Journal: Shock and Vibration 16 (2009) 45–59
Authors: Robin Alastair Amy, Guglielmo S. Aglietti (E-mail: [email protected]), and Guy Richardson
Places of work of the authors: Astronautical Research Group, University of Southampton, School of Engineering Sciences, Southampton, UK
Surrey Satellite Technology Limited, Guildford, Surrey, UK
Copyright 2009 Hindawi Publishing Corporation. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Annotation. In the future, it is expected that all modern electronic equipment will have ever-increasing functionality while maintaining the ability to withstand shock and vibration loads. The reliability prediction process is hampered by the complex response and failure characteristics of electronic equipment, so current methods are a compromise between accuracy and cost.
Reliable and fast prediction of the reliability of electronic equipment during its operation with dynamic loads is very important for the industry. This article shows the problems in predicting the reliability of electronic equipment, slowing down the results. It should also be taken into account that the model for calculating reliability is usually built taking into account a wide range of equipment configurations for a number of the same type of components. Four classes of reliability prediction methods (reference methods, test data, experimental data, and modeling of the physical causes of failure - failure physics) are compared in this article to choose the possibility of applying one or another method. It is noted that most failures in electronic equipment are caused by thermal loads, but this review focuses on failures caused by shock and vibration during operation.
Translator's note. The article is a review of the literature on this topic. Despite its relatively old age, it serves as an excellent introduction to the problem of assessing the reliability of various methods.
1. Terminology
BGA Ball Grid Array.
DIP Dual In-line Processor, sometimes known as Dual In-line Package.
FE Finite Element.
PGA Pin Grid Array.
PCB Printed Circuit Board, sometimes known as a PWB (Printed Wiring Board).
PLCC Plastic Leaded Chip Carrier.
PTH Plated Through Hole, sometimes known as Pin Through Hole.
QFP Quad Flat Pack - also known as gull wing.
SMA Shape Memory Alloys.
SMT Surface Mount Technology.
Note from the original authors: In this article, the term "component" refers to a specific electronic device that can be soldered to a printed circuit board, the term "package" refers to any component of an integrated circuit (generally any SMT or DIP component). The term "attached component" refers to any composite printed circuit board or component system, emphasizing that the attached components have their own mass and rigidity. (Crystal packaging and its impact on reliability are not discussed in the article, so hereinafter the term "package" can be perceived as a "package" of one type or another - approx. transl.)
2. Statement of the problem
Shock and vibration loads applied to a printed circuit board cause stresses on the printed circuit board substrate, component packages, component traces, and solder joints. These stresses are due to a combination of bending moments in the PCB and the mass inertia of the component. In the worst case, these stresses can cause one of the following failure modes: PCB delamination, solder joint failure, lead failure, or component package failure. If any one of these failure modes has occurred, then a total failure of the device is likely to follow. The failure mode experienced during operation depends on the type of package, the properties of the printed circuit board, as well as the frequency and amplitude of bending moments and inertia forces. Slow progress in electronic equipment reliability analysis is due to the many combinations of input factors and failure modes that must be considered.
In the rest of this section, we will try to explain the complexity of considering different inputs at the same time.
The first complicating factor to consider is the wide range of package types available in today's electronics, as each package can fail for different reasons. Heavy components are more susceptible to inertial loads, while the response of SMT components is more dependent on PCB curvature. As a result, because of these basic differences, these types of components have widely differing failure criteria based on mass or dimensions. This problem is further exacerbated by the constant emergence of new components available on the market. Therefore, any proposed reliability prediction method must adapt to new components in order to have any practical application in the future. The PCB's response to vibration is determined by the stiffness and mass of the components, which affect the local response of the PCB. It is known that the heaviest or largest components significantly change the board's response to vibration in their installation locations. PCB mechanical properties (Young's modulus and thickness) can affect reliability in ways that are difficult to predict.
A stiffer PCB may reduce the PCB's overall response time under load, but at the same time, the bending moments applied to components may actually increase locally (Also, in terms of thermally induced failures, it is actually preferable to specify a more compatible PCB since this reduces the thermal stresses imposed on the packaging - ed.). The frequency and amplitude of local bending moments and inertial loads imposed on the package also affect the most likely failure mode. High-frequency, low-amplitude loads can lead to structural fatigue failure, which can be the main cause of failure (low/high cyclic fatigue, LCF refers to failures in which plastic deformation predominates (N_f < 10^6 ), while HCF stands for elastic deformation failures , usually (N_f > 10^6 ) to failure [56] - ed.) The final layout of the elements on the printed circuit board will determine the cause of failure, which can occur due to stress in a single component caused by inertial loading or local bending moments. Finally, it is necessary to take into account the influence of the human factor and production features, which increases the likelihood of equipment failure.
When considering a significant number of input factors and their complex interaction, it becomes clear why an effective method for predicting the reliability of electronic equipment has not yet been created. One of the literature reviews recommended by the authors on this issue is presented in IEEE [26]. However, this review focuses mainly on fairly broad classifications of reliability models, such as the method of predicting reliability from reference literature, experimental data, computer simulation of failure conditions (Physics-of-Failure Reliability (PoF)), and does not touch upon failures in sufficient detail. caused by shock and vibration. Fouchet et al. [17] follow a similar outline of the IEEE review as there is a strong focus on thermal failures. The previous brevity of the analysis of PoF methods, especially in relation to shock and vibration failures, deserves further consideration. An IEEE-like review is in the process of being compiled by the AIAA, but the scope of this review is not yet known.
3. Evolution of reliability prediction methods
The earliest reliability prediction method, developed in the 1960s, is currently described in MIL-HDBK-217F [44] this method, using a database of electronic equipment failures, obtain the average life of a printed circuit board, consisting of certain components. This method is known as a method of predicting reliability according to reference and normative literature. Although Mil-Hdbk-217F has become increasingly obsolete, the reference method is still in use today. The limitations and inaccuracies of this method have been well documented [1995], leading to the development of three classes of alternative methods: computer simulation of physical conditions of failure (PoF), experimental data, and field test data.
