EM simulation for 4 layers PCB

[wisnukusuma]

I have set up a simulation using freecad plugin, I want to see how the efield or hfield propagates through my PCB. My PCB layer stack up is 4. and after more than an hour of calculation and putting all the available vtk's to paraview I saw nothing change. My expectation should work like the video attached where I used your 2 layers PCB on the youtube video. I just wondered how to setup a efield or hfield box on the freecad, should it wrap all the PCB objects, or used a thin box (10um) and placed under the copper or in the fr4. I also attached my setup file if anyone can help to see what's wrong with my setup. thank you.

hfield.mp4

setupmacro_openEMS_simulation.zip
freecadsetup

[bobb-uk]

panire3 (over on YouTube) has shared his demo code and the priority for his "air" entry was the lowest. Try swapping priorities for your FR4 and Air entries.
Bob. (PS: am not an expert in openEMS)

[wisnukusuma]

Already swab with the same result.

[LubomirJagos42]

I'll try to look on this if not forget, glad to see that you used that FreeCAD addon :)

[wisnukusuma]

also thankyou for your hardwork creating very helpful tools like this. @LubomirJagos42

[LubomirJagos42]

Hi, so I looked at your example and you are trying to simulate board from KiCAD as it is, this is OK, openEMS is definitely able to provide some simulation results, but here are some obstacles:

• FDTD algorith us using rectangular grid, therefore mesh must be really dense to properly catch spaces and traces on PCB board
• for curved shapes and this kind of simulation HFSS would be easier to create mesh since it's using FEM method and therefore tetrahedral adaptive mesh

Anyway, since here we are using openEMS, I did this things to got some simulation setup:

• make grid less dense to got faster simulation, you used spacing in 0.1mm and for air also 0.1mm, I was getting 20.000.000 cells what is too much
• gridlines in Z ax weren't crossing copper layers
  • for this I used hack, I put auxiliary rectangle in ZY plane which is aligned from top layer to bottom and put there grid count to 4, this way there are generated 4 lines in Z and since they are space euivalent they are crossing exactly your copper layers
  • changed grid spacing in XY to 0.17mm to get in total 2.000.000 what at least simulate at reasonable speed and in debug PEC I checked that geometry is generated and PCB traces shape is there
  • changed spacing for air in Z direction to 1mm to reduce cells count
  • your E field was box, I used plane with zero thickness to get 2D E field monitor to be able directly load it into paraview and got similar animation as you put here
• it showed me message unusued material in FR4, therefore whole substrate what are 3 layers of substrate were as one object in FreeCAD and my suspicion is that they were not solid but shells therefore it was maybe hollow and not crossing any gridline therefore
  • I draw some boxes around these layers and used boolean operations to separate each of them and then solidify and assign them separately into simulation
  • NOTE: now looking on that freecad file, you can also used substrate objects from imported KiCAD, they are already there each layer separate
• I change maximum timesteps to 100.000 previous ten thousand was too less
• excitation changed from step excitation to gaussian with wideband fc, for step simulation there is no f0 and fc and I have not too much experience with it so at the moment I don't know how to calculate S11 from step excitation so I put this to get some animation and possible S11 calculation

I put that KiCAD board importer since I was thinking about these simulation that they will be nice and it's OK to do them, now I see there are some pitfalls:

• substrate layers importing, better is used each layer separately and ensure they are solids
• mesh definition could be hard since for complicated shapes it's hard to make not dense shape :(

Notes to my simulation modification:

• on my laptop with "Intel i7-8850H CPU @ 2.60 GHz, 6 Cores" it takes quite time at start to create FDTD operator like 20min, so yes this simulation is running but due it generates 3mil cells it takes some time at beginning
• average calculation for timesteps one line output from openEMS core takes around 20s
• you put there excitation cca 8000V/m in simulation in paraview there are really high amplitudes I see 800kV/m, this is also due there is substrate FR4 with er=4.6 which amplifies excitation input
• the model contains just copper traces which are unloaded at the end, I could be maybe good to add some load on the end of excited trace

Hope this helps. Attaching my FreeCAD simulation with settings in zip file.

modified_signal_propagation_setup.zip

[VolkerMuehlhaus]

Hi @wisnukusuma,

the rule of thumb is that the signal on the line will not "see" objects that are more than 3 * linewidth or 3 * substrate height away. So you can cut out that region, but make sure that you keep ground path connection froms vias etc.

This is how I would cut the model to analyze your line segment:

[wisnukusuma]

Nice suggestion @VolkerMuehlhaus! I already made it like you but for the sake of reduce the sim calc time, but that still took 5+ hours. I thought because my setup geometry was bad like @LubomirJagos42 analyze it. Also I don't know the dimension rule how to cut the board for simulation. And yeah, I got some insight from you now. Thankyou very much God Bless you!

[wisnukusuma]

Hi @LubomirJagos42 ! in your setup file, you set grids to 0.17, 0.17, and 0.43 mm XYZ axes respectively, how do you get those? do you have a cheatsheet to get the number? or do you just tune the parameter until you get somewhat close to 1/3 2/3 rules?
also for z copper layer lines, why 13 counts?

[LubomirJagos42]

to be honest I just put there random number which come to my mind, I looked at your mesh and saw it's too dense so I tried 0.17 and it seems to be less dense and still there were some lines in spaces and it looks fine, so I haven't follow any rule
for Z layers I needed at least 4 lines each one for copper layer and then I wanted to have 3 lines through substrate so I calculate it like 4+3*3=13
maybe mesh could be less or more dense I don't know, but this from debug PEC shows that at least the shapes are meshed

you can try other values, maybe it will be better

[wisnukusuma]

Hi @LubomirJagos42, when we tune the grid numbers, how to know how many cells we generate from the setup without running create FDTD Operator cause it takes a long time to get this done. Is there any command line that I can inspect number of cells?