You are on page 1of 4

RELIABILITY AND QUALITY TESTING OF MATERIALS

B.K. Jones

Department of Physics, Lancaster University,


LANCASTER, LA1 4YB, UK
b.jones@lancaster.ac.uk

The quality and reliability of electronic devices are now routinely determined by various
destructive and non-destructive tests. Although electronic systems are usually easier to
study using electrical tests there is a good possibility to use similar methods on materials
and mechanical systems. Such methods are found to be valuable to increase the reliability
of devices by selecting very good devices and rejecting bad devices. This enables the
manufacturing process to be improved and also has the advantage of increasing the yield.

Keywords: Reliability indicators, yield, statistical process control, quality.

INTRODUCTION
There is a general need to improve the quality and reliability of all materials and products. This
leads to general customer satisfaction as well as increased yield in the production and manufacture
process. Since final inspection, with rejection or correction, and the later customer returns are very
expensive, the monitoring of the material and product at early stages of manufacture is important and
cost-effective. The quality and reliability of electronic devices of all sorts are now routinely
determined by various destructive and non-destructive tests. This is very successful. A very high
reliability is necessary for each component to obtain a low system failure rate since electronic systems
contain a very large number of components and the system failure rate is proportional to the number of
components and their individual failure rates. We describe here the methods that are used for this
quality test and discuss the ways that similar methods can be used for non-electronic systems. It should
be noted that most of the failure mechanisms for electronic systems are, in fact, physical and even
mechanical with such familiar effects as cracks, hot spots, corrosion, crystal growth, etc..
Although electronic systems are usually easier to study, using electrical tests, there is a good
possibility to use similar methods on materials and mechanical systems. The main extension would be
to introduce suitable transducers for the tests. Such methods have been found to be valuable to increase
the reliability of devices by selecting very good devices and rejecting bad devices. As well as the use to
select special devices for high reliability applications. This enables the manufacturing process to be
improved and also has the advantage of increasing the yield.
We need to define roughly what we need from our tests. ‘Quality’ is a production quantity with
some minimum acceptable value of each parameter and then, usually, a defined small statistical spread
of the values in order to obtain consistency rather than trying to achieve the maximum which can be
attained. ‘Reliability’ is a users’ term which implies that there should be no failure or significant
change in the important quantities throughout the normal lifetime and under normal use.
TESTS
There are many types of test which can be used to evaluate the performance of samples.
A functional test is a test for its fitness for its purpose. It is the minimum acceptable if the
customer is to receive a working device. It could just be a try to see whether it performs to some
specification and would then be a ‘Pass/Fail’ test. If a high resolution measurement can be made of
critical parameters then there can be pass/fail criteria on the most important quantities but production
histograms can also be produced on all the measured quantities and the mean and standard deviation
can be produced for each batch or day’s production. This then leads to Statistical Process Control
(SPC) which can be used to monitor the process. Thus for resistors the simplest test would be a
measure of the resistance which has to be within a certain tolerance of the prescribed value. A more
precise measurement would give the value for each sample which would be labelled with the
appropriate preferred value and tolerance. When all the values of production are recorded and a
statistical analysis done then one has a very good understanding of the consistency of the process and
can detect departures from the ideal with early warning to allow rapid correction of the production
process. For a transistor there may be 10 or 20 parameters measured and then some selection for
different particular qualities such as gain or leakage current and some parameters are found to be
particularly sensitive and appropriate for sensing the quality of the process. For a complex device such
as a digital integrated circuit it is impossible to do a full functional test so that tests of critical parts are
performed on a sample basis and also more general tests which we will discuss later.
Gross manufacturing faults are frequent causes of failure to pass the functional test. It is also
commonly found that these lead to failure in the first few hours/days/months of use. One way to allow
for this is to give a return warranty to customers. Another method is to burn-in each unit. Then it is
operated for a period by the manufacturer to reveal these early problems. This is expensive so that
now strong attempts are made to eliminate manufacturing faults completely by tight monitoring of
process parameters by SPC.
In the limit the individual test parameters can be taken up to a high value at which there is
permanent change or failure. This stress to failure is destructive and can only be performed on samples.
Since the failure point is well above the normal operating value this gives only a measure of some
quality of production but is valuable as a process control parameter.
A more realistic time to failure is a test to see how long a sample will keep within its
specifications while being used within the manufacturer’s use conditions. This is then a measure of its
useful life for the customer. However this is only on a sample basis and will take an extremely long
time so it is of little practical use. Normally an accelerated time to failure is used in which the
parameters of use are increased to, or above, the specified operating limits to decrease the life to a
usable time. To obtain the normal lifetime the accelerated value is extrapolated to the normal working
stress by doing experiments at varying stress levels. This will still take a very long time.
Many of the problems of destructive tests on samples which take a long time can be decreased
by the use of reliability indicators (RIs). These are quantities which are found to vary rapidly with a
decrease in quality well before any failure limit is reached. They therefore provide an indication of
quality in an extension to the functional test and their changes indicate a degradation before failure for
the lifetime or failure tests. The advantage is that they are non-destructive and so can be used on all
production devices as well as on a sample basis. The problem is to identify a suitable parameter and to
be able to measure it with sufficient precision and confidence. Once some experience has been gained
it is possible to use these changes to detect early changes in the degradation and hence extrapolate these
to a low stress or faster accelerated time to failure test. For semiconductor devices these indicators are
not difficult to find and two are found to be very sensitive; the leakage current of a junction and low
frequency noise.

