Reliability Engineering: Predictions and Extrapolation

Reliability

• Electronics manufacturing will always need Ph.D.s with common sense who can live with imperfection.
• Most practitioners are unaware of the limitations of their knowledge.
• Many Ph.D.'s struggle with making 'best guesses', conservative estimates, and assessing risks.
• A purely empirical approach to reliability only provides knowledge about the tested samples.
• Predictions about similar samples in the same test might be made, but surprises are common.
• True understanding allows for making predictions and extrapolating data outside of the tested range.

Mechanical/Materials vs. Industrial Eng

• Failure is sensitive to combinations of many imperfectly understood factors.
• Databases, neural networks, and general empirical trends analysis can be useful in this context.
• It is important to make good use of what is already understood.
• One doesn't need to be an expert to apply expert knowledge practically.

Accelerated Testing

• In accelerated testing of Ni/Sn solder joints, the intermetallic bond layer grows thicker by an artificial example.
• The intermetallic bond layer growing thicker results in a weaker and having a higher resistance.

Intermetallic Bond Growth

• Failure happens when the IMC reaches a certain thickness in terms of (strength, resistance), XF
• J = -D * dC/dx
• dC/dx = Csn /x
• Intermetallic growth rate is dx /dt ~ K*D/x
• xdx ~ KD*dt
• Integrate: x²/2 ~ KD(t + to)
• Approximate to ~ x²/2 = 0
• Thickness x ~ {K*D}1/2 *t1/2
• Time to fail t₁ ~ XF² / (K * D)

Arrhenius Dependence

• Time to fail te ~ XF2 (K* D)
• Diffusion coefficient depends on temperature T(Kelvin !!!): D = Do * e-∆E/kT
• Where k is Boltzman constant and ∆E is activation energy for mechanism (diffusion) that controls IMC growth rate
• tf(T) = G * e∆Ε/ΚΤ
• In our example, concern is with total growth by diffusion through layer.
• Most cases involve diffusion through (growing) polycrystalline layer
• Diffusion on grain boundaries is much faster than through lattice, but boundaries are much narrower (and T dep weaker)
• Two parallel diffusion mechanisms/paths
• D = D1 + D2
• D₁ = D01 * e-∆Ε₁/ΚΤ
• D2 = D02 * e-∆Ε2/kT
• The sum is not an exponential
• Actual time to fatal thickness = 'combined'
• One or the other mechanism dominates, depending on T
• With accelerated testing, the changes in slope are unknown and extrapolation can be untrustworthy.

Populated PCB

• Printed circuit boards have many solder joints.
• It is important to ask "how reliable are they?"

SnAgCu Solder Joints

• Sn with a few percent of Ag and Cu is the most common solder alloy used.
• We spend a significant part of the semester on this and alternatives (also illustrating parts of a more general approach to reliability).
• Start with how work is presented to industry (the ones who needs it, whether they know it or not).
• First, the terminology they all know:

Thermal Mismatch

• All packaging engineers are familiar with thermal cycling:
• The component and the substrate are connected via solder joints and that is how they support the shear stresses:
• Zero stress is at the middle, the highest stresses are at the edges

DNP

• The neutral point is located at the center.
• Corner joints damage and fail faster

Shear Strain

• Shear strain = ∆x/h
• Same CTE mismatch gives smaller strain on taller joint
• Taller joints damage more slowly (BGA vs LGA)

Bending/Vibration/Drop

• In the last couple of decades expensive electronics also increasingly subject to bending such that they are;
• Supported by two edges
• Vibrating
• Vibration is makes PCB bend back and forth, typically 500 times per second (500 Hz)
• Bending compresses center joints and stretches corner joints
• Corner joints typically wear out and fail faster than others

Accelerated Testing

• Automotive electronics typically has to survive 20 years
• High vibrations and variations in temperature must be accounted for

Accleration Factor

• Acceleration factor = cycles/time to fail in use/ cycles/time to fail in test

Thermal Cycling

• Thermal cycling more complicated, even in simplified test:
• Tmax, tmax, Tmin, tmin, heating and cooling rates are all variables

Accelerated Testing of Solder Joints

• It doesn't Have to be a Waste of Time and Money

Solder Reliability

• Commonly, the ultimate long term life of a high end electronics assembly is limited by the fatigue and failure of a solder joint
• That is why 'everybody' continues to do a massive amount of accelerated testing

Interpreting Reliability Tests

• The most commonly studied are area array (BGA, CSP, ...) joints

Do you need a model?

• Together, the industry has established enormous experience & data base on accelerated thermal cycling results
• Experienced practitioners with 'common' sense ('more stress is worse') can do what most models do: 'predict' accelerated test results.
• Variation of life in test with DNP2, Tmax, tmax, and less with Tmin, tmin (and even less with ramp rates?)

My Thermal Cycling Model?

• It can't necessarily predict accelerated test results better
• It can make justified predictions to outside the range accessible to accelerated testing
• It can tell us if/when predictions are only conservative, and what consequences are for predictions and test protocols
• Accelerated tests are too simple, real life is more complex (varying amplitudes, combinations w. vibration)
• identify major interactions and how to test for 'worst case'

Outline

• What we want/need
• Failure of common models
• Tutorial on microstructure and damage in thermal cycling
• Mechanistic model for thermal cycling
• Failure of current constitutive relations
• Conservative acceleration factor expression
• Isothermal cycling (vibration,..) – simple model and semi-empirical damage accumulation rule – 'worst case scenario'
• Combinations - 'worst case scenario'
• IMC failure and cratering?
• Other alloys – high T and low T
• Solder alternatives?
• Justified shear test?

Reliability of Solder Joints

• Life under repeated (cyclic) loading is commonly determined by one of several competing factors.

