Re: [Help-gnucap] Time step control

[al davis]

On Monday 03 December 2007, a r wrote:
> Still I have a feeling that playing with options is a hack.
> Why gnucap does not recognize that the state of the circuit
> (bsim3 transistors) changes and it should simulate the
> transition more precisely?

You might consider to try the exact Spice model.  All Berkeley 
variants of the BSIM3 model are in the models-bsim package.  If 
you are not sure which to use, the most recent "BSIM3v330" is 
probably the best choice.

To use the Berkeley model, build that library, 
("make" should build the whole tree)
then "attach -------".  You can do the attach as "-a" on the 
command line.


There could be a bug in the ".model" port of the model.  I do 
know that modelgen has the "hidden state" bug that people 
complain about in Spectre.  I also know that the capacitance 
portion of the port (d_mos8.model) has not been tested very 
well.

Now that gnucap can use the exact Spice model, the ".model" 
files that contain ported model code will be phased out.  In 
the future, a new version of modelgen will accept Verilog-A as 
the source language.  My effort has been directed toward that, 
rather than fixes to code that will be removed.



I am not aware of any simulator that gets the step control 
correct over a region change.  I know that Spice does not, that 
anything based on "ADMS" does not, and that anything that uses 
the Berkeley models as they come does not.  Gnucap has the 
mechanism for doing it, but the model needs to call it.  Spice 
doesn't even have the mechanism.


In the future, the "attach" command will be able to manage the 
compiling of a plugin, so you will be able to attach a source 
file or directory without worrying about the compiling details.