IPC-D-279 EN.pdf - 第51页

where L is the wetted length of the solder joint. In addition, besides substantial shear stresses at the interface between the solder joint and the base material to which it is wetted, even lar ger peeling stresses occur…

−20°C to +20°C, in which the change from stress- to

strain-driven solder response takes place, do not follow the

damage mechanism described in Eqs. #1 and #2 [Ref. A-9:

34]. The damage mechanism is different than for more

typical use conditions and is likely dependent on a combi-

nation of creep-fatigue, causing early micro-crack forma-

tion, and stress concentrations at these micro-cracks caus-

ing faster crack propagation during the high stress cold

temperature excursions, as well as recrystallisation consid-

erations.

A-3.5 CAVEAT 3 — High-Frequency/Low-Temperatures

For high-frequency applications, f>0.5 Hz or t

<l s, e.g.,

vibration, and/or low temperature applications, T

< 0°C,

for which the stress relaxation and creep in the solder joint

is not the dominant mechanism, the direct application of

the Coffin-Manson [Ref. A-9: 14] fatigue relationship

might be more appropriate. This relationship is

(50%)=

[

f’

∆γ

]

−1

[Eq. A-5]

where ∆γ

is the cyclic plastic strain range and c ≈ −0.6.

It has to be noted, that the determination of ∆γ

depends on

the expansion mismatch displacements and the separation

of the plastic from the elastic strains.

For loading conditions of this character, it is possible that

high-cycle fatigue behavior may be observed.

A-3.6 CAVEAT 4 — Local Expansion Mismatch

For applications for which the global thermal expansion

mismatch is very small, e.g. ceramic-on-ceramic or silicon-

on-silicon (ﬂip-chip solder joints), the local thermal expan-

sion mismatch becomes the primary cause of fatigue dam-

age. Equation A-4 does not address the local thermal

expansion mismatch. This reliability problem needs to be

assessed using an interfacial stress analysis [Ref. A-9: 35]

and appropriate accelerated testing.

For leaded surface mount components with lead materials

that have CTEs signiﬁcantly lower than copper alloy mate-

rials, e.g., Kovar

or Alloy 42, the results from Eqs. A-1

and A-2 will be optimistic, since the fatigue damage con-

tributions from the solder/lead material CTE-mismatch, the

local thermal expansion mismatch, are not included.

It has shown that the interfacial stresses resulting from the

local expansion mismatch follow [Ref. A-9: 35].

τ∝L (α

Solder

−α

Basr

)(T

max

− T

min

)

[Eq. A-6]

Figure A−2 Solder Joint Pull Strengths for Gullwing Leads Consisting of Alloy 42 from Different Vendors and Copper

[Ref. A-9: 31]

July 1996 IPC-D-279

where L is the wetted length of the solder joint. In addition,

besides substantial shear stresses at the interface between

the solder joint and the base material to which it is wetted,

even larger peeling stresses occur. Both of these stresses

are proportional to the parameters given in Eq. A-6.

From Eq. A-6 it is quite clear, that for leads consisting of

Alloy 42, the wetted length of the solder joint, that is the

length of the lead foot should be minimized to reduce inter-

facial stresses. That, of course, is contrary to the good

practice that the foot length should be at least three times

the lead width for optimum solder joint quality. However,

since in most applications, the local expansion mismatch

results in contributory damage to the more important dam-

age caused by the global expansion mismatch, this contra-

indication can be ignored without suffering catastrophic

consequences.

From the available experimental data, the damage term, to

be used in Eq. A-1, for the local expansion mismatch alone

∆D(local)=

[

L∆α∆T

]

[Eq. A-7]

where the parameters are the same as in Eq. A-6 and

=0.1 mm, a scaling wetted length. The local expansion

mismatch is then treated as an additional loading condition

(see sections A-3.9 & A-3.10).

