6.3.1. Choice of physical constraints

All previously presented advantages of aggravated tests are not real unless the product is correctly stressed. The choice of physical contributions is therefore essential. Several studies were conducted (IES 1984) on this subject, and their results can be summarized in Figure 6.5 (below).

As it is impossible to simultaneously generate constant temperature and have thermal cycling, the dominant condition is kept, namely, the thermal cycling. In fact, the most important physical contributions (thermal cycling, vibration and power cut) were grouped in a single machine, greatly simplifying these tests, particularly in terms of time and costs.

6.3.2. Principle of HALT

In order to find the operational limits of a product, the approach illustrated in Figure 6.6 is used.

Here, the technological limit is identified as being the maximal values guaranteed by the manufacturer of the components. For example, for a component that is guaranteed up to +85°C, the technological limit of the product will be reached when the temperature level of the aggravated test exceeds this value, even if the product is still fully operational.

HALT testing involves the following process:

• – finding the low-temperature operational limit (LTOL). The starting point is the ambient temperature, which is then decreased in steps of generally 10°C for 10 min. If the product is operational, the temperature is further decreased by 10°C until an operational failure occurs. This can be illustrated by Figure 6.7 (below);
• – finding the high-temperature operational limit (UTOL). The starting point is the ambient temperature, which is then increased in steps of generally 10°C for 10 min. If the product is operational, the temperature is further increased by 10°C until an operational failure occurs. This can be illustrated by Figure 6.8 (below).

NOTE.– It is not surprising that destructive low-temperature limits are rarely reached by electronic materials. They are indeed more sensitive to high operational temperatures than to low operational temperatures.

• – Finding the vibration operational limit (UVOL). The starting point is the vibration in Grms, which is increased in steps of generally 5 Grms for 10 min. If the product is operational, the vibration level is further increased by 5 Grms until an operational failure occurs. This can be illustrated by Figure 6.9 (below).

It is important to note that vibrations take place along 6 degrees of freedom (one translation and one rotation per axis) and that the product being tested is at the ambient temperature. The duration of one step is of minimum 10 minutes, but it may take longer to conduct the required tests. Vibrations generally range between 2 Hz and 5 kHz.

• – Finding the operational limits by a combination of vibrations and thermal cycling.

A combination of constraints, such as vibrations and thermal stresses, can be used to reveal latent weaknesses of the equipment, which do not occur under a single constraint (e.g. temperature only or vibration only).

NOTE.– Moreover, during each of these steps, the package is monitored (controlling the operational tolerance drift and recording the constraints and their responses at the card level) so that defects due to temperature change are detected.

• – Unless it is checked by other tests, it is useful to check the cold start of the equipment at various steps. This is an additional test, which involves checking the start of equipment under stabilized low temperatures, in order to determine its limit. It is distinct from ON/OFF cycles, as the latter are conducted on an initially operating equipment, whose components have a higher temperature than the environment, especially if it is very dissipative.

Figure 6.10 illustrates this process.

NOTE.– The above example of combined cycle does not take into account thermal inertias, which can be generated by a collapse. The thermal profiles (temperature slope/step duration) must be corrected to fit the constraints. The time needed to initialize the equipment and the monitoring software must also be taken into account.

• – The application tolerances of the acceptable constraints are generally of ±3dB in vibration about the rated profile and ±2°C in thermal cycling.

6.3.3. Specific or additional constraints

The nature of the physical contributions or specific stresses to be applied is suited to the tested equipment. It is therefore difficult to formulate a generic list. Moreover, when non-electronic equipment is involved, the constraints and the stresses to be applied through a testing process are defined case by case.

For electronic materials, the following specific tests can be taken into consideration during a testing process, if their relevance is approved during the stage of constraint selection:

• – power cycle;
• – cycle of various supply voltages combined with previous sequences;
• – finding the limit alternative voltages as well as the limit frequencies;
• – finding the maximum direct supply voltages.

For mechanical materials, the following constraints can be taken into consideration:

• – continuous increase of mechanical load;
• – increase of speed or frequency of mechanical loads;
• – incremental increase of the load acceleration.

The above list is just an example, and is not exhaustive. Its purpose is to show the diversity of constraints that can be implemented during an aggravated testing process. The preliminary stage of analysis and determination of constraints and constraints to be implemented during a process is therefore very important.

6.3.4. Number of required samples

According to the referenced literature (McLean 2009), due to the dispersion of results according to the tested products and the representativeness, it is recommended to use four products per robustness test. However, for highly expensive products and provided that the dispersion of the various operational limits observed is very weak, a single sample may be used per type of test.

6.3.5. Operational test, diagnosis and identification of weaknesses

It should be possible to stimulate and test the product throughout the testing process so that reversible malfunctions and breakdowns are detected. Test coverage should be sufficient, and it should be possible to observe the evolution of temperature performances, if the function depends on it. It is highly recommended that we conduct a preliminary analysis of the range of temperatures specified for all the components and their expected performances according to the datasheet provided by the manufacturer. This will help in identifying the operational chains that will probably be impacted first during temperature margin tests and to make sure that the test covers these functions.

