Chapter 7 Grain boundary effect in the coarsening of polycrystalline solid films 7.1 Introduction In this chapter, we investigate the impact of grain boundary diffusivity on the kinetics of spinodal decomposition and coarsening in polycrystalline solid films. These films are used extensively in the electronic, magnetic, chemical, photonic and microelectromechanical devices and systems. In films, as in bulk materials, the grain boundaries act as fast diffusion paths due to their high density of defects. Here, we include these fast diffusion paths by tying mobility to a specific phase field variable used to distinguish grain boundaries. There exists a great deal of literature for modelling phase transformation with either constant mobility or mobility that depends on composition or order-parameter [81, 82, 22, 23, 24, 25]. In this chapter, we investigate a variable mobility field owing to grain boundaries. We consider a free-standing polycrystalline solid film with both fixed and migrating grain structures. The mobility in the CH model is assumed to be a function of the grain structure, where the magnitude of mobility is much higher on the grain boundaries than in the grain. The organization of this chapter is as follows. In Sections 7.2, the governing equations for composition evolution and grain boundary migration are given. In Section 7.3, the numerical method as well as an analysis of the relative time scales for grain boundary migration and composition diffusion is presented. Simulation results are presented in Section 7.4. 7.2 Governing equations We adopt a phase field or a generalized Cahn-Hilliard (CH) equation for phase separation. The general form of this model for a binary alloy is given in Eqn. (2.13) as µ µ ¶¶ ∂f (C) ∂Eel ∂C = ∇ · B(C, Ψ(x, t))∇ + − β∆C , ∂t ∂C ∂C (7.1) 7.2 Governing equations 119 where C is the composition, B(C, Ψ(x, t)) is the mobility of the atomic species described by C, f (C) is the stress-free chemical free energy density, Eel is the elastic energy density, and β is a measure of interfacial penalty for the rapid variation of C. We choose f (C) to be a multiwelled function. Also, β is assumed to be constant. If the surface energy is composition-independent and the body has no mass exchange with the environment, then we have the following boundary conditions: ∂C = 0, ∂n where JCH = −ρB(C, Ψ(x, t))∇ (7.2) JCH · n = 0, and µ ∂f (C) ∂Eel + − β∆C ∂C ∂C ¶ (7.3) is a flux derived from the gradient of effective diffusion potential and ρ is an atomic density. Also, n is the outward normal of the body surface. In a periodic case, these boundary conditions need to be adjusted appropriately. In the expression for the mobility B(C, Ψ(x, t)) in Eqn. (7.1), we have used Ψ(x, t) (possibly a vector) to describe a general phase field (order parameter) variable. In this chapter, Ψ(x, t) will be used to describe the grain boundary structure. Note that in the literature, special forms of composition dependent mobility, such as B = 1 − αC 2 (Langer et al. [81]) or B = C(1 − C)(Cm1 + (1 − C)m2 ) (Zhu et al. [24]), have been considered independent of grain boundary structure. However, in our model we consider a dependence of the mobility on the crystallographic grain boundaries (as described by a particular choice of Ψ(x, t)) and assume the mobility is independent of the composition profile C(x, t). The grain structure in a polycrystalline body is associated with a distribution of diffu- sivity, where the diffusivity on grain boundary, DGB , is orders of magnitude greater than the diffusivity in the grain, DL especially at temperatures below 0.6Tm , where Tm is the melting point of matrix. Kaur et al. [83] gives the ratio DGB /DL as at least 105 . Further, the mobility B in the CH equation can be shown to be connected with the diffusivity D via a simple model as follows. Ignoring the elastic and interfacial effects, the two fluxes: J CH in the CH model (see Eqn. (7.3)) and JF from Fick’s law are JCH ≈ −ρB∇ µ ∂f (C) ∂C ¶ ≈ −ρB · ¸ ∂ 2 f (C) ∇C, ∂C 2 (7.4) and JF ≈ −D∇C. (7.5) 7.2 Governing equations 120 Since these two fluxes, JCH and JF both describe the rate of change of composition, B and D must be connected by a certain scaling. Here, since we are not interested in the composition dependence of B (or D), we may assume that B follows D in its dependence on the grain structure. That is, for fixed grain boundaries, we assume BGB /BL = DGB /DL . For the case where grain boundaries are allowed to migrate, a grain boundary structure is projected from Ψ(x, t) and the mobility B is set as in the fixed grain boundary case. The migrating grain boundary case is further discussed in Section 7.2.1. 