( SNUG 10 Item 10 ) --------------------------------------------- [04/15/10]
Subject: What 238 users think of IC Compiler
DOUBLE TROUBLE: From the many user comments, Synopsys IC Compiler is a
nightmare of 10,000 switches to run plus a sea of bugs -- yet SNPS makes a
mint off of ICC and (from this survey) it has a ballpark 52% marketshare
vs. Cadence 24% vs. Magma 14% vs. Atoptech 3% vs. Olympus 2%???? WTF???
Is the current state of commercial P&R tools really THAT bad?
On the plus side, the user data reports that if your design does survive
the 10,000 switches and the many many bugs of ICC, it'll correlate
excellently with PrimeTime and Calibre.
With regards to IC Compiler, our group's experience is (choose
ALL that applies):
1a- Love it! ICC is fairly painless and easy to use.
: ########### 11%
1b- ICC has 10,000 switches; you need a PhD to run it.
: ############################### 31%
2a- It's robust code; not many bugs in ICC
: ###################### 22%
2b- It's a nightmare; ICC is a hornet's nest of bugs!
: ##################################### 37%
3a1- ICC correlates WELL with PrimeTime
: ############################# 29%
3a2- ICC correlates POORLY with PrimeTime
: ## 2%
3b1- ICC correlates WELL to Calibre
: ############### 15%
3b2- ICC correlates POORLY to Calibre
: 0%
4- We use (name tools & co's) P&R instead.
Cadence Encounter : ######################## 24%
Magma Talus & Blast : ############## 14%
Atoptech Aprisa : ### 3%
Mentor Olympus SoC : ## 2%
Comments?
---- ---- ---- ---- ---- ---- ----
PrimeTime correlation is achieveable with ICC, but it's tedious to dial
in because of the 10,000 switches.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC had some weird results but we got around them. 10,000 switches.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Too many bugs in ICC. Of course Synopsys rakes in a ton of dough on
service and support to its top customers. I know. I saw the bill.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC is a buggy tool that struggles to get timing closed.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC (buggy!!)
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Fix ICC UPF support, fix ICC PNS, fix ICC CTS
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC is MUCH more robust than Astro. Still buggy but less than expected.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC has been cleaned up quite a bit over past 3 years. It has become a
decent tool. Haven't evaluated other tools in recent past to compare,
though.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC is a pain, but from our experience its competitors are similar. In
this arena the products are always beta at best (and very $$$)
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We have a voodoo priest for ICC P&R.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We are considering switching to different tools away from ICC.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC works, but (of course) you need to know what to do, which switch to
use when, etc. But even though we are a fairly small site, we receive
pretty good support on this from SNPS.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC is a pretty good tool. I haven't had too many problems with it.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Every P&R tool has its weaknesses and needs specific AE services.
Better ICC supportiveness would be helpful.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Maybe not a hornet's nest of ICC bugs, but we've certainly run into quite
a few problems. Seems like every time we turned around, our CAD group
was using an SP release for some part of the flow or for some blocks.
There are a *lot* of ICC options available. Our CAD group puts wrappers
around most of the stuff, but this often leads to sub-optimal results.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Modify ICC to be able to deal better with non-timing elements and with
preliminary P&R models without timing models -- this is a problem that
forces us to make dummy models for certain blocks/cells or to "change"
them to pure power macros.
We have had a physical-only tester block which really did not need any
timing (fuses and some other special devices with pad pins), but we had
to label all the pad pins as powers or grounds where they are really
like that but tester signals that will only draw a certain current until
the fuses or other test structures blow or go into steady state. That
required making 63 different "powers"; there really was only one ground.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Make ICC more intuitive and add a point-to-point router
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Throw out UPF and switch to CPF on day 1.
Fire the PNS (power network synthesis) team on day 2.
Completely re-design clock building in ICC and integrate it into
first placement and optimization stage.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Make ICC behavior like a multi-task tool instead of like a point tool.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Rebuild the Milkyway/DB format from scratch. ICC takes a 20 year old
logical/timing database and munges it with a 20 year old physical
database and everyone expects it to be seamless.
They need to bite the bullet, ditch all backwards compatibility, and
build a new database from the ground up. There is too much baggage in
Synopsys tools due to a million incremental changes in their databases
over many years.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We're moving from Astro to IC Compiler.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We're still using Astro & learning how to use ICC.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Astro, then Encounter
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We use Encounter from Cadence.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We use Cadence Encounter. Some groups here beginning to move to ICC.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
My group was fortunate enough to be "allowed" to use Cadence (Encounter)
for floorplanning since ICC-DP was so absolutely pathetic and broken when
we needed it. However, I'm sure that access won't last...
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We use (Magma) P&R instead.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We use Magma P&R instead. We'll be forced to use the whatever-its-called
Cadence P&R tool, in the next few weeks.
I really liked working with Magma tools: Slightly buggy, but with great
results, ease-of-use, extensive user scalability and controls and with
this independent/underground/arthouse touch. Cadence digital P&R sucks.
To say the least.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We finished a large tapeout in 65 nm process. We had to contact Synopsys
for just one or two feature related issues. No tool crashing, no periodic
required updates, no license changes, etc. True, it has that many
switches. But default, standard switch list is sufficient for majority of
designs. We had to just turn on a few special switches for our design.
Learning curve was not an issue here at all.
We use both Cadence and Synopsys for P&R. Different tool set works well
with different design styles. Neither excel each other significantly.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
ICC is not as good as the others, especially at top level.
We use Cadence Encounter and Mentor Olympus P&R instead.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Magma Talus seems to be improving a lot, hearing movement away from CDN.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We used Magma some time ago. We switched from SNPS to Magma and then
back to SNPS. Pretty painful process. :-(
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We use Magma BlastX and Cadence VDIO.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Personally I think that ICC is a great tool, but I'm not too deep
into it. I do hear colleagues complain, which for sure is because
of bugs, but often also about the complex configuration, etc.
We've been doing Magma for several years, but suffered from their
"QA" and many many unsynchronized fast track releases. The basic
software concept and the data model were/are good though. Part of
the decision to go with ICC for sure is related to new contract
opportunities we had.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We use Atoptech.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
Best-in-class tools is company philosophy -- Atoptech.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We're using an internal Intel tool.
- [ An Anon EDA User ]
---- ---- ---- ---- ---- ---- ----
We don't use P&R tools anymore. Too expensive.
- [ An Anon EDA User ]
Sign up for ESNUGs! Fun!
Index
Next->Item
|
|
