Validate Spice models before use
By Ron Mancini -- EDN, March 31, 2005
I was discussing the pros and cons of using Spice as an analytical tool with some analog-design engineers at a recent conference. The consensus was that vendors have improved their implementation of Spice such that the program is now usable for any engineer who can check out the dc conditions. The biggest problems that Spice analysis has faced and still incurs are the models. Spice vendors don't verify the performance of IC models; they may ensure that a series of models functions with their program, but they never check accuracy.
Semiconductor vendors are notorious for handing out Spice models that don't work or that fail to represent the circuit. My Jan 16, 1997, EDN article, "Don't count on your Spice models until you validate them ," provides an in-depth analysis of the Spice models available at that time. That article shows that the performance of some Spice models has no relationship to corresponding IC performance. When I researched that article, I found Spice models that had twice the IC bandwidth, some with unrealizable gain/frequency curves and some with more problems than solutions. The semiconductor companies that refused to take responsibility for the models they generated exacerbated the poor model situation. Their response to a request for help with the models sounded like the whine of a cheap 10-kW generator as they attempted to push the cause of the poor performance back to the Spice vendor. This scenario was a lose-lose situation, so I recommended using Spice as a last resort.
|
Today, most semiconductor companies take models seriously, and, as a group, today's Spice models are head and shoulders above those available in 1997. Furthermore, semiconductor companies are starting to hand out limited-capability copies of Spice programs that include Spice models. This handout seems like a great freebie until you search the Web and find out that at least 10 free versions of Spice are available. The semiconductor companies are handing out the freebies to get designers hooked on designing with their ICs because the Spice programs and models work together the first time, all the time. Spice-model/program compatibility is critical because for the first time, designers have one source for answers to questions.
Spice designs ICs with an IC designer's help, and not every IC works the first time. Thus, engineers must assume that the modeling process has inherent errors that they can never eliminate. IC designers use workstations and powerful computers to simulate IC designs, and the average analog designer lacks the time or resources to complete an analysis with these design tools. So, the analog designer must use a simulated model or macromodel. The macromodels are less accurate than those that semiconductor designers use, which results in an accuracy loss in the development of the models that the semiconductor companies distribute. I evaluated new models from several vendors, and, although a definite improvement exists in the models, a lot of errors also exist. All Spice users must evaluate the models they use to determine what error components they are willing to live with, which will be the subject of my next column.
| Author Information |
 Ron Mancini is a staff scientist at Texas Instruments. You can reach him at 1-352-569-9401, rmancini@ti.com.
|
-
I agree with Ron.
In our work some troubles arised from model issues and
prototypes work differently as simulations and sometimes prototypes don't work at all.
Our guidelines for evaluating spice software gave as result that LTspice is powerful (maybe better than ather IC vendors spice programs) but it takes too much time to learn comparing to commercial competitors. We made a real test during training sessions.
Anyway, almost any commercial spice is good enough for our designing purposes when models from MANY semiconductor companies are validated!
As Ron claims, model validation is the challenge not spice comparison and free spice programs seem out of the game, since they work (...if they work) good only with models provided from the specific company.
Gerard
Kadme, Norway
Gerard Donberg - 2005-9-9 03:10:00 PDT -
Ron,
With regard to your comments about free SPICE
handouts from IC vendors, there is one free SPICE
program from an IC vender that is an exception to
the advice you promote in your article. That is
SwitcherCAD III/LTspice from Linear Technology
Corporation. It is not a limited SPICE program
but in fact an unlimited, full-featured, high
performance SPICE program that outperforms the
standard commercial SPICEs licensed out of the
EDA industry. A designer can run his 3rd party
generic SPICE models in LTspice and enjoy it's
improved performance. And unlike generic SPICE
programs, it has extensions to handle SMPS
simulation. If it had been possible to write
SMPS models for generic SPICE programs, that
would have been the natural approach. But
generic SPICE programs are not powerful enough
for the job, so a custom simulator is required.
It is not an attempt to lock people into one
simulator, but provide the means to do the
simulation whatsoever.
In 30 years of writing simulators, I've learned
that the best simulators are not for sale; i.e.,
license; but developed by the concerns that need
them. LTspice is no exception to this rule.
You can not buy a better SPICE program.
And frankly, I've been under the impression that
LTspice is the envy of TI because I was contacted
by TI to write them a clone of LTspice for TI(I
am the author of LTspice). From that, I might
guess that your comment's agenda is a disingenuous
denigration of LTspice because you can't compete
with it.
Regards,
--Mike
Mike Engelhardt - 2005-31-3 10:37:00 PST