7.2.1 Grain boundary migration Grain boundary migration (grain growth) occurs because of a decrease of the energy owing to shrinking the grain boundaries. Grain boundary migration is common in polycrystals and has been treated by a number of authors using a variety of theoretical and experimental approaches [84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94]. There has also been much work done to numerically simulate grain boundary motion. Kinetic Monte Carlo Potts model (see [55] for a review), Phase field models have also been used to simulate the evolution of grain boundaries. For example, Chen et al. performed 2D simulations of the grain-growth with a large number of nonconserved order parameters [56, 57]. Fan and Chen [95] studied the effect of ratios of grain boundary energies to interfacial energy on the microstructure in the isotropic case. Here, isotropic means that the grain boundary energy and the mobility for the migration are both independent of the crystallographic orientation of the grain and the boundary orientation. Kazaryan et al. [58, 59] investigated the dynamics and morphology of grain growth for anisotropic grain boundary energy and anisotropic grain boundary mobility using a generalized phase-field model. In this work, we consider only the case of isotropic grain boundaries. We denote the crystallographic orientations of the polycrystals, using a total number of p = 36 order parameters ηi (i = 1..p). Roughly, this means that we divide the 2D orientation space of the crystal into 10◦ intervals. We assume the following free-energy density functional [56]: f P (η1 , η2 , · · · , ηp ) = p X p p h X X c1 c2 i ηi2 ηj2 , − ηi2 + ηi4 + c3 2 4 i=1 (7.6) i=1 j>i where ‘P’ means phase field and c1 , c2 , c3 are phenomenological parameters. We choose the following values: c1 = 1.0, c2 = 1.0, c3 = 1.0. The energy in Eqn. (7.6) then ensures that for each (x, t), at most only one element in ηi (i = 1..p) has value close to ±1 and the other elements are all close to 0. With the free-energy density in Eqn. (7.6), the kinetics of the evolution of the order 7.2 Governing equations 121 parameters are given by the Allen-Cahn equation [56]: p X dηi = −µi −c1 ηi + c2 ηi3 + 2c3 ηi ηj2 − λi ∇2 ηi dt j6=i (i = 1, 2, · · · , p, no sum), (7.7) where the µi are the kinetic coefficients and the λi are energy penalties owing to grain boundaries. Note in Eqn. (7.7), no summation rule is employed. Since we consider the isotropic case, we set µi = µ̄ and λi = λ̄ for all i, where µ̄ and λ̄ are reference values. In [56], a grain structure (described by Ψ(x, t)) based on the ηi is given as Ψ(x, t) = p X ηi2 (x, t). (7.8) i=1 The variable Ψ(x, t) has low values on the grain boundaries and is close to 1 inside grains. Hence we propose the following form of the mobility for our CH model: M (Ψ(x, t)) ≡ B/B0 = R, 1, |1 − Ψ(x, t)| ≥ Th (on grain boundaries) |1 − Ψ(x, t)| < Th (inside grains), (7.9) where B0 is a reference mobility and is chosen to be the mobility in the grain BL , R is set to BGB /BL (assumed a constant) and Th is some threshold (typically taken to be 0.2). Equations (7.8) and (7.9) qualitatively describe the features of the grain boundaries and their effect on the mobility in the CH model. 7.2.2 Contribution of elasticity The term ∂Eel /∂C in the evolution equation (7.1) is associated with an elastic energy contribution arising from composition dependence of the lattice parameter of the film. This term has been considered extensively in both film and bulk phase transformations. However, the elasticity problem associated with the polycrystal structure is very complicated because of the inhomogeneity of elastic moduli introduced by the grain boundaries. While it is possible to assume isotropic elasticity for the film (so grain boundary rotation does not lead to inhomogeneity), or to use a homogenized equivalent medium, neither option seems particularly physical. Hence we choose to neglect elasticity. 