( DAC 02 Item 15 ) ---------------------------------------------- [ 9/10/02 ]
Subject: Monterey Dolphin, Sonar, & Aristo IC Wizard
COMING OUT, GUNS A BLAZING: In the 4 years since its founding, Monterey
has been the laughing stock of the physical synthesis race because they had
no know user tape-outs. Oh, Monterey played the game all right, issuing
press releases and appearing in occassional stories in the trade rags -- but
*anyone* can say *anything* in a press release. News stories full of quotes
from Monterey employees mean nothing. Nobody took them seriously. Then
about 3 months ago, I started receiving a number of detailed Monterey user
letters. ESNUG 396 #2 and ESNUG 397 #6 had 2 Infineon, 2 Canon, and 2 Zoran
tape-outs. Below you'll find 5 more tape-outs or near tape-outs from users
responding to my DAC Trip Report survey. These aren't B.S. press releases.
They're not news "stories" with hand picked customer VPs who have no actual
hands on experience with the Monterey tools. Instead, they're actual
Monterey users bitching & moaning about the bugs they ran into while using
Dolphin and Sonar. In short, Monterey now has *working* physical synthesis
and these hands-on user letters tell me that it's real.
"Of course there were some bugs in Monterey's Dolphin like other tools:
- Generating unconnected lines when dragging out IO port
from Macro cell.
- Reading out undefined constraints in SLF when outputting
SDF after Dolphin Layout.
There were other bugs, but Monterey's quick response fixed them. Plus
they cope with our enhancement requirements very well.
Canon's standard Language is VHDL, which Dolphin cannot read, but this
is easy to do convert through Synopsys Design Compiler. Of course I
hope Dophin will be able to deal with VHDL in the future.
Dolphin's upside is that timing closure solution is especially good.
Monterey released Dolphin Ver.2 which includes "useful skew" feature at
the end of last year. This feature enables max skew time to be set to
a certain value with only one command, for example:
useful_skew_setting -synthesis on -set_max_skew 4.5
As you may know "zero skew" is a general method for timing closure,
but the "zero skew" method sometimes can't provide proper netlist even
when timing is met. Zero skew gives many critical paths and of course
it takes much time. I don't know any other tool which includes "useful
skew" except celestry's ClockWise other than Dolphin.
Anyway, good things of Monterey's tool are, in short, Dolphin includes
every function: logic optimization, power routing, clock tree
synthesis, placement, routing, extraction and timing analysis, and it
does not require very much script writing. Our design team was off the
ground in the beginning of 2001, only 3 engineers (now 6 engineers).
They are not EDA engineers but ASIC designers. So we needed "Push
Button Solution" as a back-end design, if possible. We were able to
install Dolphin, set up our production design flow, learn to use it,
and tape out our first chip in less than six months. We already taped
out 2 chips with Dolphin."
- Hiroyuki Nakamura of Canon
"We taped out *two* 0.15u multimillion gate hierarchical designs -- more
than 5m total placeable objects with both internally and externally
developed custom IP blocks -- using nothing but IC Wizard, Sonar and
Dolphin for floorplanning and P&R. Synopsys Design Compiler was used
for synthesis to generate the gates fed to Dolphin. We used Simplex
VoltageStorm for IR analysis. Simplex extraction and PrimeTime were
the timing signoff tools. A customized version of "magic" was employed
for custom layout and flip chip work. Did LVS and DRC with Calibre."
- [ An Anon Engineer ]
"Currently, we're taping-out a chip for Netlogic MicroSystems in using
Monterey tools. Here's our feedback based on 8 weeks of tool usage:
Strengths:
1) Ease of Use:
ICWizard + Dolphin -- we were able to go thru first-pass P&R in
less than 2 weeks. Dolphin is easy to use, and due to its TCL
interface, it's easily customized.
2) QoR:
Dolphin's setup as well as max-trans optimization are quite good.
They seem to have a good engine under the hood. Even the router
is pretty good. We took our first-pass P&R results thru Hercules
and we were surprised to see clean LVS with a very few DRC's
(only on some arrayed-vias which could be tech-file related).
Their clock-tree-synthesis engine is also very good. We tried
both flat as well as hierarchical CTS (by specifying insertion
delays of lower-level blocks), and we got predictable results.
Dolphin provides a very good interface to path tracing, and
creates a cool "critical-path schematic" which you cross-probe
into layout.
3) Database mgmnt:
Both IC Wizard and Dolphin read standard formats (LEF, DEF, SDC
constraits etc.), which makes it easy to start a new design. I
liked their concept of "checkpointing", which allows a user to
"snap-shot" a solution at any given point in design. All the
necessary files (netlist, floorplan, constraints etc.) are
encrypted and put in a separate directory for a given checkpoint,
so all it takes for another user or a Monterey AE to look at
someone's work, is to ask for the checkpoint directory.
Weaknesses:
1) A lot of Dolphin steps (e.g. placement, max-trans optimization,
setup-opt, CTS, hold-opt etc.) are bundled in "macro-commands".
(For example, their all encompassing command: "run".) At times
I felt at loss of "knobs and switches" for cutomization. I
needed a Monterey AE to onion-peel some of their macro commands
to give you the level of control I wanted.
