chapter 2 mechanism of moisture diffusion,...

Chapter 2Mechanism of Moisture Diffusion, HygroscopicSwelling, and Adhesion Degradation in EpoxyMolding Compounds

M.H. Shirangi and B. Michel

2.1 Introduction

In order to design and manufacture robust electronic packages, it is important tounderstand the response of materials and interfaces to the conditions to whichthey will be subjected. Moisture diffusion in epoxy molding compounds (EMCs) isone of the major reliability concerns in plastic encapsulated microcircuits (PEMs),because many failure modes observed in these devices are believed to arise from thediffusion of moisture during manufacturing, storage, or operation [1–5].

Interfacial delamination between EMC and copper-based leadframe in PEMs isa common failure problem in semiconductor packages. Despite extreme demandsfrom industrial sectors for the use of simulation tools instead of expensive andtime-consuming qualification tests, most of the attempts for predicting interfa-cial delamination by numerical methods have failed, because the plastic packagesundergo complex failure modes arising from the package internal stresses andapplied thermo-mechanical loads. It is often believed that failure of the plastic pack-ages during the solder reflow process or accelerated stress testing is due to degradingeffects of moisture, such as adhesion loss, hygroscopic swelling, and vapor pressure[6–9].

Using epoxy-based encapsulating materials for protecting semiconductorsagainst environmental attacks is believed to be a turning point in electronic packag-ing industry. PEMs have many advantages such as lower cost, lighter weight, andbetter performance over hermetic packages. They are generally applied in all indus-trial areas including automotive industry, consumer electronics, military, and spaceapplications. Despite all of their advantages, one important disadvantage of PEMsis that the EMC absorbs moisture when exposed to a humid environment [1–10].EMCs are composite materials made up of an epoxy matrix that encompasses silicafillers, stress relief agents, flame retardants, and other additives [4]. The commonresin in epoxy molding compounds used in electronic packaging is epoxy cresol

M.H. Shirangi (B)e-mail: [email protected]

29X.J. Fan, E. Suhir (eds.), Moisture Sensitivity of Plastic Packages of IC Devices,Micro- and Opto-Electronic Materials, Structures, and Systems,DOI 10.1007/978-1-4419-5719-1_2, C© Springer Science+Business Media, LLC 2010

30 M.H. Shirangi and B. Michel

novolac (ECN) and the common hardener and filler are phenolic novolac (PN) andfused silica (FS), respectively [5].

Moisture behavior of EMC is mainly dominated by the diffusion of water throughepoxy resin. However, the amount of filler and its shape can influence the mois-ture diffusivity. Diffusion of moisture in epoxy resins is affected by several factors;however, surface topology and resin polarity are the primary aspects that affectthe equilibrium moisture uptake. Soles and Yee [10] found that water traverses theepoxy network through a network of nanopores that is inherent in the epoxy struc-ture. They determined the average size of nanopores diameter to vary from 5 to 6.1 Åand account for 3–7% of the total value of the epoxy material. Since the approximatediameter of a kinetic water molecule is just 3.0 Å, moisture can easily traverse intothe epoxy via the nanopores. They found that the volume fraction of nanopores doesnot affect the diffusion coefficient of water in any of the resin studied and arguedthat polar groups coincident with the nanopores are the rate-limiting factor in thediffusion process, which could explain why the diffusion coefficient is essentiallyindependent of the nanopore content.

There are many speculations on the state of water molecules in polymers.Adamson [11] proposed that moisture can transfer in epoxy resins in the form ofeither liquid or vapor. Tencer [12] suggested it is also possible that vapor watermolecules undergo a phase transformation and condense to the liquid phase. Thecondensed moisture was reported to be either in the form of discrete droplets onthe surface or in the form of uniform layers. These water layers are often quantifiedin terms of monolayers of water necessary to initiate and support corrosion of themetallization in PEMs.

Moisture diffusion in a polymer can be analyzed using the so-called thermal–moisture analogy. The method has been developed by a number of researchers[7, 13] to overcome the discontinuity problem of moisture concentration across thebi-material interfaces using normalized variables. More recently, a direct concen-tration approach (DCA) has been proposed by Xie et al. [14] to study the moisturediffusion with varying temperature and humidity conditions such as in solderingreflow.

Prediction of the problems associated with moisture in PEMs requires a fullunderstanding of the mechanism of moisture diffusion in these materials. In thischapter a detailed analysis of the role of moisture in EMC performance is presented.Moisture diffusion in epoxy molding compounds will be first investigated quantita-tively by weight gain measurements of plastic packages as well as standard bulkEMC samples. Then the characteristics of moisture absorption will be studied byperforming moisture desorption and re-sorption tests at various baking conditions.

Another objective of this chapter is to understand the mechanism of moisture-induced volumetric expansion of molding compounds. This phenomenon is knownas hygroscopic swelling and is responsible for an additional mismatch betweenepoxy molding compound and other package materials.

The influence of moisture on the adhesion, one of the most crucial reliabilityconcerns of the epoxy molding compounds, will also be investigated. When theadhesion between a polymer and a substrate like a leadframe is considered in terms

2 Mechanism of Moisture Diffusion in Epoxy Molding Compounds 31

of interfacial fracture toughness, the interface is initially under residual stresses.Depending on the amount of cure shrinkage of the EMC during polymerizationand the coefficient of thermal expansion (CTE) of the EMC and leadframe, theinterface may be under tension or pressure. The situation becomes more complex,when the epoxy molding compound expands due to the moisture absorption, whileother components are not affected by moisture. Neglecting any of the mentionedmechanisms may lead to a completely wrong understanding of the effect of mois-ture on interfacial adhesion. A fracture test setup will be presented and the effectof moisture diffusion on the interfacial fracture toughness will be discussed. Themoisture absorption and desorption curves of the bulk EMC samples will be used toexplain the fracture results. Moreover, the mechanism of the adhesion loss and itsrecoverability upon subsequent baking will be discussed in detail.

2.2 Moisture Diffusion in Plastic Encapsulated Microcircuits

The objective of this section is to study the diffusion of moisture in plastic encapsu-lated devices. Four types of plastic packages were selected to investigate how theirmoisture content changes with time at 85◦C/85% RH (relative humidity). Table 2.1lists the type and initial dry weight of these packages. All of the packages werestored more than 1 year in unprotected conditions after their molding process. Sincethese packages absorb moisture during storage, they should be baked first to reachthe dry state. The bake-out condition is believed to depend on the geometry andstorage time of the packages. However, a typical and standard baking condition inelectronic packaging is 24 h baking at 125◦C. This baking condition may not neces-sarily lead to a fully dry state as will be discussed in the next sections. Despite thisfact, the bake-out condition for all packages was assumed to be as standard (24 h at125◦C) in order to facilitate a consistent comparison between the moisture contentsof these devices.

Table 2.1 Four types of plastic IC packages for the investigation of moisture diffusion

Sample name Package typePackage dryweight (mg)

EMC used inpackage

Package dimensions(mm3)

P1 MO-188 2,209.01 MC-1 14 × 14 × 2.75P2 SOIC-24 672.13 MC-2 15.4 × 7.5 × 2.45P3 SOIC-16 451.18 MC-2 10.3 × 7.5 × 2.45P4 PLCC-44 2,540.47 MC-3 16.5 × 16.5 × 3.8

Package P1 is an MO-188 type which uses solder for the die bonding, while otherpackages use an epoxy die-attach for bonding the die on leadframe. The packages P1and P2 use molding compounds MC-1 and MC-2, respectively. Packages P2 and P3are both SOIC packages that use the same type of molding compound (MC-2). Theonly difference between P2 and P3 is that P2 is larger and has more pins comparedto P3. Package P4 is a PLCC package and is the largest among the four packages. In


0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5M

oist

ure

mas

s, m

g

(time)1/2, (hour)1/2

P1P2P3P4

(a)

0 5 10 15 20 25 30 35 40

P1P2P3P4

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35


0 5 10 15 20 25 30 35 40

Mas

s up

take

, %

(b)

Fig. 2.1 Moisture absorption of four plastic IC packages at 85◦C/85% RH: (a) mass of moisture(mg) in package with square root of time and (b) mass uptake (%) with square root of time

order to measure the dry weight of the plastic parts, all plastic packages were firstbaked at 125◦C for 24 h, then their weight was measured using an electronic balance(0.01 mg), and finally they were placed in a moisture chamber (85◦C/85% RH). Theplastic parts were removed periodically from the moisture chamber, their weight wasmeasured after reaching the room temperature, and they were placed in the chamberfor further sorption. Assuming an initial dry sample at the start of the sorption tests,the weight gain of the packages during the sorption experiment corresponds to theweight of moisture available in the package. Figure 2.1a shows the mass of moisturevs. square root of exposed time. In package P1, the epoxy molding compound is theonly organic material that absorbs moisture. The other packages use epoxy die-attach for bonding the silicon die to the leadframe. However, the thickness of thedie-attach is very low (20–50 μm) and has therefore no significant influence on theoverall mass gain of the plastic package. Hence, it is reasonable to assume for allfour packages that only the EMC is responsible for the moisture absorption of thepackage and other components are whether impermeable to moisture or have nosignificant contribution to the package mass gain.

The mass uptake of the plastic parts at time t can be determined from

Mass uptake (% ) = M(t) − MDry

MDry× 100, (2.1)

where M(t) is the weight of the sample at time t and MDry is the dry weightbefore moisture preconditioning. In order to ensure that the result of bulk mate-rial diffusion can be compared reasonably with that of the package, the weightof the molding compound of each plastic package was estimated using the dataprovided by the manufacturer and used in equation (2.1), meaning that the dryweight in equation (2.1) represents the weight of the molding compound of thepackage only and not the whole weight of the package. Using equation (2.1), themass uptake of the packages was calculated and the results as a function of the


square root of exposure time are shown in Fig. 2.1b. These graphs are more suitablefor a meaningful comparison between the moisture uptakes of different packages,because they provide comparable information about the moisture uptake of moldingcompounds.

From the graphs in Fig. 2.1b it is not possible to find the moisture diffusioncoefficient because usually a bulk material is needed to find the moisture-relatedmaterial properties of the polymers. However, there are some important results bycomparing the package diffusion with standard bulk EMCs, which will be discussedin the next sections.

2.2.1 Moisture Diffusion in a Package vs. in Bulk EMC

Using the same transfer molding process, bulk samples of EMCs (MC-1 andMC-2) were provided by the manufacturers in the form of molded disks of 100 mmdiameter and 2 mm thickness. Their moisture uptake was also measured using thesame method described above. Figure 2.2 shows the moisture uptake curve of aplastic package compared to the moisture uptake curve of the bulk epoxy moldingcompound used in that package.

The results of Fig. 2.2 suggest that the plastic packages and their respective bulkEMCs manifest similar behavior in terms of moisture absorption. For both IC pack-age and bulk EMC, a linear part of moisture absorption curve at the early stages ofsorption is followed by a slight plateau and finally a second linear phase. However,there are some differences between a plastic package and its bulk EMC in terms ofthe diffusion rates, which suggest that for plastic packages additional mechanismsmay exist which are explained in the following sections.

2.2.2 Interfacial Moisture Diffusion

From Fig. 2.2a and b, it can be postulated that the diffusion of moisture in bothpackages is faster than that of their respective bulk molding compounds. This canbe attributed to the three-dimensional nature of diffusion in the package and also toa higher diffusion rate through the leads/EMC interfaces, while for the thin plates,the diffusion is almost one-dimensional.