7.3 Numerical analysis 122 7.3 Numerical analysis 7.3.1 Nondimensionalization The nondimensionalization of Eqn. (7.1) follows as in the previous chapter (Section 6.2.1). We introduce length scale L̄ and time scale T̄ , and nondimensionalize all spatial and temporal variables. We also introduce ² = β/β0 and M = B/B0 (see also Section 7.2.1) as the nondimensional measures of interfacial penalty and mobility respectively, where β 0 and B0 are reference values. We assume the stress-free chemical free energy f is the same as in Eqn. (6.8). By introducing Γ = (2C − Cα − Cγ )/(Cγ − Cα ), f in Eqn. (6.8) can be nondimensionalized as f = W/K 4 f¯, where f¯ = (Γ2 − 1)2 and K = 2/(Cγ − Cα ). By substituting all the nondimensional variables into Eqn. (7.1) and setting β0 B0 T̄ = 1, L̄4 and W L̄2 = 1, K 2 β0 (7.10) we get the length scale L̄ and the time scale T̄ as: L̄ = µ β0 W ¶1/2 K, and T̄ = β0 K 4. B0 W 2 (7.11) The nondimensional governing equation and boundary conditions then can be written as ∂ f¯ Q = ∂Γ − ²∆Γ, J = −M ∇Q, ∂Γ = −∇ · J, ∂τ ∂Γ = 0, J·n=0 ∂n (7.12) on boundary, where Γ, Q, J are nondimensional composition, effective diffusion potential and flux respectively. Note also that periodic boundary conditions can be assumed instead of those given in Eqn. (7.12). We then consider the evolution of grain orientation order parameters. Using the time and length scales given in Eqn. (7.11), Eqn. (7.7) is nondimensionalized to dηi = Rτ λL ∇2 ηi − −c1 ηi + c2 ηi3 + 2c3 ηi dτ p X j6=i where Rτ = µ̄T̄ , and ηj2 λL = (i = 1..p, no sum on i) (7.13) λ̄ . L̄2 (7.14) 7.3 Numerical analysis 123 Note µ̄ is of dimension [T −1 ] hence Rτ is dimensionless. In the next subsection, we will show that Rτ is a measure of the relative mobility of grain boundary migration to composition diffusion. Also, Rτ is estimated based on experimental data. 7.3.2 Determination of Rτ Fan and Chen [96] show that for isotropic grain boundary energy and relaxation constants, within the diffuse-interface description, a grain boundary moves at velocity V due to its mean curvature κ as V = µ̄λ̄κ, (7.15) where µ̄ and λ̄ are reference values for µi and λi introduced in Section (7.2.1). For a half-loop constant driving force technique [93], Gottstein and Shvindlerman showed V = 2Ab , a (7.16) where a is the diameter of the half-loop and Ab is a reduced mobility which can be measured experimentally. In this half-loop technique, κ in Eqn. (7.15) is 1/a, so µ̄λ̄ = 2Ab . (7.17) Introducing Sτ ≡ Rτ λL , and using Eqns. (7.14), (7.17) and (7.11), we have Sτ = µ̄λ̄ T̄ K2 = 2A . b B0 W L̄2 (7.18) Using Eqns. (7.4) and (7.5), we have a rough relation between B0 and the bulk diffusivity D0 as in Eqn. (6.23): B0 ≈ D0 K 2 , W Rd (7.19) where Rd is the mean value of the second derivative of the stress free chemical energy to composition at its wells. In our model, Rd is roughly 8. With Eqn. (7.19), Eqn. (7.18) leads to Sτ ≈ 2Rd Ab Ab ≈ 16 . D0 D0 (7.20) By Eqn. (7.20), Sτ (hence Rτ ) measures the relative mobility of grain boundary migration and diffusion. From [93], Ab generally has a wide range of values depending on factors such as species, orientations, etc. The typical range for Ab is 10−16 ∼ 10−8 m2 /s. From standard texts on diffusion (such as [97]), the typical range for D0 is 10−16 ∼ 10−13 m2 /s. Thus, we expect that Sτ can have a range of values from 10−2 ∼ 109 . For specific materials 7.3 Numerical analysis 124 and environments, one can determine a narrower range for Sτ . In this work, we choose λ̄ such that λL = h2 /2 (see Eqn. (7.14)), where h is the uniform nondimensional spatial step; hence Rτ in Eqn. (7.13) is Rτ = Sτ /λL = 2Sτ /h2 . A typical range is Rτ ∈ (1, 1010 ). Note that by Eqn. (7.20), if Ab ≈ D0 , one expects that the grain migration and bulk diffusion have roughly the same time scale. In this situation, S τ ≈ 16 and Rτ ≈ 800 with h = 1/5 used in our simulation. The role of Rτ in the coupled process of diffusion and migration is shown in the simulation results in Section 7.4.3. 