Re: [ng-spice-devel] personal note on ACS and NG-SPICE (fwd)
On Mon, 11 Sep 2000, Alan Gillespie wrote:
> ...............How "credible"
> as a re-vamped ACS versus a Spice rework ? I don't mean to be
> cheeky, Al, it's just a matter of perceptions, i.e. Spice is an industry
> standard, a benchmark.
Spice is an industry standard. That can't be denied. There are
those who love it and will never change. There are those always
looking for change. Spice has its problems, and I hear about them
every day. Some of this is from the industry leaders.
It is not all transistor circuit design. One problem I heard of
today is from a connector company making simulations of their
product. It took Spice many hours to do a run, and the connector
wasn't all that big. Only 40 pins, but they wanted a full crosstalk
analysis in the time domain. What about a 4000 pin connector. These
things actually exist. You don't do that in Spice. I told him that
ACS would help in this case. It is a lot faster for this type of
problem. This is just one. Just another bit of evidence that a
change is long overdue, and won't come from the corporate world.
As to what is actually in use ..... Some of the large semi companies
use their own simulators. They need source. They need real support
and features that the EDA vendors cannot provide. The mid-sized and
smaller ones, and the "fabless" semi companies can't afford this, and
need it just as much. Free software is a way they can get it. It is
also a way that the creators of Free software might be able to make
some money.
Of these in-company simulators, some are Spice derivatives. Some are
not. Looking at the simulator products for sale, the same is true,
although some that are not Spice derivatives still have Spice in the
name. Some of those that are Spice derivatives have been completely
rewritten, and have no Berkeley code remaining.
Of the commercial simulators, both for sale, and in-house, the ones
with the highest reputation are the ones that deviate the farthest
from Spice, although they may have been based on Spice at the
beginning.
There are specialized simulators, that focus on a particular task.
These are usually not Spice derivatives, and they make a big deal of
it. They openly point out what is wrong with Spice.
They may be made by the same company that makes a general purpose
simulator, but by a different division. Often the politics prevents
them from sharing even with another division of the same company, as
if they consider them to be competitors. Sometimes they do not
disclose source, even within the company, resulting in a lot of
duplicated effort. There is no incentive to really extend the state
of the art. It is all driven by marketing.
This is where Free software can really show its best. It is driven
by engineers, not marketers. We can share, even if we are working on
a different project.
>> Do we have the resources to do both?
Doing both is the only way. We have Spice, which is old,
traditional, and standard. We have ACS, in which I intend to extend
the state of the art, publish the work, etc.
The thing old Spice needed most was to collect stuff that was done
for it. Berkeley stopped at BSIM2, but distributed 3, 4, and SOI
separately. There is that other BJT model. What Spice needed most,
to keep it useful yet traditional, was to collect these and make a
package that had it all in one place, so someone could get the
package rather than going to the dozen other web sites and getting
all this one by one, combining, doing minor edits, etc. This is
NG-SPICE. I think this is the path NG-SPICE should take. There is
still more to do.
Moving on with the algorithms, and advancing the state of the art is
where ACS comes in. If this was a big EDA company, they would
continue both products. (Example: Cadence has PSPICE and Spectre.)
Some things must be done for both. There is a big demand for
something like the model compiler, that can portably generate models
for any simulator from the same imput language. Doing it with both
ACS and SPICE proves that it can be done. Doing it for one or the
other is "yet-another ...". If we can do it, it has the potential to
become an industry standard, possibly even an official one,
sanctioned by EIA and supported by all major simulators.
Another thing that came up is the need for simulators to be able to
call external compiled models. These models could be shipped as
object modules, and called by the simulator. It is easy to do this
for one simulator. It is hard to do it in general. This has never
been done. We can. There is BIG demand for this. Again, if we can
do this, we score big points, possibly even funding from the big
guys, and possibly an OFFICIAL standard that the big EDA companies
would support.
Just today, I was at a committee meeting on one of those official
standards. (IBIS). Everyone involved was happy that a Free
simulator supporting IBIS will be available, including the
representatives of the big EDA companies. ACS will support IBIS
soon. With a little help from someone here, NG-SPICE can, too.
So, how credible will a non-spice simulator be, if it does all this?
Remember that for a Free simulator, there is no risk. I hear the
demand for it often.
Somebody from a big EDA company said at the open-source birds of a
feather session that there would be no real innovation in open
source. We need to prove him wrong.
Partial thread listing:
- Re: [ng-spice-devel] personal note on ACS and NG-SPICE (fwd), (continued)