Reliability of Solder Joints

• The long term life of a properly optimized joint is almost always limited by fatigue crack growth through the solder
• Damage in isothermal cycling progresses through Sn grain (scales with total inelastic work per cycle)
• Damage in thermal cycling occurs by recrystallization driven crack growth (scaling with high T work), usually delayed until precipitates have coarsened (varies with stress)

Practical Consequences

• SnAgCu failure distributions cannot be Weibull, and we can't count on Weibull to be conservative
• Interlaced twinning structure with superior thermal and isothermal fatigue properties achievable for small volumes (can't count on it: At least 0.1% of joints won't be)
• Harder, so more likely to cause cratering or IMC failure, but that is counteracted by shape of LGA vs. BGA joints.

Practical Consequences

• Miner's rule and other damage accumulation rules overestimate life in random vibration strongly
• Life depends not only on combination of amplitudes but which order they occur in
• Modified Miner's rule helps identify 'worst combination' for testing

Practical: Thermal Cycling

• Faster cooling from reflow gives longer life in accelerated thermal cycling:
  • Effect may be missed in highly accelerated test
  • Stronger in less accelerated test, may be dominant in long term service!
  • Benefit depends on (is reduced by) aging
• Benefit depends on (is reduced by) aging -Benefits of optimized design, materials and process may be reduced or eliminated -In highly accelerated testing -By aging -Often much greater in long term service

Practical: Thermal Cycling

• Generalization of trends observed in accelerated testing to long term service: -Indeed strongest effects of Tmax, tmax, less of Tmin, tmin -tmax dependence -independent of DNP dependence, -weaker at longer times (but not as soon when approaching service – lower T and stress)
• Dependence on DNP2 in test, stronger in service

Practical: Thermal Cycling

• Predicting life in service: -Shall offer highly conservative life prediction in thermal cycling -Acceleration factors are different for QFNs and passives, but overall trends similar -Don't trust current constitutive relations for realistic solder joints

Practical: Combinations

• There is growing concern with respect to sequential or simultaneous thermal cycling + vibration
• Limited amount of test results with no real plan ('look-see') or interpretation
• The worst combination (missed so far?) is high temperature vibration before, or during early stages of, thermal cycling.

Practical: ESS

• Environmental Stress Screening protocols vary
• The 'Best' method is Vibration only, but account for new acceleration factor.

Practical: Other Alloys

• Adding Bi (or Pb) leads to special effect:
• Solder first hardens in aging/annealing or thermal cycling.
• Eventually, it does start to soften.
• This strongly delays recrystallization, contributing to extension of life in thermal cycling. -Effect is much stronger in milder cycling (long term service)

Practical: Shear Test

• Shear testing of individual solder bumps is fast and cheap, but attempts to correlate with thermal cycling have always failed.
• propose and 'justify' a shear strength test that correlates roughly with thermal cycling performance by:
• --Compare alloys, designs, ... in terms of how it takes for shear strength to drop to given strength in anneal
• This does also lend support to the use of shear strength testing after various numbers of cycles

Assessment of Life

• Commonly acknowledged that that most reliability assessments are meaningless or worse if they do not reflect performance in service!
• The overwhelming majority of 'engineering testing' is not focused on predicting actual life.
• Aims at optimizing or comparing - performance in service

Assessment of Reliability

• That brings us to the even 'larger' question of generalizing our experiences and observations: do your test results only apply to the specific samples tested (and in this specific test)?

Assessment of Reliability

• Want to know if solder joints in 'new' assembly can survive 100 drops? Simply try!
• Long term reliability concerns
• -Thermal cycling
• -Isothermal cycling (vibration)
• -Aging (followed by load) – incl. can 'old' assembly survive 100 drops?
• Predictions are difficult, particularly about the future!

Example

• In thermal cycling, the strain on a joint increases with CTE mismatch, DNP and ΔT.
• t decreases with increasing joint height.
• BGA version of a component survives longer than the LGA version:

LGA Vs. BGA

• The case of SnPb solder joint they do (although not by as much as predicted based on the standoff alone
• Unless the resistance of the joint to deformation can be ignored)

LGA vs. BGA

• expected BGAs to last much longer than their LGA version in thermal cycling
• With SAC305 the difference was very small, absent or (most often) opposite
• We shall return to this and explain it (materials properties vary with size).

What we know?

• More importantly, for now, we did at least catch this issue by accelerated testing
• That doesn't happen for other ones (later)
• The behavior is reproducible, and should apply in service (but benefits may be reduced by long term aging)

Generalizing

• We all need to generalize our observation and results.
• Doing this empirically requires very broad and comprehensive testing programs
• This is often observed, but HDPUG (2011) found cases where subsequent life was slightly longer, Cisco found negative or no effect

Comparing Materials

• Which is the better SnAgCu alloy may depend, among other, on dwell time?

Models

• The thermal testing may effect the designs
• Is the a consistent trend, will a design last even longer with a different amount of rigid material

Compairing

• The component that fails first in test may not even be the one you should be concerned with (fail first in service):
• Sensativity to high temperature dwell time also varies with joint/pad size

Accelerated Testing of Lead Free

• Of course, there is more to extrapolations to realistic service conditions:
• We might even need to combine different types of loading (vibration, thermal cycling, ...) ...!

Damage Accumulation

• Explicitly or implicitly we commonly assume Miner's rule:
• CDI = ∑ ni / Ni = 1
• (If we cycle to x% of life under one set of conditions, remaining life under any other set of conditions is also reduced by x%).
• Shear fatigue life of 18 mil SAC405 joints after cycling with higher peak load
• Remaining life is 11x lower than predicted!!

What We Know?

• It is important to understand tests to explain results and to know what to check for when necessary.