Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Copper in heat exchangers wikipedia , lookup
Radiator (engine cooling) wikipedia , lookup
Passive solar building design wikipedia , lookup
Underfloor heating wikipedia , lookup
Thermoregulation wikipedia , lookup
Hyperthermia wikipedia , lookup
R-value (insulation) wikipedia , lookup
Thermal conductivity wikipedia , lookup
Solar air conditioning wikipedia , lookup
Thermal Fatigue Life Prediction of Solder Joints in Avionics by Surrogate Modeling – a Contribution to Physics of Failure in Reliability Prediction Jonas Arwidson Norrköping 2013 Linköping Studies in Science and Technology. Dissertations. No 1521 Thermal Fatigue Life Prediction of Solder Joints in Avionics by Surrogate Modeling – a Contribution to Physics of Failure in Reliability Prediction Jonas Arwidson Copyright © 2013 Jonas Arwidson, unless otherwise noted. Department of Science and Technology Linköping University Campus Norrköping SE-601 74 Norrköping, Sweden Saab AB Business Area Electronic Defence Systems Avionics Division Box 1017 SE-551 11 Jönköping, Sweden ISBN 978-91-7519-618-3 ISSN 0345-7524 Printed in Sweden by LiU-Tryck, 2013. Abstract Manufacturers of aerospace, defense, and high performance (ADHP) equipment are currently facing multiple challenges related to the reliability of electronic systems. The continuing reduction in size of electronic components combined with increasing clock frequencies and greater functionality, results in increased power density. As an effect, controlling the temperature of electronic components is central in electronic product development in order to maintain and potentially improve the reliability of the equipment. Simultaneously, the transition to lead-free electronic equipment will most probably propagate also to the ADHP industry. Compared to well-proven tin-lead solder, the knowledge about field operation reliability of lead-free solders is still limited, as well as the availability of damage evaluation models validated for field temperature conditions. Hence, the need to fill in several knowledge gaps related to reliability and reliability prediction of lead-free solder alloys is emphasized. Having perceived increasing problems experienced in the reliability of fielded equipment, the ADHP industry has suggested inclusion of physics-of-failure (PoF) in reliability prediction of electronics as one potential measure to improve the reliability of the electronic systems. This thesis aims to contribute to the development of reliable ADHP systems, with the main focus on electronic equipment for the aerospace industry. In order to accomplish this, the thesis provides design guidelines for power distribution on a double-sided printed circuit board assembly (PBA) as a measure to improve the thermal performance without increasing the weight of the system, and a novel, computationally efficient method for PoF-based evaluation of damage accumulation in solder joints in harsh, non-cyclic field operation temperature environments. Thermal fatigue failure mechanisms and state-of-the-art thermal design and design tools are presented, with focus on the requirements that may arise from avionic use, such as low weight, high reliability, and ability to sustain functional during high vibration levels and high g-forces. Paper I, II, and III describes an in-depth investigation that has been performed utilizing advanced thermal modeling of power distribution on a double-sided PBA as a measure to improve the thermal performance of electronic modules. Paper IV contributes to increasing the accuracy of thermal fatigue life prediction in solder joints, by employing existing analytical models for predicting thermal fatigue life, but enhancing the prediction result by incorporating advanced thermal analysis in the procedure. Papers V and VI suggest and elaborate on a computational method that utilizes surrogate stress and strain modeling of a solder joint, to quickly evaluate the damage accumulated in a critical solder joint from non-cyclic, non-simplified field operation iii temperature profiles, with accuracy comparable to finite element modeling. The method has been tested on a ball grid array package with SnAgCu solder joints. This package is included in an extensive set of accelerated tests that helps to qualify certain packages and solder alloys for avionic use. The tests include -20°C to +80°C and 55°C to +125°C thermal cycling of a statistically sound population of a number of selected packages, assembled with SnAgCu, Sn100C, and SnPbAg solder alloys. Statistical analysis of the results confirms that the SnAgCu-alloy may outperform SnPbAg solder at moderate thermal loads on the solder joints. In Papers VII and VIII, the timeframe is extended to a future, in which validated life prediction models will be available, and the suggested method is expected to increase the accuracy of embedded prognostics of remaining useful thermal fatigue life of a critical solder joint. The key contribution of the thesis is the added value of the proposed computational method utilized in the design phase for electronic equipment. Due to its ability for time-efficient operation on uncompressed temperature data, the method gives contribution to the accuracy, and thereby also to the credibility, of reliability prediction of electronic packages in the design phase. This especially relates to applications where thermal fatigue is a dominant contributor to the damage of solder joints. iv Populärvetenskaplig sammanfattning Tillverkare av elektronik för försvar, flyg, rymd och andra tillämpningar med krav på hög prestanda och tillförlitlighet står för närvarande inför flera utmaningar. Över tid så har storleken på elektronik kontinuerligt minskat medan effektförbrukningen har bibehållits eller ökat. Detta medför krav på mer avancerad teknologi för att föra bort värmen från elektroniken för att den inte skall överhettas och/eller tillförlitligheten påverkas negativt. Vidare kommer samma nisch av elektronikutvecklande industri sannolikt ta steget över till blyfri elektronik som en följd av ett EU-direktiv om minskning av skadliga ämnen, RoHS-direktivet. Sedan lång tid har en legering av tenn och bly använts som lod för mekanisk och elektrisk förbindning av elektronikkomponenter mot deras omgivning. Denna legering är därför väl beprövad vad gäller tillförlitlighet, och stor erfarenhet finns tillgänglig. För blyfria lod råder dock för närvarande starkt begränsad tillgänglighet på motsvarande data. Ett sätt att angripa bristen på tillförlitlighetsdata för blyfria lod är att beräkna tillförlitlighet och livslängd genom fysikaliska modeller av utmattning och andra sorters fel: den så kallade Physics-of-Failure (PoF)-metodiken. Målsättningen med avhandlingen är att bidra till utvecklingen av tillförlitliga elektronikprodukter framför allt avsedda för flygtillämpningar. Detta görs genom att närmare studera möjligheterna att hålla kontroll över temperaturen på elektroniken, samt genom att presentera en ny, beräkningseffektiv metod för att uppskatta förbrukad livslängd i en lödfog som utsatts för temperaturvariationer. Teori kring fel som kan uppstå på grund av olika temperaturbelastningar följs av en genomgång av tillgängliga verktyg för termisk design. Exempel på genomtänkt, avancerad termisk design av elektronikapparater redovisas med fokus på kravbilden som finns i flygande tillämpningar, såsom låg vikt, hög tillförlitlighet och höga vibrationsnivåer. De tre första artiklarna i avhandlingen utgör en detaljerad undersökning av effekten av att styra placeringen av effektutvecklande elektronikkomponenter på dubbelsidiga mönsterkort; ett sätt att hålla kontroll över temperaturen utan att tillföra varken extra vikt eller kostnad. Övriga artiklar riktar sig mot möjligheten att uppnå ökad precision i skattningen av förbrukad livslängd i en lödfog som funktion av temperaturrelaterade belastningar. Första steget mot ökad noggrannhet kan tas genom att integrera avancerad termisk analys i metoder som tillämpar enkla modeller för utmattning av lödfogar; ytterligare höjd precision kan åstadkommas genom numeriska beräkningar som dock kräver omfattande datorkraft och/eller lång beräkningstid. Därför presenteras en beräkningsmetod som är mycket resurseffektiv jämfört med numeriska lösningar av v finita element (FE)-modeller, och samtidigt levererar resultat helt jämförbara med FEberäkningar. Dessutom redovisas en omfattande serie experiment, vars syfte är att utvärdera två varianter av blyfria lod i kombination med olika komponenttyper, avseende användbarhet i flygtillämpningar. I experimenten har ett antal olika typer av elektronikkomponenter monterats på mönsterkort med dels de två blyfria loden och även ett ”vanligt” blyat lod för referens, varefter de utsatts för två olika nivåer av temperaturcykling. Antalet cykler till felutfall har jämförts mellan de olika loden och komponenttyperna. Resultaten bekräftar andra undersökningar i det att de blyfria loden uppvisar bättre tillförlitlighet vid måttliga påfrestningar på lödfogarna, medan det blyade lodet är tåligare vid mer extrema belastningar. Avslutningsvis blickas framåt i tiden, då möjligen den nya beräkningsmetoden skulle kunna tillämpas i sammanhanget kontinuerlig bedömning, prognostisering, av kvarvarande liv i elektronik som används. Det viktigaste bidraget från denna avhandling är värdet av beräkningsmetoden tillämpad i utvecklingsfasen av en produkt, där den kan bidra till ökad noggrannhet och därmed vunnet förtroende för predikteringen av livslängd för elektronik. Som en följd därav kan man närma sig att kunna designa ”rätt från början” och minska antalet dyrbara omtag i produktutvecklingen. vi Acronyms and Abbreviations ADHP BGA CDI CFD COTS CTE DSB FCF FE FEM FNM IMC LCM MEA PBGA PoF PBA PCB RoHS RUL SAC305 SEDcr TIM TRN Aerospace, defense, and high performance Ball grid array Cumulative damage index Computational fluid dynamics Commercial off the shelf Coefficient of thermal expansion Double-sided PCB First component to fail Finite element Finite element method Flow network modeling Intermetallic compound Life consumption monitoring More electric aircraft Plastic ball grid array Physics of failure Printed circuit board assembly Printed circuit board Restriction of hazardous substances Remaining useful life SnAg3.0Cu0.5 solder alloy Creep strain energy density Thermal interface material Thermal resistance network vii viii List of Papers The following publications are included in this thesis: Paper I: J. Johansson, I. Belov, and P. Leisner, “CFD Analysis of an Avionic Module for Evaluating Power Distribution as a Thermal Management Measure for a Double-sided PCB,” in Proc. Semi-Therm 2007, San Jose, CA, 2007. Paper II: J. Johansson, I. Belov, and P. Leisner, “An Experimental Setup for Validating a CFD Model of a Double-sided PCB in a Sealed Enclosure at Various Power Configurations,” in Proc. EuroSime 2005, Berlin, Germany, 2005. Paper III: J. Johansson, I. Belov, and P. Leisner, “Investigating the Effect of Power Distribution on Cooling a Double-sided PCB: Numerical Simulation and Experiment,” in Proc. 2005 ASME Summer Heat Transfer Conference, San Francisco, CA, 2005. Paper IV: J. Johansson, P. Leisner, J. Lee, D.W. Twigg, and M. Rassaian, “on Thermomechanical Durability Analysis combined with Computational Fluid Dynamics Thermal Analysis,” in Proc. ASME International Mechanical Engineering Congress and Exposition (IMECE 2007), Seattle, WA, 2007. Paper V: J. Johansson, I. Belov, E. Johnson, and P. Leisner, “A Computational Method for Evaluating the Damage in a Solder Joint of an Electronic Package Subjected to Thermal Loads,” Engineering Computations, accepted for publication. Paper VI: J. Johansson, I. Belov, R. Dudek, E. Johnson, and P. Leisner, “Investigation on Thermal Fatigue of SnAgCu, Sn100C, and SnPbAg Solder Joints in Varying Temperature Environments,” Microelectronics Reliability, submitted. Paper VII: J. Johansson and P. Leisner, “Prognostics of Thermal Fatigue Failure of Solder Joints in Avionic Equipment,” IEEE Aerospace and Electronic Systems Magazine, vol. 27, no. 4, Apr. 2012. Paper VIII: J. Johansson, I. Belov, E. Johnson, and P. Leisner, “An Approach to Life Consumption Monitoring of Solder Joints in Operating Temperature Environment,” Proc. EuroSime 2012, Lisbon, Portugal, 2012. ix Author’s contribution to the papers: Paper I: 70% of the analytic work, and 80% of the writing. Paper II: All experimental work, and 90% of the writing. Paper III: All experimental work, and 90% of the writing. Paper IV: All planning of the collaborative efforts, 50% of computational experiments, and all of the writing. Paper V: All computational experiments, and 90% of the writing. Paper VI: 30% of the planning of experimental work, 80% of computational experiments, and all of the writing. Paper VII: The complete paper. Paper VIII: All computational experiments, and 90% of the writing. Publications not discussed in this thesis: J. Johansson, I. Belov, K. Säfsten, and P. Leisner, “Thermal Analysis of an Electronic Module with a Double-sided PCB Housed in a 2-MCU Enclosure for Avionic Applications,” in Proc. IMAPS 2004, Long Beach, CA, 2004. J. Johansson, I. Belov, K. Säfsten, and P. Leisner, “Thermal Design Evaluation of an Electronic Module for Helicopter Applications,” CEPA2 Workshop, Paris, France, 2004. J. Johansson, “Tools and Methods for Simulation in Thermal Design of Electronics for use in Harsh Environments,” in Proc. SsoCC ´04, Båstad, Sweden, 2004. x Acknowledgements First of all I would like to thank my supervisor Professor Peter Leisner at the School of Engineering, Jönköping University, for always being positive and optimistic, able to supply a boost of energy when most needed. Also, many thanks must be directed to his colleagues, my co-supervisors Dr. Ilja Belov, for continuously reminding me about the importance of being thorough in every action taken in research, and Dr. Mats Robertsson, for pursuing this work to the finalized stage, although long time has passed and his current work situation would in practice not admit this kind of assignment. I would like to thank many people at Saab AB, though especially M. Sc. Bengt Rogvall and Dr. Ingemar Söderquist for initiating this project and believing in my capability of performing this kind of work, and M. Sc. Mats Johansson for always listening and being supportive, and for providing valuable feedback during the writing of the thesis. Doctors Håkan and Kristina Forsberg have been perfect role models and provided inspiration when mostly needed. The financial support from the KK foundation and Saabs Teknikråd is gratefully acknowledged, as this enabled the studies. My brother, Docent Jonny Johansson, has been more important to me than he could imagine, all the way through. I would also like to thank my parents for reminding me in times of trouble, that there might be other things in life that are more important than completion of this work. Ten years have passed since I initiated the work that is summarized in this thesis. It might be that a sense of emptiness will emerge after concluding the work, which has from time to time, and especially in the final stage, been quite intense. However, during these years a wonderful woman came into my life, and subsequently two amazing little copies of both of us. I believe these three incredibly important people could earn increased attention from now on, and replace the potential sense of emptiness with meaning and delight. Shortly before defending this thesis, we decided on using Marias family name for all of us, which should explain the inconsistency with “Johansson” and “Arwidson”. xi xii Contents 1 INTRODUCTION ............................................................................................. 1 1.1 1.2 1.3 2 THERMAL CHALLENGES FOR AVIONIC EQUIPMENT ......................................... 2 RELIABILITY OF LEAD-FREE ELECTRONICS ..................................................... 3 OBJECTIVE, SCOPE, AND DELIMITATIONS ........................................................ 4 THERMAL MANAGEMENT OF AVIONIC EQUIPMENT ...................... 7 2.1 2.2 2.3 2.3.1 2.4 2.5 3 HEAT TRANSFER FUNDAMENTALS ................................................................... 7 THERMALLY INDUCED FAILURE MECHANISMS.............................................. 10 THERMAL DESIGN TOOLS .............................................................................. 12 Uncertainty in Simulations ................................................................... 15 STATE-OF-THE-ART THERMAL MANAGEMENT IN AVIONICS ......................... 16 MANAGING POWER DISTRIBUTION ................................................................ 21 RELIABILITY PREDICTION ...................................................................... 27 3.1 3.2 3.3 3.4 3.4.1 3.4.2 ADVANCED THERMAL ANALYSIS INCORPORATED IN THERMAL FATIGUE LIFE PREDICTION ................................................................................................... 29 PHYSICS OF FAILURE IN RELIABILITY PREDICTION OF SOLDER JOINTS ......... 33 ACCELERATED THERMAL CYCLING TESTS OF SURFACE MOUNT PACKAGES WITH SAC, SN100C, AND SNPB SOLDER PASTE............................................ 36 SURROGATE MODELING OF DAMAGE ACCUMULATION IN SOLDER JOINTS ... 37 Data Preparation .................................................................................. 40 Algorithm for Estimation of Accumulated Damage .............................. 42 4 PROGNOSTICS OF REMAINING USEFUL LIFE OF ELECTRONIC EQUIPMENT................................................................................................... 49 5 CONCLUDING SUMMARY AND CONTRIBUTION TO THE FIELD . 53 5.1 5.2 5.3 THERMAL MANAGEMENT .............................................................................. 53 PHYSICS OF FAILURE IN RELIABILITY PREDICTION OF SOLDER JOINTS ......... 54 FUTURE WORK ............................................................................................... 55 REFERENCES ................................................................................................................. 57 THE PAPERS ................................................................................................................... 