Introduction To Statistical Process Control: Primary Knowledge Unit Participant Guide
Introduction To Statistical Process Control: Primary Knowledge Unit Participant Guide
Introduction To Statistical Process Control: Primary Knowledge Unit Participant Guide
This Primary Knowledge (PK) unit provides an overview of Statistical Process Control (SPC)
and how it relates to MEMS fabrication. Statistical Process Control, often referred to as SPC, is
a set of tools used for continuous improvement and quality control of an active manufacturing
process. There are two (2) suggested activities that reinforce the material presented in this PK as
well as a final assessment.
In this unit you learn the basics of SPC, its terminology, and some of the tools used to help
ensure a quality production line.
Objectives
• To explain process variation and the need to identify special cause variation.
Outcomes
You should be able to describe why Statistical Process Control is needed when manufacturing a
product and you should be able to apply the basic tools of statistics and Shewhart rules to
interpret a control chart.
We can all agree that when manufacturing a product, it is desired to produce a “quality” product.
This can be said whether we are talking about cars, food, medicines, or microsystems. There is
no universally accepted definition for "Quality"; it is a subjective term full of meanings and
connotations. Given the needs of a customer, we can say that quality is the realization and
control of characteristics that determine whether the product will in fact satisfy those needs.
Realization includes the design of the product. Control includes the control of deficiencies in the
product minimizing the variation around desired nominal values or "targets".Reference Mike Leeming
Statistical methods are used throughout the life cycle of a product, which are aimed at the
realization and control of certain product characteristics. For example, methods of statistical
experimental design or Design of Experiments (DOE) may be used in the design phase of the
product life cycle or in efforts to improve the control of certain product characteristics. This is
achieved by identifying key factors (e.g., thin film thickness, process temperature and pressure,
line widths) that need to be controlled during the manufacturing process of the product.
Statistical Process Control (SPC) is used real-time during the manufacturing process where in-
line data is attained from the processes that produce the products. Statistical methods are then
used to assess whether or not the process is in a state of control. This statistically based process
information can provide a greater understanding of the process by providing a graphical
interpretation of the variation in the process. All processes have some variability over time, as
illustrated below. The graph could represent the variation in oven temperature, photoresist
thickness, or number of defective die on a wafer. Have you ever cooked a soft-boiled egg? Is
the outcome “exactly” the same every single time you cook it or is there a little variation?
Variation is a natural and commonly occurring phenomenon but not all variation is created equal.
A process may contain variation that is common or inherent to the process and, there may also
exist variation that is NOT common or inherent to the process. Variation that is NOT common
would be a result of a special cause outside of the normal process conditions.
Statistically, each piece of equipment shown in the previous graphs applies a target of a 50
micrometer (µm) thickness of photoresist to a wafer, but as the graphs show, the final resist
thicknesses vary differently for each machine.
Studying process variation can provide insight into the sources of variation and ways to
minimize the variation in the manufacturing process. This knowledge can help lead to greater
consistency in the final product and less deficiencies or defects. The use of statistics makes good
sense in quality, because even when all seems to be running well, there are many uncontrolled
production factors that can affect product characteristics. When manufacturing a product, most of
the factors are unknown, can vary, and may not affect the process all of the time. The unknown
factors provide the ingredients of a probabilistic environment and natural "background noise" so
that it is impossible to predict or calculate exactly how products and their deficiencies will vary.
Under these circumstances methods of probability and statistics are applied so that predictions
can be made and those involved in the manufacturing of the product know what to expect.
Controlling quality is a science, and the mathematics of quality is probability and statistics.
A person does not have to be a statistician in order to correctly use and interpret the various SPC
tools. However, one needs to understand and correctly apply statistics terminology and notation
when using SPC for quality control. It is important to be mindful of accuracy when collecting
and interpreting data. Understanding the statistical tools used in the quality control of a
manufacturing process helps to formulate data-based predictions or decisions rather than just
"The long-range contribution of statistics depends not so much upon getting a lot of highly
trained statisticians into industry as it does on creating a statistically minded generation of
physicists, chemists, engineers, [technicians], and others who will in any way have a hand in
developing and directing the production processes of tomorrow." - Dr. Walter E. Shewhart, 1939
Variation
All products, whether being man made or nature made, are not exactly created equal. There is a
natural or inherent variation in all processes. In a field of 3 leaf clovers, you won’t have to look
too hard to find either a 2, 5, 6, or even 4 leaf clover. When chickens lay eggs, the size,
thickness of shell, color of the yoke, number of yokes, and the color of the shell all vary from
egg to egg.
