... architectures1.1
To handle different byte ordering schemes and the differences between $ 32$ and $ 64$ bit CPUs.



























































... suite1.2
The former TMA suite



























































...2.1
Depending on the runtime environment (dynamic linker, libraries) dynamic instantiation is directly performed via the dl interface (dlopen) or via a programmatic construct (switch). In the latter case the instantiation is not really dynamic in the narrow sense, because all possible implementations must be known at compile time. The mechanism is, however, transparent to the application.



























































... problem2.2
The choice of which algorithm is used strongly depends on the absolute number of point locations that are performed during the simulation. For few point locations the preprocessing time that is necessary to build the oct-tree will be dominating, whereas for a relatively large number of point locations the search time will be the limiting factor.



























































... database3.1
The actual location (file or database) of the data is transparent.



























































... values3.2
Concrete values (e.g. double, string, ...) are derived from the abstract value interface. Therefore, any data type of the used programming language (C++) can be used as a value.



























































... code3.3
If the constructor is used then the include file that contains the definition of this Reader implementation must be included in the application program.



























































... attribute3.4
The actual differences are hidden by the use of interfaces.



























































... handle3.5
The WAFER-STATE-SERVER uses a simple garbage collection mechanism (handles) to avoid memory leaks. A 0 assignment to an instance of a handle class results in a call of the destructor on that object if there was exactly one handle to the object. Please refer to Section 3.7.1 for a description of the handle class.



























































... ignored3.6
Note that a mixture of different spatial dimensions in one file is neither useful nor legal. It was nevertheless decided to issue only a warning instead of a fatal error as this helps in detecting several such errors at once as opposed to only one error per Reader invocation. This is useful when WSS files are generated by hand.



























































... managed3.7
While the time spent in the reading process of a three-dimensional simulation run is negligible, it certainly does matter for visualization. This is even more relevant if not one but many simulation results are stored and are to be viewed from a browser-like program.



























































... address3.8
Multiuser and multitasking operating systems need to manage additional information such as the user and process IDs the memory region belongs to.



























































... omitted3.9
This information is determined at compile time; the overhead is totally eliminated.



























































... time4.1
The programming language FORTRAN77 does not offer a standardized, platform independent, dynamic memory allocation mechanism.



























































... conforming4.2
This can be the case if the Wafer was already generated with interface conforming grids in a previously performed simulation.



























































... triangles4.3
The etch front is represented by refined cells only. The triangulation of the etch front takes place by splitting the interface between material and vacuum into two equal sized triangles.



























































... questionable5.1
If the simulation failure occurs during the computation of the gradient, the step direction will be computed incorrectly



























































...5.2
On WINDOWS machines all jobs will have the same nice level, thus the regular rank $ R_i$ is used.



























































... ignored5.3
They are realized as comments.



























































... two5.4
A crossover with two parents is called a sexual crossover operation. This mimics a biological reproduction mechanism. There are, however, also algorithms that combine three or more individuals (asexual crossover).



























































... number5.5
In the inverse modeling example that is presented in Section 5.6 an analytical model to approximate the dopant profile of a transistor is used, where the genes directly correspond to the parameters of the model.



























































... optimizer5.6
The programs use the standard input and standard output channels that are associated with every UNIX program or WINDOWS console application.



























































... clusters5.7
The ratio $ I/I_*$ is called interstitial supersaturation.



























































... grids7.1
The grids of a Wafer need not be boundary conforming.



























































...7.2
Note that in this case the smallest possible grid that is necessary to represent the geometry should be used to minimize the overhead that is introduced by representing a segment without any quantities via the hull of a grid.