65 xiii 1 Introduction The history of aviation had a remarkable beginning. Only six years after the first successful flights of the Wright brothers in December 1903, the Frenchman Henri Blériot flew across the English Channel. Another five years from that, scout planes performed reconnaissance in World War I, shortly followed by the creation of small forces of fighter planes that engaged in aerial combat and bombing raids [1]. Simultaneous to the first long jumps of aviation, the evolution of electronic equipment was initialized when the Englishman John A. Fleming invented the Fleming valve, which today is called a diode. Three years later, in 1906, the American Lee D. Forest developed the diode further to what is currently called the triode, which amplifies the voltage of an incoming signal, for example from a radio antenna. This invention eventually enabled radiotelephone equipment for use in aircrafts; although heavy and voluminous, this represented the first electronic equipment for use in aircrafts, named avionics. After the initial steps of the evolution of electronics, not much happened with the technology on which radio (1910´s), television (1930´s), and the electronic computer (1940´s) was based. By the time jet propulsion was emerging as the dominating technique to enable forward motion of aircrafts, in 1947 the transistor was invented [2]. The transistor was more reliable, smaller, lighter, and consumed less power than the vacuum tube, and had soon replaced the vacuum tube in electronic equipment, including avionics. This represented path-breaking development of technology and the beginning of an avalanche in utilization of electronics. In the early 1960´s, the single transistor was extended by the invention of the integrated circuit, and in 1965 Gordon Moore made his famous prediction that the number of transistors on a single chip would double every year the following ten years. With some adjustments, Moore’s prediction extended much longer than to 1975, and is still valid, and even extended to “More than Moore", which includes diversification of the parts that can be integrated in one electronics package [3]. Currently, the size of a transistor on a silicon integrated circuit is in the scale of 101 nanometers, and single-molecule transistors are produced, which are made of carbon nanotubes and silicon nanowires with diameter of less than ten nanometers [4], [5]. Thus, the expression avalanche in electronics design might be justified considering a technology developing from vacuum tubes to nanotubes in just about half a century. With regards to airborne electrical equipment, a modern aircraft may be equipped with hundreds of avionic units, with functionality ranging from passenger entertainment, to safety-critical flight control systems. 1 Chapter 1 The downsizing of electronics has also led to the need of a multi-scale approach in product development, both considering time and geometry. To minimize the power dissipation in a transistor, the switch rise and fall time is optimized, gaining a few nanoseconds, while the component must prove a lifetime of many years. The size of the transistors on an integrated circuit is in the range of 10 nanometers, whereas the length scale of the entire component package is several centimeters. Another challenge currently arises from the increased complexity in product development, since multiple technology areas need to develop simultaneously and collaboratively. New materials are continuously developed, and in the development of integrated circuits, chemistry, metallurgy, physics, electronics, and mechanics engineers need to cooperate. Profound requirements are put on the quality of the product and the design process, including robustness to altered boundary conditions, and reliability. A large number of tests are to be performed on every component and system, including thermal, mechanical, moisture, vibration, shock, chemical, and electrical tests, as well as combinations hereof [3]. Each active transistor within an electronic component generates heat due to switching losses and electrical resistance in the semiconductor. Hence, the continuing reduction in size of electronic components, concurrent with increasing clock frequencies and greater functionality, result in increasing volumetric heat generation and surface heat fluxes in many products. As an effect, keeping temperature levels of electronic components within their maximum ratings is central in electronic product development, affecting all levels of the electronic product hierarchy, from the chip to the complete system. The restriction of hazardous substances (RoHS) directive [6] limits the use of lead in the manufacture of various types of electrical and electronic equipment. The knowledge about field operation reliability of lead-free solders is still limited, as well as the availability of models for prediction of service life of the solder joints. Combined with low-volume manufacturing of potentially safety-critical equipment, many manufacturers of electronics for the aerospace, defense, and high performance (ADHP) equipment are therefore currently exempt from the RoHS directive. However, manufacturers of commercial off the shelf (COTS) electronic components have since several years converted to producing lead-free electronics. Since ADHP manufacturers often utilize COTS electronic components, the transition to lead-free electronic equipment will most probably propagate also to manufacturers of electronic equipment for avionic applications. Accordingly, the need to fill in several knowledge gaps, including reliability and reliability prediction of lead-free solder alloys, is emphasized. 1.1 Thermal Challenges for Avionic Equipment The thermal challenges that emerge from the downsizing of electronics are common to most electronic equipment. For avionic applications, the measures to overcome these challenges are however constrained by the obvious and strong driving forces to minimize weight, and maximize reliability. Avionics applications can be divided into military and civilian applications, where military applications have historically represented the driving force in the 2 Introduction evolution of avionics. In the past, design of military electronic equipment embodied the cutting edge in technology, as military applications represented more than 50% of the total semiconductor market in the 1960´s [7]. Currently, ADHP applications represent approximately 2% of the semiconductor market [8] and thus render no interest of major semiconductor manufacturers to make such parts. Therefore, COTS electronic components designed for computers, consumer, and telecommunications applications, which today represent more than 75% of the semiconductor market, are used to a growing extent in ADHP applications. From a thermal perspective, high-performance COTS components are often cooled by adding a fan-cooled heatsink with spring-loaded attachment to the top surface. This is adequate for stationary computers that are subjected to insignificant vibration loads during its lifetime. However, for military and/or avionic applications, this would not be a feasible design due to multiple reasons such as weight, volume, and vibration levels. Instead, focus might be set on minimizing contact resistance to a cooling surface, optimizing heat conduction, and affecting the power distribution on the printed circuit board assembly (PBA) in collaboration with electronic design engineers. For avionic applications, the trend of more electric aircraft (MEA) generates additional thermal management challenges, as an increasing amount of avionic systems are employed in harsh environments, subject to large variations in ambient temperatures, external thermal loads generated by nearby high-power dissipating equipment, and inherently transient internal power dissipation [9]. 1.2 Reliability of Lead-Free Electronics Facing the transition to Pb-free electronics, the ADHP industry currently needs to know that utilized electronic packages can be qualified for field operation with sufficient reliability and lifetime. In the research community, large effort has been put on determining whether lead-free solder can replace the well-proven tin-lead solder with regards to thermal fatigue life [10]–[14]. This process is however difficult to complete, due to lack of field data on the large number of lead-free solder alloys that exist on the market, variation of the alloy composition of a specific solder alloy between different suppliers, and high dependence of initial microstructure, and thus reliability, on manufacturing parameters. During product development for avionic applications, the electronic packages must be qualified for field operating conditions that vary depending on the specific location of the equipment (for example in zones with controlled or uncontrolled environment in an aircraft). In qualification of a certain package type for different thermal environments, accelerated testing is the predominating method, although acceleration factors for lead-free solder joints are still largely unknown. Emphasized by observations of degraded reliability during operation testing of U.S. Army systems, incorporation of physics-of-failure (PoF) in reliability prediction of electronics currently attracts a lot of attention by the military and aerospace industry [15]. Until recently, the physics-of-failure contribution to reliability prediction defined in industrial handbooks such as MIL-HDBK 217F [16], extended to calculations of 3 Chapter 1 thermal cycles to failure, based on simplified analytical correlations. However, in a recent industrial standard [17], a number of physical failure mechanisms are defined, as well as suggestions on how to evaluate them. Specifically for thermal loading of solder joints, it is recommended to employ finite element (FE) analysis in order to evaluate damage accumulation during one thermal cycle, and subsequently calculate the number of cycles to failure [17]–[20]. Obviously, this assumes reduction of field conditions to cyclic loading, which might result in reduced accuracy of the predicted lifetime. Evaluation of damage accumulation in solder joints for field conditions without temperature data reduction is a task that still poses a large challenge for fullscale FE analysis, in terms of computational resources. In case the expected operating environment is known, but not explicitly expressed as a combination of thermal cycles, rainflow cycle counting can be applied to convert the application environment to temperature cycles and half-cycles [21]. Subsequently, constitutive laws can be applied along with FE analysis and fatigue laws to predict the thermal cycling reliability [22], [23]. However, the cycle-counting approach contains assumptions regarding the ratio between dwell time at a certain temperature level and the ramp time for each counted cycle [24], [25]. Hence, compared to the cycle-counting approach, improved accuracy of the computed damage in solder joints would be achieved using uncompressed temperature data as input to the same lifetime prediction method. As a consequence of the increased transient variation of thermal loads expected from the MEA trend, it may furthermore be beneficial to employ advanced thermal analysis such as computational fluid dynamics (CFD) to supply detailed temperature data as input for the FE analysis. However, direct usage of FE analysis for evaluation of accumulated damage in a solder joint from uncompressed temperature data would lead to long computational time, and require large computational resources. Therefore, there is a need for a computationally more efficient method to assist designers in quick evaluation of the accumulated damage in solder joints for non-modified operating temperature profiles. 1.3 Objective, Scope, and Delimitations The power density of electronic components is ever escalating [26], thereby continuously creating new challenges to the avionics industry with regards to reliability. The use of commercial off the shelf (COTS) electronic components in avionic applications brings lead-free electronics, which has been insufficiently verified with regards to long-term reliability, closer to the avionic industry. Today, the impact of this evolution on field reliability of avionic equipment is at large unknown. One measure that has been proposed in order to increase the knowledge in this area is to introduce physics of failure (PoF) in reliability prediction [17]. The objective with this thesis is twofold: To study and analyze thermal challenges for avionic equipment and investigate power distribution on an avionic printed circuit board assembly (PBA) as a thermal management measure, and to address the identified need of physics of failure in reliability prediction in the product design phase of avionic equipment by testing the following hypothesis: 4 Introduction It is possible to develop a method that provides rapid evaluation of damage accumulation in solder joints in field temperature conditions, such that: It utilizes non-cyclic, non-compressed temperature data of anticipated field operation temperature environment. It provides accuracy comparable to three-dimensional finite element (FE) analysis. Its computational efficiency is significantly higher than FE analysis. The thermal challenges are addressed in chapter 2, by presenting fundamentals of heat transfer, thermal fatigue failure mechanisms, and state of the art thermal design and design tools, with focus on the requirements that may arise from avionic use. For example, such requirements comprehend low weight, high reliability, and ability to sustain functional during high vibration levels and high g-forces, which all affect the feasibility of some thermal management measures. Paper I, II, and III provide an indepth investigation utilizing advanced thermal modeling of power distribution on a double-sided PBA as a measure to improve its thermal performance. The need in the design phase for quick evaluation of damage accumulation in solder joints is addressed in chapter 3. With regards to thermal fatigue of solder joints, the current state-of-the-art physics-based reliability prediction comprises simplifying the thermal loads, to cyclic loading with assumptions regarding the ratio between dwell times and temperature ramp times. Following this, analytical life prediction models are employed to calculate the number of cycles to failure. Hence, one step to take in order to increase the potential accuracy of the prediction would be to employ the same analytical life prediction models, but enhance the accuracy by including advanced thermal analysis in the procedure (paper IV). Reliability engineers would benefit from a life prediction model that was validated for non-cyclic temperature variations at field operation temperature levels. The maximum possible accuracy of the life prediction would then be achieved by 3-D finite-element (FE) modeling of stress and strain during a number of anticipated typical field operation temperature profiles. Even without fully validated life prediction models, this kind of calculation would today bring higher credibility to the lifetime prediction, and supply increased understanding of the impact of different temperature profiles. However, FE modeling can require large computational resources and long calculation time. Therefore, a computational method is proposed that utilizes surrogate stress and strain modeling of a solder joint, to quickly deliver damage evaluation for a solder joint subjected to a non-cyclic, non-simplified field operation temperature profile. This is treated in Papers V and VI. Much research is ongoing to create constitutive laws for the emerging lead-free solder alloys [19], [27], [28]. Extending the horizon further, to a future when validated life prediction models will be available, the suggested computational method is expected to enable increased accuracy of embedded prognostics of remaining useful thermal fatigue life of a critical solder joint. Chapter 4 presents an overview of prognostics of electronics, and a brief description of the potential prognostics 5 Chapter 1 application of the suggested method, with more details provided in Papers VII and VIII. There is a multitude of failure modes, mechanisms and locations in all levels of avionic equipment. With regards to evaluation of damage accumulation, this work has however been limited to focus on thermal fatigue of solder joints. A concluding summary of the work, reflecting on the contributions to the field, ends this thesis. 6 2 Thermal Management of Avionic Equipment 2.1 Heat Transfer Fundamentals Heat transfer is the science that seeks to predict the transfer of energy that takes place between material bodies as a result of a temperature difference [29]. There are three modes of heat transfer: conduction, convection, and radiation. Conduction is the transfer of energy within a body due to a temperature gradient. The energy is conducted from the high-temperature region to the low-temperature region (see Figure 1). In its simplest form, one-dimensional heat flow by conduction is calculated as: q cond tA , l where l is the length of the conducting path, A is the area of the conducting path, the thermal conductivity of the body, and t is the temperature difference. Figure 1. One-dimensional, steady-state conduction heat flow [30]. 7 (1) is Chapter 2 For the three-dimensional case, conduction heat transfer is expressed as an energy balance for an infinitesimal element, such that (in Cartesian coordinates): x T x y T y z T z q cp T t (2) where the left hand side describes the net transfer of thermal energy into the control volume and the energy generated within the infinitesimal element, and the right hand side is the change in thermal energy storage in the element. Convection is the energy transfer from a body to a fluid, which might be gas or liquid. A difference is noted between forced convection, where the fluid is propelled by a fluid accelerating device such as a fan, and natural convection, where the motion of the fluid is initiated by a change of density due to heating of the fluid (see Figure 2). The force that arises from this phenomenon is called buoyancy force. The energy transferred by convection is calculated as: q conv hA(Ts Tamb ) (3) where h is the convection heat transfer coefficient that depends on the properties and velocity of the fluid, A is the area of the heat dissipating surface, Ts is the temperature of the surface, and Tamb is the temperature of the ambient air. Figure 2. Temperature and velocity distributions for natural convection in air near a heated vertical surface. Upward movement of hot air (a). Distributions at arbitrary vertical location (b). is the distance at which the velocity and the temperature reach ambient surrounding conditions [31]. 