September 27, 2002
OpenAccess API Could
Link Design, Manufacturing
By Richard Goering
EE Times |
|
|
OpenAccess API
Could Link Design, Manufacturing
MONTEREY, Calif. — A powerful coalition
of chip makers, mask makers, equipment providers
and EDA vendors will gather here Monday (Sept.
30) to call for a single design-through-manufacturing
data model based on the OpenAccess application
programming interface.
While holding on to that long-range vision,
the coalition will also bring out an immediate
replacement for the venerable GDSII layout
format. That new format promises a tenfold
data reduction, according to the Semiconductor
Equipment and Materials International (SEMI)
IC design/photomask data path task force,
which represents some 30 companies. But the
group's more ambitious — and controversial
— mission is to create a "universal
data model," or UDM, that can extend
from design all the way through photomask
and device manufacturing.
Backers say that together, the new layout
format and UDM can save billions of dollars
in mask-making and manufacturing costs, perhaps
permitting the design and manufacturing of
sub-90-nanometer devices that would otherwise
be prohibitive. Moreover, the UDM promises
to bring manufacturing data back to chip designers,
who must start to account for process variations
at or below the 90-nm node.
Skeptics note that OpenAccess, based on
technology developed by Cadence Design Systems
Inc., still lacks broad acceptance in the
electronic design automation community —
and would need a lot of work to serve the
needs of manufacturing. SEMI's endorsement
throws considerable weight behind the nascent
standard but potentially puts Cadence competitors
such as Synopsys Inc. and Mentor Graphics
Corp., which have yet to embrace OpenAccess,
in an awkward position.
In the end, economics may drive the decisions.
"What's happened is that our data inefficiencies
have reached a $4 billion to $6 billion annual
cost," said Thomas Grebinski, chairman
of SEMI's data path task force and a contractor
at Micronic Laser Systems AB. "We think
we can knock that down quite a bit."
A UDM may be what's needed to continue down
the silicon process road map, said Steve Schulz,
newly appointed president of the Silicon Integration
Initiative (Si2), which oversees the OpenAccess
Coalition. "It the UDM could make economically
feasible advanced silicon processes that would
otherwise not be feasible, because yields
would be too low or the time to test too long,"
he said.
A UDM that links design and manufacturing
may take a long time to unfold, but the GDSII
replacement format — unnamed as yet,
but initially called New Stream Format (NSF)
— should be available on an open-source
basis by March, Grebinski said. And there's
no controversy here: Cadence, Synopsys and
Mentor Graphics all plan to quickly support
it.
Toothache, headache
The SEMI task force was launched in 2001
to tackle serious data inefficiencies in the
design-to-manufacturing link. Mitch Heins,
vice president of IC manufacturing and strategic
relationships at Cadence, said the task force
quickly found it was facing a "toothache
and a headache."
The toothache, Heins said, was the need
for immediate relief from the cumbersome GDSII
format — hence the creation of an NSF
working group. The headache is the desire
to provide more "intelligence" than
the polygons provided by either GDSII or its
successor. That's where the UDM comes into
play, and so the SEMI task force also created
a UDM working group.
Generations of chip designers have taped out
designs using the GDSII format, developed
by Calma in the mid-1970s. But GDSII files
are becoming unwieldy, especially with the
addition of resolution enhancement technologies
(RETs) such as optical-proximity correction,
which exponentially drives up the data requirements.
"Some of our members have told us they're
dealing with files as big as 50 Gbytes, and
there's no end in sight," said Kurt Wampler,
chairman of the SEMI task force's NSF working
group and distinguished engineer at ASML MaskTools.
"The GDSII stream is extremely inefficient
in the way it handles geometric figures —
it takes 64 bytes to represent a single rectangle,
which is extremely costly."
The new format, Wampler said, is routinely
showing a 10x reduction in data size, although
in some cases the figure may be much higher.
It also removes a number of GDSII restrictions,
such as 16-bit and 32-bit fields, and provides
a 64-bit-capable format. Wampler said it's
considerably more flexible in terms of cell
names, properties, the number of vertices
in a polygon and many other areas.
One
of the most important aspects of the new format
is that it may reduce or eliminate the conversions
that usually take place after GDSII is created,
such as fracturing into a pattern file format
such as Mebes. Peter Buck, engineering manager
at DuPont Photomasks, noted that today, mask
shops generally receive data that's already
been fractured — data that loses the
hierarchy information in GDSII.
