( ESNUG 455 Item 5 ) -------------------------------------------- [06/29/06]
Subject: Apache NSPICE vs. Synopsys HSIM & HSPICE vs. Cadence UltraSim
OLDE GUARD -- Ask any engineer what his/her first EE SW experience was and
it'll invariably be some story about SPICE and SPICE decks. With something
like 10 major and minor flavors of SPICE on the market; each has been
optimized to its own special sub-niche. Apache NSPICE is no different.
"How does Apache NSPICE stack up against its sea of SPICE and near
SPICE rivals? What are its strengths? What are its weaknesses?"
We evaluated the NSPICE tool from Apache for signal integrity and high
speed I/O design. We benchmarked it against HSPICE for accuracy and
against HSIM for speed and memory usage. For us a smooth interface to
s-parameter is very important for our broadband design, so we must use
an s-parameter model for our substrate package.
In the beginning we had some problems w/ NSPICE due to timing accuracy.
The problem was due to the diode models they used in their tool. This
was solved. NSPICE is a very good tool. It is much faster than Avanti
HSPICE for the same accuracy (10X faster). NSPICE can handle very huge
netlists that HSPICE can not even read in. From speed point of view it
is comparable with HSIM. But HSIM still can not handle the s-parameter
interface. (HSIM claims that they can handle the s-parameter interface
but even for a very trivial circuit we had problems. I have an e-mail
from HSPICE admitting that they have a bug for their s-parameter
interface. They are working on that.)
We would rather have HSIM be able to handle s-parameter interface as we
have tons of HSIM licenses and we could get a better deal from them and
people already use it. But for SSN and SSO, DCAP and I/O design we must
use NSPICE for the time being.
- Cyrus Afghahi of Broadcom Corp.
NSPICE is not a general SPICE simulator. It can be used only on Std
cells and small custom blocks. I would still use fast SPICE simulators
like HSIM or NanoSim on the larger blocks to characterize the power.
- [ An Anon Engineer ]
We don't use NSPICE.
- [ An Anon Engineer ]
Better/Faster and stable than HSPICE. Its strength is its better
convergence, better use of S-param models. Its weakness is that
it needs a NSPICE encription (available for free) to get models.
- Syed Huq of Cisco Systems
NSPICE can handle large netlist, which HSPICE cannot. I do not know
other SPICE.
- [ An Anon Engineer ]
Didn't compare to anything else but seems to do the job.
- [ An Anon Engineer ]
I think NSPICE is geared for analyzing large RLC circuit, such as power
grid with active devices. Do not forget that this is one of the main
strengths of the founders of Apache (Anagram, ADM and the continued work
in StarSim at Avanti.)
- [ An Anon Engineer ]
NSPICE accuracy is very good from what we have seen in house.
- [ An Anon Engineer ]
Apache ran some SPICE analysis to give us jitter numbers for some
clock nets. We were told this was PsiWinder but we didn't know
that it was running NSPICE rather than some other SPICE engine.
- [ An Anon Engineer ]
No idea. It does the job for me when characterising APL for instance.
I'm not a SPICE guru anyway.
- [ An Anon Engineer ]
NSPICE works very well for power grid specific jobs. Without this the
run time performance may not be feasible.
- [ An Anon Engineer ]
The accuracy portion of our eval involved DVD waveform comparison to
HSIM (utilizing virtually identical VCD/vector-stimuli). Apache's
RedHawk analysis employed NSPICE-characterized current waveforms,
resulting in current and instance-voltage waveforms which compared
quite favorably with HSPICE's waveform results.
- [ An Anon Engineer ]
Index
Next->Item
|
|