PoF methods predict reliability analytically, without resorting to the use of previously collected data. All PoF methods have two common characteristics of the classical method described in Steinberg [62]: first, the vibration response of the printed circuit board to a specific vibration impact is sought, then the failure criteria of individual components after vibration exposure are checked. An important advance in PoF methods was the use of distributed (averaged) board properties to quickly create a mathematical model of a printed circuit board [54], which significantly reduced the complexity and time spent on an accurate calculation of the vibration feedback of a printed circuit board (see Section 8.1.3). Recent developments in PoF techniques have improved failure prediction for surface mount technology (SMT) components; however, with the exception of the Barkers method [59], these new methods are only applicable to very specific combinations of components and printed circuit boards. There are very few methods available for large components such as transformers or large capacitors.
Experimental data methods improve the quality and capabilities of the model used in the reliability prediction methods according to reference and normative literature. The first method based on experimental data for predicting the reliability of electronic equipment was described in the 1999 work on the HIRAP method (Honeywell In-service Reliability Assessment Program), which was created by Honeywell, Inc. [20]. The method of experimental data has a number of advantages over the methods of predicting reliability according to reference and normative literature. Recently, many similar methods have appeared (REMM and TRACS [17], also FIDES [16]). The method of experimental data, as well as the method of predicting reliability according to reference and regulatory literature, does not allow us to satisfactorily take into account the layout of the board and the working environment of its operation in assessing the reliability. This shortcoming can be corrected by failure data from boards of similar design, or by boards that have been in similar operating conditions.
Experimental data methods depend on having an extensive database containing failure data over time. Each type of failure in this database must be correctly identified and its true cause determined. This method of reliability assessment is suitable for companies that produce the same type of equipment in large enough batches so that a significant number of failures can be processed for reliability assessment.
Reliability testing methods for electronic components have been in use since the mid-1970s and are generally classified into accelerated and non-accelerated tests. The main approach is to conduct hardware test runs that create the expected operating environment as realistically as possible. The tests are carried out until a failure occurs, which will allow the prediction of MTBF (mean time between failures - time between failures). If the MTBF is estimated to be very long, then the duration of the test can be reduced by accelerated testing, which is achieved by increasing environmental factors and using a well-known formula to relate the failure rate in the accelerated test to the failure rate expected in operation. Such testing is vital for high failure risk components as it provides the researcher with data that has the highest level of confidence, however, it would be impractical to use it for PCB design optimization due to the long time of one study iteration.
A cursory review of papers published in the 1990s suggests that this was a period when methods using experimental data, test data, and PoF methods competed with each other to replace the obsolete methods of predicting reliability from reference literature. However, each method has its own advantages and disadvantages, and when used correctly, gives valuable results. As a result, the IEEE recently released a standard [26] that lists all of the current reliability prediction methods in use today. The purpose of the IEEE was to provide a guide that would provide the engineer with information about all available methods and the advantages and disadvantages inherent in each method. While the IEEE approach is still at the beginning of a long evolutionary journey, it appears to have its own merits as the AIAA (American Institute of Aeronautics and Astronautics) is following it with a guideline called S-102 which is similar to the IEEE but also takes into account the relative quality of data from each method [27]. These manuals are intended only to bring together the methods that circulate in all the world's literature published on these subjects.
4. Failures caused by exposure to vibration
Most of the past research has mainly focused on random vibration as a PCB load, however the following research specifically looks at impact related failures. Such methods will not be discussed in full here, as they fall under the classification of PoF methods and are discussed in Sections 8.1 and 8.2 of this article. Hin et al. [24] created a test board to test the impact resistance of BGA solder joints. Lau et al. [36] described the reliability of PLCC, PQFP, and QFP components under in-plane and out-of-plane impact. Pitarresi et al. [53,55] reviewed computer motherboard shock failures and provided a good overview of the literature describing electronic equipment under shock loading. Steinberg [62] gives an entire chapter devoted to the design and analysis of electronic equipment subjected to an impact, considering both methods for predicting the impact environment and methods for ensuring the operability of electronic components. Suchir [64,65] described errors in linear calculations of PCB response to impact loading applied to board fixtures. Thus, in reference methods and experimental data methods, shock-related equipment failures can be considered, but in these methods, "shock" failures are described implicitly.
5. Reference methods
Of all the available methods described in the manuals, we will limit ourselves to only two that consider vibration failure: Mil-Hdbk-217 and CNET [9]. Mil-Hdbk-217 is accepted as the reference by most manufacturers. Like all methods from manuals and handbooks, they are based on empirical approaches that aim to predict the reliability of a component from experimental or laboratory data. The methods described in the reference literature are relatively easy to implement because they do not require complex mathematical modeling, using only part types, part counts, board operating conditions, and other readily available parameters. The input data is then entered into the model to calculate the time between failures - MTBF. Despite its advantages, Mil-Hdbk-217 is becoming less and less popular [12, 17,42,50,51]. Consider an incomplete list of limitations of its applicability.
- The data is becoming increasingly out of date since it was last updated in 1995 and is not relevant to new components, there is no chance of revisiting the model as the Defense Standards Improvement Board decided to let the method “die a natural death” [26].
- The method does not provide information about the failure mode, so the PCB layout cannot be improved or optimized.
- Models assume that failure is design independent, ignoring the layout of components on the PCB, however, component layout is known to have a large impact on failure probability. [50].
- The collected empirical data contains many inaccuracies, data from first generation components are used with an unnaturally high failure rate due to erroneous records of operation time, repair, etc., which reduces the reliability of reliability prediction results [51].
All these shortcomings indicate that the use of reference methods should be avoided, however, within the limits of the admissibility of these methods, a number of requirements of the specification should be implemented. Thus, reference methods should only be used when appropriate, i.e. in the early stages of design [46]. Unfortunately, even this use must be approached with some caution, as this kind of method has not been revised since 1995. Therefore, reference methods are inherently poor predictors of mechanical reliability and should be used with caution.
6. Test data methods
Test data methods are the simplest reliability prediction methods available. The prototype of the proposed PCB design is subjected to environmental vibrations reproduced on a laboratory bench. Next, the destruction parameters are analyzed (MTTF, shock spectrum), then this is used to calculate the reliability indicators [26]. The test data method should be used considering its advantages and disadvantages.
