DffnRltd Bhvr f ld n Thn Fl t - Springer · 2017. 8. 25. · thn vn t. Dffnrltd bhvr f hrd ldppr t nd f fn ldppr t hld thrfr nt b ndrd dn tl n tr f thr f th t dffrnt pr: dft pth dffn

Diffusion-Related Behaviour of Goldin Thin Film Systems

M. Robert PinnelBel! Laboratories, Columbus, Ohio, U.S.A.

The process of diffusion in the solid state is one aspect of behaviour in

gold thin film systems which has received increased attention as gold

layers have had to be made significantly thinner without sacrificing their

reliability. This article reviews the characteristics of this process for gold

over the various one- and two-layer base metal systems which are in

common use by the electronics industry. A number of interesting

phenomena are described and related to the service problems which they

may cause in various devices.

The electronics and telecommunications industriesrequire high reliability in electrical connections.Thus, gold and gold alloys have long served as thestandard for the contact or surface material in applica-tions such as separable connectors, relays and leadframe metallization. Their low resistivity for elec-trical conduction, high chemical passivity forresistance to film formation, high bondability for easeof thermocompression and other forms of joining andtheir Base of application by standard manufacturingprocesses such as electroplating, sputtering, weldingor cladding are ideally suited to the requirements ofthese devices.

Historically, gold was used in relatively thick layersdeposited over the entire contact region, hence therewas little concern over the substrate metal and its in-teraction with the gold or the environment? The sur-face region would remain as pure gold with itsfavourable properties through the device lifetime.However, the significant increases in the cost of goldand the concurrent economie pressures of this decadehave stimulated appreciable efforts toward replace-ment of or reduction in gold usage per device. Themore conservative use of gold has been the approachusually pursued. In this, a reduction from classicallyspecified thicknesses had to be accomplished withoutany sacrifice in reliability. Thus, careful considera-tion had to be given to the behaviour of gold in thinfilm systems involving more than one metal.

Several potential mechanisms which can degradethe performance of a thin gold over base metal systemare possible when the gold is thinned significantly.Thin gold films are more likely to be porous (1) andinvite corrosion of the substrate metal despite thenobility of gold. They are subject to more rapidremoval by wear (2), which exposes the base metal

substrate with its inherent unreliable performance.Thin gold films are also subject to penetration by thesubstrate materials by diffusion (3), which can placeoxidizable base metals at the contacting surface anddegrade performance due to increased electricalresistance. All of these potential hazards involving theinteraction of gold and base metals in thin filmsystems have been the object of considerable attentionover the last five to ten year period. This article willconcentrate on a review of the diffusion behaviour ofgold/substrate systems, in particular those wherenickel and/or copper are the substrate materials.

Basic Diffusion MechanismsThere are three mechanisms pertaining to diffusion

behaviour which need to be considered and they areshown in a highly schematic form in Figure 1 for atypical gold over copper substrate layered structure.The first is classical bulk or lattice diffusion whichinvolves vacancy-atom exchange (4). It is by this pro-cess that the atoms in the layers intermingle so thatthe latter become of increasingly similar compositionwhile totally in the solid state. Bulk diffusion ishighly temperature dependent, however, andtemperatures approaching the melting point of thelower melting element are usually necessary for thereaction to near completíon in a practically significanttime. The rate constant which defines this process ata specific temperature is the lattice diffusion coeffi-cient D. Bulk diffusion takes on importance in suchfabrication steps as thermocompression bonding,which will be considered later.

As depicted in Figure 1, the lattice diffusion pro-cess is two-way. In the example shown, gold diffusesinto copper as well as copper into gold. Although thelatter step is the one most often considered, since one

62

Fig. 1 Schematic representationof the various diffusion-relatedreactions induced by thermalageing of gold/copper struc-tures. Bulk diffusion (coeffi-cient D) corresponds to thetransfer of copper into gold andgold into copper through the lat-tice. Defect path diffusion (coef-ficient D') involves transferalong lattice defects such asdislocations and twin or grainboundaries — only the latter isdepicted here. Ordered Cu3Au,CuAu and CuAu3 phases form atstoichiometric positions alongthe concentration gradient. TheKirkendall porosity at thecopper-gold interface illustratesthe faster rate of diffusion ofcopper into gold as comparedwith that of gold into copper

)LDEVER

Cu Au

)PPERJBSTRATE

is usually concerned with degradation of the reliabili-ty and properties of the gold by the inclusion of basemetal in it, the fact that the reaction is two-way canhave important consequences. If the rates of diffusionare unequal, as is often the case, excess vacancies areleft behind in the layer of the faster diffusing element,which will coalesce into a line of porosity known asKirkendall porosity (5). This can have an influenceon the mechanical integrity of the system and haspractical implications for the gold/copper system.

