next up previous contents
Next: 3 Discretization of the Up: 8 Numerical Handling of Previous: 1 Discretization of the


2 Discretization of the Simple Extrinsic Diffusion Model

We take the functions $ w(\mathbf{x})$, $ v(\mathbf{x})$ and $ u(\mathbf{x})$ defined on bounded open domen $ \Omega$. Starting from Green's theorem, a simple relationship can be proved,

$\displaystyle \int\limits_\Omega \nabla v \nabla w u d\Omega + \int\limits_\O...
...w v d\Omega=\int\limits_\Gamma v u  \frac{\partial w}{\partial n} d\Gamma,$ (141)

where $ \Gamma$ is the boundary of the domain $ \Omega$. We write (3.33) as

$\displaystyle \nabla (h(C_A)D)  \nabla C_A + D h(C_A)  \triangle C_A = \frac{\partial C_A}{\partial t}.$ (142)

By multiplying (3.79) with the basis function $ \varphi_i = \varphi_i(\mathbf{x})$ and integrating over the domen $ \Omega$ we have,

$\displaystyle \int\limits_\Omega \nabla (h(C_A)D)  \nabla C_A  \varphi_i d\O...
...mega = \int\limits_\Omega \frac{\partial C_A}{\partial t}  \varphi_i d\Omega.$ (143)

Applying (3.78) on (3.80) gives,

$\displaystyle D\int\limits_\Gamma h(C_A)  \varphi_i \frac{\partial C_A}{\part...
...Omega = \int\limits_\Omega \frac{\partial C_A}{\partial t}  \varphi_i d\Omega.$ (144)

Assuming zero-Neumann boundary conditions on $ \Gamma$ we obtain the weak formulation of the equation (3.33)

$\displaystyle \int\limits_\Omega \frac{\partial C_A}{\partial t}  \varphi_i d\Omega+D\int\limits_\Omega h(C_A)  \nabla C_A  \nabla \varphi_i d\Omega = 0.$ (145)

By introducing time discretization with time step $ \Delta t$ and writing the last equation for the single element $ T\in T_h(\Omega)$ we obtain,

$\displaystyle \int\limits_T (C^n_A- C^{n-1}_A)  \varphi_i d\Omega+ D\Delta t \int\limits_T h(C_A^n)  \nabla C_A^n  \nabla \varphi_i d\Omega = 0.$ (146)

The scalar functions $ C_A(\mathbf{x},t_n)$, $ \nabla C_A(\mathbf{x},t_n)$, and $ h(C_A(\mathbf{x},t_n))$, for $ t_n=n\cdot \Delta t$, are linearly approximated on the element $ T\in T_h(\Omega)$,
$\displaystyle C_A^n=C_A(\mathbf{x},t_n) = \sum_{i=1}^{4} \varphi_i C^{n}_{A,i},$     (147)
$\displaystyle \nabla C_A^n = \nabla C_A(\mathbf{x},t_n) =\sum_{i=1}^{4} \nabla \varphi_i C^{n}_{A,i},$     (148)
$\displaystyle h(C_A^n)= h(C_A(\mathbf{x},t_n))= h(C^{n}_{A,m}).$     (149)

The $ C_1^n, C_2^n, C_3^n, C_4^n$, are the values of the concentration at the nodes of the element $ T$, $ C^{n}_{m}$ is the concentration value at some point inside the $ T$. Normally, for $ C^{n}_{A,m}$ we use the following simple approximation,

$\displaystyle C^{n}_{A,m} = \frac{1}{4} \sum_{i=1}^{4} C^{n}_{A,i}.$ (150)

We can define discrete operator $ \ell^e$ for each $ T\in T_h(\Omega)$. Since we have to deal only with a single partial differential equation and not with a system of equations, we can omit the second index in (2.27) by defining the operator,

$\displaystyle \ell_{i}^e(C_{A,1}^n,C_{A,2}^n,C_{A,3}^n,C_{A,4}^n)=\frac{1}{\Del...
...{ik} C_{A,i}^n+D h(C_{A,m}^n)\sum_{i=1}^4 K_{ik} C_{A,i}^n,\quad k=1,2,3,4.$ (151)

The nucleus matrix is than defined as

$\displaystyle \mathbf{\Pi}(T)=\frac{\partial (\ell_{1}^e,\ell_{2}^e,\ell_{3}^e,\ell_{4}^e)}{\partial (C_{A,1}^n,C_{A,2}^n,C_{A,3}^n,C_{A,4}^n)}.$ (152)

and the residuum vector as,

$\displaystyle R_k^e=\ell_{k}^e(C_{A,1}^n,C_{A,2}^n,C_{A,3}^n,C_{A,4}^n)-\frac{1}{\Delta t}\sum_{i=1}^4 M_{ik} C_{A,i}^{n-1},\quad k=1,2,3,4.$ (153)

The nucleus matrix for this simple case can be also calculated analytically. In this case is also $ rank(\mathbf{G})=N$.


next up previous contents
Next: 3 Discretization of the Up: 8 Numerical Handling of Previous: 1 Discretization of the

J. Cervenka: Three-Dimensional Mesh Generation for Device and Process Simulation