( ESNUG 420 Item 6 ) -------------------------------------------- [10/22/03]
Subject: ( ESNUG 419 #9 ) 5 Users Report On The Magma Users Group Meeting
> Seeing that the Magma user's group (FUSION'03) met a few weeks ago, I'm
> wondering if the users who attended could send me a copy of their trip
> report that I could share with the ESNUG readers? (And remember, if
> you need to be anon, just say "make me anon" at the beginning of your
> email.) Thanks.
>
> - John Cooley
> the ESNUG guy Holliston, MA
From: Pallab Chatterjee <pallabc=user domain=siliconmap.net>
Hi, John,
Magma User Group Conference held 9/18 & 9/19, 2003 in Santa Clara.
In a quick summary: Over 120 qualified attendees, 14 quality papers, about
10 partners at the Partners' Expo and lots of good feedback from attendees
for regional meetings. The content & the message of Jim Hogan's lunchtime
speech were great. Best paper winner was Sandeep Mirchandani of Broadcom,
and two runner-up papers were presented by Howard Landman of Riverrock and
Hari Puravankara of WiPro in India.
As chairman of the Magma User Group (Fusion), I was very pleased at
attendance (up from the the first year -- we had ~110-120 people) and the
quality of the presentations given. An interesting customer demographic
which is indicative of the times -- it was attended about 90% by North
American clients. I guess the Asia and European travel restrictions by
most of the big semi guys are still in place. A nice twist -- these folks
actually ASKED if there was going to be a regional event for them to offset
this travel issue. There is some discussion in place at addressing these
global/regional issues in the near future -- so don't be surprised if you
see the scope of the user group expand to Asia and Europe soon.
A new thing for Magma was the enthusiasm of the partners to participate at
the show. Last year it was a bit like pulling teeth to get folks to want
to setup and display at yet another technical event on the calendar.
However, this year not only did the partners expo almost double in size,
but there were 3 partner papers presented (Veritools, Fishtail DA, and TSMC)
and a partner tutorial (Silicon Metrics). These papers were actually
filled with technical content on how they have interface and flow solutions
in the Magma environment rather than just marketing fluff. It was a really
nice change.
The only blemish on the show was the sound system for the lunchtime keynote
from Jim Hogan of Artisan could have been better. Copies of his keynote are
on the conference CD and it was videotaped. We are working on a logistic to
make this available in some media for folks who wanted to hear & see his
talk. The upside on lunch was it a less controversial choice of speaker
than last year, so the folks who hear him got to focus on the content rather
than WHY he was chosen as speaker.
Anyway here are some of the stats on the show:
Attendees= 120
Tutorials & Chat sessions = 5
Keynotes = 2
Papers = 14
There was an award for the best paper this year. Last year all presenters
got an Inaugural gift. The best paper winner (and receiver of the $250
prize) is Sandeep Mirchandani of Broadcom for his paper entitled "Hacking
Magma for Fun & Profit". Basically the paper was full of cool tips and
tricks you can do with the unified data model and their MTCL interface to
be able to check on the progress of a design and automatically implement
fixes. A full copy of this presentation is attached so you can post it on
DeepChip. It was pretty cool of Broadcom in this age of content paranoia
to actually allow one of their working chip guys to present scripts and
steps that could be used and interpreted by other folks. It shows that
marketing and legal have not yet ruined ALL aspects of engineers being able
to talk to each other.
The two runner-up papers were "Stability Analysis of RTL to GDS Flow" by
Howard Landman and "Taping out WASSP Using Magma Tool Suites" by Hari
Puravankara of WiPro in India.
Howard's paper prompted the most questions. He presented an engineering
statistical review of the Magma tools from both a repeatability and
performance stand point. His goal on the stability analysis was to
identify if the tools performed Randomly, Predictably or in a Deterministic
manner. He also presented that while most engineers THOUGHT they needed a
deterministic tool, they ACTUALLY needed a predictable tool. He also made
an interesting point that caught the Magma Markteting staff by surprise,
with all the emphasis on "delay stability" he found that the AREA is the
most stable in a design using the gain-based synthesis. The observations
Howard came up with are:
- When timing gets harder to meet, the fluctuation in area is less
than the fluctuation in delay.
