Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Frequency dependent capacitance

Please login with a confirmed email address before reporting spam

Hi,

for quite a while now, I am trying to model the frequency behaviour of a capacitor (3D) but hadn't had success yet. So what I would like to do, is to get the capacitance depending on the frequency. So it would be nice to get a curve, but to get the capacitance at a certain (fixed) frequency would also be sufficient in the beginning. I also the forums for several weeks and read the docs without getting much further. It looks like as if several people where trying the same but without any success.

Simulating the DC capacitance was not a big deal since there are some nice examples in the model library and on the web, so here are some questions concerning an AC analysis:

First, I'm not sure whether I can only choose electrostatics (es) as physics, since I would like to have a frequency or time dependent result on the one hand, but es contains also frequency/time dependent studies on the other hand.

Can I do this with a preset study ("time dependent") and by putting a time dependent signal at one port? Or do I need to use the "frequency domain" study for that? So far, I tried both ways in many different variations but always end up with some error messages. When I used the time dependent study, I applied a sinusoidal signal at one port, when I used frequency domain I entered the desired frequencies.

Also, is it correct that you have to divide the imaginary part of the admittance by the angular frequency to get the capacitance? I could only find examples for versions 3.x on how to do this, but not on 4.x and since the syntax has changed I'm stuck here as well.

Thanks a lot in advance!

9 Replies Last Post 17 juin 2011, 09:47 UTC−4

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 16 juin 2011, 10:05 UTC−4
I'd like to add some facts to my problem - and some questions.

First I have a general question: When capacitive structures are being modeled in v3.5a, you would usually model the electrodes, then add the surrounding material (e.g. a box of air) and then subtract the electrodes from the air box. In v4 you just add the air box without subtracting the electrodes. Why?

I was able to simulate my capacitor in a time-harmonic analysis in version v3.5a. Calculating the capacitance with the formula given in the microstrip example (C11=imag(Y11)/omega) gives me always the same capacitance, no matter what frequency I apply at the input port. I checked the input signal by drawing it versus the time, so I really see a sinusoidal signal with the desired frequency. This can't be correct, since the capacitance is frequency-dependent. Btw, I went up to 1 GHz.

Then I tried a frequency domain analysis in v4.0a. Again, I simulated up to 1GHz and I calculated the Capacitance with the same formula. Again, I get a frequency-independent capacitance which can't be correct. I plotted the Admittance Y11 which shows a perfectly linear increase with the frequency, but I would expect something else due to fringing fields etc.

So to summarize: No matter if I try a DC analysis, a time-harmonic analysis or a frequency domain analysis, I always get the same capacitance, no matter which frequency I apply at the input port.
In fact I would expect something like on this page: 74.86.124.120/$sitepreview/sonnet-dev.com/images/id-cap.gif where you can see a non-linear increase of the capacitance of an interdigitated capacitor versus the frequency.

Any ideas??
I'd like to add some facts to my problem - and some questions. First I have a general question: When capacitive structures are being modeled in v3.5a, you would usually model the electrodes, then add the surrounding material (e.g. a box of air) and then subtract the electrodes from the air box. In v4 you just add the air box without subtracting the electrodes. Why? I was able to simulate my capacitor in a time-harmonic analysis in version v3.5a. Calculating the capacitance with the formula given in the microstrip example (C11=imag(Y11)/omega) gives me always the same capacitance, no matter what frequency I apply at the input port. I checked the input signal by drawing it versus the time, so I really see a sinusoidal signal with the desired frequency. This can't be correct, since the capacitance is frequency-dependent. Btw, I went up to 1 GHz. Then I tried a frequency domain analysis in v4.0a. Again, I simulated up to 1GHz and I calculated the Capacitance with the same formula. Again, I get a frequency-independent capacitance which can't be correct. I plotted the Admittance Y11 which shows a perfectly linear increase with the frequency, but I would expect something else due to fringing fields etc. So to summarize: No matter if I try a DC analysis, a time-harmonic analysis or a frequency domain analysis, I always get the same capacitance, no matter which frequency I apply at the input port. In fact I would expect something like on this page: 74.86.124.120/$sitepreview/sonnet-dev.com/images/id-cap.gif where you can see a non-linear increase of the capacitance of an interdigitated capacitor versus the frequency. Any ideas??


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 16 juin 2011, 11:00 UTC−4
I don't think I can be of much help. However, I noticed that in the figure you linked to, the capacitance has no frequency dependence below 1 GHz whatsoever, just like in your case. So why do you stop your frequency analysis at 1 GHz instead of going higher? After all, the capacitance should be constant in the low-frequency limit, whatever "low frequency" may mean.