The main advantage of test data methods is the high accuracy and reliability of the results, therefore, for equipment with a high risk of failure, the final stage of the design process should always include vibration qualification testing. The disadvantage is the long time of manufacturing, installation and loading of the test sample, which makes the method unsuitable for design improvements in equipment with a high probability of failure. For an iterative product design method, a faster method should be considered. Loading times can be reduced by accelerated testing if valid models are available for subsequent calculation of actual service life [70,71]. However, accelerated test methods are more suitable for simulating thermal failures than vibrational failures. This is because it takes less time to test the effect of thermal loads on equipment than to test the effect of vibration loads. The action of vibration can manifest itself in the product only after a long time.
As a consequence, test methods are generally not applied to vibration failures unless there are extenuating circumstances such as low voltages leading to very long times to failure. For examples of data validation methods, see Hart [23], Hin et al. [24], Li [37], Lau et al. [36], Shetty et al. [57], Liguore and Followell [40], Estes et al. [15], Wang et al. [67], Jih and Jung [30]. A good general overview of the method is given in IEEE [26].
7. Methods of experimental data
The experimental data method is based on failure data from similar printed circuit boards that have been tested under specific operating conditions. The method is only correct for printed circuit boards that will experience similar loads. The experimental data method has two main aspects: building a database of failures of electronic components and implementing the method based on the proposed design. To build an appropriate database, there must be relevant failure data that has been collected from similar designs; this means that failure data for similar equipment must exist. Faulty equipment must also be analyzed and statistics collected appropriately, it is not enough to indicate that a given PCB design failed after a certain number of hours, the location, failure mode and cause of failure must be determined. If all previous failure data has not been carefully analyzed, then a long period of data collection will be required before the experimental data method can be used.
A possible workaround for this limitation is to implement a highly accelerated test life cycle (HALT) for the purpose of rapidly building a failure rate database, although accurately reproducing the environment parameters is difficult but vital [27]. A description of the second step in implementing the experimental data method can be found in [27], which shows how to predict the MTBF for a proposed design if the design under test is obtained by modifying an existing board for which detailed failure data already exists. Other reviews of experimental data methods are described by various authors in [11,17,20,26].
8. Computer Simulation of Failure Conditions (PoF)
Computer modeling of failure conditions, also referred to as stress and damage models or PoF models, is implemented in a two-stage reliability prediction process. The first stage includes the search for the response of the printed circuit board to the dynamic load imposed on it, at the second stage, the model response is calculated to ensure a given reliability indicator. Most of the literature is most often devoted to both the response prediction method and the process of finding failure criteria. These two methods are best understood when described independently; therefore, in this review, these two stages will be considered separately.
Between the stages of predicting the response and searching for failure criteria, the data set created in the first stage and used in the second is passed to the model. The response variable has evolved from using the input acceleration on the chassis [15,36,37,67], through the actual acceleration experienced by the component to account for the different vibrational responses of different PCB layouts [40], and finally to considering local deflection [62] or local bending moments [59] experienced by the PCB local to the component.
It has been noted that failure is a function of the location of components on the PCB [21,38], so models that take into account local vibrational response are likely to be accurate. The choice of which parameter (local acceleration, local deflection or bending moment) is decisive for failure depends on the particular case.
If SMT components are used then curvature or bending moments can be the most significant for failure, for heavy components local accelerations are usually used as failure criteria. Unfortunately, no research has been conducted to show which type of criteria is most appropriate in a particular set of input data.
It is important to consider the suitability of any PoF method used, as it is not practical to use any PoF method, either analytical or in the form of an FE (finite element method), that has not been validated by laboratory test data. In addition, it is important to use any model only within the limits of its applicability, which, unfortunately, limits the applicability of most current PoF models to use only in very specific and limited conditions. Good examples of discussion of PoF methods are described by various authors [17,19,26,49].
8.1. Response prediction
Response prediction involves using the geometry and material properties of the structure to calculate the desired response variable. It is expected that at this stage only the overall response of the underlying circuit board will be obtained, and not the response of individual components. There are three main types of response prediction method: analytical, detailed FE models and simplified FE models, described below. These methods focus on incorporating the effects of stiffness and the mass of the added components, however it is important not to lose sight of the importance of accurately modeling the rotational stiffness at the PCB edge as this is closely related to model accuracy (this is discussed in section 8.1.4). Fig. 1. An example of a detailed model of a printed circuit board [53].
8.1.1. Analytical Response Prediction
Steinberg [62] provides the only analytical method for calculating the vibration response of a printed circuit board. Steinberg states that the amplitude of oscillation at resonance of an electronic assembly is equal to two times the square root of the resonant frequency; this claim is based on unavailable data and is unverifiable. This allows the dynamic deflection at resonance to be analytically calculated, which can then be used to calculate either the dynamic load from a heavy component or the PCB curvature. This method does not directly give local PCB response and is only compatible with the deflection-based failure criteria described by Steinberg.
The validity of the transfer function distribution assumption based on amplitude measurements is questionable, since Pitarresi et al. [53] measured a critical attenuation of 2% for a computer motherboard, while using Steinberg's assumption would give 3,5% (based on natural frequency 54 Hz), which would greatly underestimate the board's response to vibration.
8.1.2. Detailed FE Models
Some authors demonstrate the use of detailed FE models to calculate the vibration response of a PCB [30,37,53] (Figures 57,58-1 show examples with a higher level of detail), however, the use of these methods is not recommended for a commercial product (unless only an accurate prediction of the local response is not absolutely necessary), since the time required to build and solve such a model is excessive. Simplified models produce data of adequate accuracy much faster and at a lower cost. The time required to build and solve a detailed FE model can be reduced using the JEDEC 3 spring constants published in [4-33], these spring constants can be used in place of the detailed FE model of each wire. In addition, the substructure method (sometimes known as the superelement method) can be implemented to reduce the computational time required to solve detailed models. It should be noted that detailed FE models often blur the lines between response prediction and failure criteria, so the work referenced here may also fall under the list of works containing failure criteria.
8.1.3. Distributed FE Models
Simplified FE models reduce model creation and solution time. The added component mass and its stiffness can be represented by a simple simulation of an empty printed circuit board with increased mass and stiffness, where the effects of mass and stiffness are turned on by a local increase in the Young's modulus of the printed circuit board.
Fig. 2. An example of a detailed model of a QFP component using symmetry to simplify the modeling process and reduce solution time [36]. Fig. 3. An example of a detailed FE model of J-lead [6].