7.3.3 Numerical scheme The time marching scheme for Eqn. (7.12) follows as in the previous chapter (Section 6.2.3). However, we introduce three sets of grid in the physical domain. To simplify the description, suppose the spatial step in x1 , x2 , x3 are all h. The physical domain is discretized as (i, j, k)h, where i = 0..L1 /h, j = 0..L2 /h, and k = −Hf /h..Hf /h. The grid for the flux coincides with the this physical grid; while the composition Γ and potential Q are computed on a grid (i + 21 , j + 21 , k + 21 )h. Hence the Laplacian operator ∆ is taken on the Γ grid, the gradient operator ∇ is taken on the Q grid and generates flux on the J grid, and the diver- gence operator ∇· is taken on the J grid and produces results on the Γ (Q) grid. Since the gradient and divergence operators have different object size, we denote their components e i· (as in x1 , x2 and x3 direction) as follows: the divergence operator ∇· is denoted as ∇ e e i (i = 1, 2, 3). Then, the following (i = 1, 2, 3), and the gradient operator ∇ is denoted as ∇ time marching scheme is used: µ ¶ ¡ ¢ Γn+1 − Γn e n+1 2 n+1 n ei · M∇ e i [4(Γ =∇ ) + Ac − ²D∆ ]Γ +R ∆τ (i is summed), (7.21) where D∆ is the discrete Laplacian operator operating on the composition Γ, and R n = −(4 + Ac )Γn is the contribution from time level n. Note the nondimensional mobility M should be evaluated on the same grid as J. If the Γ grid points are gathered into a vector by a lexicographic ordering of the nodes, e e i, ∇ e i are gathered into matrix form (still denoted by and similarly the operators D∆ , M , ∇ the same symbol), the scheme in Eqn. (7.21) can be written as h I − ∆τ [DM ][diag(4(Γn+1 )2 ) + Ac I − ²D∆ ] i£ ¤ Γn+1 = [Γn ] + ∆τ [DM ]Rn , (7.22) where DM is a mobility dependent pseudo Laplacian operator: e e iM ∇ ei DM = ∇ (i is summed). (7.23) 7.4 Simulation results 125 The grain boundary migration system in Eqn. (7.13) is integrated explicitly in time using a standard fourth order Runge-Kutta marching scheme. Periodic boundary conditions for ηi (i = 1..p) are used. 7.4 Simulation results In this section, simulation results are presented to show how enhanced grain boundary diffusion affects phase separation and coarsening in solid films. We consider first a fixed grain boundary structure in two dimensions. Then, we consider a fixed three-dimensional columnar grain boundary structure obtained by extending a two-dimensional one into the third dimension. We then introduce migrating grain boundaries in two dimensions. The following nondimensional parameters apply to all cases: ² = 0.1 and unless otherwise stated the magnitude of the ratio MGB /ML of grain boundary to bulk mobility is taken to be in 103 (recall from Section 7.2 that this ratio can be as high as 105 ). This choice qualitatively describes the phenomena occurring in the evolution without sacrificing numerical performance. Specifically, choosing too high a value of MGB /ML may cause some numerical issues due to the large range of time scales. The simulations are performed on a computational domain of 51.2 × 12.8 (with a grid 256 × 64) for two-dimensional cases and 12.8 × 12.8 × 3.2 (with a grid 64 × 64 × 16) for three-dimensional cases. All plots are rendered with grain boundary structure and composition overlapped. The white and black areas represent the phases of Γ ≈ +1 and Γ ≈ −1 respectively. The underlying grain boundaries are shown in gray and can be easily seen. 7.4.1 Results in two dimensions with fixed grain boundaries We first consider a two-dimensional case with fixed grain structure Rτ = 0. The two phases are distributed in such a way that one of the phase (Γ ≈ +1) is on the grain boundaries, while the other (Γ ≈ −1) resides in the grains, as shown in Figure 7.1 (the gray-scale scheme is shown in Figure 7.2). The average composition is Γ0 = −0.8062. The background grain boundary structure is taken from a binary image of an experimental picture from a graphics website. It is slightly modified to be periodic in the horizontal direction and also thickened digitally so that the boundaries themselves are a few grid points wide. Figure 7.2(a) shows a case with the relative mobility ratio MGB /ML = 1000. The Γ ≈ +1 phase on the grain boundary diffuses immediately toward the triple junctions and coarsens. To compare, Figure 7.2(b) depicts a case with MGB /ML = 1. It is clear that in Figure 7.2(a), the grain boundaries are the main paths for coarsening and the resulting particles are retained in the triple junctions, with a shape conformal to the junction. On the 7.4 Simulation results 126 Figure 7.1: An initial composition profile for 2D simulation. The two phases are distributed in such a way that one of the phases (Γ ≈ +1) is on the grain boundaries, while the other (Γ ≈ −1) resides in the grains. The gray-scale scheme is shown in Figure 7.2. (a) (b) Figure 7.2: Composition evolution for a 2D case with initial profile in Figure 7.1. (a) The relative mobility ratio MGB /ML = 1000. The Γ ≈ +1 phase diffuses immediately toward the triple junctions and coarsens. (b) The mobility ratio MGB /ML = 1. Standard coarsening occurs. 