( ESNUG 451 Item 2 ) -------------------------------------------- [02/08/06]
Subject: Readers critique the "new" and "improved" Hercules announcement
> During his keynote address at Boston SNUG, Aart talked up how "new" and
> "improved" his Hercules now was. Although I couldn't get Aart on video
> about it, I did get a Herc support droid and a Herc marketing droid to
> talk in detail about Aart's claim here.
>
> http://www.deepchip.com/demos/hercules.fhtml
>
> I'd love to hear what users (and even Aart's EDA rivals) think about
> this annoucement.
>
> The Synopsys claim in detail is in first 5 minutes of the interview; the
> remaining 19 minutes of the interview is me probing the specific details.
> Tell me what you think.
>
> - John Cooley
> ESNUG/DeepChip.com Holliston, MA
From: [ 99 Bottles of Beer on the Wall ]
Hey John,
Every time I hear the words "new" and "improved", I can't help thinking,
"what kind of crap were they serving me in the first place?" Is this a
confession of sorts from Aart regarding what he thought of the original
Hercules? :) Anonymous.
- [ 99 Bottles of Beer on the Wall ]
---- ---- ---- ---- ---- ---- ----
From: Mo Casas <modesto.casas=user domain=transeda spot calm>
Hi John,
Remember that many of the "droids" have families and when they show their
kids that they got on your newsletter, they may have to explain why you
call them droids. (I woke up sensitive in Paris today)
- Mo Casas
TransEDA
---- ---- ---- ---- ---- ---- ----
From: [ 98 Bottles of Beer on the Wall ]
Hi, John,
Too much fluff and no essence of any technical details. Most of their
claims are stories and cannot be verified. A few things that stink:
1. Short finder -- a nice feature that Calibre had in 1999 and
Assura in 2000. I know because I worked with them to implement it.
2. Foundry support -- each vendor who wants TSMC to put a deck for a
technology on their web site is creating the decks and provide
TSMC for QA. It is not TSMC who writes them for the users, is the
vendor who does the job. Cadence, Mentor, Synopsys and probably
now Magma are the ones who write them.
3. Sign off tool at TSMC -- at DAC 2005 Mentor said the same thing
about Calibre, the tool of choice. When I asked at DAC the TSMC
representative: What about Hercules, Assura, and Hercules? The
answer was "We never said that we do not support others..." Go to
TSMC web site and all the tools "are" the tool of choice.
4. Performance in distributed networks -- maybe you can bring the
developers/architects from Magma, Cadence, Synopsys and Mentor to
a round table and let them try to sell the concept of each of the
tools. If TI is choosing Hercules is probably for the "good enough"
versus "price" not BEST IN CLASS tool. In today's market this is
luxury that not too many can afford. When I asked at DAC 2005 to
see a demo of Hercules I found out that it was not even available,
no demo at DAC? Can you believe such a "successful tool" without
a marketing splash and a super demo? They did not bring a Hercules
demo at DAC for the last 3 years... Draw your own conclusions...
5. I think that you got 23 minutes of marketing fluff and 1.5 minutes
of technical details.
If you want you can use my comments but make it anonymous.
- [ 98 Bottles of Beer on the Wall ]
---- ---- ---- ---- ---- ---- ----
From: Mike O'Reilly <mjoreilly=user domain=comcast.net>
Hey John,
Interesting marketing spin. One benchmark is not enough... nor is two or
three. Get some real customer feedback. Unfiltered would be best... and
oh by the way... just because "They" have the "fastest" today doesn't mean
squat tomorrow. What's with speaking about 45 nm and then backing off to
90 nm decks?
- Mike O'Reilly
Cadence Marketing
---- ---- ---- ---- ---- ---- ----
From: [ 97 Bottles of Beer on the Wall ]
Hi, John,
I spent 2 years going through comparisons of Hercules and Calibre and these
two clowns have really blown my weird BS-o-Meter off the scale. To begin
with, Synopsys could not implement more than 30% of the DRC rules for 90 nm
and 65 nm, even after 2 years of opportunity to do it.
