( DAC 02 Item 18 ) ----------------------------------------------- [ 9/10/02 ]
Subject: Apollo/Astro, Plato NanoRoute, Silicon Ensemble, Pulsic, Columbia
GET RICH QUICK: A few years ago, the big get rich quick scheme for EDA
start-ups was to make a Verilog simulator because Synopsys, Mentor, Avanti,
and ViewLogic were trying to compete against Cadence's Verilog-XL. It
appears that in recent times the EDA acquisition market has been routers of
one flavor or another. Synopsys bought Everest some time ago and this
year's big Christmas present for SNPS has been acquiring Avanti. Cadence's
Silicon Ensemble (and Warp Route) have been running out of steam, so they
bought Plato's Nanoroute this year. Now both CDN and SNPS have been trying
to figure out exactly what they've bought, how it overlaps with their R&D,
what to keep?, what to throw away?
"Pulsic has some amazing shape-based routing, but the capacity is too
small to suit today's ever increasing sizes."
- [ An Anon Engineer ]
"Pulsic is an interesting one. Their technology core looks good and the
self running demos were impressive. Not sure if there is a business
model for the Pulsic, though. I went to the Pulsic booth 3 times by
myself and with others, and got three different pitches for the same
product. It's an analog router (that does not do spatial topology
import - so I'm not sure the application). It is module builder and
standard high capacity APR tool - timing driven I think. I only got
that answer 2 of the 3 visits. Other constraint driven - not sure
(1 of 3 yes answers). It is a top level build tool - gridless route
at the top is good - however - they weren't consistent in answering
issues on core I/O, building the peripheral rings, global line balance,
etc. I still have to lean towards either Plato or Columbia for the top
level nod. Pulsic has some good technologists. If only they had
marketing and a target client base they could be successful. I do not
see Pulsic as a player in the analog/mixed signal space. Their tools
are NOT as good as Mentor/Synopsys/Cadence's current offerings."
- Pallab Chatterjee of SiliconMap
"Pulsic sells a shaped based router (could it be like Plato Nanoroute?)
It does not do placement, although it will do small placement
adjustments. Interestingly, the salesman tried to convince me this was
their first year at DAC, even though I remember them (him, in fact)
from the last two years. I assume this means they feel they came
before they were really ready but they're ready now and want a fresh
start (same story I heard from Monterey Design). Heck, we all deserve
a fresh start. Anyway, their tool is for custom, analog and top level
routing (not routing of big blocks). It has full RC extraction and can
fix crosstalk problems on its own. Pulsic plans to release a tool for
doing routing of blocks in 1Q03."
- John Weiland of Intrinsix
"Cadence's Virtuoso has 3 parts to it:
a) The polygon editor.
b) VCR: This is the basic router of the original CCT.
c) CAR: Chip Assembly Router, which is VCR with global router,
power router, and a host of other good features.
There are a number of very powerful features to the Cadence VCR/CAR
suite. The CCT router was a very powerful area router and excelled
in chip and cell level assembly. It has a number of useful
interfaces to Virtuoso XL, that allows you to assemble at the
transistor level.
I would not like to expand on the Cadence suite or write a
comparison table.
I will instead devote my note on the value of the Jupiter/Columbia
suite. A reader can draw his own conclusions at the end of it:
i) An extraordinary level of integration
- the Milkyway database allows an enormous amount of
integration of the tools: Jupiter, Columbia, Astro,
Hercules, Star-RCXT,...
- Since Columbia is used at the full chip level, there are
no artificial boundaries between the tools. This is
important for the transitions between floorplanning
(block and signal planning), power and clock routing,
signal routing and chip assembly is very good.
- I do not have to support multiple technology files.
- I do not have to support multiple constraint files to
control the router and the same timing driven features.
ii) Productivity in floorplanning and chip assembly is highly
dependant on the GUI. When you have about 10000 signals
and about 30000 pins and several hundred pads to be worked
on, GUI is critical.
By having a uniform GUI between all the tool suites, Columbia
is definitely very useful. I do not have to write a set
of scripts for floorplanning and another for assembly.
iii) The router is an area router and a very powerful one. I can
go into a number of details on the value of this router, but
to make a long story short, Columbia can be used to manipulate
many polygon-level issues (traditionally these edits were done
by hand after routing).
iv) The degree of control in the router is critical to a successful
route in high performance designs. At my frequencies, my initial
assumption was that I would end up routing the wires by hand.
I have yet to start my manual edits using Columbia.
v) Transition between hierarchies is very easy. There are no "hard"
integration options that forces the internal data of lower levels
of hierarchy to be visible at the top. And vice versa.
vi) Speed of the Columbia router. It is very fast and the route
quality is excellent. Controls to the router are many and it
can be scripted.
And finally, they have a solid technical team that has been very
customer-oriented in my experience."
- Ram Sunder of Specular Networks
"When Synopsys bought Avanti they got the Columbia router. This is a
shape based router for doing top level routing and probably a more
capable tool than Synopsys Flexroute - don't be surprised if Flexroute
(the old Everest router) goes away."
