AMD,
UMC to codevelop advanced process control technology for 300-mm
chipmaking
The
world's second-largest manufacturer of microprocessors and the second-biggest
foundry have set their sights on establishing industrywide 300-mm
standards for advanced process control (APC). Trading on their respective
strengths in the technology, AMD and UMC plan to develop APC methods
that both manufacturers will put to the test at a fab the companies
will operate together in Singapore.
The chipmakers hope to
take advantage of APC's real-time monitoring capabilities to reduce
the impact of yield-killing errors in expensive 300-mm processes.
The jointly run Singapore fab, Au Pte., is scheduled to begin production
in 2005. UMC also plans to use the new technology at its Fab 12A
in Tainan, Taiwan.
AMD has considerable experience
with the technology, having begun using it in 1996 for high-volume
production at Fab 25 in Austin, TX. In addition, the chipmaker's
Fab 30 plant in Dresden, Germany, features APC-based processes that
combine real-time closed-loop control and run-to-run control to
minimize process drift. The technology's ability to detect and classify
faults acts as a warning to shut down process equipment before significant
damage can occur to yields, AMD points out.
UMC says its foundry operations
have led the world in the adoption of 300-mm fab automation. "By
leveraging our combined manufacturing expertise, we can accelerate
the proliferation of advanced control technologies required for
automated wafer fabrication," says Chris Chi, senior vice president
of fab operations and president of UMC. A primary goal of the collaboration
is establishing a new standard for automated wafer fabrication,
Chi points out.
The partners will put together
a budget for their collaboration next year, says Tom Sonderman,
director of advanced process control for AMD in Austin. The companies
plan to work together "immediately" in Tainan, he notes. That work
will be transferred to the Singapore joint venture.
The budget for the joint
effort "encapsulates a lot of things," Sonderman emphasizes. In
particular, the APC infrastructure encompasses several elements,
including manufacturing execution systems. "There's a lot of peripheral
infrastructure that allows the tools in the fab to provide the necessary
information for control. What rides on top of that is our APC framework
together with technology provided by KLA-Tencor through its product
called Catalyst, plus the cost of putting all that in place."
For a typical 200-mm fab,
the cost of installing APC technology "is in the $3 million range,"
Sonderman says. An APC setup for a 300-mm factory "is some percentage
higher than that."
AMD requires "APC-enabled
functionality" on critical tool sets from manufacturers such as
Applied Materials and Novellus, he notes. Key functions are data
acquisition and dynamic recipe management, "which means you have
to be able to modify the recipe at the tool level on a lot-by-lot—or
for 300-mm [processing]—a wafer-by-wafer basis." Most equipment
suppliers offer that functionality, Sonderman says, but he adds,
"a lot of the debate from the standards point of view involves the
right way to do it.'"
The software used with
the APC hardware holds equal importance. Sonderman says AMD has
an APC-baseline technology, "which is a configured version of the
Catalyst product I mentioned. Over time, AMD has built more capability
into that control system." As for the software, the chipmaker faces
decisions on "what to bring back into our commercial products."
AMD distributes the outcome through a "third-party–slash–internal-development
strategy."
Sonderman offers two examples
of how AMD's manufacturing benefits from APC technology. One has
to do with drive-current control for microprocessor performance.
"We use a very complicated gate CD process that takes information
from the photo-step and postphoto metrology and etch-step and postetch
metrology, [and] pulls all this together in a complex algorithm
to tune the etcher and the stepper in order to achieve the desired
target that allows us to maximize the...speed grade of the microprocessor.
"Another example is in
the fault-detection of the rapid thermal anneal process, one of
the few processing areas where you don't have a physical measurement
you can make," he continues. "It's an off-line measurement. Multivariate
fault detection can be used to assess the tool health on a lot-by-lot
basis. After each lot we determine whether or not the tool ran as
it should. It's kind of like a virtual sensor."
Sonderman says there are
two ways to think about the partners' expressed goal of setting
the pace in 300-mm APC technology. "We want to come up with a standard
infrastructure for doing control in a 300-mm facilty. For AMD that
is less important than it is for UMC, which plans on building many
300-mm fabs. What they want to do is to replicate that [APC capability]."
AMD wants to "establish a highly automated facility...and all the
associated infrastructure and communication protocols so that our
solution can then become the de facto standard."
Sonderman
puts AMD's quest to lead the pack in another way. "We like to say
we don't want to become beta in a VHS world."