Mentor came in and implemented the full 90 nm and 65 nm rule set within 40
days. Maintaining updates was a snap. The biggest problem with Hercules
was the poly and base-layer implementation. In fact, we used the Calibre
deck to find places where rules in our design were missed. On the easy
rules implemented in Hercules, Calibre was typically 2.5 times faster. This
was on custom block, data path block, and full-chip comparisons. I ran
against every version of Hercules in house. Synopsys had every opportunity
to produce a so-called "new" Hercules and it did not materialize.
So the guys in the video talked up short finding. I don't see how their
assertions about short-finding is any more improved over Calibre LVS. But
one place that Hercules was clearly deficient in was in setting up the
cross-reference data base for StarRC extraction. Typically, 45% to 50% of
the nets extracted could not be cross-referenced through Hercules LVS.
With Calibre LVS, we were able to produce DSPF that was completely cross-
referenced back to the original schematic.
We gave 8 months of opportunity to Synopsys to fix the problem and they
could not. Calibre LVS was up and fully supportive of Star-RC extraction
within a few weeks which included writing the LVS deck for our devices.
I'm not working at my old company now, so my data is about 2 months old.
Synopsys was big on hype and promises for 2 years. I doubt a miracle
happened in the last 2 months.
Please keep my comments anonymous on ESNUG
- [ 97 Bottles of Beer on the Wall ]
---- ---- ---- ---- ---- ---- ----
From: Ram Ramaswamy <ram.ramaswamy=user domain=baesystems spot calm>
Hi, John,
Cadence does have an LVS tool called Assura (an overhauled version of their
old tool Diva). I cannot believe Synopsys would not be aware of it. If
companies are using Cadence in their design flow, it would be hard to
imagine that Cadence would offer only DRC. The entire tape-out depends upon
successful run of DRC & LVS. I am sure designers have taped out using the
full Cadence flow.
BTW, as a Hercules user, I do agree it is a powerful tool & has a nice
integration with their extraction tool (Star-RCXT). There are always some
minor quirks like with any other tool.
- Ram Ramaswamy
BAE Systems
---- ---- ---- ---- ---- ---- ----
From: [ 96 Bottles of Beer on the Wall ]
John, please keep me anonymous.
I work with SNPS, MENT and Magma for sign-off DRC and LVS tools.
I found the SNPS Herc interview to be amusing to say the least.
First, SNPS has been the hardest of all Physical Verification EDA vendors to
work with (Herc team). They effectively have stalled on any/all attempts
for us to do a head to head comparison to Calibre using Hercules. They
refuse to support development of the Herc rundecks needed to do this. They
just want us to trust them.
I do like the Herc data-distributed processing concept, though I have doubts
about the implementation. From conversations I have had with them on this
it appears they are still limited to distributing the processing on a layer
basis.
I do believe that Mentor is the furthest behind for developing a true
distributed capability. They were great at multi-threading but have fallen
behind in dist processing with MT Flex. We still cannot get MT Flex to
significantly improve runtimes and it is also a hassle to get set up and
working.
The best performance and ease of use I have seen anywhere is the Mojave DRC.
It is a true dist processing capability that partitions data vertically into
small design pieces. The runtimes are at least 6x better than Calibre MT
Flex using the same number of CPUs. Magma is soon to be out with their LVS
solution. We have looked at an early version of the tool and features and it
looks excellent. Much better debugging features than SNPS or Calibre by a
long shot.
- [ 96 Bottles of Beer on the Wall ]
---- ---- ---- ---- ---- ---- ----
From: Jared Bytheway <jbytheway=user domain=cirque spot calm>
Hi, John,
It's too bad that you don't have more experience with Magma, Cadence,
Mojave and Calibre so that you could have verified or refuted all of
their "not to our knowledge" comments, which were many.
(I don't have experience with any of them so I can't comment)
- Jared Bytheway
Cirque Salt Lake City, UT
---- ---- ---- ---- ---- ---- ----
From: [ 95 Bottles of Beer on the Wall ]
Hi John,
Please keep me anonymous - just in case...
I was not able to see Aart's announcement because it requires something
called Windows, but I have had the opportunity to test the new Hercules
and compare it to Calibre. I'm not convinced (yet). OK, I'm a bit biased
because I know Calibre, but Hercules is still far from the "Uh, I wish
I had a Hercules license" shape.
Here's a brief overview of my impressions.
Hercules DRC has indeed become very fast compared to Calibre and compared
to Hercules in 2001. Hercules LVS seems still to be quite slow, but I have
not had time to check it thoroughly yet.