8 Thermal Management of Avionic Equipment Thermal radiation is when a surface emits electromagnetic radiation as a result of its temperature. The frequency of the radiation depends on the absolute temperature of the radiating device; however for the temperature range possible for electronics to operate within, thermal radiation mainly occurs in the infrared frequency range. The energy transferred between two bodies by radiation can be calculated as: q rad F FG A(Ts 4 4 (4) Tamb ) where F is an emissivity function, FG is a geometric “view factor” function, is the Stefan-Boltzmann constant (5.669 10-8 W/m2 K4), A is the area of the radiating surface, Ts is the radiating surface temperature, and Tamb is the temperature of the receiving surface. Depending on the application, any of the three energy transfer modes can be the dominant mode for removing the energy, and thus the heat, from the electronic equipment. In Table 1, a coarse estimation of thermal conductivity and convective heat transfer coefficient values is provided. Table 1. Approximate values of conductivity and convection heat transfer coefficients for different heat transfer modes. Heat transfer mode Conduction in solids Thermal Conductivity (W/(m K)) 0.13-2000 Heat transfer 2 coefficient (W/(m K)) Reference [32] Natural convection in gases 5-15 [33] Forced convection in gases 15-250 [33] Natural convection in liquids 50-100 [33] Forced convection in liquids 100-2000 [33] Boiling liquids 2500-35000 [29] In ground applications, e.g. stationary computers, the critical components may be cooled by conduction to a heatsink, which is in turn cooled by forced convection. Other options exist for removing the heat dissipated by current high-performance processors, but ultimately the heat is transferred to the surrounding air by forced or natural convection. Ground applications can of course be divided into a vast number of categories, ranging from the automotive industry, through military equipment, to handheld devices, such as mobile phones, but the general principle of conduction plus convection cooling remains. Avionics applications may utilize the same cooling principles as ground equipment. In avionics, however, constraints are posed on the cooling solutions in terms of e.g. minimized weight, extreme reliability requirements, and environmental requirements such as harsh temperature levels, high levels of vibration, and lowdensity cooling air at high altitudes. In space, the main principle for cooling of electronic equipment differs completely from the previously mentioned applications, due to the absence of air. Radiation is the only mechanism that transports heat to and from a spaceship or a satellite. Therefore, control of the temperature of the electronics is realized by carefully utilizing radiating and reflecting surfaces on the hull of the vehicle. 9 Chapter 2 2.2 Thermally Induced Failure Mechanisms In order to encounter the problems that arise in avionics as a result of thermal loading, a basic understanding of failure mechanisms is needed. A failure mechanism is the mechanical or chemical mechanism that causes the failure mode, which is often a short or open electric circuit, at a specific location – the failure site. Although failure modes are often identified on system level, the failure mechanisms are active on the lowest hardware level – the failure site may often be located within an electronic component, or at the interface between the component and the PCB. The connections between electronics reliability and thermal loading are extremely complex and cannot in general be represented by analytical correlations, which provide an unambiguous answer to the lifetime of a certain component, subsystem, or system. On all failure sites of an avionic system (from on-chip to system of systems), a diversity of failure mechanisms and failure modes exist, which may eventually cause breakdown of system functionality. In Table 2, the failure mechanisms are shown, which account for the majority of failures in electronic systems, classified according to the packaging level where the respective failure mechanism may occur. Table 2. Failure sites, modes, and mechanisms at different packaging levels of electronic systems (modified from [34]). Packaging level Failure site Die metallization Failure mode Short circuit Open circuit Breakdown Level 0 (chip and on-chip sites) Level 1 (parts and components that cannot be disassembled and reassembled with the expectation that the item would still work) Gate-oxide Die Transistor Between die and molding compound Level 3 (enclosure, chassis, drawer, and connections for PBAs) Level 4 (entire electronic system) Level 5 (multi-electronic systems and external connections between different systems) Change of leakage current Crack Short circuit Delamination Bond wire Open circuit Encapsulant interface Capacitors Delamination Short circuit Open circuit Short circuit Open circuit Short circuit Solder joint Level 2 (printed circuit board (PCB), and interconnects connecting the components to the PCB) Short circuit Printed-through hole/ Via Printed circuit board (PCB) Metallization shorts Lead pad Loss of polymer strength Open circuit Trace Open circuit Connection Open circuit (often intermittent) 10 Failure mechanism Electrochemical migration Electromigration Electrical overstress (EOS) Electrostatic discharge (ESD) Time-dependent dielectric breakdown (TDDB) Hot carrier Crack initiation and propagation Contact migration Crack initiation and propagation; popcorning Bond lift due to mechanical overstress Corrosion Corrosion Dielectric breakdown Thermal fatigue, vibration fatigue Tin whisker growth Thermal fatigue Electrochemical migration Conductive-filament formation Electrochemical migration Glass transition Corrosion Corrosion Electromigration Mechanical wearout, corrosion, fretting Thermal Management of Avionic Equipment Some of the failure mechanisms seen in Table 2 are not clearly dependent on temperature, for example corrosion [35]. However, below follows short descriptions of the most clearly thermally induced failure mechanisms. Delamination The reliability of general multilayer microelectronic devices is influenced by the interfacial strength (adhesion) and resistance to fracture (debonding) of the many bimaterial interfaces. Residual stresses, thermo-mechanical cycling and mechanical loading may drive time dependent fracture in the multilayer structures, allowing for ionic contaminants to cause corrosion-induced failures, or immediate electrical failure by sheared or cratered wire bonds [36]. When two joined materials are subject to thermal loads, stresses can be produced at the material interfaces due to coefficient of thermal expansion (CTE) mismatch between the materials. These stresses may cause delamination of the materials and hence affect overall reliability of the system [37]. Apart from differences in CTE, defects in the attachment layer may cause interfacial debonding. One common defect is voids embedded in the packaging of the component. Voids can form from melting anomalies associated with oxides or organic films on the bonding surfaces, trapped air in the attachment, local non-wetting, outgassing, and attachment shrinkage during solidification [38]. For plastic encapsulants, which are inherently hygroscopic, moisture present in the layers can lead to premature failure and reduced lifetimes of these devices. The “popcorning” failure mechanism appears at high temperature, such as during the reflow soldering assembly process, when the moisture vaporizes and expands, generating high stresses that may cause delamination [39]. Bond wire fatigue Mainly due to the differences in CTE mentioned above, the bond wires that electrically connect the chip to the leadframe are subject to mechanical fatigue when exposed to thermal cycling. The interfacial strength between the bond wire and the bond pad can be reduced by diffusion-induced brittle intermetallic phases or Kirkendal voids at the interface. Thermal fatigue of solder joints Thermal fatigue of solder joints occurs when the CTE of the PCB differs from that of the electronic component attached to the PCB, and the assembly is subject to temperature variations. Shear strain imposes stress in the solder joint, and the slow process of temperature variation leads to stress relaxation primarily by creep strain within the solder joint. Repeated stress relaxation leads to solder fatigue, crack initiation, and crack propagation until electrical failure is a fact. Figure 3 shows schematically the principal failure sites in a solder ball subject to thermal fatigue loads. Evidence of thermal loadings can be found in the microstructure of the solder joint. Both grain coarsening and growth of intermetallic compounds (IMC) between the copper pad and the bulk solder can be indicators of thermal cycling material degradation [40], [41]. Figure 4 shows an example of a BGA solder ball with Sn3.0Ag0.5Cu solder alloy, before and after harsh thermal cycling between -55°C and 11 Chapter 2 +125°C. Before thermal cycling, no grains can be distinguished, whereas after thermal cycling, grain boundaries are visible using differential interference-contrast filtering on an optical microscope. Figure 3. Schematic presentation of main failure sites in a solder ball subject to thermal fatigue loads. 50µm 50µm Figure 4. Optical microscopy image of SAC305 solder ball after manufacturing (left), and after 2257 thermal cycles between -55°C and +125°C (right). Differential interference contrast filtering enhanced by increased contrast level reveals that there are no visible grain boundaries before thermal cycling, while grains are clearly visible after thermal cycling. 2.3 Thermal Design Tools In the different product development phases, various types of thermal evaluations are required. In the quotation phase, there is a need for a quick approximation of the thermal performance of a device; during the conceptual design phase a tool is needed that can make time-efficient estimates for comparing different thermal designs. In the detailed design phase, there is a need for more detailed thermal simulations, which are able to identify potential hot spots and evaluate different ways of mitigating these. For use in these design phases, six classes of tools have been identified: Hand calculations Spreadsheets 12 Thermal Management of Avionic Equipment Flow network modeling (FNM) Computational fluid dynamics (CFD) Finite element method (FEM) Experiments A summary of the identified thermal design tools, with examples of tools in each classification as well as significant features, is presented in Table 3. In the following, the variety of tools is described somewhat more in detail. Table 3. Classification of thermal design tools. Classification Hand calculations Tool Thermal Resistance Network (TRN) Spreadsheets Microsoft Excel Flow Network Modeling (FNM) MacroFlow Computational Dynamics (CFD) Coolit, Flotherm, Icepak, MacroFlow, Thermal Desktop, ANSYS, Flovent Fluid Finite Element Method (FEM) ANSYS, Matlab, Multiphysics Comsol Experiments Physical prototypes, burn-in tests Features Rapid solutions. Renders a good overview of influence of different designs on thermal performance. Handles advanced mathematical and engineering functions. Macros Graphics capabilities Data table formatting Efficient for solving complex TRN:s. Design of different types of air- or liquidcooled electronics systems. Quick evaluation of system-level thermal design. Flow constraints important. No component temperature predictions – only system level. Detail-level simulations and analyses of heat transfer and fluid flow. Extensive calculation times when modeling with high level of detail. Detail-level analysis of primarily conduction cooling. Thermal stress analysis. The classic way of evaluating a product design. Requires knowledge of measurement techniques. Hand Calculations In an early stage of product design, in the stage of the company trying to win an order to develop a certain product, a quick estimation of the thermal performance of the product would be appreciated. Here, hand calculations can be utilized. Given the total dissipated power, a rough estimation of the mechanical design of the avionic device, and the thermal boundary conditions, thermal resistance networks (TRN) that represent the different heat transfer paths within the system can be modeled. A TRN may provide rather accurate predictions of relative heat transfer efficiency when comparing different design alternatives. 13 Chapter 2 Spreadsheets When TRN:s get too complex to calculate by hand – many thermal resistances in parallel that lead to areas of different temperatures is the most common case for avionic equipment – there is an option to use spreadsheets to handle the large number of equations that should be solved simultaneously. Spreadsheets handle advanced mathematical and engineering functions, and provide graphical capabilities for displaying results. Flow Network Modeling A tool that could be used for more detailed analyses, still in the early phase of product design, is flow network modeling (FNM). FNM is a generalized methodology to calculate system-wide distributions of flow rates and temperatures in a network representation of a cooling system. Practical electronics cooling systems can be considered as networks of flow paths through components such as screens, filters, fans, ducts, bends, heat sinks, power supplies, and card arrays [42]. To be able to use FNM, empirical correlations of the impact on the flow from these components are a prerequisite. This technique does not give the detail level results for predicting component temperatures, but in order to quickly estimate average temperatures of subsystems and compare the heat transfer efficiency of different designs, FNM is a useful tool. For avionic applications, FNM may prove useful in evaluation of the cooling air system in an aircraft, while it may not add much value in thermal design evaluation of a single avionics enclosure. Computational Fluid Dynamics Computational fluid dynamics (CFD) is a tool that performs calculations of flow, heat, and mass transfer in complex geometries, at any level ranging from electronic component design to heating, ventilation, air conditioning and refrigeration (HVAC) flows in e.g. buildings. The main issue with CFD is that extensive calculation times are required when the detail level gets high. At present, much effort is put on developing systems for transferring CAD-data directly to the CFD software, including automatic simplifications of detailed geometries in order to accelerate the CFD calculations. Finite Element Method The main use of finite element method (FEM) has traditionally been analysis of mechanical stress and deformation, although conduction heat transfer can also be included in the calculations. Software packages are also available, which combine FEM and CFD solvers, in order to model coupled physical phenomena that can be described by partial differential equations. For example, thermoelectric cooling coupled with all three modes of heat transfer could be analyzed. Experiments By many considered the only true way of estimating temperatures within a system, experimental measurements might however not be the obvious way to retrieve correct data. The environmental specifications have to coincide with the actual operating conditions, and the experimental setup should correspond to reality. 14 Thermal Management of Avionic Equipment The final design tool, classified under experiments, is the burn-in test, in which the finalized device is exposed to severe environmental loading while operating. The purpose of the burn-in test is to identify errors that might occur in the initial stage of the bathtub curve [38], which is further described in chapter 3 below. Burn-in should be carefully balanced though; firstly, as this procedure is a kind of accelerated test it is of vital importance to ascertain that failure mechanisms that may occur during the burn-in also may occur in real operating conditions. Secondly, caution must be taken not to damage fully functional parts and shorten the lifetime of the device. With regards to thermal fatigue of solder joints, currently available lifetime prediction models may supply an adequate estimation of the consumed life during a number of burn-in cycles. A lot of experimental data has been published, and constitutive laws for creep rate have been derived, which comprise harsh thermal loading that is the case during burn-in testing. However, addressing the impact on lifetime of the simultaneous thermal and vibrational loading is not possible with the information available today. In practice, all of the tools above except the burn-in test can be used for dynamic as well as static temperature predictions. The precision in transient calculations varies accordingly with the tool used, and the design stage in which the calculations are performed. 2.3.1 Uncertainty in Simulations When a CFD simulation is initiated to predict airflow and heat transfer in an electronic system, many approximations as compared to real-life physical components are introduced. Together, these approximations may cause the result to differ substantially from reality. Lasance [43] summarizes the errors from the input data and from the numerical setup to a final error, which is in the order of 20%. It should be noted, that when speaking of percentage errors in temperature, the temperature rise relative the ambient temperature or the temperature of the cooling air must be referred to. In [44], a methodology to follow before initiating a CFD simulation is proposed as seen in Figure 5. 