MOISTURE-INDUCED GOLD BALL BOND DEGRADATION OF POLYIMIDE- PASSIVATED DEVICES IN PLASTIC PACKAGES

Transcription

1 MOISTURE-INDUCED GOLD BALL BOND DEGRADATION O POLYIMIDE- PASSIVATED DEVICES IN PLASTIC PACKAGES C Glenn Shirley Intel Corporation 52 NE Elam Yong Pkwy MS: AL3-5 Hillsboro, OR 9724 Abstract A new mechanism of gold ball bond degradation in plastic packages has been demonstrated Polyimide die topcoat and moistre stress accelerate bond degradation The mechanism is frther accelerated by mis-targeted bonds and "low" bonding parameters Introdction Growth of intermetallic phases between gold and alminm and incidence of Kirkendall voiding are major factors inflencing the strength and reliability of gold ball bonds to alminm bond pads Intermetallic growth and bond degradation have been widely stdied with accelerated life tests in both hermetic and plastic 2 packages The progression of growth of intermetallic phases depends on the environmental conditions (ie, temperatre and hmidity), on the initial as-bonded morphology as determined by bonding parameters (ie, ltrasonic power, dynamic bond force, temperatre, and time), on the type of bond pad metallization, and on imprities and contamination on the bond pad srface, 3,4 or a given degree of intermetallic formation, the strength of a bond also depends on geometrical factors sch as ball bond interfacial area and bond placement 5 or example, a bond may be mis-targeted so that it overlaps the edge of a bond pad Polyimide passivation is known to improve die reliability in plastic packages by mechanical protection of the die srface The incidence of thin film cracking, for example, is greatly redced However, new reliability jeopardies are possible, particlarly relating to wire bonding: Polyimide etch resides or polyimide hydrolysis prodcts generated in a hmid ambient may inflence intermetallic growth Mis-targeted bonds overlapping polyimide may also interact with polyimide by either affecting the local bonding conditions, or by mechanical interactions dring environmental stress To get a worst case look at these concerns, we ran small wire pll and preliminary bond shear experiments or the wire pll experiment, 2 polyimide-passivated nits were bonded, half with centered bonds, and half prposely bonded 25 off of the bond pad They were then assembled in 84 lead PQP packages and sbjected to either hors of nbiased 56 C, 85 relative hmidity (56/85) HAST or an hor dry bake at 56 C After stressing, molding compond was removed from the die and wires were plled Cmlative distribtions of the pll strengths for and 25 off-pad bonds following these stresses are shown in igre 9 9 C 7 m 3 Melissa Shell De Gzman Intel Corporation 22 Mission College Blvd PO Box 589, MS: SC2-24 Santa Clara, CA HAST O-CENTER HAST CENTERED BAKE CENTERED BAKE O-CENTER Pll orce (gm) ig Bond strength of polyimide topcoated plastic packaged nits after hors of 56/85 HAST or 56/ bake Open symbols: Ball lifts Points: Wire breaks Two modes are indicated by plot symbols: A open symbol indicates a separation at the ball / bond pad interface, whereas a plotted point indicates a wire break It can be seen that, withot hmidity, no intermetallic fractres occrred The strength distribtion characterizes wire breaks (mostly neck fractre) only, and is the same as is observed in nstressed nits In contrast, samples sbjected to the same time-temperatre conditions, bt with 85 relative hmidity moistre added, 2 to 6 of the bonds fail by intermetallic fractre at the ball/pad interface Bonds mis-targeted by 25 have the higher proportion (6) of sch failres Moreover, of the bonds mis-targeted by 25 of the ball diameter were zero-strength (The pll tester registers abot gram even for nattached wires) A preliminary shear test experiment was also performed by stressing nits at 56 C for varios times and at (bake), 65, and 85 relative hmidity A nmber of nits and wires sfficient to establish the median shear strength were tested, with the reslts shown in Table I It is clear that bonds are virtally ndegraded after bakes for as long as 2 hors at 56 C, whereas addition of moistre to the stress leads to appreciably weaker bonds Table I Mean shear strength of interfacial fractre modes in grams after stress at 56 C for varios times and relative hmidities Polyimide-passivated test dice at 56 C and x Relative Hmidity x RH hr hr 2 hr 6 hr 2 hr 2 g 4 g 9 g g 26 g 7 g g 26 g 9 g