- John Weiland of Intrinsix
"The good news about Astro is it is high capacity. If we design over
2M gates, Apollo/Saturn's capability is not enough. Up to 5M gates,
Astro's optimization capability is very efficient in a flat layout
style. The down side is because it's a new tool, I think current
Astro has several functions which are not completed. So we use both
Apollo and Astro. If Astro has some problems, we can use Apollo
for work around. Both tool has same database, Milkyway. So it is
easy to switch each other.
Over 5M gates, it is difficult to design by flat layout style anywhere.
So we are benchmarking floor planning tools, Jupiter-XT and
First Encounter. We're using a 12M gate design for this benchmark,
but unfortunately it's not finished yet.
I don't know which way is better, traditional hierarchical layout like
Jupiter-XT or 'virtual' flat layout like First Encounter, Floorplan
Compiler, or Ammcore. If hierarchical layout design would become
mainstream, routing tools like Pulsic might be necessary in the future.
Regarding database, I think Milkyway is the best."
- Zenji Oka of Ricoh
"I don't know anything about the Synopsys PhysOpt tool, but I've been
looking at Avanti. We currently use Apollo. I think Astro will be a
good tool, and will get the job done. I don't think Avanti has synthesis
as tightly coupled into the physical design yet as, say, Magma does."
- Cordell Prater of Prairie Communications
"To date, we have completed several tape-outs from Astro through
Enterprise (staying within the Milkyway database the whole way, with
Star and Hercules for signoff parasitics and DRC), and all our new
Avanti work is being started in Astro rather than Apollo/Saturn. It
will be the primary P&R for the majority of our designs.
If you can make a tool more complex, I haven't seen it yet. Astro has
inherited virtually all of the functionality that grew up in Apollo and
added timing optimization at nearly every phase of the process from
pre-placement to post-routing. In Astro they've worked on parasitic
estimation/calculation, clock-tree synthesis, high fanout net collapse
and resynthesis, database integration, standard input and output
formats, real-time hierarchy maintenance, and the twisty little
passages of impossible design rules. Most of it actually works.
About the only thing that missed the boat along the way for Astro was
handling of CTS simultaneous with initial detail placement -- though
this is promised to return in future updates.
What can't Astro do? Really big flat. SOC-E with FE placement and
Nanoroute routing may have a better chance in that arena, but they
don't seem to be there yet -- certainly not at the level of the Astro
integration. Magma is making noises about big flat too, but I haven't
seen it yet either. And so is Ammocore.
I like the way Astro has been going so far. I just hope it keeps
going more or less the same way after the acquisition. I don't know of
a more responsive company than Avanti."
- [ An Anon Engineer ]
"Monterey and Magma had no solid solution yet when we started our eval
back in 2000. Astro was not there then either. Avanti Saturn is trash
as we have seen in our 4-5 test cases. Every case hit fatal bugs in
Saturn. The timing engine and SDC parser is Avanti's big trouble. As
for Astro, the router is even worse than Apollo II in our test cases.
Avanti salesperson even told us not to use some fragile fuction in
Astro in March.
The big problem of Apollo is the release quality. We always have to
keep more than 2 versions to complete a job. Buggy in every version.
Old bugs can pop out in new version. Astro can not prove to be usable
here by now. So we will stay on Apollo until next year."
- J.S. Yang of Sunplus
"As of version 3.5.0.3, Astro offers all the features a P&R person needs
to get blocks of up to 150,000 stdcells in 0.13 um technology done.
It's Database/Visualization/Editing are as good as they were with
Apollo or better. (Better ways of selecting which route type one wants
to see, placement map/global route map/error viewing/editing/CEL
viewing/... you name it, Astro works as one wants it to work.)
It would be nice to have "geImposeCell" as an option in the display
settings. You can impose 2 versions of CEL's or CEL and FRAM view to
check whether they match. Comes in handy if you don't have the
Calibre/Astro interface and want to load/view Calibre GDS error files
on top of your CEL view.
Verilogin/hierarchical Verilogout finally works (have not tried it
with "fix module boundary" or "scan-chain reordering" yet.)
Astro's SDC-constraint based timer and TLU+ based extraction give
progressively better corellation data going thru placement/global/
detail route/PARA view (StarXT-data). You can check the SDC
constaints right after loading them and the pre-placement gives you a
good feeling on whether your initial netlist will meet timing with
these constaints. The supported SDC list gets longer and there are
only a few discrepancies with regard to clocks going thru XOR's where
PrimeTime and Astro disagree.
Astro has all the timing optimizations that one could ask for and they
do work. The pre-placement/in-place/post-placement/in-detail route/
post-detail route optimizations are as good as the SDC constraints and
your timing models/capacitance data allow them to be. Astro's gated
clk-tree insertion works as advertised. Skew is usually OK and double
with double spacing clk-nets is supported.
Antenna removal works if your library/techfile provider did his
homework and corellated it with your runset.
Using StarRCXT to generate the PARA view gives timing analyzer results
that are the same as PrimeTime's.
