When we introduce a new chip, we plan and execute a comprehensive reliability qualification plan. This plan will be based on many different reliability stresses addressing infant mortality rate test, early life failure rate test, long term life test failure rate prediction based on a small population of samples pulled from early production lots .
Due to the fact of limited device sample sizes, we are trying to assign a confidence level to our failure rate predictions using “industry standard” chi-square adjustment in the hope, our prediction will be closer to real field failure rates.
This is a “standard” approach of the semiconductor industry because testing very large sample sizes of chips is economically not feasible, especially for small-and fabless semiconductor companies.
IBM Corp.’s Semiconductor Division calls the above practice “finding the tip of the iceberg  “only indicating if there are major catastrophic failure mechanisms”. IBM and major semiconductor manufacturers are stressing large sample sizes in ongoing reliability testing of the outgoing device population.
Above approach requires capabilities and facilities for ORT (ongoing reliabiliy testing) of tens of thousands of devices per year. Only major dedicated manufacturers do this
(like Intel, National Semiconductor, Micron, etc. )
In the course of 2-3 years of intensive ongoing reliability testing of samples of the outgoing population combined with field failure information will one be able to make reliability assessment and meaningful prediction of the maturing semiconductor product.

An adaption of the Functional Safety standards IEC 61508 and IEC 26262 by the European Union brought a new life into slowly fading activity of reliability prediction. Both reliability prediction and reliability demonstration are now key parts of many product development programs, however despite phonetic similarity those two have little in common as well as the result they generate. 

While reliability prediction is an analytical activity often based on mathematical combination of reliabilities of parts or components comprising the system; reliability demonstration is based on product testing and is statistically driven by the test sample size.  Therefore the obtained results could drastically differ.  For example, a predicted system failure rate of 30 FIT (30 failures per billion (109) hours) would corresponds to a 10 year reliability of 99.87% (assuming 12 hours per day operation).  In order to demonstrate this kind of reliability with 50% confidence (50% confidence is considered low in most industries) one would need to successfully test 533 parts (based on binomial distribution) to the equivalent of 10 year field life.  Needless to say that this kind of test sample is prohibitive in most industries.  For example in the automotive electronics the test sample size of 23 is quite common, which roughly corresponds to 97% reliability with 50% confidence. 

The natural question is: how do you reconcile the numbers obtained from reliability prediction with the numbers you can support as part of reliability demonstration?

The answer is: I don’t believe that you can.

You can make an argument that reliability demonstration produces the lower estimate values.  Additionally the test is often addresses higher percentile severity users, thus the demonstrated reliability for the whole product population will likely be higher.  However, in most of the cases the gap will remain too wide to close.  This is something, which reliability engineers, design teams, and most importantly customers need to be aware of and be able to deal with as part of the product development reality.

What does the audience think? We’d love to hear your opinions on this.

Andre Kleyner

This is an application where two Zener diodes were placed in series, in a back to back configuration.  They were placed across the primary winding of a transformer used to apply modulation to an AM transmitter.  The modulation was in the form of a single frequency, and the modulation level was not to exceed 30% by specification. The modulation level showed some variation with temperature, so that the diodes were selected to limit the voltage across the transformer primary to ensure that the 30% modulation limit would not be exceeded.

The drive for the transformer/limiter combination was from a low impedance source, adjusted to provide 30% modulation peaks during final test.  The pole mounted transmitter was required to operate in all weather, unsheltered conditions, at any airport in the United States.

The designer, who was an outside consultant, made some assumptions:

  1. The peak voltage for 30% modulation was 12 Volts, so two 11 Volt diodes were placed in back to back series configuration where one would provide an 11 volt drop in the reverse direction, and the other would provide a 1 volt (approximately) drop in the forward direction, meeting the 12 Volt requirement;
  2. The diodes operate at low dissipation because they are non-conducting, except for excessive peak conditions;
  3. The Reliability and Component Engineering functions of the company could be bypassed because they found too many things wrong, and their input cost too much.

