|
November 11, 2005
OASIS In the Valley
By Ed Sperling
Electronic News |
|
|
OASIS in the
Valley
Electronic News
sat down to discuss a new file format that spans
chip development to photomask creation, called the Open
Artwork System Interchange Standard (OASIS), with Bob
Gleason, engineering manager for Intel’s mask operation
automation and services; Carl Vickery, senior member of
Texas Instruments’ technical staff for design to
manufacturing and ASIC product development; Kurt
Wampler, distinguished engineer at ASML’s MaskTools
subsidiary; Richard Gladhill, in charge of advanced mask
data prep automation at Toppan Photomasks, and Thomas
Grebinski, CEO of OASIS. What follows are excerpts of
that discussion, held in front of a live audience at the
recent GSPx show in Santa Clara.
Electronic
News: Why do we need
OASIS? What’s the problem that has to be solved?
Gleason: It is one open format that
will replace a multiplicity of formats we have to deal
with today in tape-out and mask preparation. Today, GDS
stream is our primary format for data and hierarchical
form, as it comes through design. And as we go through
tape-out and mask data preparation, the file sizes begin
to explode. At that point, that’s where we’re no longer
able to deal with the large files and we’re forced into
a variety of proprietary formats. With OASIS, we have a
format that can represent both hierarchical and flat
data, so you have one format all the way through the
process. The other thing about OASIS is that it
eliminates a lot of the restrictions such as length of
cell names, number of digits used represent integers,
and so forth. We want to deal with those problems before
we hit those problems.
Wampler: Not only do GDS stream files
become extremely large and unwieldy, the format itself
breaks internally. There are a number of fields that are
16 bits or 32 bits, and design complexity has reached
the point where we’re pushing against or exceeding those
limits. One of the design goals of the OASIS format was
to completely relax all of those restrictions and allow
numbers and pointers in the files to be as many bits as
they need to be to scale to arbitrarily large amounts of
geometric information, regardless of hierarchy. We also
wanted to provide a richer information palette to carry
more information about the design than the GDSII stream
format was capable of carrying, along with the cells and
polygons.
Vickery: We also need to support a
richer set of data abstractions -- for example, tables
of contents and bounding boxes. These are things that
don’t naturally exist in GDSII and which typically
result in EDA software reading in GDSII data and
building its own internal data structures before it can
even start doing any useful work. By creating a format
that gives EDA software leverage over the data we can
cut through a lot of this re-formatting and
non-value-added translation and get a leg up on what we
pay EDA vendors for, namely the processing of the data
itself.
Gladhill: We were using the GDSII
format internally and having great difficulty doing the
very large files. By switching to OASIS, we were able to
reduce our disk consumption and improve our processing
speed. We’re able to work with data we weren’t able to
work with before because of the file sizes.
Electronic
News: Where and when
will OASIS become important?
Grebinski: The driving force behind the
development of the format was the need to develop a more
efficient mask layout capability. That includes mask
data prep, fracturing, being able to have available new
types of constructs that allow tools to run far more
efficiently than they do today, not only in terms of bit
efficiency but also in terms of run time. OASIS has its
value in mask data prep, layout and mask manufacturing,
but also upstream of design tape-out in place-and-route
and DRC and anywhere where there is a geometric element
in design. OASIS can replace these GDS-like constructs
and make that design flow into mask manufacturing far
more efficient.
Electronic
News: New standards
often take a lot longer than initial backers believe to
become widely accepted and adopted. What will happen
with OASIS?
Gleason: Our intent is to begin using
it in 2006. We will use it before we desperately need
it. The last thing we want is for a tape-out to crash
because we’ve run into an integer limit.
Vickery: Our timeline is also 2006
because 65-nanometer technology has begun to put a
significant amount of pressure on the data mask flow
that we didn’t see at 90 nanometers. If you like to roll
the dice at tape-out time, you can do that, but there
are some steps like adopting OASIS and some other
scalability issues that will let you sleep better at
night.
Electronic
News: How much will
this reduce time-to-market and potentially reduce the
cost of chip development?
Grebinski: I think the issue is
preserving time-to-market. The industry is driving in
the direction of greatly lengthened design cycles
because things were bogged down at the back end. Having
the OASIS format, which is an order of magnitude more
compact than the GDSII stream, increases the headroom
for design complexity to grow without incurring
proportional delays.
Electronic
News: One of the
biggest problems in a disaggregated industry is that
data isn’t being shared back and forth between the
foundries and the design houses. Does OASIS help solve
this, or any of the myriad technical issues we have to
deal with?
Grebinski: The format does allow a more
collaborative effort. It’s unprecedented in the
involvement of foundries, competing EDA companies and
mask shops. After we developed the format, that
collaborative effort continued. A lot of the
interoperability issues we had with GDSII will not be as
large as we had with OASIS, and our ability to
communicate with each other company-to-company will
improve.