15 Chapter 2 1. Problem definition 2. Physical/computational domain 3. Computational grid 4. Model selection 5. Material selection and properties 6. Boundary conditions (BC) 7. Solution strategy 8. Post-processing Objectives of model? Degree of accuracy needed? Model part of, or full system? Geometry created in CFD or imported from CAD? Two- or three-dimensional or axi-symmetric? Can symmetry be used? Type of mesh? Regions in need of finer mesh? Isothermal, conjugate, or conduction only problem? Laminar/turbulent, steady/unsteady, compressible or incompressible flow? Near-wall function? Influence of buoyancy and/or radiation? Thermal properties independent of absolute temperature? PCB:s - use effective thermal conductivity? Isotropic or orthotropic properties? Simplification by symmetry, cyclic or periodic BC? Are all BC:s known? Convection BC:s possible? Algorithm? Higher order interpolation scheme needed? Convergence time? Do the results make sense? Refinement needed? Problem objectives satisfied? Figure 5. General methodology before running CFD [44]. 2.4 State-of-the-Art Thermal Management in Avionics Examining different cooling technologies, it is easy to start to compare heat flux capacity. However, it must not be forgotten that the critical component might be a lowpower component, which due to its placement on the PBA gets too hot, faces too large temperature gradients, or experiences too high thermally induced mechanical loading due to its size. Since the significant metrics a thermal design engineer must meet are junction and solder joint temperatures, the main feature to compare between different cooling technologies is not heat flux capacity, but applicability for the current cooling requirement [45]. This chapter explores important thermal and practical attributes of current state-of-the-art cooling schemes. In Table 4, a reviewing list of various cooling 16 Thermal Management of Avionic Equipment technologies is provided, with significant features of each technology identified. Technology maturity and the applicability for use in avionics applications are graded from 1 to 5, where 1 means emerging or unsatisfactory, and 5 denotes mature or outstanding. Table 4. Review of cooling technologies. Cooling technology Heat pipes Spray cooling Jet impingement Immersion cooling Phase transformation solid-fluid Phase transformation for thermal energy storage Forced convection direct air cooling Thermoelectric cooler Hybrid cooling Managing power distribution Significant features Functionality independent of orientation Highly efficient heat removal. Wetting of electronics however requires a need for electrical isolation. Used in some thermal interface materials to minimize thermal contact resistance Useful for capacitive storage of heat at transient temperature extremes Possible only when the cooling air is noncontaminated Electronic refrigerator with low efficiency Multiple thermal interface resistances Distribute heat sources as evenly as possible Maturity 4 Avionic use 3 Reference [33] 3 3 [29] [46], [47] 4 3 [48] 4 4 3 2 5 5 4 1 [45] 5 5 [45] 5 5 [49] Heat Pipes Heat pipes belong to the passive cooling technologies. A heat pipe is an evacuated, vacuum-tight envelope, outer diameter from 3 mm and up, with a wick structure on the inside (see Figure 6). It contains a small amount of working fluid, which in the isothermal state is uniformly distributed over the wick structure by capillary forces. When heat is applied anywhere on the heat pipe, the working fluid at this location vaporizes. Since the envelope is evacuated, the vaporized working fluid spreads immediately over the entire volume inside the heat pipe to establish uniform pressure in the contained volume. At the condenser area, the vaporized working fluid is condensed into the wick structure releasing its latent heat of vaporization, and is returned to the evaporating area by capillary forces. Heat pipes operate independent of gravity, although the performance is maximized when the orientation is vertical and the capillary forces are assisted by gravity to transport the working fluid back to the area of vaporization [50]. Functionality can be negatively affected by g-forces exceeding 5 g [51], wherefore implementation in avionics should be treated with great care. 17 Chapter 2 Figure 6. Heat pipe function [52]. Spray Cooling/Fluid Jets Spray cooling and fluid jets both uses fluid impingement directly onto the electronics for heat removal. Spray cooling utilizes small droplets that evaporate from the heat-dissipating surface, whereas jet impingement consists of a number of jet streams that create a continuous fluid film on the surface, thereby transporting the heat away by convection to the fluid. Comparisons made in [46] indicate a higher efficiency of jet streams in terms of the ratio between dissipated heat and power consumed by the fluid pumping system. Using deionized water as a coolant, heat fluxes as high as 300 W/cm2 at 80 C surface temperature could be removed from the 5.0 8.7 mm2 surface of a diode. Water can be used as the coolant provided that the components are electrically isolated from the fluid, but in return water has much higher heat capacity than the dielectric fluids that may be impinged directly onto electrically conductive surfaces. Both these technologies represent high-efficiency and high-complexity cooling solutions, which may be utilized in avionic applications unless no other option provides adequate cooling. Reliable fluid pumping systems are required, and filtering techniques to avoid clogging of nozzles, which could result in catastrophic temperature levels. Immersion cooling A PBA can be fully submerged in a container of dielectric fluid. The need for electrical isolation excludes water as coolant, which otherwise would have been highly efficient considering heat removal capacity. As an example, the power supply unit (PSU) for a radar array for the F-18 fighter is liquid-cooled in a flow-through cooling system design. The total power dissipation of the PSU is 400 W, and the maximum temperature of the device is 75 C at an inlet temperature of the cooling fluid of 15 C [48]. Phase transformation solid-fluid Due to the increased volumetric heat flow in electronics, continuous research is carried out to minimize thermal contact resistance between the heat source and the cooling system. Thermal interface materials (TIM) are available, which when heated change phase from solid to liquid. This means that when the heat source starts heating up, the TIM wets the surface of the heat source, enabling optimal thermal contact. It is 18 Thermal Management of Avionic Equipment vital to realize that the contact pressure between the cooling system and the heat source should preferably remain constant at both phases. This may be enabled by a spring-loaded attachment of the heat source to the cooling system, which may be possible in avionic systems in case the heat source has relatively low weight, such as switching transistors in power converters for electric motor drive. Phase transformation for thermal energy storage The power dissipated by an avionic system may not be uniform over time, as is neither the environmental conditions. Phase transformation of for example a polyalcohol material or paraffin can be used as a capacitive storage of heat dissipated under worst-case conditions. Research is currently ongoing to improve the low thermal conductivity of such materials, which is limiting the usage of this technology in avionic applications [49]. Forced convection direct air cooling Direct air cooling of the electronic components and the PCBs, is an efficient, light-weight, and cheap cooling solution. Drawbacks for avionic use are the requirements put on the cooling air such as content of moisture, particles, and oil, as well as reduced cooling capacity at high altitudes, and the common operational mode with loss of cooling air. Furthermore, since the PBAs do not require mounting on a card carrier for conductive heat removal, which also acts as mechanical stabilizer, the PBAs may need to be mechanically reinforced to reduce the impact of vibration. Historically, thermal design for cooling electronics at high altitudes has been a matter of design tolerance. With a large margin between the operating temperatures and the maximum temperature, it has been accepted that the margins are smaller at high altitudes, but still within the allowed design space. As design margins are shrinking, this is currently not good enough. The problem can be overcome by introducing fully pressurized cooling systems for all the critical electronics, but this adds great complexity, and increased cost and weight. Instead, higher flow of air through the cooling system has to be admitted when the air density decreases [53]. This way the impact of altitude on the cooling capacity is reduced, though not kept constant. Thermoelectric cooler A thermoelectric cooler (TEC) uses the temperature difference that arises when an electric current flows through a circuit in which two different metals are joined (Peltier effect). In electronics cooling applications, this effect can be used primarily to cool point sources of heat. One way of rating a TEC is to study the coefficient of performance (COP) of the TEC, defined as the ratio of transferred heat to input power [33]. The COP of a TEC is often low, in the range of 1 [54], compared to larger scale machines such as air conditioners or refrigerators with a COP of 3 to 5. Hence, TEC have nearly no use in avionic equipment since the total power dissipated in the unit will increase significantly. 19 Chapter 2 Hybrid cooling Hybrid cooling is a comprehensive term for cooling schemes, which incorporates more than one cooling technology, e.g. liquid cooling in combination with liquid-to-air heat exchanger. Examples of hybrid cooling schemes are shown in Figure 7 [45], [55]. For all of these designs, the liquid may be replaced by air, providing greatly reduced cooling capacity, but also substantially less weight, complexity, and cost. The leftmost solution is the most economical design, providing fair temperature gradients at moderate power loads. For example, at approximately 50 W dissipated into a heat sink/card carrier with the approximate dimensions 250x200x4 mm3, a total temperature rise in the range of 7°C can be expected from the wedge clamp; 2-3°C in the contact resistance between card carrier and the card-edge heat exchanger, and 45°C in the card carrier. Since this design assumes a solid card carrier, commonly aluminum, this detail supplies thermal mass, which helps to reduce the transient temperature rise in case the cooling system temporarily is disabled. The other designs provide more efficient cooling of the electronics, but also greater complexity of the system, and reduced capability to handle loss of cooling, which is a common requirement to sustain for a limited time in avionic applications. Figure 7. Avionics hybrid cooling schemes [55]. Price and Short [56] describes the thermal design of an airborne computer chassis situated in an electronics pod, which is suspended from the fuselage by pylons. The computer consists of 24 PBAs that are mounted in card slots in the chassis (see Figure 8), dissipating in total 400 W. The PBAs are cooled by conduction to pin fin heat sinks integrated in the top and in the bottom of the chassis, cooled by forced convection 20 Thermal Management of Avionic Equipment obtained by a fan included in the unit. The design approach utilized to enable operation at low air pressure was to define the maximum allowable cooling air supply temperature, ranging from 55 C at sea level, to 26 C at 13,700 m altitude. Figure 8. Exploded view of air-cooled computer chassis [56]. 2.5 Managing Power Distribution In contrast to the elaborated computer enclosure briefly discussed above, a costfree and weight-free measure to improve the thermal performance of electronics, regardless of air density and all other environmental loads, is available by managing power distribution on a double sided PBA. Papers I, II, and III provide an in-depth investigation of this thermal management measure. The avionic unit studied in this context consists of three PBAs, housed in a 2 Modular Concept Unit (MCU, ARINC 600 standard) sealed enclosure, including one double-sided PBA (DSB). The PBAs in the unit are cooled by thermal conduction to the walls of the enclosure. Ultimately, the heat is removed from heatsinks integrated in the enclosure walls by forced convection cooling air (see Figure 9). 21 Chapter 2 Figure 9. Avionic enclosure studied in the context of managing power distribution (left), and double-sided PBA, mounted to enclosure side wall, with 24 individually controlled power sources on each side of the PCB (right). Evaluation of different power configurations has been formulated as a problem of determining non-dominated designs, as exemplified in Figure 10. Assume that the system performance depends on r discrete design variables q1,…, qr representing a point q = (q1,…, qr) of an r-dimensional space D. Let Q = QT QS be a finite set of feasible alternatives or designs, where QT D is a set of designs to be explored satisfying the design variable constraints based on technical specification due to production and/or application reasons, and QS D is a set of designs for which state variable constraints are satisfied, e.g. thermal constraints. Let also Y ={ym , m = 1,...,M} be a finite set of attributes (e.g. performance criteria), which are considered to be minimizing. A design p Q is called a non-dominated design [57], [58] if there exists no design q Q, such that ym (q) ym ( p) for all m 1,..., M , (5) and ym0 (q) ym0 ( p) for at least one m0 1,..., M . 22 (6) Thermal Management of Avionic Equipment Max case temperature (fluid side), oC (HS side [W], fluid side [W]) = (24, 6) 95 94 93 8 92 5 1 91 90 2 89 4 7 88 86 87 88 3 89 90 91 92 93 94 95 o Max case temperature (HS side), C Figure 10. Domination graph: arrows are directed from dominating designs toward dominated designs. An experimental setup has been created that enables full control of the power dissipated by each component placed on the DSB, as seen in Figure 11. The initial investigations have been conducted on DSB with uniform power configuration, that is, the same power is applied to every component on one side of the DSB. For the nonuniform power configurations shown in Figure 12, high-power components on the primary side of DSB are placed opposite to low-power components on the secondary side, and vice versa. Such a configuration is considered reasonable to keep the components on each side of DSB as cool as possible. In this study, power configurations including components with equal power forming relatively large groups on the board have been analyzed. The highlighted power configuration in Figure 12 results in the lowest maximum temperature on the DSB. CFD simulated temperatures of each power configuration are shown in Figure 13 and a comparison between measured and simulated temperatures for the power configuration highlighted in Figure 12 are provided in the contour plots seen in Figure 14. In an environment of 55°C cooling air supplied at a rate of 6.7 g/s to the 2MCU enclosure, with 39 W of power dissipated by other PBAs in the enclosure, and 36 W dissipated by DSB, managing power distribution between the sides of the PCB results in up to 5.5°C decrease of the maximum temperature of the included components. 23 Chapter 2 Figure 11. Screenshot of software for power distribution control. high-power components on HS side high-power components on fluid side Figure 12. Evaluated DSB symmetrical power configurations with 30 W dissipated on HS side and 6 W dissipated on the fluid side of the PBA. The arrow highlights the most preferable power configuration with regards to maximum temperature. 24 Thermal Management of Avionic Equipment Air flow direction Figure 13. Surface temperature of the DSB fluid side for symmetrical power configuration with 30 W on HS side and 6 W on fluid side. Figure 14. Contour plots of measured case temperatures (left), versus simulated temperatures (right) on the DSB fluid side with the most preferable power configuration with regards to maximum temperature. Having identified the importance of utilizing advanced thermal analysis in order to calculate temperature distribution within electronic equipment, the next step of the work for this thesis aimed to incorporate CFD analysis in physics-based lifetime prediction. This is covered in the next chapter. 25 Chapter 2 26 3 Reliability Prediction The IEEE defines reliability as “The ability of an item to perform a required function under stated conditions for a stated period of time.” [59]. Reliability predictions for avionic equipment are often based on traditional methods, such as presented in Military Handbook (MIL-HBK) 217 [16]. In this type of methods, reliability is determined by calculating a constant failure rate, or its reciprocal the mean time between failures (MTBF), of each component and sub-system comprising the total system. The “bathtub curve” shown in Figure 15 illustrates the failure rate as a function of time. Initially, manufacturing defects and/or defect electronic components may cause high failure rate. These imperfections are identified and eliminated as a result of the burn-in test briefly discussed in Section 2.1; a test that every unit is exposed to prior to shipping out to the customer. Subsequent to its early failures, a robust product will thrive in a period of low, near constant failure rate with intrinsic failures that occur randomly. Finally, mechanical, chemical, or electrical failures due to wear-out start to increase as the product approaches its lifetime. Failure rate Burn in Useful life Wear out Time to wearout failure Constant failure rate Time Figure 15. The “bathtub curve” illustrating the failure rate over time, with high initial failure rate due to imperfections in the electronic components or in the manufacture of the equipment, followed by a period of constant failure rate due to random failures during the useful life of the equipment, and increasing failure rate when approaching the end of life of the equipment. 