The Evolution of Intel’s Copy EXACTLY!

Technology Transfer Method

Chris J. McDonald, Intel SEMATECH

Index words: copy exactly, technology transfer

Abstract Introduction
Semiconductor manufacturing is characterized by very com- Table 1 shows the typical technology transfer approaches
plex process flows made up of individual process steps, many used over the last ten years or so. At the 1.5-micron genera-
of which are built to very close tolerances. Furthermore, there tion, process flows were much simpler than they are today. A
are complex interactions in these process flows, whereby small band of technical experts would typically be employed
each process step can affect many other steps, and each final to orchestrate a successful technology transfer. Generally
device parameter might be determined by the results from there would be few ground rules. Since there is always a
many inputs. This level of complexity is increasing with each lengthy “certification” or “qualification” exercise to prove
new technology generation. Items that were once consid- product quality and reliability, the transfer from R&D to manu-
ered second-order effects, such as barometric pressure and facturing, or to a new factory, offered the opportunity to
ultra pure water temperature, are now important variables introduce improvements to the equipment and process. The
affecting process results. latest model equipment or even a new vendor might be cho-
sen. Process recipes could be changed to improve them. In
The costs of technology development and capital equipment
the case of an existing factory picking up a new process flow,
for production are very high and are increasing with each
changes were made to match existing processes and meth-
generation, thus making technology transfer very important.
ods to improve efficiency and productivity. Sometimes, a
Once a new process flow and product portfolio have been
wafer size conversion would even be made at the same time,
developed, it is essential that the technology transfer to mass
involving many changes. Overall, however, the number of
production take place as quickly as possible, without disrup-
variables was relatively small, which made it simple to trouble
tive quality issues, and with the highest possible yield. No
shoot any results that did not come out as expected.
time is available to debug new problems that occur during
the transfer.
The traditional technology transfer approach often allows
many equipment and process changes to be made. These
changes are intended as improvements in the process, or
they are for the convenience of the production factory, which
may be already producing other products. As semiconduc-
tor technology becomes more complex, these changes have
resulted in unforeseen problems that cause production start-
up delays and inferior results.

The Copy EXACTLY! philosophy and systems were devel-

oped [6] in order to minimize the time required for a technol-
ogy to be transferred and to ensure product quality and yields Table 1: Technology transfer strategies
are not compromised. The methodology has been improved
and refined, and has become an important element in Intel’s
overall manufacturing strategy [1]. This paper describes the For the one-micron generation, technology transfer started
Copy EXACTLY! methodology and the increase in technol- to get more complicated. A structured methodology was
needed, whereby each process step would be measured to
ogy transfer performance that it has brought about. Some
ensure it matched a target value or complied with a set of
side benefits of this methodology are also discussed.
specification limits [2] [3]. Most projects, however, only fo-

