Critical Decisions

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

P R o c e s s I m pr o v e m e n t

Critical Decisions

O
rganizations measure the process capability of final product charac-
strategy ensures teristics to ensure compliance to specifications, report to customers
corrective action and management, and assess opportunities for improvement.
Quality professionals often encounter situations in which the process
fixes the problem control parameters and in-process outputs show compliance to process
documentation and specifications. As products move downstream, however,
the first time the magnitude of nonconformance increases, resulting in low yield and
out-of-control situations.
This always perplexes engineers and can indicate a poor understanding
of the relationship between upstream critical-to-process (CTP) parameters
and downstream critical-to-quality (CTQ) characteristics.
Low yield costs more at the end of a process than at the beginning, mainly
because of value additions in terms of material, labor and overhead that
occur during every step. But there is a way to identify the exact process step
where corrective action is needed using a strategy grounded in traditional
thinking on the importance of understanding CTP-CTQ relationships.
This approach also adds an emphasis on confirming those relationships
with respect to time. Understanding such relationships is key to reducing
By Govind the proportion of expensive, nonconforming, out-of-specification products
downstream; reducing overall internal failure costs; and freeing capacity that
Ramu, JDS may otherwise have been tied up processing potential failures.
JDS Uniphase Corp. used a step-by-step method in its semiconductor
Uniphase Corp. wafer process plant to identify time-dependent relationships between CTP
parameters and CTQ characteristics. The knowledge obtained through this
process was used by the process engineering team for yield improvements
(yielded dies per wafer).
Although the actual descriptions of the process variables, quality char-
acteristics (optical specifications) and yield improvements made are the
organization’s intellectual property, the graphical data minus the process-
specific labels (in Figures 2 and 3, p. 18) come directly from the shop floor.
In this case, the approach was used on an existing process. But the exercise
can also be performed during new process development. In fact, I highly
recommend it.1

1. Identify CTQs, CTPs

Assemble a team consisting of product-line management, account man-


agement, R&D, quality and manufacturing engineering, and customer
representation. If customer representation is not practical, the quality
department, marketing, sales or account management should provide sup-
port on behalf of the customer. Assign this team the task of identifying CTQ
characteristics.

six sigma forum magazine I May 2010 I 15


Crit ical Decisions

As products move downstream, the magnitude of nonconformance

increases, resulting in low yield and out-of-control situations.

This always perplexes engineers.

For the same product line, assemble a second team Table 1. CTPs and CTQs by process step
to identify CTP parameters. Include manufacturing,
quality, equipment and production engineers, and Process Process Process Process Final test
invite R&D engineers when it’s applicable. Using the step 1 step 2 step 3 step n step
same R&D and quality and manufacturing engineer- CTP1 CTP3 CTP4 CTP n CTQ1
ing representatives from the CTQ team in the CTP CTP2 CTP5 CTQ2
identification exercise provides continuity. Make sure CTQ3
CTQ identification involves strong customer represen- CTQ4
tation and that CTP identification relies on stronger CTQ5
engineering representation. CTP = critical to process
CTQ = critical to quality
The CTQ team should consider using voice of the
customer tables, customer surveys, scorecards, mar-
ket needs, engineering specifications and historical 2. Create a CTQ-CTP relationship matrix
failure data of comparable legacy products. The team
should further brainstorm any new expectations for For existing products under regular review, you should
the product. already have CTQs and CTPs identified, and should
Next, the CTP team starts a brainstorming discus- understand the relationships between them. If you
sion. The inputs come from a combination of histori- have not formally documented them, hold a brain-
cal data, equipment, process knowledge and experi- storming meeting to develop a relationship matrix
ence with the product line. The output should be a set (see Figure 1). For new products, you should be able
of CTP parameters for each process step, as illustrated to create the relationship matrix as you brainstorm and
in Table 1. document CTQs and CTPs for the first time.

