( ESNUG 555 Item 3 ) -------------------------------------------- [01/22/16]
Subject: Dean on IC Manage Envision -- his Big Data Tapeout Predictor
DAC'15 Troublemakers Panel in San Francisco, CA
Cooley: Dean.
Dean: Yes.
Cooley: You recently just launched some sort of tapeout predictor
tool with the words "Big Data" splashed all over it...
How's it going?
Dean: Yes, so we launched IC Manage Envision. As you know, at
IC Manage we have a lot of data, because we have the kind
of the database that keeps track of the hierarchy, and the
design data, and the changes, and who did what.
And all of the design teams have all these log files of their
tool runs, their DRC runs, their LVS runs. What we did,
basically, over the last three years, is worked on some
technology to put that all into a Big Data database -- and
then perform analytics on it to solve three key problems.
Also I think last year I talked about big data a little bit...
Cooley: Right. You gave a foreshadowing on it.
Dean: "Hey it's coming! Save your data!" But the 3 key problems
we that we're choosing to address -- and there are lots of
problems you can address with Big Data -- we're kind of focused
on stuff we think is key. One is design progress. And that's
basically automated, real-time analysis of how far along is
your design. It doesn't do you any good if it's not automated.
The methodologies that people use today are very manual, very
time intensive, very people intensive. We want to get that
automated so people can tell.
Cooley: Yeah, I saw that when I read your write up in ESNUG 550 #5,
you were like Big Brother sniffing of all the "use" records,
and so no engineer even put in reports. It was just sniff
your "tool use licenses" and things...
Dean: Correct. It's also focused - it's key to tie it into the
hierarchy of the design and the changes going on in design,
i.e. the "change number".
The second key problem we're addressing is resource allocation.
You want to know which blocks are behind, which blocks are done,
so that you can basically find the long pole in the tent,
reallocate resources, make sure your resources are where they
need to be. But then also to know how much resources you're
using to explain to the finance guys.
We had a presentation earlier today where Xilinx presented kind
of some of their success of being able to answer to the finance
people where all the money is going.
And then the third, is basically tape out predictions. So you
can kind of predict when things are actually going to get out the
door. And in order to do that, you need have one run behind you
historically, so that you can compare and see how you did last
time -- and start to map how that changes.
So those three problems are what we're addressing with Envision.
And I think it's one of the first really good applications of
Big Data into the EDA world.
Cooley: What's the difference between your stuff and - hell - when I
drove here, taking the taxi from the airport, I saw billboards
that were companies doing Big Data things and cloud things and...
Dean: Well, you know we are in Silicon Valley or you know in San
Francisco where Big Data is big. I think our stuff is different
because we're applying it to a EDA problem; to the chip design
problem. I think you're going to see other applications of Big
Data technology -- you know -- i.e. Big Data databases and kind
of this heavy analytics to other problems in chip design.
Amit: There's a lot, by nature a lot of data collection aspects.
There's prediction technology, and data reduction using machine
learning technologies. Solido, for example, we're handling the
Big Data problem in the SPICE simulation world -- being able
to take all that data and predict what are the areas of issue
in the design.
Cooley: Are you saying that your tools actually do data mining?
Amit: Yea, absolutely.
Cooley: Oh, OK. I know Calibre did something similar to that.
Are your... Yea, your other tool... [ looking at Sawicki ]
Sawicki: My other tool... Yeah, I mean in the test phase...
I love to say the words. We've actually been applying aspects of
multi-die, multi-design, multi-wafer, and fail log information
collected over six months. And then we get to do Bayesian
Statistics. I just love saying that word. I'm going to do it
again. "Bayesian Statistics" -- to try to pull out from the
noise what you get from diagnostics to go find out...
It's like literally, we have had dozens of customers, we went in
and said you have a via-5/via-6 problem in your manufacturing
line. And then they go in the PFA, there it is. There'll be a
voiding issue. There'll be a dishing issue. It's amazing what
you can do. And it's all about statisitics. It's a powerful
process that we're just starting to apply in this space.
Dean: [ off mic ] I think there is a lot of opportunity.
Cooley: Microphone.
Dean: Yes. There's a lot of opportunity. And one of the things we want
to do at IC Manage is to take our hierarchal database and be able
to map it on top of the runs and analysis that the engineers are
doing -- so they know which portion of the design and which
version they are working on -- and so we see our role as
partnering with a lot of these folks.
Join
Index
Next->Item
|
|