The Evolution of Intel’s Copy Exactly! Technology Transfer Method 1

Intel Technology Journal Q4’98

cused on matching device and final product parameters. As

long as these were correct, changes would still be introduced
as a part of the transfer process.
For the sub-micron generation, the above approach has had
its share of problems. There are many more process steps
today, and many of them are made up of several components.
For example, a typical metal or dielectric layer is now a sand-
wich of multiple layers of different materials and composi-
tions. Very fine device structures are subject to different
effects, such as inter-layer stresses and adhesion. Phenom-
ena that were once considered second-order effects now have
a significant effect on the process result. Among these are
barometric pressure and ultra pure rinse water temperature.
In general, the process is manufactured with much closer
tolerances, increasing the importance of process control. Even
Figure 1: Traditional technology transfer method for the
the length of an electrode cooling hose has had a catastrophic
1.0-micron generation
effect, but this is a very subtle variable to find. With larger
die sizes, defect control becomes even more critical, and the Eventually the yields did converge. The second and third
way the process is actually run becomes a more important factories, which were already manufacturing different pro-
factor. An example is preventive maintenance intervals and cess flows, made many changes to the process and equip-
workmanship. Many of the factors just mentioned are diffi- ment and used their existing manufacturing methods. It took
cult to measure and quantify, which makes them dangerous several months of task force actions to catch up, by which
“unknowns” during a technology transfer project. When time the first factory was moving further ahead. In effect, the
many changes are made, the risk of something going wrong same learning curve was repeated independently by every
is greatly increased. Moreover, if something does not come factory. For the 0.8-micron generation, a new factory start up
out as expected, the number of variables that have to be and a new product introduction were delayed by three months
studied when trouble shooting the problem is greatly in- while a device reliability problem was solved, and it took
creased. The amount of experimentation and therefore, time, approximately one year to obtain equivalent yields [4].
required to find the problem increases as a power function of
the number of variables involved. If the problem is a Method
showstopper, for example affecting the product reliability,
For sub-micron technology, it was realized that a fundamen-
the end result is a costly delay. Even if this is not the case,
tally new approach would be needed in order to accomplish
yields may be depressed for an extended period.
an “order of magnitude improvement” in the effectiveness of
Figure 1 shows one example of using the traditional approach technology transfer. The Copy EXACTLY! philosophy and
for the 1.0-micron technology generation. The die per wafer systems were developed [5] [6] for the 0.5-micron generation,
yield is one of the most important variables in wafer fabrica- and they have become a key part of Intel’s manufacturing
tion and is used in this example. (The graph is normalized for strategy [1]. The capital letters, underline, and exclamation
die size). Other factors affecting product performance or point emphasize the paradigm shift that is required to trans-
manufacturing efficiency showed similar trends. The first fer technology using this method.
production factory in this example, which was a brand new
facility, obtained results that reasonably matched the parent Copy EXACTLY! Philosophy
R&D line. You can see the yields improve further as improve- Stated in its simplest form, “everything which might affect
ments were made and the organizations moved down their the process, or how it is run” is to be copied down to the
learning curves. Note, however, the divergence at the end of finest detail, unless it is either physically impossible to do
Year 2. The technology transfer results were good [2]; how-
so, or there is an overwhelming competitive benefit to intro-
ever, yields diverged as the R&D line focussed on yield im-
ducing a change. This philosophy differs greatly from the
provement, and the manufacturing line concentrated on in-
traditional method. In practice, there are always some issues
creasing volume.

The Evolution of Intel’s Copy Exactly! Technology Transfer Method 2

Intel Technology Journal Q4’98

that crop up and prevent an exact copy being made, so it was • Secondly, data is collected at the process step output
important to provide for making some changes in a controlled level on parameters such as film thickness, resistance,
fashion without opening the field too much. For example, in composition, etc., and they are compared to results at
the R&D site.
Europe the supply voltage and frequency can be different
than those in the U.S., so these had to be accommodated. • Thirdly, a comparison is made at the module level, using
Moreover, engineers are typically trained and rewarded for test structures such as oxide capacitors and metal ser-
making improvements, which in the semiconductor industry pentine patterns.
implies orchestrating change. Even the educational system • Fourthly, the actual product characteristics are measured
stresses independent work, and copying is seen as cheating. and matched.
Making a philosophical statement is obviously much easier Formal statistical tests are used at each level. If the match
than implementing it within a large team of R&D and manu- passes these tests, then we proceed to the next level and so
facturing engineers. Therefore, a comprehensive set of sys- on. If the match does not pass the tests, the root cause must
tems was put in place to ensure it would be implemented, and be found and eliminated. If it can’t be found, trouble shoot-
this set of systems is discussed in the next section. ing occurs to find out which of the previous level inputs is
responsible because, despite best efforts, something may
Systems have been overlooked. It is vitally important to avoid the
The systems that were implemented are as follows: temptation to make a compensating adjustment. Due to the
complexities involved, an adjustment may result in an inter-
Four level matching: Traditionally, it has been considered
acting parameter, possibly something not measured, being
acceptable if the final product parameters are matched be- mismatched.
tween the R&D line and manufacturing. However, the Copy
EXACTLY! approach requires four levels of matching. These A change control system: Most factories have some kind of
are illustrated in Figure 2. approval process for making changes to a production pro-
cess, either in the form of a sign-off list or a formal change-
control committee. Generally there is some kind of record of
the data showing the benefits of the change. The R&D line
continues to make improvements to finish off the technology
development and, in many cases, they may also run some
level of samples and production output. With Copy EX-
ACTLY! change control is started before technology trans-
fer, and all changes are implemented directly into both the
R&D and production lines within one week, or according to
an approved schedule. The pace of R&D work is not allowed
to slow, so careful planning is required to ensure the new line
is ready to accept the changes in real time. Any engineer
from the manufacturing line who has a good idea for im-
Figure 2: Four-level matching provement is encouraged to pursue it. The only difference
from the traditional approach is that the idea must be imple-
mented simultaneously at all sites. The change control board
• Firstly, the physical inputs have to be matched. These is responsible for the smooth operation of the system, which
are the energies and materials supplied to the process includes ensuring that the additional requirements do not
chambers: for example, gas flows, temperatures, pres- slow down the rate of improvement.
sures, RF power, and so forth. These might be supplied
Equipment difference form: In the Copy EXACTLY! system,
to the equipment by external sources or be generated
each first piece of equipment in the new factory or on the new
within the equipment itself. Everything about the equip-
process flow in the existing factory is treated as a change,
ment and its installation must be an exact copy down to
subject to change control. Audits are conducted and an
the diameters of piping and the number of bends, board
Equipment Difference Form is prepared from each. This form
revisions, software, etc. The settings for these param-
documents the actual difference, what risks it might pose,
eters and anything that might affect them are copied.
and the corrective action plan. This is formally reviewed by
Standards are generated to allow measurement and com-
management.
parison, and the values are measured and matched.