A review of the diffusion rates available in literature reveals that, in general,the diffusion rate along the interface of adhesive/substrate is faster than that of thebulk adhesive and it becomes more critical to the lifetime of the adhesive jointsas the strength of the interface decreases. Vine et al. [15] studied the moistureuptake of an epoxy bonded to aluminum substrates with various surface treat-ments. They observed faster diffusion in three-layer sandwich specimens thanpredicted based on mass-uptake experiments performed on bulk diffusion speci-mens. They attributed this behavior to the presence of micro-cavities in the adhesivelayer.


0

0.050.

1

0.150.

2

0.250.

3

0.35

Mass uptake, %

(tim

e)1/

2 , (

hour

)1/2

P2

MC

-2

05

1015

2025

3035

40

(a)

0

0.050.1

0.150.2

0.250.3

0.35

(tim

e)1/

2 , (

hour

)1/2

05

1015

2025

3035

40

Mass uptake, %

P1

MC

-1

(b)

Fig

.2.2

Moi

stur

eab

sorp

tion

ofa

plas

ticIC

pack

age

com

pare

dto

itsbu

lkE

MC

:(a)

P1vs

.MC

-1an

d(b

)P2

vs.M

C-2


Davis et al. [16] utilized the technique of electrochemical impedance spec-troscopy and investigated epoxies bonded to aluminum with various surface prepa-rations that resulted in either a weak or strong interface and observed that the rate ofcrack growth was slower for strong interfaces. However, for weak interfaces crackgrowth was detected almost immediately as moisture appeared at the interface andresulted in a fast rate of crack growth. Zanni et al. [17] compared the calculated dif-fusion rates between non-bonded adhesive specimens and bonded adhesive joints.They observed that the interfacial diffusion coefficient was greater than that of bulkadhesives and hypothesized the phenomenon of “capillary diffusion,” where thehigher surface energy of the dry adhesive effectively pulls moisture along the inter-face. They used Fickian diffusion to measure water diffusion into bonded joints andrelated enhanced rate of moisture ingress to water diffusion occurring rapidly in theinterface region.

These examples all suggest that the diffusion of water into the adhesive jointsis not a simple process and may involve several pathways for the moisture ingress,dependent on the system chemistry and interfacial morphology.

2.2.3 Moisture Accommodation at Interfaces

An important observation in Fig. 2.2 is the higher amount of maximum moisturecontent in a package compared to its respective bulk molding compound. Thisdifference can be attributed to the accommodation of water molecules at the inter-faces between molding compound and the leadframe. This hypothesis is supportedby the study of Chan et al. [18] who performed Fourier transform infrared spec-troscopy (FTIR-MIR) technique measurements that showed the maximum moistureconcentration at an interface is higher than that at the saturation state of a bulkmolding compound.

There are many other evidences in the literature on the accumulation of waterat the interfaces [19]. Buchwalter [20] suggested that for weak interfaces wheresecondary bond forces dominate the adhesion, failure occurs almost immediatelyas water contacts the interface. O’Brien [21] suggested that it is possible for thewater to be present in the bulk and interface of the adhesive, yet the integrity ofthe adhesive bond can be preserved if the interface is strong. This is the case wherecovalent bonds are present. For strong interfaces, the role of interfacial diffusionbecomes less important and the rate-limiting step for failure becomes the chemicalreaction at the interface.

Wu et al. [22] used neutron scattering by D2O to show high concentration ofwater at polyimide interfaces. They showed that the concentration of water at inter-faces without adhesion promoters was significantly higher than with them. Nguyenet al. [23] used infrared spectroscopy to measure the accumulation of water at anepoxy/SiO2 interface and correlated it to weakened adhesion. They detected signif-icant diffusion at the interface for poorly adhered adhesive systems where adhesionforces are governed by secondary interactions. Bowden and Throssell [24] found


that on aluminum, iron, and SiO2 surfaces, the layer can be up to 20 molecular lay-ers thick at ambient temperatures and humidity. Takahashi [25] used AC impedancespectroscopy to show that at a relative humidity of 80% the interfacial capacitanceincreased abruptly, suggesting the formation of water clusters in the bulk and at theinterfacial regions.

Another possible reason for a slightly higher amount of moisture content in plas-tic packages compared to that of bulk molding compounds may be the presence ofresidual stresses in packages which causes some nano-scale damages along the inter-faces. Buchwalter [20] studied the effect of applied stresses on the moisture-induceddamage at interfaces of a plastic package and used a wedge test to externally loadthe adhesive bonds and concluded that mechanical stresses can work in concert withmoisture to cause interfacial damage other than that caused by moisture alone.

Since there are many uncertainties when a package is used for the understand-ing of moisture diffusion in EMCs, for the rest of this work the focus will be onthe moisture absorption of bulk molding compounds. MC-1 which is a commercialmolding compound will be used further for a systematic investigation of moistureabsorption and desorption in EMCs. Moreover, other samples required for the char-acterization of hygroscopic swelling and interfacial adhesion will be produced fromthe material MC-1.

2.2.4 Fickian Moisture Diffusion

Diffusion of water in polymers has been widely investigated and for most cases therate of diffusion has been assumed to be constant (Fickian diffusion). Fick’s secondlaw can be applied to describe the moisture diffusion process in many polymericmaterials as follows [26]:

∂C

∂t= ∇ · (D∇C), (2.2)

where D (mm2/s) is the diffusion coefficient, C (g/mm3) is the moisture concen-tration, t (s) is the time, and x (mm) refers to Cartesian coordinates. For isotropicmaterials, Fick’s second law can be simplified as follows:

∂C

∂t= D

(∂2C

∂x2+ ∂2C

∂y2+ ∂2C

∂z2

). (2.3)

If the one-dimensional case of an infinite plate of thickness l with appropriateboundary conditions is considered, the analytical solution, giving the temporal andspatial moisture concentration, C, at time t and distance x from the mid-plane isgiven by


Ct − C0

C∞ − C0= 1 − 4

π

∞∑n=0

( − 1)n

2n + 1exp

(−D(2n + 1)2π2

4 l2t

)× cos

((2n + 1)π

2 lx

).

(2.4)

Here C∞ is the maximum equilibrium moisture concentration, C0 is the initialmoisture content, and D is the Fickian diffusion coefficient.

Since it is not possible to measure the moisture concentration at a point experi-mentally, the above expression is integrated over the thickness of the bulk film andthe fractional mass uptake of the specimen as a function of time is [26]

Mt

M∞= 1 − 8

π2

∞∑n=0

1

(2n + 1)2exp

(−D(2n + 1)2π2

4 l2t

), (2.5)

where Mt is the mass of moisture after absorption time t and M∞ is the mass ofsaturated sample. Equation (2.5) implies that the moisture mass of a polymericsample exposed to a humid environment obeys an asymptotic behavior, for whicha saturation state exists. In other words, if two samples with different thicknessesare exposed to a humid environment, the final moisture uptake defined by equation(2.5) is the same for both samples; however, the time to reach this maximum valuedepends on the thickness of the sample as shown in Fig. 2.3.

∞== CtlxC ),(

l

∞== Ctx

x

C ),0(

21 ll <∞MMt

t(a) (b)

Fig. 2.3 One-dimensional Fickian diffusion in a plate

2.2.5 Non-Fickian Dual-Stage Moisture Diffusion

As observed in Fig. 2.2a and b, there is no equilibrium of moisture absorption evenafter long-period moisture sorption. It was suspected that this phenomenon may bedue to the larger thickness of the bulk samples (2 mm). However later investiga-tions showed a dual-stage moisture absorption for all sample geometries. In order toinvestigate the role of geometry in the moisture performance of bulk EMCs, twodifferent sample geometries of material MC-1 were produced. The sample withthickness 1 mm has a diameter of 50 mm and the sample with thickness 2 mm has a


0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0 5 10 15 20 25 30 35 40

Mas

s up

take

, %

(time)1/2/l,(hour)1/2/mm

thickness: 2 mm

thickness: 1 mm

Fig. 2.4 Effect of samplegeometry on the non-Fickianbehavior

diameter of 100 mm, meaning that for both samples the one-dimensional moisturediffusion can be assumed. Figure 2.4 shows the percentage of moisture content as afunction of

√t/l, where t is the time of exposure to moisture and l is the thickness

of the bulk EMC.It can be observed that the rate of mass uptake per thickness was the same for both

sample geometries at the beginning of the exposure to moisture, which is a typicalbehavior of Fickian moisture diffusion in polymers. However, after the

√t/l ratio of

around 13, the curves started to deviate from each other. This ratio corresponds to 1week of sorption for the sample with 1 mm thickness and approximately 3–4 weeksfor the sample with 2 mm thickness, respectively. This sorption time will be denotedas virtual saturation in this work and will be used as a virtual border between thetwo phases of the moisture absorption.

It is important to note that in equation (2.3) the diffusivity D in Fickian diffusionis assumed to be independent of moisture concentration C. This assumption maynot hold true for many polymers and, consequently, equations (2.3), (2.4), and (2.5)cannot be used directly for the case of non-Fickian diffusion. A prominent featureof non-Fickian diffusion is that there is no characteristic equilibrium mass uptake.There are various types of non-Fickian moisture absorption reported in the litera-ture. Chen and Zhao [27] reported that the moisture absorption in some moldingcompounds can be characterized by linearly decreasing diffusivity as a function ofaverage moisture content. However, Celik et al. [28] found that for highly non-Fickian diffusion of some organic substrates a power-law relation between thediffusivity and moisture content exists.

Non-Fickian behavior may be the consequence of a relaxation process in polymermolecules and/or the result of an irreversible reaction between polymer and moisturesuch as formation of hydrogen bonds. Weitsman [29] proposed various types ofnon-Fickian moisture absorption and used a combined damage/diffusion model tointerpret the non-Fickian moisture uptake of fiber-reinforced polymeric composites.Loh et al. [30] suggested that non-Fickian diffusion is generally considered to occur


when the relaxation of the polymer influences the uptake behavior. Such responsesare conventionally divided into two groups. One is known as class II (class I beingFickian diffusion) and generally occurs when the relaxation rate controls the uptake.The other group is termed as anomalous uptake and generally occurs when diffusionand relaxation have comparable rates. Two-stage or dual-uptake diffusion has beenobserved in many polymeric materials and is one of the most common types ofanomalous moisture uptakes.

It must be noted that the aging at 85◦C/85% RH (MSL1) is one of the most severemoisture sensitivity level (MSL) reliability test conditions. Figure 2.5 shows themoisture absorption curves of the material MC-1 in two aging conditions. The rateof diffusion and the maximum mass uptake at MSL1 are higher than that at MSL3condition (30◦C/60% RH). The dual-phase moisture absorption can be observed atthe MSL1 condition more significantly. However, there seems to be a slight perma-nent increase in the mass uptake curve at 30◦C/60% RH, which suggests that thesecond phase of moisture absorption may exist even at low temperatures.

0

0.05

0.1

0.15

0.2

0.25

0.3

0 5 10 15 20 25 30 35 40

Mas

s up

take

, %

(Time)1/2, (hour)1/2

85°C/85%RH

30°C/60%RH

Fig. 2.5 Effect of sorptionconditions on the non-Fickianbehavior

Another mechanism leading to a non-Fickian behavior of moisture absorp-tion may be the swelling of the polymer matrix, which increases the number ofactive sites available for sorption. Also a chemical degradation of epoxy resinsand/or epoxy/filler interface may cause the non-Fickian diffusion. In the latter case,Lekatou et al. [31] observed that the water diffusion initially follows the Fickianmodel, but then the deviations from the ideal behavior are explained by the flow ofwater molecules along the filler–matrix interface followed by diffusion into the bulkresin and transport of water by microcracks.

