Electronics Forum / CAD / October 2005

Perhaps it was too subtle for you, so I'll highlight it for you.
http://groups.google.com/group/sci.electronics.cad/browse_frm/thread/fd9f4d78f03
7c5d1/c2037bd721acf107?q=anasoft+SuperSpice+Kevin-Aylward

Kevin is not only an engineer; not only a SPICE user;
he in fact PRODUCES a well-known variant.

Because he *maintains* it (read: bug reports),
he is very aware of the shortcommings of it
and he tweaks his software to adapt to those as they arise.

Yes, I know all about Bob. He is misguided on this.

I have made my own opinion. Its based on being both an analogue ic and
board designer for er.. some years, and knowing how its actually done in
practise.

I use to

Of courses Spice has its limitations, just as a screwdriver does.
However, this doesn't mean that Spice shouldn't be used as the
fundamental design tool for analogue design.

You are obviously a newbie on this so I'll point out one or two issues.
Lets take analogue ic design. How do you propose to design a 1000
transistor circuit? A 10,000 transistor circuit? What's the fab cost?
Turn around time? You reckon that you can solve the equations by hand?

This is the deal. 10,000s of analogue ic designers, that is *all* of
them use spice as *the* number one de-facto method of designing
circuits. Period. It cant be any other way, today. Its simply not
possible to reliably design such circuits without spice. The designs are
two large and complex and cost too way much to fab. Its typically a 40
hour day, 5 days a week of solid simulation. This *is* the way it is.
Its quite common for people to design large analogue circuits, and have
them work 100% with first pass silicon. Some even get a $50k bonus on
that condition. Those that suggest that Spice is a side line tool, are
on a par to claiming that a Bible is just a superficial add-on to the
x-tian religion. They are quite oblivious to what practising analogue ic
designers use as a matter of course on a daily basis.

For the most part, designs don't work because of simply neglecting to do
a specific simulation, rather then the simulation itself not reflecting
real life. Its hard to think of all operating conditions. However, most
spices have various feature that allow worst case analysis to be
performed and other such multyruns. One might typically do 10,000
variations of a circuit. How do you propose to do such checking in the
real world?

Spice is like anything else, GIGO. Realistically, there is no
alternative. Even a 1 transistor circuit has no exact analytical
solution. The key is getting good models, and understanding the model
failings and compensating for that in the design.

Of course designs have to be physically checked on the bench, but if you
do know what you are doing this checking can be very, very, minimal.
Down to just producing a data sheet for example. I am sure Jim T. could
give us a few examples of right first time:-)

Kevin Aylward
informationEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

[. . . .]

Heh.  Another reason why open-source EDA tools are preferable over
secret-source ones: No obsolescence.  In particular, no obsolescence
based upon stupid political or marketing considerations.

http://geda.seul.org/
http://ngspice.sourceforge.net/

SDB

I think a great many (most ?) problems with SPICE and other simulation
programs in general are actually due to problems of the "Floating
Point" data type. AFAIK the total reason for being of the floating
point data type was to get a reasonable range and precision using as
little memory as possible. Today memory is not a problem anymore, and
one can use a fixed point number format with the desired range and
precision necessary for any simulation. A typical construct in many
simulations are:

(x0-x1)/k where x0 and x1 are almost equal. This causes problems in
floating point. If x0 an x1 are say 1.0 and 1.0001 then it is not a
problem. If it is 1000000000.0 and 1000000000.0001, then it bombs out.

I personally think that with todays systems, the use of floating point
should be banned, and in stead large fixed point numbers should be
used. The only disadvantage compared to floating point is that it uses
more memory. (And the little problem that almost no currently used
languages supports them as standard)

Regards
 Anton Erasmus

I'll add my $0.02 here; perhaps it is useful.

On the SPICE vs. no SPICE debate:  Designing modern analog ICs [1]
would be well-neigh impossible without SPICE due to modern circuit
size and complexity.  Other posters have already pointed this out.
Also, when designing an IC, you control nearly
all parameters of the components you use, and you have highly accurate
models of your transistors available.  Therefore, SPICE can do a good
job predicting circuit behavior.

