Hi!
I like to use OpenEMS for PDN design so I started with a simple setup and later added one decoupling capacitor.
First I only simulate two planes and observe the impedance.
I expect the impedance to look capacitive for low frequency, reach a minimum and then the impedance to look inductive and finally to observe modal resonances. However, at low frequency the impedance is almost flat and don't follow the expected 1/(2*pi*f*C) behavior. After about 30MHz the impedance profile follow the expected behavior.
One simulation with only the two planes and one simulation with the added lumped C.
How can I improve the performance at low frequencies? Why is there a disagreement at lower frequencies?
I simulate the C with AddLumpedElement that extend between two vias in the xdirection. Is this a good way to simulate an added capacitor between the via:s? To include mounting inductance maybe it will be better to have two transmission lines between the vias with a very small gap in the middle where I place the lumped Celement?
Attached are the .m file I used for my simulation.
Best regards
Joakim.
Simulating impedance between two planes with decoupling
Moderator: thorsten
Simulating impedance between two planes with decoupling
 Attachments

 PDN_PlaneVia.m
 Simulation setup for impedance between 2 planes with decoupling capacitor
 (6.22 KiB) Downloaded 175 times
Re: Simulating impedance between two planes with decoupling
Hi Joakim,
sorry for the late answer but I was on a short holiday
Unfortunatly nobody else seemed to have an idea.
I would need to have a closer look to give a more detailed answer, but maybe you already have some new informations?
My guess would be that you have to let the simulation run longer. From the curves you showed even at high frequencies it does not look fully converged yet.
And for very low frequencies the simulation needs to run for a long time. Afterall, one period for 30MHz is a 1000 times longer than for 30GHz. I'm not even sure if you get good result for this low frequencies if your excitation goes up to this high frequencies in the first place... What I would do in your case only simulate for high frequencies (maybe >100MHz?) and just extrapolate for low frequencies, afterall for these small frequencies (much much lower than any resonance frequency of the device) everything should be nice and easy?
regards
Thorsten
sorry for the late answer but I was on a short holiday
Unfortunatly nobody else seemed to have an idea.
I would need to have a closer look to give a more detailed answer, but maybe you already have some new informations?
My guess would be that you have to let the simulation run longer. From the curves you showed even at high frequencies it does not look fully converged yet.
And for very low frequencies the simulation needs to run for a long time. Afterall, one period for 30MHz is a 1000 times longer than for 30GHz. I'm not even sure if you get good result for this low frequencies if your excitation goes up to this high frequencies in the first place... What I would do in your case only simulate for high frequencies (maybe >100MHz?) and just extrapolate for low frequencies, afterall for these small frequencies (much much lower than any resonance frequency of the device) everything should be nice and easy?
regards
Thorsten
Re: Simulating impedance between two planes with decoupling
Hi Thorsten
I hope you enjoyed the holiday
I tried three different steps, but no noticeable difference at the low frequency end.
change the excite frequency from 5GHz to 2GHz.
change convergence limit from 40dB to 60dB.
improved the grid resolution from lambda/16 to lambda/64.
Not a big problem but nice to know how to improve performance and learn if it is setup problem, model problem or some limitations in the simulator?
My goal is to import a hyperlynx file for some powergnd pairs, add some decoupling capacitor and investigate the performance. Before that I just want some simple simulations to run to verify setup ideas.
Best regards
Joakim
I hope you enjoyed the holiday
I tried three different steps, but no noticeable difference at the low frequency end.
change the excite frequency from 5GHz to 2GHz.
change convergence limit from 40dB to 60dB.
improved the grid resolution from lambda/16 to lambda/64.
Not a big problem but nice to know how to improve performance and learn if it is setup problem, model problem or some limitations in the simulator?
My goal is to import a hyperlynx file for some powergnd pairs, add some decoupling capacitor and investigate the performance. Before that I just want some simple simulations to run to verify setup ideas.
Best regards
Joakim