6.3.6. Monitoring specification

The set of parameters to be measured and therefore to be monitored during the various testing sequences should be defined before the HALT tests. Indeed, it may prove necessary to adapt certain testing methods and even the equipment itself in order to have or allow a proper monitoring.

It is important to differentiate between:

• – Operational monitoring, which is the monitoring of product functionalities. Operational monitoring of the equipment is required for deciding on the proper operation of the electronics at any moment of the aggravated test. Defects can be evidenced by monitoring analysis, and it is important to know which operational constraint generated which defect. This makes it possible to identify the electronic functions that failed under constraint and to rapidly define the failing component.
• – Recording the stresses perceived by the equipment: a proper instrumentation is a must for the analysis of resulting defects, notably the identification of defects due to reaching the technological limits.

For each test monitoring sequence, the following must be defined:

• – parameters to be monitored;
• – operational constraints to be applied;
• – functions covered by the tests;
• – frequency of tests (every 30 seconds, etc.);
• – type of test (intrusive, such as the injection of breakdowns, specific or mobile probe);
• – number of defects based on which a breakdown is confirmed;
• – inhibition conditions;
• – behavior of the associated product.

6.3.7. Instrumentation

In order to properly grasp the levels of involved constraints, the product subjected to aggravated tests must be instrumented. This instrumentation (thermocouples, accelerometers, constraint meters, etc.) can also be used to instantaneously measure the response of the product to constraints. The implementation of this instrumentation is defined based on risk analyses and focuses on the most sensitive components, therefore, the electronic components. For example, on electronic cards, accelerometers are placed on vibration-sensitive components in order to check their maximal displacements during testing.

Vibrational and thermal profiles must be measured at points where it is possible to reconcile the operational defect of the equipment and the applied constraints, or even to estimate the card damage. These profiles must be recorded by accelerometers and temperature probes whose location guarantees a proper application of the constraint during burn-in, in view of proper traceability.

6.3.8. Root cause analysis, corrective actions and breakdown management

Root cause analysis is an integral part of the aggravated testing process. It offers the certainty that the malfunction is related to the test during which the defect appeared, and not to previous stresses, or that the defect is not related to anything else (damage related to the rig, manufacturing defect, manipulation error, etc.).

The root cause analysis process must be documented with the failure mode, the exact cause of failure or its suspected cause, in case of uncertainty.

Failure analysis involves two significant stages:

1. 1) electrical diagnosis, which confirms and characterizes the electric defect and guides the physical analysis;
2. 2) physical analysis, which identifies the defect location and determines its root cause.

By identifying the root cause of the failure, it is possible to devise appropriate corrective actions, depending on the margin being found. Depending on the type of weakness detected during the tests and characterized by the analysis, corrective actions may involve:

• – changing the reference of a component;
• – modifying the location of a component;
• – locally reinforcing a component (collage, underfilling);
• – reviewing the mechanical design of the card (thermal drain, mechanical maintenance).

A test is considered successful if it highlights as many potential defects of the material as possible. It is important to consolidate all this information, upstream and downstream of the operation, in order to optimize the initially chosen aggravated test. This requires implementing a product to collect information related to failures. This product must be related to the failure modes observed:

• – during the aggravated test;
• – after the aggravated test (material acceptance, client refusal).

Therefore, the information collection may be composed of:

• – the testing reports for each material;
• – the results of the aggravated tests of the materials;
• – the return of experience (RETEX).

A copy of the testing report is attached to each unit subjected to an aggravated test. It contains at least the following elements:

• – identification of the material (designation, reference, serial number);
• – the test result.

For each breakdown observed, the following elements should also be mentioned in the report:

• – the type of cycle during which the defect occurred (e.g. vibration or thermal cycle);
• – the findings;
• – the defective components;
• – corrective actions;
• – the number and type of cycles throughout the stages;
• – the date.

It is important to have a follow-up of corrective actions. Hence, a file containing the results of aggravated tests should be edited regularly. It indicates the number of objects subjected to aggravated tests for each type of material and mentions the causes of breakdowns as well as the actions taken in order to repair the material. In principle, this report should be drawn up by the quality assurance in manufacturing department.

Finally, a report of the equipment returns must also be edited regularly. For each piece of equipment returned, the report should indicate:

• – the conditions under which the failure occurred;
• – the operating mode of the equipment;
• – the age of the equipment;
• – the operating hours, the number of kilometers, etc.;
• – the serial number of the equipment;
• – the place and date of the incident;
• – the conditions of the return to the workshop (transportation mode, operations conducted, etc.);
• – the examination result;
• – the cause of failure (if confirmed);
• – the failed component(s), if applicable.