The Evolution of Intel’s Copy Exactly! Technology Transfer Method 3

Intel Technology Journal Q4’98

Supplier education: Equipment and materials’ suppliers are mittees to set the strategic direction and manage the technol-
constantly improving their products in response to demands ogy. For the 0.35- and 0.25-micron technology, this has been
from the semiconductor industry for improvement. These expanded: a Steering Committee at the plant manager level
changes are still desirable; however, with Copy EXACTLY! sets the overall direction, Joint Engineering Managers’ Teams
they are first introduced into the R&D line and from there manage the technology, and individual Joint Engineering
transferred to production. The suppliers are a vital part of Teams work the details at the process and equipment level.
the technology transfer and need to be thoroughly educated Similar structures exist in other areas, such as Manufacturing
on the new concept and systems. Operations and Automation, and a “Joint Synergy Team”
manages the overall system.
Audits: An audit is a formal procedure whereby engineers
from R&D and from production audit both lines. These au-
Results
dits are required and scheduled as part of the technology
transfer and are ongoing for a period thereafter. A report is Figure 3 shows some typical results obtained from Copy
written for each audit, detailing plans to correct all differ- EXACTLY! Two new factories were successfully brought on
ences found. line with the same yield results as the parent R&D line. Fur-
thermore, all three lines were able to improve their yields
Joint specifications: Since the equipment, process recipes, together by implementing improvements simultaneously.
and procedures are all the same, there is no reason why the Other parameters such as product quality and reliability and
documents provided for training and manufacturing opera- manufacturing efficiency also matched very closely.
tions cannot be the same. These are not copies; they are the
same documents, either paper or electronic.
Questions and answers (Q&A): Different engineers tend to
interpret the Copy EXACTLY! message in different ways.
For example, some engineers might say, “Surely if I make sure
the pressure is the same, then it doesn’t matter if I use a
different pump with less bends in the vacuum line.” The
answer to this particular question is “Yes, it does make a
difference, and no, it’s not ok to make a change.” The ratio-
nale is that you might be able to get the same result under
ideal conditions, but the only way to guarantee you will al-
ways get the same results, both steady state and transients,
under all possible conditions of environment, age, etc., is to
copy the configuration exactly. To deal with this type of
question, a detailed Q&A list was prepared and communi-
cated to all engineers involved on the project.