Table 2. Semiconductor wafer process team’s relationship summary

Screening design of experiments Time-dependent


Relationship Cause and effect Correlation
(DoE) using engineering judgment relationship
CTP1 and CTQ2 No DoE significance No Low correlation ?
CTP1 and CTQ4 No DoE significance Yes Low correlation ?
No DoE experiment, engineering
CTP2 and CTQ1 Yes High correlation ?
judgment is yes
CTP2 and CTQ5 Not known High correlation ?
CTP3 and CTQ3 Significant factor Yes High correlation ?
CTP3 and CTQ5 Not known Moderate correlation ?
CTP4 and CTQ2 No Moderate correlation ?
CTP5 and CTQ1 Significant factor Yes High correlation ?
CTP5 and CTQ3 No DoE significance No Low correlation ?
CTP5 and CTQ5 Not known Moderate correlation ?
CTP = critical to process
CTQ = critical to quality

16 I May 2010 I W W W . AS Q . ORG


C r i t i c a l D e c i sions

Initial inputs for the relationship matrix come Figure 1. Relationship matrix
from the outputs of quality tools, such as cause and
effect diagrams, and from statistical methods, such
as correlation and design of experiments (DoE). 5
Confirmation for the selection of variables for the

Critical-to-quality characteristics
screening experiment will come from engineering
judgment. This is based on a combination of knowl- 4
edge of and experience with the product. Then,
create a relationship table to identify the source
of data, information and available knowledge (see
3
Table 2 for the relationship summary created by the
semiconductor wafer process team).
As you and your team identify relationships
between CTQs and CTPs (Table 2), you need to 2
agree on the degree of that relationship. A good way
to do this is via a relationship matrix (see Figure 1),
classifying each one as significant (concentric circle), 1
moderate (circle) or weak (triangle). This is similar
to the relationship matrix used in quality function 1 2 3 4 5
deployment.2 Critical-to-process parameters
While assigning significance, make sure the team
understands the difference between correlation and Significant Moderate Weak
causation.3 Causation happens as an effect of cause.
Mathematical correlation alone does not mean cau-
sation exists. Mathematical
correlation can only become
causation by a validated theory. Table 3. CTP2 and CTQ1 in the measured sequence
3. Comparing process Device # Date (1) Time (1) CTP2 Device # Date (2) Time (2) CTQ1
behaviors 0045 04/11/2008 08:00 5.85 0048 04/14/2008 11:00 -50
0046 04/11/2008 09:00 5.75 0045 04/14/2008 12:00 -49.5
Up to this point, the tools used 0047 04/11/2008 10:00 5.80 0050 04/14/2008 13:00 -50
to establish relationships— 0048 04/11/2008 11:00 5.85 0047 04/14/2008 14:00 -50
cause and effect diagrams, 0049 04/11/2008 12:00 5.80 0046 04/14/2008 15:00 -50.5
screening DoE and correla- 0050 04/11/2008 13:00 5.85 0049 04/14/2008 16:00 -50
tion—have been independent CTP = critical to process
of time. Now, you can turn CTQ = critical to quality
your attention to confirming
whether the identified rela-
tionships are time dependent.
In the example in Table Table 4. CTQ1 arranged in CTP2 process sequence
2, you still don’t know if the
time-dependent process Device # Date (1) Time (1) CTP2 Device # Date (3) Time (3) CTQ1
behavior of CTQ1 follows 0045 04/11/2008 08:00 5.85 0045 04/14/2008 12:00 -49.5
that of CTP2. The sample 0046 04/11/2008 09:00 5.75 0046 04/14/2008 15:00 -50.5
information in Table 2 shows 0047 04/11/2008 10:00 5.80 0047 04/14/2008 14:00 -50
a cause and effect relation- 0048 04/11/2008 11:00 5.85 0048 04/14/2008 11:00 -49.5
ship, high correlation of data 0049 04/11/2008 12:00 5.80 0049 04/14/2008 16:00 -50
and agreement in engineer- 0050 04/11/2008 13:00 5.85 0050 04/14/2008 13:00 -49.5
ing judgment between CTP2 CTP = critical to process
CTQ = critical to quality
and CTQ1. To see if the CTP2

six sigma forum magazine I May 2010 I 17


Crit ical Decisions

Figure 2. CTP2 control chart


Individual measurement of CTP data
6

5.9 UCL = 5.909


A
5.8 B
CTP data

C
AVG = 5.725
5.7 C

5.6 B
A
LCL = 5.541
5.5

11 22 33 44 55 66 77 88 99 110 121 132 143


Sample
CTP = critical to process
UCL = upper control limit
AVG = average
LCL = lower control limit

Figure 3. CTQ1 control chart (modified time sequence)