EXAMPLES
As an example of what can be done in electronic systems we will use the failure mechanism of
electromigration. This is the open circuit failure of the fine metal lines which connect the transistors in
an integrated circuit. The current density is very high so that in the resistance process momentum is
given to atoms by the electron current. Where there is an atom flux divergence there will be a local loss
of atoms and eventually a void will form and grow to break the conductor. The normal industrial test
of performance is to measure the lifetime under extremely high current stress at an elevated
temperature. This is destructive, but fast.
Various reliability indicators can be found from high resolution resistance measurements.
Figure 1 shows a typical resistance change with time under moderately high temperature and current
stress. Naturally it is not practical to reduce the acceleration to a very low level but a life of about two
weeks is acceptable. It can be seen that the resistance rises so that the rate of resistance change near the
beginning of the stress can be used as a reliability indicator which can be extrapolated to the change at
the end of life, a few percent. As well as this fairly uniform change in resistance over the whole of the
life there are violent changes near the end of the life. These are dynamic void effects. Excess electrical
noise is in fact the fluctuations of the resistance so that electrical noise is a good reliability indicator. It
can predict the later failure of the sample and can also indicate a badly made sample. It is found that
whereas the resistance may change by a few percent the noise will change by many orders of
magnitude. This type of noise, which is indicative of failure or poor quality, normally has a spectrum
which increases towards low frequencies. It is thus a very sensitive reliability indicator. Since these
are very early and sensitive indicators they are non-destructive in the sense that changes can easily be
detected after a stress which is low enough that very little damage has been done and the sample is well
within its specification. .
The range of reliability indicators for electromigration is not complete. The best method of
measurement of the resistance change is to use an AC bridge. If the bridge signal amplitude is large
there will be Joule self-heating at twice the AC drive frequency. The size of the resulting harmonic
signal can be large and give more information of the degradation process. Also if the frequency is
comparable to the reciprocal thermal decay time for heat dissipation from the metal through the
insulating layer to the semiconductor substrate then any phase-shift or delay change is an indication of a
change in the thermal contact. It is a measure of the delamination of the metal from the substrate.
Some examples of such events are shown in Figure 2.
It is apparent that sensitive tests are easier for electronic systems than mechanical ones and also
that the cost and time needed to make extensive measurements and analysis are more effective for high
value products than low value ones. However this should not deter people from attempting to improve
the quality and reliability of products. Unfortunately it is not possible to generalise and each product
will need different experimental methods.
The simple, basic measurements for SPC are probably obvious and a decision needs to be made
of the most useful ones and the value of the extra cost needed to make them and what use can be made
of the results. We will therefore concentrate on possible reliability indicators.
Electrical measurements are easiest but if the electrical properties do not reflect the desired
mechanical properties then the use of some appropriate transducer will be needed. As with electrical
systems ‘noise’ can be valuable. By noise we mean random or semi-random fluctuations in a quantity.
Randomness is a measure of uncertainty and this is related to quality, reliability and stability. For
mechanical systems the acoustic noise is most obvious.
Non-destructive methods such as the acoustic noise which can be valuable in a first test of an
engine or bearing. It is also of value in the first stages of breaking in a metal or a composite structure.
As small breaks occur the strain energy is released as acoustic vibrations. Similar discontinuous
variations of other quantities such as the resistance or capacitance can be expected. A spectral analysis
can diagnose the important frequencies for the degradation. The thermal conductance of joints can be
used to detect breaks. Electrical fluctuations in the resistance between a metal and an electrolyte can be
used to monitor the corrosion process. Defects well below the surface can be detected by eddy currents
and ultrasonic methods. Perhaps the noise in the signal could be studied when the sample is subjected
to light loads.
R amplitude changes (% )
6

5 FFT C
4
FFT B
3
FFT D
2

1 FFT A

-1
0 200 400 600 800 1000 1200 1400 1600 1800

Stress time (ks)

Figure 1. The resistance variation during the full lifetime of sample P2D6(2) stressed with J=2.51
MA/cm at 2560C. The slow resistance variation indicates gradual degradation and can be used as a
2

predictor of the time to failure. The various time periods shown are used to calculate an FFT spectrum
as a measure of the resistance fluctuations. It can be seen that these increase in intensity with time and
hence damage.

6 40

4 20
/ (%)
R/R (%)

2
0

0
-20
0 20 40 60 80 0 20 40 60 80
stress time (ks) stress time (ks)

150
delay(s)

100

50

0 20 40 60 80
stress time (ks)

Figure 2. Changes during the full lifetime of sample P3A20 to show that there are large changes in the
amplitude of the second harmonic and these and the delay in the second harmonic is not reflected in
changes in the resistance. The change in the delay or phase-shift is ascribed to delamination.

You might also like