2 These observations prompted s to carry ot a larger experiment to explore these isses in greater depth Specifically, we considered the effects a polyimide topcoat, mis-targeting of bonds by varios amonts, and ball bond process parameters Since qantitative predictions of reliability reqire a mathematical model to describe these phenomena, destrctive ball shear data were acqired for several accelerated environmental stress conditions, sfficient to extract probability distribtions of shear strengths, and acceleration factor formlae which can be sed to extrapolate the predictions of the model to conditions which actally occr in service There are at least for methods for stdying bond degradation mechanisms: A simple electrical test for opens, Kelvin resistance measrements 6, a bond wire pll test, and a ball shear test Electrical continity tests are not necessarily the best choice to determine ball bond integrity in plastic packages becase even a separated bond can be held in contact with the bond pad by compressive stresses in the package Kelvin measrements provide convenient information abot kinetics of growth of intermetallic phases and voids at the A/Al interface, bt provide no direct information abot the bond strength Wire pll tests are convenient and qick, bt only provide information abot pathologically weak bonds, since wire strength is only abot -2 grams We therefore chose to se the bond shear test since it provides mch more information abot the bonds, in a mode similar to the way bonds are actally stressed in the plastic package In Section 2 we describe the fabrication and stress flow which prodced the data In Section 3 we describe the distribtion of fractre modes as a fnction of the experimental variables to develop a qalitative nderstanding of the main effects of the variables Section 4 presents an analysis of bond shear strength data leading to a qantitative reliability model for bond strength as a fnction of time, temperatre, hmidity, etc Section 5 is an analysis of the effect of bonder parameters on A/Al intermetallic integrity The main conclsions are collected in Section 6 2 abrication, Stress, and Test low This stdy tilized SRAM die in lead PQP packages bonded on a K&S 484XQ bonder The passivation consisted of 6 micron of plasma silicon nitride Additionally, some of the experimental nits also had a for micron polyimide topcoat Each die contained a nmber of repetitions of an off-pad bond matrix: or consective bond pads were bonded with varying degrees of intentionally mistargeted bonds; the bonds overlapped the pad edges by, 5, 5, and 25 percent of the ball diameter The polyimide was overetched to clear the pads, so only the 25 mistargeted bonds actally overlapped polyimide As noted by Hnd, proper choice of bond parameters is vital to obtain reliable ball bonds In the present stdy, ball bond time and temperatre were fixed The ball shear test was sed to choose the best ltrasonic power and dynamic bond force windows meeting the strength and failre mode reqirements as described in reference 7 This methodology ensred that optimal ball bonds were made at the mid-point ("nominal") power/force bonder settings Polyimidepassivated nits with the off-pad bond matrix were bonded at the low power/low force ("low") corner of the bond window and at the high power/force ("high") corner as well as the mid-point settings This was done to determine the sensitivity of mis-targeted bonds on polyimide to bond process window variations Only the data from the midpoint bond settings were sed to extract the reliability model The effect of bond parameters will be discssed in Section 5 The material fabrication and environmental testing flows are shown in ig 2 Some nmolded lead frames were taken from the prodction flow so that as-bonded bond strength data cold be acqired The packaged nits were sbjected to simlated boardmont stress ("preconditioning") consisting of 24 hors of bake at 25 C, followed by 5 cycles of temperatre cycling condition "C", a 68 hor soak at 85/3, and 3 cycles of vapor-phase solder reflow This was followed by a series of highly accelerated temperatre/hmidity/time