the step thereby forming a kink and optimizing its coordination. The migration mechanism possesses two transition states, corresponding to bridge-sites between the adjacent FCC- and HCP-sites. The intermediate minimum corresponds to the configuration when the stepatom passes through the HCP-site. The whole process is described by a separate reaction coordinates for each transition state. The two barriers of this diffusion jump are 0.10 and 0.08 eV, respectively. It is worth noting that according to our simulation the dimer prefers to dissociate at the step, with one of the atom diffusing down the step at once, instead of both atoms diffusing down simultaneously. Similar to simulations of the dimer, molecular dynamics simulations at 300 and 600 K indicate that a trimer diffuses very fast on the island. Its diffusion down the {1 1 1}-faceted step is more complex, and consists of two events. At the first event (referred to as process 1), one of the three atoms moves down the step by an exchange mechanism (similar to that of the dimer), forming a kink-site. The trimer thereby dissociates and the remaining dimer starts to diffuse on the island. At the second event the dimer itself diffuses down the step as described before or returns to the kink-site and by another exchange process a double kink is formed (referred to as process 2); both processes are observed in the dynamic simulations. For the first event – one of the three atoms diffusing down the facet – the potential energy of 90 M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 Fig. 3. Atomic trajectories of a dimer at 600 K: (a) a top view, and (b) time dependence of positions of the two atoms. The open circles represent atoms in the island, and the thin solid line and the thick dotted line correspond to the two atoms, respectively. the trimer is shown in Fig. 5(a) (the potential energy of the initial configuration is related to the formation energy of the trimer). Again two transition states occur, when the step atom passes through the bridge-sites as described before. The two energy barriers are 0.10 and 0.12 eV, respectively. For the second process – the remaining dimer diffusing down the facet – either of the two alternatives may occur. The energy barriers for the first alternative have already been described above. The potential energy curve for the other alternative – process 2 – is shown in Fig. 5(b). In this case the relevant step-atom is not free to move through the adjacent HCP-site, which is partially blocked by the kink at the neighboring FCC-site. At the same time a complex motion of the dimer-atoms is observed. This explains the pronounced shoulder in the potential energy curve and the higher single M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 91 Fig. 4. Diffusion of a dimer down the {1 1 1}-faceted step: potential energy along the path, with energy barriers and configurations indicated. The energy at the initial configuration corresponds to the formation energy of the dimer at this position. energy barrier is 0.26 eV. The insets in Fig. 5 indicate the intermediate atomic configurations along the diffusion paths. All these barriers are much smaller than the 0.47 eV for kink incorporation, 0.62 eV for corner breaking, and 0.80 eV for edge evaporation [15]. It is interesting to note that the energy gain of 0.63 eV in process 2 is much larger than the 0.04 eV in process 1. As a result, the reverse process of process 2 has an activation energy of about 1.0 eV. In contrast to the much lower barriers of other processes, such an event is highly unlikely. On the other hand, the inverse process of process 1 has a much lower barrier of about 0.15 eV and indeed, this process is observed in our dynamic simulations. When part of the clusters (one of the dimer atoms or two of the trimer atoms) diffuse down the step, the