AnsweredAssumed Answered

How to define explicit ports in momentum and get a valid DC solution

Question asked by Kobi on Jul 19, 2017
Latest reply on Jul 26, 2017 by Kobi

Like • Show 0 Likes0
Comment • 8

I have a layout structure that implements GSG pads for transistor characterization.

The GSG pad is connected to a co-planar transmission line that eventually connects to the device. Thus, I defined the Momentum ports as explicit ports where the Signal pad is the positive terminal of the port and the 2 ground pads are the negative terminals of the port. When I run the simulation with the active device the DC operating point is not valid. Only when I set the Momentum ports to implicit i.e. single-ended and referencing the global ground do I get a valid DC solution. Unfortunately this is not the correct way to represent the layout and the EM problem and I wonder if anyone knows how can I use explicit ports and still get a valid DC solution.

No one else has this question

Outcomes

8 Replies

volker_muehlhaus Jul 20, 2017 12:38 AM
Mark Correct
Correct Answer
Hello kobi,

Have you used RF mode or Microwave mode?
Do the Y/Z parameters look smooth / reasonable down to DC?

What is the structure size in wavelength - any change to extract a well behaved physical model from the S-params?

Best regards
Volker
Like • Show 0 Likes0
Actions
Kobi Jul 23, 2017 9:09 PM
Mark Correct
Correct Answer
Thank you for your reply Volker.

Actually, I think that than issue I am describing is more fundamental than solution smoothness.

Let me describe the issue in more detail.
There is a structure of 2 x GSG probe pads that connect to a transistor via short transmission lines as depicted below.
The probes have a bias T option to provide DC bias voltage to the gate (left GSG pads) and drain (right GSG pads) of the device.
The transmission lines are implemented as co-planar TL structure.

In Momentum port assignment it makes sense to use differential ports for each signal pad in the following way. Pins 1 and 2 are defined with explicit reference negative pins in order to make sure that the distance between the ports terminal will both exceed on tenth of a wavelength in the max simulation frequency.
Port1: P1 positive terminal, P2 and P3 negative terminals
Port2: P4 positive terminal, P5, P6 negative terminals.

After the EM simulation is complete I use the S-parameter ports in the following way in order to capture the actual current return path in the co-planar structure. Each S-parameter port has a DC component of say 0.6V.

The transmission coefficient (S21) between the input pad to the DUT terminal looks fine as depicted in the plot below but the DC operating point is not valid because it has negative component on the ground plane (-150mV).

The intuitive solution (for me) is to reference the ground plane to ground in the following way:
When I measure the transmission between port 1 and port 2 I get the following:

I don’t understand this result.
Like • Show 0 Likes0
Actions
- volker_muehlhaus @ Kobi on Jul 24, 2017 6:15 AM
  Mark Correct
  Correct Answer
  Hello Kobi,
  
  I see ... I was thinking you have issues with small leakage currents from small precison issues with low frequency EM results, but this is indeed more fundametal!
  
  When you define the differential port between the GSG pads, this defines a specific mode: all current flowing into the signal pad must return through the ground pad. The 2-port results contain no information about the global ground/substrate path, and the GSG path is truely floating due to this port definition.
  
  This means the voltage between your local port ground and global ground is undefined. But ADS needs to assign a useful value to run simulation. In this case, ADS assumes a differential feed at the port, with symmetric voltage relative to global ground. That is why you see all port voltages symmetric to global ground, including DC, with +1/2 Vport at one pad, and -1/2 Vport at the other pad.
  
  The potentials against global ground might look surprising, but the voltage between the GSG pads is your desired voltage. So your transistor bias works just fine, if connected between your circuit's signal and circuit ground. Only the voltage marker values are unexpected, because they refer to that global (not local) ground. If you connect each port/term grounds to global schematic ground, you should see the "expected" DC voltages.
  
  Have a look at the two schematics below, the only difference is the extra global ground node at the left port (Term2).
  You can see that the expected voltages at that left circuit side are now displayed relative to global ground. But hey, what's on the right side? That's what I wrote above: the S-parameters only have information on the desired mode, i.e. the voltage between signal and ground on each pad. There is no additional information on other modes, so these differential 2-port S-params don't provide information that there's a connection from left GSG ground to right GSG ground. That's why we would also need to connect the other port's ground to global ground, if we wish to display the DC voltage against global ground.
  
  And here's something that might not be intuitive: You can connect your local grounds with global ground, and your return currents will still flow only through the EM ground path, as desired, because that is what you enforced with your differential port definition.
  