2) IC Wizard still uses some Aristo format files, which Monterey is
in the process of obseleting.
3) Dolphin CTS does not provide a whole lot of control on
"insertion-delay". We had a scenario where there was a block
that received a late clock, and CTS needed to really reduce the
insertion delay, even at the expense of worsening the skew a tiny
bit. But we could not find a way to do that. Even worse, there
are no "cts levels" controlling knobs like in Apollo where you
can specify how many levels of buffering you want, etc. Since
mis-match in insertion delays for blocks translate to skew at the
next level, it is extremely important for P&R tools to perform
CTS with both insertion-delay and skew as equal importance
constraints. Also, over 400 Mhz, most about everyone wants to
customize the clock-tree, which requires P&R tools to provide
features like controlling the levels and number of buffers/type
at each level as well as handling pre-routed clock-trunks etc.
This seemed to be absent in Dolphin.
4) The Dolphin router is relatively slow. We want it faster.
5) The Dolphin router does funky jogs when routing to hard-macro
pins, where you would expect it to do straight shot routes.
Monterey describes this behaviour as off-grid pins, but we are
not convinced. They're working on this.
6) Dolphin's Seascape GUI has troubles selecting dense/overlapped
geometries while clicking due to what they describe as "not on
grids" etc., which we find hard to believe.
7) The Seascape GUI doesn't have good set of pull-down-menu commands
for common tasks like selecting and zooming to selected objects,
or selecting and viewing properties of multiple selected objects
by cycling thru each selected object etc.
8) Dolphin's abstract generator (LEF generation) creates extra pins
for some shapes near the boundaries during block abstraction,
which caused some pain for hierarchical methodology.
Overall we're mostly happy with Monterey. They've been very responsive
to our issues. Their pluses outweight their minuses."
- Bharat Gohil of SiNTEGRA, Inc.
"I did see Monterey's Sonar/Dolphin tool at DAC this year. I very much
appreciated Monterey's approach of getting a real user that gave their
experience using the tool set. The user gave both the good and the bad
and it was clear the good far outweighed the bad. Magma, Synopsys and
Cadence have good backend tools, but they still appear to do all the
layout tasks sequentially. Monterey's approach impressed me and that
is why we decided to adopt their tool set for our COTs flow. As soon
as we get thru our first design using their tools, in ESNUG tradition,
I will be able to let you know the gotcha's we found."
- Kevin Guy of Astek Corporation
"Regarding IC Wizard:
1) The GUI is not very intuitive, and it alphabetically lists it
commands, as opposed to being a task-oriented listing.
2) Their files management uses a library/folder/portfolio/cabinet
mechanism, which we found a bit clumsy. The database saving
mechanism is not very efficient, since it creates versions of your
designs, as opposed to user-defined names for designs.
3) Most bind-keys are counter-intuitive and are much different than
Apollo, Virtuoso or other main-stream tools causing a lot of
troubles. For example, if you happened to have selected a set of
pins, and you wanted to PAN, if by mistake you click on <-- or -->
key, instead of panning, your selected objects will MOVE !
4) Cannot select and interactively move and manipulate pins so you end
up hacking ACF (Aristo Constraint) files."
- Bharat Gohil of SiNTEGRA, Inc.
"We are currently using Monterey Sonar and Dolphin tools on 0.13 um ST
process. Our Designs (two 1M gate blocks + RAM's among a 8M+ gates)
have made Synopsys/Avant tools to dive into congestion issues and
timing problems.
The "Nice" things we see in Monterey vs. Synopsys/Avanti solution is:
1. Dolphin is a single design platform from netlist to GDS -- no need
to worry about passing data to/from tools. So we get rid of funny
things like Avanti tools not understanding timing exceptions like
Synopsys, modelling issue for tristate cells ...
2. Clocktree synthesis is part of the timing optimisation -- no need
to overdesign based on skew budgets, hold fixing is part of
optimisation run, no extra loop post routing needed -- which often
breaks some setups.
3. The prediction feature in Sonar is working -- at least if you do
not push the design to the limit -- and saves time, speeding the
iteration process. Really helpful to debug the timing aspects.
4. Scripting is easy and reduced. It was 3 pages versus 15 pages.
The "Not so Nice" things:
1. Today's version is consumes more memory than the competition.
2. Integration of IC Wizard floorplanning with Sonar/Dolphin is not
yet done.
3. Although it's useful, Sonar's prediction feature is not 100%
trustable when pushing the design to the limit -- but I do not
know of any tool able to do this today.
4. some features exist which are not so useful -- like terminal
maps of IR drop. It's not precise enough.
Overall I strongly believe that Monterey is ahead what we have seen so
far. How it proliferates is another story as it is a small company."
- Francois Remond of STMicroelectronics
"I have not seen any real market push or killer technology from
Monterey. The Avanti/Synopsys Aristo stuff is the same since 2 years
ago. Same customer testimonials and benchmark data. I am not sure
who is looking at them. All the folks I talked to only mentioned
SPC, InTime and Magma. Monterey would not let me into their suite
this year (not sure why) so I only saw the product on the floor. It's
OK, I guess. Seems like a lot of money for just OK."
