| Author |
 Topic  |
|
|
Darius
Canada
1 Posts |
 Posted - Aug 12 2014 : 19:50:26
|
Hello Enrico,
We bought the license for FasterCap to use it in our research at IQC, University of Waterloo.
I am trying to calculate the mutual capacitance between two transmission lines (metal plates) sitting on top of a substrate (silicon block). I have the following questions:
(1) Does having two different dielectric constants (air:1 and Si:11.7) cause any convergence problem? If it does what is the solution?
(2) GMRES Iteration stops sometimes after reaching to 999. Is there any way to increase the maximum number from 999 to e.g. 2000? Does it help at all to increase this max.GMRES.Iter number?
(3) What is the dimension of the Mesh relative refinement (-m) value in the Manual Setting? i.e. are they grid size? If yes what is their unit? meter micrometer?
For a structure with 2mm long metal line what is the best value to start with?
(4) Do you have any sample file or example showing what suitable parameters are when we use “Manual” setting instead of “Automatic Settings”?
(5) Does FasterCap take effective dielectric effect into account?
For example in the following structure ( a metal line on top of a substrate) since most of the electric field is in the air, it looks as if the transmission line is already embedded in a material with different dielectric material of smaller value i.e.
(#949;eff = (#949;+1)/2 + (#949;-1)/2 * f(W/H)) where W and H are line width and substrate thickness, respectively.
Your help is sincerely appreciated. If I could know how to send you the picture of our structure that would be very helpful.
Sincerely,
Darius
|
|
|
Enrico
550 Posts |
Posted - Aug 13 2014 : 17:02:33
|
Let's start from a basic question, i.e. are your tx lines really 3D, or are you happy with the tx line parameters of C and L per unit length?
In the first case, i.e. 3D, I do expect that the cross-section is not uniform along the length and/or the line has bends, vias, etc. However in this case there is an underlying limit in FasterCap / FastHenry2, i.e. the quasistatic assumption, so your model will be accurate only up to a certain frequency, when you can consider your structure as lumped.
In the second case, i.e. 2D, the cross-section will be uniform along the length, while on the section itself the dimensions will be small w.r.t. the maximum wavelength of interest. In this case the quasistatic assumption is valid on the cross-section, and you can happily simulate the line to obtain the parameters per unit length.
Now you have two ways to get these parameters. The first is to use a 3D model, with a line long enough for the fringing fields on the ends to be negligible. But this is a sort of overkill. The second, preferred way is to use the 2D capabilities of FasterCap, that were designed for this goal. You define only the cross-section, and get (much faster) the per-unit values.
Clarified the above, I will tackle your questions in this context:
quote:
(1) Does having two different dielectric constants (air:1 and Si:11.7) cause any convergence problem? If it does what is the solution?
No I do not expect issues in having two dielectric constants. Actually FasterCap is designed to allow multiple mediums. Convergence problems are usually related to errors in the geometrical description or by high geometrical ratios, but it is difficult to give a generic answer to this question. You should share your input file.
quote:
(2) GMRES Iteration stops sometimes after reaching to 999. Is there any way to increase the maximum number from 999 to e.g. 2000? Does it help at all to increase this max.GMRES.Iter number?
No, convergence should be reached long before that. Looks like there are issues.
quote:
(3) What is the dimension of the Mesh relative refinement (-m) value in the Manual Setting? i.e. are they grid size? If yes what is their unit? meter micrometer? For a structure with 2mm long metal line what is the best value to start with?
No, they are not grid size, and this is a relative parameter, that has no dimension in itself. Anyway, it ultimately controls mesh fineness. In your case you should probably be more interested in the '-d' parameter, that sort of defines the level of compression of the internal potential interaction matrix. This could be lowered up to 0.1 in case of high geometrical ratios (there is an example in the samples directory for the parallel plate capacitor, see the comments inside the file). Do not lower this value routinely, however, as for standard geometries it is useless to go below 1.0 and you only increase (significantly) the simulation time.
If you want to go manual, you can start with a very high -m value (e.g. 10000) and then pick up the output value at the end of the simulation, and use it as the real starting point:
Max Mesh relative refinement value: 0.0335302
quote:
(4) Do you have any sample file or example showing what suitable parameters are when we use “Manual” setting instead of “Automatic Settings”?
See above. Moreover, if you run any of the samples in automatic mode, you can see the parameters set by FasterCap, and you can imitate them.
quote:
(5) Does FasterCap take effective dielectric effect into account?
Actually not. FasterCap uses the permittivity values of the different dielectrics, not the effective permittivity.
Best Regards,
Enrico
P.s. sorry but for security reason the Forum does not allow to upload pictures. We had a lot of spam in the past unfortunately.
|
 |
|
| |
Topic |
|
|
|
simulating Coplanar Waveguides
