Fatigue Model based on Average Cross-Section

0 downloads 0 Views 3MB Size Report
Fatigue Model based on Average Cross-Section Strain of Cu Trace Cyclic Bending. D. Farle/, A. ... in the copper traces, near the connection to the solder pad.
Fatigue Model based on Average Cross-Section Strain of Cu Trace Cyclic Bending

l 2

4 2 3 4 D. Farle/ ,A. Dasgupta ,Y. Zhou ,l.FJ. Caers ,and l.W.C. De Vries Delft University of Technology,Building 36 (EWI),LB 01.480,Delft,the Netherlands

CALCE Electronic Products and Systems Consortium,University of Maryland,College Park,MD 20742 3 Microsoft Research 1 Microsoft Way,Redmond,WA 98052 4 Philips Research,High Tech Campus 7,WDX-3A,Eindhoven,the Netherlands

1.1. Background

Abstract

3D

This study focuses on quasi-static mechanical cycling durability of copper traces on printed wiring assemblies

multitechnology,

multi-functional,

multichip

packaging technology, usually referred to as a "SiP"

PWA specimens populated with Land Grid

(system-in-package), has emerged as one of the leading

Array (LGA) components on copper-defined pads were

methods of miniaturization. These packages can be found

(PWAs).

cycled to failure under zero-to-max, three-point bending.

anywhere

Failure is defmed in terms of electrical opens due to

electronics, to

miniaturization

fatigue damage propagation through the entire cross­

Opto-Electromechanical Systems (MOEMS),biochemical

smart

section of the trace. Failure statistics were collected and

'lab-on-a-chip'

failure analysis was conducted to identify fatigue failures

electronic controls [6, 7].

in the copper traces,near the connection to the solder pad. Cyclic bending of this assembly was modeled with 3D, elastic-plastic, finite

deformation

(geometrically

is

needed, from

portable

biomechanical implants, Micro­

sensors, and

even

home

appliance

Additionally, these packages

frequently forego leads and even solder balls for further miniaturization.

This generally results in stiffer, more

robust,

interconnects.

solder

Consequently,

the

nonlinear) finite element analysis. Due to the complexity

mechanically weakest point in the chain becomes the

of the geometry, a two-step global-local approach was

traces on the PWA [8-13]. When

used to identify the cyclic strain history and the mean­

electronic

stress at the copper trace failure site. A generalized strain-based fatigue model is proposed,

assessing

the

mechanical

assemblies, the

dominant

durability failure

site

of is

generally believed to be in the solder interconnect [12].

to characterize the fatigue durability in terms of the

However, in many package styles, such as BGAs (ball

amplitude of cyclic strain and the cyclic mean of the

grid arrays), LGAs (land grid arrays), MLFs (micro lead

The strain and stress values are

frame) and QFNs (quad flat no-lead), the Cu trace

averaged over the entire cross-section, to be consistent

emanating from the solder pad may be the weakest failure

with the failure criterion defined above.

Average cross­

site, especially if the solder joint is copper-defined rather

sectional model constants are iteratively estimated by

than mask-defined, and if there is a sudden neck with

ensuring that they are simultaneously compatible with

sharp re-entrant corners where the copper trace emanates

hydrostatic stress.

both the durability test data and the copper stress-strain

from the solder pad [11, 13, 14]. This is particularly true

curves used in the FEA (finite element analysis).

in situations with cyclic mechanical loading, such as

To

make better use of time, a Response Surface (RS) was

cyclic

created after a study determined certain key constants

drop/shock [5,8,9,11-13,15,16].

varied reasonably linearly with each other from model to The important impact of this study includes insight cycling, a quantitative model to predict its occurrence,

bending, vibration

and

repetitive

1.2. Motivation

model. This RS was used for the actual iteration. into copper trace failures in PWAs under mechanical

quasi-static

This paper investigates the cyclic stress and strain histories in the Cu traces and attempts to quantify the accumulated fatigue damage in terms of these parameters. Prior research by the author had focused on the solder

and validated guidelines to prevent it by design.

interconnects for this package style [17, 18].

1.

present study the focus is on the Cu traces.

Introduction

In the

The literature reveals many examples where the Cu

With the consumer electronics industry constantly seeking out new ways to decrease the size of its products,

trace is the weakest point in the system and the dominant

many methods of miniaturization of electronic circuits are

failure site [10, II].

being explored [1-5]. With new methods, architectures,

been found to be easily preventable by revision to the

and structures come new reliability concerns.

trace design. Studies show that redesigned Cu traces can

While

Additionally, these failures have

reliability and failure studies generally focus on the solder

indeed survive repetitive mechanical loading and that the

joints of electronics, an increase in other failure sites is

failure site shifts elsewhere [10,12,19,20].

being noticed.

