( ESNUG 453 Item 2 ) --------------------------------------------- [03/01/06]
Subject: Mentor, Cadence, Synopsys, Magma, PDF Solutions
Q: As a competitor, what do you think about these specific DFM companies?
In plain English, what exactly do they do? And how do you view them?
Mentor
Mike -- DFM is all about the link between design and manufacturing. You
must fuse these two disciplines together to improve the quality of your
chip. Mentor has strong backend mask data prep tools, but no viable
strategy to link those tools into your design flow. They're DFM orphans.
Their new Calibre LFD appears to be a yield scoring tool that uses their
old Calibre OPC engines (post-GDSII). That approach won't work because
manufacturing tools are too slow and too cumbersome to be used by the
designer. This is just repackaging their old tools from what I can see.
They're only protecting their installed Calibre base. It's not clear
how Mentor will add value upstream, in the design creation flow.
Atul -- Mentor acquired OPC Technologies in 1998. Mentor is the most
aggressive in DFM. They have a superb Calibre platform for DRC and also
RET/OPC but it is all post-GDSII. Any post-GDSII tool is doomed to
failure for DFM. They take days per layer; far too long. These tools
can't generate the guidelines to fix the problems they find. And they
require waaaay too much information (i.e. the design specific OPC recipes
that none of the fabs want to give out.) Their new Calibre LFD is not
going to be fast enough, or handle full chip or large blocks. In order
to get any accuracy, Calibre LFD will need the fab's OPC recipes which
the fabs won't release.
Cadence
Mike -- Lithography-aware design is a good idea. Designers must begin to
understand manufacturing effects (like litho variation) earlier. I
believe the source of the information must be tied to production-proven
tools and accurate process-calibrated models. I don't see much evidence
of that part at Cadence. There's a blind spot in the Cadence view. The
fabs aren't using Cadence OPC nor Cadence litho simulation (because Cadence
has neither) so there's no way that Cadence can provide accurate DFM info
calibrated to a specific process to the designer.
Atul -- Cadence is the dark horse of the big EDA vendors. Currently they
don't play in the RET/OPC space, however Fister is not afraid to take
risks. He understands that the leverage comes from doing things
predictively on the *design* side and not in simply trying to gravitate
OPC/CMP upstream in some version of its current form.
Synopsys
Mike -- Of the Big 3, Synopsys has the largest collection of tools to
address DFM. I don't see the required integration, front to back, however.
Linking TCAD, mask data prep, yield analysis and design creation into a
unified product offering would be great; when will we see it? I don't see
Proteus and SiVl data feeding PhysOpt. That would be pretty powerful if
they could pull that off, but I don't see it. With a large company that
has a large existing install base, it's hard to make all the changes across
the spectrum of tools to do good DFM.
Atul -- Synopsys acquired Avanti, Numerical Tech, and recently HPL. These
are the basis of all their RET/OPC revenue. Like Cadence, Synopsys has
all the implementation and analysis tools in their entire flow also, but
unlike Cadence their hands are full protecting their current post-GDSII
turf from Mentor, Magma, and the start-ups. I disgree with Mike; it's NOT
a tool integration issue -- you just can't connect data from Proteus
(post-GDSII) back into PhysOpt (pre-GDSII) -- these worlds just can't mix!
Magma
Mike -- The physical verification product from Mojave is a good start.
There's still a lot to do. They lack OPC, litho sim, critical area
analysis (random defects), parametric variability, and a strategy that
builds this into one flow. It can't be tacked on piecemeal.
Atul -- Magma is currently #4 in OPC in a 2-horse race? They definitely
have their work cut out for them. Are they attacking the OPC market or
are they addressing the DFM in their design flow? From what I read now
they are trying to go after the OPC market first. If it were up to me at
Magma, I'd go after the whole DFM problem instead of only the OPC part.
PDF Solutions
Mike -- PDF lets you harvest the manufacturing data so you can understand
yield and DFM issues. It was a smart move for PDF to team up with Magma
because it gives them access to design implementation. It was smart for
Magma to team up with PDF because it give them access to manufacturing
intelligence. People believe in EDA tools that are calibrated to real
manufacturing data. That's why they're on my watch list.
Atul -- I disagree with Mike. I don't know which Magma tools have been
calibrated to any manufacturing data. I think there are none. They're
only trying to impact the random yield problem through better standard
cell selection. I think this is too cumbersome for a designer to deal
with and its low impact. It doesn't address any of the systematic DFM
variation issues. PDF generates almost all their revenue in the fab.
They're not on my watch list because their technology really doesn't
go upstream well.
Index
Next->Item
|
|