Each individual process step has many different parameters or variables. Each of these
parameters can vary or drift during the process. For example, in the coat process of the
photolithography step, it is desired to have a specific thickness of photoresist. The resulting
thickness depends upon or is a function of the spin speed of the chuck on which the wafer sits,
and the actual viscosity of the photoresist deposited on the wafer’s surface. If one or both of
these variables change during the coat process of a batch of wafers, then the final thickness of the
photoresist will change from one wafer to another. Small changes in these variables may be
acceptable as a natural or inherent variation of the process. Any change outside of this inherent
Types of Variation
With any process there are two types of variation: variation that is inherent to the process and
variation that is not inherent to the process. For example, let’s return to these two process graphs
again. Assume that these two graphs or control charts represent the variation of photoresist
thickness in a microsystems process over time. However, this time, let’s assume that the graphs
are from the same piece of equipment, but on two different days. Has something changed?
As you’ll soon learn, the control chart on the left is “in control” and the variation that you see is
expected because it has been statistically determined that this is the variability inherent to the
process. This is the variation due to the “background noise” that we mentioned earlier. This
variation is referred to as “common cause variation”. However, the control chart on the right
tells a different story. Notice the two points outside of the dashed lines. These points represent
variation that is neither predictable nor inherent to the system. This type of variation is due to an
“assignable cause” that can be known, or most often is unknown. Such variation is referred to as
“assignable cause variation” or “special cause variation”. A problem solving team should be
able to determine the cause of this variation and eliminate it from the process. As you can see
from this example, control charts used in SPC help determine which type of variation (common
or special cause) is present. As a result, control charts can help determine the best course of
action.
By monitoring the process this way, it is easy to pick up on a process change such as the
temperature increase. After an investigation, an example of a root cause for this sudden jump in
temperature may be due to not enough coolant in the cooling system. The cooling system could
be checked, if a coolant is needed, than it would be added to alleviate the problem.
So would one ever want to decrease the variation inherent to the system? Of course you would!
However, inherent variability cannot be reduced until the process is in control and all special
cause variation has been eliminated.
How much variation is too much? That depends on the product. There are some cases where
limited variation is intentionally built into the process. Can you think of a product where
variation is desired? Many hand made products are valued because of the inherent variation,
such as hand carved furniture, pottery, and jewelry. There are also products for which a certain
amount of variation is acceptable or not recognizable by the consumer, such as the color of a car,
the weight of a box of cereal, or the thread count in a set of sheets.
Microsystems, on the other hand, seeks to have as little variation as possible. Microsystems deal
in micrometers and nanometers, not centimeters and inches. Because of the extremely small
feature sizes (1 µm to 100 µm) and precise alignment (sometimes in the nano range),
microsystems fabrication needs to be on “target” from step to step, over hundreds of steps.
Remember the example of the micro pressure sensor process? There are so many steps involved
in microsystems fabrication that when variation occurs at one step, it can perpetuate and affect
other steps along the way.
Statistical Process Control (SPC) assists in the proper management of variation. Certain types of
variation require action by management. Other types of variation often do not require
management intervention and can (depending on company policy and training) be remedied by
people who are not in management (e.g., equipment operator or technician or both). Companies
normally leave special cause variation in the reliable of hands of the people that are in daily
contact with the process – operators, technicians, supervisors, and engineers. However, reducing
inherent variation requires extensive knowledge of the process, the product, the equipment, and
all of the factors that together, create this variation. It may also require process changes and
even product redesign; therefore, reducing inherent variability is normally left in the hands of
management; although, operators and technicians may be asked to be on the problem solving
team.