Individual measurement of CTQ1 data
-43

-45

-47
CTQ1 data

-49 A UCL = -48.67


B
-51 C AVG = -51.21
C
B
-53 A
UCL = -53.76
-55
11 22 33 44 55 66 77 88 99 110 121 132 143
Sample
CTQ = critical to quality
UCL = upper control limit
AVG = average
LCL = lower control limit

manufacturing sequence is an indicator for CTQ1, Impulsively analyzing statistical process control
follow these steps: of CTQ1 in the measurement sequence of “Date
1. Plot CTP2 according to the time sequence of man- (2)” and “Time (2)” is an incorrect way to com-
ufacturing “Date (1)” and “Time (1)” (Figure 2 is pare the quality characteristic to the process
the semiconductor wafer process control chart). parameter. From Table 1, we know the previous
step parameters have no leverage on CTQ1. We
2. Use manufacturing database traceability to
retrieve the CTQ1 data corresponding to CTP2. know that only CTP2 and CTP5 have leverage on
Often, lot number, batch number or device serial CTQ1. In our exercise, we are trying to discover
number is a connecting field. As you see from the if CTQ1 would behave the same way as CTP2, so
example of the partial manufacturing database in we arrange CTQ1 in the same manufacturing
Table 3 (p. 17), CTQ1 is measured in a different sequence as CTP2 (see Table 4).
sequence (refer to device number) than CTP2. 3. Arrange CTQ1 in the same process time sequence

18 I May 2010 I W W W . AS Q . ORG


C r i t i c a l D e c i s i ons

Make sure the team understands the difference between correlation and

causation. Causation happens as an effect of cause. Mathematical


correlation alone does not mean that causation exists.

as CTP2, and rename the columns “Date (3)” and exist that will affect the nature of the relationships
“Time (3)” (see Table 4). you identify.
4. Plot CTQ1 next to CTP2 using the “Date (3)” and For example, any given CTQ-CTP relationship may
“Time (3)” time sequence (see Figure 3). depend upon the use of certain equipment or certain
environmental conditions. You may need to assess
5. Look for process patterns and behaviors—such relationships for multiple conditions across multiple
as increases or decreases in variation, mixtures, process steps to thoroughly understand.
cycles, clusters and trending—between the two Additionally, you may encounter situations when
charts. If CTQ1 demonstrates matching patterns observing process parameters in combination leads
and behaviors to CTP2 at the time period, we can you to different conclusions than when observ-
conclude CTP2 is an early predictor of CTQ1.4 ing the parameters individually. You may need to
Remember, this is on top of an already-estab- observe two or more CTPs jointly to understand a
lished relationship using cause and effect analy- CTQ.
sis, statistical correlation, DoE and engineering
judgment. Acknowledgement
A similar exercise for CTP5 will reveal whether
it is also an early predictor for CTQ1. If two or The author wishes to thank Catherine Harwood for assisting in data col-
lection, statistical process control analysis and relationship identifications
more process parameters have patterns that match during the implementation of this method.
CTQ1, use your engineering judgment to pick the
most appropriate one or study them in combina- References and notes
tion using a three-way or four-way chart to make
downstream predictions. 1. For more details on steps one and two, see Govindarajan (Govind)
Ramu, “Metrics that Trigger Actionable Discussions: Prioritize
A powerful combination of quality tools can help Process Improvements Using Gauge R&R and SPC Capability,” www.
derive relationships, match patterns and behaviors asq.org/sixsigma/2007/12/metrics/metrics-that-trigger-actionable-
discussions-prioritize-process-improvements-using-gauge-rr-and-spc-
between process parameters and quality characteris- capability.pdf.
tics, and identify the most appropriate upstream test 2. Yoji Akao, Quality Function Deployment, Productivity Press, 1990.
3. Tom Kubiak and Donald W. Benbow, The Certified Six Sigma Black Belt
parameters for final test characteristics. Monitoring Handbook, second edition, ASQ Quality Press, 2005.
the stability and variations of these CTP parameters 4. It is possible to write sophisticated algorithms to compare the charts.
This is beyond the scope of this article. In the interest of keeping it
will provide early insight into patterns and behaviors simple, I suggest practitioners visually compare the patterns and behav-
downstream, helping reduce the expensive noncon- iors. Full-factor design of experiments can be used to understand the
effect of combinations of process parameters.
forming, out-of-specification products and overall
internal failure costs.

Challenges and lessons learned

You may not see a mirror reflection between the


comparison charts because there are parts that are
scrapped at intermittent steps.
Some of the implementation scenarios are always What do you think of this article? Please share
more complex. One of the considerations to
your comments and thoughts with the editor by e-mailing
take into account with this proactive approach to
improvement is whether any conditional constraints godfrey@asq.org.

six sigma forum magazine I May 2010 I 19

You might also like