#163 BJT modeling error

0.0.18
closed
nobody
None
5
2014-08-23
2014-06-05
Dwight Wildon
No

Hi,

I am an engineer building a HF transistor library NXP SiGe_C transistors
the published models have the transit time parameters:
Tf="1.3236e-12" Xtf="43.92" Vtf="-17.68" Itf="89e-3"
QUCS gives Range ERROR for Vtf, and and aborts Sim.
Have manually computed the Tff value from Tech paper, and results are quite reasonable. I beleive the range checking code needs modification.

Discussion

  • Guilherme
    Guilherme
    2014-08-15

    Maybe we can change the Error into a Warning. Can you provide further details?

    The error is about Vtf orTf`?

    Currently Qucs expects Vtf > 0.

    It might happen that in order to fit the measured devices to the bjt model (Spice Gummel-Poon I believe) the manufacturer is tweaking the model parameters.

     
    • Dwight Wildon
      Dwight Wildon
      2014-08-19

      I'm sure tweeking was what it was.
      Error to Warning sounds reasonable, as long as it conyinues the
      simulation.

      Attached is a zip file of pertinent data:

      NXP model source
      My library
      test project
      pdf data sheet for one of the transistors.

      You should note that NXP publishes the exact same model for all of the
      transistors in the family.The "area" factors were added by me to obtain
      a complete model for all of the series except for BFU710F which the
      original model fits with an area of 1.0.

      Dwight Wildon dwildon@cfl.rr.com

       
      Last edit: Guilherme 2014-08-19
      Attachments
  • Guilherme
    Guilherme
    2014-08-19

    The attached patch allow the simulation to continue and issuing a warning instead of failing. Something like:

    creating netlist...
    NOTIFY: SW1: running netlist for Ibb = 5e-06
    NOTIFY: SW2: running netlist for Vdd = 0
    WARNING: Unphysical model parameter Vtf = -17.68 in BJT `BFU790F.SUB1.chip_790'
    NOTIFY: DC1: creating node list for DC analysis
    NOTIFY: DC1: solving DC netlist
    NOTIFY: DC1: convergence reached after 10 iterations
    

    Are you compiling from source or using a distributed binary?

    I see that the BUF710 has Vtf=0 while the others three models have Vtf=-17.68.

    May I ask how do you calculate the Vtf parameter? When you say 'original model fits', are you fitting to the datasheet curves?

     
    • Dwight Wildon
      Dwight Wildon
      2014-08-23

      Sorry, I changed all of the Vtf params to "0" so that I could optimize
      the Area param. I must have forgotten to change the BFU710 back after
      testing the DCs

      Vtf param does not affect the DC characteristics, so it can be made 0 to
      run DC's for area calculation testing.

      I took Vtf from the NXP model without change. When I said the original
      model ran correctly, I meant the DC's matched the published typical
      curves without scaling. Iin Spice the Area factor is normally part of
      the model call, that is it is supplied with the net list insertion. In
      qucs, it is part of the model. I tink this is better for discrete part
      simulation, but the spice way probably works better for IC transistor
      level modeling.

      Thanks for your help, will try patch when I get a chance.

      Am currently running from compiled code, but I can re-compile if needed.
      Version Ubuntu 14.04, qucs-0.0.18.131128

       
      Last edit: Guilherme 2014-08-23
  • Guilherme
    Guilherme
    2014-08-23

    • status: open --> closed
     
    • Dwight Wildon
      Dwight Wildon
      2014-09-02

      Please find attached the final version of the NXP SiGe family of
      transistors. The devices have two emitter connections. the "fat" one has
      lower inductance. At low <10GHz frequencies if simply tied together
      adequate modeling will result. At higher frequencies, the delay and
      parasitics in the connection must be considered.

      Have a good day!

       
      Last edit: Guilherme 2014-09-02
      • Guilherme
        Guilherme
        2014-09-02

        Thank you for sharing!

         
  • Guilherme
    Guilherme
    2014-08-23

    The patch is now merged into the repository.
    It will soon be available in the next snapshot or stable release.
    If you wish to contribute and to have the component library integrated into Qucs, please attach its final version.