7.4 Simulation results 127 (a) (b) Figure 7.3: Two grain structures with the same initial composition profile for a simulation with Rτ = 1. (a) A relative large grain structure; (b) A very large grain structure. The gray-scale scheme is shown in Figure 7.4. (a) (b) Figure 7.4: The evolution of a particle system in a polycrystalline solid film with large grains. MGB /ML = 1000. The grain boundary migration is very slow, with Rτ = 1, thus emulating a ‘fixed’ grain boundary. Coarsening occurs with preference on the grain boundary. Note the shaded color indicates channels with fast diffusion rate. (a) A relative large grain structure; (b) A very large grain structure. 7.4 Simulation results 128 contrary, standard coarsening behavior occurs in Figure 7.2(b) and there is no preferential coarsening direction. Also, note that coarsening with grain boundary diffusion is much faster than without grain boundary diffusion. We also consider a case in the migrating grain scenario (Section 7.4.3) but with R τ = 1, so the grain boundary is essentially fixed. Figure 7.4 depicts two such cases for large grain films. The initial composition profile and grain structures are shown in Figure 7.3. Note that in the figures (also in the following figures in this chapter), the part of the particles with Γ ≈ +1 over the grain boundary is shaded due to the contrast of colors. These shaded regions hence can be treated as an indicator of channels with fast diffusion rate. One observes that coarsening is accelerated along the grain boundaries. For phases inside a grain, there are only very small variations of composition due to the low bulk mobility. Γ 1.10 τ = 0.0000000 Γ 1.10 0.55 0.55 0.00 0.00 -0.55 -0.55 -1.10 -1.10 τ = 0.0000000 Figure 7.5: A three-dimensional initial profile with small range random composition around Γ = 0. The average composition is Γ0 = 0. The plot is overlapped with a threedimensional fixed grain boundary. The grain boundary structure is extended from a twodimensional grain boundary profile. On the left is a top view, on the right a side view. 7.4.2 Results in three dimensions with fixed grain boundaries A columnar grain boundary structure in three-dimensions is obtained by extending the two-dimensional fixed grain boundary structure into the third dimension (thickness direction). Though more complex grain boundary structure can be generated, for example, by Voronoi tessellation or other methods, we don’t consider them in this work. We investigate a case with a random initial composition profile with a small range around the critical composition Γ = 0, as depicted in Figure 7.5. The average composition is Γ0 = 0. The relative mobility ratio MGB /ML = 1000. The evolution process is shown in Figure 7.6 using both a top view and a side view. Figure 7.6 shows that, as in the two-dimensional case, the grain boundary and its junctions serve as fast channels for phase separation and coarsening. Spinodal decomposition 7.4 Simulation results Γ Γ Γ Γ 1.10 τ = 0.0000700 129 Γ 1.10 0.55 0.55 0.00 0.00 -0.55 -0.55 -1.10 -1.10 1.10 τ = 0.0007100 Γ 1.10 0.55 0.55 0.00 0.00 -0.55 -0.55 -1.10 -1.10 1.10 τ = 0.0036300 Γ 1.10 0.55 0.55 0.00 0.00 -0.55 -0.55 -1.10 -1.10 1.10 τ = 0.0079500 Γ 1.10 0.55 0.55 0.00 0.00 -0.55 -0.55 -1.10 -1.10 τ = 0.0000700 τ = 0.0007100 τ = 0.0036300 τ = 0.0079500 Figure 7.6: Composition evolution in a free-standing film with a fixed columnar grain structure in three dimensions. MGB /ML = 1000. Spinodal decomposition occurs in the plane and along the thickness direction on the grain boundary. The initial composition profile is in Figure 7.5. On the left is a top view, on the right a side view. 7.4 Simulation results 130 occurs first along the grain boundaries, followed by coarsening. 