Vickery: For us it’s just one more tool
in the toolbox. Its main redeeming value is that it can
let logistics and data representation recede into the
background of problems you have to deal with. If you’re
trying to stick a property on a piece of data to
communicate some information it has all the mechanisms
to do that. In GDSII you can spend quite a bit of time
and effort just handshaking between two tools to convey
a simple piece of information. It won’t solve any
problems unilaterally, but it can ease the pain.
Gleason: Problems with design are not
where OASIS is going to see it’s main application
benefits. It’s going to be in the processing of data
after design. The opportunity to communicate design
intent back to that process and improve it is enhanced
by things like the lifting of restrictions on data types
and layers.
Electronic
News: Does that
change with the compression of the back-end and
front-end processes?
Wampler: You’ve got to separate out
design methodology from the infrastructure that supports
that methodology. For example, computation geometry
algorithms are where GDSII can be improved. That helps
with these more advanced methods of design. From that
aspect, OASIS is an enabler and does allow for more
advanced designs.
Electronic
News: This is rev
1.0 of a standard. What comes next?
Vickery: The first generation of tools
that support reading and writing have not yet tapped all
the potential we put into the data standard to represent
information and to represent it compactly. I anticipate
several tool generations built on the 1.0 standard
before we tinker with the standard in any meaningful
way. That is a testament to the generality of the scope
of the standard and the wide spectrum of thought and
intellectual property that went into defining the
standard.
Gleason: This is a standard. The last
thing we want to see is are widespread revisions of
OASIS.
Electronic
News: What will the
real benefit of OASIS be from a technology standpoint?
Gleason: The big advancements in new
chips are going to come from design and process
technology, not from the data format. The advantage of
the data format is that it’s an open standard and it
gets around the restrictions that might inhibit us from
getting products out once they’re design. It doesn’t
solve those kinds of problems.
Vickery: By and large, OASIS solves
logistics and communications problems. They seem to be
persistent, but they’re not the fundamental things
designers struggle day in and day out.
Electronic
News: How many
companies are involved in OASIS now and how many need to
be?
Grebinski: The companies involved today
are the large IDMs, foundries, mask shops and EDA
companies. Europe is adopting OASIS as a region, which
means Infineon and the Crolles alliance. Also, companies
that have internal CAD tool expertise are adopting
OASIS. I can’t say we need more participants. What will
follow is that the second-tier companies that are a few
years away from 65 nanometers will adopt it in a couple
years. It’s really based on the node of the technology.
Wampler: I did an informal survey and
counted 15 commercial implementations of OASIS
reader/writer software in EDA, and another 7
implementations inside integrated device manufacturers.
There are many implementations under way and many
already released to the market that have undergone
extensive testing, so there’s a fair amount of adoption
already.
Electronic
News: Who gets the
real value? Is it the leading edge companies at 65
nanometers, because not everyone will necessary get to
that node and beyond?
Vickery: That leaves fewer of us to
slice the pie. That sounds like a really good plan.
That’s part of what happens at the bleeding edge. The
leading-edge foundries will deploy OASIS at 65
nanometers and 45, and as that gets proven out it will
start to filter back up.
Wampler: Any time a customer wants to
make a switch for a particular design from GDSII to
OASIS, that transition will be easy. Once they’ve made
that move, they’ll never go back to GDSII Stream with
that data.
Electronic
News: What’s the
genesis of OASIS?
Grebinski: It was started in 2001 by a
couple people in the EDA community talking with IDMs and
said, ‘This is something we need to do.’
Electronic
News: Will this help
the EDA companies from a revenue standpoint?
Gleason: In the short-term, it’s
another format the EDA suppliers will need to support.
We need EDA suppliers to supply us with software that
can handle something bigger than 32 bits. If they say
there’s an inherent limitation in the data format and
they can’t handle it, or they’ve run out of characters
for cell names, which has happened in the past, at some
point that limit will be in the data format itself. If
the EDA companies had to deal with that themselves,
rather than with the backing of Semi, the IC
manufacturers and the mask industry, I think they’d have
a hard time doing that.
Vickery: I don’t think you can charge much for this. But
it’s ubiquitous like air. You don’t want to live without
it.
Electronic
News: The mask
makers have become the punching back of the industry.
Will OASIS help control costs?
Gladhill: No. The best you can hope for
is that it will allow us to operate more efficiently and
handle larger amounts of data without investing in more
hardware and other computational resources necessary to
handle it. But it’s not going to reduce the need to
handle massive amounts of data. We’ll still need
resources to do the necessary computation. That cost is
not going to go away. At the same time, there are
studies indicating we are severely under-investing in
the mask industry, which would make mask prices a lot
higher than they are.
