( SNUG 04 Item 17 ) ---------------------------------------------- [08/11/04]
Subject: Nassda HSIM, Avanti HSPICE, NanoSim, Mentor Eldo, Cadence Ultrasim
THE SPICE OF LIFE: The two big players in the SPICE and fast-SPICE market
are Nassda HSIM and Avanti HSPICE/Synopsys NanoSim.
Dataquest FY 2002 IC SPICE Market (in $ Millions)
Nassda HSIM #################### $29.3 (45%)
Avanti HSPICE ############### $22.8 (35%)
Silvaco ######## $11.7 (18%)
others # $1.3 (2%)
Synopsys NanoSim did not have a serious presence 2 years ago (hence it's
not in these DQ stats.) And the two new players (also not in the stats)
in this niche are Apache NSPICE and Cadence Ultrasim. Even though the
Mentor Eldo tool is older than all of them, my guess is that Eldo must
be in the "others 2%" number. Regardless, Nassda HSIM seems to be the
most popular with the users in this category.
11.) Do you use NanoSim, Nassda HSIM, Avanti HSPICE, Apache NSPICE,
Cadence Spectre, or Mentor Eldo? How do they rank against each
other in your eyes?
Hey John you are mixing up different kinds of tools.
Synopsys NanoSim, Nassda HSIM and Cadence Ultrasim (you forgot to
mention the last one) are "fast-SPICE" simulators for chip/block level
verification and maybe for digital custom design.
On the contrary are Avanti HSPICE, Cadence Spectre and Mentor Eldo
accurate SPICE simulators for real analog design.
We use Cadence Spectre for analog design. (I would rank HSPICE 2nd and
Eldo 3rd.)
We have been using Nassda HSIM for block verification of high speed
digital circuits and some analog stuff, too. As we are not completely
satisfied with HSIM, we are currently evaluating both Ultrasim and
NanoSim, in the hope to find something better.
Ultrasim seems to work very well with complex analog circuits involving
not only MOSFET but also Verilog A models. It's a real nightmare for
block level digital circuits simulation. Ultrasim's full of bugs and
does not still support a lot of features which all other competitors do.
I think it's not really ready for release yet.
NanoSim is definitely much more comfortable. It has finally got a lot
of useful features for digital circuits simulation. However NanoSim has
its troubles when it comes to simulating complex analog circuit with
additional Verilog A models. NanoSim takes an eternity to complete
such simulations.
After all those trouble with NanoSim, and above all with Ultrasim, I
must say that Nassda HSIM is not so bad....
In this area there is definitely lot to do yet. The current tools are
not mature. At the present HSIM and NanoSim are by far better than
Ultrasim. But things could change.
- Marcello Vena of Xignal Technologies AG
We tested Nassda HSIM on 3 mixed-signal chips. In all 3 projects, we
successfully performed a full-chip transistor-level simulation. It
only took about 2 hours for each engineer to run their evaluation;
from simulation start through verification of the main items we needed.
The 3 chips were:
1.) a 12V, 400 gates (digital) + 2,500 transistors (analog) chip.
We fully simulate this chip in only 2 hours with HSIM. We could
not even do full-chip simulation using standard SPICE due to
convergence issues and crashes.
2.) a 2.7V to 5.5V, 7,000 gates (digital) + 5,000 transistors (analog)
chip. We did a full-chip simulation of this chip in only 4 days
with HSIM, whereas standard SPICE took us 21 days. Our initial
intent was to also perform a post-layout simulation on the chip
with HSIM, to see if HSIM could indicate matching problems in the
layout. Due to independent issues with HSIM (labeling of layout),
we didn't succeed during our limited evaluation timeframe.
Nevertheless, HSIM's netlist simplification was clearly apparent:
initially the netlist had 200,000 lines, and after HSIM RC
reduction there were only 5,000 lines left.
3.) a 2.7V to 5.5V, 12,000 gates (digital) + 500 transistors (analog)
chip. Full chip simulation with HSIM only took 20 minutes.
Full-chip simulation with standard tools was not possible for this
chip, for the same reasons as the first chip. HSIM also easily
detected floating nodes.
We have found that errors are very explicit with HSIM while this is not
the case with other fast SPICE tools. Another good point we see in HSIM
is that it supports Verilog-A, which makes it useful for modeling.
My only warning regarding HSIM is that the results obtained were very
dependent on the parameters we used for the simulation. A very fast
simulation can give completely wrong results! Engineers need to have a
good idea of what the expected result should be. They need to fine tune
these parameters for speed, analog level of simulation for analog part
and digital part, resolution, etc. to get maximum accuracy in a
minimum of time.
Based on the simulations we did in our evaluation, we found:
- HSIM could find most of our design bugs.
- HSIM's simulation accuracy is sufficient for our designs.
For us, HSIM is necessary for any big netlists (2,000+ transistors) and
for full-chip simulations at transistor-level.
- Patrick Besseux of Microchip Technology
We use Nassda HSIM and Spectre. HSIM is very useful in speeding up
simulation of large custom designs.
- Joe Dao of Aeluros, Inc.
Avanti HSPICE - started to be too slow
Cadence Spectre - started to gain value...
Using both
- [ An Anon Engineer ]
Nassda HSIM is far ahead of NanoSim in term of usability and "ease
to success". Complex simulation with NanoSim is a big challenge!
Mentor Eldo is definitely the best. Avanti HSPICE the worst and
Spectre has the best DC convergence, but a very poor usability.
Cadence is late but catching up here.
- Jean-Paul Morin of STMicroelectronics
We use Spectre and Avanti HSPICE all the time. I've never found an
appreciable difference.