In-line variation reduction is a major goal of Statistical Process Control. Interestingly, the way in
which products and process characteristics vary about some desired target value can actually
provide valuable information that can be used to understand, control, and reduce the variation.
With predictable processes that also have less variation, the producer is better able to change or
control things to provide improved products that are more consistent and on target. More
consistent products that are on target require less inspection, scrap, and rework. An erratic
process full of variation is similar to the automobile that has a lot of play in the steering
mechanism. You have little control and less freedom in the driver's seat as you have to keep
adjusting and hoping you will stay on the road.
In order to understand the basics of Statistical Process Control (SPC), there are several statistical
concepts you should be familiar with. These concepts are used to provide basic analytical
descriptions of data. These statistical calculations include sample median, sample mean, sample
variance, sample range, and sample standard deviations. Here, we briefly discuss these statistical
concepts that you need to understand in order to employ SPC.
There are also many different types of charts that help to describe data in different ways in order
to achieve quality control. Some examples include Run Charts, Frequency Distribution Tables,
Histograms, Stem and Leaf Plots, Box Plots, Scatter Plots and Control Charts. This lesson
discusses Control Charts and its uses in Statistical Process Control. For more information on the
other charts mentioned, please review the Basics of Statistics provided as part of the Statistical
Process Control Learning Module.
Data Samples
When employing Statistical Process Control, statistical analysis is usually done on a sample of
data, hence the statistics which we cover here are for a sample set of data. All calculations use
the data from the sample.
One set of calculations of SPC identifies the central tendency of the sample data. This set
includes the sample median and mean. Remember, the word "sample" is used because the
median and the mean are the median of the data in the sample set and the mean of the data in the
sample set.
• The Sample Median represents the data value that is “physically” in the middle of the
sample set when arranged in numerical order. (Examples provided below.)
Another set of calculations addresses variability or dispersion in the sample data. The statistics
that are usually calculated are the sample range, sample variance, and the square root of the
sample variance which is called, sample standard deviation.
• Sample Range is the spread of the distribution of the data, or more simply put – the
difference between the maximum value minus the minimum value. For example, given
the following sample of voltage measurements, what would be the sample range?
• Sample Standard Deviation is a measurement of how the data values are distributed
around the sample mean and within the range of values. This is one of the most
important calculations for SPC analysis and is used to develop control charts. Both
common cause variation and special cause variation rely on the standard deviation for
analysis. This will be discussed in detail shortly.
Let’s get specific about our sample size now and define variables. Take a random sample of size
n from a population of x’s (x1, x2, …,xn).
For example, say you have a random sample of five values from a population of x’s. These
values are (x1, x2, x3, x4, x5). For this set of data, your n would be equal to five (n=5). For the
following calculations, let us use a numerical example to demonstrate the necessary calculations.
Say you have obtained a random sample of voltage measurements, (2,4,9,1,4) volts.
Sample Median
To obtain the sample median, first put the numbers into ascending order as follows: (1, 2, 4, 4,
9). The sample median is the middle number, 4. The sample median is 4 volts. Had the sample
been of even sample size, the median would be the average of the two center numbers. For
example, suppose the sample was (2, 4, 1, 5, 1, 3). Ordering the data we have (1, 1, 2, 3, 4, 5).
The two middle numbers are 2 and 3; thus, the median would be the average of 2 and 3, or
2.5.The sample median is a random variable 𝑋. The numerical value of the sample median is
represented by 𝑥.
The arithmetic mean, µ, of a set of data is the sum of the data values, xn, divided by the number
of data values (n):
Σ𝑥!
𝜇=
𝑛
When a set of data is identified as a sample, the arithmetic mean is calculated for the set of data,
but can sometimes be represented as 𝑋.
(2 + 4 + 9 + 1 + 4) 20
𝜇=𝑋= = = 4 𝑣𝑜𝑙𝑡𝑠
5 5
A mean of 4 volts tells us that the central location of the distribution based on the sample data is
4 volts. In the case of the following equations, µ will be used as the sample mean, but keep in
mind that it can also be referred to as 𝑋 when referring to the arithmetic mean of a sample.
