Workshop Report: Southwest Region
A Success Story in Statistical Process Control
at Texas Instruments, Austin, TX
Leading edge technology and increasingly higher quality targets offer plenty of
challenge to TI's PWB Group at Austin. Thanks to the training and involvement of
all employees in the prevention of defects, their performance continues to
improve in quality, delivery, and WIP control.
Peter W. M. John, Ph.D.
o wonder the PWB (Printed
N
Wiring Boards) group at the
Austin, TX facility of Texas Instruments is proud. They brought their
manufacturing operation from the
brink of disaster to world-class in
three years, a story they shared in
the July 1989 workshop. The keys
to their successful turnaround are
Statistical Process Control (SPC)
and Statistical Ouality Control
(SOC).
The PWB group makes printed
wiring boards (printed circuit
boards); see accompanying box.
Much of their product goes to the
U.S. Department of Defense. In late
1985 and 1986, their manufacturing
process deteriorated so badly that
more than half their lots were being
rejected. WIP (Work-in-Process)
stacked up. More than half their orders were delinquent. Waste,
scrap, and rework were at unbelievable levels. It was time for radical change-time to go to SPC/
sac.
They had dabbled in SPC/SOC
before. Beginning in 1982 they had
outside experts talk to various
groups of workers and give training
courses. But, apart from x-bar
charts on some shop and office
walls and a small start on quality
circles, there was little or no implementation until the catastrophe of
1986 got their attention like a
whack with a 2 x 4. Then manage-
34
ment made the decision to go
wholeheartedly for SPC/SOC. It required a complete cultural change
in their system.
What Did TI Do?
First and foremost, higher
management became actively involved in a highly visible way.
There was no doubt that the people at the top were solidly behind
SPC/SOC.
They confronted four vital
truths:
• You cannot inspect quality into a
process. You must focus on prevention of defects, rather than
detection.
• To focus on prevention, you need
to know more about your process, taking nothing for granted.
• You must train everybody from
managers to the rawest recruits
in the principles and philosophy
of SPC/SOC.
• You must control variability. If
you do not, it will destroy you
You must strive to reduce itrelentlessly.
In hindsight, these truths are
self-evident. Like many ideas, they
are not obvious until circumstances
force them to your attention.
Results: Complete Turnaround
The payoff has been a complete turnaround in the manufacturing process. It is now a model of
quality and profitability. Two exam-
pies from the plating shop illustrate
this improvement.
The plating shop runs three
shifts, five days a week, and is responsible for six different plating
lines. The amount of scrap has fallen to about 15 percent of the level
that prevailed in 1986 when the
SPC/SOC program began.
In one plating stage, the activator palladium concentration has
been brought under control. The
operators now make the decisions
to add more chemical by looking at
their x-bar charts, rather than by
intuition. The saving is $50,000 per
year in that step alone. A plating
scrap graph is shown in Fig. 1.
Even better ideas coming out
of the group's quality circles have
resulted in savings of $328,000 in
the first five months of this year
How TI Did It
TI management jumped right in
and made a commitment to SPC/
sac from the top down. They appointed two full-time quality control
facilitators with responsibility,
among other things, for a serious
program to train everybody. That
part was obvious.
They also set about understanding what really went on in
their manufacturing process. They
identified, in some cases for the
first time, the factors that controlled their processes and learned
how to reduce their variability. To
Target
•
PWB Plating Scrap
10
-
9
8
-
7
6
-
5
4
-
3
2
-
1
,
o
,
1985
,
1987
1986
1988
1989
Fig. ~. This graph represents the retative dollars scrapped for plating defects,
normalized to t 986= 1O.
do this they conducted numerous
designed experiments. They did
not confine themselves to the simple Taguchi experiments. They
used more advanced experimental
designs, such as Box-Behnken designs, to investigate quadratic response surfaces and locate optimum levels. They learned that the
variability in some measurements
was much worse than they had believed and had to be reduced significantly.
In the plating example, mentioned earlier, they learned that
measurement variability accounted
for 95 percent of the recorded variability of the palladium concentra-
1600 01
KOH
14,1089
r===========118
10
16,8678
,-
6,38434
f'b ~.
1.95140
i
i·t~!l Ii !r\~ . !
14
TIME SERIES
XBAR
•
,;1
1-
-, 1.93791
------j
tion! ImprOVed techniques reduced
that figure to 14 percent.
Real Time Process Control
The most striking control feature is the new RTPC systemReal Time Process Control. Each
operator has a monitor displaying
about 20 or 30 miniature windows,
one for each important factor in
that stage of the process. A yellow
light - or worse, a red light - on a
window prompts the operator to
touch that window on the screen,
which calls up the full screen display for that variable. An example
is shown in Fig. 2. Moreover, each
manager can monitor the process
himself in the same way.
Each display has four charts.
On the left are the traditional x-bar
and R charts for the last 50 samples or so. On the right, the last
100 observations are plotted batt>
in time sequence and in a histogram.
There are also some statistics.
The mean and the standard deviation are calculated from the latest
observations. They are not the
same as the values used to construct the control lines in the charts
on the left. Last, but not least,
comes the house benchmarkCpk.
Two Key Statistics: Cp and Cpk
Cp and Cpk are two commonly-used measures of process capability' Both compare the process
variability to the specification limits
and give the engineer an indication
of how small the fraction of defectives will be. They are defined as
follows:
Cp = USL - LSL
6<7
'
Cpk = mean - ~~sest SL
HC"C;;:.
1fs384
,830539
----1 1. 11464
'======~Hrr:IS"'fTiVOG"'RAffi'M======"- ~
Fig. 2. This printout is of a screen from the TI Austin Real Time Process Control
System. Using a touch screen, the operator can access this Information lor all
variables being monitored on the process.
where USL, LSL are the upper and
lower specification limits, and <7 is
the process standard deviation.
