|
June 25, 2005
EDA’s Shakeup
By Ann Steffora
Mutschler
Electronic News
|
|
|
EDA’s Shakeup
Electronic News
sat down at the recent Design
Automation Conference to discuss innovation in design
methodologies with Nitin Deo, Senior VP of marketing at
Ponte Solutions; Srinivas Raghvendra, senior director of
business development and marketing for DFM Solutions at
Synopsys; Thomas Grebinski, CEO of Oasis Tooling; and
Michael Naum, president, CTO and co-founder of Silicon
Dimensions. What follows are excerpts of that
conversation.
Electronic
News: Where is the
methodology shift occurring and having the biggest
impact in the design flow?
Raghvendra: Right now it is in the back
end and the proof is in the revenues. It’s definitely a
problem here and now for people who are going to 90nm
and 65nm. That’s not to say that front-end is not an
issue -- there will also be a change there -- but if you
take a snapshot today, and look at the more pressing
problem, it is in the back-end.
Deo: It’s not in the back end -- it’s
at every end. Wherever any design phase starts and ends,
that’s where the issue is. So, be it very high level
design, when somebody is designing at an architectural
level of description, from that point to RTL, there is
an issue because of complex functionality and being able
to foresee what’s going to happen next. So, it’s not
about ‘this end’ or ‘that end;’ it is about
predictability.
Naum: What we are seeing is that there
has got to be a way for logic designers to think
physically. DFM is the caboose on the train but there’s
an engine on that train somewhere at the top. The
problem is that when you get to the caboose, it’s too
hard of a problem to go fix. We can’t back up all the
way to the engine and start fixing the problems, it’s
too expensive. A lot of problems start to culminate
early in the design cycle -- it’s like an avalanche
going down a hill -- if we don’t break that avalanche
from coming down the hill, we end up chewing up all of
our NRE dollars, and things get really expensive. The
problem is everywhere and because we’ve got very complex
chips and because the caboose doesn’t talk to the
conductor, everything falls apart. What we hear people
screaming about is, ‘If I only knew about this earlier,
it wouldn’t have been a problem.’
Grebinski: DFM is more of a tactical
issue when it comes to predictability. The
characteristics and behavior of silicon, and the
bandwidth in predictability aren’t the same as they used
to be, say, 10 years ago. So when an Intel or an IBM
comes up with a new architecture at the transistor
level, the bandwidth to be able to predict what will
happen with that particular transistor is very narrow.
That causes problems for the EDA community because it is
a single-variant, non-statistically based design
environment that doesn’t take these kinds of factors
into consideration. DFM addresses a lot of these issues
-- it gets in there and says, ‘This is what we can
predict, but we need this much to do a real design.’ How
do we build mathematical algorithms to bolster what we
get from an IBM or an Intel? There is a lot of internal
development going on because the issues are so
complicated at that process level, that companies like
IBM are building these tools before they license them to
anyone else. The IBMs of the world are already at 45nm
node and are working beyond that, but they are not
giving that technology to the EDA community. From a DFM
perspective, it is really important to get the
predictability to work very closely with these companies
and to have strategic partnerships to the point where
the trust is there enough so that they will share what
those secrets are, and how those predictability factors
play out even in that short bandwidth.
Electronic
News: Guard-banding
is often used as a way to mitigate manufacturing
defects, but many companies have learned the hard way
that this strategy won’t work for 90nm and below. What
is the alternative to using these rules?
Deo: The way you do that is by not
using rules. It’s the rules that cause the problems
because they don’t predict what is happening in the
process. That’s why model-based analysis has to be the
way to go. Models have to be built so that they are
closer to reality since they depict the reality of the
process a lot better than rules. Once those models are
built, they can then be used upstream. The way they are
used at the routing stage, will not be the same way they
are used at the synthesis stage but there is some
information that can be boiled down to simple cell
selection and determine if a particular cell is a
problem.
Electronic
News: Since the
models have to be based on real manufacturing data, is
this kind of development possible by the EDA vendors or
will it happen downstream at the manufacturer?
Deo: The IDMs have had this kind of
model-based analysis for years. The only thing is that
they have not given it to anyone outside. That’s why
there are new models of EDA coming in and there is a
methodology change in bringing these models and
model-based tools out to the market.
Electronic
News: If customers
have these capabilities, what can the EDA vendors bring
to the table?