Then there was real life:

  1. Field returns with discoloration on the circuit boards under the diodes;
  2. Field returns with no modulation;
  3. Field returns with modulation intermittently greater than 30%;
  4. Field returns with modulation intermittently low;
  5. By the time the field returns were on the receiving dock, the consultant was long gone.

The returns were turned over to the Failure Analysis lab of the Components Engineering function, and the design was examined by a Reliability Engineer and a Components Engineer.

The diode characteristics were determined to be:

VZ = 11V ± 5% @ 23 mA and TJ = 25°C;

PD = 1 Watt, maximum;

Temperature coefficient of voltage = +0.06%/°C typical;

VF = 1.2 V @ 200 mA, maximum.


Since the diode forward voltage drop would be expected to be considerably lower at low current, the diode forward drop could reasonably be assumed to be approximately 1 volt making the total drop of the diode set approximately 12 volts in either polarity.  At first glance, the initial design assumptions appear to be reasonable.

A simple tolerance analysis begins to show the problem.  The ±5% tolerance on the zener voltage equals 550 mV, placing the zener voltage in the range of 10.45V to 11.55V.  If we assume that the diode forward drop remains constant at 1V, the series combination can have a total voltage drop range of 11.45V to 12.55V.

A further complication is the temperature coefficient of the Zener voltage.  The operating temperature range over all US airports is on the order of -55 °C to +55 °C.  The temperature coefficient applies to the value of the zener voltage at the tolerance extremes, yielding two values of Zener voltage at each temperature extreme.

The calculation for the zener voltage over temperature is straight forward:

VZ(at temp) =  V+ Tempco * VZ *  ΔTemp

Calculation at low and high tolerance and low and high temperature yields four values:

Zener Voltage over Tolerance and Temperature




Zener Voltage







The forward biased diode is also affected by temperature.  It has a temperature coefficient of voltage of -2mV/°C, which yields a change in the forward voltage drop that is opposite in polarity to the change in the zener diode.

Combining the forward voltage over temperature with the zener voltage over tolerance and temperature yields the clipping voltage:

Clipping Voltage over Tolerance and Temperature




Zener Voltage







Based on this, the new clipping voltage range is 11.108V to 12.698V.

Further conditions that affect the temperature range over which clipping begins are internal temperature rise in the box, and direct heating of the box by sunlight.  Considering these would unnecessarily complicate this discussion.

Production test:

In order to reduce costs, the production boxes were only tested at room temperature.  Two engineering units were successfully temperature tested for type approval, and on that basis, full temperature testing was waived on the production units.

Boxes with diodes that had clipping voltages of 12 V and higher sailed through test without problem, since the modulation voltage could be set to peak at 12 V, meeting the 30% modulation requirement.  In the field, boxes that operated at elevated temperatures operated normally, with no diode failures.  These same boxes, when operated at low temperature, depending upon the exact clipping voltage, frequently failed for low modulation because their temperature coefficients forced the clipping voltage below 12 V.  In a few cases, the self-heating of the diodes due to the dissipation from clipping allowed their clipping voltages to reach thermal equilibrium near 12 V, allowing the boxes to operate satisfactorily.  The modulation level was a function of the ambient temperature, causing intermittent failures on some boxes.

Boxes that contained diodes with clipping voltages below 12 V, for the most part, also sailed through test without problem.  This seems unreasonable, since it should not have been possible for those boxes to be set at 12 V.  Here, the positive temperature coefficient of the zener voltage came to the rescue.  As the test technicians increased the modulation toward 30%,  diodes that broke down below 12 Volts began to conduct, thereby warming up, raising their breakdown voltage.  With enough drive, many of the diodes could be driven hard enough to reach thermal equilibrium at 12 Volts.  Typically, most of these diodes were heavily over dissipated.  When these boxes reached the field, they operated for a time, but eventually, the diodes began to get leaky, leading to increased dissipation and subsequent short.  Some of these boxes also displayed intermittent failure of the modulation level, due to the changes in ambient temperature.