- Pallab Chatterjee of SiliconMap
"I did see a demo of Monterey Dolphin at DAC, and I heard their customer
presentation by Zoran, as well. The most impressive thing to me about
Monterey is their business model. You pay for the s/w at tapeout. The
Zoran guy made a big deal about it. It aligns Monterey's interests
with mine. I like that. Our company is a small start-up and would
benefit from the lower cost of ownership as well.
As for Dolphin features, I'm looking for a tool that does crosstalk
aware routing, power optimization, and is easy to use. It looks like
Monterey can meet these needs. It's hard to really know in a demo.
I think Magma may have unique idea with their gain-based synthesis.
This could produce a more optimum design faster. We'll probably have
to do an eval on both tools. If they have similar performance, as far
as I'm concerned Monterey will take the lead on the business issues!"
- Cordell Prater of Prairie Communications
"It's far too early to say anything definite. But our first impression
of Dolphin has been positive. So far we have been playing with small
(30 kgate) blocks to familiarize ourselves with the tools. The routing
gets done into timing. Our clock-tree (which has clock gating) seems
to complicate their CTS.
The tools seems easy to use and should be easy enough for the frontend
designers to get feeling of implementation issues (can't be bad thing).
As for the gotchas so far:
- The Monterey tech-files are not available from the foundries. This
is not very suprising since Monterey is not the market leader. But
this is not a big issue. Monterey seems to support tech-file
generation quite well.
- Support for HP-UX (our 2nd choice) is not as good as for Solaris,
and the HP version will be discontinued. Linux (our 1st choise)
support is not yet in place.
In 2-3 of months we'll try to make a real tapeout with these Monterey
tools. When that is done I'll have realistic view of things."
- Mikko Laiho of u-NAV Microelectronics
"We're using Cadence (FirstEncounter/SE) and Magma (BlastPlan/Fusion)
for ASIC chip implementation. We don't use PhysOpt in our design
center but we support DEF/PDEF interface for ASIC users who run
PhysOpt. Comparing with Cadence and Magma, Monterey's strength are
- Concurrent process for chip implementation
- Multithread process architecture
As for concurrent process, Monterey has big benefit to reduce time
for chip implementation. Currently if we use Cadence tools, we have
to go through serial processes by engineer:
placement -> routing -> extraction -> delay calculation
-> timing optimization -> signal integrity verification
-> SI fixing -> extraction -> delay calculation -> timing....
And it is very painful to use many point tools go through because of
translations. If it is true what Monterey sales talk, our job will be
easier a little bit.
Another Monterey strength is scalability because of multiple CPU usage.
No one can accept to have 10 times run time because of 10 times complex
SoC chip implementation. Nice talk.
Anyway, I don't trust claims by any EDA vendor presentation. I trust
benchmarks. I don't have a benchmark at this moment."
- Hideki Yamada of Toshiba
"From their DAC presentation, Monterey appears to have a more seamless
solution in getting from floorplan to final layout than Synopsys
PhysOpt. I think they both do a good job at considering all the design
constrains during placement and optimization. Things such as setup,
hold and congestion. Although my gut feel is that Monterey is more
congestion aware than PhysOpt.
As I understand things here Monterey has some difficulties fitting
into our Avanti signoff flow. PhysOpt would have similar issues except
that we only use placement data from it and then let Avanti do the
detailed route. We also use some internal tools to help close timings
that are in our sign off flow that neither can address. But, with
Monterey I don't feel we would need to use these internal tools as the
design would close timings as it comes out of Monterey.
This is from their DAC suite. Reality may be very different."
- [ An Anon Engineer ]
"Based strictly on their DAC demos, Monterey seems to have a nice,
integrated flow. The feature I most liked is they can generate
feedthoughs. I want to have "abutted" blocks. The only thing I use
routing channels for is clock trees. The Monterey tool looked like it
would work well. It also looked like it was highly integrated. While
at DAC I tried to talk to others who were using it. The gossip I heard
was that when it worked, it was great. I also heard others tell me it
was their last choice. It seems like Dolphin works well for certain
designs but falls on its face for others. Again this is just the
gossip I heard. I have no real experience. We're partners with ST.
They really like IC Wizard but are still evaluating Sonar/Dolphin."
- Maynard Hammond of Scientific-Atlanta
"We have First Encounter/PKS in house, evaluated Monterey and Magma. In
my opinion, Magma has the best integration and user interfaces and runs
fast on Linux. Monterey's strength is at it's multithreaded engine and
64-bit code, so potentially it could handle very big design, but we did
not test that yet. Monterey's weakness is integration. It's not easy
to use IC-Wizard together with Dolphin and has many problems need to be
iron out, which means a user will have hard times in using Monterey's
"hierarchical" flow. Another Monterey strength is its Signal Integrity
features are build-in. Monterey does not have Linux port yet. Their
synthesis engine is not a full-blown synthesis engine, it seems doing a
reasonable job during optimization though."
- Tien Ying of Tvia, Inc.
|
|