The stiffness increase factor can be calculated by physically cutting out the attached member and applying bending test methods [52]. Pitarresi et al. [52,54] considered the effect of simplifying the added mass and stiffness provided by the components attached to the PCB.
The first paper considers a single case of a simplified FE-model of a printed circuit board, verified on the basis of experimental data. The main area of interest in this article is the definition of distributed properties, with the note that an accurate model requires high accuracy in torsional stiffness.
The second article looks at five different completed circuit boards, each modeled with several different levels of simplification of its composition. These models are compared with experimental data. This article concludes with some instructive observations of the correlation between mass-stiffness relationships and model accuracy. Both of these papers use only natural frequencies and MCE (Modal Assurance Criteria) to determine the correlation between the two models. Unfortunately, the error in natural frequency cannot give any information about the error in local accelerations or bending moments, also the MCE can only give an overall correlation between the two natural forms, but cannot be used to calculate acceleration or curvature error percentages. Using a combination of numerical analysis and computer simulation, Cifuentes [10] makes the following four observations.
- Simulated modes must contain at least 90% vibrating mass for accurate analysis.
- In the case where board deviations are comparable to board thickness, a non-linear analysis may be more appropriate than a linear one.
- Small errors in component placement can cause large errors in response measurements.
- Response measurement accuracy is more sensitive to mass errors than stiffness.
8.1.4. Border conditions
The rotational stiffness coefficient of the edge of the printed circuit board significantly affects the accuracy of the calculated response [59], and depending on the specific configuration, is much more important than the added mass of the component and the stiffness. Modeling the edge's rotational stiffness as zero (actually just a maintained condition) usually produces conservative results, while modeling as rigidly clamped usually underestimates the results because even the most rigid PCB clamping mechanisms cannot provide a fully clamped edge condition. Barker and Chen [5] validate the analytical theory with experimental results to show how the rotational stiffness of the edge affects the natural frequency of the PCB. The main conclusion of this work is a strong correlation between the rotational stiffness of the edge and the natural frequencies, consistent with the theory. This also means that large errors in edge rotation stiffness modeling will lead to large errors in response prediction. Although this work has been considered in a particular case, it is applicable to modeling all types of boundary condition mechanisms. Using experimental data from Lim et al. [41] gives an example of how the rotational stiffness of an edge can be calculated for using FE in a PCB model; this is achieved using a method adapted from Barker and Chen [5]. This work also shows how to determine the optimal location of any point in the structure in order to achieve the maximum increase in natural frequencies. Works that specifically consider the effect of modifying the boundary conditions to reduce the vibrational response also exist by Guo and Zhao [21]; Aglietti [2]; Aglietti and Schwingshackl [3], Lim et al. [41].
8.1.5. Shock and vibration impact predictions
Pitarresi et al. [53-55] use a detailed PCB FE model to predict the shock and vibration response of a board with components represented as 3D blocks. These models used experimentally determined constant damping factors to improve response prediction at resonance. Shock response spectrum (SRS) and time-sweeping methods have been compared to predict impact response, both methods being a trade-off between accuracy and solution time.
8.2. Rejection Criteria
The failure criteria take a measure of the board's response and use it to derive a failure metric, where the failure metric can be mean time between failures (MTBF), cycles to failure, probability of failure, or any other measure of reliability (see IEEE [26]; Jensen [ 28]; O'Connor [47] for a discussion of failure metrics). The many different approaches to generating this data can be conveniently divided into analytical and empirical methods. Empirical methods generate failure criteria data by loading component test samples to the required dynamic load. Unfortunately, due to the large range of inputs (component types, PCB thicknesses and loads) that are possible in practice, the published data is unlikely to be directly applicable, as the data is only valid in very special cases. Analytical methods do not suffer from such shortcomings and have a much wider applicability.
8.2.1. Empirical failure criteria
As stated earlier, the limitation of most empirical models is that they are only applicable to configurations involving the same PCB thickness, similar component types, and input loading, which is unlikely. However, the available literature is useful for the following reasons: it provides good examples of failure test execution, highlights various failure metrics, and provides valuable information regarding failure mechanics. Lee [37] created an empirical model to predict the reliability of 272-pin BGA and 160-pin QFP packages. Fatigue failures in the conductors and in the package body are investigated, and the experimental results are in good agreement with the stress-based damage analysis calculated using the detailed FE model (see also Li and Poglitsch [38,39]). The process yields cumulative damage for a given level of vibrational acceleration of the vibration input.
Lau et al. [36] evaluated the reliability of specific components under shock and vibration loading using Weibull statistics. Liguore and Followell [40] considered the failures of LLCC and J-lead components by changing the local acceleration in service cycles. Local acceleration is used as opposed to chassis input acceleration, and the effect of temperature on test results was also investigated. The article also makes reference to a study of the effect of PCB thickness on component reliability.
Guo and Zhao [21] compare component reliability when applying local torsional curvature as a load, in contrast to previous studies that used acceleration. Fatigue damage is simulated, then the FE model is compared with the experimental results. The article also discusses the optimization of the location of components to improve reliability.
Ham and Lee [22] present a test data method for the problem of determining lead solder stresses under cyclic torsional loading. Estes et al. [15] considered the gullwing component failure problem (GOST IEC 61188-5-5-2013) with applied input acceleration and thermal load. The components studied are the chip package types CQFP 352, 208, 196, 84 and 28, as well as FP 42 and 10. The article is devoted to the failure of electronic components due to oscillations in the orbit of a geostationary Earth satellite, the time between failures is given in terms of years of flight on geostationary or low earth orbits. It is noted that failure of the gullwing wires is more likely in places in contact with the package body than in a solder joint.
Jih and Jung [30] consider equipment failures caused by inherent manufacturing defects in a solder joint. This is done by creating a very detailed FE model of the PCB and finding the Power Spectral Density (PSD) for various manufacturing crack lengths. Ligyore, Followell [40] and Shetty, Reinikainen [58] suggest that empirical methods produce the most accurate and useful failure data for specific configurations of connected components. These kinds of methods are used if certain inputs (board thickness, component type, range of curvature) can be assumed to be constant throughout the design, or if the user can afford to perform actual tests of this kind.