Sample Variance
When we are looking at a sample set of data, it is desired to quantify how spread apart the data
values are with respect to each other. Understanding how far the numbers are from the mean is
just one of the statistical characteristics which can help us to understand the nature of the
process.
Where𝜎 ! is sample variance, xi are the individual sample data values, 𝜇is the sample mean, and n
is the number of sample data values. Using our sample voltages, (2,4,9,1,4) volts and sample
mean (𝜇 = 4), we will begin our calculation by looking at the differences between the individual
sample data values, xi, and the sample mean, 𝜇.
Note that each of the resulting numbers shows how far above or below each data value is from
the sample mean. The sum of the differences (deviations from the sample mean) should always
!"
sample variance= 𝜎2= ! = 9.5 𝑠𝑞𝑢𝑎𝑟𝑒 𝑣𝑜𝑙𝑡𝑠 (𝑉 ! )
Remember that each of the data values represents a voltage reading; therefore, the squared value
of the variance has the unit of V2.
The standard deviation is simply represented as the square root of the sample variance. Sample
standard deviation is an estimate of the root mean square deviation from the mean.
!
− 𝜇)!
!!!(𝑥!
𝜎= 𝜎! =
𝑛−1
Using our sample data, the square root of the sample variance, 9.5 volts2 is 3.08 volts. Therefore,
the sample standard deviation 𝜎, is 3.08 volts.
How does this statistical data from the sample help us create our control charts? Let’s revisit the
control charts we spoke of earlier. If you recall, the control chart monitoring the process
temperature looked like this.
There are many types of charts used to monitor different types of data. It is very common in
industry to base the limits for controlling the process on the sample mean, µ or as you recall, 𝑋.
This type of chart is commonly referred to as a Mean or 𝑋-Chart. When this type of chart is
used, the middle line, or the centerline is the sample mean, µ. The dotted lines are the upper
control limit (UCL) and lower control limit (LCL). The upper control limit can be calculated or
set many ways, but most commonly 𝜇 + 3𝜎 is used for the upper control limit and 𝜇 − 3𝜎 for the
lower control limit.
As you can see, setting up a control chart can be very easy and practical, you just have to be
familiar with basic statistical calculations and have access to the process data.
To learn more about the specifics of control charts and how to interpret the results, please read
Control Chart Basics.
Statistical Process Control (SPC) is a scientific method that can provide much information about
a process and how specific process parameters are varying with time. In order to produce a
quality product, the amount of variation should be understood and controlled. SPC is a common
set of tools used in industry to help manage this variation. Control charts help to identify
variation and identify process issues before they begin to affect the product.
All processes have common cause or inherent variation. Most processes also experience some
special cause variation. In order to maintain a quality product, it is important to be able to
identify special cause variation in order to reduce or eliminate it. Statistical Process Control
(SPC) can help to identify and monitor all types of variation.
Understanding the basic mathematical concepts involved with SPC is important in developing
and utilizing these tools correctly. Some basic statistical concepts that you should be familiar
with are sample median, sample mean, sample range, sample variance, and the square root of the
sample variance, the sample standard deviation. Using these basic calculations will assist you in
creating a SPC chart to monitor process variation.
References
1. Dr. Michael Leeming, University of Arizona
2. Dr. Richard Prairie, Statistical Professor, University of New Mexico
3. Montgomery, D.C. (1985). Introduction to Statistical Quality Control (2nd edition). New York:
Wiley.
4. Devore, J.L. (1995). Probability and Statistics for Engineering and the Sciences (4th edition).
New York: Duxbury Press.
Disclaimer
The information contained herein is considered to be true and accurate; however the Southwest
Center for Microsystems Education (SCME) makes no guarantees concerning the authenticity of
any statement. SCME accepts no liability for the content of this unit, or for the consequences of
any actions taken on the basis of the information provided.
Support for this work was provided by the National Science Foundation's Advanced
Technological Education (ATE) Program through Grants. For more learning modules related to
microtechnology, visit the SCME website (http://scme-nm.org).