2 Adaptivity of the Finite Difference Method
The solution of the atomic mixing model for a material
system of n species involves the solution of n
partial differential equations.
In practical situations, the equations that arise can only
be solved numerically and hence the developing solution
will always contain numerical error. The finite difference
or finite element discretisation can be applied in the
spacial coordinate, it is natural to apply
a dose-stepping method along the other axis.
In general,
the finer the mesh (in both space and dose)
the greater the accuracy of the method, but the slower the
processing speed.
In reference [4],
the implicit FDM that solves
the partial differential equations in the mixing region is
described in detail. Unlike the explicit method that is employed
in the original IMPETUS code, the explicit method
has no stability restriction on the dose step. One further
difficulty with the explicit method is that halving the
size of the spacial interval automatically required the
dose step to be reduced to a quarter of its size. Hence, in
cases where a finer spacial mesh is required, the dose step
needs to be severely reduced and thus the processing time
greatly increased.
In IMPETUS II, the size
of the dose step is restricted only by accuracy considerations.
There is no direct relationship between the size of the
dose step and the spacial mesh size. Thus the implementation
of the implicit method also gives us much greater freedom to
control or adapt the method to the problem in hand. IMPETUS II
contains methods for adjusting the length of both the spacial interval
and the dose step.
2.1 Adjusting the Spacial Interval
In our experience with the method on a variety of material
systems, it is generally found that an inter-nodal distance of 2
or 3 angstroms is generally sufficient to give a satisfactory
solution. However, in the case of a thin layer (for example 20 angstroms
or less), it is natural to refine the mesh so that the
layer is well-represented. Atomic diameters are generally of
the order of a few angstroms and hence there is a physical lower limit on
the realistic layer thickness. If we assume that around ten nodes
are sufficient to represent a thin layer and the layers are
at least say 5 angstroms thick then the minimum inter-nodal
distance expected is around 0.5 angstroms.
In the IMPETUS II algorithm, sharp interfaces are only
allowed into the domain of the FDM when they have become
sufficiently smoothed by the application of the linear diffusion methods.
Once they enter the domain of the FDM, the mesh may need
refining to satisfactorily represent the interface.
In the execution of IMPETUS II, the domain of the
FDM progresses through the extent of the
material (see figure 1 of reference [4]).
In IMPETUS II the spacial mesh is always uniform. The
adaptivity is achieved simply by continually halving the inter-nodal
distance should a thin layer or sharp interface exists in the
window at any particular time until the mesh is
sufficiently fine to represent the material distribution.
The inter-nodal distance is
increased when the thin layers or interfaces become more
diffused. The control of the mesh size is not directly based
on a mathematical analysis of the error, rather it is
guided by the desire to obtain a good node-to-node
representation of the material distribution within the
range of the FDM.
2.2 Adjusting the Dose Step
In IMPETUS II a measure of the
accuracy of the numerical method from dose step to dose step
is obtained through simply summing the concentrations
in the material distribution at the new dose and
observing the drift from unity. (Note that in IMPETUS II
the concentrations are adjusted after each dose step so that
they do sum to unity, and the method for carrying this
out is described in section 5.) The drift
from unity is used to control
the dose step; if the measure of accuracy is larger than
expected then the dose step is decreased and if it is smaller the
dose step is increased.
2.3 Solution to Test Problems
The test structure of figure 1 is used to demonstrate the facility
of IMPETUS II to deal effectively with thin layers. The
structure consists of four layers of pure Germanium of thicknesses
40, 20, 10 and 5 angstroms with sharp interfaces in pure Silicon.
The bombardment energy is 5keV and the angle of incidence is 45 degrees
to the normal. The resulting yield-dose curves are given in figure 2.
Figure 1. Material consisting of thin layers.
Figure 2. Yield-depth curve for Germanium arising from the structure of figure 1.