Ch. Pichler, R. Plasun, R. Strasser, and S. Selberherr: High-Level TCAD Task Representation and Automation
next up previous
Next: Applications Up: High-Level TCAD Task Previous: System Components

Task Encapsulation

A class of evaluable entity (EVE) objects has been defined to provide uniform access to basic services like process flow simulation and RSM evaluation as well as to complex tasks specified by the user. For example, the minimization of the bulk current for a given process is encapsulated in an object that is evaluated for a set of initial value vectors for the optimizer generated by the DoE module or by a simple LISP loop. Figure 2 shows the basic application of an EVE object to represent a process flow.

 figure49
Figure 2:   Encapsulation of process flow with evaluable entity (EVE) objects.

After specifying those process parameters that are to be used as input variables (controls) for subsequent analysis tasks, the EVE object hides all evaluation details of the underlying application. All process parameters in a flow as well as all simulator input parameters are directy accessible by specifying the name of the parameter and the label of the step in the flow. A variety of clients can request evaluations for certain input parameter sets via the uniform interface established by this encapsulation (Figure 3).

 figure61
Figure 3:   EVE objects establish a uniform interface for evaluation requests by internal and external clients and applications.

Interface Agents provide an interface between client tools like optimizers or RSM tools. The interface agent talks to these clients using standard protocols. For each type of client a seperate protocol, ie. a set of messages and responses, is defined in order to enable communication between the TCAD shell and the client. The communication is handled via standard input and standard output of the client. Thus users are able to plugin their own implementation of an optimizer, for example, without having to modify the framework itself, provided that the tool talks the protocol. Figure 4 illustrates the communication between control shell an optimizer and an RSM tool. The interface agents also resolve error conditions arising from tool errors (malfunctions of clients are recognized).

 figure72
Figure 4:   Communication between the TCAD shell, the optimizer, and the RSM tool.

Starting from existing tasks, more complex ones can be built by combining EVE objects in a nested manner (Figure 5).

 figure83
Figure 5:   EVE objects can be combined to build complex tasks from simpler ones.

In addition to providing a standardized interface, EVE objects also assist in the management of all instances that have been generated (Figure 6).

 figure95
Figure 6:   Evaluation instances (EVI) objects form part of an EVE object.

In the GUI, an EVE and its pertaining instances are represented as a spreadsheet with direct access to plot tools, Doe generation, and RSM generation (Figure 7).

 figure108
Figure 7:   A spreadsheet-GUI (in the lower part of the window) provides a convenient interface for initiating new evaluations and experiments, generate 1D plots, and build and view response surface models. The upper part of the display gives direct access to the simulation runs that generated the extracted data. Colors are used to indicate the state of experiments.

For all experiments generated from an EVE object, response surface models (RSM) can be built, viewed, and used as proper EVE objects for further analysis. Figure 8 shows the GUI for exploring RSM data.

 figure120
Figure 8:   Graphical user interface for viewing response surface models. One or two parameters can be selected as plot axes, the remaining ones are parameters of the plot.

One-dimensional and two-dimension cuts of the N-dimension parameter space can be selected and displayed using the gnuplot tool (Figure 9) or any other plot program. (Access to an external plot tool is again encapsulated by an agent object; therefore, any number of different programs can be easily integrated using the standardized agent interface).

 figure132
Figure 9:   Two-dimensional plot of six-dimensional response surface model.


next up previous
Next: Applications Up: High-Level TCAD Task Previous: System Components

Ch. Pichler, R. Plasun, R. Strasser, and S. Selberherr: High-Level TCAD Task Representation and Automation