README for NGSPICE
------------------
This long message describes what NGSPICE may become in the (near ?)
future. I used a question mark because, as you will read, most of the
features of ngspice are found on Hi-quality commercial products and
(which is the true reason) I have no idea on how can be implemented.
** Why resurrecting Berkeley's Spice ?
Berkeley's spice can be considered the father of most circuit simulators
available today. It is an old but still good piece of software. It may not
be the fastest or the most reliable, but it's free, it is available in
source code, and most of the electrical simulators inherited it's syntax.
On the more technical side, spice3f4(5) uses good numerical algorithms
(commercial implementations have only strengthened them), implements most
of the models for MOSFET submicron designs and has a powerful set of
analyses. On the more "social" side: it's well introduced in the
academic environment.
** What does NGSPICE mean ?
It stands for Next Generation Spice but that's not the official name of
the projest. This projects still lacks a name. NGSPICE is a temporary
name.
** What will NGSPICE be ?
Berkeley's Spice lacks in three directions:
a) Graphical user interface (I prefer to say "the framework").
b) Documentation.
c) Features in the code.
* The framework:
Spice is (and should continue to be) acommand line or a text tool, but
this makes very difficult to design large circuits. To overcome this
difficulty, a schematic entry tool and a waveform viewer tool are
needed. Nevertheless, there are other tools that can be useful:
a parts database, an editor which higlights the syntax, a symbol
editor, etc. Most of these program already exists in the open source
world, so they need ony to be integrated in a common EDA environment.
This is the first direction of development.
* Documentation: Commercial simulators (hi-end) have very good manuals
with tutorials, models equation explained, example of use, suggestions,
etc. This line of development has the task of provinding the final
spice user with an orderd and comprehensive set of information on the
program and it's features. The documentation should be useful for the
student as well as for the circuit professional.
* Improvements to the spice code: This is the hard part. The target of
this direction is to make ngspice a commercial grade simulator. This
means improving it's speed, it's numerical robustness, include the
latest models available and some other important features. I will
describe some of them briefly:
- Analyses -
Network analysis: given four nodes, extract z,y,s and the other
double bipole parameters.
Monte Carlo analysis: statistical simulation based on device
parameters tolerances.
Worst Case analysis: find the worst case of operation of a given
circuit based on device parameters tolerances.
Parametric analysis: repeat an analysis when one or more parameters
assumes different values.
- Devices -
Behavioral device: enhance the B device of spice3 to accepts IF THEN
ELSE conditions, and digital keywords like DELAY, HIGHV. LOWV, etc to
simulate simple digital circuits.
Dynamically loading of device: reduce the memory occupied by the
simulator using shared object code for devices. Each device
is a .so library that is inserted only if the circuit contains
an element modeled by the device. If we are simulating CMOS,
we do not need BJT or SOI (in most of the situations).
Code Level Modeling: Let users write their devices in C and use
them in the simulator. I have discovered a couple of standars
for doing this at the Sematech ftp site.
Improving device: Include additional parameters to some devices:
HDIF, LDIF, etc.
- Numerical Algs -
Integration: include (if necessary) more stable algorithms for
integration. What about Runge-Kutta Methods ?
How if we check LKC at every node as convergence test ?
Linearization:
Are there better algorithms for nonlinear equations the the Newton
raphson ?
- Sparse Matrix -
Faster handling of sparse matrices: new processors offers generous
caches and vectors operations.
- Options -
Possibility to mesure circuit pameters, like the delay between two
nodes, etc.
... and others that will emerge during the project.
Paolo Nenzi (pnenzi@ieee.ing.uniroma.it)
NOTES:
------
This version builds correctly on redhat linux 6.0, and most probably
on any glibc2-based linux distribution. The executables have hardly been
tested, all I know is that they build. The main goal of this release is
to get autoconf to work.
The goal is to put all possible configurations flags into the autoconf support.
It should build on most other unices (especially POSIX ones), please report to me
if it does, or if you know how to fix it if it doesn't.
As the -Wall flag is turned on during compilation, you will notice a HUGE
amount of warnings - our first job should be to work on those to 'sanitize'
the existing code.
MAILING LISTS:
-------------
There are three mailing lists dedicated to the development of ngspice.
ngspice-users@lists.sourceforge.net: This list is the list for the users of the
ngspice simulator.
ngspice-devel@lists.sourceforge.net: ngspice development issues. Developers and
"want to be" developers should
subscribe here.
ngspice-frontend@list.sourceforge.net: issues related to development of
frontends for ngspice.
To subscribe the list(s), send a message to:
<ngspice-users-subscribe@lists.sourceforge.net>
<ngspice-devel-subscribe@lists.sourceforge.net>
<ngspice-frontend-subscribe@lists.sourceforge.net>
WEB SITE:
--------
This project is hosted on Sourceforge.
The home page is http://ngspice.sourceforge.net
Manu (emmanuel.rouat@wanadoo.fr)
Paolo (p.nenzi@ieee.org)