Besides the data passed to and from TCAD tools, there is a lot of data to be kept internally in the framework related to the high-level tasks mentioned in the previous section. This data may or may not be visible to the TCAD user, either directly or indirectly.
The TCAD system has to keep track of the data flow occurring inside the framework. The data level of a TCAD environment must therefore be able to manage and archive simulation sequences in order to ensure the reproducibility of the results and easy backtracking through the simulation history. This capability of the data level is called a history and versioning mechanism. It must be possible to determine the data modification times, and which tools modified which data objects. This mechanism is especially important for all high-level tasks running multiple tools or sequences of tools. It is therefore a prerequisite for the successful implementation of PFRs (Process Flow Representation) on the task level of the TCAD system.
For high-level tasks like device characterization and response surface modeling, the need for non-geometrical point data occurs, often referred to as point tuples. These point tuples are collected in so-called point clouds, the points of which may have any number of coordinates. The physical units of these coordinates need not necessarily be spatial ones. Consider the case of a drain current versus gate voltage characteristic: Data points would be stored as two-dimensional point tuples in a point cloud whose first coordinate has a voltage unit (e.g. V), and the second coordinate has an electrical current unit (e.g. mA). Especially for these tasks it is desirable to allow tools to run in parallel on a cluster of computers. A precondition for this is the multiprocessing ability of the application interface, to enable parallel simulator runs using the same data set as well as ensuring clusterwide access to the data.
Optimization and sensitivity analysis require a different feature of a data level: In both cases, single input parameters or point coordinates have to be varied, and the data level must be able to support this detailed manipulation of a tool's input data. Due to the inhomogeneity of the input processing of conventional tools, this is a tough requirement.