These elements can be used in the intervention on design, manufacturing operations or on the effectiveness of aggravated tests.

6.4.1. Estimation of robustness margins

This chapter explains how to determine the operational margins of a product by aggravated tests. In fact, many documents recommend the estimation of the robustness margins of a product but they do not explain how to check that they are sufficient. This chapter proposes a method to meet this objective.

Margins are determined once the operational limit is estimated through HALT tests. The margin is then defined as follows:

For example, at a high temperature, a product can be specified at 85°C and have an operational limit at 108°C. The resulting margin is therefore 108 – 85 = 23°C. This notion of the margin is obviously applicable to all the physical contributions implemented in the aggravated test.

6.4.2. Sufficient margins

The fact that a design should be reviewed when the margins are insufficient is often mentioned. This requires a definition of what a “sufficient margin” means. Moreover, this may depend on the manufacturing constraints inherent to the product concerned, and what is acceptable for one application may not necessarily be so for another.

Hence the use of the “resistance/constraint” method, well known in mechanics, is proposed here. Here, the constraint is the specification required by the system manufacturer, while the resistance is the operational limit estimated during the HALT test.

Resistance R and constraint C are considered as two random variables following a normal distribution, which leads to:

where

μ_c is the level of the physical contribution of interest given in the specification;

σ_c is obviously zero;

μ_r is the average of operational limits of products having undergone a HALT test;

σ_c is the standard deviation of the operational limits of products having undergone a HALT test.

It is possible to estimate failure probability, meaning the probability that the resistance R is weaker than the constraint C, leading to a product failure. The margin is then considered as sufficient when the probability P(R < C) < P_target, where P_target is the objective probability depending on the application. The difficulty is then in estimating the target probability.

Given the previous hypotheses, it can be shown that the probability of having a failure (in fact, that the resistance is weaker than the constraint) is given by:

[6.1]

where ϕ is the distribution function of the standard normal distribution. Since σc is zero, this finally leads to:

[6.2]

The target probability for the two types of manufacturing applications should therefore be estimated.

6.4.2.1. The reliability objective is the probability to fulfill the mission

Considering all the elements involved in the calculation of this probability as independent, if “n” is the number of elements, then:

[6.3]

Referring to Chapter 1, this probability can also be written in terms of the type of failures as follows:

[6.4]

where R_j, R_c and R_v are the proper operation probabilities of youth, catastrophic and aging failures, respectively. It is possible to formulate a hypothesis according to which, thanks to burn-in, there will be no youth failures or, more reasonably, their influence will be negligible or R_j(t)~ 1. Based on predictive reliability methods, an estimation of the proper operation probability of catastrophic failures can also be made.

Finally, the analysis of components with limited service life, which was developed in this chapter, can be used to give an estimation of the proper operation probability of the aging failures. Hence, based on all these estimations, it is possible to deduce a value of the probability P_target. To illustrate the proposed approach, consider the following example:

Assume that a product is specified at a maximal temperature, of +85°C. It can be readily deduced that:

• – μc = 85°C;
• – σc = 0.

Assume that the HALT test provided a temperature operational limit of +90°C, or μr = 90°C (the failure during the HALT test was observed at Ta = 95°C). If several products were subjected to the HALT test, the standard deviation of the temperature operational limit can be estimated. If a single product was subjected to the HALT test, then a standard deviation of 4°C can be considered, according to (McLean 2009). Let us formulate this last hypothesis so that σr = 4°C.

The failure probability given by equation [6.2] can then be calculated, Pdef ~ 8.8 10-5.

Let us assume that the objective of product reliability at the end of mission of duration Tm = 10 years is Rp(Tm) = 95% and that the catastrophic failure rate is λc = 2.21 10^-8.

The proper operation probability of catastrophic failures is therefore at the end of mission Rc(Tm) = 98.083%. Consequently, if there are no youth failures, no aging failures, the probability P_target must be:

[6.5]

Or numerically: R_target ≥ 96.857%. The failure probability was estimated at 8.8 10^-5 or a proper operation probability of 99.379%.

The reliability objective is therefore met in this purely virtual example.

6.4.2.2. The reliability objective is an MTBF

For this type of application, reasoning directly in terms of proper operation is no longer applicable, since the specified reliability objective is an MTBF. For this type of application, the maintenance is generally corrective (the failed component is instantaneously replaced by a new one). The notion of rate of occurrence of failure (Rocof) must be introduced, which is generally a time-dependent quantity. However, according to the hypothesis of catastrophic failures, it is known that Rocof is constant and equal to the failure rate λc of catastrophic failures.

For this type of application, repair times are generally negligible compared to the proper operation time so that MTBF ~ MTTF, and therefore .

Resuming the previous hypotheses, it can be reasonably expected that youth and aging failures are not observed, leaving only catastrophic failures so that:

Therefore, λ_HALT ≤ λ_target.