rest is just the adatom diffusion at the step with one kink (dimer) or double kinks (trimer). There are two possibilities of the adatom diffusion: (1) the adatom diffuses down the kink-step following the first part of the clusters and (2) the adatom migrates away from the kink-step. The calculated energy barriers of the first possible diffusion are 0.11 and 0.16 eV for dimer and trimer, respectively. Indeed, in our molecular dynamics simulations, we observed these two kinds of diffusion of the adatom. Because the adatom has a very small migration barrier on the (1 1 1) surface (<0.1 eV) and small Schwoebel barrier (<0.1 eV) at the steps, it is likely to stay on the island, migrate away from the kink-step, and finally diffuse down the steps by exchanging mechanism. This mechanism has been discussed in detail in [13,14]. For completeness we have also calculated the migration barriers of dimers and trimers diffusing down {1 0 0}-faceted steps, following similar diffusion mechanisms as at {1 1 1}-faceted steps. The results are summarized in Table 1. Consistent with the dynamics simulations, the energy barriers clearly show that diffusion down {1 0 0}-faceted steps is much more difficult than {1 1 1}-faceted steps; the minimum barrier is 0.40 eV for {1 0 0}faceted steps in contrast to 0.10 eV for {1 1 1}faceted steps. Apart from the diffusion down a step, the dynamic simulations also indicate that the clusters have preference to reside near the steps. Based on the calculated formation energies at various locations on the island, we estimate that a dimer or trimer has about 0.1 eV lower formation energy near the steps than on the flat island. 92 M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 Fig. 5. Diffusion of a trimer down the {1 1 1}-faceted step: potential energy along the path (a) when the first atom diffuses (process 1), and (b) when the second atoms diffuses (process 2); the corresponding energy barriers and atom configurations are indicated. The energy of the initial configuration corresponds to the formation energy of the trimer and dimer with kink in this configuration. Table 1 Energetics of a cluster diffusing down a (1 1 1) island Diffusion barrier (eV) Energy gain (eV) Facet h1 1 0i=f1 0 0g h1 1 0i=f1 1 1g h1 1 0i=f1 0 0g h1 1 0i=f1 1 1g Dimer Trimer (process 1) Trimer (process 2) Last atom in the dimer Last atom in the trimer 0.39 0.38 0.13 0.15 0.16 0.10/0.08 0.10/0.12 0.26 0.11 0.16 0.28 0.09 0.65 1.08 0.94 0.27 0.04 0.63 0.95 0.92 For a trimer, one atom diffuses down in process 1, and the second atom follows in process 2. When double barriers exist for one diffusion process, both are listed. M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 4. Conclusions and discussions Using the molecular dynamics and molecular statics methods, we have studied the processes of dimer and trimer diffusing down a (1 1 1) island in aluminum. Our study shows: (1) Diffusion down the h1 1 0i=f1 1 1g-faceted steps is much easier than the h1 1 0i=f1 0 0g-faceted steps. (2) A dimer dissociates during the process of diffusing down the h1 1 0i=f1 1 1g-faceted step. One of the two atoms diffuses down the step by an exchange mechanism. Two transition states are encountered along the path with barriers of 0.10 and 0.08 eV. The other atom remains on the island as a single adatom; if it diffuses down the step following the first, the barrier is 0.11 eV. (3) A trimer also dissociates in the process of diffusing down the h1 1 0i=f1 1 1g-faceted step in two separate diffusion events. At first, one of the three atoms diffuses down the facet with double barriers of 0.10 and 0.12 eV. The other atoms remain on the island as a dimer; if the second atom diffuses down the facet following the first one, the barrier is 0.26 eV. The third atom remains on the island as a single adatom; if it diffuses down