stresses sing a pressrized nsatrated temperatre/hmidity (HAST) chamber and a satrated steam test Sample sizes varied at each readot; the short drations of the mild stresses had the largest sample sizes to increase the probability of observing failres The nits intended for the 56/65 stress were inadvertently stressed at 56/85 for 7 hors before the mistake was noticed This additional stress was taken into accont in the extraction of a model 85/85 hr 2 hr Wafer abrication No Polyimide Topcoat Wafer abrication Polyimide Topcoat Assemble to wire bond,, 5, 5, 25 off pad Use low, nominal and high force bonder settings (Low and high settings sed only for material stressed at 56/85) Mold, dejnk, plate, trim and form leads 2/ 336 hr hr Precondition 35/85 hr 6 hr 2 hr Sample strips for shear test Sample nits at EOL for shear test 7hr 56/85 56/ /65 hr 7 hr 2 hr hr hr Bake dry, decap, test by bond shear or pll ig 2 abrication, environmental stress, and bond shear test flow for reliability model extraction Eight bonds per device were sheared, two of each degree of mistargeting The bonds failed either by dctile fractre of the ball, leaving gold on the bond pad ("gold" failre), or by fractre at the interface between the gold ball and the alminm bond pad ("intermetallic" or "interfacial" failre) Virtally every bond sheared failed by one of these modes 3 ractre Mode Analysis In this section we analyze the distribtion of bond fractre modes obtained in the flow of ig 2 as a fnction of presence or absence of polyimide, degree of bond mis-targeting and the conditions and dration of environmental stress In ig 3 we show the wire pll data acqired in the experiment It is apparent that a combination of polyimide and mis-targeted bonding leads to significant bond degradation We shall explore these effects more comprehensively, inclding the effect of environmental stress and time, by analysis of shear mode distribtions

3 C m POLYIMIDE O-CENTER NO-POLYIMIDE POLYIMIDE CENTERED & O CENTERED CENTERED Pll orce (gm) ig 3 Polyimide Vs non-polyimide and 25 off-pad (bonded at nominal bonder settings) wire pll strengths after hors of 56/85 HAST Open symbols: Lifts de to interfacial fractre Points: Wire breaks Althogh or primary focs is on the evoltion of bond strength dring environmental stress, it is interesting to stdy the variation in strength throgh the assembly process This is shown in Table II Table II Ball shear strength and modes for nits with polyimide topcoat immediately after bond (Strip), at end of assembly (EOL), and after hors of 56/85 HAST stress Bonds with strength < 25 grams failing by either mode (in practise, only the Interface mode) are recorded as "Weak" Strength (gm) ailre Mode Cont Case Off Aver Std Ball Interface Weak -age Dev Shear Strip 68 5 Strip EOL EOL HAST HAST When bonds are initially formed ("Strip" in Table II), of failres are interfacial fractres with a mean strength of abot 67 grams At the end of assembly ("EOL" in Table II), the thermal processing throgh molding compond cre leads to an increased strength of the interface so that most bonds now shear by dctile deformation of the gold ball As expected, mistargeted bonds show a lower incidence of dctile ball shear than centered bonds (66 verss 93, respectively) inally, after hors of HAST stress, bonds again fail by interfacial fractre, bt in contrast to the predominant mode jst after assembly, 9 are "weak" (<25 grams) 8 This seqence of modes is cased by strengthening of the A/Al interface by growth of intermetallic phase dring the mild thermal history of cre, then sbseqent degradation of the interface as Kirkendall voids form and coalesce dring the more severe HAST stress When bonds were sheared, the maximm shear force and the failre mode were recorded The way in which the two observed failre modes were distribted as a fnction of the experimental parameters is shown in Tables IIIA-E The tables show the incidence of interfacial fractre for each degree of bond mistargeting, and for polyimide verss no-polyimide The tables also show the incidence of "weak" bonds arbitrarily defined as bonds with shear strength less than 25 grams or example, Table IIIB shows that for polyimide at 336 hors of steam (2/), /36 25 off-pad bonds failed in shear test by interfacial fractre