The generation pass includes all steps that are necessary to execute modal and static analyses, extract displacement and eigenvector information, and create a reduced order model of the structure. The reduced order model generation procedure is time consuming but it only has to be done once. After a reduced order model is established, you can perform any type of analysis with speed typical of system or circuit simulators and accuracy typical of finite element models. The generation pass consists of the following steps.
The following sections describe each step.
The jobname takes on special significance in reduced order model generation. By using jobnames effectively, you can eliminate much of the file handling inherent in a three-pass analysis. You should specify different jobnames for the generation pass and the use pass (for example, GEN and USE). /FILNAME,GEN will give the jobname GEN to all the files produced by the generation pass.
In this step, you assign the model database (prepared in the model preparation phase), the dimensionality of the model, and the primary operating direction of the device.
You assign names to pairs of conductors to represent lumped capacitances. Conductors that interact in the operation of the device should be assigned as conductor pairs.
If nonzero boundary constraints, temporary zero boundary constraints or structural nodal forces will be applied in the use pass, you must declare nodes used as ROM master nodes. Furthermore, ROM master nodes are necessary to attach other elements to the ROM model (for example, COMBIN40) or to simply monitor nodal displacements during the use pass. There can be up to ten ROM master nodes representing the displacement in the operating direction. Master node displacements in the operating direction will be stored as UX degrees of freedom.
To assist the program in determining which eigenmodes of the device are important in characterizing the structural response of the system under operating conditions, you should run a static analysis with a "test" load which deforms the structure in the operating direction of choice. The loads should drive the structure to a typical deformation state, which is representative of most load situations seen in the use pass. The amount of applied loads, the resulting displacements and even the accuracy of the computed results are not important because only ratios between modal coordinates are evaluated. The simplest test load could be in the form of imposed displacements. Alternatively, if you cannot define a test load, the modes and their amplitude range will be determined with respect to the linear modal stiffness ratios in the operating direction (see RMMSELECT).
The difference between using or not using a test load can be illustrated by a model of a beam clamped at both ends and suspended above a ground plane. For example, a voltage test load applied on the movable structure excites only symmetric eigenmodes in the operating direction. The RMMSELECT macro would select the symmetric modes in the order that corresponds to their displacement amplitudes. On the other hand, if no test load is specified, the RMMSELECT macro would select the lowest symmetric and asymmetric modes in the operating direction.
After you run a static analysis for a test load, you need to extract the neutral plane displacements.
Note: The neutral plane nodes were grouped into a node component named NEUN in the model preparation phase.
If the device is subjected to gravity loads, or pressure loading, you must run a static analysis for each individual element load prior to creating the reduced order model. The effects of the element loading are considered in the mode selection for the reduced order model. Additionally, the element loads may be applied in the use pass when their effects on the device response are required.
Each individual element load must be run as a separate load case in a multi load-step static analysis. Up to five element loads can be imposed in the generation pass. Later, in the use pass, the loads can be scaled and superimposed using RMLVSCALE.
After you run the analysis, you need to extract the neutral plane displacements.
Note: NLGEOM must be OFF for linear and stress-stiffened structural models unless prestress is relevant. Here, the element loads must be moderate so that no deflection dependent change of stiffness occurs. The rule of thumb is that the resulting displacements must be between 0.001 and 0.1 times the device thickness.
Next, you perform a modal analysis (ANTYPE,MODAL) with modal expansion (MXPAND) for the desired range of modes to be considered. The modal analysis captures modes of the device that will characterize the structural response. The ROM method assumes that the lowest modes dominate the structural response. You may need to constrain the device motion in order to ensure that the dominant modes are captured as the lowest modes in the modal analysis.
You then extract the eigenvectors of the neutral plane nodes (component NEUN).
Selection of the pertinent modes and their operating range is an essential step in
the efficient and accurate determination of the reduced order model. You can use the
results of the modal analysis and the test load and element load static analyses to
determine the most appropriate modes to characterize the structural response. To
perform an automated mode selection that uses those results, issue
RMMSELECT with Method =
TMOD.
The following are important points to remember at this step:
Modes considered for use in the ROM are classified as "DOMINANT” or RELEVANT.” Dominant modes are those with expected large displacement amplitudes. Their amplitudes interact with all system parameters derived from the strain energy and capacitance functions. Either one or two dominant modes are allowed. Relevant modes are those with expected small displacement amplitudes. Their behavior is strongly influenced by the amplitude of dominant modes but the interaction between the relevant modes can be neglected. Such a simplification is valid for most MEMS devices and it makes the following data sampling procedure faster. The ultimate goal is to select the fewest possible number of modes to sufficiently characterize the deformation of the structure for the intended operating conditions. The fewer the modes, the shorter the time will be to generate the reduced order model.
The Dmin and Dmax
arguments of the RMMSELECT command are the lower and
upper bounds of the total deflection range of the structure. They should be
large enough to cover the operating range in the use pass.
The automated mode selection performed by the RMMSELECT command may be manually changed or overridden. In some instances, specific knowledge of the device behavior and required modes may be already known, in which case you have the flexibility to select and modify the appropriate mode selection.
You can use the RMMRANGE command to define and edit the modal parameters.
The following are important points to remember at this step:
The computed displacement operating range for each mode can be
modified by the Min and
Max arguments of the
RMMRANGE command. Note that if the mode was
previously classified as "UNUSED" by either the
RMMSELECT or the RMMRANGE
commands, and you are issuing RMMRANGE to activate
this mode for ROM, the Min and Max parameters will be interpreted as the
total deflection range. Here, RMMRANGE will find the
lower and upper bounds for the newly added mode, and calculate its
contribution factor based on the information about all the active modes.