The basic design approach to limiting the modulation level was flawed.  The diode clipper was intended to limit the temperature variability of the modulation source.  Instead, it induced additional temperature sensitivity and a high failure rate.  This approach was deemed to be less expensive than a design where the modulation level was actively sensed, and feedback applied to control the modulation source.

The initial assumptions did not take into account the zener diode voltage tolerance and temperature coefficient.  This is the root cause of all of the failed units returned from the field.  It is interesting to note that whether the symptom was low modulation, high modulation, or burned boards and shorted diodes, the source of the failure is traceable to the same source.  Inattention to and/or lack of understanding of the basic operating parameters of the zener diode was that source.

Lessons learned:

  1. Read and understand the datasheet;
  2. Bypassing oversight is more expensive in the end;
  3. Using engineering units for type approval carries the risk that the units may not represent the production items;
  4. Validation of the qualifications of your consultant is paramount to success.

Zener diodes are among the most misunderstood and misused parts in electronic systems.  How can this be?  You might ask.  The diode only has two connections, run some current through it, and it simply clamps to a specific voltage.  Couldn’t be easier.  End of story.


This is the tale of a 50 mHz. low noise crystal controlled oscillator intended for a commercial spacecraft application.  The electrical specifications of the oscillator were tight; the size, weight and power budgets were even tighter.

Extreme stability requirements necessitated the use of a heater.  To save on power, weight and size, the manufacturer opted to heat only the crystal.  The main oscillator circuit was contained in a small hybrid package.  The hybrid, the heater controlled crystal and some tuning capacitors were mounted on a small printed wiring board.  The oscillator was powered by a simple shunt regulator using a 9.1 volt Zener diode mounted inside the hybrid, with a dropping resistor mounted on the board, feeding from a 12 volt regulated power supply.

Very simple, run some current through the diode and it clamps at the specified voltage.  Couldn’t be easier.

Another characteristic of the oscillator was that it had to turn on at -20⁰C and reach stability within fifteen minutes.  Reaching stability in the allowed time was no problem as the mass of the crystal and the heater assembly was small, and the operating temperature was reached in less than two minutes.

The operating frequency, on the other hand, was an issue.  In order for the oscillator to operate at 50 mHz., the crystal had to operate in overtone mode.  Overtone crystals are notorious for starting at some frequency other than the expected one.  With careful tuning, the oscillator could be made to start at the correct overtone.  Some crystals have spurious modes very near the expected frequency, and can start at one of those frequencies.  Once started, the crystals do not shift to their expected frequencies unless power is removed and the oscillator restarted.  There is no guarantee that the oscillator will come up at the expected frequency on restart.

As a result of this characteristic, each oscillator had to be tuned to match each crystal.  Since operation with a frequency error of even a few ppm would be catastrophic in the application, the spacecraft manufacturer required that each oscillator be tested for turn on and stabilization at -20⁰C.

The oscillators performed flawlessly on turn on and stabilization time at the manufacturer’s facility, so they were shipped.   The user retested the turn on and stabilization characteristics prior to installation in the next higher assembly.  Again, the oscillators performed flawlessly.


To ensure that there were no discontinuities in the temperature performance, the user brought the temperature up in 5⁰ increments while monitoring the oscillator output.  At approximately -5⁰C there was a sudden large burst of noise superimposed on the oscillator output.  This noise was evident until the oscillator temperature reached 0⁰C, whereupon the output noise dropped back into specification.  When the temperature was decreased, the noise reappeared.  This was very repeatable.

So, what happened?

The designer, pressed for power, decided to take advantage of the fact that the oscillator was to be powered from a regulated supply.  He knew that the Zener diode test current was specified to be 20 mA.  He also knew that the reason for biasing the diode at the test current was to minimize the Zener impedance, so that current variation would have minimal effect on the diode voltage.  He reasoned that with a regulated power supply, he didn’t have to concern himself with current variation due to changes in the input voltage, and further, the load was constant, so that he didn’t have to worry about current changes from that source.