throgh the A/Al intermetallic The remaining 25/36 bonds failed in shear test by dctile shear throgh the gold ball Moreover, none of the 336 hor steam, polyimide, 25 off-pad bonds (/36) were weaker than 25 grams, irrespective of failre mode In the tables, whenever a bond was weaker than 25 grams, it failed in test by the interfacial fractre mechanism Several conclsions can be made by inspection of Tables IIIA- There is no statistical difference in failre mode incidence among, 5, and 5 off-pad cases for any environmental stress, with, or withot polyimide The 25 off-pad case, however, prodced more interfacial fractres and weaker bonds than the rest of the poplation Notable examples inclde 85/85 with polyimide in Table IIIA, and steam with polyimide in Table IIIB Therefore, in the analysis of shear strengths in Section 4, the, 5, and 5 off-pad data are groped together, and the 25 off pad data are treated as a separate grop The strength of the bonds decreased with increasing off pad as shown in Table III for a cople of examples, with the 25 off-pad bonds being significantly weaker than the others 2 At early stress drations, there is a clear difference in the incidence of interfacial fractre between polyimide and nopolyimide This is evident at the hor 85/85 readot where the polyimide case of 25 off-pad has 5/78 interfacial fractres, whereas the non-polyimide case of 25 off-pad has /48 interfacial fractres This is also evident at all steam readots, at the 2 hor 35/85 HAST readot, and the hor 56/65 and 56/85 HAST readots The strength distribtions for a cople of early readots shown in Table III clearly show that bonds in the polyimide case are weaker than in the non-polyimide case SEM analysis (ig 4) showed that Kirkendall voids formed only on polyimide-passivated dice with all degrees of off-pad bonding following moistre stress at early readots The dice withot polyimide topcoat, by contrast, did not show sch voids This reslt sggests that a polyimide etch reside or hydrolysis prodct activated dring the moistre stress may be responsible for the void formation and conseqent bond weakening or longer stress times, a majority of bonds, with and withot polyimide fail in test by interfacial fractre, and we will see in Section 4 that there is little difference between the strength distribtions in polyimide and no-polyimide cases 3 A high incidence of the interfacial fractre mode does not necessarily correspond to a high incidence of "weak" bonds Nevertheless, as stress dration increases, interfacial fractre is a precrsor to weak bonds In fact, the initial formation of intermetallic phase actally strengthens the A/Al interface Bt with increasing stress dration and frther growth of the intermetallic phase, Kirkendall voids form and coalesce, ltimately weakening the bond This effect is seen in Table IIID

4 Table IIIA 85/85, nominal bonder settings Polyimide No Polyimide Off 2 2 Interfacial Bonds < 25 grams Sample Size Table IIIB Steam (2/), nominal bonder settings Polyimide No Polyimide Off Interfacial Bonds < 25 grams Sample Size Tables IIICa 35/85 HAST, nominal bonder settings, polyimide Polyimide Off Interfacial Bonds < 25 grams Sample Size Tables IIICb 35/85 HAST, nominal bonder settings, no polyimide No Polyimide Off Interfacial Bonds < 25 grams Sample Size Table IIID 7 hors of 56/ /65 HAST, nominal bonder settings Polyimide No Polyimide Off hors 2 hors hors 2 hors Interfacial Bonds 9 5 < 25 grams Sample Size Table IIIE 56/85 HAST, nominal bonder settings Polyimide No Polyimide OP 7 7 Interfacial Bonds < 25 grams Sample Size Table III Ball shear strengths for selected HAST stresses, nominal bonder settings Polyimide No Polyimide OP Mean Std Dev Mean Std Dev 7 56/ hr 56/ hr 56/