- The area traded off for timing is small compared to Design Compiler.
- Slightly over constraining BlastRTL and BlastFusion can be a good
thing, but over 5% should require a change in the RTL.
- Leave yourself room, if the congestion gets high, the timing suffers.
I am also sending you Howard's presentation to post on DeepChip for your
readers to reference.
Hari's paper focused on a design of a custom chip of a couple of million
gates that they built in TSMC's 0.13G process with Artisan Libs. The
highlights of the paper were the detailed levels of the flow steps that
were used and their reliance on third-party signoff tools. Since this was
their first major Magma design -- they due diligenced the design with the
extra steps of third-party signoff. The paper shows how they integrated
the Magma flow with PrimeTime, the Simplex tools, Assura and Celtic. It
also gave a wish list of items to be updated to better support their
hierarchical methodology. Once again, I am sending you Hari's presentation
to post on DeepChip for your readers to reference.
The partner paper that drew the most interest and questions was from Ajay
Daga of Fishatil Design Automation (new company, new product). His
presentation was titled "Timing issues and False Paths/Multicycle Paths".
Basically they have a new tool that auto identifies these paths so when you
run P&R you are only trying to close timing on the real constraints of the
design. The big application target is when individuals who were NOT the
RTL designers and may not know their way around the code do the P&R. The
tools can work and identify most of these issues, so handlers of "blind"
netlist blocks are not spending most of their time on un-real things. I am
sending you a copy of Ajay's presentation to post on DeepChip.
Other quick topics covered: Visualization in the RTL-GDS design flow from
Veritools, library qualification by TSMC, tips and tricks on BlastFusion
by Calix, an industry overview paper about ASIC vs other models and the
role of EDA from Open Silicon, and several user success stories including
a concept to tapeout in 3 months and a 9 M gate design from Airgo & Renesas.
- Pallab Chatterjee
Chair, Fusion'03
SiliconMap, LLC Livermore, CA
---- ---- ---- ---- ---- ---- ----
From: [ Stacy's Mom Has Got It Going On ]
Hi John,
Keep me anon please.
Here's what I saw at Fusion. It's still pretty darn small, with one paper
presented at a time.
It's clear that the users like Magma compared to their alternatives. From
what I see, for the most part Magma users have evaluated other tools and
found that for what they are doing, Magma is the best choice. That's the
way it's been where I've worked recently.
Unfortunately, a lot of the time that people were presenting their papers
was taken up with "why we chose Magma". As users, we really don't care;
we've already made the decision to go with Magma, just like the presenter
has. We want to know the cool things you *did* with the tool. Don't waste
15 minutes of your allegedly 30-minute presentation on telling us why you
picked the same tool your audience did!
Where it showed the most signs of life was in the R&D discussion. There
was a lot of good back-and-forth that I think will lead to a better, and
certainly better-documented, tool.
In the meantime, I'll dream up interesting paper ideas for next year.
- [ Stacy's Mom Has Got It Going On ]
---- ---- ---- ---- ---- ---- ----
From: Philippe Sarrazin <philippe=user domain=teradiant spot calm>
Hi John,
I left Magma about a year ago to join an IC design house startup as Physical
Implementation lead expert of ... the Magma tools and flow, of course.
For the first time last week I attended the Magma user group event, as a
simple user. The first evidence is that this second event was bigger than
last year. This one attracted more users. I didn't count but easily, an
extra 30% attended. [ Editor's Note: 120 users attended this year. ] Also
the number of sponsors and partners was really much higher. This seems to
the sign of a dynamic growing community.
I have been globally pleased with what I saw. From the papers presented by
the users I expected more about hints and tricks or about the ways to work
around the tools limitations. But not really.
