Riad KACED wrote, on 04/30/08 00:15:
> Hi Ahmad,
>
> The ideal QA tool will check the integrity of every single file and
> re****ts any single change between PDK versions. This ideal tool is
> unfortunately a utopia that does not exist in a real world. We have
> tried to make this with 2 or 3 of my colleagues in my former company
> and we managed to get something very powerful but it does not prevent
> from leaking and letting couple of bugs going through …
>
> The duties of such a tool is to run all the flows sup****ted by the
> PDK, not only on inverters but real-case designs as well.
>
> In other words, you have to make modules for :
>
> Digital/Analog/AMS/RF FE simulation (DC/AC/Tran/Noise …)
> DRC
> LVS
> RC Extraction
> Dummy filling
> Layout fini****ng
> …
>
> Instantiate all the sup****ted devices, with the most significant
> parameter sets, insure that they instantiate well, netlist well, call
> the pcells well, run the callbacks well, check the CDFs, the Model
> files, pass the DRC, well exctracted by the LVs engine, well extracted
> by the Parasitic extractor engine … and That’s only a little bit !
>
> Run the DRC flows, the LVS Flows, RCX, …
>
> Well, the list is very long and It’s very hard to achieve.
> I learnt over my experience that the best way in QA-ing a PDK is to
> run it on real designs (recent tapeouts for example).
>
> The QA tool we made was a huge Perl engine which runs bits of Skill,
> Shell, Eldo (I was not using ocean/spectre at this time)
>
> That’s a pretty heavy stuff to tackle!
>
> Good luck anyway!
>
> Riad.
You might also want to take a look at:
http://www.cdnusers.org/****tals/0/cdnlive/emea2007/customdesign/1.8GiroudFCSR.pdf
Whilst the QA aspect wasn't the main focus of the paper, it does talk a
bit
about this.
Note that OCEAN _is_ SKILL. It's just a SKILL API to simply control
simulation
and evaluate results. It certainly makes sense to use OCEAN functions as
part of
your SKILL code to QA the PDK.
I've done this by auto-generating schematics, containing instances of the
devices, with controllable sources on each pin of the device - to at least
check
things like the fact that they netlist and simulate with no errors, and
then to
validate that the operating point access works, and so on. More complex
tests
are needed of course to validate that the models behave properly, but
that's
perfectly doable.
Cadence also has a tool "STEP" which is for doing PDK testing automation.
You
might want to contact your Cadence account manager for details.
Regards,
Andrew.
|