the step following the first two, the barrier is 0.16 eV. (4) Diffusion down the other faceted step, h1 1 0i=f1 0 0g, is more difficult. In particular, the diffusion barrier of the first atom is about 0.2 eV higher than down the h1 1 0i=f1 1 1g step. For practical purposes, this process can be ignored in surface evolution of aluminum below half of the melting temperature. As in any classical molecular dynamics simulations, the reliability of the results depends on the interatomic potential. The EAM potential of aluminum [23] used in this paper provides an accurate description of surface formation energies, adatom formation energy and migration energy on the (1 1 1) surface [9,23]. Therefore, we expect that (a) the identified mechanisms are reliable, and (b) the calculated energetics on the (1 1 1) surface are reliable at least in relative magnitude. Further, diffusion processes strongly depend on the under- 93 lying crystal structure, further ensuring the reliability of the identified mechanisms and the calculated energetics. More accurate determination of the absolute magnitude of the energetics is being carried out using ab initio methods. Acknowledgements The work described in this paper was substantially supported by a central research grant from the Hong Kong PolyU (G-V943), partially by a grant from Deutsche Forschungsgemeinschaft under SFB 292 ‘‘Multicomponent Layered Systems’’, and partially by grants from the Research Grants Council of the Hong Kong Special Administrative Region (PolyU 5146/99E and Group Research project: Computer Aided Materials Engineering). The joint financial support from the Hong Kong RGC and the German DAAD is also gratefully acknowledged. References [1] P. Ruggerone, C. Ratsch, M. Scheffler, in: D.K. King, D.P. Woodruff (Eds.), Growth and Properties of Ultrathin Epitaxial Layers, The Chemical Physics of Solid Surfaces, vol. 8, Elsevier, Amsterdam, 1997, pp. 490–544. [2] H. Huang, G.H. Gilmer, T. Diaz de la Rubia, J. Appl. Phys. 84 (1998) 3636. [3] H. Huang, G.H. Gilmer, J. Comput. Aided Mater. Des. 6 (1999) 117. [4] G.H. Gilmer, H. Huang, T. Diaz de la Rubia, C. Roland, Comput. Mater. Sci. 12 (1998) 354. [5] G.H. Gilmer, H. Huang, T. Diaz de la Rubia, J.D. Torre, F. Barumann, Thin Solid Films 365 (1999) 189. [6] H. Huang, G.H. Gilmer, Comput. Mater. Sci. (in press). [7] H. Huang, G.H. Gilmer, J. Comput. Aided Mater. Des. (in press). [8] W.C. Liu, C.H. Woo, H. Huang, J. Comput. Aided Mater. Des. 6 (1999) 311. [9] H. Huang, G.H. Gilmer, T. Diaz de la Rubia, to be published. [10] O. Pankratov, M. Scheffler, Surf. Sci. 309 (1994) 1001. [11] O. Pankratov, M. Scheffler, Phys. Rev. Lett. 75 (1995) 701. [12] A. Bogocevic, P. Hyldgaard, G. Wahnstroem, B.I. Lundqvist, Phys. Rev. Lett. 81 (1998) 172. [13] R. Stumpf, M. Scheffler, Phys. Rev. Lett. 72 (1994) 254. [14] R. Stumpf, M. Scheffler, Phys. Rev. B 53 (1996) 4958. 94 M. Bockstedte et al. / Computational Materials Science 23 (2002) 85–94 [15] A. Bogicevic, J. Stroemquist, B.I. Lundqvist, Phys. Rev. Lett. 81 (1998) 637. [16] S. Ovesson, A. Bogicevic, B.I. Lundqvist, Phys. Rev. Lett. 83 (1999) 2608. [17] B.G. Liu, J. Wu, E.G. Wang, Z. Zhang, Phys. Rev. Lett. 83 (1999) 1195. [18] S. Vaidyanad, A.K. Sinha, Thin Solid Films 75 (1981) 523. [19] H. Onoda, M. Kageyama, K. Hashimoto, J. Appl. Phys. 77 (1995) 885. [20] S.C. Wang, G. Ehrlich, Surf. Sci. 301 (1990) 239. [21] J.M. Wen, S.L. Chang, J.W. Burnett, J.W. Evans, P.A. Thiel, Phys. Rev. Lett. 73 (1994) 2591. [22] M. Bockstedte, A. Kley, J. Neugebauer, M. Scheffler, Comput. Phys. Commun. 107 (1997) 187. [23] F. Ercolessi, J.B. Adams, Europhys. Lett. 26 (1994) 583. [24] M.S. Daw, M.I. Baskes, Phys. Rev. Lett. 50 (1983) 1285. [25] M.S. Daw, M.I. Baskes, Phys. Rev. B 29 (1984) 6443. [26] M. Parrinello, A. Rahman, J. Appl. Phys. 52 (1981) 7182.
© Copyright 2026 Paperzz