Cp is the simpler of the two.
The PWB group uses its reciprocal
as a measure of capability. Cp
measures the capability if your
process is on target - if the process average is indeed halfway between the specification limits.
When your process is not exactly
C>
Winter 1989
35
Fixing the Vital Few, Then the Trivial Many
The table below helps demonstrate how ambitious the 1.3 Cpk target is.
Cpk
0.05
0.67
0.80
0.90
1.00
1.33
1.67
4.00
Number of defects
133,600 ppm (part per million)
71,800 ppm
16,400 ppm
6,900 ppm
2,700 ppm
66 ppm
<1 ppm
<1 ppb (part per billion)
Data obtained from R&M 2000 Variability Reduction Program
As higher Cpk values are attained, there are fewer defects to measurE.
improvement or changes against, thus making the diagnostic journey longer.
If getting to 1.3 has taken five years, 1.5 may take another five years. This is
an example of fixing the vital few and now attacking the trivial many.
Flg.l
on target, the bias must be taken
into account by using the second
criterion, Cpk. As Cpk increases,
the percentage of defectives decreases.
The Cpk statistic is the PWB
group's benchmark. All improvement is measured against it. The
target for this year is to have every
Cpk > 1.3, which corresponds to a
defective rate of about one in
10,000. So far they have made 70
percent of their variables meet this
Cpk requirement. They plan to increase the requirement to Cpk >
1.5 next year and even higher in
the future; see Fig. 3.
The Rapid Reaction Facility
The PWB group built a small
shop, the Rapid Reaction Facility,
for the expeditious handling of
small orders of 12 or fewer panels.
They aimed to reduce the cycle
time to two weeks or less. It is
already down to a week and a half.
They estimate that the facility will
payoff its cost in 16 months. Their
steps to reduce cycle time include:
• Dedicated facility - no shared
equipment or people
• High level of multifunctional people brought in from existing PWB
facility
36
• Work-In-Process managed at 8
to 10x daily output goals
• Reduced lot sizes
• Setup time reduction in key processing areas.
In such a facility, flexibility is
essential. The operators must be
cross-trained on multiple processes. Again we see the need for continual training in the work force.
How Are the Changes
Manifested?
There is no doubt that higher
management is solidly behind the
drive for SPC/SQC. People are regularly taken off the lines for training. The benchmark, Cpk, crops up
everywhere. So do charts - the
usual x-bar and R charts, plus
others showing the status of improvement. The latter charts are
present~d in formats resembling
the traditional QC charts, reinforcIng the dominant SPC/SQC theme.
The cultural change is pervasive, almost to the extent of religious fervor. The operators now
exercise much more control of their
own processes. The shops are
spick-and-span. Operators proudly
tell visitors of the progress in
waste water disposal, and of the
new machine from SWitzerland, and
of TI's active participation in setting
the speCifiCations for it to do what
they wanted. The whole group is
clearly involved to the hilt.
Where Does TI Go From Here?
There remain more production
steps to be integrated into the program. There is the self-imposed increase in the Cpk target to 1.5 to
meet. There are suppliers to be
trained and certified. Training the
work force must not slacken. Can
they sit back after that? No!
.
One reason is that quality gets
Into your blood. There are two
other less subjective reasons. You
must press forward because your
competitors on both sides of the
ocean are improving their manufacturing capabilities too. Furthermore
the design engineers continually ,
Impose new and much more complicated demands upon the manufacturing engineers. The improvements made by PWB in three years
are all the more impressive when
we recall that 1989 boards are
much more advanced than 1986
boards.
What was good enough for last
year's products will not be compet-
About T' Austin Products •••
The cirCUit boards manufactured at the Austin facility are primarily multilayer
in design. Many mechanical and electro chemical processes are used in
making a circuit board. Basically copper clad epoxy laminate is photo
imaged and chemically etched, bonded together in a high
pressure/temperature press, drilled using an NC (numerical control)
machine, photo imaged a second time, plated with copper and tin lead
metals, chemically etched to remove excess copper, and finally routed into
discrete part form.
The product mix in Austin varies from standard multilayer, flex, rigid flex
to high tolerance microwave circuit boards and state-of-the-art
polymer-on-metal surface mount circuit boards.
Target
itive next year. As Satchel Paige
said, "Don't look back. Something
may be gaining on you."
Would TI have made its wholehearted improvement effort without
the catastrophe of 1986? Perhaps
so, but not for a while. The moral is
not to wait until you are hit by a 2
x 4.
The members of the PWB
team have learned how to make
boards more reliably, more qUickly,
and more cheaply. If we measurE'
quality by such yardsticks as reduced waste or increased Cpk,
they have fulfilled their mission. But
have they improved quality in the
sense of building better boards?
Have they also increased the
MTBF (Mean Times Between Failures)?
The workshop did not address
that question. But it must be understood that we cannot be content
with process improvement alone.
We must also strive continuously
for product improvement. Manufacturing engineers cannot sit back
and leave that to the design engineers; it is the mission of the manufacturing engineers too.
'Gupts, P., and others, "A Systematic Ap
proach to SPC Implementation," Quality
Progress, April 1987. p. 22-25.
Author:
Dr. Peter W.M. John is a professor of
mathematics and statistics, Department of Mathematics, University of
Texas, Austin, lX. He is a consultant
to industry on quality assurance and
design of experiments. He has written
two books and numerous papers on
the design of experiments. A third
book, on statistics and quality assurance for engineers, will be published
by Wiley next year.
o