The oscillator circuit drew 15 mA without the Zener diode.  It seemed a shame to waste more current in the diode than the circuit required for operation, and since the source voltage and load current were both constant, the designer squeezed out a power saving by reducing the entire current to 20 mA., 15 for the oscillator and 5 for the diode.

So far, so good.  Everything worked.  The oscillator turned on correctly at cold temperature.  Ship!!!

What did the designer miss?

He neglected to find out how much current the oscillator required at -20⁰C.  The oscillator current rose as the temperature went down, until the point was reached that all the current was flowing in the oscillator, and none in the diode.

Zener diodes have a peculiar, but not too well known, characteristic.  They are inherently noisy, but in the vicinity of the turn-on knee, they become extremely noisy.  The so called “noise diode” is nothing but a certain type of Zener reference diode that has been tested to have a specified noise spectrum under specified bias conditions.  Some series of Zener diodes include noise information on the datasheets, but only at the test current.  The diode used in this application had no noise specification.  Even if it had one, operation at the turn-on knee would not have been controlled.

The oscillator was quiet between -20⁰C and -5⁰C because the Zener diode was turned completely off due to lack of bias current.  At about -5⁰C the diode was just crossing the threshold of operation, generating significant noise.  By the time that 0⁰C was reached, the diode was on and operating normally.  This was confirmed by operating the oscillator as room temperature and reducing the supply voltage until the diode dropped out.  The same noise characteristic was noted at the threshold point.  Several oscillators were temperature tested, with the same results.  The only variation being the temperature at which the noise burst occurred.  Several oscillators were tested at reduced voltage at room temperature.  They were found to work well, even down to five volts.

Lessons learned.

  • Temperature testing  involves sweeping the entire range, not just testing at the extremes.
  • Understand the temperature performance of each part of your product.
  • The care and feeding of the Zener diode is paramount for a successful application.  A well fed Zener is a happy Zener!

Potting compounds are often used on printed circuit boards to improve reliability.  In spite of this added protection, adhesive failures, otherwise known as delamination, can occur and lead to substantial problems from moisture ingress.  Root causes of adhesive failures may be surface contaminants, inherently weak bonding between solder mask and potting material or thermal stresses that develop during temperature cycling.

In an adhesion analysis a combination of chemical and mechanical tests is often needed to determine the root cause.  Chemical testing includes surface analyses such as XPS and SIMS; both reveal chemical groups available for bonding at the surface.   Contaminants may be revealed during these tests.  Basic tests such as FTIR, TGA and EDS provide information on the materials including fire retardants whose loadings are typically quite high, affecting adhesive bonding.  Mass spectrometry can be used to identify suspected contamination areas.  Ion chromatography is often used to identify weak organic acids, a byproduct from incomplete volatilization of no-clean flux.   For determining an inherent compatibility of bonding surfaces, contact angle measurements using standard solvents, and surface tension measurements using the pendant drop method, can be very valuable.

Chemical testing must be done in conjunction with mechanical testing.  The aim of the mechanical test is to reproduce the failure mode in the laboratory under controlled conditions.  The easiest mechanical tests to perform are lap shear or peel strength tests.   The lap shear test is straightforward to fixture and provides a method for comparing bond strengths among different boards or potting materials.  A more sophisticated technique, the four-point bending test using pre-notched specimens, can be used to quantify adhesion energy, Gc.


For more information see:

Firas Awaja, Michael Gilbert, Georgina Kelly, Bronwyn Fox, Paul J. Pigram, Adhesion of polymers, Progress in Polymer Science 34 (2009) 948–968

Study of interfacial adhesion energy of multilayered ULSI thin film structures using four-point bending test, Zhenghao Gana, S.G. Mhaisalkara, Zhong Chena, Sam Zhangb, Zhe Chenc, K. Prasad, Surface & Coatings Technology 198 (2005) 85–89