The second mechanism which must be consideredis that of rapid diffusion along defect paths. Thisprocess has a lesser temperature dependence than lat-tice diffusion and usually becomes dominant at lowertemperatures, and at room temperature for manysystems. Defects like grain boundaries, dislocationsand twin boundaries, which are common and prac-tically unavoidable in metals, can serve as rapidtransport pipes through a metallic layer. This factor isparticularly relevant in thin gold films where thetechniques of application such as electrodeposition(6,7,8), sputtering (9) and vapour deposition (6) allprovide very fine grained structuren of high defectdensity. The rate constant defining this process isusually given by the symbol D' and is called thedefect path diffusion coefficient. D' has been found tobe typically four to six orders of magnitude largerthan the lattice diffusion coefficient D at half themelting temperature in kelvins and below (10,11).Defect path diffusion is of primary concern in rela-tion to the formation on gold layers of surface films ofbase metal oxides which can degrade such propertiesas electrical contact resistance and bondability.

The third mechanism relating to interdiffusion isthe possible formation of ordered intermetallicphases between gold and a substrate metal. Thesecan take the form of layers growing at the originalgold-base metal interface as depicted in Figure 1.Such is the case for the gold/copper system wherethree ordered phases, Cu 3Au, CuAu and CuAu3 canform and are stable below about 400°C (12). Thelayer growth rate is relatively slow at these lowtemperatures and usually has no influence on the goldsurface properties. However, ordered layers are oftenbrittíe and their formation can influence themechanical integrity of the system. Susceptibility ofthe gold surface film to fracture and flaking off mayresult if mild bending stresses are applied undersuch conditions.

In the following sections, these basic mechanismsrelating to diffusion will be highlighted in greaterdetail for particular gold/base metal systems of cur-rent widespread practical application.

The Gold Film/Copper Substrate System

A gold layer on copper or a copper alloy substratehas been a most common combination for electricalcontacts and connectors for decades. Lattice diffusionin the gold-copper system at temperatures above the400 to 500°C range was extensively evaluated as farback as the 1930's. These investigations (13 to 16)were carried out on fine (pure, unalloyed) gold andwere aimed primarily at determining the lattice diffu-sion coefficient of gold in copper and that of copper ingold or the interdiffusion coefficient D, reflecting anaverage of the two-way diffusion process. The results

63

TEMPERATURE°C

975 750 500 400 250 200 150 100 50 25

• ARCHBOLD,KING (15)

♦ MARTIN et al. (16)

O MATANO et al. (14)

I PINNEL et al. (17)

X HALL et al (18)

o JOST (13)

10

11

12

13

lo14

0

15

16

17

18

19

1.0 1.4 1.8 2.2 2.6 3.0 3.4

10 3 l T. K -t

Fig. 2 Summary of the values measured by various authors indeterminations of the lattice interdiffusion coefficient D in thegold-copper system

of these studies are summarized in Figure2 and show good consistency amongthemselves and also with more recent data(17,18) acquired using more modernphysical techniques such as electron pro-be microanalysis (EPMA) and Augerelectron spectroscopy (AES) (18).These techniques permit the analysis ofvery small diffusion zones and data can beacquired in a much lower temperaturerange approaching that of practical in-terest where the diffusion rate is relativelyslow. Actual operating temperatures forcomponents such as connectors in circuitpacks having a high density of power-dissipating devices currently exceed 80°Cand are anticipated to become as high as135°C. Also, some metallizations are ex-posed to temperatures near 300°C forhours in order to cure polymericmaterials assembled within the same com-ponent. Thus, the importante of suchdata is in determining the time attemperature over which a given thicknessof gold will survive before beingpenetrated by sufficient copper to alter itssurface properties. As demonstrated in(17), one month exposure at 250°C issufficient to allow copper to diffuse exten-sively through 2.5 .tm of fine gold and at500°C only three days is required forsimilar penetration of 25 pm of gold.Analogous determinations, based on thediffusion coefficients given in Figure 2,

i be made for any combination of time,nperature and layer thickness.

line

Fig. 3 SEM micrographshowing the degradation of a25 pm thick pure gold elec-trodeposit by oxidation oftopper diffused from the

line substrate after 6 hours ageingat 750°C. The intensity pro-file of the CuKa X-ray emis-sion scans clearly shows thehigh concentration of copperat the surface of the gold andin stringers perpendicular tothat surface. Also as a resultof diffusion, the sharp inter-face has been replaced by aTine smooth transition from agold-rich coating to a copper-rich substrate

64

Once copper has penetrated to the gold surface, itsoxidation is unavoidable. At very high temperatures,copper oxidizes both as a surface layer and also aslong stringers penetrating far back into the originalgold layer (Figure 3). As can be observed from copperconcentration EPMA profiles, the original copper-gold interface fades completely as in diffusion bon-ding and the original pure gold layer becomes agold-copper alloy of varying composition as a result ofhigh temperature lattice diffusion.