A.3.7 CAVEAT 5 — Very Stiff Leads/Very Large Expan-

sion Mismatches

Equations A-3 and A-4 differentiate between surface mount

solder attachments that are leadless and those with compli-

ant leads. Leadless solder attachments presume substantial

plastic strains due to yielding prior to creep and stress

relaxation, whereas Eq. A-4 assumes that the compliant

leads prevent stresses in the solder joints to reach levels

where substantial yielding, and thus plastic strains prior to

creep and stress relaxation, can take place.

However, there is an intermediate region that is not cov-

ered by these assumptions. For very stiff, non-compliant

leads (e.g., SM connector headers), perhaps at lead stiff-

nesses K

> ~90 N/mm and/or for very large thermal

expansion mismatches (e.g., ceramic MCMs on FR-4)

resulting in strain ranges ∆γ > ~10%, the damage estimates

in Eq. A-4 can be substantially in error, because the

assumptions underlying Eq. A-4 are violated.

For very stiff leads the stresses calculated in Eq. A-4 can

exceed the yield strength of the solder. Since yielding will

not permit stresses signiﬁcantly higher than the yield

strength, these calculated stress ranges will overestimate

the cyclic fatigue damage and thus result in substantially

underpredicted fatigue lives. To prevent this analytical

error, the stress range in Eq. A-4 needs to be bounded by

the yield strength of solder in shear.

For very large thermal expansion mismatches the full dis-

placements will not be transmitted to the solder joints,

because the leads will accommodate displacements by plas-

tic deformations of the lead material. Possible exceptions

are situations where very stiff leads are also involved, in

which case the solder joint reliability is best estimated

using Eq. A-1 for leadless solder attachments. The strain

range that can be accommodated by creep and stress relax-

ation in the solder joints can be signiﬁcantly exceeded by

the displacements resulting from very large thermal expan-

sion mismatches and the cyclic fatigue damage would be

signiﬁcantly overestimated. Under these conditions FEA is

required to determine the split in the accommodation of the

displacements between the lead and the solder joints.

Under these circumstances, Eqs. A-3 and A-4 will provide

lower and upper bounds for the reliability estimates,

respectively. The higher the lead stiffness, the closer the

expected results will be towards the results given by Eq.

A-3 for the leadless—‘inﬁnitely stiff leads’—solder attach-

ments. Very high lead stiffnesses can occur in the case of

through-hole component leads converted to surface mount

and for connector headers where the male header pins have

been simply bent into a gull-wing lead foot without any

reduction in the lead cross-section. Very high thermal

expansion mismatches occur primarily in accelerated test-

ing and in extraordinary environments like storage and

transport for products that are designed for benign operat-

ing environments.

A.3.8 Statistical Failure Distribution and Failure Prob-

ability

While the physical parameters deﬁne the median cyclic

fatigue life from physics-of-failure considerations, solder

attachment failures for a group of identical components

will follow a distribution—like all fatigue results—which

typically is best described by a Weibull statistical distribu-

tion [Ref. A-9: 36]. Given the statistical distribution, the

fatigue life at any given failure probability for the solder

attachment of a component can be predicted as long as the

slope of the Weibull distribution is known. Thus, the

fatigue life of surface mount solder attachments at a given

acceptable cumulative failure probability per component, x,

is —assuming a two-parameter (2P) Weibull statistical

distribution—given by

(x%)=N

(50%)

[

1n(1 − 0.01x)

1n(0.5)

]

[Eq. A-8]

where β = Weibull shape parameter or slope of the

Weibull probability plot; typically β≈3 for

fatigue tests, from low-acceleration tests of

stiff leadless solder attachments β≈4 and ≈2

for compliant leaded attachments.

IPC-D-279 July 1996

Experimentally, β can be found to be quite variable with

more severely accelerated reliability tests resulting in

tighter failure distributions and thus giving larger values

for β. Values of β in the range of 1.8 to 9.0 have been

observed.