If you disagree with the automatically calculated parameters for this
mode, you can overwrite them by issuing RMMRANGE one
more time.
The Nstep argument of the
RMMRANGE command specifies the number of
equidistant steps for the coming data sampling procedure. Dominant modes
should be sampled with 8 to 11 steps, relevant with 3 to 5. For three
steps, the considered mode is linearized at the operating
point.
The default damping ratio is 0.05 for all modes. This number can
be changed by the Damp argument of the
RMMRANGE command for any mode at any time (even
in the use pass). Special consideration should be given to this damping
parameter, as it represents the effects from fluidic damping of the
structure.
The Scale argument of the
RMMRANGE command is necessary to overcome
convergence problems when computing the response surface. It should
be:
Scale = max{abs(Min),abs(Max)}-1
You can use RMMLIST to call a status report at this point to check your mode specifications.
At this point you should save your ROM database. The RMSAVE command saves it as an ASCII file. It will be used in the use pass and the expansion pass.
The next step is to run multiple finite element solutions on the structural domain and the electrostatic domain to collect sample points of strain energy and capacitance data for ROM response curve fitting. The model database must include the "STRU" and "ELEC" physics files and node components for the neutral plane nodes ("NEUN") and conductors ("CONDi") (see Model Preparation). A ROM database is also required. The program performs the multiple finite element runs automatically with no user intervention.
The following are important points to remember here:
The number of finite element solution runs is dependent on the number of modes selected and the number of steps chosen to characterize each mode. A "finite element solution set” consists of a single structural analysis, and a set of electrostatic analyses, one for each conductor pair defined (see RMCAP command). For example, consider the following scenario of number of modes selected and number of steps specified:
Mode 1: Dominant; 8 steps specified
Mode 3: Dominant; 5 steps specified
Mode 5: Relevant; 3 steps specified
The total number of "finite element solution sets" would be 8 x 5 x 3 = 120.
The Nlgeom flag must be set to ON in case of
stress stiffening or prestress. Capacitance data can either be calculated
from the charge voltage relationship (Cap flag
set to CHARGE) or from the derivatives of the electrostatic field energy
based on the CMATRIX macro. The
Cap flag must be set to CMATRIX if far field
elements are involved. The CMATRIX method is only recommended if significant
electric field leakage occurs to the open domain, and capacitance effects of
this leakage are significant.
The results are stored in files called jobname_ijk.dec whereby a separate file is written for each relevant mode k. The files contain all the information necessary to evaluate the behavior of the relevant mode k with respect to the dominant modes i and j.
In this step, you specify the polynomial orders for the modes that were selected for the ROM using RMMSELECT for use in function fitting the strain energy and capacitance data.
Make sure that the order of each mode is less than
Nsteps specified by RMMRANGE but
at least two. Polynomials with order eight and higher tend to oscillate and should
be avoided.
In the run sample point generation step, the strain energy and capacitance data were computed at different linear combinations of all involved modal basis functions. In this step, you find mathematical functions that represent the dependency of the acquired data with respect to the modal coordinates. A least squares fit algorithm determines these mathematical functions. You can chose among four different polynomial trial functions, which are either inverted or not. The polynomials are later used to interpolate the energy and capacitance data between sample points and to compute their derivatives with respect to the modal coordinates to establish the system matrices.
Keep the following recommendations in mind:
The argument Type = LAGRANGE is required if
only one dominant mode or two dominant modes and no relevant modes are
available. Otherwise try to use Type = PASCAL or
even one of the reduced polynomials since those require fewer coefficients
and enable essential speed up in the use pass.
You should not invert strain energy functions. Capacitance functions should be inverted if the gap between conductors changes significantly during the operation. This happens for parallel plate arrangements where the conductors move perpendicularly to their surface. For comb drive systems, the capacitance function should not be inverted since conductors move tangentially to each other.
The next step is to perform a fitting procedure for all ROM functions based on modal data and functional data generated via RMSMPLE and options defined by RMROPTIONS.
Polynomial coefficients for the response surfaces are stored in files called jobname_ijk.pcs that correspond to the sample data file jobname_ijk.dec.
Response surface plots help you verify that the fit functions to the expected behavior. If necessary, you can try different surface options to improve the fit results.
Response surface plots might also help you recognize oscillations. However, oscillations are usually not visible at the response surface itself but become obvious at the second derivative plots. To overcome oscillations, you should reduce the polynomial order or try another polynomial type. If both fail, you should increase the number of data points in the appropriate mode direction.
Next you should generate a status report that will help you assess the quality of the response surface.
In this step, you may export the ROM model to an external VHDL-AMS compatible simulator. The export procedure creates the necessary files to run the ROM model in the system simulator.
Element loads are considered if an arbitrary scale factor was applied via RMLVSCALE prior to executing RMXPORT.
RMXPORT generates a set of VDHL-AMS input files that contain the following:
Problem specific constants (Initial.vhd)
Strain energy functions (S_ams_ijk.vhd)
Capacitance functions (Cxy_ams_ijk.vhd)
ROM in VHDL language (Transducer.vhd)
Note: The VHDL-AMS transducer model is similar to a black-box model with terminals relating electrical and mechanical quantities. A further system description file is necessary to specify the external circuitry (voltage sources, controller units), structural loads (nodal forces, element loads) and run time parameters (time step size, total simulation time).