Systems Synergy
Figure 3: Copy EXACTLY! technology transfer method for
The scope of copy EXACTLY! for the 0.5-micron technology
the 0.5-micron generation
was for the most part limited to anything that might have an
impact on the process, or how it is run. The motivation was to
guarantee equivalent yields, starting with the first wafer, and As always with projects of this magnitude and complexity,
to ensure there were no reliability problems to delay produc- there were some issues encountered along the way. For ex-
tion. One recommendation [7] from this program was that the ample, a very subtle problem affecting the integrity of the
concept could be extended into other areas as a way to fur- sub-micron metal lines was found. However, since the pro-
ther accelerate new factory start-ups and the manufacturing cess had been copied so precisely, trouble shooting became
ramp of new-generation technologies. an exercise in revisiting the exceptions that had been made
The 0.35- and 0.25-micron generation technology transfers and auditing to look for unforeseen errors in copying. In this
example, two variables were identified as suspects, and a
[8] took the Copy EXACTLY! method a step further into what
simple experiment on test wafers identified the cause within a
has been described as total “Systems Synergy,” where al-
week. A simple typographical error had been made in enter-
most every aspect of the fabs are identical at multiple geo-
ing a process recipe. The problem was very subtle and would
graphic sites. The 0.8- and 0.5-micron generations both had a
have taken many weeks to identify if a traditional transfer
“virtual factory” organization structure and a series of com-
approach had been used. In addition, areas for improvement

The Evolution of Intel’s Copy Exactly! Technology Transfer Method 4

Intel Technology Journal Q4’98

in the technology were known and found in both sites. Since production line engineers earn the right to take a leadership
no new problems were introduced as a result of the technol- role in making improvements. In the meantime, ideas are still
ogy transfer, the number of engineers and other resources welcome, but they are implemented through the R&D organi-
available for basic improvement work was greatly increased. zation and at the same time at both sites. The R&D engineers
Moreover, the overall technology transfers to two new 0.5- also need to make some sacrifices. To make changes they
micron factories were accomplished in record time with very now need the support of the production line engineers.
few problems along the way.
The results obtained clearly show the merit of the Copy
Figure 4 shows the results obtained on the 0.25-micron tech- EXACTLY! philosophy and systems. The process flow was
nology generation, using copy EXACTLY! and Systems transferred to two new factories in record time with equiva-
Synergy. lent yield and other indicators, and with no product quality
issues. The new lines were able to precisely intercept the
technology learning curve.
A number of other benefits were also realized and are as
follows:
Customer acceptance: Many major customers for integrated
circuits are well aware of the risks in changing manufacturing
plants and will typically demand the opportunity to re-qualify
a second source. If the supplier has a high credibility rating,
this may simply require a study of all the data from the new
line. However, in many cases the customer may want sample
devices to submit to his/her own testing, or to a third party
laboratory. He/she may also require a site visit and an oppor-
tunity to audit the new line. In all cases there will be addi-
tional costs and delays in time to market. However, once the
customers understood the Copy EXACTLY! method, many
Figure 4: Copy EXACTLY! technology transfer method of these concerns, costs, and delays were eliminated.
for the 0.25-micron generation
On-going mutual synergy and shared learning: In the ex-
It is now routine for new Intel factories or new technologies ample outlined here, the R&D line continued to manufacture
that were transferred to obtain equivalent yield starting with the new products along with the two manufacturing lines. By
the first check-out wafer. Production quantities of products keeping the process in lock step at all three sites it was pos-
are started immediately since there is such confidence the sible to share the improvement projects among them. Im-
product will be good. provements were characterized in one site and transferred to
the others with minimal effort. In effect, the number of engi-
Discussion neers per process step or per area for improvement is in-
The difficulties in implementing this new philosophy and creased, as is the number of improvement ideas generated.
system are not to be underestimated. Any major project,
Manufacturing flexibility: With three sites running the exact
such as a new factory or new process flow in an existing
same process, products were easily transferred back and forth
factory, is started with the ambition to get the best ever re-
with no re-qualification, other than checking the mask set.
sults. Engineers are trained and rewarded for doing improve-
Using free capacity at another site has also solved manufac-
ment projects, and product re-qualification affords them such
turing bottlenecks.
an opportunity. The natural tendency is thus to use the new
start-up as an opportunity to implement improvements. To
Conclusion
change the mind set of a large body of technical experts
requires a very simple message, consistently delivered, and The Copy EXACTLY! method has proven itself as a tech-
backed up by a set of systems that make it difficult to behave nique for semiconductor technology transfer. A new process
differently from the desired state. With Copy EXACTLY!, flow and products can be introduced to production in mini-
the message to the production engineers was to achieve the mum time with equivalent yields and without the introduc-
best ever replication in the fastest possible time, and it will be tion of product-quality issues. Both manufacturer and cus-
considered the “best ever.” Once the new products are up tomers can reduce their time to market. This approach could
and running, with good stability and in high volume, the equally be employed in other industries where the technol-
ogy is complex and has many interacting variables affecting

