First Published in EOS/ESD Technology Europe
Spring 1990
Cost Benefit
Analysis:
How Much is Enough?
Just how
thorough does staticcontrol have to be? Here's how to find
the costbenefit ration of staticcontrol efforts.
Richard Y. Moss
HewlettPackard Co.,
Palo Alto, CA, USA
Two of the questions most
asked by manufacturing managers are "How good must my ESDcontrol
program be?" and "How much money can it save me?"
These are reasonable questions, considering that managers must
justify their decisions in terms of returnoninvestment or payback
period.
Unfortunately, the first question
is difficult to answer and the second almost impossible. One rarely
knows the extent of ESD damage in a process until after it has
been eliminated.
Basics
ESD is a statistically variable
phenomenon. Subjecting a device or circuit to ESD pulses even
in a controlled experiment using the best equipment produces
wildly variable results. The voltage at which measurable damage
occurs is unpredictable, as is the exact nature of the damage
and the location of the discharge path. In addition, the damage
can be hard to identify as ESD damage per se.
One way to deal with statistically
variable phenomena is to use statistical methods. For example,
if you drive an automobile, you know that rocks thrown up by tires
of passing vehicles can randomly strike your windshield. Damage
sometimes occurs and sometimes does not, depending on many variables.
However, if you could dig into some insurance company files, you
could calculate the average number of damaged windshields per
1,000 cars in different geographical areas. Here's how that might
be done.
Probabilities
We'll
call the average number of "hits" the "expected
value," and use the symbol "y" to represent it.
Using the Poison distribution, we can then calculate the probability
of a certain number of hits, "r," in a particular stretch
of road over a period of time. The equation looks like this:
P(r) is the probability of
exactly r events, and y is the average number of occurrences of
that event (Ref 1).
Assuming that ESD events are
as random as the stones that hit your windshield, we can apply
this formula to an ESDcontrol program and try to calculate the
probability of zero ESD hits, or r = 0. The formula then becomes
very simple:
P (0) = e^{y}
The next step is
to make this useful. First, suppose that y is the average number
of electrostatic discharges in your facility or process without
ESD control. This is a definition of how "bad" your
environment is. Second, define factor "C," the percentage
of ESD hits the control program prevents, which is a measure of
its effectiveness. Finally, compute the factor "S,"
the percentage change in the probability of ESD damage. This is
a measure of the savings a program generates. The resulting equation
will look like this:
Fig 1 is a plot of
this equation for various values of y the average number of
ESD "hits" in the process. We can see that even in an
ESDfree environment (where y < 0.1) we need an ESDcontrol
program with C = 90% effectiveness to achieve a savings of S =
90% of ESD losses. In a situation where there is a high number
of hits on the average (y = 20), a 99.5% effective program is
needed to achieve 90% savings.
The Payoff
Even if we are not
able to calculate the exact value of y, the message is clear.
A halfhearted ESDcontrol effort won't do, especially in the
face of conditions conducive to ESD events. Such conditions include
low humidity, extensive use of plasticpackaging materials and
stateoftheart electronic components that are easily damaged.
The relationship
between how much you spend on ESD control and how much you are
able to save is a nonlinear one, and unless you produce a program
that is more than 90% effective, you won't get the kind of savings
you desire. But a person that takes ESD control seriously will
find the rewards can be financially plentiful.
The
Answer is Yes

Given the nonlinear
relationship between the cost of ESD control and the savings
that can result from it, some will ask whether it's worth
the effort to establish the thorough control program that
the statistics suggest is needed. The answer will almost
always be yes, and to show why, here are the following examples.
IC Manufacturing. Assemblyandtest
yield of a highspeed, bipolar LSI device rose from 22%
to 100% when conductive containers and operator wriststrap
grounders were adopted.
Assemblyandtest yields losses of NMOSLSI circuits were
cut by 50%, and an 800% returnoninvestment was realized
in the first year of ESD control.
Meanwhile, MOSIC yield losses were reduced by 50% at final
test through an ESDcontrol program.
Electronics Manufacturing (Stores and
Incoming). After implementation of an ESDcontrol
program, the storage of bipolar transistor arrays in triboelectrically
chargeable packaging was found to have caused 13% production
failures and 1.5% fieldwarranty failures.
Introduction of an ESDcontrol program at incoming test
cut the reject rate for all lowpowerSchottkyTTL ICs from
0.35% to 0.11%.
Incoming inspection reject rates for linear ICs (op amps)
fell from 2.6% to 0.26% when both vendors and users adopted
ESD control.
Electronics Manufacturing (Assembly and Test).
Failure analysis showed that 60% to 80% of production rejects
of CMOS logic may have been due to ESD damage.
When an ESDcontrol program was adopted, 79% of beamlead
diode failures in hybrid circuitry disappeared.
Reject rates on bipolar digital logic dropped from 0.18%
to 0.11% after an ESDcontrol program was adopted in a computermanufacturing
area.
Electronic Hardware (Field Reliability).
Fieldwarranty failures of a microwave product dropped
30% in the first nine months with ESDcontrol.
Warranty failure rate of a product using a large ratio of
CMOS and lowpowerSchottkyTTL ICs dropped 50% after a
twodiode protection circuit was added to every input and
output.
An independent telephone system supplier experienced a 66%
reduction in fieldfailure rate after service and installation
personnel adopted ESD control.

References
1. C. Lipson and N.J. Sheth,
Statistical Design and Analysis of Engineering Experiments,
McGrawHill, 1973 (pp. 5459).