7.4.3 Results in two dimensions with migrating grain boundaries Grain boundaries and phase interfaces are distinct, i.e., in one grain, there can be two phases; while in a single phase, there can be two crystallographic orientations. Hence grain boundaries can migrate at the same time that the composition evolves. As described in Eqn. (7.9), the migrating grain boundaries (Ψ(x, t)) generate a time-dependent mobility M for composition diffusion, hence depending on the magnitude of the relative mobility Rτ , migrating grain boundaries can have a strong effect on phase separation. This can be viewed as follows. As grains grow, fast diffusion channels are established or removed between adjacent particles. When the channel is present, coarsening occurs at a faster rate; when it is not present, the coarsening rate drops significantly. The interaction between these two mechanisms gives complicated behavior for the composition evolution. An initial composition profile is generated for a polycrystalline grain structure as shown in Figure 7.7. The average composition is taken to be Γ0 = 0. Figures 7.8(a) and 7.8(b) depict two cases with the same initial conditions except Rτ = 1K and 10K respectively. In both cases, MGB /ML = 1000 and ² = 0.1. As expected, the grain boundary structure moves faster as Rτ increases. Figure 7.7: A particle initial composition profile in a polycrystalline solid film. The grayscale scheme is shown in Figure 7.8. Figure 7.8(a) depicts the evolution for Rτ = 1K. As stated earlier, this corresponds to the situation where the grain migration and bulk diffusion occurs at roughly the same rate. The grain structure shows noticeable changes over the elapsed time. For such a rate, enough time is allowed for particles to coarsen and the topology of the shaded regions shows smooth transitions. Recall that the shaded region is the part of phase Γ ≈ +1 which stays over the grain boundary (hence is subject to a fast coarsening rate). When Rτ increases, grain evolution occurs faster than the composition evolves. Figures 7.8(b) depicts the case for Rτ = 10K. Since the fast diffusion channels move through the domain quickly, the composition is essentially “frozen”. Compared to Figure 7.8(b), the topology of shaded regions in Figure 7.8(a) shows smoother transitions. 7.4 Simulation results (a) 131 (b) Figure 7.8: The evolution of a particle system in a polycrystalline solid film. For both cases, MGB /ML = 1000. (a) Rτ = 1K. Noticeable change of the grain boundary structure occurs. (b) Rτ = 10K. Grain boundaries move faster. The fast growing grains freeze some of the composition when the grain boundary moves away quickly. 7.5 Summary and discussion 132 7.5 Summary and discussion In this chapter, we use a CH model to investigate how variable mobility owing to grain boundary structure affects phase separation and coarsening in polycrystalline solid films. The mobility, which is roughly proportional to the diffusivity, is taken to be much greater on the grain boundaries than in the grain. Both fixed and migrating grain boundary structure are considered. For grain boundary migration, a phase field model is adopted and a grain structure is projected from the order parameters. In this case, the relative mobility of the grain boundary migration to bulk diffusion is found to be a key factor. Numerical simulations are performed in two dimensions for a fixed grain boundaries and in three dimensions for a fixed columnar grain structure. The grain boundaries are observed to accelerate the coarsening, especially at the triple junctions. Simulations of migrating grain boundaries are performed in two dimensions. If the relative mobility of grain migration is large enough, then the migrating grain structure can freeze the composition on the time scale of grain boundary motion. For the migrating grain boundary cases, we assume specific forms for both the phase field model describing the grain structure, and the dependence of the mobility on the grain structure. For the fixed grain structures, the width of the grain boundary is taken to be relatively thick, and the ratio of grain boundary diffusivity to bulk diffusivity is limited by the numerical implementation. Finally, other grain boundary related mechanisms, such as grooving, pinning, solute segregation, etc., are not considered. Some of these features will be addressed in future investigations.