( DAC 04 Item 16 ) --------------------------------------------- [ 02/09/05 ]
Subject: Magma Blast Fusion
GARY'S 98 PERCENT -- When I talk with Gary Smith of Dataquest about his
EDA marketshare numbers, he tells me he puts something like 98% of all
Magma revenue under what you'd call the P&R market. (Gary actually calls
it "implementation", but that's another story...) That is, Magma makes
98% of its income off of Blast Fusion sales. This might get more complex
this year if you look at what users are saying about Magma's power
optimization flow. How much will the Blast Power embedded in a Blast
Fusion flow change things?
The Blast tools for 90 nm TSMC was impressive. They demo-ed better
than any tool at DAC.
Upside:
- running on the same database, with the same extractor and timing
engine is very powerful.
- being able to synthesize and time across different PVT points for
power, size and speed all concurrently is the best in class.
- optimize power across different power domains
Downside:
- crappy support; the buzz around DAC is that Magma's support is
great during pre sales, but falls away quickly after the deal
is signed. By the time you get to production work, you're talking
with only help line personnel.
- DC Experts; we have a lot of DC experts and they view their
knowledge of how to get DC to give great results as their value
to the projects. Nobody doing synthesis at our company wants to
change and our group a great relationship with Synopsys.
If true, Magma's low power solution was equal or better than Cadence
and they blew Synopsys' presentation away.
- [ An Anon Engineer ]
For my own education I was checking the low power flows out. I was
most impressed with Magma's v4.2 story. It looked to be the most
comprehensive. It has the usual retention flops, voltage islands, but
also includes auto Vt optimization and decoupling cap insertion,
although most folks now flood free space with caps anyway.
The Magma power analysis looked comprehensive. It provides for both
static and dynamic analysis, including transient analysis of surges
and voltage drop induced timing delays.
I was also interested that Magma picked up QuickCap. I saw it as Magma
taking high performance and low power markets seriously. Actually it's
getting hard to avoid detailed analysis. I like QuickCap's niche of
high accuracy for up to 1,000 nets. There are plenty of cases of
critical paths and cells that fall into that category. The tunable
accuracy is a great feature. It will be interesting to see how they
utilize this technology in the rest of the product line. I hope they
extend it to full RCLK extraction.
- Grego Sanguinetti of Tektronix
Internally, we evaluated Blast Power and were impressed by the
consistency on leakage power performance from different initial
netlists. That's its biggest difference from Sequence and Synopsys.
But Blast Power runtime didn't gain much over them.
Magma has covered most power-aware features, including automatic level
shifter insertion, block-level power plane synthesis and analysis, and
automatic retention flops inference.
We will further look inside to see what their real capability is.
- Jian-Dai Pan of Faraday
I saw Magma's DAC booth demos for power integrity and DFM. Their
solutions are tightly integrated to Magma's flow. These functions are
higher level from a technical point of view, but they are expensive at
the stand alone tool point of view.
Magma's power integrity and DFM solutions appear advanced based on
their demos, but I have not done a real benchmark to confirm this.
- [ An Anon Engineer ]
Magma:
1. The tools can do optimization using timing variation due to both
SI and IR.
2. During placement Blast Fusion will respect placement blockages but
these same blockage areas can be designated for use as optimization
buffers placement areas only.
3. Blast Fusion also identifies and displays the SI and IR drop delays
in the fail paths for easier analysis.
4. They do multi mode, multi corner analysis.
5. They have a 45-degree router built-in.
Comparison with Cadence:
1. Similar to First Encounter in claims of RTL to GDS. The Magma GUI
is much more intuitive and better laid out.
2. The Cadence flow is more widely adopted in the industry and TSMC
reference flow. They have better parasitic extraction capabilities.
Magma Strengths: Very good GUI. Good analysis capabilities have been
added to the user interface which allows the designer to interactively
look at effects like SI and IR Drop. Good "what if" analysis for
Blast Rail and Blast Noise.
Magma Weaknesses: The common database for the entire flow, which is a
good idea in general, can be limiting because it doesn't allow the end
user to introduce point tools into the entire physical design flow.
This assumes Magma to have "best in class" tools for the entire flow,
which clearly is not the case now.
Another weakness is that the Magma flow is not part of the TSMC
reference flow which usually is an indicator of "sign-off" quality
in a tool suite.
- Dan McConnell of ATI Technologies
I had a good impression of Magma's tool. You can get timing closure
with one tool in the layout flow. I think Magma's weak point is RC
extraction and routing. It had some stability problems, but I haven't
seen the latest version of the tool.
- [ An Anon Engineer ]
We became interested in Magma due to their claims of improved ease of
use and claims of improved productivity. Designs coming into my group
are typically 4X in size from our previous generation. We're looking
to simply do more with the same number of people.
Magma's platform is built on a single executable. This means that
when you purchase a Magma license or receive upgrades, all of their
functionality is included with one license purchase. A college degree
is not required to license these tools. This simplifies things
drastically for management of tool procurement; not to mention the
execution of the tools. When the tool is used, everything is loaded
and available to the user in one cohesive environment.
The biggest technical win is the embedded methodology of how Magma
tools are to be run. This includes their Gain Based synthesis, but
in general terms it's that Magma's methodology doesn't perform
optimization on cells without having detailed data to optimize on.
This makes exploratory synthesis fast -- and we found it correlates
to post-placed results nicely. This approach has proven itself to
offer us improved productivity, less license usage, quick turn times,
and superior usage of low-Vt cells.
The interface to the tools is also nicely done. For remote users, the
tools can be run locally but the CAD database can be remote. Their
Mantle tool has the ability to connect to a remote session and transfer
only "Mantle data" back & forth - greatly improving remote performance.
Our remote users raved about this.
- [ An Anon Engineer ]
We've been using Magma Blast Fusion (and Blast Create) for more than
a year now, and had working silicon on two different chips.
- Joseph Dao of Aeluros
Magma needs to clean up their act in the physical synthesis domain.
They are so much constraint driven that if the constraint for one path
is loose, wrong or simply not there, they will unravel it, thus causing
timing violations. Validating all the constraints is a monumental task
and requires a more formal way of checking them. This is why tools
like Fishtail focus are a must.
- Himanshu Bhatnagar of Conexant
As opposed to the old standard of Astro/PhysOpt, I've recently used
Magma to complete a 0.13 u tapeout. From preliminary unsanitized
thru to finalized constraints, Magma's physical synthesis results were
predictable. The timing issues traced to non-optimal floorplanning
were easily resolved thanks to a new graphical timing display (v4.2.x),
usable scripting interface, and intuitive help system.
There still remains some issues to cleanup before we could claim that
Magma has a viable multi-voltage physical synthesis to GDSII solution.
Just like the other tools, a little 'dont_touch', strict power railing,
and regioning finess allowed completion.
At least for 0.13 u, Magma has my vote.
- Marvin Dong, Consultant
Index
Next->Item
|
|
