Devices are known to pass qualification testing, then fail in the field. Does that suggest the test specifications are inadequate?

“Experiment without a theory is blind. Theory without an experiment is dead.” – Unknown reliability physicist

Shortening a product’s design and development time in today’s industrial environment typically precludes time-consuming reliability investigations. To get maximum reliability information in minimum time and at minimum cost is the major goal of an IC package manufacturer. On the other hand, it is impractical to wait for failures, when the lifetime of a typical electronic package today is hundreds of thousands of hours. Accelerated testing is therefore both a must and a powerful means in electronics manufacturing.1

The major AT types are summarized in Table 1. Product development testing (PDT) is a crucial part of design for reliability (DfR). A typical example is shear-off testing, when there is a need to determine the most feasible bonding material and its thickness.

[Ed.: To enlarge the figure, right-click on it, then click View Image, then left-click on the figure.]

Highly accelerated life testing (HALT) (see, e.g., Suhir et al2) is widely employed, in different modifications, with an intent to determine the product’s design and reliability weaknesses, to assess its reliability limits, to ruggedize the product by applying stresses (not necessarily mechanical and not necessarily limited to the anticipated field stresses) that could cause field failures, large (although, actually, unknown) safety margins over expected in-use conditions. HALT is a “discovery” test, while it is the qualification testing (QT) (see, e.g., Suhir3) that is the “pass/fail” one and, as such, is the major means for making a promising and viable device (package) into a reliable and marketable product.

QT brings to a “common denominator” different manufacturers and different products. When it comes to manufacturing, however, mass fabrication generates, in addition to desirable-and-robust (“strong”) products, also some amount of undesirable-and-unreliable (“weak”) devices (“freaks”), which, if shipped to the customer, will most likely fail in the field.

Burn-in testing (BIT) is supposed to detect and eliminate such “freaks,” so that the final bathtub curve of a product that underwent BIT does not contain the infant mortality portion. In today’s practice, BIT, which is a destructive test for the “freaks” and a nondestructive test for healthy devices, is often run within the framework of and concurrently with HALT.

Despite all the above AT effort, devices that passed the existing QT often fail in the field. Are these QT specifications and practices adequate? If not, could they be improved to an extent that for a product that passed the QT and survived the appropriate BIT, there is a quantifiable and sustainable way to assure that it will perform in a failure-free fashion in the field? It has been suggested4 that probabilistic design for reliability (PDfR) concept is used to create a “genetically healthy” product. The concept is based on recognition that reliability starts at the design stage, that nothing is perfect, and that the difference between a highly reliable and an insufficiently robust product is “merely” in the level of the probability of its failure. If one assesses, even tentatively, the probability of failure in the field and makes this probability sufficiently low, then there will be a reason to believe that a failure-free operation of the device will be likely (“principle of practical confidence”). With this in mind, a highly focused and highly cost-effective failure-oriented-accelerated testing (FOAT), which is the heart of the PDfR concept, should be conducted in addition to and, in some cases, even instead of HALT. FOAT is a solid experimental foundation of the PDfR approach. The prediction might not be perfect, but it is still better to pursue it than to turn a blind eye to the fact that there is always a non-zero probability of the device failure. 

Understanding the underlying reliability physics is critical. If one sets out to understand the physics of failure in an attempt to create a failure-free product, conducting FOAT should be imperative. Accordingly, FOAT’s objective is to confirm usage of a particular more or less established predictive model (PM), to confirm (say, after HALT is conducted) the underlying physics of failure, to establish the numerical characteristics (activation energy, time constant, exponents, if any, etc.) of the particular reliability model of interest.

Here are some well known FOAT models:

Arrhenius’ equation and its numerous extensions and modifications used when there is evidence that the elevated temperature is the major cause of the material or the device degradation (lifetime of electrical insulations and dielectrics, solid state and semiconductor devices, inter-metallic diffusion, batteries and solar cells, lubricants and greases, thermal interface materials, plastics, etc., as well as reliability characteristics other than lifetime, such as, e.g., leakage current or light output).
Boltzmann-Arrhenius-Zhurkov’s (BAZ)5 can be used when the material or a device experience combined action of elevated temperature and external loading; Crack growth models are used to assess the fracture toughness of materials in the brittle state. Inverse power law is used in numerous modifications of the Coffin-Manson’s semi-empirical relationships aimed at the prediction of the low cycle fatigue lifetime of solders that operate above the yield limit. Miner-Palmgren’s rule is used to address fatigue when the elastic limit is not exceeded. Weakest link models are used to evaluate the lifetime in extremely brittle materials, like Si, with highly localized defects. Stress-strength interference models are widely used in various problems of structural (physical) design in many areas of engineering, including microelectronics. Eyring-Polanyi’s equation is used to evaluate the lifetime of capacitors and electromigration in aluminum conductors. Peck’s equation is used to evaluate the lifetime of polymeric materials and the effect of corrosion. Black’s equation is used to quantify the reliability in electromigration problems, to evaluate the lifetime of hetero-junction bipolar transistors and the role of humidity. It is important to emphasize that all these models can be interpreted in terms of the probability of failure under the given loading conditions and after the given time in operation. A bathtub curve is a good example of a FOAT. If this curve is available, then many useful quantitative predictions could be made (see, e.g., Suhir6). As another example, let us consider an IC package whose steady-state operation is determined by the Boltzmann-Arrhenuis law Here τ is the lifetime; τ0 is the time constant; U is the activation energy; T is the absolute temperature and k is Boltzmann’s constant. The probability of the package non-failure can be found, using an exponential law of reliability, as Solving this equation for the T value, we have:  . Addressing, e.g., surface charge accumulation failure, for which  assuming that the FOAT predicted time factor τ0 is τ0 = 2x10-5 hours, that the probability of failure at the end of the device’s service time of τ = 40,000 hours should not exceed Q = 10-5, the above formula yields: T = 352.3⁰K = 79.3⁰C. Thus, the heatsink should be designed accordingly. More complicated examples of FOAT and design decisions based on it can be found in Suhir3-8.