8.2.2. Analytical failure criterion
SMT models of gussets
Various researchers looking at SMT corner lead failures suggest that this is the most common cause of failure. Articles by Sidharth, Barker [59] complete an earlier series of papers by presenting a model for determining the deformation of SMT corner leads and contour lead components. The proposed model has an error of less than 7% compared to the detailed FE model for the six worst case scenarios. The model is based on the formula previously published by Barker and Sidharth [4], which modeled the deflection of an attached part subjected to a bending moment. Suchir's article [63] deals analytically with the stresses expected in stack terminals due to locally applied bending moments. Barker and Sidharth [4] build on the work of Suheer [63], Barker et al. [4], who consider the effect of leading rotational stiffness. Finally, Barker et al. [7] used detailed FE models to study the effect of lead dimensional variations on lead fatigue life.
It is appropriate to mention here the work on lead spring constants by JEDEC, which greatly simplifies the creation of models of lead components [33-35]. Spring constants can be used instead of a detailed model of lead connections, the model will reduce the time to build and solve the FE model. The use of such constants in the component FE model will prevent direct calculation of lead local stresses. Instead, the total lead strain will be given, which then must be related to either local lead stresses or lead failure criteria based on the life cycle of the product.
Material fatigue data
Most of the failure data for materials that are used for solders and components are mainly related to thermal failure, and there is relatively little data related to fatigue failure. The main reference to this area is provided by Sandor [56], who provides data on the mechanics of fatigue and failure of solders. Steinberg [62] considers the failure of solder samples. Fatigue data for standard solders and wires are available from Yamada [69].
Fig. 4. The usual disclaimer position from the QFP component manual, close to the package body.
Modeling solder release failures is challenging due to the unusual properties of this material. The solution to this question depends on the component to be tested. It is known that for QFP packets this is usually not taken into account, and the reliability is evaluated according to the reference and normative literature. But if the soldering of BGA, PGA components of large size is calculated, then lead connections, due to their unusual properties, can affect the failure of the product. Thus, for QFP packages, lead fatigue properties are the most useful information. For BGA, information on the durability of solder joints subjected to instantaneous plastic deformation is more useful [14]. For larger components, Steinberg [62] provides solder joint pull-out stress data.
Failure Models for Heavy Components
The only failure models that exist for heavy components are presented in an article by Steinberg [62], which considers the tensile strength of components and gives an example of how to calculate the maximum allowable stress that can be applied to a lead joint.
8.3. Conclusions on the applicability of PoF models
The following conclusions have been made in the literature regarding PoF methods.
The local response is critical to predicting component failure. As noted in Li, Poglitsch [38], components at the edges of the PCB are less prone to failure than those located in the center of the PCB due to local differences in bending. Therefore, components at different locations on the PCB will have different failure probabilities.
Local board curvature is considered a more important failure criterion than acceleration for SMT components. Recent papers [38,57,62,67] indicate that board curvature is the main failure criterion.
Various types of packages, both in terms of the number of pins and the type used, are inherently more reliable than others, regardless of the specific local environment [15,36,38].
Temperature can affect component reliability. Liguore and Followell [40] state that fatigue life is highest in the temperature range from 0 ◦C to 65 ◦C, with a noticeable decrease at temperatures below -30 ◦C and above 95 ◦C. For QFP components, the location where the wire joins the package (see Fig. 4) is considered the main fault location, not the solder joint [15,22,38].
Board thickness has a certain effect on the fatigue life of SMT components, as BGA fatigue life has been shown to decrease by a factor of about 30-50 if the board thickness is increased from 0,85mm to 1,6mm (while keeping the overall curvature constant) [13]. Flexibility (compliance) of component terminals significantly affects the reliability of peripheral lead components [63], however, this is a non-linear dependence, and the terminals of the intermediate connection of elements are the least reliable.
8.4. Program methods
The Life Cycle Center of Excellence (CALCE) at the University of Maryland provides PCB vibration and shock response software. The software (named CALCE PWA) has a user interface that simplifies the process of running the FE model and automatically enters the response calculation into the vibration model. The assumptions used to create the FE response model are absent, and the failure criteria used are taken from Steinberg [61] (although the Barkers method [48] is also supposed to be implemented). In order to provide general recommendations for improving the reliability of equipment, the described software gives good results, especially since it simultaneously takes into account thermally induced stresses and requires minimal special knowledge, however, the accuracy of the failure criteria in the models has not been experimentally confirmed.
9. Methods for improving the reliability of equipment
This section will discuss post-project modifications that improve the reliability of electronic equipment. They fall into two categories: those that change the PCB's boundary conditions and those that increase the damping.
The main purpose of modifications to the boundary conditions is to reduce the dynamic deflection of the printed circuit board, this can be achieved by stiffening ribs, additional supports, or reducing the vibration of the input environment. Stiffeners can be useful as they increase the natural frequencies, thereby reducing the dynamic deflection [62], the same applies to the addition of additional supports [3], although the location of the supports can also be optimized, as shown by JH Ong and Lim [40]. Unfortunately, ribs and supports usually require layout redesign, so these techniques are best considered early in the design cycle. In addition, care should be taken to ensure that modifications do not change the natural frequencies to match those of the supporting structure, as this would be counterproductive.
The addition of isolation improves product reliability by reducing the impact of the dynamic environment transmitted to the equipment and can be achieved either passively or actively.
Passive methods are usually simple and less expensive to implement, such as the use of cable insulators [66] or the use of pseudoelastic properties of shape memory alloys (SMA) [32]. However, it is known that poorly designed isolators can actually increase the response.
Active methods provide better damping over a wider range of frequencies, typically at the expense of simplicity and mass, so they are usually intended to increase the accuracy of very sensitive precision instruments rather than to prevent damage. Active vibration isolation includes electromagnetic [60] and piezoelectric methods [18,43]. In contrast to the methods of modifying the boundary conditions, damping modification is aimed at reducing the peak resonant response of electronic equipment, while the actual natural frequencies should change insignificantly.
As with vibration isolation, damping can be done either passively or actively, with similar design simplification in the former case and higher complexity and damping in the latter.
Passive methods include, for example, very simple methods such as gluing the material, thereby increasing the damping of the printed circuit board [62]. More sophisticated methods include particle damping [68] and the use of broadband dynamic absorbers [25].
