September 26, 2002
New Photomask Data Format
Takes Aim at Rocketing Mask-set Costs
By J. Robert Lineback
The Semiconductor Reporter |
|
|
New Photomask
Data Format Takes Aim at Rocketing Mask-set Costs
One of the biggest bottlenecks in photomask
design and production is about to be broken
open wide open--or at least that's the hope
an industry taskforce representing nearly 30
IC manufacturers, tool companies, electronics
design automation suppliers, and mask-making
shops.
On Monday, at the annual BACUS Symposium on
Photomask Technology in Monterey, Calif., a
new 64-bit capable version of data-file interchange
between design and mask-production systems will
be rolled out by the IC Design/Photomask Data
Path Taskforce of the Semiconductor Equipment
and Materials International (SEMI) trade group.
The new format will be proposed as a replacement
for the industry's widely employed but aging
16-bit GDSII Stream format.
According to SEMI's taskforce leaders, GDSII
has finally reached the point of being "intolerable"
for many tools and systems now used to produce
mask sets for complex ICs.
"A number of constraints that people used
to live with are now becoming real limiters,
in terms of cost of ownership and productive
output for a number of different kinds of tools--not
only mask writers and inspection systems, but
also design automation tools," said Tom
Grebinski, who is SEMI's Data Path Taskforce
chairman.
Cobbled together and limping along
A major problem with GDSII is also the fixes
and tweaks that many companies are adding to
help overcome the problems of 16-bit files doing
the work of what could be 64-bit computing.
"The problems that have come up in the
last several years relate to interoperability
issues or data inefficiency issues between tools
and between design and manufacturing,"
warned Grebinski, who is also in charge of strategic
partnerships and alliances for mask pattern-generator
supplier Micronic Laser Systems AB.
"We use a lot of different kinds of formats
to translate different languages to be able
to represent data in different ways… Tools
are not able to communicate with each other
without other translators. The whole thing has
started to become a big bottleneck," he
said.
Some industry leaders are hoping the new format--the
name of which is still being kept secret until
Monday--will not only break huge bottlenecks
in transferring finished IC designs to photomask
shops but also reduce time and costs in preparing
files for reticle production. The new format
could also increase the mask-writing time of
tools while lowering the overhead of file storage,
according leaders on the SEMI taskforce.
When the effort started in 2001, SEMI's taskforce
and chip consortium International Sematech in
Austin, Tex., found single GDSII Stream files
that were 50 gigabytes in size or greater.
"Our goal was to achieve at least an order
of magnitude in the reduction of the file size
for the same data that is represented in the
GDSII stream format," explained Kurt Wampler,
chairman of the SEMI Data Path Taskforce's new
stream format working group. "We have done
quite a bit of benchmarking with live data from
a number of sources, and we have achieved or
exceeded that goal for a great number of cases,"
said Wampler, who works for ASML MaskTools Inc.
in Santa Clara, Calif.
SEMI's task force hopes the new format will
eventually increase efficiency and result in
data compaction ability that's 10-50 times greater
than today's GDSII Stream format. The group
will introduce the 1.0 draft of the new format's
specification at the symposium in Monterey,
and SEMI's members will vote on the proposal
during the next several months.
Fighting photomask inflation
One hope among backers of the new format is
to target the rocketing costs of photomasks,
which are already above $600,000 per mask set
for 0.13-micron process technologies and pushing
beyond the $1 million barrier in the next 90-
nm process node, according to many experts.
In fact, the price of system-on chip (SoC) photomask
sets could reach $1.5 million, according to
Shang-Yi Chiang, senior vice president of R&D
at Taiwan Semiconductor Manufacturing Co. Ltd.
(TSMC).
At the $1-to-$1.5 million price tag for completed
photomasks, many IC engineering managers are
becoming more cautious and apprehensive about
releasing new designs for tape outs, Chiang
noted during a panel discussion at Semicon Taiwan
last week in Taipei. If the design doesn't work
or yield well in production, "they could
lose their jobs," he said.
While there are many different issues at play
in mask costs and wafer-fabrication yields,
some industry experts believe one obvious fix
is updating GDSII, which has its roots in the
old Calma Corp. layout workstations in the early
1980s. The GDSII format was considered a proprietary
jewel of Calma, which was acquired by General
Electric Co. in the 1980s and then later purchased
from GE by Valid Logic. Eventually, Cadence
Design Systems Inc. in San Jose gained ownership
of GDSII when it bought Valid Logic.
"Cadence owns the copyrights to GDSII Stream
format, but it has pretty much dropped all of
the access restrictions to the format,"
noted SEMI working group chairman Wampler. "In
the 1980s, it became popular and filled a need
for hierarchal data interchange in spite of
the proprietary attitude at Calma. That gradually
relaxed… Now you can find it on the Web
if you want to see how GDSII files are written."
Open from the start
The openness has been one aspect that SEMI's
task force wanted to maintain in the next-generation
replacement of GDSII. "What we are working
on will be an open standard form the get-go,"
Wampler said.
Also from the get-go, the new stream format
will be flexible, based on the current draft
of the 1.0 specification. While the proposed
new format has been expanded to accommodate
64-bit computing requirements, it is also designed
to be "32-bit safe."
"We have removed all of the restrictions
[16-bit fields in the data structure of GDSII
and some 32-bit limits in the addresses]. Any
pointer, integer or structural field value can
exceed 32-bit when its needs to, but all of
the integers in the format are adjustable or
variable in width," Wampler explained.
"They only use as many bytes as they need
to represent the data."
The primary challenge in creating a GDSII replacement
candidate was technical, and for the most part
the 30 companies represented in the taskforce
efforts worked well and quickly together.
"I was surprised as to how much agreement
we had because we had people from integrated
device manufacturers, tool manufacturers, EDA
companies, and the merchant mask makers,"
Wampler told The Semiconductor Reporter. "Each
of those representatives brings somewhat of
a different perspective to the problem, but
we all learned through the process and I think
everyone is pretty much happy with the results."