5 ig 4 Cross section of 25 off-pad ball bonds after hors of 56/85 HAST Top: With polyimide topcoat Bottom: Withot polyimide topcoat Kirkendall voiding is associated only with the polyimide case 4 Reliability Model Extraction In this section we analyze qantitative shear strength data acqired in the flow shown in ig 2, with the objective of extracting a model that can be sed to predict bond reliability Althogh wire pll data show the effect of polyimide, mis-targeting and moistre in a qalitative way, the data is not as sefl as shear data in extracting a qantitative reliability model There are two reasons for this: () The interfacial fractre mode of interest is severely censored by the strength of the wire That is, only very severely degraded A/Al interfaces will fractre when tested by this method - so we are deprived of information abot early stages of degradation (2) In plastic packages, the main stress applied to bonds by the molding is a shearing stress 9 Ths, a model based on shearing data is closely related to the actal stress on the bonds in a plastic package When each bond is shear tested, the strength and failre mode are recorded If the interface between the gold ball and the alminm pad is sfficiently strong, the ball will fail by dctile shear throgh the gold ball, leaving gold on the pad The shear strength recorded for this dctile ball shear mode provides only a lower bond on the strength of the interface On the other hand, if the A/Al interface is weaker than the stress reqired to case dctile shearing of the gold ball, then the interface will fractre by shear, and the strength recorded is an actal measre of the strength of the interface We wish to know the distribtions of the A/Al interface strengths as if there were no censoring by the dctile gold ball shear mechanism obscring observation of stronger interfaces We employed Nelson's method of hazard plotting analysis to separate the strength distribtions of the two competing, mtally censoring failre modes We chose to plot the shear strength data on lognormal probability plots since this avoids the isse of extrapolation to negative bond strengths, which wold be physically implasible We groped the, 5, and 5 off-pad data since, as shown in the mode analysis in Section 3, there was no observable dependence on the degree of mis-targeting p to 5 off pad The 25 off pad data were treated as a separate grop 52 lognormal distribtion plots were made with the separated gold ball shear and interfacial fractre distribtions sperimposed Straight lines were fitted by the least-sqares method throgh each of the distribtions to determine the vale of the median shear strength and sigma A range of vales for sigma was fond for each shear fractre mechanism We fond σ = 7 ± 2 (A ball shear mechanism) σ = 7 ± 7 (Interfacial fractre mechanism) We assmed one vale (the median vale) of sigma for each mechanism, and made slope-constrained least-sqares fits to the distribtions sing the median vales of sigma Examples of distribtion plots, and sigma-constrained distribtions are given in ig 5 for the case of 25 off-pad, polyimide-passivated material stressed at 35/85 The vales of the median shear strengths of interfacial fractre,, were ths determined as a fnction of temperatre, hmidity, time in stress, degree of passivation overlap, and whether or not the die had a polyimide overcoat The next step was to determine a fnctional dependence for the time variation of the median interfacial shear strength We noticed from a log-log plot of verss time that is approximately inversely proportional to time at relatively long stress times However, cannot be inversely proportional to time for all times, since for short times this wold predict a strength increasing withot limit, in violation of the fact that the interface between the gold ball and the alminm pad will have a finite strength at zero time in stress We explored several fnctions which had the appropriate asymptotic behavior for long and short times, choosing the simplest The polyimide data are well described by the following fnction: = 2 ( at) + b 2 (Polyimide) () where a is the time acceleration factor, and b is the limiting initial bond strength This is shown in ig 6, where polyimide data (solid 2 symbols) are seen to fall on straight lines on a plot of /( ) verss t 2 for a 35/85 HAST stress This was also observed for other stresses It is also clear from ig 6 (and other similar plots) that the time dependence of for long stress times was similar for polyimide and non-polyimide, bt different for 25 mistargeted bonds and < 25 mis-targeted bonds