Distributed detail routing nicely speeds up the routing stage. 2 GHz,
dual CPU, 2 GB memory Linux machines give you the fastest runtimes as
long as your block (under 150,000 stdcells) fit's in 32-bit address
space. Beats HP/SUN's runtime by a factor of 2.
Overall: Astro/Calibre/StarRCXT/Verplex/PrimeTime give you a nice
toolset. You just have to figure out which floorplanner to use."
- [ An Anon Engineer ]
"Avanti: We had several Avanti tools in house. With the exception of
Star-RCXT which is an EXCEPTIONAL tool, 2 years (< 5 bugs), Avanti
support sucks BIGTIME. It's like dealing with an organization with
a collective bipolar personality when it comes to support. Irrational
and illogical. P&R tools require extensive support and Oki did not
need any additional migranes so Avanti NEED NOT APPLY.
SI was a weak area for Cadence but CadMOS was a good addition and that
solved the X-talk issue. Their SE-SI tool was really bad with IR drop.
Cadence purchased Simplex (Oki was evaluating Voltage storm anyway) and
that solves the IR drop problem. For antenna checks Oki used Calibre
(Mentor) but Cadence came up with an enhancement to Warp route to fix
antenna's with diodes. Oki was also looking at the next generation
router. (Guess what Cadence acquires Plato.) It fits right into our
roadmap. Oki also had recently replaced the archaic Star-DC with Delay
Storm as the new delay calculator and that fell into the Cadence camp
with the Simplex acquisition as well.
The other weak area was power routing for Cadence and they recently
released Silicon Ensemble 5.4 which has substantial improvements in
power routing."
- Syed Ahmed of Oki Semiconductor
"The real deciding factor was with SE-PKS, we got the physical synthesis
capability, and we were able to integrate CTS, floorplanning, and final
routing with the same tool(s). Once you purchase PKS, it wasn't too
much more for SE."
- Dave Garrett of Bell Labs Research Australia
"I think Apollo's router gave better results than SE Warp Route but
that's an impression, not based on hard data."
- [ An Anon Engineer ]
"We continue to use Cadence's Silicon Ensemble Warp Route and it does a
good job signal routing our PhysOpt-built designs. We could continue
using Warp Route for that task for the forseeable future. However, if
Plato NanoRoute proves itself superior in wall time, results or both,
we'll switch to it as appropriate.
The biggest hole we have in the physical design is power grid routing.
We used SE Sroute to do this for our chip, but boy was it painful
figuring out how to drive it. This coupled with our big desire to roll
floorplan iterations quickly have us very interested in the physical
prototyping tools."
- Mark Wroblewski of Cirrus Logic
"Plato's router seems to be super fast but then again, what matters at
the end of the day is working silicon and fastest time to tapeout;
which means DRC/LVS and timing correct designs, not just speed.
Nothing about databases. We as customers will have to live with both
the Synopsys standards and the Cadence standards though both claim
they make theirs open."
- Jai Durgam of SiImage
"We're currently evaluating Cadence SoC Encounter, and one of the best
parts of it is the NanoRoute router. NanoRoute rocks! - it blows away
everything else. I'm in the process of using it to tape out a 1.2
million instance design, routed flat, that would fail in any other
router. Nano is very fast, very high capacity, and easy to use
although there are a few minor bugs I had to work around."
- Eric Despain of Micro Linear
"From the Cadence customer point of view, I think the purchases of
Silicon Perspectives and Plato was a good idea that brought good
technology into the company. At the same time I think it says
something about the state of Cadence's R&D efforts. The big issue
is how well Cadence does at integrating these new tools with their
existing tools."
- [ An Anon Engineer ]
"Cadence bragged a lot about the Nanoroute router they got when they
bought Plato. The tool can divide a job between multiple processors
and they say the time to do routing varies linearly with the number
of processors (this would require some smarts dividing up the netlist
and maybe the right type of netlist). They say it does on-the-fly
crosstalk analysis and prevention and does shielding better than
Silicon Ensemble, in addition to being a lot faster. They said they'll
continue to support Silicon Ensemble for routing but Nanoroute will
become their first choice for new customers.
Apex Design Systems also sells a placer that can be distributed over
multiple processors. Rather than dividing up the chip by blocks or
stripes, they have a very clever way of actually doing multiple
iterations of the placement at the same time. They claim their
benchmarks show the results as more predictable than other placers."
- John Weiland of Intrinsix
"NanoRoute is the best stand-alone cell-based router on the market
today. Cadence is also leading in custom IC/analog routing tools.
Did you say "Genesis"? I think today it is called "OpenAccess".
The success of Cadence with their standalone routing tools is in
sharp contrast with their lack of success in the OpenAccess database.
A customer once said: "To work around tool bugs, you need ASCII data
files. And tools always have bugs". So I don't bet on this Cadence
database stuff at all."
- [ An Anon Engineer ]
"Our users of SE find it slow and lacking features and are asking for
Astro. We also aren't impressed by SE router performance, particularly
for incremental ECO routes."
- John Webster of Intel
"I am an Ambit BuildGates and NC-SIM user, but our backend team is
actually using Avanti Astro."
- Garret Shin of Lincom
|
|