No comparable data are available for the commonlyused cobalt- or nickel-hardened golds. This is a conse-quence of the instability of these films at elevatedtemperatures. It has been pointed out (19) that thepronounced blistering and porosity which develop insuch electrodeposited golds when heated above about400°C are probably due to thermal decomposition ofincorporated organic contaminants. A typical exam-ple is shown in Figure 4 of a 0.3 per cent cobalt-hardened gold after exposure at 500°C for 24 hours.The extensive porosity throughout the gold is ob-vious. However, the fact that much of this porosity isconcentrated at the original gold-copper interface, soas almost to cause delamination, reduces diffusionacross this boundary to near zero and precludesdetailed study of the diffusion process at suchelevated temperatures.

At less elevated temperatures, between roomtemperature and 250 to 300°C, where defect path dif-fusion becomes the predominant process, porosity ofhard gold does not develop to any appreciable degree.Thus, defect path diffusion does take place and can bemeasured. This observation is highly relevant forseveral reasons. Hard golds are commonly used indevices on account of their superior wear propertiesand resistance to cold-welding in comparison to finegold. Increased hardness is mostly attributable torefined grain size (20) and consequent increaseddefect density (7,8). This has obvious effects on theamount of copper (or other base metal) which can betransported via these paths to the surface regionwithin a given time. Diffusion-related behaviour ofhard gold/copper systems and of fine gold/coppersystems should therefore not be considered as iden-tical in terms of either of the two different processes:defect path diffusion and lattice diffusion.

Some of the available data on the diffusion rate ofcopper along defect paths in hard or pure gold areshown in Figure 5. These results were all obtainedusing AES on cobalt-hardened gold (21,23) or on puregold (22,24). The greatly enhanced rate of defect pathdiffusion over that of lattice diffusion is apparent.The former process occurs with surprising rapidityeven at room temperature as shown by the data of (22)or by the extrapolation of the other data to 25°Cassuming no change in the process mechanism. These

.i d R g

^s ^i •f es k i. ^. K` , Y^ y .f`^$5^0. S in i ^ •^N

^.^is3a ré...rlïs^`á^3

Copper Goldsubstrate layer

Fig. 4 SEM micrograph showing the thermal degradationof a 25 pm thick 0.3 per cent cobalt-hardened gold elec-trodeposit on copper after 24 hours exposure at 500°C inair. Note the virtually continuous Kirkendall porosity nearthe original gold-copper interface

TEMPERATURE,°C

750 500 400 250 200 150 100 50 25

i

GOLD-COPPER

11 LATTICE DIFFUSION

12 r\ \R

o

13

21)\ \R

14 \(

(23) \15 \

16

17\\(24) (22)Q

18

19

1.0 1.4 1.8 2.2 2.6 3.0 3.4

10 3 1 T,K -1

Fig. 5 Summary of the values measured by variousauthors in determinations of the defect path diffu-sion coefficient D' in the gold-topper system. Someof the latttce interdiffusion data of Figure 2 are alsoindicated for comparison purposes

65

Table 1Approximate Thicknesses of Cu 3Au and CuAu3

in Thermally Aged 3 000 A Gold /4 500 A Cop-per Couples. After (29)

Ageing Ageing Cu3Au CuAu3temperature time thickness thickness

°C hours A A

160 1 200 1 350 2 700

180 50 325 850180 200 1 300 3 500

200 20 450 1 050200 50 1 100 2 700

220 5 525 950220 20 1 800 3 500

results would predict the arrival of appreciableamounts of copper at the gold surface in normaldevice lifetimes. However, it is also clear that there iswide scatter in the data. In addition to the potentialdifferences in the measured apparent diffusion rateresulting from differences between hard and fine goldmicrostructures, other factors can produce such varia-tion. These have been discussed in detail (21,25) andthey include the sample boundary conditions, themethod of extracting a diffusion coefficient (from firstarrival time as opposed to surface accumulationmeasurements) and the surface sinking reaction.