  Confusing? Yes, I know these local grounds are confusing, especially when using look-alike symbol that seem to enable all sorts of connections between various nodes. If you use the full 6-port EM data, there is more information in the EM results to described all these other modes/current paths. The 2-port data created with the differential ports is much more restricted, and not complete as far as voltage to global ground is concerned.
  
  Best regards
  Volker
  
  PS: An old appnote that I wrote on this topic a few years ago:
  Dr. Mühlhaus Consulting & Software GmbH » The ground pin confusion in EM transmission line models
  Like • Show 0 Likes0
  Actions
  - Kobi @ volker_muehlhaus on Jul 24, 2017 9:02 PM
    Mark Correct
    Correct Answer
    Hello Volker,
    
    Thank you for sharing the app note, it is very instructive and relevant.
    I agree with your statement in the app note that at least for SMD devices it is better in most cases to use the global ground and not restrict the current to be purely differential between the differential ports and thus not neglect any shunt effects.
    However (as you also mentioned) there are some case where you have no other choice but using an explicit ground reference for example:
    1. Using SMD devices when there is no global ground plane available as in the case a single layer design. This is a trivial case.
    
    2. Using co-planar transmission line. In this case the assumption that the reference ground on both sides of the transmission line will carry most of the returning current and will be used by the EM field is valid although we still lose some information regrading any shunt effects due to the presence of a global reference ground as in the case of GCPW structure.
    
    3. In cases where the global reference ground is electrically far from the ( + ) terminal of the port. In this case using the global reference ground as the ( - ) terminal will violate the restriction of electrically short distance between the port terminals and might result in potentially nonphysical S-parameters file.
    
    In my original question I have a combination of the cases #2 and #3. I have a CWP transmission line and my global reference ground is electrically far from the metal layer that the CPW is implemented. This is a case of a very high frequency application. Thus I guess that I don't have any other choice but defining explicit reference terminal for each EM port. Do you agree?
    
    In the example that you shared in your reply you showed that the DC solution for the floating ground case is +500mV and -500mV. I agree that it does really matter as long as the potential between the signal pad and the reference ground is what we defined in the DC voltage source.
    But what happens if I short the ground pads on both sides of the structure? we would expect that the DC solution will arrive at the same potential for the ground. But this is not what I get in the simulation.
    In the picture below, each signal pad was defined as differential port where the 2 adjacent ground pads were used as the ( - ) terminal. When I run the DC simulation I get very strange results. The DC solution on the left Ground pin is -300mV while the DC voltage on the right ground pin is -600mV. These ground pins are sharing the same metal plane. how is this possible?
    
    Also When I connect my device to the ports dedicated to it (in the middle of the transmission line) I get nonsense.
    
    Do you have any thoughts to share based on your experience on what is the correct way to define EM ports for this structure?
    
    Thanks,
    Kobi
    Like • Show 0 Likes0
    Actions
    - volker_muehlhaus @ Kobi on Jul 25, 2017 2:04 PM
      Mark Correct
      Correct Answer
      Hi Kobi,
      
      I fully understand what you are talking about, because I'm working as an EM specialist on these exact RFIC modelling topics for many years. You might have missed some details in my answer, so let's have a look again.
      
      I agree that GSG is the right excitation, but I also pointed out that these 2-port GSG results are only a subset of the full 6-port solution. EM solvers like Momentum or Sonnet internally calculate the 6-port solution for the 6 pins, and then internally reduce this to 2-port data based on your pin/port grouping. So technically, there is no accuracy advantage in using the reduced data. You could also use the 6-port data, and then wire at schematic level to achieve the same result.
      
      I agree that GSG from a 2-port simulation is fine, if you accept the reduction in information/data that you get from this simulation. You loose data on the path from your localground to global ground (that's why ADS voltage readout references to global ground doesn't work as you expect) and you loose information on the path from one local ground to the other local ground (that's why connecting one port's local ground does NOT create a path to the other local grounds). This is not an ADS issue, it's by definition of the 2-port S/Y/Z-parameters.
      
      You asked: "In the picture below, each signal pad was defined as differential port where the 2 adjacent ground pads were used as the ( - ) terminal. When I run the DC simulation I get very strange results. The DC solution on the left Ground pin is -300mV while the DC voltage on the right ground pin is -600mV. These ground pins are sharing the same metal plane. how is this possible?"
      That's exactly what I described above: the 2-port S/Y/Z-parameters give you information about voltages & currents at port 1 (referenced to port 1 local ground) and port 2 (referenced to port 2 local ground). There is no information about voltage between the local grounds.
      Does this 2-port data tell you if the grounds are connected? No! If you have an impedance (or gap) in that path, you can't tell from the 2-port data if that is in the signal conductor, or the ground conductor, or both. So the S2P results doesn't connect the grounds, that's by definition of the 2-port S/Y/Z-parameters.
      