At first I was impressed by the speed of Hercules, but after looking into
the differences(!) in the results between Hercules and Calibre I became
a bit worried that the speed had a price and I think I might be right.
The default way of measuring distances is different between the two tools
and if that's a change for Hercules I don't know. Calibre measures the
Eucledian distance as default, whereas Hercules only measures 'opposite'
i.e. they don't measure what lies ouside the endpoints of an edge. See:
This is actually quite significant wrt speed. I tried to change Hercules
rules for the 8 metal layers I had, and it gave an increase in runtime of
5%. I have not had the time to change all distance measuring rules to use
'real' distance checking, but the metal checks were only about 5% of the
possible distance checks that could be modified. All in all I would expect
Hercules to lose it's advantage if rules were written to check exactly the
same thing.
This brings me to my main worry -- distance checking should always be
Eucledian unless special rules call for simpler checking!!
This figure has width violations:
Shown here:
which the default setting of Hercules will not flag.
- [ 95 Bottles of Beer on the Wall ]
---- ---- ---- ---- ---- ---- ----
From: John Ferguson <john_ferguson=user domain=mentor spot calm>
Hi John,
I watched the Hercules video that you posted with great interest. There
were a number of bold claims made, many of which we in the Calibre team
disagree with. But, rather than address them, I'd prefer to see the
response from your readers.
There were, however, two specific statements that I think need some
further clarification, involving Hercules's relationship with TSMC.
In the video, Synopsys claims that Synopsys engineers conduct benchmarks
against Calibre on-site at TSMC. This is not true. There are very
stringent contract terms in place between TSMC and Mentor, as I'm sure
they have with all their vendor/partners, that clearly indicate what
their Calibre licenses can be used for. This clearly does not allow the
use of Calibre by our competitors for benchmarking. I have discussed
this claim with TSMC, and they have verified that they have not been
involed in any of the claimed benchmark activities, nor have they
enabled Synopsys to conduct such tests on their own.
In addition, Synopsys claims that Hercules is run on every design coming
into TSMC for 0.18 and 0.15 micron, and that Hercules is the internally
used sign-off tool for TSMC. This is not true. Both Calibre and
Hercules are supported at those nodes, but it is Calibre that is
considered the sign-off standard. This means that Hercules results are
compared against Calibre for validation. Not only is this true at 0.18
and 0.15, but continues through 65 nm.
Neither Calibre, nor Hercules is run at TSMC on customer data for
physical verification. Instead, the job of verifying that designs are
DRC and LVS clean is entirely on the user (except in rare cases where
TSMC runs Calibre as a service to their users). As the internal
sign-off tool within TSMC, Calibre rules are written and validated on
test structures, but are also used to validate all of TSMC's internally
generated IP, run to re-verify designs that have unexpected yields, and
used in the characterization of new process nodes, Mentor customers can
feel confident in the accuracy of the results. Also, because Calibre is
used in the process of qualifying new process nodes, Calibre rules are
always available first for new processes.
Keep in mind, of course, that I'm just a Mentor marketing droid. You
really should get TSMC's side of the story on this, John.
- John Ferguson
Mentor Calibre DRC Marketing Wilsonville, OR
---- ---- ---- ---- ---- ---- ----
From: [ 94 Bottles of Beer on the Wall ]
Greetings John,
I would greatly appreciate if you can keep this message anonymous, as I'd
probably be fired if word got back to my employer that I sent this.
LOVED the video with all the Hercules claims, in fact I'm gonna watch it
again, just to make sure I didn't miss anything. Now for your readers who
haven't seen it yet, I'd like to offer a viewers guide which can help them
determine, by observing body language, if a speaker is being truthful.
I think it'll provide some enjoyment, too, 'cause there's a couple points
in the video where these fellers look as nervous as a pair of long-tailed
cats in a room full of rocking chairs...
The following will help your readers to decide if these gentlemen are
indeed above board on their claims:
http://www.blifaloo.com/info/lies.php
Thanks John. I like these videos you're producing, because you CAN see
peoples' body language. There just ain't any nuance in a datasheet or
a press release...
- [ 94 Bottles of Beer on the Wall ]
Index
Next->Item
|
|