Active vibration control is usually achieved through the use of piezoceramic elements bonded to the PCB surface [1,45]. The use of hardening methods is case specific and should be carefully considered in relation to other methods. Applying these methods to equipment that is not known to have reliability issues will not necessarily increase the cost and weight of the design. However, if a product with an approved design fails during testing, it may be much faster and easier to apply a structural strengthening technique than to redesign the equipment.
10. Opportunities for the development of methods
This section details the possibilities for improving the reliability prediction of electronic equipment, although recent advances in optoelectronics, nanotechnology, and packaging technologies may soon limit the applicability of these proposals. The four main reliability prediction methods cannot be used at the time of device design. The only factor that could make such methods more attractive would be the development of fully automated, low-cost manufacturing and testing technologies, since this would allow the proposed design to be built and tested much faster than at present, with minimal human effort.
The PoF method has a lot of room for improvement. The main area where it can be improved is in integration with the overall design process. The design of electronic equipment is an iterative process that brings the developer closer to the finished result only in conjunction with engineers specializing in electronics, manufacturing and thermal engineering, structural design. A method that automatically addresses some of these issues at the same time will reduce the number of design iterations and save a significant amount of time, especially when considering the amount of interdepartmental interaction. Other areas for improving PoF methods will be divided into types of response prediction and failure criteria.
Response prediction has two possible development paths: either faster detailed models or improved simplified models. With the advent of more and more powerful computer processors, the time to solve detailed FE models can become quite short, at the same time, thanks to modern software, the assembly time of the product is reduced, this ultimately minimizes the cost of human resources. Simplified FE methods can also be enhanced with an automatic FE model generation process similar to those offered for detailed FE methods. Automated software (CALCE PWA) is currently available for this, but the technology is not well proven in practice and the simulation assumptions made are unknown.
Calculating the error inherent in various simplification methods would be very useful, which would allow useful fault tolerance criteria to be implemented.
Finally, a database or method for stiffening attached components would be useful, where these stiffenings could be used to improve the accuracy of the response models. The creation of component failure criteria depends on slight variation of similar components from different manufacturers, as well as on the possible development of new types of packaging, since any method or database for determining failure criteria must take into account such variability and changes.
One solution would be to create a method/software to automatically build detailed FE models based on input parameters such as lead and package dimensions. Such a method may be feasible for generally uniformly shaped components such as SMT or DIP components, but not for complex irregular components such as transformers, chokes, or custom components.
Subsequent FE models can be solved for stresses and combined with material failure data (S-N plasticity curve data, failure mechanics or similar) to calculate component life, although material failure data must be of high quality. The FE process should be related to real test data, preferably in as wide a range of configurations as possible.
The effort involved in such a process is relatively low compared to the alternative of direct lab testing, which must perform a statistically significant number of tests at different PCB thicknesses, different load intensities and directions, even hundreds of different types of components are available for even a few types of boards. In terms of simple laboratory testing, there may be a method to increase the value of each test.
If there were a method for calculating the relative increase in stresses due to changes in some variables, such as PCB thickness or lead dimensions, then the change in component life could subsequently be estimated. Such a method can be created using FE analysis or analytical methods, which will eventually lead to a simple formula for calculating failure criteria from existing failure data.
Ultimately, it is expected that a method will be created that combines all the different tools available: FE analysis, test data, analytical analysis, and statistical methods to create the most accurate failure data possible with the limited resources available. All individual elements of the PoF method can be improved by introducing stochastic methods into the process, which allow taking into account the influence of variability in the materials of electronic equipment and the stages of its production. This would make the results more realistic, perhaps leading to a process for building equipment that is more robust to variability while minimizing product performance degradation (including weight and cost).
Ultimately, these enhancements may even enable real-time evaluation of equipment reliability during the design process, instantly suggesting safer component, layout, or other reliability recommendations while including other issues such as electromagnetic interference (EMI), thermal and industrial.
11. Заключение
This review introduces the complexities of predicting the reliability of electronic equipment, traces the evolution of four types of analysis methods (by reference literature, experimental data, test data and PoF), leading to a generalization and comparison of these types of methods. Reference literature methods are noted to be useful only for preliminary studies, experimental data methods are only useful if extensive and accurate time data are available, and test data methods are vital for design qualification testing but not sufficient for optimization. designs.
PoF methods are discussed in more detail than in previous literature reviews, dividing the study into the categories of predictive criteria and failure probability. The Response Prediction section reviews the literature on distributed properties, boundary condition modeling, and levels of detail in FE models. It is shown that the choice of a response prediction method is a trade-off between accuracy and time for creating and solving an FE model, while again emphasizing the importance of the accuracy of the boundary conditions. In the "Fracture criteria" section, empirical and analytical failure criteria are considered; for SMT technology, reviews of models and heavy components are given.
Empirical methods are applicable only in very specific cases, although they provide good examples of reliability testing methods, while analytical methods have a much wider range of applicability, but are more difficult to implement. A brief discussion of the existing methods of failure analysis based on special software is given. Finally, conclusions are made about the future of reliability prediction, considering the directions in which reliability prediction methods can develop.
Literature[1] GS Aglietti, RS Langley, E. Rogers and SB Gabriel, An efficient model of an equipment loaded panel for active control design studies, The Journal of the Acoustical Society of America 108 (2000), 1663–1673.
[2] GS Aglietti, A lighter enclosure for electronics for space applications, Proceeding of Institute of Mechanical Engineers 216 (2002), 131–142.
[3] GS Aglietti and C. Schwingshackl, Analysis of enclosures and anti vibration devices for electronic equipment for space applications, Proceedings of the 6th International Conference on Dynamics and Control of Spacecraft Structures in Space, Riomaggiore, Italy, (2004).
[4] DB Barker and Y. Chen, Modeling the vibration restraints of wedge lock card guides, ASME Journal of Electronic Packaging 115(2) (1993), 189–194.
[5] DB Barker, Y. Chen and A. Dasgupta, Estimating the vibration fatigue life of quad leaded surface mount components, ASME Journal of Electronic Packaging 115(2) (1993), 195–200.
[6] DB Barker, A. Dasgupta and M. Pecht, PWB solder joint life calculations under thermal and vibrational loading, Annual Reliability and Maintainability Symposium, 1991 Proceedings (Cat. No.91CH2966-0), 451–459.
