( ESNUG 433 Item 2 ) -------------------------------------------- [10/20/04]
Subject: ( ESNUG 399 #7 ) One User's Critique Of Sequence's Columbus-AMS
> We extracted with both Star-RCXT as well as Sequence's Columbus tool and
> actually discovered that we were using the wrong Tech file for Star-RCXT
> since the numbers weren't correlating. Once we fixed that, we got
> correlation to within 5% between SPEF from Star-RCXT and Columbus (as
> determined by using PrimeTime to do timing.)
>
> - Sudhanshu Jain
> Broadcom Corp. Milpitas, Ca
From: [ I Want My MTV ]
Hi, John,
Anon please.
We use Sequence tools for all high-speed mixed-signal or analog
circuits now. The results have been very good. Before we used Cadence
Diva and Dracula (more than two and half years ago.) Columbus-AMS is
much more accurate and offers more functionality. The technology files
were easy to setup. We don't intend to do large logic designs with
Columbus-AMS, at least now.
The bottle-neck in extraction now is the data-reduction. We can
extract enormous amount of RLCM quickly but most simulators choke on
the input because data (RLC nets) is not reduced enough. Columbus-AMS
attempts to do some reduction but more functions and automation &
guidance is required in this area. We can, however, find quickly the
resistive drops of power or clock nets, and we have control over the
size of minimum extracted resistance.
Once we include coupled caps and inductance into extraction the size of
extracted data grows very quickly and slows down the simulation.
Typically we do not have very large block sizes in mixed signal designs
so this has still been manageable by simulating through server batch
jobs.
I believe other tools also have the same limitation in net reduction.
Via and contact farms have been a problem with RC-extraction because of
numerous small extracted parasitic components/nets. It would be nice
to have reducing technique that gradually trades accuracy with the
complexity of extracted nets, eventually dropping into a pi-model, such
as a Savarino-O'Brien approach, if necessary. So you could do a quick
and dirty simulation followed by more accuracy later once you have
frozen the layout.
One of the challenges in analog/mixed-signal is the 'exotic' components
extraction that you do not see in digital world. These are RF
varactors, inductors, caps and resistors etc. All these have to be
extracted by your device extraction and excluded in LPE.
This means that we have to develop the device extraction code ourselves
for Diva. Sequence provides then good scripts to link Diva extraction
into Columbus-AMS LPE.
Having to use external device extraction is a weakness, though. They
depend on other tools in this respect. Other flows than Diva based
extraction are also available, but we are not currently using them in
production work (although we are testing another extraction flow now).
If you don't run Diva LVS, it might be difficult sometimes to map
complex layout nets to your schematic nets. Here is where Sequence
could offer more help. We have customized the tool environment with
our own additional GUI to integrate the design flow and provide checks
and helpful functions on the layout etc. For those that do not have
CAD-department Sequence GUI is adequate for all the extraction and
probing part of the work.
In the latest revision, Columbus-AMS includes a nice viewer for intra-
net components or net-to-net coupling that works better than the
Cadence parasitic probe feature. I like this new feature a lot and
designers seemed to welcome it for layout debugging purposes.
I am currently promoting the idea that analog designers will have to be
deeply familiar with the layout. Using Columbus-AMS seems to get them
more involved with the details of the analog HF layout. So, note that
we do not limit tool use to technology service or layout area work but
have the designers actively looking their designs with the tool.
Starting to use the tool has not been a problem and parasitic
component/net viewer helps them identify problems. TCL-scripting is
available for advanced users who want more functions for analyzing the
data.
Columbus-AMS extracts inductance and mutual inductance and was first to
do so. Their technology library generation is well automated although
slow for 6-metal technology if you only have one library builder
license and cannot run parallel processes for this. You spend better
part of the week, if not longer, waiting for the library building to
finish.
Technology setup is well thought including latest process features,
such as non-rectangular cross-sections of metal lines etc. I was
missing only one feature, which was conformal coating option, but a
capping layer can approximate this.
- [ I Want My MTV ]
Editor's Note: There's an online demo of Sequence Columbus-AMS at
http://www.deepchip.com/demos/demo0433-02.fhtml -- I would love
to hear what other users have to say about it in ESNUG. - John
|
|