Most likely, the dual-phase moisture absorption in epoxy molding compoundscan be explained with the two mechanisms of absorbing water by polymers. Duringthe early stages of the sorption, the absorbed molecules reside in the free volumes inthe polymer, but their transfer to a bound state with a different energy level requiresovercoming some energy barriers and occurs relatively slowly. Possibly the two


bound and unbound mechanisms act simultaneously. However, due to the faster dif-fusion in free volumes at the beginning of the sorption, the effect of the bondingmechanism can be observed later, when the first mechanism slows down by reach-ing its saturation state. In this case, most of the bonded water causes non-Fickiandiffusion behavior that typically causes a gradual increase of moisture uptake withtime. These statements are supported by the results of desorption tests after varioussorption times, which will be discussed in Section 2.3. It will be shown that the non-Fickian behavior during the absorption process is responsible for a non-reversiblesorption process. This supports the hypothesis that there is an energy level for thetransition from free volume diffusion to molecular bondings.

By using equation (2.5), the diffusion coefficient of bulk materials can be foundfrom the slope of the initial linear part of the moisture uptake curve togetherwith the sample weight at saturation state. The initial stage of moisture absorption(Mt/M∞ < 0.5) can be simplified as follows:

Mt

M∞= 4

(Dt

πl2

)1/2

. (2.6)

However, the problem of moisture diffusion in molding compounds is that fromFigs. 2.4 and 2.5, no specific saturation point can be observed. Hence, an estima-tion of the diffusion coefficient by using Fick’s law is not possible. By consideringa virtual saturation level after 168 h, a first value of diffusion coefficient can beestimated. Using a longer soaking time as saturated level will result in obtaininga smaller Fickian diffusion coefficient. This is repeated for various sorption timesuntil a relation between D and M∞ can be obtained. After having the correspond-ing values of D and M∞, a transient finite element (FE) analysis is performed foreach pair of data and the average moisture content at the end of diffusion pro-cess is calculated. Figure 2.6 shows the relation between the apparent diffusion

0.0E+00

4.0E–07

8.0E–07

1.2E–06

1.6E–06

2.0E–06

8.0E–6 1.0E–5 1.2E–5 1.4–5 1.6E–5 1.8E–5

Diff

usio

n co

effic

ient

, mm

2 /s

average moisture concentration, g/mm3

MC-1

MC-2

Fig. 2.6 Diffusion coefficient of two bulk EMC materials as a function of average moisture contentat 85◦C/85% RH condition


coefficients with the average moisture content in the sample. It indicates clearlythat the diffusion coefficients of both materials decrease with increasing moisturecontent.

A three-dimensional transient finite element analysis was performed using thethermal–moisture analogy by the simulation tool ANSYS. The dashed lines inFig. 2.7 show the Fickian simulation of the weight gain process by assuming afinal saturated level at the end of absorption process (1,000 h sorption at 85◦C/85%RH). For this Fickian simulation, D is assumed to be the last point in Fig. 2.6 andC∞ is the assumed saturation moisture concentration, which can be found fromC∞ = M∞/V , V being the specimen volume.

0 5 10 15 20 25 30 350

0.05

0.1

0.15

0.2

0.25

Mas

s up

take

, %

0 5 10 15 20 25 30 350

0.05

0.1

0.15

0.2

0.25

(time)1/2, h1/2

Mas

s up

take

, %

Experimental Data, MC−1

Fickian Simulation, MC−1

Non−Fickian Simulation, MC−1

Experimental Data, MC−2

Fickian Simulation, MC−2

Non−Fickian Simulation, MC−2

Fig. 2.7 Fickian and non-Fickian simulations of the moisture absorption in two EMC plates


In order to determine the non-Fickian dual-stage diffusion properties, the fol-lowing method proposed by the current authors [1] is used: A first Fickian diffusionwith the diffusion coefficient D1 and saturated content level of C1∞ is followed by asecond Fickian (D2 and C2∞ = C∞−C1∞) after a specific time (t2 = 168 h). Threeindependent variables (D1, D2, and C1∞) are found by using a least mean-squareapproach to produce the best fit to the experimental curve. The Fickian (D and C∞)and non-Fickian (D1, D2, and C1∞) diffusion parameters are listed in Table 2.2.

Table 2.2 Fickian and non-Fickian absorption parameters

DFickian (mm2/s) C∞ (g/mm3) D1 (mm2/s) D2 (mm2/s) C1∞ (g/mm3)

MC-1 0.7e–6 5.4e–6 1.9e–6 1.5e–6 3.7e–6MC-2 0.8e–6 4.53e–6 2.3e–6 1.2e–6 3.1e–6

As shown in Fig. 2.7 the non-Fickian model that is composed of two parallelFickian diffusions can be used to obtain the mass uptake of the molding compoundsvery exactly. The fitting parameters D1, D2, and C1∞ are so far used only to get thebest fit of the experimental results and may not necessarily have physical meanings.However, they could be further investigated in the future to achieve a meaning-ful interpretation of sorption phases and their contribution to the overall moistureuptake of molding compounds.

2.3 Moisture Desorption

Moisture desorption in epoxy molding compounds takes place at reflow process.Moreover, it is important to understand the mechanism of moisture desorption, sinceduring the assembly of PEMs, the packages undergo baking to remove moisture andthus reduce the probability of moisture-induced failures such as popcorn crackingand interfacial delamination.

Experiments on the moisture desorption behavior of the plastic packages listed inTable 2.1 were performed at various temperatures. The packages were stored in anunprotected environment before the sorption tests and thus initially contained mois-ture due to long-period storage in air. The plastic parts were first baked at 125◦C for24 h to remove the initial moisture content. The weight after baking was consideredas the dry weight; however, results found from desorption curves of the packagesand also from desorption of bulk EMCs revealed that the condition of 125◦C bakingfor 24 h may not fully remove all the moisture content. After baking, the pack-ages (three packages for each test category) were placed in a moisture chamber at85◦C/85% RH for a sorption time of 168 h. Afterward, the desorption experimentswere performed in four infrared chambers at four baking temperatures: 75, 110, 160,and 220◦C. Since the glass transition temperature of the EMCs used in the packagesrange from 95 to 130◦C, the aim was thus to run the desorption experiments at twotemperatures below and two temperatures above the glass transition temperature.The desorption period (baking time) was chosen to be 96 h for baking at 75◦C, 70 hfor baking at 110◦C, 48 h for baking at 160◦C, and 4 h at 220◦C, respectively. The


–0.5

0

0.5

1

1.5

2M

oist

ure

mas

s, m

g

(time)1/2, (hour)1/2 (time)1/2, (hour)1/2

(time)1/2, (hour)1/2(time)1/2, (hour)1/2

P1, 75°CP1, 110 °CP1, 160 °CP1, 220 °C

P4, 75°CP4, 110 °CP4, 160 °CP4, 220 °C

P3, 75 °CP3, 110 °CP3, 160 °CP3, 220 °C

P1, 75°CP1, 110 °CP1, 160 °CP1, 220 °C

–0.2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Moi

stur

e m

ass,

mg

)b()a(

–0.4–0.3–0.2–0.1

00.10.20.30.40.50.6

Moi

stur

e m

ass,

mg

00.5

11.5

22.5

33.5

44.5

0 2 4 6 8 10

0 2 4 6 8 10

0 2 4 6 8 10

0 2 4 6 8 10

Moi

stur

e m

ass,

mg

)d((c)

Fig. 2.8 Moisture desorption of four plastic IC packages at various temperatures: (a) MO-188, (b)SOIC-24, (c) SOIC-16, and (d) PLCC-44

weight measurements of the plastic parts were performed periodically by removingthe parts from the oven and measuring at room temperatures similar to the pro-cedure explained in Section 2.2. Figure 2.8a–d shows the temperature-dependentdesorption results of the packages P1–P4, respectively.

The results of Fig. 2.8 indicate that for all packages studied, the baking condi-tion at 75◦C cannot lead to a complete removal of moisture content. Even bakingat 110◦C seems to be inappropriate for the removal of moisture from the packages.Interestingly, for the package P3 which is the smallest part among the packages, thefinal weight of the sample at the end of bake-out was lower than the apparent initialdry weight, which explains why the moisture mass in Fig. 2.8c was a negative valueat longer baking times. This can arise from two possible reasons. The first reasonmay be the out-gassing of the polymer during the baking condition. This assump-tion was proved not to hold true due to the results found from thermal gravimetricanalyzer (TGA) measurements of dry bulk EMC samples that showed insignificantout-gassing even at 220◦C. The second reason is that, principally, the assumption ofgaining a dry package upon 24 h baking at 125◦C might have been wrong. It seemsthat this baking condition did not lead to a complete removal of the moisture contentand the apparent dry samples contained moisture in reality. It seems to be difficultto understand the mechanism of moisture desorption from the results of baking of


plastic IC packages, because other phenomena like interfacial diffusion and mois-ture accumulation may lead to a misinterpretation of the desorption results. That iswhy standard bulk samples were later used to understand the intrinsic mechanismof moisture desorption.

Samples of bulk EMC (both geometries with 1 and 2 mm thickness of the mate-rial MC-1) were placed in humid conditions (85◦C/85% RH). Some samples wereremoved after 2 weeks of sorption while the others were removed after 4 weeks ofsorption for a subsequent baking in an infrared oven. Polymers lose their moisturecontent when they are exposed to dry environments at high temperatures. However,not all the moisture may escape from them even upon exposure to high tempera-ture. In this work a systematic approach was chosen to find out at which state of themoisture absorption the formation of the residual moisture content may take place.Figure 2.9a and b illustrates the weight loss of samples at 110 and 160◦C, respec-tively. A higher initial moisture content in these figures indicates that the sample was

0

0.05

0.1

0.15

0.2

0.25

0.3

0 2 4 6 8 10

Mas

s lo

ss, %



Baking at 110°C after virtual saturation (thickness = 1 mm)Baking at 110°C after 2nd phase absorption (thickness = 1 mm)Baking at 110°C after virtual saturation (thickness = 2 mm)Baking at 110°C after 2nd phase absorption (thickness = 2 mm)

(a)

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0 2 4 6

Mas

s lo

ss, %

Baking at 160 °C after Virtual Saturation (thickness = 1 mm)Baking at 160 °C after 2nd Phase absorption (thickness = 1 mm)Baking at 160 °C after Virtual Saturation (thickness = 2 mm)Baking at 160 °C after 2nd Phase absorption (thickness = 2 mm)

(b)

Fig. 2.9 Desorption of an EMC as a function of exposure time to dry environment: (a) desorptionat 110◦C and (b) desorption at 160◦C


aged for a longer time in humid condition prior to baking and hence the moistureabsorption had reached the second absorption phase. This longer aging in a moistenvironment has a direct influence on the subsequent baking curves. From thesedesorption curves the following results can be postulated:

• A complete removal of moisture was not achieved for any of the samples withinthe time period investigated. Samples with higher initial moisture content show ahigher final residual content, suggesting that a longer time of exposure to mois-ture leads to a higher amount of “non-reversible” moisture content at a certaintemperature. Baking at elevated temperatures (e.g., 160◦C) leads to more mois-ture release and a lower amount of residual moisture content. This may meanthat for the debonding of hydrogen bonds between water molecules and poly-mer chains a certain amount of energy is needed, which is not available at lowertemperatures (e.g., 110◦C).

• The thicker samples (2 mm thickness) are shown to retain a higher amount ofresidual moisture content when compared to the thinner samples (1 mm thick-ness). This is reasonable, as thicker samples normally need more time to reacha certain amount of moisture content during the moisture absorption. A longerexposure to moisture results in a higher amount of non-reversible hydrogen bondsand consequently a higher value of residual moisture contents. This suggests thathydrogen bonding is active from the early stages of exposure to moisture; how-ever, its influence is more dominant when the Fickian mechanism deceleratesupon reaching a virtual saturation.