Sufface sinking is particularly relevant. Gold doesnot participate in the reaction at the surface.However, oxidation or any other film-forming reac-tion of the base metal can have a direct effect on thecourse of diffusion. If the reaction at the surface,which essentially removes copper from the diffusionprocess, is slower than the diffusion step, the gold dif-fusion paths will become clogged with base metal anddiffusion will slow down to the rate of removal at thesurface rather than proceed at its intrinsic rate. In theabsence of an environment, as is the case in vacuum,this surface removal step is surface diffusion. It hasbeen shown (21) that in the system discussed here,surface diffusion is least effective and , corrosion incontaminated air (such as with chlorine) one of theprocesses most effective in removing copper from thesurface. Oxidation in relatively clean air falls betweenthese extremes. The nature of these surface reactionshas been examined in detail (26) and is of highrelevante to gold/copper devices in electronic ap-paratus, since the formation of surface films leadingto an increase in resistance is a primary mode offailure. Thus, a number of factors, including both thediffusion rate and oxidation rate, must be consideredin determining whether these processes are a potentialhazard for a particular device application.

The single exception to the statement that the goldlayer does not participate in the surface reaction is forhard electrodeposits which include base metals suchas cobalt or nickel. Part of these additions exist as freeatoms in gold-based solid solution and can diffuse tothe surface, like the substrate material does. Indeed,the formation of layers of cobalt oxide (23,27) andpotassium oxide (28) has been observed and can bedeleterious to the properties of the gold surface.

The last diffusion-related mechanism to be discus-sed for the gold/copper system is the reaction whichtakes place at the gold-copper interface. During ex-posure, even for short times, to temperatures belowabout 400°C, the three ordered gold-copper phasesare in equilibrium and can develop since thestoichiometric compositions are attained in specificranges across the diffusion profiles. In very thin filmsof fine gold on copper it has been demonstrated (29)that the Cu 3Au phase is formed initially and followedby the CuAu 3 phase at temperatures as low as 160 to220°C and in times as short as hours. The CuAuphase does not appear in the early stages of the pro-cess. It is first observed after 60 hours at 200°C andthereafter continues to grow at the expense of theCuAu 3 phase. Metallographic observations showedthat these phases formed as layers parallel to theoriginal gold-copper interface. Typical thicknesses ofthe layers formed at various times and temperatures(29) are given in Table I. It is Clear that even thoughthe Cu 3Au layer appeared first and continued to growwith time at temperature, the CuAu 3 layer wasthicker by at least a factor of two for almost all timeand temperature conditions measured.

100

Zo 75

wa

0 50LLJ

3LUáo 25U

o_COPPER f- — GOLD

Fig. 6 This CuKa X-ray emission scan reveals thepresente of Cu3Au and CuAu only at the diffusioninterface between a 25 pm thick 0.3 per cent cobalt-hardened gold electrodeposit and the coppersubstrate. The sample was aged 4 months at 250°C

66

Gold 1

5pm

CuAu

10µm

porosity

10 pm

4

Cobalt-hardened gold, aged 8months at 250°C

Fine gold, aged 6 months at250°C

Cobalt-hardened gold, aged 4days at 400°C

Fig. 7 SEM back-scattered images showing the effeets of ageing time and temperature on the respective formation of layersof ordered phases and of Kirkendall porosity in cobalt-hardened or fine gold/copper systems. Ordering is slow below250°C and not possible at or above about 400°C. The formation of Kirkendall voids is greatly accelerated by increased dif-fusion temperature and time

Other studies involving thicker layers (25 kim) offine and cobalt-hardened gold, aged up to 3 years attemperatures ranging from 150 to 400°C, showedconsistent behaviour (30). In these cases, ageing timeswere. sufficiently long for the CuAu layer to totallyconsume the CuAu 3 layer and only Cu 3Au and CuAuremained. In some instances, the layer thicknesses ex-ceeded 5 as determined by X-ray profiling with awavelength dispersive spectrometer in a scanningelectron microscope (SEM) on specimen cross sec-tions. A profile of the CuK a X-ray line for a sampleaged four months at 250°C exhibits the above men-tioned features (Figure 6). The layers have a composi-tion range consistent with the equilibrium diagram(12). They are also readily distinguished in the SEMbackscattered images shown in Figure 7 for samplesaged at 250 and 300°C. The ordered phases are notdiscerned after ageing at 400°C as CuAu is barelystable at this temperature and the upper limit ofstability of Cu 3Au is about 390°C (12).