27 Chapter 3 The models applied for traditional prediction of reliability due to thermal failure mechanisms are based on the Arrhenius law, which states that the failure rate depends exponentially on the steady-state temperature of the analyzed component, such that p E a kT Ae (7), where p is the failure rate, A is a constant, Ea is the activation energy for the analyzed failure mechanism, and k is Boltzmann´s constant. From this, the “10 C rule” has been derived, which implies that a decrease of the steady-state temperature by 10°C increases the life of an electronic component by a factor of two. This might be true for some failure mechanisms at a limited range of temperatures. However, as general statement about the component failure rate, which depends on a number of failure mechanisms and wear-out mechanisms, the 10 C rule does not apply. Some failure mechanisms have a temperature threshold below which the mechanism is not active at all, while others are even suppressed at elevated temperatures [60]. In the temperature range -55°C to +150°C, most of the reported failure mechanisms are not due to high steady-state temperature. They either depend on temperature gradients, temperature cycle magnitude, or rate of change of temperature [38]. Hence, a more elaborated method would be required to increase the confidence in the reliability predictions, which in a better way captures the physics behind each failure mechanism. This can be achieved by utilizing a life prediction model to assess the time to wearout failure as seen in Figure 15. This will be further covered in chapter 3.2 below. One step towards physics-based life prediction was taken in the MIL-HBK 217, when thermal cycling fatigue was included by employing the Coffin-Manson model [61], 1 Nf 1 2 2 c (8), f where Nf is the mean number of temperature cycles to failure, is the cyclic strain range, f = 0.325 is the fatigue ductility coefficient, and c = -0.442 is the fatigue ductility exponent for standard eutectic SnPb solder. The cyclic strain range depends on the difference in coefficient of thermal expansion (CTE) between the component and the PCB, the cyclic temperature extremes, distance from neutral point (commonly the center of the package), and stand-off height. In a multitude of papers (e.g. [58], [62], [63]), variants of the Coffin-Manson thermal fatigue prediction model have been experimentally validated for different types of electronic components, hole-mounted as well as surface-mount attached, different PCB composition, etc. The predominating modification of the CoffinManson model has been made by Engelmaier [64] in that he made the constant c variable such that, for SnPb solders, 28 Reliability Prediction c 0,442 6 10 4 Tsj 1,74 10 2 ln 1 360 td (9) where Tsj is the mean solder joint cycling temperature, and td dwell time (in minutes) at the temperature extremes. However, most real geometries cannot be accurately represented by this model. Many effects like deformation of board and component, multiaxial distribution of strain, the part of the strain that results from local CTE mismatch between the solder alloy and copper pads, and the complicated inelastic character of solder deformation can probably not be included in a similar model. Hence, since this analytical technique does not fully take into account the physics behind the failure mechanisms, correlation to thermal cycling tests is required for each package type under consideration [65]. Furthermore, the non-periodic temperature fluctuations experienced during field usage of a product cannot be captured by this model. Instead, periodic behavior must be extracted from the measured data [66], and unknown distortion of the input data is thus introduced. 3.1 Advanced Thermal Analysis incorporated in Thermal Fatigue Life Prediction In particular when power cycling plays a vital role for the temperature of the equipment, advanced thermal analysis may improve the accuracy of physics-based lifetime prediction. Paper IV presents research collaboration with Boeing Phantom Works, in which a lifetime prediction tool has been extended to allow analysis of electronics in a complex thermal environment. An option has been added to read ‘snapshots’ of the result from a transient CFD analysis and use them as a basis for the thermal fatigue analysis. The tool calculates the fatigue life of a printed circuit board assembly (PBA) in an avionic unit under combined vibration and cyclic thermal and power loads. Individual component lifetime under combined operation environments are evaluated in terms of component cumulative damage indices (CDI). CDI for each individual environment is the ratio of number of cycles required for that environment to the life cycles that the solder joint can sustain under that specific environment. The CDI evaluation does not distinguish the sequence of damage between thermal and vibration, so the total CDI is obtained by summing the CDI due to vibration and the CDI due to thermal cycling (Miner’s rule) [67]. The Engelmaier model is applied to calculate the thermal fatigue life of the solder joints in the assembly. As an example, the process utilized to analyze the lifetime of an avionic PBA is presented. The initial step is a transient thermal analysis of the unit in which the PBA is located. The different operating modes of the unit are modeled and analyzed using a commercially available CFD tool, which generates a time history of the temperature at all points within the unit and PBA, as shown in Figure 16. 29 Chapter 3 Figure 16. Typical transient temperature profile at various points of the avionic unit. The second step comprises exporting temperatures from the transient temperature analysis to the lifetime prediction tool. The temperatures calculated by the global analysis are mapped to the PBA mounted within the box, yielding the temperature distribution of the PBA as functions of time, see Figure 17. The lifetime prediction tool utilizes Equations (8) and (9) together with the transient temperature profile converted to temperature cycles and half-cycles, to assess the lifetime of each lead and solder joint included in the PBA. The thermomechanical fatigue level of leads and solder joints within the unit are reported as a cumulative damage index (CDI). CDI from one of the operating modes defined for the PBA is shown in Figure 18. Figure 17. Temperature of PBA at different points of time. 30 Reliability Prediction Figure 18. CDI from thermomechanical fatigue. Lifetime prediction of solder joint due to vibration is performed separately. The environment can be seen in Figure 19, in which the continuous random vibration spectrum specified for the location of the unit is presented. This vibration environment would be applied to the PBA for 5 hours, according to the specification. Subjected to this vibration spectrum, 16 modes of natural frequencies have been identified causing board deflection to take into account for the fatigue calculations. The 5 first modes can be seen in Figure 20. The CDI due to vibration, shown in Figure 21, is added to form an overall CDI based on Miner’s rule. The total CDI is presented in Figure 22. Figure 19. Vibration environment for PBA. 31 Chapter 3 Figure 20. Natural frequencies of PBA. Figure 21. CDI from vibration. Currently, tools are at hand that incorporate thermal analysis embedded in the lifetime prediction software. However, the thermal analysis models are quite basic, and generally do not include CFD analysis to correctly model fluid flows. In more complex cases, such as presented in Paper IV, it may be convenient to integrate advanced thermal analysis software for lifetime prediction purposes. Hence, the lifetime prediction tool has been modified to use result files from any thermal analysis tool as input data for thermal fatigue prediction. 32 Reliability Prediction Figure 22. Total CDI from vibration and thermomechanical fatigue. There is a drawback with incorporating advanced thermal analysis in lifetime prediction tools, besides the obvious increase of computational load. Operating any progressive software tool requires substantial knowledge about the mechanisms that are simulated. Hence, the operator of such a comprehensive tool should preferably be experienced in a number of engineering disciplines such as mechanical fatigue, lifetime prediction, and thermal modeling. Furthermore, a general word of caution needs to be acknowledged as the computational tools get increasingly powerful: The results are never more accurate than the input data provided to the tool. Considering the case presented above, the environments to which the unit is assumed subject are based on the specified operating environments and loads. Great care needs to be taken to spread the understanding of the difference between assumed and real life environmental loads. 3.2 Physics of Failure in Reliability Prediction of Solder Joints In section 2.1 of the thesis, a notion of the multitude and variation of thermally induced failure mechanisms is conveyed. In order to conceive the impact of these mechanisms on the factual reliability, understanding the physics behind the failure mechanisms is of vital importance. In the mid-1990-s, CALCE Electronic Products and Systems Center defined a new life prediction method utilizing the Physics-ofFailure process [68]. An overview of the Physics-of-Failure process is shown in Figure 23, where the chain can be seen of retrieving detailed information of all geometries, materials, production processes, field environments, etc., to use as input for performing analyses of manufacturability, lifetime, and performance, which result in risk mitigating solutions to enable a more reliable product. 33 Chapter 3 The Physics-of-failure (PoF) Design & Qualification Process Design-Fix-Build-Test methodology Stakeholders Inputs Internet Databases Integrators Suppliers Environments Storage, transport, usage Materials PWB, solders, adhesives, heat sinks Assemblers Parts ASICs, memories, FPGAs, processors, etc. Researchers Regulatory Agencies Product Analysis Outputs Manufacturability analysis “Risk Mitigation Solutions” Laminate formation, solder joint formation, etching, plating Architecture Production Process “Stress” analysis Global Temperature, vibration, shock, moisture, EMI Operational Parameters Life Cycle Profile Local Lead, solder, PTH/vias, trace, laminate Failure analysis Defect identification Failure identification Stress management Design tradeoffs CFF, delamination, fiber debonding, lead fatigue, solder fatigue, via fatigue Soldering, laminating, screen printing, plating, etching, etc. Product and supply chain evaluation Performance analysis Failure models Cross-talk, delay time, failed circuit identification Screening conditions Process Wearout, overstress Tests Sensitivity analysis HALT, HASS, ESS, etc. Evaluate product failure due to variations in environmental and architectural parameters Accelerated test Health management solutions Figure 23. Physics-of-Failure process [69]. With regards to the impact of temperature variations on lifetime, mathematical models have been developed to estimate the number of thermal cycles to failure. However, these always contain simplifications to a varying degree, and since there are a variety of failure mechanisms as previously shown, great care should be taken to understand the background of the models before applying them in reliability prediction. It is crucial to remember a few factors that make life prediction of solder joints a complex task [23]: Solder joints differ from each other (distance from neutral point, height, geometry, metallurgy). Initial microstructure of the solder alloy varies. Substrate finishes and component finishes. Intermetallic compounds (IMC) vary with soldering process. Many possible failure mechanisms: Grain/phase coarsening, grain boundary sliding, matrix creep, micro-void formation and linking, resulting in crack initiation and crack propagation. Non-linear, and temperature dependent mechanical behavior of solder. Strictly speaking, creep processes take place above -40 C, which is approximately half of the homologous temperature, which the is absolute temperature melting point. 34 Reliability Prediction Failure criterion might vary (intermittent electrical opens that are difficult to detect; mechanical or electrical damage). The failure location in a specific solder joint may vary depending on the above variations. The existing life prediction models for specifically thermal fatigue of solder joints can be classified in three groups, listed in the order of increased complexity [70]: Analytical models, Constitutive law + fatigue law models, and Damage mechanics based models. The analytical models, primarily represented by the Engelmaier model [64] and its modifications, are simple to implement, but they include too little detail to utilize for higher accuracy reliability prediction purposes. The Constitutive law + fatigue law models, such as the strain energy partitioning method (SEPM) [27] and the Accumulated strain energy density (SEDcr) approach [18], [19] which will be further discussed below, provide more accurate predictions with less restriction than the analytical models. The damage mechanics based models, including the Disturbed state concept (DSC) [28] and the computationally more efficient Reduced accelerated DSC (RADSC) [71], may provide the highest prediction accuracy. However, the implementation and computational effort is considerably higher for the damage mechanics based models than for the other two classes of models. Secondary creep is often assumed the main damage mechanism for thermal fatigue of solder alloys [18]–[20], [23]. FE calculated volume-averaged SEDcr based on secondary creep, is in a number of publications used to quantify the damage accumulation process in solders [18], [19], [70], [72], [73]. Constitutive laws for creep strain rate as a function of stress and temperature include double-power law and hyperbolic-sine law [23]. The power law can be expressed as follows: d s dt AII T n II exp Qa , II kT AIII T n III exp Qa, III kT , (10) and the hyperbolic-sine law: d s dt where n C III , IV sinh exp d s is the steady state creep strain rate, dt Qa , kT (11) is the applied stress, k is the Boltzmann’s constant, T is the absolute temperature, Q is the activation energy, n denotes stress exponents, A and C are constants, and prescribes the stress level above which the sinh-dependency dominates the calculated creep strain rate. Both constitutive laws capture the change in creep mechanism that takes place at certain stress levels denoted with the roman numbered subscript II through IV. Implementing either of these models in FE calculations, the accumulated creep strain can be estimated, and utilized in a fatigue law. Currently, SEDcr is widely accepted as damage metric for life prediction. SEDcr is calculated by multiplying the creep strain by the applied stress. 35 Chapter 3 In the recent standard for physics of failure in reliability prediction ANSI/VITA 51.2 [17], a model is advised that belongs to the constitutive law + fatigue law class of models. It utilizes SEDcr as damage metric [18], and a very simple fatigue law to evaluate the lifetime. This method is recommended for life prediction, even though it has not been validated for field operation temperature levels. Furthermore, since the standard suggests expressing the anticipated field temperature in terms of cycles, the temperature data will be simplified with regards to ramp rates and dwell times. Even though this standard represents a large step towards physics-based reliability predictions, it is apparent that much work remains until thermal fatigue life prediction of solder joints can be performed with full confidence in the results. One way to approach completion of the process is to perform various kinds of reliability tests and publish as much details from the tests as possible, as has been done in Paper VI, and further described in section 3.3 below. Ideally, a life prediction model would be available that was validated for noncyclic temperature variations. Indeed, even without fully validated life prediction models, this kind of calculation might bring higher credibility to the prediction results, and supply increased understanding of the impact of different temperature profiles. However, comprehensive FE modeling of stress and strain in a solder joint in noncyclic temperature environment can require large computational resources and long calculation time. Therefore, a computational method has been proposed that quickly evaluates the damage accumulated in a solder joint exposed to a non-cyclic, nonsimplified field operation temperature profile with accuracy comparable to FE modeling. 3.3 Accelerated Thermal Cycling Tests of Surface Mount Packages with SAC, SN100C, and SnPb Solder Paste Accelerated thermal cycling tests have been performed with the twofold aim to contribute to development of models for evaluation of damage accumulation, and to provide further experimental data to assist in qualification of certain package types and lead-free solder alloys. Paper VI describes the tests, which have been performed on a range of surface-mount electronic component packages that could be considered for avionic applications, mounted with SAC305, SN100C, and Sn62Pb36Ag2 solder paste. The composition of each tested solder alloy is provided, as well as detailed manufacturing parameters and geometries of each electronic package. From the experimental results and FEM assisted interpretation thereof, engineering guidelines have been generated regarding selection of packages and solder alloys for avionic applications. One of the 48 PBAs subjected to the tests is shown in Figure 24, showing the ten different component types that have been included in the tests. 