There is some, unfortunately as yet inadequate, evidence

that for lower failure probabilities a three-parameter (3P)

Weibull distribution, postulating a failure-free period prior

to ﬁrst failure [Refs. A-9: 32,37], may be applicable. From

physics-of-failure and damage mechanism considerations, a

failure threshold as provided by a 3P-Weibull distribution

makes sense, since the fatigue damage in the solder joints

has to accumulate to crack initiation and complete crack

propagation. While the 2P-Weibull distribution may be

overly conservative for designs to very small acceptable

failure probabilities (x < ~0.1%), a too liberal choice of the

failure-free period is deﬁnitely non-conservative. This area

requires more work.

Also, when designing to low failure probabilities, the vari-

ability in the quality of the solder joints may no longer be

negligible; also solder joints with latent defects that made

it into the ﬁeld will have in impact on the actual failure

experience of a product in the ﬁeld.

A-3.9 Multiple Cyclic Load Histories

The loading histories over the life of a product frequently

include many different use environments and loading con-

ditions [Refs. A-9: 38,39]. Multiple cyclic load histories

(e.g., ‘‘Cold’’ temperature fatigue cycles combined with

higher temperature creep/fatigue cycles (see Table A-1)

combined with vibration and local expansion mismatches)

all make their contributions to the cumulative fatigue dam-

age in solder joints. Under the assumption that these dam-

age contributions are linearly cumulative—this assumption

underlies Eqs. A-1 and A-2 as well—and that the simulta-

neous occurrence or the sequencing order of these load

histories makes no signiﬁcant difference, the Palmgren-

Miner’s rule [Ref. A-9: 40] can be applied.

Frequently the initial reliability objective is stated as an

allowable net cumulative damage ratio (CDR). The CDR is

calculated as the sum of the ratios of the number of occur-

ring load cycles to the fatigue life at each loading condition

and is

CDR =

j=1

[Eq. A-9]

where

= actual applied number of cycles at a speciﬁc cyclic

load level j,

= fatigue life at the same speciﬁc cyclic load level j

alone.

The fatigue life is frequently not completely speciﬁed and

is normally taken to be the mean cyclic fatigue life. Equa-

tion A-8 can be used with the allowable CDR signiﬁcantly

less than unity to provide margins of safety, or more accu-

rately, margins of ignorance.

Because the failure of solder joints results from wearout

due to fatigue, the failure rate is continuously increasing.

This is in stark contrast to the reliability design philosophy

of MIL-HDBK-217 [Ref. A-9: 41] which presumes a con-

stant failure rate. These increasing failure rates are properly

represented by an appropriate statistical failure distribution.

Thus, to assure low failure risks, the fatigue life should be

speciﬁed at the acceptable cumulative failure probability at

the end of the design life as per Eq. A-3. Thus, Eq. A-9 is

more appropriately written as

CDR(x%)=

j=1

(x%)

= 1

[Eq. A-10]

where

CDR(x%)= cumulative damage ratio resulting in a cumu-

lative failure probability of x%,

(x%) = fatigue life at the cyclic load level j and a fail-

ure probability of x% .

This approach works very well for the design of the solder

attachment for a single component. However, it is inad-

equate for a reliability analysis of the whole assembly.

See section 3.1.13 for a discussion of non-linear or over-

load effects.

A-3.10 System Reliability Evaluation

Equations A-1 through A-10 address the reliability of the

SM solder attachment of individual components. Systems

consist of a variety of different components most of which

occur in multiple quantities. Further, as shown in Table

A-1, many use environments cannot and should not be rep-

resented by a single thermal cyclic environment, and accu-

mulating fatigue damage from other sources, such as cyclic

thermal environments as described in Caveats 2 to 4 as

well as vibration, needs to be included also.

For a multiplicity of components, i, in the system, the

effect of the various components on the system reliability

can be determined from

∑

(N)=1 − exp

{

1n(1 − 0.01x)

i=1

[

j=1

f,i,j

(x%)

]

}

[Eq. A-11]

July 1996 IPC-D-279