Raghvendra: If you think about any new
development in EDA, it’s always been pioneered by the
leading IDMs. In fact, I would go as far as saying that
if somebody is not already doing it, it’s not worth
doing. So, it’s not unprecedented that we will have to
go in after they’ve actually done something, because it
comes down to ROI. They have to have a captive team
doing this, and in the days of expense control and
watching every dollar, sooner or later they will realize
that the EDA tools are at a point that it makes sense to
use them. We just have to prove to them the ROI is
better giving us those dollars than doing it themselves.
Electronic
News: Does the EDA
industry have what it takes to reach $7.5 billion by
2009, as predicted by Gartner Dataquest?
Deo: Yes, and the way it will be
realized is not through technologies; business models
have to change. The way the business models are today,
it is never going to get to $7 billion -- heck, it is
never going to get to $5 billion. If three vendors are
fighting for a piece of business, customers know how to
play vendors against each other.
Electronic
News: If this
industry is truly an outsourced business, doesn’t the
customer know what the value is to them?
Naum: A lot of customers actually don’t
understand what the value is. What they think is that
EDA software is as cheap as printing a CD. What’s not
realized is the size, headcount and the amount of money
that is poured into developing an EDA tool. The problem
is that we are not just any outsourcing business, we are
outsourcing times a thousand because we are solving
everyone’s problems.
Electronic
News: Isn’t the
responsibility then on the EDA companies to show the
value they bring?
Naum: It is, if we get out of the
pricing wars. We talk about all these changes to make
things happen. If we, as an EDA industry, keep selling
tools for the price of maintenance, to keep the other
guy out, how can we pay to develop this entire solution
set, if our gross margins go down to nothing? Not even
giants like Synopsys or Cadence will be able to do it.
Grebinski: I’m not so sure that the
model that exists today will change. I think the
companies that have internal CAD development capability
will continue to build their own tools for the first
go-around until they have them stable and then they will
license them out as a way to build the rest of their
products. How close the EDA community is to those
customers that build these new transistors and circuits
will determine how successful those companies are. I
think that’s where the value-added is for the EDA
community.
Electronic
News: What is the
best way for the EDA industry to innovate for the
methodology shift?
Deo: It comes down to manufacturability
or getting the silicon information into the design flow
-- that’s where the opportunity is. The reason the whole
methodology shift is happening is because of the
complexities of 90nm and 65nm, but more importantly
because there were no other tools available. All the
IDMs were doing their own tools but there were no
commercially available tools or technologies. What has
to happen now that we are building this bridge between
design and manufacturing is that the EDA industry has to
learn to deal with the manufacturers. Until now, we
dealt with them in a one-way fashion -- they gave
information and we used it. Now we have to deal with
them in a two-way fashion; we have to have interactive
data information infrastructure going back and forth,
creating a silicon infrastructure, and then making that
platform available for the design teams.
Grebinski: I would reiterate the
significance of partnerships with the device
manufacturers. Back in the early 80s, when the device
manufacturing side of the business -- the ones who were
working in process -- were suffering through similar
types of problems, the equipment industry went off and
did their own thing and left the device manufacturers
behind. Then, all of a sudden, the device manufacturers
came back and said, ‘Hey wait, you are going off into an
esoteric area. Come back and let’s work more closely
together.’ There was a lot of consolidation back then
with the equipment companies because not every device
manufacturer can work with every equipment company out
there. Critical to survival of the EDA community is the
relationship with the customer and also recognizing that
the EDA industry does not solve problems as do the IDMs
that are building these tools internally; and that
basically, the EDA community is a
manufacturing/outsourcing arm.
Naum: We definitely have to partner up.
If we think we can invent it ourselves, we are fooling
ourselves. We are looking at DFM avoidance while we wait
for the models to come up and collide in the middle.
Raghvendra: I think the way to think
about it is to look at our customers’ issues; that’s
where you start from. The insight we’ve had is that what
our customers do has changed over the years. It used to
be our customers were mostly designing for fast
computers, and supercomputers and all sorts of whiz-bang
things, which were fewer in numbers but nicer in margins
and they could afford to pay their suppliers well.
What’s happening now is most of them are going into
consumer electronics where cost is king. You have to
empathize with them that their costs are paramount and
think about what it takes for them to be successful. It
boils down to three things: cost, predictable time to
results, and quality of results, which means, if you say
it is going to be 130MHz PCI bus, it had better be. If
we get to the point where we internalize what our
customers are doing, which is serving the market where
every penny matters, and give them the tools so they can
serve their customers well, then perhaps we can partake
in the feast.
