+ Post New Thread
Results 1 to 8 of 8
  1. #1
    Full Member level 4
    Points: 1,228, Level: 7

    Join Date
    Jan 2018
    Posts
    201
    Helped
    1 / 1
    Points
    1,228
    Level
    7

    How to reduce Harmonic Balance simulation time in Spectre

    Hi, I am running Harmonic balance simulation for my circuit at different frequencies to find loadpull values but it takes lot of time sometimes even half a day to finish with all the frequencies. Especially it is slow at low frequencies in my case , for example at 4 GHz or 8 GHz. Is it usual to take this long? or what parameters I should consider changing in my circuit like biasing,Circuit stability, voltage source values, and transistor width and sizes?. I chose Number of Harmonics =11, Oversample Factor =9 and in loadpull section, I chose 40 number of steps for rho and phi. Thanks

    •   AltAdvertisement

        
       

  2. #2
    Advanced Member level 5
    Points: 39,705, Level: 48

    Join Date
    Mar 2008
    Location
    USA
    Posts
    6,413
    Helped
    1872 / 1872
    Points
    39,705
    Level
    48

    Re: How to reduce Harmonic Balance simulation time in Spectre

    Minimizing the number of distinct nodes and devices is key.

    At my last employer they struggled with this too, and got a
    lot of improvement by having one of the CAD group folks write
    a Perl script that would combine parallel FET fingers into a
    single m=XXXX device (rather than XXXX of them, which is
    what comes from a M0<1023:0> type instantiation).

    You might look to your schematic style and the netlist form,
    to see whether you are dragging that kind of baggage. May
    not be practical to do the same coding, but maybe you can
    represent your circuit in a way that creates less solution
    overhead. That is,

    M0 NS NG NS NB w=xxx l=0.0yy nf=1024

    instead of

    M0<127:0> NS NG NS NB w=xxx l=0.0yy nf=8

    or some such. In many technologies it's not practical
    to make super high finger count FETs, tap rules force
    you to break them up, and literal-minded designers will
    represent the physical form N distinct active area
    regions, hend N FETs) in the schematic even if a single
    FET instance is just as realistic.

    RF switches are one such critter. CMOS PAs, another.


    1 members found this post helpful.

  3. #3
    Advanced Member level 5
    Points: 30,406, Level: 42
    BigBoss's Avatar
    Join Date
    Nov 2001
    Location
    Turkey
    Posts
    4,458
    Helped
    1340 / 1340
    Points
    30,406
    Level
    42

    Re: How to reduce Harmonic Balance simulation time in Spectre

    Oversample Factor is too much that's why your simulation is so slow..
    Since the waveform does not have extreme shape ( ringing,overshoot,hardly squared etc. ) you don't have to increase Oversample Factor much.
    It should be 3-4 max. for sinusoidal-like waveforms.



    •   AltAdvertisement

        
       

  4. #4
    Full Member level 4
    Points: 1,228, Level: 7

    Join Date
    Jan 2018
    Posts
    201
    Helped
    1 / 1
    Points
    1,228
    Level
    7

    Re: How to reduce Harmonic Balance simulation time in Spectre

    Quote Originally Posted by dick_freebird View Post

    M0<127:0> NS NG NS NB w=xxx l=0.0yy nf=8
    Thanks. does M0<127:0> represent number of multipliers?. In my case I have 4 multipliers and number of fingers 125. But If I change it to just the number of fingers =125*4 and multiplifer =1, the output power that I get is decreased.



  5. #5
    Advanced Member level 5
    Points: 16,998, Level: 31
    pancho_hideboo's Avatar
    Join Date
    Oct 2006
    Location
    Real Homeless
    Posts
    2,612
    Helped
    701 / 701
    Points
    16,998
    Level
    31

    Re: How to reduce Harmonic Balance simulation time in Spectre

    Quote Originally Posted by circuitslave View Post
    Thanks. does M0<127:0> represent number of multipliers?.
    No.
    This is a repeated instance.
    However if you generate layout from schematic, this generates same layout as multiplier.

    Quote Originally Posted by circuitslave View Post
    In my case I have 4 multipliers and number of fingers 125.
    But If I change it to just the number of fingers =125*4 and multiplifer =1,
    the output power that I get is decreased.
    Surely consider physical device structure.
    They give different layout.



  6. #6
    Advanced Member level 5
    Points: 16,998, Level: 31
    pancho_hideboo's Avatar
    Join Date
    Oct 2006
    Location
    Real Homeless
    Posts
    2,612
    Helped
    701 / 701
    Points
    16,998
    Level
    31

    Re: How to reduce Harmonic Balance simulation time in Spectre

    Quote Originally Posted by circuitslave View Post
    I chose Number of Harmonics =11,
    Oversample Factor =9
    Why do you set Oversample Factor =9 ?

    Can you understand Harmonic Balance Analysis ?
    Can you unsderstand a meaning of "Number of Harmonics" and "Oversample Factor" ?

    Show me your netlist portion regarding analysis, option and save statment.

    If such large "Oversample Factor" is truely required, use Shooting-Newton-PSS instead of HB-PSS.

    Quote Originally Posted by circuitslave View Post
    Especially it is slow at low frequencies in my case , for example at 4 GHz or 8 GHz.
    Show me simulator logfile.



  7. #7
    Full Member level 4
    Points: 1,398, Level: 8

    Join Date
    Jan 2019
    Location
    Belgium
    Posts
    199
    Helped
    83 / 83
    Points
    1,398
    Level
    8

    Re: How to reduce Harmonic Balance simulation time in Spectre

    What are you simulating that requires an oversample factor of 9?
    Here are some guidelines from Cadence website: https://community.cadence.com/cadenc...balance-part-1



    •   AltAdvertisement

        
       

  8. #8
    Full Member level 4
    Points: 1,228, Level: 7

    Join Date
    Jan 2018
    Posts
    201
    Helped
    1 / 1
    Points
    1,228
    Level
    7

    Re: How to reduce Harmonic Balance simulation time in Spectre

    Quote Originally Posted by circuitking View Post
    Hi, I am running Harmonic balance simulation for my circuit at different frequencies to find loadpull values but it takes lot of time sometimes even half a day to finish with all the frequencies. Especially it is slow at low frequencies in my case , for example at 4 GHz or 8 GHz. Is it usual to take this long? or what parameters I should consider changing in my circuit like biasing,Circuit stability, voltage source values, and transistor width and sizes?. I chose Number of Harmonics =11, Oversample Factor =9 and in loadpull section, I chose 40 number of steps for rho and phi. Thanks
    I realized that it is because my circuit is unstable at the frequencies where I am doing loadpull.



--[[ ]]--