[7] DB Barker, I. Sharif, A. Dasgupta and M. Pecht, Effect of SMC lead dimensional variabilities on lead compliance and solder joint fatigue life, ASME Journal of Electronic Packaging 114(2) (1992), 177–184.
[8] DB Barker and K. Sidharth, Local PWB and component bowing of an assembly subjected to a bending moment, American Society of Mechanical Engineers (Paper) (1993), 1–7.
[9] J. Bowles, A survey of reliability-prediction procedures for microelectronic devices, IEEE Transactions on Reliability 41(1) (1992), 2–12.
[10] AO Cifuentes, Estimating the dynamic behavior of printed circuit boards, IEEE Transactions on Components, Packaging, and Manufacturing Technology Part B: Advanced Packaging 17(1) (1994), 69–75.
[11] L. Condra, C. Bosco, R. Deppe, L. Gullo, J. Treacy and C. Wilkinson, Reliability assessment of aerospace electronic equipment, Quality and Reliability Engineering International 15(4) (1999), 253–260 .
[12] MJ Cushing, DE Mortin, TJ Stadterman and A. Malhotra, Comparison of electronics-reliability assessment approaches, IEEE Transactions on Reliability 42(4) (1993), 542–546.
[13] R. Darveaux and A. Syed, Reliability of area array solder joints in bending, SMTA International Proceedings of the Technical Program (2000), 313–324.
[14] NF Enke, TJ Kilinski, SA Schroeder and JR Lesniak, Mechanical behaviors of 60/40 tin-lead solder lap joints, Proceedings – Electronic Components Conference 12 (1989), 264–272.
[15] T. Estes, W. Wong, W. McMullen, T. Berger and Y. Saito, Reliability of class 2 heel fillets on gull wing leaded components. Aerospace Conference, Proceedings 6 (2003), 6-2517–6 C2525
[16] FIDES, FIDES Guide 2004 issue A Reliability Methodology for Electronic Systems. FIDES Group, 2004.
[17] B. Foucher, D. Das, J. Boullie and B. Meslet, A review of reliability prediction methods for electronic devices, Microelectronics Reliability 42(8) (2002), 1155–1162.
[18] J. Garcia-Bonito, M. Brennan, S. Elliott, A. David and R. Pinnington, A novel high-displacement piezoelectric actuator for active vibration control, Smart Materials and Structures 7(1) (1998), 31 -42.
[19] W. Gericke, G. Gregoris, I. Jenkins, J. Jones, D. Lavielle, P. Lecuyer, J. Lenic, C. Neugnot, M. Sarno, E. Torres and E. Vergnault, A methodology to assess and select a suitable reliability prediction method for eee components in space applications, European Space Agency, (Special Publication) ESA SP (507) (2002), 73–80.
[20] L. Gullo, In-service reliability assessment and top-down approach provides an alternative reliability prediction method. Annual Reliability and Maintainability, Symposium Proceedings (Cat. No. 99CH36283), 1999, 365–377.
[21] Q. Guo and M. Zhao, Fatigue of SMT solder joint including torsional curvature and chip location optimization, International Journal of Advanced Manufacturing Technology 26(7–8) (2005), 887–895.
[22] S.-J. Ham and S.-B. Lee, Experimental study for reliability of electronic packaging under vibration, Experimental Mechanics 36(4) (1996), 339–344.
[23] D. Hart, Fatigue testing of a component lead in a plated through hole, IEEE Proceedings of the National Aerospace and Electronics Conference (1988), 1154–1158.
[24] TY Hin, KS Beh and K. Seetharamu, Development of a dynamic test board for FCBGA solder joint reliability assessment in shock & vibration. Proceedings of the 5th Electronics Packaging Technology Conference (EPTC 2003), 2003, 256–262.58
[25] V. Ho, A. Veprik and V. Babitsky, Ruggedizing printed circuit boards using a wideband dynamic absorber, Shock and Vibration 10(3) (2003), 195–210.
[26] IEEE, IEEE guide for selecting and using reliability predictions based on ieee 1413, 2003, v+90 C.
[27] T. Jackson, S. Harbater, J. Sketoe and T. Kinney, Development of standard formats for space systems reliability models, Annual Reliability and Maintainability Symposium, 2003 Proceedings (Cat. No. 03CH37415), 269–276.
[28] F. Jensen, Electronic Component Reliability, Wiley, 1995.
[29] JH Ong and G. Lim, A simple technique for maximizing the fundamental frequency of structures, ASME Journal of Electronic Packaging 122 (2000), 341–349.
[30] E. Jih and W. Jung, Vibrational fatigue of surface mount solder joints. iThermfl98. Sixth Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (Cat. No.98CH36208), 1998, 246–250.
[31] B. Johnson and L. Gullo, Improvements in reliability assessment and prediction methodology. Annual Reliability and Maintainability Symposium. 2000 Proceedings. International Symposium on Product Quality and Integrity (Cat. No. 00CH37055), 2000, -:181–187.
[32] M. Khan, D. Lagoudas, J. Mayes and B. Henderson, Pseudoelastic SMA spring elements for passive vibration isolation: part i modeling, Journal of Intelligent Material Systems and Structures 15(6) (2004), 415–441 .
[33] R. Kotlowitz, Comparative compliance of representative lead designs for surface-mounted components, IEEE Transactions on Components, Hybrids, and Manufacturing Technology 12(4) (1989), 431–448.
[34] R. Kotlowitz, Compliance metrics for surface mount component lead design. 1990 Proceedings. 40th Electronic Components and Technology Conference (Cat. No. 90CH2893-6), 1990, 1054–1063.
[35] R. Kotlowitz and L. Taylor, Compliance metrics for the inclined gull-wing, spider j-bend, and spider gull-wing lead designs for surface mount components. 1991 Proceedings. 41st Electronic Components and Technology Conference (Cat. No. 91CH2989-2), 1991, 299–312.
[36] J. Lau, L. Powers-Maloney, J. Baker, D. Rice and B. Shaw, Solder joint reliability of fine pitch surface mount technology assemblies, IEEE Transactions on Components, Hybrids, and Manufacturing Technology 13(3) (1990), 534–544.
[37] R. Li, A methodology for fatigue prediction of electronic components under random vibration load, ASME Journal of Electronic Packaging 123(4) (2001), 394–400.
