Re: initial_step and final_step in DC sweep

Geoffrey,

I have taken a look at your MOS11 model: I see that you have totally 
abstained from using initial_step and final_step, but also from the 
analysis functions. I assume that you're relying on compiler optimizations 
that let the separate sequential blocks only be run when none of the 
module-scope variables they use changes value. Is that correct? Doesn't 
that introduce wildly varying run times in the various simulators? 
Wouldn't analog conditional statements be a more obvious guard against 
unnecessary execution of these blocks?

Marq


Marq Kole
Competence Leader Analog Simulation, Philips ED&T









"Geoffrey.Coram" <Geoffrey.Coram@analog.com> 
Sent by:
owner-verilog-ams@server.eda.org
25-08-2006 15:22

To

cc
verilog-ams <verilog-ams@server.verilog.org>
Subject
Re: initial_step and final_step in DC sweep
Classification







Marq -

Ack!  Those Silvaco models ... they were a true mess when I 
first saw them (they had capacitance models implememented
with a $sit_get_prev(), where sit = silvaco internal task,
so they weren't charge-conserving).  I didn't actually code
the model myself; I gave Silvaco some feedback and they put
my name in the header as though I was the author.

If you look at my MOS11 model, also on the Designer's Guide
site, you'll see that I do not use initial_step.

I'll admit, I was forced to use initial_step for one RF
simulator, where we were only interested in transient
performance, and it was not smart enough at the time to
build the dependency tree.

-Geoffrey


Marq Kole wrote:
> 
> Geoffrey,
> 
> :-))  <---- really big grin
> 
> In my most recent case it is a BSIM4 model that you coded, available 
from Silvaco, linked through by the Designer's Guide. Admittedly, it is 
from March 2004 (says the header - it's version 4.3.0 so it can't be older 
than 2003 when that model was released in C code). I've been updating it 
to version 4.4.0.
> 
> Rest assured that you're far from the only one; among the many others 
I've always been doing it myself. I'll to change my ways ... :-)
> 
> I'm more concerned about implementers having this implemented without 
considering the nitty-gritty detail of the standard.
> 
> Marq