"With the new format, fracturing can
move back to the mask shop," Buck said.
"It just removes one of the steps in
the process that appears to be redundant."
Motorola Inc. is seeing a 10- to 20-times
reduction in data size from the new format,
said Al Reich, the company's manager of process
and optics enhancement technology. Moreover,
he said, the new format can carry some non-geometric
information. "We would like to tag things
with information to improve RET processing,"
Reich said.
Wampler, the NSF working-group chair, said
that chip designers won't be affected by the
new format and that all EDA vendors need do
is develop readers and writers — no
more complex than supporting GDSII. Si2 will
announce an OpenAccess reader/writer for the
new stream format, but any database can support
it.
Karen Bartleson, director of quality and
interoperability at Synopsys., said her company
will "absolutely" support the new
format. Joseph Sawicki, general manager of
Mentor Graphics' design-to-silicon version,
said that Mentor will announce Calibre support
for the new format on Monday. Sawicki noted
that Mentor's own Standard Layout Format specification
provided a foundation for SEMI's new stream
format.
SEMI's NSF working group will present its
"best effort" on the new format
at the Monday meeting, said Grebinski, but
it will then need to be approved by the entire
SEMI task force. After that step, it goes
to the International Standards Committee for
Microlithography, and then back to SEMI for
approval as a real standard.
The SEMI task force's UDM working group
that Grebinski heads will announce its endorsement
of OpenAccess on Monday. "What we're
doing is adopting the OpenAccess format and
trying to bring it downstream into photomask
manufacturing and then into device manufacturing,"
Grebinski said. He acknowledged, however,
that OpenAccess "needs a lot of work"
to get there.
It's important to note that SEMI is not
calling for a single, unified database that
would serve design and manufacturing. That
could happen someday, Grebinski said, but
for now, the mission is to transfer data with
a consistent data model, using the OpenAccess
API to massage any differences in formats.
As Si2 and the OpenAccess coalition have long
maintained, any database could theoretically
be accessed underneath the OpenAccess API.
But first, Grebinski said, OpenAccess needs
acceptance in the EDA community. "The
second step is to start developing the new
stream format with OpenAccess and open up
the data pipe that pushes OpenAccess down
into manufacturing," he said. The new
stream format, he noted, will still be needed
for geometric data.
A UDM can tell a mask shop what's important
about the geometric data. Today, said Cadence's
Heins, mask shops just get a bunch of polygons
— resulting in "really stupid things"
like trying to fix a defect in a logo or an
area fill. What UDM can do, he said, is bring
design intent into the manufacturing world.
"The way I view that UDM paradigm,"
said Motorola's Reich, "is that we wouldn't
be passing circuit designs down an assembly
line. Rather, it would be a repository that
each group would operate on using transactions.
So all of the data that's available at design
time would also be available at mask-making
time."
Perhaps most interesting to chip designers
is that, with the UDM, data can flow from
manufacturing to design. With a "common
pipe for parasitics," Grebinski said,
designers will have access to process information
for the first time, and they'll be able to
make intelligent decisions about process variations.
"Lithography groups are already placing
restrictions on designers," said DuPont's
Buck. "At some point mask making is going
to place restrictions on designers as well.
UDM allows the transfer of information, both
upstream and downstream, so that designs are
both lithography- and mask-friendly."
'Just a vision'
There are, however, skeptics. Gary Smith,
chief EDA analyst at Gartner Dataquest, said
OpenAccess doesn't have much chance of going
beyond a Cadence standard in the 130- to 90-nm
semiconductor generation. A standard like
UDM may be "essential" at 70 nm,
Smith said, but SEMI must be careful to avoid
"tying two standards together."
Synopsys' Bartleson described UDM as "still
just a vision and a proposal. Since it's so
long-term, there's no immediate impact on
Synopsys or its customers." She termed
the creation of UDM a "daunting"
task and said she'd personally rather see
multiple databases, with bridges that allow
for communication of data.
Mentor's Sawicki called UDM an "interesting
concept," but said it faces "significant
barriers" in terms of organizational
dynamics. One hurdle is intellectual-property
protection. Some people don't even want to
send hierarchical GDSII data to external mask
shops, let alone design data, he said.
"UDM right now is like the joke about
the four blind guys around the elephant,"
Sawicki said. "What you think it is depends
on where you're standing."