• An interesting observation is the parallel initial linear parts of the desorptioncurves of the samples with similar thickness. For a certain geometry at a constanttemperature, the desorption curves of samples with different histories are in par-allel. A longer exposure to moisture results in a higher amount of non-reversibleabsorption mechanism; however, the desorption rate at a constant temperaturedoes not depend on the sorption history and depends only on the baking con-dition. This means that the non-reversible mechanism arising from the secondabsorption phase has influence only on the residual moisture content and not onthe desorption coefficient.

In contrast to the results presented in this section, there are some other stud-ies that suggest a complete removal of moisture content upon baking. He and Fan[32] performed in situ measurements of moisture absorption/desorption on thinfilm bismaleimide-triazine resin/glass fiber laminates and observed repeatable mois-ture absorption. It is possible that in these cases only the first phase of moistureabsorption was activated within the time period of the recycled test.

The Fickian model fails to describe the desorption behavior of such polymerswith a dual-stage diffusion, because of two reasons. First, it over-estimates themoisture escape from the polymeric materials and, second, it does not consider theability of the materials to keep a certain amount of water after a long-term baking.It is worth noting that upon 1 week baking at 110◦C, a residual moisture contentof around 40% of the saturated level was observed in the thick samples. However,


not enough effort has been made on solving the failure of Fickian model for pre-dicting these residual moisture contents. A simple non-Fickian moisture desorptionmodel is suggested in this work, which can solve this problem with conventional FEsimulation tools.

In this model, two parallel diffusion simulations are performed. The first one is adesorption process with the diffusion coefficient D1 and boundary conditions similarto a Fickian desorption (initial nodal moisture concentration of all nodes being C =C∞ and applying C = 0 on the external surfaces of the sample). The second one is aparallel absorption process of an originally dry sample with the diffusion coefficientD2 and the boundary conditions being similar to a Fickian boundary condition withC = Cresidual on the external surfaces. The nodal moisture concentration at eachspecific time of the first analysis should be added to that at the corresponding timeof the second analysis. The result is a non-Fickian moisture desorption with thecontrolled desorption coefficient and, more importantly, with the residual moistureconcentration Cresidual.

The residual moisture content can be simply found from experimental data byCresidual = Mresidual/V, where Mresidual is the residual moisture mass in the sampleat the end of desorption process and V is the volume of the sample. The parametersD1 and D2 can be found by fitting the results to the experimental curve, similar tothe approach described in the previous section. The non-Fickian desorption modelis compared with both experimental data and Fickian model in Fig. 2.10. Table 2.3summarizes the fitted parameters of the non-Fickian model and compares them withthe values of Fickian desorption.

It must be mentioned that the desorption experiments were performed in aninfrared oven, with possibly some humidity in the air. Some of the residual moisturecontent in the materials could be the result of a competing parallel absorption dueto the moisture available in the air. Another source of error arises from a short timeneeded for the weighing of the specimens after they are removed from the oven.

0 5 10 15 200

0.05

0.1

0.15

0.2

0.25

0.3

timeo.5 (houro.5)

Wei

gh

t L

oss

(%

)

Experiment, T = 110°C

Fickian Simulation, T = 110°C

Non−Fickian Simulation, T = 110°C

Fig. 2.10 Comparisonbetween experimental,Fickian, and non-Fickiansimulations of MC-1 at110◦C


Table 2.3 Fickian and non-Fickian desorption parameters

T (◦C) DFickian (mm2/s) D1 (mm2/s) D2 (mm2/s) Cresidual (g/mm3)

110 0.7e–6 3.5e–6 5.5e–6 2.2e–6160 2.5e–5 8.5e–5 8.5e–5 1.5e–6220 15.5e–5 20e–5 19e–5 0.6e–6

Since the electronic balance scale is sensitive to the heat, a time of approximately5 min is needed, so that the hot samples reach the room temperature. Within thistime some moisture uptake from the atmosphere is possible.

Assuming Fickian desorption, the desorption coefficient can be estimated fromthe initial linear part of the experimental data by using equation (2.6). Thetemperature-dependent desorption coefficient of the material MC-1 is shown inFig. 2.11. It can be observed that the desorption coefficient fulfills the followingequation:

D = D0 exp

(−�E

kT

). (2.7)

The diffusion coefficient D depends on an initial diffusion constant, D0, temperature,T, and activation energy, �E, where k is Boltzmann’s constant. For the sample MC-1, the activation energy can be estimated as 0.55 eV.

There are some researchers that proposed direct measurements of the diffusioncoefficient from the plastic packages. Although these methods are not precise, theyenable reasonable comparison between different EMCs and provide more flexibilityfor the selection of a proper EMC for a specific application. Teverovsky [33, 34]proposed a simple rapid technique for the estimation of temperature dependencyof moisture diffusion characteristics directly on the plastic packages of PEMs. Thesuggested technique is based on substituting the moisture sorption kinetics withthe temperature domain measurements of moisture desorption from a flat polymersample or flat package of PEM that has been pre-saturated with moisture in humidity

2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8

× 10−3

−16

−14

−12

−10

−8

1/T(1/K)

ln(D

), m

m2 /

s

ln(D) = −6400*(1/T) + 3.5

Fig. 2.11 Desorption coefficient of MC-1 against adverse temperature


chamber. Assumptions for this technique are as follows: (1) temperature dependenceof diffusion coefficient follows Arrhenius law, (2) Fickian diffusion holds true, and(3) the temperature in the package is uniform. Using this technique, the diffusioncoefficient of a QFP package was found as a function of temperature and the resultis shown in Fig. 2.12. It was also found that the diffusion coefficients determinedfrom the packages were 10–60% larger than those found directly from the bulkmolding compounds, which is in agreement with the results of this work.

y = 0.1416e–5.148x

1.E–09

1.E–08

1.E–07

1.E–06

1.E–05

2 2.25 2.5 2.75 3 3.25 3.5

D, c

m2 /

s

1000/T, 1/K

Quad flat package QFP-144 Fig. 2.12 Arrhenius-likediffusion coefficient founddirectly from a QFP package.The least-squares fitcalculations yield D0 =0.14 cm2/s and U = 0.42 eV[35]

2.4 Second Run of Absorption (Re-sorption)

The complex mechanism of moisture absorption suggests that a second run ofmoisture absorption after an absorption/desorption cycle would be even more com-plicated. In order to achieve logically comparable re-sorption curves, the first runof absorption was stopped after 1 and 2 weeks of sorption for thin and thick sam-ples, respectively. This sorption time corresponds to the time needed to reach thefirst plateau in sorption curves and will be denoted as “virtual saturation.” Afterthe first sorption, the EMC samples were removed from the humidity chamber andplaced in two infrared dry ovens at temperatures of 110 and 160◦C to release theirmoisture at a constant temperature. After reaching a “virtual dry state” (the finaldesorbed state in Fig. 2.9a and b), the samples were placed in the same humiditychamber again for the second run of moisture absorption at 85◦C/85% RH. Sincethe samples had undergone an absorption/desorption cycle, they had an initial mois-ture content (residual moisture content at the end of desorption) as described in theprevious section. This initial content was low, because the absorption was stoppedas soon as the samples reached the virtual saturation and was not considered in themass uptake during the re-sorption process. Figure 13a and b shows the compar-isons between the first run of moisture absorption and the second run after bakingat 110 and 160◦C for the thin and thick samples, respectively.

From these curves the following results can be postulated:

• After the desorption at 110◦C, the second absorption curve before virtual satura-tion was found to be almost identical to the first run for both 1 and 2 mm samples.


0

0.05

0.1

0.15

0.2

0.25

0.3

Mas

s up

take

, %



first absorption

2nd absorption after desorption at 110°C


(a)

0

0.05

0.1

0.15

0.2

0.25

0 5 10 15 20 25 30

0 5 10 15 20 25 30 35 40

Mas

s up

take

, %

1st absorption



(b)

Fig. 2.13 Re-sorptionexperiment of bulk samplesafter two baking temperaturesof 110 and 160◦C: (a) thinsamples (thickness 1 mm)and (b) thick samples(thickness 2 mm)

This means the moisture absorption due to the first phase of moisture diffusionis mostly repeatable. The moisture desorption at 110◦C happens below the glasstransition temperature, at which the relaxation of polymer chains and the changeof free volumes are not significant.

• However, the desorption at 160◦C affected the second run of moisture absorp-tion significantly. The rate of moisture absorption at the second run was foundto be higher than that at the first run. The increase in the rate of moisture uptakecan be attributed to the expansion of free volumes in the polymeric materialsdue to the storage of the material, which was baked at 160◦C, well above theglass transition temperature. For the thin samples (Fig. 2.13a) the differencebetween the first and second runs of moisture absorption is much less than thethicker samples (Fig. 2.13b), because of less exposed time to elevated tempera-ture. Consequently, the formation of new free volumes is less, as this is dependenton the time exposure to high temperatures.


The phenomenon of increasing volumes of EMCs and PEMs was also reportedby Teverovsky [34], who observed that after baking, the volume of some PEMsincreased by 0.06–0.27%. In order to evaluate how the bake temperature affectsthe results of the re-sorption, QFP144 packages, which manifested normal volumereduction behavior during baking, were baked at different conditions: 125◦C for96 h, 165◦C for 25 h, and 205◦C for 2 h. After baking at each condition, the sampleswere moisturized at 85◦C, 100% RH for 168 h, and then baking was repeated atthe same temperature as before. The results of these measurements are shown inFig. 2.14 [34].

–0.6

–0.5

–0.4

–0.3

–0.2

–0.1

0

0.1

0.2

0.3

Initial bake 85°C/100%RH/168hr

bake

dM

, (d

V),

%

dV

dm

205°C 2hr

125°C 96hr

165°C 24hr

Fig. 2.14 Effect of bakingcondition on moisture uptakeand volume deviation in aQFP package [35]

The phenomenon of non-reversible moisture content was reported by Lin [35],Shirangi et al. [1, 2], and Xie et al. [36]. However, Teverovsky [9] reported thatbaking resulted in virtually a complete removal of moisture as shown in Fig. 2.14.This may be due to many possible reasons, including short sorption time, lack ofmeasurement accuracy, or type of epoxy used in the molding compounds. However,as shown in Fig. 2.14 the volume variations increased with the bake temperature.The 2-h bake at 205◦C caused more than 0.4% decrease in the volume compared toonly 0.1% after 125◦C bake.

2.5 Hygroscopic Swelling

Water molecules in polymeric materials have been identified to have two distinctstates. A “free” or “unbound” state of water is attributed to water molecules thatare present in voids and nanopores of the material [7, 21, 37] and can easily movethrough the free volumes of polymer. Another state is formed by water molecules,which disrupt inter-chain polymer ties and are called “bound” water molecules.The “bound” water molecules react with the polymer chains via hydrogen bond-ing. It is often believed that the moisture-induced swelling of polymers is due to thehydrogen bonding and the presence of water molecules in free volumes of the poly-mers has a less contribution to the hygroscopic swelling of polymers [19, 38–42].


This identification is further supported by measurement of the ratio of hygroscopicvolume expansion to the volume of absorbed water which is less than unity [38],indicating that some of the absorbed water does not contribute to swelling. Thehygroscopic mismatch strain at the interfaces between molding compounds andmetallic components in a package could be as high as the thermal mismatch strains[7, 38–40].