- Gord Allan of Carleton University (Canada)
We use Avanti HSPICE but are also trying Nassda HSIM. Seems to be
good from user feedback.
- Abraham Si of Maxim Integrated Products
We use Avanti HSPICE. We have heard Nassda is better but the cost did
not justify it.
- [ An Anon Engineer ]
We use Avanti HSPICE.
- Andrew Bell of PMC-Sierra, Inc.
Tried Spectre some years ago - yuck! Currently use NanoSim, HSPICE,
and internal SPICE. Also use various "custom" SPICE engines with
characterization tools such as "dynaSPICE" and "smartSPICE". Oh yeah,
those are the same thing, aren't they? Generally, NanoSim is great
for high-speed simulations. Avanti HSPICE and internal SPICE both
great for detailed simulations, but HSPICE is pretty darned pricey
compared to internal so we continue using internal at the cost of one
full-time engineer to maintain/update/develop.
- [ An Anon Engineer ]
I think our analog guys used PSPICE in the past, now they are moving
to Cadence Spectre.
- Juan Carlos Diaz of Agere
I have several years of experience in using Nassda HSIM for memory
design. It's an excellent tool for memory designers. I use if for
functional verification, timing characterization, critical path
analysis, race condition checking, power simulation, leakage
analysis, etc. The best part of HSIM is its hierarchical simulation
capability. It makes it possible to simulate the full memory directly
from the hierarchical netlists without building a separate model for
critical path and timing simulation, like what we do for HSPICE
simulation. We have parasitic resistance built into the hierarchical
schematics and found a way to short these resistors in LVS. With HSIM
we only need one set of schematics/netlists to be used for everything
(LVS, functional verification, power simulation, characterization,
etc.) This significantly simplified our design flow.
I've found the accuracy of timing numbers obtained from HSIM is usually
within 5% of difference compared with HSPICE. Its power numbers have
a larger discrepancy compared to other tools, especially the peak
current numbers. But I don't know which one is closer to correct
answers. My leakage simulation show pretty close results as HSPICE.
My only concern is that the memory bitline separation accuracy is
sometimes not good enough. It could report BL separation more than
10 mV different from HSPICE (which is a big deal to us.) But I didn't
dig into this to find out the reason. HSIM has a lot of accuracy
control parameters. Maybe I can get better results by adjusting
those parameters.
- Dechang Sun of LSI Logic
Don't personally use any of these, but other people at our company use
Avanti HSPICE and Mentor Eldo.
- [ An Anon Engineer ]
We use Mentor Eldo.
- Massimo Scipioni of STmicroelectronics
Avanti HSPICE is obviously the gold standard, but Spectre is as good
but faster and with more functionality, particularly Spectre RF. Eldo
just doesn't seem to be as well supported by the foundries. NanoSim
is a good fast SPICE with lots of power reporting options. HSIM may be
faster (I don't have head to head comparisons), but it has an interface
that is much more like SPICE and thus is much easier to use. Some of
the power reporting features are a little more difficult to use than
NanoSim, though. We currently use Spectre and will probably get Nassda
HSIM for our fast-SPICE.
- Brett Warneke of Dust Networks
We use Avanti HSPICE, Cadence Spectre and Synopsys NanoSim. HSPICE and
Spectre are equivalent. NanoSim is for larger simulations. Previous
comparison (2 years ago) shows Nassda HSIM to be superior to NanoSim
in performance and capacity.
- [ An Anon Engineer ]
We use Silvaco's SmSPICE for transient and AC analysis. Avanti HSPICE
is almost as good.
- Louis Morales of Innotech Systems
We use Eldo. At the time that decision was made they had the only
really usable RF solution. I haven't checked current status, and
I don't care to switch.
- [ An Anon Engineer ]
Analog Artist is the transistor level simulator for us. HSPICE is old
but good. NanoSim is fast and allow large circuits, I think it is the
best if accuracy is not your number one goal.
- Haiming Jin of Intel
We use Avanti HSPICE. We got a package deal with other tools and it
satisfied our requirement.
- Santhosh Pillai of Parama Networks
Both Nassda HSIM and Synopsys NanoSim appear to have comparable
performance and accuracy, maybe a slight advantage with NanoSim.
We don't use the other fast SPICE tools.
- [ An Anon Engineer ]
We use NanoSim, Avanti HSPICE and Spectre. Their rank is
1st HSPICE
2nd NanoSim
3rd Spectre
We are currently trying to evaluate Nassda HSIM.
- Marco Oliveira of Chipidea Microelectronica
I use Cadence Spectre. The complicated thing about Spectre is all the
programming things you can do with it (Ocean, Skill, etc.). It takes
some time until you really take advantage of all its features.
Nassda HSIM is much faster. I used it for some simulations, but
especially parasitic simulations were not 100% reliable.
- Klaus Vongehr of Philips Semiconductors
Avanti HSPICE has made some solid improvements in the last year, but
Spectre is still faster. They brought out there latest HSPICE RF,
but as an early tool it is lacking integration in the Cosmos
environment, plus not as sophisticated and easy to use as SpectreRF
Spectre is still better than HSPICE for tough DC and transient
convergence.
- [ An Anon Engineer ]
Synopsys NanoSim is very similar to Nassda HSIM. But accuracy and
usability of NanoSim is far below than HSIM. I thought NanoSim was
developed at first place, though. HSIM is much better than NanoSim.
It is bad news for the industry that HSIM might be eliminated,
nevertheless of a law suit.
- Myung Kong of National Semiconductor
|
|