36 Reliability Prediction QFN48 0402RES BGA256 BGA49 QFN72 LGA133 1206RES BGA1152 TSOP48 QFP304 Figure 24. Printed circuit board assembly subjected to thermal cycling tests. The accelerated thermal cycling tests comprise nearly 12000 cycles in -20°C to +80°C environment, and more than 6000 cycles of -55°C to +125°C temperature variation. The results confirm the feasibility of the lead-free solder alloys in -20°C to +80°C thermal cycle, as well as the benefit of softer SnPb-alloys when subjected to -55°C to +125°C temperature cycling. The effect of die size on thermal cycling reliability of a full-array BGA component has been quantified with the help of Weibull graphs and FE analysis. The accelerated test results have been related to field environments found in avionic applications. In agreement with other published results [11], [13], [23], SAC305 PBGA256 has been selected to be a viable candidate both for controlled and uncontrolled thermal environments. 3.4 Surrogate Modeling of Damage Accumulation in Solder Joints When the dependency of one or more response variables on one or more design variables is not possible to express analytically, and experiments or FE simulation is considered too time consuming for practical utilization, measures such as model reduction methods or surrogate modeling can be applied. Model reduction methods that take into account nonlinear response have been demonstrated, for example on advanced calculations of crack propagation in elastic-plastic media [74]. While these methods assume computationally efficient reduced FE modeling in the calculations, surrogate modeling takes one step further in that more simple methods are utilized for the response evaluation. Surrogate modeling has two primary applications: design optimization, and design space approximation. One way to create a surrogate model 37 Chapter 3 for design space approximation is to design a response surface by interpolating the values of the response variables between the results of a limited number of experiments with design variables set to cover the design space. Different interpolation schemes can be applied to estimate the response between the data points included in the experiments [75]. The novel computational method presented in Paper V, and further elaborated in Paper VI, provides relatively fast evaluation of accumulated damage in a critical solder joint of an electronic package on a PCB, with accuracy comparable to computationally demanding FE simulations. Linearly interpolated FE-simulation-based response surfaces are utilized to create a surrogate model for design space approximation. The method includes the following: a simulation-based design-of-experiment (DoE) procedure that utilizes thermal load limits for the solder joint, an analytic method to improve the data resolution, construction of linearly interpolated surfaces from the FE model response, and computation of accumulated damage in a solder joint for field temperature conditions. Due to unavoidable singularities in stress values delivered by numerical solution at the interface between the copper pad and solder [76], [77], a published procedure has been utilized to estimate accumulated creep strain energy density (SEDcr) in a ball grid array (BGA) solder joint [18]. Briefly, SEDcr is volume-averaged in a region comprising a 25 m thick layer of solder at package or board interface, depending on where the largest damage accumulation is expected, with at least two finite elements across the thickness. The creep strain energy density is computed for each time interval tn , tn 1 , n 0,1,... of a discretized operating temperature profile, through multiplying the calculated creep strain by the average of the equivalent stresses at the beginning and at the end of each time interval. The computation is performed for each finite element in the region of the solder joint where the damage accumulation is largest. In order to calculate both the creep strain accumulated during the time interval tn , tn 1 , and the equivalent stress at the end of the same time interval, the following input data is required: the temperature at the beginning and at the end of the time interval, and the equivalent stress at the beginning of the time interval. The calculation procedure is described with the three graphs provided in Figure 25. The graph seen in Figure 25 (a) highlights the operating temperature Tn , the neutral temperature Tsf of the solder joint, von Mises or equivalent stress n , and accumulated creep strain n , for time intervals tn 1, tn and tn , tn 1 . These time intervals are translated to FE time intervals 0,2t FE , t FE tn 1 tn , shown in Figure 25 (b) and Figure 25 (c), respectively. Temperatures T1 and T2 in Figure 25 (b) correspond to Tn 1 and Tn in Figure 25 (a), whereas T1 and T2 in Figure 25 (c) reflect Tn and Tn 1 , respectively. The neutral temperature T0 is defined such that the slope between T0 and T1 creates the equivalent stress 1 at time tFE . The equivalent stress in Figure 25 (b), represent n and n in 2 , and the accumulated creep strain Figure 25 (a), while 2 and seen in Figure 25 (c), correspond to n 1 and n 1 , respectively. 38 Reliability Prediction T Tn+1 n n-1 Tn Tsf n+1 n+1 Tn-1 n tn-1 tn (a) T2 1 0 t T T T2 T0 T1 tn+1 2 tFE 2tFE 1 T1 T0 2 0 t (b) tFE 2tFE (c) t Figure 25. Transition from operating temperature (a) to FE time interval; transition from time interval t n 1 , t n (b), and from time interval t n , t n 1 (c). Depending on whether the temperature in the beginning of each time interval is below or above Tsf (see Figure 25 (a)), the electronic package either contracts or expands relative to the PCB. As a consequence, the equivalent stress in the end of each time interval may differ by orders of magnitude, even at similar initial equivalent stress magnitudes at the beginning of the time intervals, and similar temperature slopes from Tn 1 to Tn , and Tn to Tn 1 , respectively. Translated to the FE time interval as shown in Figure 25 (b) and Figure 25 (c), the equivalent stress 2 , resulting from the initial equivalent stress 1 followed by similar temperature slopes from T1 to T2 , would significantly differ for the two situations. Assumptions and simplifications accompanying the proposed method are provided in detail in Paper V. The method is realized in two steps: data preparation, in which the surrogate model is created, and, computation of accumulated damage caused by thermal loads by utilizing the surrogate model. 39 Chapter 3 3.4.1 Data Preparation The data preparation comprises the following steps: 1) Identify the solder joint that accumulates the largest thermomechanical damage by FE modeling of the electronic component of interest. 2) Assess thermal load limits, and create a simulation plan that adequately covers the identified range of thermal loads. Perform a sequence of short FE simulation runs according to the plan, and store the model response in data files. 3) Create response surfaces by linear interpolation between the data points. 4) Increase the data resolution to improve the quality of the response surfaces. From a practical viewpoint, the data preparation phase takes some time. However, most of the described processes, such as creating the simulation plan based on thermal load limits, performing the FE simulation runs, creating response surfaces, and increasing the data resolution, can be automated via appropriate scripting that enables co-operation between e.g. Matlab and the FE solver. Such automation is partly implemented for the results reported in this thesis. Specifically, Python scripting is employed to schedule the simulation runs in Abaqus 6.9, and programming in Matlab R2011b has been done to generate interpolants and increase the data resolution. The thermal load limits are defined by the extreme operating temperatures, Tmax and Tmin , the maximum temperature change T max for the chosen time step, and the maximum difference T01max between the operating temperature Tn , and the concurrent neutral temperature T0 n expected during the lifetime of the solder joint. The purpose of the simulation plan is to generate a sufficient number of data points to accurately compute the damage metric. The results of the simulations are stored in one data file for each finite element within the region of the solder joint where damage accumulation is largest. In every simulation run, the FE model is computed with a three-temperature profile (T0 , T1, T2 ) , as can be seen in Figure 25 (b) and Figure 25 (c), returning the associated output quantities 1 , 2 and , which are unique for every finite element within the evaluated region. A simulation flow for data generation is shown in Figure 26. The resulting data points are represented by ordered sets that correspond to N simulation runs: (T0 j , T1 j , T2 j , 1 j , 2 j , j ) , j 1,2,..., N . The thermal load limits Tmax , Tmin , T01max , T max should be utilized such that the interpolated response surfaces would be able to handle all possible operating temperature profiles that the solder joint may experience during its lifetime. Every difference between T1 j and T0 j will correspond to the difference between the current neutral temperature of the solder joint and the operating temperature, and every difference between T1 j and T2 j will correspond to the temperature increment in the operating temperature profile during each time interval tn , tn 1 , n 0,1,... . 40 Reliability Prediction Thermal load limits: T min , T max , T 01 max , T max Selection of three temperatures (T0 j ) N (T1 j ) N 1 T0 j [T1j T1 j T01max,T1j 1 (T2 j ) N 1 T01max] [ T min , T max ] T2 j [T1 j T max,T1 j j T max] 1,2,..., N Automated finite element simulation runs Collection of data points: (T0 j , T1 j , T2 j , 1 j , 2 j , j ) j 1,2,...,N Figure 26. Simulation flow for data generation. 3.4.1.1 Revised Data Structure The symmetric, “spider-like” structure of the data as depicted in Figure 26 has provided a reliable base for designing the response surfaces utilized in the original version of the method. Although redundant data points has been created due to the resulting, non-physical extension of neutral temperature levels, this has been considered feasible for the data required for a limited span of operating temperatures. However, in order to enable evaluation of thermal fatigue for an extended temperature range, the data structure had to be revised. Rather than a symmetrical structure of T01max linked to T1 j , the neutral temperature now originates from a defined range of T0 n temperatures, as shown schematically in Figure 27. The novel data structure, introduced in Paper VI, gives threefold reduction in the number of initial simulation runs, compared to the previously reported data structure. 41 Chapter 3 T0 T1 T2 Figure 27. Schematic representation of new data structure for three-temperature data points. 3.4.2 Algorithm for Estimation of Accumulated Damage The proposed method calculates the volume-averaged accumulated SEDcr within the evaluated region of the solder joint, during arbitrary thermal loading within the above estimated limits. A flowchart of the computational algorithm is presented in Figure 28. The computational process begins with setting parameters n 0, 0, m 0, and Wm 0 , m 1,2,...M , to assume stress-free conditions and no accumulated creep strain energy density in the identified solder joint at time t0 = 0. Index n represents a discrete time point tn where accumulated damage in the solder joint is evaluated, n = 0,1,2,… . For each time interval tn , tn 1 , n 0,1,2,... the three-temperature points (T0 n, m , Tn , Tn 1) are formed. For n > 0, the current element-wise stress-free int int temperature T0 n, m is set by evaluating interpolants T pos or Tneg for each element m from the evaluated region, at the query location (Tn , n ,m ) . Which interpolant to evaluate depends on whether the element-wise stress-free temperature T0 n 1, m is higher or lower than the temperature Tn at the end of the previous time interval. Subsequently, the von Mises stress n 1, m and the accumulated creep strain are retrieved by evaluating interpolants S int and E int , respectively, at the query location (T0 n, m , Tn , Tn 1) . Notice, while temperatures Tn and Tn 1 are shared by all M n 1, m finite elements, the current stress-free temperature T0 n, m is unique for each element m, m 1,2,...M . The SEDcr accumulated during the time interval tn , t n calculated, such that Wn 1,m ( n 1,m 2 n,m ) 1 is then (12) n 1, m . 42 Reliability Prediction The total element-wise SEDcr Wm accumulated by the time tn 1 is summated, and subsequently applied to calculate the volume-averaged total SEDcr WVA , accumulated by the time t n 1 , WVA Wm Vm m 1,2,..., M Vm m 1,2,..., M , (13) where Vm is the volume of finite element m. Initialize algorithm with n 0, 0, m 0, Wm 0, m = 1,2,…,M. Form three-temperature point interpolant evaluation, int (Tn , n, m ), if T pos T0 n, m T0 n,m , Tn , Tn 1 , by Tn T0 n 1, m int (Tn , n, m ), if Tn Tneg T0 n 1, m Tn , if n &n 0 &n 0 0 m = 1,2,…,M. Retrieve equivalent creep strain and equivalent stress by interpolant evaluation, and calculate accumulated SEDcr for time interval t n , t n 1 , n 1, m E int T0 n, m , Tn , Tn 1 , n 1, m S int T0 n , m , Tn , Tn 1 , Wn 1, m m n 1, m n, m 2 n 1, m , 1,2,..., M . Calculate total accumulated SEDcr in each element, Wm Wl , m l 1, 2,..., n 1 , m = 1,2,…,M. Calculate volume-averaged accumulated SEDcr, and proceed to next time step, WVA n Wm m 1, 2,..., M Vm Vm m 1,2,..., M , n 1 Figure 28. Computational algorithm for calculation of volume-averaged accumulated SEDcr. 43 Chapter 3 As can be seen in Equations (10) and (11), the creep strain rate depends nonlinearly on the absolute temperature and the equivalent stress. Therefore, linear interpolation between the data points might not result in acceptable accuracy, unless the data is adequately resolved. Thus, in order to generate adequate data resolution without requiring an impractical number of simulation runs, an analytical method to increase the number of data points has been developed. Within the estimated thermal load limits reported in this work, the equivalent stress in the solder joint has been found near-linearly dependent on temperature, due to linearly elastic properties of all materials in the FE model except for solder. A new data point (T0new , T1new , T2new , 1new , 2new , new ) can therefore be enabled, where 1new and new 2 are generated by linear interpolation between two numerically calculated data points, where only one of the three temperature parameters differs. Subsequently, Equation (10) or (11) can be applied to deliver new . Details on the procedure for improvement of data resolution are provided in Paper V. In order to identify the solder joint that accumulates the largest damage, FE simulation of a relatively mild accelerated test with a lead-free PBGA256 component attached to an FR-4 PCB as seen in Figure 29 has been performed. The results have been compared with the failure location observed in diagonally cross-sectioned components that have been exposed to a thermal cycling test (see Figure 30). Failure analysis has furthermore served to reveal the failure mode, thereby increasing the confidence in the assumptions and simplifications made in the FE model. A coarsely meshed quarter-symmetry 3-D FE model with equal mesh on all solder joints identifies the solder joint with the largest damage accumulation correctly. The highest concentration of accumulated creep strain energy density is located in the solder at package interface, in agreement with the experimental results. Next, a detailed octant-symmetry 3-D FE model is created, with local mesh refinement on the solder joint that accumulates the largest SEDcr (see Figure 31). Figure 29. PBGA256 component under thermal cycling test. 44 Reliability Prediction Figure 30. Optical microscopy image of crack in solder joint coinciding with die corner; arrow indicates crack location near the solder joint-to-component bond pad interface. Figure 31. Octant symmetry FE model, including mesh and composition; zoom in on detailed mesh on critical solder joint, with 25 m thick evaluated region; SEDcr distribution in lower part of the figure confirms failure location seen from experiments. 45 Chapter 3 A number of problems have been encountered and mitigated during the development of the suggested method. More pronounced singularity-behavior has been observed in the solder at the package interface, which has led to non-physical stress relaxation at high temperature dwell, and troubles in obtaining stable values from the surrogate model of T0 n, m for some of the finite elements. Furthermore, insufficient precision of the FE-calculated volume-averaged stress data obtained from the simulation runs in the data preparation phase, has led to miscalculation of the stress in vicinity of the neutral temperature. Details on the measures taken to mitigate these phenomena are provided in Paper V and Paper VI. The final result is that SEDcr is computed more than two orders of magnitude faster than FE simulations with the same operating temperature profiles. Comparisons between SEDcr calculated with the proposed method and FE simulation for an example of field temperature profile is shown in Figure 32. Figure 33 shows the same for the -20°C to +80°C thermal cycling test, as well as the difference in accumulated SEDcr between FE and the proposed method. It can be seen, that the agreement between the results predicted with the suggested computational method and the FE simulations, converges to approximately 4%. Operating Temperature, C 80 70 60 50 40 30 20 0 2000 4000 6000 8000 10000 12000 14000 10000 12000 14000 Time, sec Acc. energy density, J/m3 300 250 Surr. model FE 200 150 100 50 0 0 2000 4000 6000 8000 Time, sec Figure 32. Example of field temperature profile (top), and accumulated creep strain energy density: FE simulation vs. surrogate modeling method (bottom). 