Another important feature visible in Figure 7 is thedevelopment of Kirkendall porosity along the copper-rich side of the couple, which increases in abundantewith the time and temperature and hence the extentof diffusion. The formation of this porosity reflectsthe more rapid transport of copper into gold than viceversa (17). It is apparent that such a line of porositycould lead to delamination of the gold layer. This wasfound to be the case in the evaluation of thermocom-pression bonds between gold plated copper leadframes and gold metallized thin film circuits (31).The pull strengths of the leads decreased measurably

(in many instances to zero) when aged between 200and 300°C for times to only 2000 hours. Analysis ofthe fracture surfaces by X-ray and SEM techniquesrevealed the presente of Cu 3Au. The development ofthis layer of limited ductility together with Kirken-dall porosity probably caused the degradation of thisparticular system.

The Gold Film/Nickel Substrate SystemLattice diffusion in the gold-nickel system has also

been studied (32), but primarily above 812°C wherecomplete solid solubility exists (33). It was found thatthe rate of interdiffusion is two orders of magnitudegreater in gold-rich solutions than in nickel-rich solu-tions, which is certainly the direction unfavourable tothe properties of the gold surface layer as was alreadydescribed for copper in gold. Fortunately, theseelevated temperatures are outside the realm of prac-tical interest for common devices incorporating a goldlayer over nickel and at lower temperatures the cir-cumstances change appreciably. A broad miscibilitygap occurs in the gold-nickel system which at 300°C,for instante, limits the solubility of nickel in gold toless than 8 atomic per cent and that of gold in nickelto less than 2 atomic per cent (33). Phase separationgreatly restricts lattice diffusion and by comparisonmakes it a much slower process than in the copper-gold system. This is presumably the reason why nobulk diffusion has been observed in the lowertemperature range and why nickel enjoys its reputa-tion as an outstanding diffusion resistant substrateunder gold.

67

Table IIComparative Thicknesses of Copper andNickel Measured at the Free Surface of a2.5 µm Coating of Cobalt-Hardened Gold

After Ageing at 150°C

Ageing Equivalent Equivalent

time copper thickness nickel thickness

days A A

0.17 18± 4 —5 206±40

10 461±90 —18 — None*

40 — 8±2

80 — 38±8

*Less than the detection limit of about 0.1 monolayer

If defect path diffusion is considered, however, therestrictive nature of the miscibility gap is not so clear.Available data (25) for exposure to temperaturesunder 250°C indicate that such diffusion does occurand at a rate nearly equal to or only moderately slowerthan that of copper through gold. As a result, nickeloxide or other nickel-based corrosion products canform on the surface of gold deposits and therebydegrade the properties related to resistivity andbondability. A comparison of the rate at which thisprocess takes place, relative to copper and in terms ofsurface accumulation of base metal, is given in TableII. In this instante, the nickel defect path diffusion andsurface oxidation rates are moderately slower thanthose for copper, but still rapid enough to warrantconsideration in some gold over nickel applications.

To degrade the surface properties of gold, nickelmust react with the environment to form a resistivefilm. It has been shown in a study of nickel oxidationout of gold solution (34) that the gold does not par-ticipate in this film formation, which was to be ex-pected due to its nobility. The nickel was found to

1000.5,um GOLD / 2 AND 4,um NICKELICOPPER

ó 0.5 Nm GOLD / 0.5,um NICKELL•COPaW 80

v Og

60Nm GZ

40COPp R

1-zX 20

W1 1 1 1 1

1 2 5 10 20 50 100 200 500 1000

HEATINGTIME, HOURS

F.ig. 8 Reliability vs. ageing time at 125 °C of elec-trical contacts consisting of 0.5 pm pure gold oncopper with or without a nickel intermediate layer.The acceptance criterion was a contact resistancelower than 1.0 mQ. After (37)

form a limiting film thickness (34,35,36), which is afunction of temperature, relative humidity and at-mospheric contaminants, but which for comparableconditions is substantially thinner than the con-tinuously growing film of copper compounds. In thissense, the gold/nickel system may be considered assuperior to the gold/copper system.

The gold-nickel metallurgical system does not havethe potential for ordered phase formation and thus nolow temperature mechanical degradation of the layer-substrate bond is expected.

The Gold Film/Nickel Underlayer/CopperSubstrate System

Frequently, both copper and nickel are presentunder a gold or gold alloy surface layer in electroniccomponents and other products. The copper or a top-per alloy serves as the primary inexpensive substratewith good mechanical spring properties, solderabilityand formability. A layer of electrodeposited (or withnewer techniques also sputtered, vapour deposited,etc.) nickel is then applied to provide improved wearresistance, due to its hardness, and also as a diffusionbarrier to inhibit the penetration of copper to the goldsurface (3) as discussed in an earlier section of thisreview. Common wisdom prescribed nickel as areasonably effective barrier to copper diffusion andsome accelerated ageing studies supported this belief(37). Results showed that fine gold over copperdevices failed more rapidly than similar samples witha nickel barrier when aged up to 1000 hours at 125°C(Figure 8). While the reputation of nickel as an effec-tive barrier is well founded with respect to latticediffusion, recent AES results on defect path diffusionthrough gold (25) are only marginally in favour ofnickel when compared to copper.