In order to recognize the impact of the hygroscopic swelling on the overall shapeof plastic packages, the surface topography of a plastic IC package was investigated.The measurements were performed via an FRT MicroProf with a chromatic sensor(CWL). The chromatic sensor illuminates the sample by using a white light sourceand measures the wavelength-dependent (chromatic) distribution of the reflectedlight and determines the absolute height information. Figure 2.15 shows the out-of-plane topography of the top surface of a TQFP-epad package which uses themolding compound MC-1 introduced in the previous sections. The surface topog-raphy of a dry package at room temperature is shown in Fig. 2.15a. The samplewas then placed in a humidity chamber at 85◦C/85% RH for 168 h (MSL1). Afterthe sorption, the sample was removed from the humidity chamber and exposed toambient temperature. Upon cooling to room temperature, the warpage was againmeasured as shown in Fig. 2.15b. By comparing the warpage of a dry and amoisture-preconditioned package, it becomes clear that the hygroscopic swellingof EMCs alters the overall stress balance in plastic packages. The warpage directionof the package changed from a “smiling” shape to “crying” after moisture precon-ditioning. At room temperature, the warpage of a dry package is dominated by twofactors: The first mechanism is the cure shrinkage of epoxy molding compoundsdue to the cross-linking of the polymers during the transfer molding. The secondmechanism is the mismatch between the coefficient of thermals expansion (CTE) ofdifferent materials. When the packages absorb moisture, the hygroscopic swellingof the EMC acts as a third mechanism and affects the warpage direction of the pack-age by introducing a moisture-induced expansion to the epoxy molding compound.

Fig. 2.15 Topography of the upper surface of a TQFP-epad package: (a) dry package and (b) after168 h aging at 85◦C/85% RH


The amount of hygroscopic strain found from the dimensional change is normallyassumed to be linearly proportional to the moisture concentration as follows [1–3,6–8, 38–42]:

εh = βC , (2.8)

where εh is the hygroscopic strain, β (mm3/g) is the coefficient of hygroscopicswelling (CHS), and C (g/mm3) is the moisture concentration.

There has been a long debate and some discrepancy between the different stud-ies over a suitable method in order to find the coefficient of hygroscopic swelling[38–42]. The aim of all these works is to find the hygroscopic swelling strain andto relate it to the local moisture concentration. The next sections of this studyprovide useful information about the direct effect of hygroscopic swelling by mea-suring the deflection of specially designed bi-material beams and comparing theresults with those found from a well-established characterization method, namelythe TMA/TGA approach.

2.5.1 Characterization of CHS by Warpage Measurementof Bi-material Beams

Since the direct impact of the hygroscopic swelling on the plastic packages is affect-ing their warpage, this work investigates the swelling of the EMCs by measuring thedeflection of bi-material beams during moisture absorption. Bi-material beams ofcopper/EMC with two different EMC thicknesses were designed and manufacturedas shown in Fig. 2.16a and d. The samples were used later for the characterization ofinterfacial fracture toughness between the EMC and the copper leadframe. In orderto manufacture the bi-material Cu/EMC beams via transfer molding process, coppersubstrates were first machined into 50 × 10 × 0.4 mm3 strips. After cleaning withacetone the substrates were placed in the cavity of a molding machine. Pellets of acommercial epoxy molding compound (MC-1 in the previous sections) were intro-duced into the cavity of a pre-heated mold at about 175◦C and kept under a pressureof 60 bars for 90 s; the molding compound was dispensed automatically on the cop-per surface at 175◦C. After molding the samples were placed in an infrared chamberfor post-mold curing at 175◦C for 6 h in order to complete the polymerization pro-cess of the epoxy molding compound. Three samples from each type were used togain a statistically acceptable CHS value. Figure 2.16b and e shows the warpageof the beams in the dry state, which arises from the cure shrinkage of EMC duringthe manufacturing process and the CTE mismatch between EMC and copper. Afterthe post-mold process, samples were placed in a humidity chamber at 85◦C/85%RH and were removed periodically from the moisture chamber. Their warpage wasmeasured at room temperature and then they were placed in the humidity chamberfor further sorption. During the moisture absorption, the warpage of the thin sam-ples (with an EMC thickness of 0.6 mm) changed from a concave to a convex shape(see Fig. 2.16b and c). The convex shape increased and reached a constant valueof approximately 185 μm after 1 week of sorption. Further exposure to moisture


EMC

Wdry = +186 µm

hCu = 0.4 mmhCu = 0.4 mm

hEMC = 0.6 mm hEMC = 1 mm

Wdry = +12 µm

Wmoist = –185 µmWmoist = +31 µm

Hygroscopic swelling changesthe stress state

Hygroscopic swelling reducesthe stresses

EMC

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 2.16 Effect of hygroscopic swelling on the warpage of the bi-material beam

did not affect the warpage significantly. The thick samples (with an EMC thicknessof 1.0 mm) had an initial warpage of 186 μm at room temperature in the fully drystate (Fig. 2.16e). During the sorption, their warpage decreased to around 31 μmand remained almost constant after 2 weeks of sorption as shown in Fig. 2.16 f.The warpage values of both samples as a function of exposure time to moisture areshown in Fig. 2.17.

–250

–200

–150

–100

–50

0

50

100

150

200

250

0 1 2 3 4 5

War

page

, µm

Exposure to 85 C/85%RH, weeks

tEMC = 0.6 mm

tEMC = 1.0 mm

Fig. 2.17 Warpage change ofbi-material beams as afunction of exposure time tomoisture at 85◦C/85% RH


The warpage changing of the bi-material beams is due to the hygroscopicswelling across the exposed surface of the epoxy molding compound. However,since the epoxy molding compounds are viscoelastic materials, stress relaxationcan also cause a change in the warpage during the moisture aging at 85◦C/85%RH [42]. This was confirmed by a separate test regarding the aging of samples ina dry condition at 85◦C in order to isolate the effect of the thermal aging and toassess the pure effect of the hygroscopic swelling at 85◦C/85% RH. Nevertheless,in order to calculate the pure hygroscopic swelling strain, one needs the exactamount of the cure and thermal strains. The authors showed in [43] that a simplemodification in the CTE values together with applying a viscoelastic model canaccurately predict the warpage of these bi-material samples. In this work, the samemodel was applied to determine the hygroscopic strain of the epoxy molding com-pound similar to that explained in [42]. The glass transition temperature (Tg) ofthe EMC was measured and found to be 109◦C and the CTE1 and CTE2 weremeasured 8.9 and 29.6 ppm/K, respectively, as reported in [43]. The cure shrink-age during the molding process of the samples was found by warpage analysisas well. The sample history was modeled with the FE program ANSYS as fol-lows. First, the cure shrinkage at the molding temperature was accounted for byapplying the cure strain to the molding compound. Afterward the thermal strainsarising from the cooling process from mold temperature (175◦C) to room temper-ature (25◦C) were taken into account. Finally, the hygroscopic swelling and thestress relaxation during the aging at 85◦C/85% RH were modeled. The hygroscopicstrain was obtained from the finite element calculation to be almost 0.061%. Themoisture content of the EMC was measured in [1, 2] to be between 4.45E–6 and6.7E–6 g/mm3, depending on the sorption history. Consequently, by using equation(2.8), the coefficient of hygroscopic swelling can be estimated at an average valueof 134 mm3/g.

Some samples were removed from the moisture chamber after the saturation andwere placed in a dry infrared oven at 110◦C. The warpage of the samples was mea-sured after 10 days of baking. This was done to investigate whether the hygroscopicswelling is reversible. As was expected, not all the moisture-induced warpage dueto moisture absorption was recovered. This suggests that the measurement of thehygroscopic swelling by any method that deals with the dimensional change dur-ing the desorption of the sample should be avoided, at least for the polymericmaterials which show such non-reversible hygroscopic swelling. In the next sec-tion another method based on the dimensional measurement of EMCs during thebaking will be investigated and the results will be compared to that from the abovemethod.

2.5.2 Characterization of CHS by TMA/TGA

Hygroscopic swelling of EMCs has been usually investigated by performing twoparallel analyses. In the first one the weight loss of a saturated sample during


desorption at a constant temperature is measured. This can be done via a TGA (ther-mal gravimetric analyzer) or running a finite element simulation of the desorptionprocess of the sample [1]. In the other one the dimensional changes of the sam-ple during desorption are measured. Usually a TMA (thermal mechanical analyzer)measurement [1, 6, 41, 42, 44] or shadow Moiré interferometry [3, 7, 40] is used forthe in situ measurement of the hygroscopic strains. In all these methods, the dimen-sional changes of the sample during the isothermal desorption are considered andit is assumed that the behavior of the polymeric materials in terms of swelling andshrinkage during the absorption and desorption is the same. However, the problemof this method is that, similar to the residual moisture content upon desorption ofthe molding compounds reported previously, a residual hygroscopic strain may alsoexist. In other words, not all the swelling during the moisture uptake may be recov-ered after desorption. This may cause a wrong correlation between the measuredstrains and the moisture concentrations.

In order to determine the CHS of material MC-1, initially moisture-precon-ditioned (up to virtual saturation) samples of 8 mm length were placed in the TMAfor the strain measurement at different temperatures. The average elastic strain alongthe measuring line was reported by Zhou et al. to be close to zero [41]. The pres-ence of elastic strain due to hygroscopic stress contributes a relatively small analysiserror in the determination of the coefficient of hygroscopic swelling [44], comparedto the error caused by non-uniform hygroscopic swelling deformation. To eliminatethe role of thermal expansion during desorption with TMA, a dry sample was usedas reference so that only hygroscopic strains were documented.

For the TGA measurement with the available equipments, samples must have amaximum length of 2 mm to fit to the equipment, which makes it difficult to couplethe moisture concentration results from TGA with strains from TMA because ofdifferent sample geometries. As an alternative for TGA, a finite element analysis ofthe desorption process was performed and used for the mass loss calculation usingthe desorption data from Table 2.3. Figure 2.18 shows the experimental results ofTMA and the corresponding TGA simulations at three temperatures.

From each pair of isotherm TMA/TGA curves of Fig. 2.18, a CHS can beobtained by fitting a linear line and calculating its slope as depicted in Fig. 2.19. Thecurves do not cross the origin due to the inaccuracy of the TMA at longer stages ofdesorption. The results show a smooth increase of CHS at higher temperatures dueto vapor pressure-induced expansion of the EMC at elevated temperatures.

The CHS results found from the TMA/TGA method are in the same order ofmagnitude as the value determined using the warpage analysis as compared inTable 2.4. However, to the authors’ best knowledge, the warpage analysis shouldbe preferred to estimate the CHS value. This is because of the fact that the hygro-scopic swelling develops during the moisture preconditioning and its direct effectis changing the beam warpage. Consequently, the method can be benchmarked byapplying the CHS value in the FE analysis and comparing the simulated warpageto the experimental one. This enables the verification of the whole stress analysis,which includes cure, thermal, and hygroscopic strains.


0 1000 2000 3000 40002

4

6

8x 10−4

stra

in

time (min)

TMA, T = 110 °C

TMA, T = 160° C

0 1000 2000 3000 40002

3

4

5

6x 10−6

C, g

r/m

m3

C, g

r/m

m3

C, g

r/m

m3

time (min)

0 200 400 600 800 1000 12001

1.5

2

2.5x 10−3

stra

in

time (min)

0 200 400 600 800 1000 12000

2

4

6x 10−6

time (min)

0 50 100 150 200 250 3003

3.5

4

4.5

5x 10−3

stra

in

time (min)0 50 100 150 200 250 300

0

2

4

6x 10−6

time (min)

TMA, T = 220° C

TGA, T = 110° C

TGA, T = 160° C

TGA, T = 220° C

Fig. 2.18 Results of TMA analysis at three temperatures (left) together with results of TGA fromsimulation at corresponding temperatures (right)

It must be mentioned that the results from Fig. 2.19 show that the CHS increaseswith increasing baking temperature. This is a problem of estimating the CHS usingthe TMA/TGA approach, as during the strain measurements by TMA, especially atelevated temperatures; the expansion of the EMC due to the high vapor pressure isalso documented. Consequently, the results at lower temperatures seem to be closerto the real hygroscopic expansion during the sorption at 85◦C/85% RH.