46 Reliability Prediction 100 Operating temperature, C 80 60 40 20 0 -20 -40 0 1 2 3 4 5 6 7 Time, sec 6 x 10 4 3 Acc. energy density, J/m 4 3T FE Surr. model FE 5 8 x 10 4 3 2 1 0 0 1 2 3 4 5 6 7 Time, sec 8 x 10 4 10 Diff. in acc. energy density, % 8 6 4 2 0 -2 -4 -6 -8 -10 0 1 2 3 4 Time, sec 5 6 7 x 10 4 Figure 33. TC temperature (top), predicted SEDcr to compare the surrogate modeling method for evaluation of damage accumulation with FE simulation (middle), and difference between FE and the surrogate modeling method in accumulated SEDcr (bottom). The method is believed to be general in terms of application to different electronic packages, as far as validated life prediction models are available. The requirement to the FE model is that it is capable to identify the solder joint that accumulates the largest thermomechanical damage, and predict the right failure location. Therefore, the quality of the results delivered by the proposed computational 47 Chapter 3 method depends on the quality of the FE model. The added value with the proposed method is that various non-cyclic field temperature profiles, taken in any sequence, can be quickly evaluated in the design phase, provided that appropriate data preparation has been performed according to described routines. Besides the revised data structure, further improvements of the method are presented in Paper VI. In order to resolve singularity-related problems, it has been found necessary to implement T0-limit control, and stress relaxation at hightemperature dwell to maintain adequate accuracy for increased temperature range. Due to its ability for time-efficient operation on uncompressed temperature data, the developed method might contribute to the accuracy of reliability prediction of electronic packages. This especially relates to applications where thermal fatigue is a dominant contributor to the damage of solder joints and where time dependent analysis, such as creep analysis, is involved. 48 4 Prognostics of Remaining Useful Life of Electronic Equipment Historically, diagnostics and prognostics of remaining useful life (RUL) have often been implemented for critical structural components of a system. Electronic equipment has been a minor part of the system, and the lifetime of the electronics has been judged much longer than the most critical mechanical components. Currently, however, the function of such equipment relies more and more on electronics, embedded in various locations of the system. Simultaneously, essentially depending on the use of commercial off the shelf (COTS) electronics in more rugged applications than the original intention, the lifetime of the electronics is decreasing. Consequently, prognosticating the remaining useful life of electronic equipment is a research area that currently attracts attention. Accuracy of life prediction tools has become critically important, due to increased reliability requirements on absolute numbers, instead of traditional relative comparison [23]. As an outcome of this, the interest increases of, for instance, aircraft manufacturers, to continuously monitor RUL of line replaceable units for avionic use, thus achieving significant potential to provide safer, more reliable, and cost-effective avionic systems [78]. Much research is currently ongoing to define methods and models to correctly predict RUL of electronics, with all levels of failure modes taken into consideration. As an outcome of the above attempts, four wide-ranging categories of implementation of prognostics can currently be found in literature: A. Monitor failure precursors: Performance parameters are monitored, such as voltage levels, and switching time constants, that may indicate impending failures [79]; B. Introduce expendable devices: Implement fuses and “canaries” into the system; a prognostic cell that fails earlier than the system, and thus provides advance warning of a failure [80]; C. Microstructural analysis: Removal and cross-sectioning of expendable, dummy component solder joints. The grain size and the rate of change of intermetallic growth may reveal the state of the solder joint [81]. D. Life consumption monitoring (LCM): use physics-of-failure models, applied on measured environmental and thermal loads, to compute RUL of the equipment [24], [82]. Each of the above categories has its individual advantages and limitations, further discussed in Paper VII. However, the concept of merging two or more of these 49 Chapter 4 approaches, achieving fusion prognostics [83], is currently of high interest in the research community [84], [85]. Similar reduction of field operation temperature data as described in the context of reliability prediction has been suggested for real-time fatigue calculation of solder joints [86]. The benefit gained by instead implementing the computational method previously presented in this thesis, is the potential of increased accuracy of the prediction of remaining useful life of the monitored system. It has to be noted, however, that this would require validated combinations of constitutive laws for calculation of the damage metric, and life prediction models in which to apply the damage metric. A prognostic unit embedded in the host system needs to be miniaturized, reliable, robust, and independent of external power supply over long periods of time. Furthermore, it should be “invisible” for the host system, not to adversely affect the traditional mean time between failures (MTBF) calculations, which supposedly accompany the initial quotation documents. The Physics-of-Failure process has been adapted for utilization in prognostics applications according to the flowchart given in Figure 34 [87]. After identification of which parameters to monitor during usage of the product, by utilizing fault tree analysis and failure mode, mechanisms, effects, and criticality analysis (FMMECA) followed by virtual reliability assessment, applicable sensors are introduced in the system. The acquired data is simplified to minimize memory space needed for storage, and make the data compatible with the requirements of each PoF model. Step 1: Conduct failure modes, mechanisms and effects analysis Step 2: Conduct a virtual reliability assessment to assess the failure mechanisms with earliest time to failure Step 3: Monitor appropriate product parameters environmental (e.g. shock, vibration, temperature, humidity) operational (e.g. voltage, power, heat dissipation) Step 4: Conduct data simplification to make sensor data suitable for stress and damage models Step 5: Perform stress and damage accumulation analysis Step 6: Estimate the remaining life of the product Is the remaining-life acceptable? Continue monitoring Schedule a maintenance action Figure 34. Physics of Failure process applied for life consumption monitoring (LCM) [87]. 50 Prognostics of Remaining Useful Life of Electronic Equipment The novel comprehensive method for prognostics of thermal fatigue failure of solder joints suggested in Paper VII, is summarized in Figure 35. Steps 1 through 3 are performed in advance, while steps 4 and 5 are executed in real time. From a global CFD simulation in the first step, and the following global reliability prediction of the PBAs included in the system in step 2, the first component to fail (FCF) is identified. A temperature sensor is introduced near FCF in the third step. Using data from the CFD simulation, and information about the state of the host unit (on/off), the correct temperature offset between the temperature measured at the sensor location, and the average FCF temperature, is calculated. This gives the true time history of the temperature of FCF. As opposed to the existing Physics of Failure-method shown in Figure 34, the suggested method for evaluation of damage accumulation utilizes unrefined temperature data as input to the real-time, in situ RUL calculations performed in the fourth step. In the fifth step, data reduction, by means of ordered overall range (OOR) method and rainflow algorithm [24], is performed to save the general temperature time history for future analysis, development of products for similar environments, and design of accelerated tests. Also, as can be seen in Figure 35, the logged temperature profile is used to verify the initial assumptions of temperature time history. The accuracy of the prognosticated remaining useful life depends on the level of details captured in the model, and the level of confidence from validation efforts. Action Output 1. Global transient computational fluid dynamics (CFD) analysis Expected time history of temperature 2. Global durability analysis First component to fail (FCF) 3. Introduce temperature sensor at FCF True time history of FCF temperature 4. Real-time local RUL calculations for FCF Remaining thermal fatigue life 5. Data reduction and storage FCF temperature history logged for future use Figure 35. Suggested method for thermal fatigue prognostics of solder joints. 51 Chapter 4 Since the prognostic system is transparent to the host system, the initial reliability calculations for the host system will remain unaffected. The proposed computational method also enables use of a low-end computer to perform the remaining useful life (RUL) calculations. Thereby power consumption is minimized, and reliability is maximized. In Paper VIII, the surrogate modeling approach for estimation of accumulated damage in solder joints has been taken one step further towards future realization in life consumption monitoring applications. Implementation routines are discussed, and effects on the prediction accuracy of data resolution and other factors have been investigated. An avionic application scenario presented for a PBGA256 component serves to highlight implementation routines, which have been found realizable under the following assumptions: Initially, all solder joints are considered stress-free. Being a frequent assumption for thermal fatigue of lead-free solders, secondary creep is the chosen damage mechanism [20], [23], [88]. Other damage mechanisms for thermal fatigue, such as time-independent plasticity and primary creep, are not addressed. The hyperbolic-sine law is employed as constitutive model for creep rate calculation [23]. Accumulated creep strain energy density (SEDcr) is used as damage metric [19], [70]. A number of lifetime prediction models show good correlation to thermal cycling experiments with this damage metric [88], [89]. Although with certain limitations, SEDcr may furthermore quantify the damage accumulation not only for termal cycling tests, but also for operating temperature conditions [19], [90]. Computed SEDcr is volume-averaged in a 25 m thick interfacial layer of the solder joint, at the solder-to-component interface [76]. Linear interpolation between data points obtained by initial simulation runs is applied to create the response surfaces. Accuracy issues are mitigated by employing an analytic method to improve the data resolution. The accuracy of prediction with the proposed method is evaluated by relating the computed results to FE simulated results. Uncertainty investigation of the deformation constants in the constitutive law is out of scope of this thesis, as is robustness to production induced variations of for example solder joint geometry and solder microstructure. The surrogate modeling approach will provide added value to life consumption monitoring by increasing the accuracy of, specifically, thermal fatigue life consumption of solder joints. A data resolution has been determined that forms the base for response surfaces that provide prediction accuracy comparable to 3-D FE simulations. Even though the suggested approach has to be carefully validated, the vision is that it can be combined with other LCM methods and deliver improved prediction accuracy of remaining thermal fatigue life. 52 5 Concluding Summary and Contribution to the Field The main contributions in this thesis to development of avionic equipment are threefold: Design guidelines for power distribution on a double-sided PBA, a novel, computationally efficient method for evaluation of damage accumulation in solder joints in harsh, non-cyclic field operation temperature environments, and the results from thermal cycling tests of electronic packages intended for avionic applications, assembled with SAC305, Sn100C, and SnPbAg solder alloys. 5.1 Thermal Management It has been concluded that the thermal performance of avionic equipment can be improved by utilizing power distribution between the sides of a double-sided PBA. Paying attention to this opportunity in an early stage of product design can render costand weight-free thermal enhancement of the system. The reported CFD-based methodology can be applied as part of concurrent development and contribute to the trend in development of design tools that enable co-design of electrical and thermal layout of printed circuit boards. Quantitatively, in an environment of 55°C cooling air supplied at a rate of 6.7 g/s to the outside of a sealed avionic enclosure, with 39 W of power dissipated by other PBAs in the enclosure, and 36 W dissipated by a double-sided PBA attached to the enclosure side wall, managing power distribution between the sides of the doublesided PBA results in up to 5.5°C decrease of the maximum case temperature of the included components. Bridging the gap between thermal management and reliability prediction, the benefit of incorporating advanced thermal analysis in physics-based lifetime prediction has been highlighted and elaborated. Results have been presented from a collaborative research project, in which a physics-based lifetime prediction tool has been extended to allow for analysis of electronics in a complex thermal environment. Signifying one step towards physics of failure in reliability prediction, this methodology is expected to be introduced shortly in the development of avionic products. 53 Chapter 5 5.2 Physics of Failure in Reliability Prediction of Solder Joints Representing the key contribution of this thesis, a novel computational method has been presented that utilizes surrogate stress and strain modeling of a solder joint to enable quick evaluation of thermal fatigue damage, such that: - It utilizes non-cyclic, non-compressed temperature data of anticipated field operation temperature environment. On the tested various temperature profiles, it provides accuracy within 4% compared to 3-D FE analysis. Its computational efficiency is two orders of magnitude higher than FE analysis. The added value with the proposed method is that various operating (non-cyclic) temperature profiles, taken in any sequence, can be quickly evaluated in the design phase. Due to its ability for time-efficient operation on uncompressed temperature data, the method gives contribution to the practicable accuracy, and hence the credibility, of reliability prediction of electronic packages. Effects of data resolution, which forms the base for response surfaces utilized in the surrogate model, on the prediction accuracy have been investigated, as well as other factors. A data resolution has been determined that provides prediction accuracy comparable to 3-D FE simulations on the tested temperature profiles. The method has been tested on a fullarray BGA256 package subjected to -20°C to +80°C thermal cycles, as well as to representative avionics field temperature profiles. It is however expected to be general in terms of application to different electronic packages, assuming that validated life prediction models are available for each solder alloy and relevant geometry of the critical solder joint that has to be identified in advance. The potential future application of the computational method as a means for embedding real-time prognostics of remaining useful life in avionic equipment has also been elaborated in the thesis. A review of prognostics and health management methods for electronics has been provided, and a procedure to enable increased accuracy of in situ, real-time prognostics has been suggested. A proposed realization in avionics has been discussed, including implementation routines. Even though the suggested approach has to be carefully validated, the vision is that it can be combined with other life consumption monitoring methods and deliver improved prediction accuracy of remaining useful life of electronics. Experimental data has been provided from accelerated thermal cycling tests for qualification of lead-free electronic packages and solder alloys for use in avionics. The tests were performed in -20°C to +80°C and -55°C to +125°C environments. The results confirm similar tests in that the lead-free solder alloys perform better in -20°C to +80°C thermal cycle, while the benefit of softer SnPb-alloys is seen at higher levels of strain in the solder joints. Employing Weibull analysis combined with FE analysis, the effect of die size on thermal cycling reliability of a full-array BGA component has been quantified. The accelerated test results have been related to field environments found in avionic applications. In the context of reliability prediction, especially assuming that the reliability and design engineering disciplines collaborate in the early stages of product development, 54 Concluding Summary and Contribution to the Field the computational method may have its strongest impact. Most clearly pronounced for applications where thermal fatigue is a dominant contributor to the damage of solder joints, utilization of the method may assist to bring product development one step closer to “First time right” design. As a consequence, this would imply a reduction of the number of expensive iterations in product design. 5.3 Future Work Full understanding and complete modeling capabilities of the physics of failure of electronic equipment is a great challenge that will require a lot of research. Small steps are continuously taken to increase the knowledge in this area that is of increasing importance to the ADHP industry. The experimental data provided in this thesis, supplemented by further in-depth failure analysis of the test vehicles, will provide one such step in that it will contribute to increased understanding of crack propagation in solder joints in different thermal environments. Concerning reliability prediction, the suggested computational method should be verified against FE modeling of multiple solder joint geometries. If, in the future, necessary understanding of the physics of failure would be established, a new design space would open concerning reliability and maintenance of electronic systems. In avionic units it could be commercially interesting to include sensors that monitor the environmental loads, and indicate when service or exchange of the unit is due. The surrogate modeling method presented in this thesis is expected to contribute to the accuracy of LCM of avionic equipment. However, since thermal fatigue of solder joints is but one out of a variety of failure mechanisms that may define the lifetime of avionic equipment, the suggested method has to be combined with other LCM methods. 