A number of intriguing aspects were uncovered in adetailed study (38) of diffusion-related behaviour inthe fine or cobalt-hardened gold/nickel/copper tri-layer systems between 150 and 750°C. Nickel layers20 µm thick under 20 µm of `fine gold and also 5 µmthick under 5 pm of cobalt-hardened gold were in-vestigated. Ageing times were extended to as long asthree years at the lowest temperatures. Interdiffusionof the elements was followed by EPMA profiling in aSEM on metallographically polished cross sections.The high temperature results exhibited the expectedbehaviour and are depicted in Figure 9. Copper andnickel being mutually soluble in all proportions, agradual concentration gradient reflecting interdiffu-sion has built up across the copper-nickel interface.Kirkendall porosity now appears at the copper-nickelinterface reflecting the more rapid penetration of cop-per into nickel than vice versa (39,40). Nickel andgold show very limited mutual solubility, due to thebroad miscibility gap previously mentioned, hence

68

Gold

X14.: fiit `

GoldI

10 fm

Fig. 9 Typical coneentration vs. distantie profileson a gold/nickel/topper system after 6 hoursageing at 750°C. The profiles were producedfrom X-ray emissions induced in a SEM

the very abrupt concentration discontinuity at theirinterface. The concentration of nickel in gold remainsvery low but this has not prevented sufficient nickelfrom diffusing to the surface to form an oxide layer.

At 500°C and below, the nature and results of in-terdiffusion are found to change significantly. As il-lustrated by Figure 10, copper and nickel no longerexhibit mutual solubility, but instead the copper-nickel and nickel-gold interfaces show similarfeatures. The nickel appears to be unaffected by thediffusion process with little penetration of it into cop-per or gold detectable by the experimental techniqueused. AES analysis of the gold surface detected thepresente of some nickel, transported there by defectpath diffusion through gold. However, the nickelseems to serve as a transport medium for goldthrough to the copper substrate. As a consequente,gold has accumulated at the copper-nickel interfaceand has gradually diffused into the copper as shownby the gold concentration gradient. Identicalbehaviour can be noted for the copper which ac-cumulated at the nickel-gold interface. This copperwill eventually diffuse to the gold surface in spite ofthe nickel 'barrier' and oxidize. The intermediatelayer will extend the life of the system by at least thetime required by copper to diffuse across it but not in-definitely. The rate of transport across the nickellayer for both copper and gold must be fairly similardue to the absence of any visible Kirkendall porosity(Figure 10) which is so clearly evident after diffusionat higher temperature (Figure 9). Diffusion behaviourof this type has been observed down to 250°C and islikely to take place down to room temperature asno change in mechanism is foreseen over this

Fig. 10 Typical concentration vs. distante pro-files on a gold/nickel/topper system after age-ing at 500°C or below. The sample shown herewas annealed 80 days at 400°C

extrapolated range. However, ageing at 200°C andbelow for times as long as three years has not resultedin sufficient diffusion to be detected by EPMA.

This unusual behaviour has been attributed to thedeep penetration of the miscibility gap in the gold-nickel binary system into the gold-nickel-copper ter-nary system (41). The rapid defect path diffusion ofgold through nickel and into the copper near thecopper-nickel interface limits the solubility of nickelin this 'alloyed' region compared to that in pure cop-per and restricts copper-nickel interdiffusion. At300°C and below, the distribution of copper in goldand that of gold in copper at the opposing nickel in-terfaces become very patchy. This is taken as suppor-ting evidente of the defect path mode of diffusion ofcopper and gold through the nickel because signifi-cant bulk diffusion would lead to even distribution ofthese metals in Bach other parallel to the nickelunderlayer. The lack of detectable concentrations ofcopper and gold in the nickel (Figure 10) further sup-ports this interpretation.

The benefit of nickel in at least retarding the out-wards diffusion of copper is apparent from the workdescribed here (38). Another study (42) supports theconclusion that nickel does have such an effect butdisagrees with the intuitive feeling that increasing thenickel thickness will increase the time for copper toeventually reach the gold surface. An additionalbenefit of the nickel layer at lower temperatures is itsability to retard the formation of the ordered gold-copper phases at the interface as well as that ofKirkendall porosity. This prevents the loss of bon-ding strength that is observed in the gold/coppersystem when aged at 200°C and above (31).