By the way, even in v4 I always subtract overlapping volumes, since I prefer my physics à la "two bodies cannot be in the same place at the same time". This may no longer be necessary, though, since Comsol 4 has these features now that allow for overriding domain settings by other domain settings.
I don't think I can be of much help. However, I noticed that in the figure you linked to, the capacitance has no frequency dependence below 1 GHz whatsoever, just like in your case. So why do you stop your frequency analysis at 1 GHz instead of going higher? After all, the capacitance should be constant in the low-frequency limit, whatever "low frequency" may mean. By the way, even in v4 I always subtract overlapping volumes, since I prefer my physics à la "two bodies cannot be in the same place at the same time". This may no longer be necessary, though, since Comsol 4 has these features now that allow for overriding domain settings by other domain settings.

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 16 juin 2011, 13:50 UTC−4
Hi,

I don't see why a capacitor should be frequency dependent in the AC/DC approximation, unless you have a frequency dependent permittivity which is not the case for air.

The picture changes in RF as soon as your structural dimensions approach the wavelength you use.

But you used AC/DC didn't you?

Cheers Edgar
Hi, I don't see why a capacitor should be frequency dependent in the AC/DC approximation, unless you have a frequency dependent permittivity which is not the case for air. The picture changes in RF as soon as your structural dimensions approach the wavelength you use. But you used AC/DC didn't you? Cheers Edgar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 juin 2011, 04:00 UTC−4
Hi John,

You're absolutely right, I also noticed that the Capacitance seems to stay very constant in lower frequencies but I guess this is just a question of resolution. I guess if you zoom in and adjust the axis scaling, the same plot would also show a quadratic increase in the lower frequency region, although its not that big. In my case, a have a perfectly constant value. Besides, I ran the simulation up to 100GHz yesterday and still did not see a difference.

Thanks for explaining the issue with the two bodies, I think I understand now. You can still create non-intersecting domains but you don't have to do it anymore, thanks a lot!!
Hi John, You're absolutely right, I also noticed that the Capacitance seems to stay very constant in lower frequencies but I guess this is just a question of resolution. I guess if you zoom in and adjust the axis scaling, the same plot would also show a quadratic increase in the lower frequency region, although its not that big. In my case, a have a perfectly constant value. Besides, I ran the simulation up to 100GHz yesterday and still did not see a difference. Thanks for explaining the issue with the two bodies, I think I understand now. You can still create non-intersecting domains but you don't have to do it anymore, thanks a lot!!

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 juin 2011, 04:20 UTC−4
Hi Edgar, thanks a lot for your reply!

First of all, yes I used AC/DC. So from what I know, the capacitance increases versus the frequency due to an increase in fringing fields between the fingers. I think coupling with the substrate (which doesn't count for me since the capacitor is sourrounded by air) is frequency dependent as well.

Now here's the thing: I'm not a hundred percent sure wether AC/DC is taking things like this into account. But there's this guy from the university of british columbia who has lots of comsol examples for electrostatic problems on his webpage:
www.mina.ubc.ca/konradw_comsol_electrostatics

There he modeled quite simple plate capacitors (DC). He claims, that the difference in capacitance between a hand calculation and the comsol model is only due to the fringing fields (e.g. check out the description below the model called "Capacitor with Uniform Dielectric" on that homepage). That's why I assumed you can model fringing influence on the capacitance with AC/DC model...


Hi,

I don't see why a capacitor should be frequency dependent in the AC/DC approximation, unless you have a frequency dependent permittivity which is not the case for air.

The picture changes in RF as soon as your structural dimensions approach the wavelength you use.

But you used AC/DC didn't you?

Cheers Edgar
Hi Edgar, thanks a lot for your reply! First of all, yes I used AC/DC. So from what I know, the capacitance increases versus the frequency due to an increase in fringing fields between the fingers. I think coupling with the substrate (which doesn't count for me since the capacitor is sourrounded by air) is frequency dependent as well. Now here's the thing: I'm not a hundred percent sure wether AC/DC is taking things like this into account. But there's this guy from the university of british columbia who has lots of comsol examples for electrostatic problems on his webpage: http://www.mina.ubc.ca/konradw_comsol_electrostatics There he modeled quite simple plate capacitors (DC). He claims, that the difference in capacitance between a hand calculation and the comsol model is only due to the fringing fields (e.g. check out the description below the model called "Capacitor with Uniform Dielectric" on that homepage). That's why I assumed you can model fringing influence on the capacitance with AC/DC model... [QUOTE] Hi, I don't see why a capacitor should be frequency dependent in the AC/DC approximation, unless you have a frequency dependent permittivity which is not the case for air. The picture changes in RF as soon as your structural dimensions approach the wavelength you use. But you used AC/DC didn't you? Cheers Edgar [/QUOTE]