2.5.3 Characterization of CHS by Archimedes Principle

Teverovsky [34] investigated the hygro-thermal expansion of two commercial mold-ing compounds. The volume measurements of molding compounds were done byArchimedes principle based on the weight measurements of the parts in air and thenafter immersion into a fluid (Galden D02), using the density of the liquid. The vol-ume and weight of the sample were measured two times: after saturation in moistureand after baking at high temperature and the CHS was then found from the followingformula:


2 2.5 3 3.5 4 4.5 5 5.5 6 6.52

4

6

8S

tra

in

0 1 2 3 4 5 61

1.5

2

2.5

Str

ain

0 1 2 3 4 5 6

x 10−6

x 10−6

x 10−6

3

4

5

x 10−4

x 10−3

x 10−3

Str

ain

Moisture concentration (C), (gr/mm3)



Strain versus C at T=110°C



CHS=129 mm3/g

CHS=146 mm3/g

CHS=168 mm3/g

Fig. 2.19 Calculation of CHS by combining the results of TMA and TGA

Table 2.4 CHS results from different approaches

Method CHS (mm3/g)

TMA/TGA at 110◦C 129TMA/TGA at 160◦C 146TMA/TGA at 220◦C 168Warpage analysis of a bi-material beam 134

CHS = 1

3

Vmoist − Vbake

Mmoist − Mbake

Mbake

Vbake.

The results of CHS at different temperatures and humidities are shown inFig. 2.20. It can be observed that the hygro-thermal expansion is not a constant,but depends on moisturizing conditions and has a trend of increasing with moistureuptake.

The results in Fig. 2.20 reveal that moisture uptake is virtually a linear functionof the relative humidity for the tested molding compounds.


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7dM

/M, %

RH, %

MG70F-0627

MG33F-0520

00.10.20.30.40.50.60.70.80.9

1

dV/V

, %

RH, %

MG70F-0627

MG33F-0520

00.10.20.30.40.50.60.70.80.9

1

0 20 40 60 80 100

0 20 40 60 80 1000 20 40 60 80 100C

ME

RH, %

MG70F-0627

MG33F-0520

(a) (b)

(c)

Fig. 2.20 Isotherms of the equilibrium moisture uptake: (a) volume swelling and (b) coefficientof moisture expansion (CME) for two epoxy molding compounds [34]

2.6 Moisture-Induced Adhesion Degradation

Water is often regarded as the main agent in reducing the service life and reliabilityof adhesive joints, electronic devices, and composite materials. The mechanism ofadhesion loss at a critical relative humidity is the subject of much speculation [21,45]. The dramatic reduction in adhesion strength has been attributed to both physi-cal and chemical changes resulting from the moisture absorption either in the bulkadhesive or at the interface between the adhesive and substrate.

Moisture can influence the interfacial adhesion through three mechanisms. Thefirst mechanism is the intrinsic aggregating effect of water molecules upon directpresence at the interfaces and degrading the interfacial adhesion by bonding tothe polymer chains [46–48]. The second mechanism is that the absorbed moisturechanges the mechanical properties of polymeric materials. For example, Dudek et al.[47] and Ferguson et al. [49, 48] reported that moisture can change the elastic mod-ulus and shift the glass transition temperature of polymers to lower values. Thismechanism leads normally to a slight difference in the mode mixity of the measuredfracture toughness of moist sample when compared to that of dry ones. The thirdmechanism is the swelling of polymeric materials upon exposure to moist environ-ments, thereby causing an additional mismatch between volumetric expansions ofsubstrate and adhesives. This is even more pronounced when the joint between apolymer and metal is investigated. Since the metallic substrate is impermeable to


moisture, only the polymeric adhesive absorbs moisture and causes a mismatchin hygroscopic strains. In order to measure the intrinsic fracture toughness of amoisture-preconditioned bi-material sample, the influence of hygroscopic swellingwhich induces an apparent change in the measured fracture toughness should beisolated.

Brewis et al. [50] proposed that degradation of the interface is the basic causefor adhesion loss. They suggested that bonds attributable to secondary forces atthe interface are broken up by moisture, hydration of the oxide surfaces, and/orwater condensation due to the lowering of the vapor pressure by the presence of saltimpurities at the interface. In contrast to their results, Lefebvre et al. [51] suggestedthat the change in the bulk properties of the epoxy adhesive is responsible for theobserved onset of adhesion loss at a critical relative humidity. They reported that thesolubility, or mass of absorbed water, as a function of relative humidity increasedabruptly at the critical relative humidity. Their study suggests that capillary conden-sation and the depression of glass transition are not responsible for the adhesion loss.They proposed that inter-chain hydrogen bonds are broken by absorption of water.However, using nuclear magnetic resonance, McBrierty et al. [52] opposed thetheory that inter-chain hydrogen bonds are broken by absorption of water. Theyshowed in epoxies based on the diglycidyl ether of bisphenol A (DGEBA) thatno disruption of the hydrogen bonding network occurs at room temperature in thehydrated epoxy resin.

Ferguson [49, 48] suggested that there are three mechanisms that contributeto water penetration at the interface in epoxy adhesive structures: bulk diffusion,wicking along the interface, and capillary action associated with micro-cracking.The first mechanism was discussed in the last sections. The second mechanism formoisture transport to interface is attributed to wicking along the interface and wasdiscussed in previous sections as interfacial diffusion. Comyn et al. [53] found thatthe rate of wicking of glass-to-lead alloy joints bonded with an epoxy adhesivecould not be accounted for by the rate at which water enters the epoxy adhesiveby bulk diffusion alone. They concluded that water must also enter the interfaceby “wicking” along debonded zones along the interface. The final mechanism formoisture transport to the interface is by capillary action associated with voids andcracks present in the epoxy or epoxy composite. Lu et al. [54] found that the addi-tion of fillers to polymers resulted in faster sorption kinetics when compared tothe bulk polymer alone. They concluded that water was absorbed not only by theepoxy but also by the interfaces inside the epoxy introduced by the addition offillers.

The intrinsic effect of moisture on the interfacial fracture toughness of cop-per/EMC interface will be investigated in this section using a combined experi-mental and numerical method. The end-notched flexure (ENF) test [54] has beenwidely used to characterize the mode II fracture toughness [2, 42, 43, 55, 56]. Atypical ENF testing setup for the bi-material interface is essentially a three-pointbending test with an initial crack of length a at one end as shown in Fig. 2.21.The bi-material samples introduced in the previous sections were used for the char-acterization of interfacial fracture toughness between copper leadframe and epoxy


Fig. 2.21 End-notchedflexure (ENF) test for themeasurement of interfacialfracture toughness

molding compound MC-1. Since the upper layer undergoes a larger deflection, thesamples were positioned with the copper side facing up.

A precrack was generated by coating the desired area of the leadframe witha special spray to prevent it from adhering to the leadframe during the moldingprocess. Later a razor blade was introduced at the end of the beam at the interfaceto facilitate precracking. After the delamination test, there was a visible trace on theregion of the copper leadframe which was precracked and the precrack length wasmeasured by using a ruler. Furthermore, the precrack length was verified for a fewsamples by C-mode scanning acoustic microscopy.

A typical load–displacement curve of an ENF test is shown in Fig. 2.22. Forperforming each test a specimen was placed on the lower supports. Then the upper

0,0 0,2 0,4 0,6 0,8 1,00

4

8

12

16

20Point B: Initiation of

interfacial delamination

upon reaching the

critical force B

A

For

ce (

N)

Displacement (mm)

ENF at 25° C for precrack length of a = 11.8 mm

C

Point C: Arrest of the crackPropagation.

Point A: Automatic start of displacement

documentation and imaging after reaching

a preload of 1N

Fig. 2.22 Load–displacement curve of an ENF fracture test


support was moved down manually and positioned just above the specimen. Nextthe loading was initiated and the displacement was read off upon reaching a preloadof 1 N (point A in Fig. 2.22).

At the early stages of applying load, an initial linear relation between force anddisplacement can be observed. The slope of the curve at this stage represents thebending stiffness of the composite EMC/Cu structure. The peak of the load pro-file (point B in Fig. 22) corresponds to the critical force and the initiation of theinterfacial delamination. This force was used later as the critical force in the finiteelement analysis. After reaching the peak load the delamination propagates until itstops at a point (point C) which is usually in the middle of the beam. During crackpropagation (from point B to point C) the compliance of the beam increases, whichmeans that the energy release rate of the ENF test is dependent on the crack length.After the crack stops at point C, the load increases again linearly. After point C,the slope of the graph represents the stiffness of the EMC only. The fracture loadat the onset of crack propagation where the load suddenly drops (point B) was usedfor the measurement of the interfacial fracture toughness via finite element analysis.For each test condition, between four and eight specimens were used, and each peakload with its corresponding precrack length was used for a separate finite elementcalculation. In order to prevent large deformations, the span between two lowersupports was fixed at 34 mm.

Based on the fracture curves obtained from the ENF tests, finite element anal-yses were carried out to determine the resultant interfacial fracture toughness. A3D finite element model was built with eight-node elements using the FEM toolANSYS. For the analysis of the ENF model, contact elements without friction wereused in cracked surfaces to avoid element penetrations. The residual stresses dur-ing the manufacturing process of the samples were taken into account as follows:First the chemical cure shrinkage during the molding process was introduced in theFEA model as explained by the authors in [43]. Then the cooling process from themolding temperature (175◦C) to room temperature (25◦C) was taken into account.Finally the introduction of a precrack, which leads to some stress relaxations in thesample, was modeled by decoupling the selected nodes at the interface.

The virtual crack closure technique (VCCT) has successfully been used to obtainthe total strain energy release rate (SERR) and the mode mixity for both homoge-nous and interface cracks based on the results of finite element analysis. VCCTrequires only one complete analysis of the structure to obtain the deformations.The method yields the total energy release rate in the direction in which the crackis extended virtually. A macro was written for the FEM tool ANSYS that allows tofind all three components of the SERR. Since VCCT is only applicable in the case oflinear elastic fracture mechanics (LEFM), it is questionable if this approach wouldbe suitable for the interface crack between the elastic leadframe and the viscoelasticEMC. The authors justified the use of VCCT for the Cu/EMC samples in [43] andsuggested that the VCCT is applicable as far as the fracture tests are performed attemperatures below the glass transition temperature of EMCs.

Bi-material samples similar to those shown in Fig. 2.16 were used for the calcu-lation of interfacial fracture toughness. At first samples with an EMC thickness of


0

20

40

60

80

100

120

0 1 2 3 4 5 6 7N

orm

aliz

ed i

nter

faci

al

frac

ture

toug

hnes

s, %

Exposure time to 85°C/85%RH, weeks

Fig. 2.23 Adhesiondegradation of EMC/Cuinterface as a function ofexposure time to moisture

2 mm were produced. The average interfacial fracture toughness was normalized tothe value of the dry samples at room temperature. The adhesion loss upon aging inhumid environment was compared with the control value at dry state. Except for thesamples measured at dry state, all samples were placed in the moisture chamber andfive samples were removed after each week. After reaching the ambient tempera-ture, a precrack was introduced in the samples and the fracture test was performedfor all five samples. By running an FE simulation of the sample sorption, it wasexpected that the adhesion is not affected during the first 2 weeks of aging and itsmaximum decrease takes place after around 6–8 weeks of sorption. However, asshown in Fig. 2.23, it was observed that most of the adhesion loss happened imme-diately within the first week of moisture sorption. The fracture tests reveal that after1 week of sorption the highest decrease in interfacial adhesion occurred, followedby a small decrease during the second week of sorption. The interfacial adhesionreached a constant value after approximately 2 weeks of exposure to moisture. Thissurprising result of rapid adhesion degradation can be attributed to the high rate ofmoisture diffusion along the interface between the EMC and the leadframe. Theseresults can be justified by the sorption comparisons in Fig. 2.2a and b, which showeda higher diffusion rates across the interfaces.