[38] R. Li and L. Poglitsch, Fatigue of plastic ball grid array and plastic quad flat packages under automotive vibration. SMTA International, Proceedings of the Technical Program (2001), 324–329.
[39] R. Li and L. Poglitsch, Vibration fatigue, failure mechanism and reliability of plastic ball grid array and plastic quad flat packages.
[40] Proceedings 2001 HD International Conference on High-Density Interconnect and Systems Packaging (SPIE Vol. 4428), 2001, 223–228.
[41] S. Liguore and D. Followell, Vibration fatigue of surface mount technology (smt) solder joints. Annual Reliability and Maintainability Symposium 1995 Proceedings (Cat. No. 95CH35743), 1995, -:18–26.
[42] G. Lim, J. Ong and J. Penny, Effect of edge and internal point support of a printed circuit board under vibration, ASME Journal of Electronic Packaging 121(2) (1999), 122–126.
[43] P. Luthra, Mil-hdbk-217: What is wrong with it? IEEE Transactions on Reliability 39(5) (1990), 518.
[44] J. Marouze and L. Cheng, A feasibility study of active vibration isolation using thunder actuators, Smart Materials and Structures 11(6) (2002), 854–862.
[45] MIL-HDBK-217F. Reliability Prediction of Electronic Equipment. US Department of Defense, F edition, 1995.
[46] SR Moheimani, A survey of recent innovations in vibration damping and control using shunted piezoelectric transducers, IEEE Transactions on Control Systems Technology 11(4) (2003), 482–494.
[47] S. Morris and J. Reilly, Mil-hdbk-217-a favorite target. Annual Reliability and Maintainability Symposium. 1993 Proceedings (Cat. No. 93CH3257-3), (1993), 503–509.
P. O'Connor, Practical reliability engineering. Wiley, 1997.
[48] M. Osterman and T. Stadterman, Failure assessment software for circuit card assemblies. Annual Reliability and Maintainability. Symposium. 1999 Proceedings (Cat. No. 99CH36283), 1999, 269–276.
[49] M. Pecht and A. Dasgupta, Physics-of-failure: an approach to reliable product development, IEEE 1995 International Integrated Reliability Workshop Final Report (Cat. No. 95TH8086), (1999), 1–4.
[50] M. Pecht and W.-C. Kang, A critique of mil-hdbk-217e reliability prediction methods, IEEE Transactions on Reliability 37(5) (1988), 453–457.
[51] MG Pecht and FR Nash, Predicting the reliability of electronic equipment, Proceedings of the IEEE 82(7) (1994), 992–1004.
[52] J. Pitarresi, D. Caletka, R. Caldwell and D. Smith, The smeared property technique for the FE vibration analysis of printed circuit cards, ASME Journal of Electronic Packaging 113 (1991), 250–257.
[53] J. Pitarresi, P. Geng, W. Beltman and Y. Ling, Dynamic modeling and measurement of personal computer motherboards. 52nd Electronic Components and Technology Conference 2002., (Cat. No. 02CH37345)(-), 2002, 597–603.
[54] J. Pitarresi and A. Primavera, Comparison of vibration modeling techniques for printed circuit cards, ASME Journal of Electronic Packaging 114 (1991), 378–383.
[55] J. Pitarresi, B. Roggeman, S. Chaparala and P. Geng, Mechanical shock testing and modeling of PC motherboards. 2004 Proceedings, 54th Electronic Components and Technology Conference (IEEE Cat. No. 04CH37546) 1 (2004), 1047–1054.
[56] BI Sandor, Solder Mechanics - A State of the Art Asssessment. The Minerals, Metals and Materials Society, 1991.
[57] S. Shetty, V. Lehtinen, A. Dasgupta, V., Halkola and T. Reinikainen, Fatigue of chip scale package interconnects due to cyclic bending, ASME Journal of Electronic Packaging 123(3) (2001), 302– 308.
[58] S. Shetty and T. Reinikainen, Three- and four-point bend testing for electronic packages, ASME Journal of Electronic Packaging 125(4) (2003), 556–561.
[59] K. Sidharth and D. B. Barker, Vibration induced fatigue life estimation of corner leads of peripheral leaded components, ASME Journal of Electronic Packaging 118(4) (1996), 244–249.
[60] J. Spanos, Z. Rahman and G. Blackwood, Soft 6-axis active vibration isolator, Proceedings of the American Control Conference 1 (1995), 412–416.
[61] D. Steinberg, Vibration Analysis for Electronic Equipment, John Wiley & Sons, 1991.
[62] D. Steinberg, Vibration Analysis for Electronic Equipment, John Wiley & Sons, 2000.
[63] E. Suhir, Could compliant external leads reduce the strength of a surface-mounted device? 1988 Proceedings of the 38th Electronics Components Conference (88CH2600-5), 1988, 1–6.
[64] E. Suhir, Nonlinear dynamic response of a printed circuit board to shock loads applied to its support contour, ASME Journal of Electronic Packaging 114(4) (1992), 368–377.
[65] E. Suhir, Response of a flexible circuit printed board to periodic shock loads applied to its support contour, American Society of Mechanical Engineers (Paper) 59(2) (1992), 1–7.
[66] A. Veprik, Vibration protection of critical components of electronic equipment in harsh environmental conditions, Journal of Sound and Vibration 259(1) (2003), 161–175.
[67] H. Wang, M. Zhao and Q. Guo, Vibration fatigue experiments of SMT solder joint, Microelectronics Reliability 44(7) (2004), 1143–1156.
[68] ZW Xu, K. Chan and W. Liao, An empirical method for particle damping design, Shock and Vibration 11(5–6) (2004), 647–664.
[69] S. Yamada, A fracture mechanics approach to soldered joint cracking, IEEE Transactions on Components, Hybrids, and Manufacturing Technology 12(1) (1989), 99–104.
[70] W. Zhao and E. Elsayed, Modeling accelerated life testing based on mean residual life, International Journal of Systems Science 36(11) (1995), 689–696.
[71] W. Zhao, A. Mettas, X. Zhao, P. Vassiliou and E. E. Elsayed, Generalized step stress accelerated life model. Proceedings of 2004 International Conference on the Business of Electronic Product Reliability and Liability, 2004, 19–25.
Source: habr.com