Re: [Ngspice-devel] Some strange behavior of bsim4 model

[Robert L. <Rob...@t-...>]

Robert Larice <Rob...@t-...> writes:

>   It seems to me BSIM4GEgpPtr is rotten in the second run.
>   This is a pointer to a matrix element, and is engineered from BSIM4gNodePrime
>   I suppose, these extra nodes (BSIM4gNodePrime etc)
>     have to be allocated for the SPmatrix,
>   and it seems to me this is done only on the first run.
>   On the second run the new matrix doesn't know of these extra nodes,
>     and the element pointer points to an unused slot of the matrix.
>   So far just guessing.

Hello again,

  in b4set.c there is:

            if ((here->BSIM4rgateMod > 0) && (here->BSIM4gNodePrime == 0))
            {   error = CKTmkVolt(ckt,&tmp,here->BSIM4name,"gate");
                if(error) return(error);
                   here->BSIM4gNodePrime = tmp->number;
                ...
            }
            else
                here->BSIM4gNodePrime = here->BSIM4gNodeExt;

  this is executed once for every `op' analysis.
  on the first run, BSIM4gNodePrime == 0,
    and the conditional if-body is executed.
  on the second run, BSIM4gNodePrime != 0,
    and is thus overwritten with an incorrect value (..gNodeExt)

  Now I don't know what the correct fix is.
  Either, the code was written with the implicit assumption, that
    the whole struct has to be bzero'ed on every new analysis.
  Or, the == 0 query should be just deleted,
  Or, CKTmkVolt is indeed meant to be called only on the first run,
    and the else-body should be guarded.

Regards,
Robert