An extension of HALT. FOAT could be viewed as an extension of HALT. It should be employed when reliability is imperative, and therefore, the ability to quantify it is highly desirable. HALT is, to a great extent, a “black box”, i.e., a methodology that can be perceived in terms of its inputs and outputs without a clear knowledge of the underlying physics and the likelihood of failure. FOAT, on the other hand, is a “white box,” whose main objective
is to confirm usage of a particular predictive model that reflects a specific anticipated failure mode. The major assumption is, of course, that this model is valid in both AT and in actual operation conditions. HALT does not measure (quantify) reliability; FOAT does. HALT can be used, therefore, for rough tuning of the product’s reliability, while FOAT should be employed when fine tuning is needed, i.e., when there is a need to quantify, ensure and, if possible and appropriate, even specify the operational reliability of the device. HALT tries, quite often rather successfully, to kill many unknown birds with one stone. HALT has demonstrated over the years its ability to improve robustness through a “test-fail-fix” process, in which the applied stresses are somewhat above the specified operating limits. By doing that, HALT might be able to quickly precipitate and identify failures of different origins. HALT often involves step-wise stressing, rapid thermal transitions, etc. Since the principle of superposition does not work in reliability engineering, both HALT and FOAT use, when appropriate, combined stressing under various stimuli. FOAT and HALT could be carried out separately, or might be partially combined in a particular AT effort. New products present natural reliability concerns, as well as significant challenges at all the stages of their design, manufacture and use.  HALT and FOAT could be especially useful for ruggedizing and quantifying reliability of such products. It is always necessary to correctly identify the expected failure modes and mechanisms, and to establish the appropriate stress limits of HALTs and FOATs to prevent “shifts” in the dominant failure mechanisms. There are many ways this can be done (see, e.g., Suhir9).


The FOAT-based approach, which is, in effect, a “quantified and reliability physics oriented HALT,” should be implemented whenever feasible and appropriate, in addition to the currently widely employed various types and modifications of the forty-years-old HALT. In many cases the FOAT-based effort can and should be employed, even instead of HALT, especially for new products, whose operational reliability is unclear and for which no experience is accumulated and no best practices exist. The approach should be geared to a particular technology and application.10


1. E. Suhir, “Reliability and Accelerated Life Testing,” Semiconductor International, Feb. 1, 2005.
2. Intertek company website,
3. E. Suhir, R. Mahajan, A. Lucero and L. Bechou, “Probabilistic Design for Reliability (PDfR) and a Novel Approach to Qualification Testing (QT),” 2012 IEEE/AIAA Aerospace Conf., 2012.
4. E. Suhir, “Probabilistic Design for Reliability,” Chip Scale Review, vol. 14, no. 6, 2010.
5. E. Suhir, “Assuring Aerospace Electronics and Photonics Reliability: What Could and Should Be Done Differently,” 2013 IEEE Aerospace Conference, March 2013.
6. E. Suhir, “Remaining Useful Lifetime (RUL): Probabilistic Predictive Model,” International Journal of PHM, vol. 2(2), 2011.
7. E. Suhir, “Predictive Modeling is a Powerful Means to Prevent Thermal Stress Failures in Electronics and Photonics,” Chip Scale Review, vol. 15, no. 4, July-August 2011.
8. E. Suhir, “Applied Probability for Engineers and Scientists,” McGraw-Hill, New York, 1997.
9. E. Suhir, “Analysis of a Pre-Stressed Bi-Material Accelerated Life Test (ALT) Specimen,” ZAMM, vol. 91, no. 5, 2011.
10. E. Suhir, “Considering Electronic Product’s Quality Specifications by Application(s),” Chip Scale Review, vol. 16, no. 4, 2012.

Ephraim Suhir, Ph.D., is Distinguished Member of Technical Staff (retired), Bell Laboratories’ Physical Sciences and Engineering Research Division, and is a professor with the University of California, Santa Cruz, University of Maryland, and ERS Co.;

Submit to FacebookSubmit to Google PlusSubmit to TwitterSubmit to LinkedInPrint Article