69

SnNi

The Gold Film/Tin-Nickel Underlayer/Copper Substrate System

An equiatomic alloy of tin and nickel which can beelectrodeposited has been advocated as a substitute oras an underplating for gold on copper in applicationssuch as printed circuit boards (43,44). Its attractivefeatures include high hardness and strong corrosionresistance, making it of interest under thin gold filmswhich are usually porous. This alloy has already foundapplication under gold layers as thin as 0.05 pm.

The behaviour of a tin-nickel layer between copperand gold at temperatures where diffusion related ef-fects come into play is also of interest and a curiosity.In one study, tin-nickel deposits 12.5 pm and 2.5 imthick, plated between 60 pm fine gold and copper,were aged at up to 500°C. It is clear from the resultsdepicted in Figure 11 that tin-nickel underlayers areunsuitable at temperatures toward the upper end ofthis range. After only 50 hours exposure at 500°C,the initially continuous layer has begun to decompose

Tin-rich stringers

1IT•,.10 pm 10µm

Aged 50 hours at 500°C Aged 1 yang at 2.50°C

Fig. 11 SEM micrographs showing the decomposition and general behaviour of a 12,5 ua thick tin-ni kel layer, deposited.hetween a copper substrate and a gold coating, wheu thermally aged. Al 500°C, SnNi deeoinposes white at 250°C only par-hal decomposition and migration of soane tin towards the gold surface are visible

10prn

Gold

`I in-richSnNi Gold stringers

H //.^H

Ti -rich stringers

10m 10 pm

tIAged 50 hours at 500°C Aged 30 days at 300°C

Fig. 12 SEM rnicrographs showing the decomposition and general behaviour of a 2.5 µm thick tin-nickel layer, depositedbetween a topper substrate and a gold coating, when thermally aged. The same effects as those depicted in Figure 11 areobserved after shorter ageing times

70

into discrete particles. Gold and copper have com-pletely penetrated the layer, as if it were not even pre-sent. AES analysis of the gold surface revealed thepresence of both tin and copper - over 60 sm awayfrom their initial position! Thus, both copper frombelow the layer and tin from the decomposed layerdiffused through the 60 pm gold layer (again pro-bably by defect path diffusion) and altered its surfaceproperties.

The decomposition of the tin-nickel alloy is pro-bably related to the modification of the metastableelectrodeposited SnNi phase into the elevatedtemperature equilibrium two-phase structure con-sisting of Ni 3Sn 2 and Ni 3Sn 4 (45). This phase change,together with intermixing of the four elements by dif-fusion, may possibly explain the apparent disintegra-tion of the interlayer. At lower temperatures, wherethe modification of the layer is less dramatic (for ex-ample, 1 year at 250°C), tin is already being releasedand is reforming as stringers of tin-rich phase in thegold surface region (Figure 11). Even these milderconditions are deleterious to the surface electrical pro-perties of the gold film/tin-nickel underlayer/coppersubstrate system.

Thinner tin-nickel electrodeposits, such as the ini-tially 2.5 sm thick layers shown in Figure 12, exhibit

similar behaviour except that they are nearly totallydissolved much more quickly. After 5 hours at500°C, only a few remnants of the original layer arevisible and the tin or tin-rich phase can be observed tobe flowing toward the gold surface in 'waves'. After30 days at 300°C, decomposition is incomplete butalready the tin-rich stringers form a network in thesurface region of the gold.

Conclusions

Gold when plated over base metals such as copperor nickel will be subject to interdiffusion. This willoccur with surprising rapidity even at roomtemperature by the process of defect path diffusion.The typical result is the accumulation of base metalor base metal oxide in the surface region, which canalter electrical properties such as contact resistance.In addition to surface phenomena, othercharacteristics of the diffusion process, which occurnear the gold-base metal interface, such as Kirkendallporosity or intermediate phase formation can alsoalter the behaviour of gold layers in thin film systemsby mechanical strength degradation. Thus, diffusionand its potential consequences must be carefully con-sidered when designing devices using gold/base metalthin film systems.