Quantitative

This study focuses on the occurrence of

of

insights

copper

are

traces

needed

into

so

failures

that

the

fatigue

copper (Cu) trace damage sites just outside the solder

durability

can

joint,and offers up a new formation of fatigue model with

predicted and design guidelines can be developed to

be

model constants for the assessment of the Cu.

prevent these failures.

The existing literature does not

have a consensus of Cu properties for electronics devices [21-24].

978-1-4577-0106-1111/$26. 00 ©20111EEE

-

1110

It is generally accepted that the Ramberg-

-

2011 12th. Int. Conf on Thermal. Mechanical and Multiphysics Simulation and Experiments in Microelectronics and Microsystems. EuroSimE 2011

Osgood material model [25] is the best representation to employ:

E- 12: .,0' ,.1 ' " &HDO ,.1 ,, DI-tn,

1

E=

i+ Gzr

(1)

where e is the strain, a is the stress, E is the elastic modulus,

K is

a material

hardening exponent.

coefficient,

and

n

is the .. �

The hardening exponent, n, was

held constant in this study.

Engelmaier [21] and Hong

.1 �--+--1---�-�--��-�-�

[26] show that the sensitivity to the cyclic changes in n is very small compared to that of

K.

Also held constant is

the elastic modulus, E, for the same reason.

Figure

1/4 1/l 1

I

shows a constitutive schematic, pointing out the various monotonic and cyclic values. The decision was made to Engelmaier's work using

E = 82.7GPa [21, 22] has been used most, and falls in the middle of the field as far as the literature goes. Constitutive

-;:-..-::-_ --:;:-;;.

Models

, .'

, ..

Figure 2: [21J Engeimaier's Coffin-Manson plot showing contributions of strains and deviations at extremely low cycles.

aspects of a material model and the differences between use Engelmaier's Ecu and ncu'

..

There are many existing fatigue models that are applied to copper fatigue.

Most prominent is Sines'

model [27], general form seen in Equation

(4).

Sines'

model, however, uses only a linear mean stress effect

Larger K

factor, which doesn't fit well with the observed behavior in this study.

Also well known in the Morrow model, a

form of the Sines model, seen in Equation (5) [28]. This model has a better mean stress effect factor, however it can clearly be seen that if the am value surpasses the a't value, the model mathematically falls apart.

This may

seem physically incorrect, however with the construct of the actual von Mises yield surface it can be seen that if the principal plan is constantly changing with the stress state, a configuration can occur in which the hydrostatic stress (along the hydrostatic axis) could have a magnitude which is greater than the single calculated equivalent stress (or von Mises) stress at the failure surface. Two others widely used are the Smith-Watson-Topper

Figure 1: General stress-strain schematic showing the hysteresis loops of the monotonic model along with the single (dotted) line of the cyclic model, as well as the effects of varying n and K.

(SWT) model [29], Equation (6) and the Walker model [30], Equation (7).

C

1

cycles to failure approach l or, in other words, as the load toward

the

cyclic

fatigue

stress,

a';;

the

k(JH = (Jf Nf b (Ja b tIm = (Jf Nf tIf

-

Engelmaier's work [21] has also showed that as the tends

Neither of these allow for empirical

correction. Clearly a new model was needed.

_

() ()

(4) (5)

corresponding monotonic values to a'; and the fatigue ductility, e'fare lower than expected. Figure 2 shows the fatigue curves tending toward a saturation value before Nt =1. As mentioned previously, the relationship between the cyclic and monotonic values in the constitutive models of Cu is shown in Figure l. Engelmaier's transfer functions between cyclic (with apostrophe) and monotonic (without apostrophe) for strength and ductility are Equations (2) and (3), respectively, which account for the dotted line in Figure 2. The cyclic fatigue constants pair of a'fand e';; must of course lie on the constitutive curve for Cu. Therefore there is a corresponding cyclic Ramberg-Osgood model in Equation are substituted for a and ' (Jf= (J f 2Nf

c.

( t ' ( Cf=C f 2Nft

(I),

K (K')

for the

when a'fand e';;

Experimentation

The copper trace fatigue durability data for this study is taken from experimental work presented elsewhere in the literature by the author [19]. This was work done in conjunction with the author of this study, but performed by co-authors Zhou and de Vries. The test methodology and fatigue results are summarized here for completeness. 2.1.

Specimen & Test Methodology

The package of interest in this study is an RF SIP component for use in portable electronic products.

It

contains multiple wirebonded dice, flip-chip dice, and

1/4

(2)

Nf =lY2

(3)

Nf=

2.

ceramic SMD passives, all on a common substrate. The wirebonded dice are attached to the laminate board with die attach.

