Re: [ng-spice] Design Document Comments


To ng-spice@ieee.ing.uniroma1.it
From Manu Rouat <emmanuel.rouat@wanadoo.fr>
Date Thu, 07 Oct 1999 22:05:36 -0400
Delivered-To mailing list ng-spice@ieee.ing.uniroma1.it
Mailing-List contact ng-spice-help@ieee.ing.uniroma1.it; run by ezmlm
Organization la Guilde
References <Pine.LNX.3.96.991007103012.23935A-100000@ieee.ing.uniroma1.it >
Reply-To ng-spice@ieee.ing.uniroma1.it
Sender rouat@war.wanadoo.com

Paolo Nenzi wrote:

> DIRECTORY ORGANIZATION
> ======================
> I like the structure organization you made, but how do you think to split
> the existing code into the new structure ?

It's work, and it will take some time..... we can start as soon as we agree 
on a tree - maybe we can do that now?
The transition will most probably be a bit chaotic, but it can be done if
we are well organised.

> Let's talk about the parser and the simulation kernel:
> 
> these are the more important and controversial parts of ng-spice: kev
> would like to make a mixed mode simulation kernel to allow for
> verilog/spice concurrent simulation.
> 
> Any documents about parser and sim kernel writing ?

There are 3 types of mixed-mode simulations that I know:


- use the analog kernel for all components (thus they all have an analog 
description)
but it is possible to have different levels of precision for different blocks 
(ie high
precision for pure analog blocks , low precision for digital blocks to speed 
things up)

- use one program that is 'on top' of an analog and a digital simulator - 
this is the
method used by Mentor Graphics - it works, but it is really heavy. The 
program simulates the
analog part and the digital part separately with the appropriate simulator 
and takes care 
of the interaction between the two. I suppose that both simulators must be 
coded to allow
this.

- use a true mixed-mode simulator like Avanti's Starsim - it is a completely 
different
method of simulation (transistor charateristics are 'precalculated' using an 
analog
simulator, and put into tables - these tables are then used by the simulator 
instead
of recalculating everything).



My feeling is that we should stick to a true analog simulator, at least in 
the beginning.


manu



Partial thread listing: