next up previous index
Next: 9.2.5 Optimization Result and Up: 9.2 Short Channel Effect Previous: 9.2.3 Process Simulation

9.2.4 Task Definition

Using the task-level encapsulation mechanisms presented in Chapter 8, the complete analysis and optimization task outline above can be defined in VISTA/SFC by a list of statements. Figures 9.6 through 9.10 show the code for batch-mode operationgif.

In Figure 9.6, the first statement defines a new project that uses a process flow description file. (The LAT implant is realized by using two subsequent implantation steps, POCKET-L and POCKET-R, of opposite implantation angles). Table 9.3 lists the relevant process parameters and responses. The second statement creates an EVE object to manage all experiments of the optimization task. The symbolic name "flow-eve" is assigned to identify this object unequivocally in the case several EVE instances exist simultaneously.

 figure2646

 table2656

The Eve-Define-Control statements (cf. Section 8.4.1) define the position of the experimental control variables in the process flow. For example, the implantation angle anginc in the process step labeled POCKET-R is assigned the symbolic name PAR. The value of PAR is derived from the value of the control variable PA by multiplying with -1, leading to symmetric implantation profiles. The Eve-Set-Control statements define default values and ranges for the control variables.

Responses are defined with the Eve-Define-Control statement. In the example the device simulation generates sets of data points for gate voltages between 0 V and the drain bias vd. Using the LISP functions last and first to derive the values of the response variables from these sets is a simple way to get the minimum and maximum drain current and bulk current values.

Once the controls and responses are defined, the Eve-Eval statement is used to simulate the nominal device. After termination of the simulation, the results can be examined (Figure 9.7). Parameter value settings can be given to study interactively the behavior of the design.

 figure2699
Figure 9.7:   The simulation of the nominal device is started by using the Eve-Eval command. Parameter value settings can be given. Note that ib represents tex2html_wrap_inline8331 times the maximum bulk current (see Figure 9.6).

Figure 9.8 shows the VISTA/SFC sequencegif statement calling three DoE steps to create two times 7 experiments, varying the gate length L from tex2html_wrap_inline8295 to tex2html_wrap_inline8297 in equidistant steps for PD set to tex2html_wrap_inline8331 and tex2html_wrap_inline8339, respectively, plus 15 experiments from a full-factorial CCI design for the three process parameters. The Eve-Stop command prevents the new experiments from being submitted for simulation immediately; Eve-Start starts evaluation for all 29 scheduled instances. They are run simultaneously by the run controller and distributed across the network.

 figure2728
Figure 9.8:   Design of experiments for screening analysis of LAT implant. The DoE module is called for the design type SA and override nominal settings for the dose parameter PD. The sequence operator is used to chain a sequence of asynchronous LISP commands.

After computation of the sample points, a response surface is built from the data generatedgif for the three responses vth, ib, and id-on. Figure 9.9 shows the sequence of statements to run the RSM generator, define a target function for the optimization process, and start the optimization for a set of initial values. From the available optimizers, the donopt program is selected. The target function for the optimizer is constructed from the threshold and maximum bulk current values to raise the threshold voltage at L=tex2html_wrap_inline8299 and keep the bulk current as low as possible.

 figure2746
Figure 9.9:   VISTA/SFC statements for RSM model generation and optimization.

If one wishes to run the optimizer directly on the simulation server, the RSM generation step is skipped and the target function is defined directly in the "flow-eve" object to provide input for the optimizer (Figure 9.10).

 figure2759


next up previous index
Next: 9.2.5 Optimization Result and Up: 9.2 Short Channel Effect Previous: 9.2.3 Process Simulation

Christoph Pichler
Thu Mar 13 14:30:47 MET 1997