Via modeling in ADS EM Simulation

[edwardev]

Hello Everyone,
I'm new to ADS EM tool.

I created an 4 layer substrate as the following link in ADS. And want to use EM to run the via of L1 to L2 and L1 to L3.
The method I added the via is draw a circle on the "Cond_M2" layer for L1 to L2, and " Cond_M3"layer for L1 to L3.

Substrate :

https://imgur.com/5yMWIUx

Q1) is my method to create the via correct ?
Q2) when I want to get the EM result for L1 to L2, I draw a transmission line on layer 1, and a transmission line on L2, and also draw a circle on "Cond_M2" layer , then de-embedded the port to the edge of the via.
When the length of the transmission line is different, it seems the result is different in high frequency ( over 50 GHz ), even though I de-embedded the port to the same position on two cases, does anyone know the reason ? And which result I should trust ?

The result for both length of the transmission line (blue is the short trace, red is the long trace )

https://imgur.com/a/NMFzEC6

Setup of the via :

https://imgur.com/468v8Aa

https://imgur.com/RiXIdEm

Q3). I want to use a simple inductor to model the via, the first one is L1 to L2, the other one is L1 to L3. The methods I used is to get the EM result for the two vias, and use a inductor to fit the curve of the S11 from the EM result. ( the setup schematic is shown in the link below )
But it seems the modeled inductance of the two vias is similar although the length difference between the two vias is different ( the via of L1 to L3 is twice length than L1 to L2 ) . Is my EM setup for the vias correct ? ( similar as question 1 ) , I exported the S2P file from the EM result and connect one pot to GND in the schematic diagram.

Schematic :

https://imgur.com/mY9la2F

Result:
https://imgur.com/24PJCu7

Thanks

 

[volker@muehlhaus]

Hello!

Q1) is my method to create the via correct ?

Yes, correct

Q2) when I want to get the EM result for L1 to L2, I draw a transmission line on layer 1, and a transmission line on L2, and also draw a circle on "Cond_M2" layer , then de-embedded the port to the edge of the via.
When the length of the transmission line is different, it seems the result is different in high frequency ( over 50 GHz ), even though I de-embedded the port to the same position on two cases, does anyone know the reason ? And which result I should trust ?

The mesh of the line itself will be different if you change line length, so the effective mesh near the via can also change. For practical purposes, the results still look very close, so I would not worry!

But if you wish to refine the mesh, you can find these settings in the mesh tab. I would in this case specify the absolute size, not cells per wavelength, but the end results is the same. Edge mesh is also very important where the current enters the via drill.

Q3). I want to use a simple inductor to model the via, the first one is L1 to L2, the other one is L1 to L3. The methods I used is to get the EM result for the two vias, and use a inductor to fit the curve of the S11 from the EM result. ( the setup schematic is shown in the link below )
But it seems the modeled inductance of the two vias is similar although the length difference between the two vias is different ( the via of L1 to L3 is twice length than L1 to L2 ) . Is my EM setup for the vias correct ? ( similar as question 1 ) , I exported the S2P file from the EM result and connect one pot to GND in the schematic diagram.

I would model it differently, because just line a piece of transmission line, the via has series inductance and shunt capacitance to the bottom ground. If you short the output port to that ground, you don't capture that properly.

For 1-port simulation of the series path, you can connect one floating Term (not TermG !) across (between) both ends of the EM model.
This also ignores the small shunt C, but is more correct from a simulation method viewpoint.

Good luck!
Volker

 

[edwardev]

Hello!



Yes, correct



The mesh of the line itself will be different if you change line length, so the effective mesh near the via can also change. For practical purposes, the results still look very close, so I would not worry!

But if you wish to refine the mesh, you can find these settings in the mesh tab. I would in this case specify the absolute size, not cells per wavelength, but the end results is the same. Edge mesh is also very important where the current enters the via drill.

View attachment 193628




I would model it differently, because just line a piece of transmission line, the via has series inductance and shunt capacitance to the bottom ground. If you short the output port to that ground, you don't capture that properly.

For 1-port simulation of the series path, you can connect one floating Term (not TermG !) across (between) both ends of the EM model.
This also ignores the small shunt C, but is more correct from a simulation method viewpoint.

Good luck!
Volker

Many thanks.

 

[edwardev]

I also got another question :
When I run the EM result for the blind via (L1 to L2) and buried via ( L2-L3), and change the height of the air. I used the same method as the previous post.
1st case : air is 0.5 cm
2st case : air is 100 cm.
the 0.5 cm case is more practical.

But the result is not a little different. At first, I think even though in the 0.5 cm case, most of the E-field should be restricted between the signal layer and the bottom layer since the height of the air is 10 times more than the substrate in the 1st case.
does anyone know is that something wrong in the 0.5cm case ?

substrate setup:

https://imgur.com/D8spYq3

Result between the two case ( red is 0.5cm case. blue is the 100cm case )

https://imgur.com/TcsQOkt

Many Thanks

 

[volker@muehlhaus]

It seems that you have built a resonator between the top and bottom ground which resonates at 40 GHz. Wavelength in air is only 7.5mm, so the mixture with FR4 and air might cause some resonace that invalidates your port calibration. The requirement for port calibration is that we have single mode propagation on the cal standards, so overmoded structures can become invalid.