      You are asking for 2-port data of a 3 conductor system, and that's where things go wrong.
      
      What to do?
      You can use the 2-port data and add missing information at schematic level, e.g. by connecting global and local grounds for voltage readout. Different from what you expected, this does not bypass your layout ground path, you can try that with a testbench. Actually, you should try that, to get a better feel for data that you can (and can't) get from local ground ports.
      
      Or you can simulate as a 6-port, and wire at schematic level. This wiring in schematic eliminates issue #3 from your port above, because for GSG path we combine pins/ports that are physically close, and don't evaluate questionable combinations of far-away terminals. For the GSG path, this will give the same results as telling the ADS port editor to do the 6port->2port reduction for you. I know that you are sceptical, so please try yourself and check that this is indeed identical for the RF path, just adding extra information on the substrate path.
      
      Best regards
      Volker
      Like • Show 0 Likes0
      Actions
      - Kobi @ volker_muehlhaus on Jul 25, 2017 1:21 PM
        Mark Correct
        Correct Answer
        Hello Volker,
        
        Your comments make a lot of sense and it seems that at this point I just have to experiment the different port combinations in order to get more confidence in these measurements.
        
        I have only one concern regarding your comment "I agree that GSG is the right excitation, but I also pointed out that these 2-port GSG results are only a subset of the full 6-port solution. EM solvers like Momentum or Sonnet internally calculate the 6-port solution for the 6 pins, and then internally reduce this to 2-port data based on your pin/port grouping".
        According to what I understand from your comment, the EM solver will always solve for the 6 ports i.e. each port will be referenced to the global reference ground and then it will reduce it to a 2-port solution. If this is correct then why did we bother in the first place to define an explicit ground terminal for each port i.e. the ( - ) terminal from each side of the signal pad? didn't we do it because the distance of the ( + ) terminal to the global reference ground was electrically long and could potentially result in non-physical S-parameter solution?
        Am I missing anything here?
        
        Thanks,
        Kobi
        
        Like • Show 0 Likes0
        Actions
        volker_muehlhaus @ Kobi on Jul 25, 2017 2:06 PM
        Mark Correct
        Correct Answer
        Hi Kobi,
        
        I understand your concern.
        
        Having worked closely with Sonnet EM solver guys, I can assure that Sonnet MoM does this internal port reduction with explicit grounds, and I'm 99% sure that Momentum does the same thing internally. If you look in ADS online help at the description of Momentum GSG ports, you will see a picture showing how S2P "recombined S-parameters" are calculated from 6 ports that reference global ground.
        
        qthelp://ads.2016.01/doc/em/Ports_for_Momentum.html
        
        But maybe you can ask Keysight support (or the Momentum developers in Gent) to confirm that.
        
        Why would we do this? I think it's to prevent the user from seeing weird data in some places of the full S-matrix, where pins/ports are at large distance and that particuluar Sij isn't physically valid/meaningfull at the upper end ofthe frequenbcy band. But that's only my best guess, there might be other reasons.
        
        Best regards
        Volker
        
        PS: One thing that comes to my mind: If you use GSG 2-port data, and don't want to pull your GSG grounds to global ground with a direct connection, because you are worried what that does to RF return, you can also try a 1MOhm resistor. That's enough to give that node a defined DC voltage.
        Like • Show 0 Likes0
        Actions
        Kobi @ volker_muehlhaus on Jul 26, 2017 2:29 PM
        Mark Correct
        Correct Answer
        Hi Volker,
        
        I like your idea of installing 1MOhm resistor to pull the GSG ground to global reference ground. It makes a lot of sense to me because this is an efficient way to restrict the RF signal to the differential GSG arrangement and let the DC operating point to be defined "artificially" by forcing the S-param engine to reference the DC solution to the global ground. Thanks.
        
        Regarding the usage of differential ports, my take from this discussion is that by restricting the current to flow differentially between 2 terminals i.e. using explicit ground, we are actually losing information. Thus I can't think of any useful scenario that one might want to use explicit ground terminal unless this is the only way to ensure that the electrical distance between the 2 terminals of a port is <1/10 of the wavelength. Even for the SMD device case the shunt effects are lost by defining a differential port.
        
        The fact that the EM engine is solving for the full S-parameter matrix using an unknown reference ground still remains a mystery to me. I wish I could get some insight into the MoM engine implementation by the MoM developers in Gent but I find it hard to believe that it will happen.
        
        I really appreciate the useful discussion on this topic with you as it allowed me to better understand the DC solution with differential ports. And it also allowed me to define for myself more accurately what I don't understand.
        
        Thanks a lot,
        Kobi
        Like • Show 0 Likes0
        Actions