55 Chapter 5 56 References 1. The Columbia Electronic Encyclopedia, Sixth Edition, [Online], Columbia University Press, 2005. 2. Encyclopaedia Britannica, [Online] Available: http://www.britannica.com.bibl.proxy.hj.se/EBchecked/topic/602718/transistor, 2013. 3. G.Q. Zhang, W.D. van Driel, X.J. Fan, Mechanics of Microelectronics. Doordrecht, the Netherlands: Springer, 2006. 4. M. Ito, K. Kobayashi, Y. Miyato, K. Matsushige, H. Yamada, "Local potential profiling of operating carbon nanotube transistor using frequency-modulation high-frequency electrostatic force microscopy," Applied Physics Letters, vol.102, no.1, pp.013115-013115-5, Jan 2013. 5. V. P. Georgiev, E. A. Towie, A. Asenov, "Impact of Precisely Positioned Dopants on the Performance of an Ultimate Silicon Nanowire Transistor: A Full Three-Dimensional NEGF Simulation Study," IEEE Transactions on Electron Devices, accepted for inclusion. 6. Directive 2002/95/EC on the restriction of the use of certain hazardous substances in electrical and electronic equipment, 2002. 7. M. Pecht, D. Das, R. Biagini, “Using Electronic Parts Outside the Manufacturer´s Specified Temperature Range,” in Proceedings of The 3rd International Conference on Quality and Reliability, Melbourne, Australia, 2002. 8. IHS iSuppli Research, http://www.isuppli.com/Semiconductor-Value-Chain/ MarketWatch/Pages/OEM-Semiconductor-Spending-Grew-to-$240-Billion-in2011.aspx, 2012. 9. T.X. Wu, J. Zumberge, and M. Wolff, “On regenerative power management in more electric aircraft (MEA) power system,” in Proc. NAECON 2011 - IEEE National Aerospace and Electronics Conference, 2011, pp. 211-14. 10. D. Hillman and R. Wilcoxon, “JCAA/JG-PP No-Lead Solder Project: -55ºC to +125ºC Thermal Cycle Testing Final Report,” Rockwell Collins Advanced Manufacturing Technology Group, 2006. 57 References 11. T. Woodrow,“JCAA/JG-PP LEAD-FREE SOLDER PROJECT: -20°C to +80°C THERMAL CYCLE TEST,” in Proceedings of SMTA International Conference, Rosemont, IL, September 24-28, 2006. 12. “NASA-DoD Lead-Free Electronics Project Joint Test Report - Final," National Aeronautics and Space Administration (NASA): Technology Evaluation for Environmental Risk Mitigation Principal Center, 2011. 13. J.C. Suhling, H.S. Gale, R.W. Johnson, M.N. Islam, T. Shete, P. Lall, M.J. Bozack, J.L. Evans, S. Ping, T. Gupta, and J.R. Thompson, “Thermal cycling reliability of lead free solders for automotive applications” in Proc. Thermal and Thermomechanical Phenomena in Electronic Systems, ITHERM '04, Vol.2, 2004, pp. 350 – 57. 14. H. Ma, M.M. Ahmad, K.C. Liu, “Reliability of Lead-Free Solder Joints Under a Wide Range of Thermal Cycling Conditions,” IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 1, iss. 12, pp. 1965-74, Dec. 2011. 15. L.E. Bechtold, “Industry Consensus Approach to Physics of Failure in Reliability Prediction”, in Proceedings of 2010 Annual Reliability and Maintainability Symposium (RAMS 2010), San Jose, CA, 2010. 16. Military Handbook (MIL-HDBK)-217 F Notice 2, “Reliability Prediction of Electronic Equipment” February 28, 1995. 17. ANSI/VITA Standard 51.2: Physics of Failure Reliability Prediction, 2011. 18. A. Syed, “Accumulated creep strain and energy density based thermal fatigue life prediction models for SnAgCu solder joints”, in Proceedings of the 54th IEEE Electronic Components and Technology Conference, ECTC, Las Vegas, NV, 2004, pp.737-46. 19. A. Syed, ”Updated life prediction models for solder joints with removal of modeling assumptions and effect of constitutive equations”, in Proceedings of Thermal, Mechanical and Multi-Physics Simulation and Experiments in MicroElectronics and Micro-Systems, EuroSimE, Como, Italy, 2006. 20. W. Dauksher and J. Lau, “A finite-element-based solder-joint fatigue-life prediction methodology for Sn-Ag-Cu ball-grid-array packages”, IEEE Transactions on Device and Materials Reliability, vol. 9, iss. 2, pp. 231-36, 2009. 21. M. Musallam, C.M. Johnson, C.Y. Yin, C. Bailey, and M. Mermet-Guyennet, “Real-Time Life Consumption Power Modules Prognosis Using On-line Rainflow Algorithm in Metro Applications”, in Proceedings of the 2010 IEEE Energy Conversion Congress and Exposition (ECCE), Atlanta, GA, 2010, pp. 970-77. 58 References 22. A.R. Syed, “Solder Joint Life Prediction Model and Application to Ball Grid Array Design Optimization”, in Proceedings of 1996 SEM conference on Experimental/Numerical Mechanics in Electronic Packaging, 1996, pp.136-44. 23. A. Schubert, R. Dudek, E. Auerswald, A. Gollhardt, B. Michel, and H. Reichl., “Fatigue life models for SnAgCu and SnPb solder joints evaluated by experiments and simulation”, in Proceedings of 53rd Electronic Components and Technology Conference, New Orleans, LA, May 2003, pp. 603-10. 24. A. Ramakrishnan and M. G. Pecht, “A life consumption monitoring methodology for electronic systems,” IEEE Trans. Components and Packaging Technologies, vol. 26, no. 3, pp. 625-34, Sept. 2003. 25. M.H. Chang, D. Das, and M. Pecht, “Interconnect reliability assessment of high power Light Emitting Diodes (LEDs) through simulation”, in Proceedings of 2nd International Conference on Reliability, Safety and Hazard - Risk-Based Technologies and Physics-of-Failure Methods (ICRESH 2010), Vashi, New Mombai, India, 2010, pp. 418-24. 26. “International Technology Roadmap For Semiconductors 2011 Edition Executive Summary,” http://www.itrs.net/Links/2011ITRS/2011Chapters/ 2011ExecSum.pdf, 2013. 27. A. Dasgupta, C. Oyan, D. Barker, and M. Pecht, “Solder Creep-Fatigue Analysis by an Energy-Partitioning Approach,” Trans. ASME Jnl. Electronic Packaging, vol. 114, pp. 152-60, Jun. 1992. 28. C.S. Desai, J. Chia, T. Kundu, and J.L. Prince, “Thermomechanical Response of Materials and Interfaces in Electronic Packaging: Part I – Unified Constitutive Model and Calibration,” Trans. ASME Jnl. Electronic Packaging, vol. 119, pp.294-300, Dec. 1997. 29. J.P. Holman, Heat Transfer, 9th ed., New York: McGraw-Hill, 2002. 30. W.H. Giedt, "Conduction (heat)", AccessScience@McGraw-Hill, [Online] Available: http://www.accessscience.com, 2013. 31. W. H. Giedt, “Convection (heat),” AccessScience@McGraw-Hill, [Online] Available: http://www.accessscience.com, 2013. 32. C. Nordling, J. Österman, Physics Handbook, Lund, Sweden: Studentlitteratur, 1996. 33. R. Tummala (ed.), Fundamentals of Microsystems Packaging, New York: McGraw-Hill, 2001. 34. J. Gu, N. Vichare, T. Tracy, and M. Pecht, “Prognostics implementation methods for electronics,” in Proc. RAMS '07, Orlando, FL, 2007, pp. 101-06. 35. J. Henriksen, Corrosion of Electronics, Korrosionsinstitutet Bulletin No. 102, Stockholm: Korrosionsinstitutet, 1991. 59 References 36. M. Lane, W. Ni, R.H. Dauskardt, Q. Ma, H. Fujimoto, N. Krishna, “Debonding of interfaces in multilayer interconnect structures,” in Proc. Symposium on Advanced Interconnects and Contact Materials and Processes for Future Integrated Circuits, San Francisco, CA, 1998. 37. S.X. Wu, C.P. Yeh, K. Wyatt, “A constitutive model of polyimide films and its integration with finite element analysis for residual stress prediction in thin film interconnects,” in Proceedings of INTERpack '97, Hawaii, 1997, pp. 1285-90. 38. P. Lall, M. Pecht, E. Hakim, Influence of Temperature on Microelectronics and System Reliability, New York: CRC Press, 1997. 39. A.A. Gallo, R. Munamarty, “Popcorning: A Failure Mechanism in PlasticEncapsulated Microcircuits,” IEEE Trans. Rel., vol. 44, no. 3, Sept. 1995. 40. P. Lall, R. Vaidya, V. More, and K. Goebel, “Assessment of Accrued Damage and Remaining Useful Life in Leadfree Electronics Subjected to Multiple Thermal Environments of Thermal Aging and Thermal Cycling,” IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 2, no. 4, p 634-649, Apr. 2012 41. H. Chen, L. Wang, J. Han, M. Li, H. Liu, “Microstructure, orientation and damage evolution in SnPb, SnAgCu, and mixed solder interconnects under thermomechanical stress,” Microelectronic Engineering, vol. 96, pp. 82-91, Aug. 2012. 42. C.L. Belady et al., ”Improving Productivity in Electronic Packaging with Flow Network Modeling (FNM),” Electronics Cooling Magazine, vol. 5, no. 1, 1999. 43. C.J.M. Lasance, “The conceivable accuracy of experimental and numerical thermal analyses of electronic systems,” IEEE Transactions on Components and Packaging Technologies vol. 25, no. 3, pp. 366–82, Sept. 2002. 44. D. Pal, G.K. Morris, “Computer-based thermal analysis,” in J. E. Sergent, A. Klum (ed.), Thermal Management Handbook, New York: McGraw Hill, 1998. 45. K. Azar, “Cooling technology options, part 1,” ElectronicsCooling, vol. 9, no. 3, 2003. 46. M. Fabbri, S. Jiang, V.K. Dhir, “A comparative study of cooling of high power density electronics using sprays and microjets,” ASME Journal of Heat Transfer, vol.127, no. 1, pp. 38-48, 2005. 47. A. Bhunia, A. Brackley, C. Nguyen, B. Brar, "Device Scale Heat Removal for High Power Density GaN Devices," in Proc. 2012 IEEE Compound Semiconductor Integrated Circuit Symposium (CSICS), La Jolla, CA, Oct. 2012. 48. D. Price, “A Review of Selected Thermal Management Solutions for Military Electronic Systems,” IEEE Transactions on Components and Packaging Technologies, vol. 26, no.1, pp. 26-39, Mar 2003. 60 References 49. J.N. Shi, M.D. Ger, Y.M. Liu, Y.C. Fan, N.T. Wen, C.K. Lin, N.W. Pu, “Improving the thermal conductivity and shape-stabilization of phase change materials using nanographite additives,” Carbon, vol. 51, pp. 365-72, Jan. 2013. 50. T. Nilsson, “P-order Heat Pipe; OEGPE-12-0014,” Saab Internal report, April 2012. 51. C. Sarno, G. Moulin, “Thermal management of highly integrated electronic packages in avionics applications,” ElectronicsCooling, vol.7, no.4, Nov 2001. 52. [Online] Available: http://www.heatpipe.nl, 2013. 53. P. Lall, “Cooling in Electronic Applications,” in F. Kreith (ed.), CRC Handbook of Thermal Engineering, Boca Raton: CRC Press, 1999. 54. Y. Zhou, J. Yu, “Design optimization of thermoelectric cooling systems for applications in electronic devices,” International Journal of Refrigeration, vol. 35, no. 4, pp. 1139-44, Jun. 2012. 55. I. Mudawar, “Assessment of High-Heat-Flux Thermal Management Schemes,” IEEE Transactions on Components and Packaging Technologies, vol. 24, no. 2, pp. 122-41, June 2001. 56. D. Price, E. Short, “Thermal Design of an Airborne Computer Chassis with AirCooled, Cast Pin Fin Coldwalls,” ASME Journal of Heat Transfer, vol. 127, no. 1, pp. 11-17, Jan. 2005. 57. R.B. Statnikov, J. B. Matusov, Multicriteria Optimization and Engineering, New York: Chapman and Hall, 1995. 58. C. Andersson and J. Liu, “Effect of corrosion on the low cycle fatigue behavior of Sn-4.0Ag-0.5Cu lead-free solder joints,” Intl. Jnl. of Fatigue, vol. 30, no. 5, pp. 917-30, May 2008. 59. IEEE Standard 1413-2010 (Revision of IEEE Standard 1413-1998), IEEE Standard Framework for Reliability Prediction of Hardware, 2010. 60. J.D. Parry, J. Rantala, C.J.M. Lasance, “Enhanced electronic system reliability challenges for temperature prediction,” IEEE Transactions on Components and Packaging Technologies, vol. 25, no. 4, pp. 533–38, Dec. 2002. 61. W. Engelmaier, “Fatigue Life Of Leadless Chip Carriers Solder Joints During Power Cycling,” IEEE Trans. Components, Hybrids, and Manufacturing Technology, vol. CHMT-6, no. 3, pp. 232-37, 1983. 62. R. Ghaffarian and N. P. Kim, “Ball grid array reliability assessment for aerospace applications,” in Proc. 1997 International Symposium on Microelectronics (SPIE), Philadelphia, PA, 1997, pp. 396-401. 63. D. A. Pietila, M. Rassaian, and K. Brice-Heames, “Design characterization of microwave antenna BGA interconnect structure using test-validated physics-of- 61 References failure methods,” in Proc. 37th annual IEEE International Symposium on Reliability Physics, San Diego, CA, Mar. 1999, pp. 347-55. 64. W. Engelmaier, "Solder Joint Reliability, Accelerated Testing and Result Evaluation," in J. Lau (ed.), Solder Joint Reliability: Theory and Applications, New York: Van Nostrand Reinhold, 1990. 65. D. Bhate, D. Chan, G. Subbarayan, and L. Nguyen, “Fatigue crack growth and life descriptions of Sn3.8Ag0.7Cu solder joints: a computational and experimental study,” in Proc. 2007 Electronic Components and Technology Conference, Reno, NV, 2007, pp. 558-65. 66. N. Vichare, P. Rodgers, and M. Pecht, “Methods for Binning and Density Estimation of Load Parameters for Prognostics and Health Management”, International Journal of Performability Engineering, vol. 2, no. 2, pp. 149-61, Apr. 2006. 67. [Online] Available: http://en.wikipedia.org/wiki/Fatigue_(material), 2013. 68. M. Pecht, A. Dasgupta, “Physics-of-Failure: an Approach to Reliable Product Development,” Journal of the Institute of Environmental Sciences, vol. 38, no. 5, p 30-34, 1995. 69. M. Osterman, T. Stadterman, “Failure assessment software for circuit card assemblies,” in Proc. 1999 Reliability and Maintainability Symposium, Washington, DC, 1999, pp. 269-76. 70. S. Ridout, C. Bailey, “Review of methods to predict solder joint reliability under thermo-mechanical cycling,” Fatigue Fracture of Engineering Materials and Structures, vol. 30, no. 5, pp. 400-12, 2007. 71. M. Rassaian, C.S. Desai, R. Whitenack, and J.C. Lee, “A Unified Constitutive Model Based on Disturbed State Concept and Multi-Domain Method for Design and Reliability in Electronic Packaging,” in Proc. ASME InterPack, Hawaii, 1999, pp. 2031-36. 72. W.W. Lee, L.T. Nguyen, and G.S. Selvaduray, “Solder joint fatigue models: review and applicability to chip scale packages”, Microelectronics Reliability, vol. 40, no. 2, pp. 231-244, 2000. 73. G. Massiot and C. Munier, “A review of creep fatigue failure models in solder material - simplified use of a continuous damage mechanical approach”, in Proceedings of Thermal, Mechanical and Multi-Physics Simulation and Experiments in Micro-Electronics and Micro-Systems, EuroSimE 2004, Brussels, Belgium, 2004, pp. 465-72. 74. F. Galland, A. Gravouil, and M. Rochette, “A Global/Local Model Reduction Approach Dedicated to 3D Fatigue Crack Growth with Crack Closure Effect,” IOP Conf. Series: Materials Science and Engineering 10 012043, 2010. 62 References 75. Z. Qian, C.C. Seepersad, V.R. Joseph, J.K. Allen, C.F.J Wu, “Building Surrogate Models Based on Detailed and Approximate Simulations,” ASME J. Mech. Des., vol. 128, no. 4, pp. 668-77, July 2006. 76. X. Fan, M. Pei, and P.K. Bhatti, “Effect of finite element modeling techniques on solder joint fatigue life prediction of flip-chip BGA packages,” in Proceedings of 56th Electronic Components & Technologys Conference, ECTC, San Diego, CA, 2006. 77. R. Darveaux and A. Mawer, “Thermal and power cycling limits of plastic ball grid array (PBGA) assemblies”, in Proc. Technical Program of SMI Surface Mount International, Advanced Electronics Manufacturing Technologies, San Jose, CA, 1995, pp. 315-26. 78. P. Lall, M. N. Islam, M. K. Rahim, and J. C. Suhling, “Prognostics and health management of electronic packaging,” IEEE Trans. Components and Packaging Technologies, vol. 29, no. 3, pp. 666-77, Sept. 2006. 79. J. P. Hofmeister, P. Lall, E. Ortiz, D. Goodman, and J. Judkins, “Real-Time Detection of Solder-Joint Faults in Operational Field Programmable Gate Arrays,” in Proc. 2007 IEEE Aerospace Conference, Big Sky, MT, 2007, pp. 19. 80. S. Mishra, M. Pecht, and D.L. Goodman, “In-situ sensors for product reliability monitoring,” in Proc. SPIE – The International Society for Optical Engineering, v 4755, 2002, pp. 10-19. 81. P. Lall, M. N. Islam, N. Singh, J. C. Suhling, and R. Darveaux, “Model for BGA and CSP reliability in automotive underhood applications,” IEEE Trans. Components and Packaging Technologies, vol. 27, no. 3, pp. 585-93, Sept. 2004. 82. G. Cuddalorepatta and A. Dasgupta, “Effect of Primary Creep Behavior on Fatigue Damage Accumulation Rates in Accelerated Thermal Cycling of Sn3.0Ag0.5Cu Pb-Free Interconnects,” in Proc. International Conference on Thermal, Mechanical and Multi-Physics Simulation Experiments in Microelectronics and Micro-Systems, EuroSimE 2007, Freiburg-im-Breisgau, Germany, 2007, pp. 121-28. 83. N. Patil, D. Das, C. Yin, H. Lu, C. Bailey, and M. Pecht, “A Fusion Approach to IGBT Power Module Prognostics,” in Proc. International Conference on Thermal, Mechanical and Multi-Physics Simulation and Experiments in Microelectronics and Microsystems, EuroSimE 2009, Delft, the Netherlands, April 2009. 84. S. Cheng, M. Pecht, “A fusion prognostics method for remaining useful life prediction of electronic products,” in Proc. IEEE International Conference on Automation Science and Engineering, 2009, Bangalore, India, Aug. 2009, pp. 102-07. 63 References 85. J. Xu and L. Xu, “Health management based on fusion prognostics for avionics systems,” Journal of Systems Engineering and Electronics vol. 22, no. 3, 2011. 86. V. Rouet, K. Moreau, and B. Foucher, ”Embedded Prognostics and Health Management Systems” in Proc. 2nd Electronics System-Integration Technology Conference, ESTC 2008, London, UK, 2008, pp. 79-84. 87. N. M. Vichare and M. G. Pecht, “Prognostics and health management of electronics,” IEEE Trans. Components and Packaging Technologies, vol. 29, no. 1, pp. 222-29, March 2006. 88. A. Syed, “Limitations of Norris-Landzberg equation and application of damage accumulation based methodology for estimating acceleration factors for Pb free solders,” in Proc. International Conf. on Thermal, Mechanical and Multi-Physics Simulation, and Experiments in Microelectronics and Microsystems, EuroSimE 2010, Bordeaux, France, Apr. 2010. 89. W. Dauksher, “Design for Reliability: FE Modeling of lead-free solder interconnects,” in Shangguan D. (ed.). Lead-free Solder Interconnect Reliability, Materials Park: ASM International, 2005. 90. T-Y. Pan, “Critical accumulated strain energy (CASE) failure criterion for thermal cycling fatigue of solder joints,” Transactions of the ASME Journal of Electronic Packaging, vol. 116, no. 3, pp. 163-70, 1994. 64 The Papers 65