Designing analog boards, on the other hand, is different.  Most of the
time, the models you have at your disposal are vendor macromodels,
which are not device-level models of the actual components you use.
Rather, they are idealizations which attempt to model the important
features of the device's performance in its operating region.
Vendors won't give you real device-level models of their components
because then you could reverse-engineer their circuits.   Therefore,
the SPICE models you use in board design are generally useful, but are
not totally accurate.  

Also, when designing boards, stray capacitances are not as well
understood or controlled as they are when designing ICs.  (Perhaps if
you purchase a $100K tool from one of the big EDA vendors you can
extract the strays from a PCB layout, but I have never seen that done
in real life.)   Therefore, the fabbed board will always act
differently from any SPICE simulation, particularly if your circuit is
sensitive to strays.

Therefore, for IC design, SPICE is indespensible.  For board
design SPICE provides good guidance, but isn't the last word in
predicting circuit performance.

As for the issue of convergence mentioned above:  My experience is
that if your SPICE simulation behaves strangely or doesn't converge,
it is likely that you have a fundamental problem with your circuit.
When a circuit doesn't converge, besides looking for floating nodes, I
always examine my circuit thoroughly looking for subtle mess-ups such
as two different current sources in series, or two different voltage
sources in parallel.  More often than not, I find that I have
committed some kind of error.

Stuart

[1]  Note bene:  I am not an IC designer, so others can speak with
more authority about this.  Nonetheless, my point is general enough to
not require detailed, expierential knowledge of IC design.

I think a great many (most ?) problems with SPICE and other simulation
programs in general are actually due to problems of the "Floating
Point" data type. AFAIK the total reason for being of the floating
point data type was to get a reasonable range and precision using as
little memory as possible. Today memory is not a problem anymore, and
one can use a fixed point number format with the desired range and
precision necessary for any simulation. A typical construct in many
simulations are:

(x0-x1)/k where x0 and x1 are almost equal. This causes problems in
floating point. If x0 an x1 are say 1.0 and 1.0001 then it is not a
problem. If it is 1000000000.0 and 1000000000.0001, then it bombs out.

I personally think that with todays systems, the use of floating point
should be banned, and in stead large fixed point numbers should be
used. The only disadvantage compared to floating point is that it uses
more memory. (And the little problem that almost no currently used
languages supports them as standard)

Regards
 Anton Erasmus

Hi Neil,

They have a demo version, free, which has these files. Try
<http://www.micro-cap.co.uk>. I haven't been there for quite a while, so
this may have changed.

Regards, BruceR

What I will say here is that a work college, and  myself on and off,
have been using LTSpice on some circuits very recently, like currently.
Sure, it converges most of the time when XSpice and Tanner spice, and
any others don't, however it still has problems on some circuits we have
been trying. This is to be compared with TISpice (internal Texas
Instruments spice). In a past life of 3+ years, it *never* failed to
converge, ever. It seemed to have 100 hundreds of algorithms to try
automatically. So, as far as robustness goes, I cant agree. Its good,
but not the best, imo. As for speed, I have never compared it to
TISpice.

Kevin Aylward
informationEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

Well, it was an LT guy that told me LT-SPICE was a PSPICE variant but I
don't have the need to trash people, (even anonymously) that some
people do so I will not mention who.  I certainly will yield to Mr.
Engelhardt on anything about LT SPICE because my LT friends say
he is THE MAN for this.  After all, we know Swanson would rather
work to death one man rather then hire a department.  Mike does the work
of two departments and I really respect that.  LTSPICE is his baby and
he has a right to be proud of all that work.

Can he tell me the relationship between SwitcherCAD and LTSPICE?
Are they the same thing?  And I was once told that LTSPICE does
not allow import of models from other vendors like ADI and National.
Sounds like bunk but it is a proprietary program after all.

BTW Mike, Bill Gross is a recently retired Vice-president and a former
IC designer so I will give him the complement tomorrow that you
consider him an apps guy like me. Swanson really must have you chained to a
workstation.