Re£erences

1 R. G. Baker and T. A. Palumbo, Plating, 1971, 58, 791-8002 A. J. Solomon and M. Antler, Plating, 1970, 57, (8), 812-8163 M. S. Frant, 'Plated Metals in Electrical Contacts', Space

Technology Laboratories Symposium on Connectors, LosAngeles, California, November 1962

4 P. G. Shewmon, 'Diffusion in Solids', McGraw-Hill, NewYork, 1963, 44-45

5 R. E. Reed-Hill, 'Physical Metallurgy Principles', VanNostrand, Princeton, New Jersey, 1964, 263-264

6 A. Gangulee, .7. Appl. Phys., 1972, 43, 867-8737 P. S. Willcox and J. R. Cady, Plating, 1974, 61, (12),

1117-11248 S. Nakahara and Y. Okinaka, ,7. Electrochem. Soc., 1976, 123,

1284-12899 R. W. Balluffi and J. M. Blakely, Thin Solid Films, 1975, 25,

363-39210 R. Butz, W. Erley and H. Wagner, Phys. Status SolidiA, 1971,

7, K5-K811 R. W. Balluffi, Phys, Status Solidi, 1970, 42, 11-3412 M. Hansen and K. Anderko, 'Constitution of Binary Alloys',

Second Edition, McGraw-Hill Book Company Inc., NewYork, 1958, 199

13 W. Jost, Z. Phys. Chem., 1932, B-16, 123-12814 C. Matano, ,cpn. .7. Phys., 1934, 9, 41-4715 T. F. Archbold and W. H. King, Trans. TMSAIME, 1965,

233, 839-84116 A.•B. Martin, R. D. Johnson and F. Asaro, ,7. Appl. Phys.,

1954, 25, 364-36917 M. R. Pinnel and J. E. Bennett, Metall. Trans., 1972, 3, (7),

1989-199718 P. M. Hall, J. M. Morabito and N. J. Panousis, Thin Solid

Films, 1977, 41, (3), 341-36119 Ch. J. Raub, H. R. Khan and J. Lendvay, Gold Bull., 1976, 9,

(4), 123-12820 C. C. Lo, J. A. Augis and M. R. Pinnel, 'Proceedings of the

4th International Conference on Thin Film Phenomena', Lon-don, September 1978, in press

21 H. G. Tompkins and M. R. Pinnel, ,7. Appl. Phys., 1976, 47,(9), 3804-3812

22 J. A. Borders, Thin Solid Films, 1973, 19, 359-36823 H. G. Tompkins, 1. Electrochem. Soc., 1975, 122,983-98724 P. M. Hall and J. M. Morabito, Surf. Sci., 1976, 59, 624-63025 H. G. Tompkins and M. R. Pinnel, J. Appl. Phys., 1977, 48,

(7), 3144-314626 M. R. Pinnel, H. G. Tompkins and D. E. Heath, Appl. Surf.

Sci., in press27 J. H. Thomas III, IEEE Trans. Parts Hybrids Packag., 1976,

PHP-12, 255-25928 J. H. Thomas III, ,7. Electrochem. Soc., 1978, 125, 898-90229 K. N. Tu and B. S. Berry, 1. Appl. Phys., 1972, 43, (8),

3283-329030 M. R. Pinnel, Bell Laboratories, Columbus, Ohio, unpubli-

shed research31 P. M. Hall, N. T. Panousis and P. R. Menzel, IEEE Trans.

Parts Hybrids Packag., 1975, PHP-11, 202-20532 J E. Reynolds, B. L. Averbach and M. Cohen, Acte Metall.,

1957, 5, 29-4033 M. Hansen and K. Anderko, op. cit., 22034 M. R. Pinnel, H. G. Tompkins and D. E. Heath, .7. Elec-

trocheni. Soc., submitted for publication35 M. J. Graham and M. Cohen, ,7. Electrochem. Soc., 1972, 119,

879-88236 S. H. Kulpa and R. P. Frankenthal, J. Electrochem. Soc., 1977,

124, 1588-159237 M. Antler, Plating, 1970, 57, (1), 615-61838 M. R. Pinnel and J. E. Bennett, Metall. Trans., 1976, 7A, (5),

629-63539 L. C. C. Da Silva and R. F. Mehl, .7. Met., 1951, 3, 155-16340 A. Nyilas, G. Betz and M. G. Frohberg, Z. Metallkd., 1973,

64, 824-82641 A. S. McDonald and G. H. Sistare, Gold Bull., 1978, 11, (4),

128-13142 J. C. Turn and E. L. Owen, Plating, 1974, 61, (11), 1015-101843 G. C. Wilson, Trans. Inst. Met. Finish.,, 1972, 50, 109-11344 M. Rao, M. Mitchell and R. Anderson, Solid State Technola,

1974, 17, (11), 47-4845 J. E. Bennett and H. G. Tompkins,1. Electrochem. Soc., 1976,

123, 999-1003

71