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 juin 2011, 05:50 UTC−4

[...] there's this guy from the university of british columbia who has lots of comsol examples for electrostatic problems on his webpage:
www.mina.ubc.ca/konradw_comsol_electrostatics

There he modeled quite simple plate capacitors (DC). He claims, that the difference in capacitance between a hand calculation and the comsol model is only due to the fringing fields [...] That's why I assumed you can model fringing influence on the capacitance with AC/DC model...


What he means by that is this: Comsol gives you the actual capacitance, including the fringe field. In the analytical solution (C = eps_0 eps_r S/d), the fringe field was neglected, hence the difference. There is no analytical solution for the value Comsol calculates, so this approximation is the only point of reference.

In his example, there is no time/frequency dependence: the electric field doesn't vary with time, therefore the capacitance doesn't either.

I haven't given it much thought, but the only way I could possibly see a frequency dependence (for constant eps_r in each domain) is if you have at least two different dielectrics. Leakage currents through the dielectrics would then redistribute electric charges over time and change the electric field.

Best,

John
[QUOTE] [...] there's this guy from the university of british columbia who has lots of comsol examples for electrostatic problems on his webpage: http://www.mina.ubc.ca/konradw_comsol_electrostatics There he modeled quite simple plate capacitors (DC). He claims, that the difference in capacitance between a hand calculation and the comsol model is only due to the fringing fields [...] That's why I assumed you can model fringing influence on the capacitance with AC/DC model... [/QUOTE] What he means by that is this: Comsol gives you the actual capacitance, including the fringe field. In the analytical solution (C = eps_0 eps_r S/d), the fringe field was neglected, hence the difference. There is no analytical solution for the value Comsol calculates, so this approximation is the only point of reference. In his example, there is no time/frequency dependence: the electric field doesn't vary with time, therefore the capacitance doesn't either. I haven't given it much thought, but the only way I could possibly see a frequency dependence (for constant eps_r in each domain) is if you have at least two different dielectrics. Leakage currents through the dielectrics would then redistribute electric charges over time and change the electric field. Best, John

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 juin 2011, 07:25 UTC−4
Hi John, thanks a lot for your input, I really appreciate it!

So I had the same idea and what I did was, I simulated the capacitor (14 microns high) on a 100um pyrex substrate (epsilonr=6.4) with another 100um air box on top (epsilonr=1).

The result is almost the opposite from what I would expect, the capacitance gets smaller (see attachment). The frequency range was from 1Hz to 100GHz by going up with a factor of ten (1Hz, 10Hz, 100Hz, ...).

Pretty strange...
Hi John, thanks a lot for your input, I really appreciate it! So I had the same idea and what I did was, I simulated the capacitor (14 microns high) on a 100um pyrex substrate (epsilonr=6.4) with another 100um air box on top (epsilonr=1). The result is almost the opposite from what I would expect, the capacitance gets smaller (see attachment). The frequency range was from 1Hz to 100GHz by going up with a factor of ten (1Hz, 10Hz, 100Hz, ...). Pretty strange...


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 juin 2011, 07:32 UTC−4
And if I now set the mesh to extremely fine, I get what I got in the beginning, a constant capacitance...
And if I now set the mesh to extremely fine, I get what I got in the beginning, a constant capacitance...


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 juin 2011, 09:47 UTC−4
Hi Jimbo,

of course COMSOL calculates the stray field (you call it fringing field). The difference between an analytical calculation and COMSOL is that COMSOL cannot calculate the stray field at large distances because the FE model must be finite.
In the AC/DC approximation the field topology does not depend on frequency, also the stray fields are frequency independent.
It is the nature of the quasistatic approximation that the relation of structural dimensions and wavelength are ignored.
You must probably use RF to see those effects you are expecting.

Cheers
Edgar
Hi Jimbo, of course COMSOL calculates the stray field (you call it fringing field). The difference between an analytical calculation and COMSOL is that COMSOL cannot calculate the stray field at large distances because the FE model must be finite. In the AC/DC approximation the field topology does not depend on frequency, also the stray fields are frequency independent. It is the nature of the quasistatic approximation that the relation of structural dimensions and wavelength are ignored. You must probably use RF to see those effects you are expecting. Cheers Edgar

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.