Now as to my SPICE experience-- well, Berkeley SPICE and Hollerith cards
yeah, a good bit of PSPICE, Intusoft ICAPs, at HP we had this thing
I think called DR Deautch or something that was supposed to converge
really well and my impression of all of them was they are crap.
But this was almost 10 years ago.

I will look for some of the oscillators and stuff that blow up or when
I get around to it publish a circuit I was using SPICE on a few months
ago and everyone can excoriate me just because I didn't know to set
abstol to something and put the .bs command in the deck.  Well duh,
whenever I have to slow down edges or loosen up accuracies to get a
convergence it just seems like a real good time to put down the
mouse and pick up a soldering iron.  Maybe I am just a scaredy-cat.

I would like to graduate past 7th grade and "you are full of sh1t"
comments so let us act like technical people and deal with facts.

Several years ago EDN magazine did a circuit and gave it to 6
SPICE vendors and Jim Williams at Linear Tech.  If I remember about half
of the programs failed to converge and the rest gave wrong results, sometimes
wildly wrong compared to Jim's real board.  Was there a memo I missed?
Have models and SPICE engines gotten that much better?

Can anyone really get any kind of mid-range SPICE to deal with non-linear
magnetics?  Does anyone trust it to design a complex flyback converter?

Are there really A to D models that give the representative data output of
the real-world signals?  Not just the math and correlations involving the
sampled-data theory but the real things going on in the analog and digital
sections? (All board-level models of course, not "real" transistor-level
models.)  Does LT offer models of those new fast converters they make?

OK, the SPICE behind National's WEBENCH uses a later version of the SPICE
engine then PSPICE.  I have heard one called level or stage two vs a three.
So what are the substantive differences?  Does PSPICE suck as much as everything
else Cadence seems to ruin?  Maybe I am complaining about my Model T when
everybody else in in a Prius.

I was at Arrowfest tonight where somebody said all SPICE does is solve a matrix.
That is what Berkeley SPICE is.  What everyone else is doing is writing
code to try and get the solution to converge when the math blows up.
I had dinner with a guy from PSPICE years ago and he said all of their work is
doing code like that, to keep things from blowing up. How comforting.  Is this
wrong?

When I was at HP we were designing automotive diagnostics.  I defy anyone to
make a good model of a spark plug gap since most attempts had real trouble
converging and then you realize the flame-front and pressure in the cylinder
affects the signal.  Did I miss that memo as well?

People, people, I am not being combative, I work in the on-line SPICE
group at National for crying out loud and really want to use it as much as
possible. Please don't jump on me like I am criticizing your religion or politics
or wife.  It just seems like every time I wade into another type of complex circuit
with SPICE I soon feel like I need a CS degree and a month of trial and
it is just so much easier to just build the thing.  Remember I am talking
board-level here, not something that you want to simulate to death since there
are 100k of masks at stake.  That is why I like our WEBENCH tool. We have
a whole department including a couple of apps guys like me to insure that we
can give good results when we run a simulation.  But we build the circuits with the
same exact components and make sure that the SPICE agrees with real-world
values so our customers don't have to. Is everybody out there designing
things with such similarity to their previous designs they know they can
trust the simulations?

Oh, if I have brought Kevin and Mike together then I guess there is
redemption in electronics after all.

Now to the important stuff, maybe I can get Pease to sign-up for Google
Groups but failing that I can at least post his reply to my lederhausen
question today:

=====================================================
*** Hello, Paul,

In reply to your comment.......

**** I do not  recall ever wearing or owning  Lederhosen, when I was in
college.  I recall
specifically that I did not. But I did wear shorts. In the winter. When
bicycling. In the
snow. When I went winter-mountaineering, up in New Hampshire, I wore
shorts plus
long-johns. Red long-johns.

*** I know  nothing about  silk-lined  Lederhosen, and I know nothing
about  the Lampoon's  parody.

*** Since I never had or wore Lederhosen, then I'm sure that some of the
ones I didn't wear were silk-lined, and
some of the ones I didn't wear were NOT silk-lined.  / rap
============================================================

Hey Mike; Tim Regen's birthday tomoroow, come over to Bldg T (The Tastey
Subs on Lawrence Expressway by Arques) and I'll buy you a beer.

Paul