Since the bi-material samples with a high EMC thickness of 2 mm and an openinterfacial diffusion path showed such anomalous results of rapid adhesion loss, newsamples with thinner EMC thickness of 0.6 mm were used for the later experiments.In the design of the mold cavity some changes were also introduced so that themolding compound covered both sides of the bi-material sample acting as a barrierfor interfacial diffusion. This mold flake was later removed by a grinding machinejust before the fracture tests. Figure 2.24 shows the intrinsic effect of aging in humidenvironment on the interfacial fracture toughness of the Cu/EMC interface. All frac-ture tests were conducted at room temperature with a displacement-controlled rateof 0.1 mm/min. In addition, the effect of moisture on the elastic modulus of theEMC was investigated using three-point bending of bulk EMC bars. It was found


58.7

28.8

44.9

26.3

0

10

20

30

40

50

60

70

80

Fra

ctur

e to

ughn

ess

J/m

2 dried after virtual saturation

dried after 2nd phase absorption

virtualsaturated

Dry

Fig. 2.24 Effect of moisture absorption and desorption on the adhesion of Cu/EMC interface

that moisture does not affect the elastic modulus significantly. Hence, all the frac-ture toughness measurement can be assumed to be at a constant mode angle as hasalso been suggested by Ferguson [48].

Using the diffusion coefficient from the experimental results the time to virtualsaturation of the samples was found numerically to be almost 2 weeks (0.21%weight gain of the whole sample). Fracture results show that the interfacial frac-ture toughness initially at 58.7 J/m2 in dry condition reduced to 26.3 J/m2 when thevirtual saturation level at the interface was reached.

In order to study the effect of an absorption/desorption cycle on the adhesion,moist samples were baked for 24 h at 125◦C reaching a virtually dry state. A longerexposure to moisture was also investigated by 4 weeks of aging in humid condi-tion (0.25% weight gain of the sample) and subsequent baking for 24 h at 125◦C.The fracture toughness of these samples was measured again at room temperatures.Some of the adhesion loss due to moisture absorption up to virtual saturation wasrecovered after drying (44.9 J/m2). However, samples that remained longer in humidconditions (4 weeks) showed almost no recoverability upon the same annealing con-dition (28.8 J/m2). This is an important result, which shows the extreme degradingeffect of the second phase of moisture absorption. The moisture absorption duringthe first phase degrades the adhesion due to the intrinsic effect of the presence ofwater molecules at the interface, which is partially reversible if a proper annealingis performed. However, the second phase of moisture absorption seems to destroythe adhesion bonds permanently, and none of the adhesion loss may be recovered ifthe moisture level at the interface reaches this critical content.

These results are in agreement with the results of Dodiuk et al. [57] who eval-uated the effect of moisture on the lap shear strength of four commercial epoxy


adhesives to aluminum. They found that the exposure of moisture caused a reduc-tion in lap shear strength; however, if the moisture concentration was below 0.3%,the strength was fully reversible after drying, indicating that adhesion loss may berecovered if the moisture content at the interface is still low. They gave no explana-tion for their results; however, the sorption results depicted in Figs. 2.5 and 2.9together with the fracture results of Fig. 2.24 can be helpful to understand themoisture-induced degrading effects. The key to understanding the adhesion lossand its recoverability or non-recoverability is the second phase of moisture absorp-tion, which causes a permanent degradation in adhesion by destroying the secondarybonds between polymer and substrate.

There are some researchers that observed a critical moisture content for the adhe-sion degradation. It has been observed that a critical concentration of water mayexist where there may be a concentration and associated humidity level below whichthe interface is not weakened. Kinloch [58] found that epoxy/mild-steel joints suf-fered no loss in adhesion from environment attack at 50% RH, even though theadhesive still absorbed water up to an equilibrium concentration. As a direct conse-quence of this observation, Kinloch proposed that a minimum, critical concentrationof water must be a requirement for the loss of adhesion due to the presence ofmoisture. Ferguson and Qu [49, 48] used a water-proof perimeter to the bi-materialtest specimens before moisture preconditioning to force one-dimensional moistureuptake through the top surface of the test specimen and prevent wicking alongthe interface. This can enable the assumption of uniform concentration at interfaceby utilizing the inherent moisture absorption characteristics of the adhesive. Theyobserved that a large portion of the loss in elastic modulus from moisture uptakewas recovered upon subsequent drying. They also observed a permanent weightincrease in epoxy samples after a subsequent baking which suggests that at leastpart of the irreversible damage resulted from hydrolysis with a greater extent occur-ring at higher humidity levels. When they investigated the adhesion under moisture,they observed significant reduction in interfacial adhesion even for low concentra-tions and concluded that the loss in interfacial fracture toughness from moisture wasnot recovered upon fully drying. They explained the permanent loss in adhesionwith adsorption theory as a primary bonding mechanism for the underfill/copperinterface.

Contrary to some results regarding the permanent adhesion loss, there are someevidences in the literature that report a full recovery in adhesion upon a subsequentdrying. The reversibility of the adhesion loss was reported for organosilicate glass(OSG) film by Lin et al. [59]. Also Shaw et al. [60] found that nearly all of thestrength lost after immersing steel/epoxy lap shear joints in distilled water for 3weeks was recovered after drying. They attributed the loss in strength after moisturepreconditioning to plasticization of the epoxy adhesive, which is generally regardedas a reversible process.

The discrepancy between the reported results in the literature may be due todifferent mechanisms available for the adhesion loss. The important result from themoisture diffusion experiments is that the adhesion tests gave reasonable predictivevalues, in agreement with the observations in moisture absorption and desorption


behavior. These realistic test conditions can enable a correlation to the response ofmicroelectronic packages to humid conditions.

2.7 Conclusion

In this study a systematic investigation of absorption and desorption of moisture inepoxy molding compounds was conducted. Absorption of moisture was found toshow a dual-stage non-Fickian behavior. The exposure of an EMC sample upon avirtual saturation (end of the first absorption phase) to a dry environment was foundto lead to an almost dry state with only slight residual moisture content at the end.However, a dry state was not achieved when the samples with higher initial moisturecontent (which were kept in humid environment for a longer time) were baked indry conditions. A residual moisture content was present which was a complex func-tion of time exposed to moisture, sample geometry, and baking temperature. Theschematic picture of the influence of sample history may be depicted as shown inFig. 2.25. Samples which reached point A (virtual saturation) show a lower residualmoisture content upon baking when compared to the samples reached point B (sec-ond absorption phase). However, the rate of desorption for both cases was the same,indicating that at least two mechanisms are active during the diffusion of mois-ture. One is a reversible mechanism that dominates the diffusion rate. The otheris a non-reversible mechanism that is a function of time, temperature, and samplegeometry.

B

A

Cres

Cres

Desorption after

reaching point A

Moisture uptake curve

Moi

stur

eC

onte

nt Desorption after

reaching point B

)hour(,time

Fig. 2.25 Schematic modelfor the residual moisturecontent upon desorption ofmoisture

The second run of moisture absorption showed also some differences from thefirst run. The sample sorption history was found to be the dominating factor. The rateof moisture absorption at the second run was found to be higher than that at the firstrun. The increase in the rate of moisture uptake can be attributed to the formation ofnew voids in the polymeric materials, which facilitates the transformation of watermolecules in the sample. Higher temperatures lead to the formation of more newfree volumes. A schematic picture of the effect of sample history on the rate of thesecond run of moisture absorption may be depicted like the one in Fig. 2.26.


1st run of absorption at 85°C/85%RH

2nd run of absorption at 85° C/85% RH

Moi

stur

e C

onte

nt

)hour(time ,

Fig. 2.26 Schematic modelof the second run of themoisture absorption after anabsorption/desorption cycle

The coefficient of hygroscopic swelling (CHS) is normally found by relatingthe dimensional changes of a saturated bulk sample to its mass loss during theisothermal desorption. However, experiments showed that the swelling of EMCsupon moisture intake was not recovered completely after their moisture desorption.This makes a correlation between the actual moisture content and the dimensionalchanges during the baking process very difficult. Consequently, other methods basedon the absorption process should be used for estimating the coefficient of hygro-scopic swelling. Warpage measurement of a bi-material beam is a better approach;however, the significant stress relaxation during the aging must be considered inthe simulation methods. A detailed analysis of the warpage via FE analysis whichconsiders the effect of cure shrinkage, stress relaxation due to viscoelasticity, andmoisture-induced swelling was performed. This method allows for the calculationof the CHS during the moisture absorption of the EMC materials.

In this study we also demonstrated that the exposure of a Cu/EMC bi-materialbeam to moisture prior to fracture tests results in a degradation of the adhesion. Thisdegradation is the result of the diffusion of water in the interface. For samples thatwere aged shortly in a humid environment the degradation was partially reversibleby applying an appropriate heat treatment at mild annealing conditions. However,long-term aging in humid condition caused a permanent adhesion loss, which wasattributed to the effect of hydrogen bonding between water molecules and polymerchains at the interface.

References

1. Shirangi, M.H., Auersperg, J., Koyuncu, M., Walter, H., Müller, W.H., Michel, B.,“Characterization of dual-stage moisture diffusion, residual moisture content and hygroscopicswelling of epoxy molding compounds,” Proceedings of the 9th EuroSime2008, Freiburg,Germany, pp. 455–462, 2008.

2. Shirangi, M.H., Fan, X.J., Michel, B., “Mechanism of moisture diffusion, hygroscopicswelling and adhesion degradation in epoxy molding compounds,” Proceedings of the 41stInternational Symposium on Microelectronics (IMAPS), Providence, USA, pp. 1082–1089,2008.

3. Fan, X.J., Zhou, J., Chandra, A., “Package structural integrity analysis considering moisture,”Proceedings of Electronic Components and Technology Conference (ECTC), Orlando, USA,pp. 1054–1066, 2008.


4. Pecht, M.G., Nguyen, L.T., Hakim, E.B., Plastic-Encapsulated Microelectronics. New York,NY: Wiley, 1995.

5. Ardebili, H., Hillman, C., Natishan, M.A.E., McCluskey, P., Pecht, M.G., Peterson, D., “Acomparison of the theory of moisture diffusion in plastic encapsulated microelectronics withmoisture sensor chip and weight-gain measurements,” IEEE Transaction on Components andPackaging Technologies, 25, 132–139, 2002.

6. Fan, X.J., Lee, S.W.R., Han, Q., “Experimental investigations and model study of moisturebehaviors in polymeric materials”, Microelectronics Reliability, 49, 861–871, 2009.

7. Fan, X.J., “Mechanics of moisture for polymers: fundamental concepts and model study,”Proceedings of the 9th EuroSime, Freiburg, Germany, pp. 159–172, 2008.

8. Shirangi, M.H., Müller, W.H., Michel, B., “Determination of Copper/EMC interface fracturetoughness during manufacturing, moisture preconditioning and solder reflow process of semi-conductor packages,” Proceedings of International Conference on Fracture (ICF12), Ottawa,Canada, 2009.