The Evolution of Intel’s Copy Exactly! Technology Transfer Method 5

Intel Technology Journal Q4’98

the end result. The concept has been successfully expanded [5] C.J. McDonald, “Fab 10 Start-Up Methodology – Copy
to cover all systems used in manufacturing. EXACTLY!,” Intel Internal Document, February 1992.

Caution [6] C.J. McDonald, “Copy EXACTLY! A Paradigm Shift in

Technology Transfer Method.” Presented at the IEEE Ad-
Copy EXACTLY! is a powerful method for technology trans-
vanced Semiconductor Manufacturing Conference, 1997.
fer, but should not be applied during technology develop-
ment. By definition, technology development means taking [7] J. Multani, C.J. McDonald et al., “P852 Virtual Factory
new processes and improving and integrating them to create Vision Through Copy EXACTLY!.” Presented at the Intel
a new generation process flow with greater capabilities. While iMEC Conference, 1994.
it may be decided that some existing process modules and
[8] G. Gimpelson et al., “Rapid Technology Transfer: BKM’s
equipment can be reused as they are, in general, technology
from the P856 Process Transfer.” Presented at the Intel iMEC
development requires great creativity and innovation. Tech-
Conference, 1998.
nology development would be dampened by the rigid disci-
pline required during technology transfer and manufactur-
Author’s Biography
ing.
Chris J. McDonald holds a B.Sc. in electronics from the Uni-
High-volume manufacturing also demands a high degree of versity of Southampton, UK and an M.Sc. in electrical engi-
change. Yields must be continuously increased, efficiencies neering from the National Technological University, USA.
improved, and costs reduced. It is vitally important that the He is a sixteen-year employee of Intel, currently assigned to
systems used to manage change strike a good balance be- SEMATECH where he is director of the Advanced Technol-
tween the discipline required to keep the factory under con- ogy Development Facility. His other positions included that
trol and the creativity and innovation required for continu- of manufacturing manager and yield/Q&R manager for the
ous improvement. Manufacturing improvement systems start-up and ramp of Intel’s high-volume manufacturing facil-
need to be very fast-moving and flexible. Multiple factories ity in Ireland and a process/equipment engineering manager
running the same process and products should remain in Albuquerque. He has twenty-four years experience in semi-
matched, but not necessarily identical at all times. New ap- conductor process development and manufacturing, and he
proaches should be tested in one site and proliferated to has worked in all areas of wafer fab engineering with a wide
others when proven. range of memory, microprocessor, and other products. His e-
Finally, the Copy EXACTLY! method is designed to match all mail is [email protected] or
factors that impact the process or how it is run. Other sys- [email protected].
tems might benefit from matching, but time and money should
not be wasted on matching factors that have no impact on
the overall process.

Acknowledgments
The author would like to recognize Tom Hartman for his lead-
ership through two new factory start-ups. The author also
acknowledges Bob Jecmen and Bruce Leising for their spon-
sorship in the implementation of these new approaches, and
for their support in the publication of this paper.

References
[1] G.H. Parker, “Intel’s Manufacturing Strategy,” presented
at the 5th International Symposium on Semiconductor Manu-
facturing.
[2] C.J. McDonald, “Fab 9.1 Start-Up Methodology.” Intel
Internal Document, June 1988.
[3] M. Bohr and S.Ruska, “P650 Transfer Methodology Pro-
posal,” Intel Internal Document, December 1989.
[4] S. Ruska, “Post-Mortem of Fab 9.2 Yield/Undercut Is-
sues,” Intel Internal Document, December 1991.

The Evolution of Intel’s Copy Exactly! Technology Transfer Method 6