There are no die stacks in this SiP.

The

-21102011 12th. Int. Conf on Thermal, Mechanical and Multiphysics Simulation and Experiments in Microelectronics and Microsystems, EuroSimE 2011

package is fairly stiff because of the many components molded together on a multi-layer substrate and has an LGA architecture, which includes interconnection pads along the perimeter, and a large thermal land in the center. A representation of the PWA specimen used in this study can be seen in Figure 3, shown in a standard 3-point flexure setup.

These are single-sided PWAs with daisy­

chained LGA components grouped into 4 columns of 3 components each, with adequate space between columns to accommodate the center roller without causing any damage.

These

PWAs

are

JEDEC

(Joint

Electron

Devices Engineering Council) standard bend test boards. The overall board thickness is 1.56mm, which is a 7-layer layup pattern with O.035mm thick metallization layers and 0.475mm thick dielectric layers. 45flm

thick

and

120flm

wide.

Figure 4: An image of a dye-stained test board with

The Cu traces are The

major

insets showing the position of land #1 and the trace

board

attached to it.

dimensions are: 48mm wide by 130mm long, with a 90mm

wide by 40mm

functional

daisy-chained

long bend test area. components

are

The test setup and equipment is a custom 3-point bend

Non­

used,

with

setup.

The board is clamped at the center roller, in an

corresponding daisy chains in the PWB, to facilitate

attempt to maintain symmetry.

electrical failure monitoring.

allow in-plane sliding, to prevent excessive membrane strains in the test specimen.

The two outer rollers

The center roller imposes

out-of-plane cyclic movement from zero to max.

Since

the loading is zero-to-max (R=O) and not completely reversed (R=-1), tests are conducted with PWAs facing both directions in the fixture. interconnects

experience

Thus, in some tests the

convex

curvature

(tensile

loading) and in some they experience concave curvature (compressive loading). The

next

steps

include:

(a)

overstress

tests

to

determine the destruct limits of the test specimen; (b) failure analysis to determine the dominant failure modes in the overstress tests; (c) selection of AST (accelerated stress test) levels that are below the destruct limits, do not cause failure mechanism shifting, but are sufficiently severe to produce timely failures;

and (d) tests and

analysis to characterize the specimen response to the

Inner Roller

entire range of excitations anticipated in the accelerated

Figure 3: '3-point' test board & bending moment diagram. Components within each group (inner & outer rows) are initially expected to experience similar strain distribution.

stress test. To characterize the specimen for mechanical cycling, a

populated

loading,

The geometry and loading of this specimen are clearly

test

board

was

subjected

to

overstress

instrumented with strategically placed strain

gauges (Figure 5) to record the resulting strain levels

symmetric about the center roller. Three point bending is

(Figure 6).

known to generate a linear gradient of bending moment

showed that a deflection amplitude of about 3mm was to

The overstress tests for the components

and curvature along the length of the specimen, as shown

sufficient

schematically in Figure 3. Consequently, the stress levels

mechanical cycles.

fail

interconnects

within

the

first

few

are highest in the interconnects of the inner columns,

were therefore kept under 3mm. The overstress tests for

The amplitudes for cyclic testing

followed by those of the outer columns. All of the Cu

the board also showed that the force-deflection curve

traces emanating from the lands have been angled so as

remained linear within the 3mm deflection envelope.

not to be perpendicular to the bend direction, except for

Finally, the board was flexed to each of the load levels

those attached to land number one, as shown in Figure 4.

meant to be used in the AST, and strain histories were

Although the components are located symmetrically with

recorded (Figure 6).

are

Cyclic fatigue testing was performed next. Eight test

asymmetric since they are located at the top right corner

boards, populated with 12 components each (schematic

respect

to

the

center

roller,

Land

1

locations

of each component (as shown later in Figure 7). Thus the

seen in Figure 5), were subjected to 3 different load

local curvature at Land 1 is different for each column,

levels, or "I) values", to generate sufficient data points on

generating 4 different loading levels in each bend test.

the S-N curve. The zero-to-max tests were conducted by

-

3 / 10

-

2011 12th. Int. Conf on Thermal. Mechanical and Multiphysics Simulation and Experiments in Microelectronics and Microsystems. EuroSimE 2011

loading both in the concave (compressive) and convex (tensile) configurations.

A "failure" was recorded when

the component registered resistance above the failure

traces emanating from the corner solder pads, just at the edge of the solder mask where the solder joint ends, as shown in Figure 7.

threshold for 10 cycles within 10% of the first failed cycled.

Meaning, for example, if the first failure was at

1,000 cycles, the tenth must be