9. Teverovsky, A., Moisture Characteristics of Molding Compounds in PEMs, Lanham, MD:QSS Group, Inc./Goddard Operations.

10. Soles, C., Yee A, “A discussion of the molecular mechanisms of moisture transport in epoxyresins,” Journal of Polymer Science, Part B: Polymer Physics, 38, 792–802, 2000.

11. Adamson, M.J., “Thermal expansion and swelling of cured epoxy resin used in graphite/epoxycomposite materials,” Journal of Material Science, 15, 1736–1745, 1980.

12. Tencer, M.,“Moisture ingress into nonhermetic enclosures and packages – a quasisteady statemodel for diffusion and attenuation of ambient humidity variations,” IEEE 44th ElectronicComponents Technology Conference, Washington DC, USA, 1994.

13. Wong, E.H., Teo, Y.C., Lim, T.B., “Moisture diffusion and vapor pressure modeling of ICpackaging,” Proceedings of Electronic Components Technology Conference (ECTC), Seattle,USA, pp. 1372–1378, 1998.

14. Xie, B., Fan, X.J., Shi, X.Q., Ding, H., “Direct concentration approach of moisture dif-fusion and whole field vapor pressure modeling for reflow process: part I – theory andimplementation”, ASME Journal of Electronic Packaging, 31(3), 031010, 2009.

15. Vine, K., Cawley, P., Kinloch, A.J., “The correlation of non-destructive measurements andtoughness changes in adhesive joints during environmental attack,” Journal of Adhesion, 77,125–161, 2001.

16. Davis, D., Krebs, A., Drzal, L.T., Rich, M.J., Askeland, P., “Electrochemical sensors fornondestructive evaluation of adhesive bonds,” Journal of Adhesion, 72, 335–358, 2000.

17. Zanni-Defarges, M.P., Shanaham, M.E.R., “Evaluation of adhesive shear modulus in a tor-sional joint: influence of aging,” International Journal of Adhesion and Adhesives, 13, 41–45,1993.

18. Chan, E., Yuen, M., “Study of interfacial moisture diffusion at Cu/Epoxy interface by FTIR-MIR technique,” Proceedings of Electronic and Component and Technology Conference,Reno, Nevada; USA, pp. 1782–1787, 2007.

19. Fan, X.J., “Moisture related reliability in electronic packaging”, Electronic ComponentTechnology Conference (ECTC), Short Course Notes; 2008.

20. Buchwalter, L.P., “Polyimides: fundamental aspects and applications,” Adhesion ofPolyimides to Various Substrates. New York, NY: Marcel Dekker, pp. 587–628, 1996.

21. O’Brien, E. P., “Durability of adhesive joints subjected to environmental stress,” Dissertationfor Doctor of Philosophy. Blacksburg, VA: Virginia Polytechnic Institute, 2003.

22. Wu, W.L., Orts, W.J., Majkzak, C.J. “Water absorption at a polyimide/silicon wafer interface,”Polymer Engineering & Science, 12, 1000–1004, 1995.

23. Nguyen, T., Byrd, B., Alsheh, D., McDonough, W., Seiler, J., “Interfacial water and adhesionloss of polymer coatings on a siliceous substrate,” Materials Research Society, 385, 57–63,1995.

24. Bowden, F.P., Throssell, W.R., “Adsorption of water vapour on solid surfaces,” Nature, 167,601–602 April 1951.


25. Takahashi, M.K., “AC impedance measurements of moisture in interfaces between epoxy andoxidized silicon,” Journal of Applied Physics, 67, 3419, 1990.

26. Shen, C.H., Springer, G.S., “Moisture absorption and desorption of composite materials,”Journal of Composite Materials, 10, 2–20, 1976.

27. Chen, X., Zhao, S., “Moisture absorption and diffusion characterization of molding com-pound,” Journal of Electronic Packaging, 127, 460–465, 2005.

28. Celik, E., Guven, I., Madenci, E., “Experimental and numerical characterization of non-Fickian moisture diffusion in electronic packages,” Proceedings of Electronic Componentsand Technology Conference, Reno, Nevada; USA, pp. 1069–1073, 2007.

29. Weitsman, Y.J., “Anomalous fluid sorption in polymeric composites and its relation to fluidinduces damage,” Journal of Composites Part A: Applied Science and Manufacturing, 37,617–623, 2006.

30. Loh, W.K., Crocombe, A.D., Abdel Wahab, M.M., Aschroft, I.A., “Modelling anomalousmoisture uptake, swelling and thermal characteristics of a rubber toughened epoxy adhesive,”International Journal of Adhesion & Adhesives, 25, 1–12, 2005.

31. Lekatou, A., Faidi, S.E., Ghidaoui, D., Lyon, S.B., Newman, R.C., “Effect of water and itsactivity on transport properties of glass/epoxy particulate composites,” Journal of CompositesPart A: Applied Science and Manufacturing, 28, 223–236, 1997.

32. He, Y., Fan, X.J., “In-situ characterization of moisture absorption and desorption in thin BTcore substrate,” Proceedings of Electronic Components and Technology Conference, Reno,Nevada; USA, pp. 1375–1383, 2007.

33. Teverovsky, A., “A rapid technique for moisture diffusion characterization of moldingcompounds in PEMs,” NEPP Report, GSFC, 2002.

34. Teverovsky, A., Moisture Characteristics of Molding Compounds in PEMs. Lanham, MD:QSS Group, Inc./Goddard Operations, NASA Technical Report, 2002.

35. Lin, Y.C., “Investigation of the moisture-desorption characteristics of epoxy resin,” Journalof Polymer Research, 13, 369–374, 2006.

36. Xie, B., Fan, X.J., Shi, X.Q., Han, D., “Direct concentration approach of moisture diffusionand whole field vapor pressure modeling for reflow process: part II – application to 3-D ultra-thin stacked-die chip scale packages,” ASME Journal of Electronic Packaging, 31(3), 031011,2009.

37. Fan, X.J., Zhou, J., Zhang, G.Q., Ernst, L.J., “A micromechanics based vapor pressuremodel in electronic packages,” ASME Journal of Electronic Packaging, 127(3), 262–267,12005.

38. Ardebili, H., Wong, E.H., Pecht, M., “Hygroscopic swelling and sorption characteristics ofepoxy molding compounds used in electronic packaging,” IEEE Transactions on Componentsand Packaging Technologies, 26, 206–214, 2003.

39. Shi, X., Zhang, Y., Zhou, W., Fan, X.J., “Effect of hygrothermal aging on interfacial relia-bility of silicon/underfill/FR-4 assembly,” IEEE Transactions on Components and PackagingTechnologies, 31, 94–103, 2008.

40. Stellrecht, E., Han, B., Pecht, M.G., “Characterization of hygroscopic swelling of moldcompounds and plastic packages,” IEEE Transactions on Components and PackagingTechnologies, 27, 499–506, 2004.

41. Zhou, J., Lahoti, S., Kallolimath, K., “Investigation of non-uniform moisture distribution ondetermination of hygroscopic swelling coefficient and finite element modelling for a flip chippackage,” Proceedings of EuroSimE, Berlin, Germany, 2005.

42. Shirangi, M.H., Müller, W.H., Michel, B., “Effect of nonlinear hygro-thermal and residualstresses on interfacial fracture in plastic IC packages”, Proceedings of the 59th ElectronicComponents and Technology Conference (ECTC), San Diego, CA, USA, pp. 232–238, 2009.

43. Shirangi, M.H., Wunderle, B., Wittler, O., Walter, H., Michel, B., “Modeling cure shrink-age and viscoelasticity to enhance the numerical methods for predicting delamination insemiconductor packages,” Proceedings of the 10th EuroSime2009, Delft, The Netherlands,2009.


44. Zhou, J., “Sequentially-coupled finite element transient analysis with hygroscopic swelling,”Proceedings of the 7th EuroSimE, Como, Italy, 2006.

45. Fan, X.J., Zhou, J., Zhang, G.Q., “Multi-physics modeling in virtual prototyping of elec-tronic packages – combined thermal, thermo-mechanical and vapor pressure modeling,”Microelectronics Reliability, 44, 1967–1976, 2004.

46. Fan, X.J., Zhang, G.Q., van Driel, W.D., Ernst, L.J., “Interfacial delamination mecha-nisms during reflow with moisture preconditioning,” IEEE Transactions on Components andPackaging Technologies, 31(2), 252–259, 2008.

47. Dudek, R., Walter, H., Michel, B., “Studies on moisture diffusion and popcorn cracking,”Proceedings of the 3rd EuroSimE, Paris, France, pp. 225–232, 2002.

48. Ferguson, T., “Moisture and interfacial adhesion in microelectronic assemblies,” Dissertationfor the Degree of Doctor of Philosophy. Atlanta, GA: Georgia Institute of Technology, June2004.

49. Ferguson, T., Qu, J., “Elastic modulus variation due to moisture absorption and permanentchanges upon redrying in an epoxy based underfill,” IEEE Transactions on Components andPackaging Technologies, 29, 105–111, 2005.

50. Brewis, D.M., Comyn, J., Raval, A.K., Kinloch, A.J., “The effect of humidity on the durabilityof aluminum-epoxide joints,” International Journal of Adhesion and Adhesive, 10, 247–253,1990.

51. Lefebvre, D.R., Elliker, P.R., Takahashi, K.M., Raju, V.R., Kaplan, M.L., “The critical humid-ity effect in the adhesion of epoxy to glass: role of hydrogen bonding,” Journal of AdhesionScience and Technology, 14(7), 925–937, 2000.

52. McBrierty, V.J., Martin, S.J., Karasz, F.E., “Understanding hydrated polymers: the perspectiveof NMR,” Journal of Molecular Liquids, 80(2), 179–205, 1999.

53. Comyn, J, Groves, C., Saville, R., “Durability in high humidity of glass-to-lead alloy jointsbonded with an epoxide adhesive,” International Journal of Adhesion and Adhesives, 14,15–20, 1994.

54. Lu, X., Hofstra, P., Bajkar, R., “Moisture absorption, dielectric relaxation, and thermal con-ductivity studies of polymer composites,” Journal of Polymer Science: Part B: PolymerPhysics, 36, 2259–2265, 1998.

55. Asao, N., Isao, H., Naotaka, T., “A new method for measuring adhesion strength of IC moldingcompounds,” Journal of Electronic Packaging, 114, 402–412, 1992.

56. Shirangi, M.H., Gollhardt, A., Fischer, A., Müller, W.H., Michel, B., “Investigation of frac-ture toughness and displacement fields of copper/polymer interface using image correlationtechnique,” Proceedings of the 41st International Symposium on Microelectronics (IMAPS),Providence, USA, pp. 917–923, 2008.

57. Dodiuk, H., Dori, L., Miller, J., “The effect of moisture in epoxy film adhesives on theirperformance: I. Lap shear strength,” Journal of Adhesion, 17, 33–44, 1984.

58. Kinloch, A., “Interfacial fracture mechanical aspects of adhesive bonded joints – a review,”Journal of Adhesion, 10, 193–219, 1979.

59. Lin, Y., Tsui, T.Y., Vlassak, J., “Water diffusion and fracture in organosilicate glass filmstacks,” Acta Materialia, 55, 2455–2464, 2007.

60. Shaw, G., Rogers, V., Payer, J., “The effect of immersion on the breaking force and failurelocus in an epoxy/mild steel system,” Journal of Adhesion, 38, 225–268, 1992.

chapter 2 mechanism of moisture diffusion,...

Documents