Users presented the way they completed their chip basically by following
the default Magma flow, flat or hierarchical, usually with a small team.
That doesn't mean that everything is push button. Magma-Tcl is still used
to complete tasks not yet performed by the tools (like pre-placemet of JTAG
cells, placement and connection of spare cells per clock domain...) Most
users are also very eager to know more about the capabilities of M-Tcl,
they usually think this is not documented enough (at all?). One paper
presented the ABCs of M-Tcl to initiate users to that "black magic".
We all knew that the Magma tools performed very well on flat designs, but at
least 2 papers demonstrated there is also now a real hierarchical top down
flow with BlastPlan or BlastPlan Pro. From the papers presented by Magma,
the most surprising to me was the one about BlastCreate, the new Magma front
end environment. Magma makes a very strong claim about run time, cell area,
timing closure and tool capacity. It is always amusing and exciting to hear
such comments. Is it real? I hope we could get comments of your readers
about that claim.
Users praised the Integrated Environment, the unique datamodel and the
advantages they get from this.
- Philippe Sarrazin
Teradiant Networks
---- ---- ---- ---- ---- ---- ----
From: Robert Knoth <robert.j.knoth=user domain=exgate.tek spot calm>
John,
This year, I attended the Fusion conference in Santa Clara. My company has
recently begun using Magma in our physical design flow, and this was the
first opportunity I had to interact with the larger Magma design community.
In the past I have been to the Avanti conference (AURORA) and the Synopsys
conference (SNUG). This was a very different experience.
CONS:
1. Smaller user group means a less diverse pool to draw papers from. At
the larger conferences, with multiple sessions at a time, you can pick
and choose papers to target the specific area you are interested in.
2. Too many case studies. After you have listened to one company's
tape-out story, they tend to blur together. In the future these should
be screened to only present the papers which have novel problems
and solutions.
3. Some of the worst acoustics ever for a keynote address. It seemed
like Jim Hogan had some interesting points about the direction of the
industry, but I'll have to wait for the video to find out.
PROS:
1. There were some very good papers. The ones which stood out:
- "Stability Analysis of the RTL to GDSII Flow" by Howard Landman
(Studying the impact of timing constraints on area/speed)
- "Low Cost ASICs" by Naveed Sherwani
(Reducing NRE and the Factory model of ASIC design)
- "Nothing to GDS in Three Months" by Michael Newman
(A case study in the need to remove guard bands from design)
2. Some very good discussions centering around the data model used by
Magma. This is one of the key differentiators from the competition.
It not only allows their tool to work on the increasingly interrelated
issues in designs, but also is allowing them to partner with other EDA
companies and integrate other tools (Veritools' "Undertow") with their
own. If only the Magma data model was better documented...
3. The interaction with the people from Magma was particularly valuable.
Not being a large customer, it is easy to slip between the EDA cracks.
This was not the case with Magma. Many were familiar with the problems
that we are currently facing at my company, and what was going on at
Magma to fix them. That kind of support is a welcome change. This is
surprising considering that at the same conference there were some much
larger customers.
4. Having the partner fair at lunch time was much better than tacking it
onto the end of an already long day.
Overall, the conference was a good experience. It will be interesting to
see how it changes as their user base grows. From what I have seen so far,
I expect it to. I also hope they are able to maintain their high level of
support and interaction. It's a great change from the status quo in the
EDA industry.
- Rob Knoth
Tektronix, Inc.
---- ---- ---- ---- ---- ---- ----
From: [ Ich Bin Ein Berliner ]
Hello John,
Last week I visited Magma's 2nd user group meeting Fusion 2003. Although,
I got not much news from a technical side. The program featured several
impressive tutorials and user presentations, with a general agreement on
the high value and quality of Magma's unified data model, and the
possibilities of Mtcl database access. Surely it was worth going there.
Please keep me anon
- [ Ich Bin Ein Berliner ]
Editor's Note: The top four user papers from the 2003 Magma User Group
meeting are #51 & #52 in the DeepChip.com downloads section. - John
|
|