6 9 hors =735 s=7 9 6 hors =627 s=7 C m a i l C m a i l =8648 s=7 Interfacial Ball Shear =6636 s=7 Interfacial Ball Shear 2 Shear Strength, grams 2 Shear Strength, grams 9 2 hors =588 s= hors Interfacial C m a i l =569 s=7 9 C m 7 a 3 i l =445 s=7 Interfacial Ball Shear 2 Shear Strength, grams 2 Shear Strength, grams ig 5 Bond shear strength distribtions for polyimide-passivated material with 25 overlap of the bond onto passivation, as a fnction of time at 35/85 HAST With increasing stress dration the distribtion of interfacial shear strengths shifts to smaller vales, whereas the gold ball dctile fractre strength distribtion remains relatively nchanged Ths, the incidence of interfacial fractres increases with increasing stress dration The fitted distribtions are slope-constrained to represent, for each mode, a single vale of sigma which is optimal for all of the strength distribtion data (48 other plots besides those shown here) It is reasonable to expect the same time-zero interfacial strength for polyimide and no polyimide Ths the polyimide and nonpolyimide cases appear to have the same asymptotic time dependence for short and long stress times On the other hand, we can see from Tables III that at intermediate stress times the incidence of interfacial fractre is greater for polyimide than for non-polyimide This corresponds to the missing non-polyimide data points (open symbols) at the earliest readot time in ig 6 One way to describe this effect is to choose a parameter n>2 in the formla = n n ( at) + b n (Non-Polyimide, n>2, n=3 chosen) (2) The data were not sfficient to determine an neqivocal vale for n, except to say that it was larger than 2 We have chosen a vale n=3 which gives a reasonable qalitative description of the nonpolyimide time dependence As a reslt of fitting all of the median shear strength data to fnctions of the form of Eqs () and (2), we fond that the zerostress strength of bonds was well represented by b = 983 grams for the 25 mis-targeted bonds, and by b = 27 grams for all the bonds mis-targeted by less than 25, irrespective of whether the passivation inclded polyimide or not The vales of the acceleration parameter, a, were also determined for each environmental stress condition, each polyimide verss no-polyimide case, and each case of bond mis-targeting We have seen in Table I that in a dry ambient (bake, relative hmidity = ) the A/Al interface strength does not degrade appreciably even at the highest stress temperatre (56 C) for the times of interest in this experiment (p to 2 hors in 56/65, and hors in 56/85) Therefore, it is reasonable to assme an acceleration parameter which goes to zero as the relative hmidity goes to zero The "Peck" formla has this property, and has been widely sed to describe the acceleration of moistre-related failre

7 mechanisms Of corse, baking ( relative hmidity) will eventally degrade the A/Al interface by Kirkendall voiding, bt at mch longer times than are observed in the present stdy This "thermal activation only" mechanism is therefore not the sbject of this paper or each case of polyimide verss no polyimide and degree of bond mis-targeting, the acceleration parameter a was fitted to the Peck formla p a = a h exp( Q/ kt) where h (<h<) is the relative hmidity and the constants a, p and q were determined by a least-sqares fit to the data 2 The constants fond are given in Table IV Table IV Acceleration model parameters in acceleration formla, Eq (3) Polyimide? Off a (gm-) - p Q (ev) Yes < 25 33E 82 9 Yes E9 5 3 No < E 2 7 No E 85 9 It is apparent that one activation energy can describe all for cases of polyimide and no-polyimide Also, a sensitivity stdy shows that the indicated variations in p of +/- 2 do not greatly affect the fit of the model Therefore, we took averages of p and Q from Table III and re-determined a for each case with the constraint that p = 98 and Q = 5 ev Therefore, the complete model describing the degradation of the median interfacial strength is embodied in Eqations (), (2), and (3) with the parameters shown in Table V Table V Interfacial strength degradation model for A/Al bonds, made at nominal bonder power/force settings in temperatre/hmidity ambients Z P is the Pth percentile of the normal probability fnction Poly? Off a Q n b (gm) (gm-) - p (ev) Yes < E 98 5 Yes E 98 5 No < E 98 5 No E 98 5 = n n [( at) + b ] / n p a = a h exp( Q/ kt) = σ Z ) P exp( P σ = 7 (3) A = a exp( Q/ kt) (4) verss h rom Eq (3) with p = 98, this shold be almost proportional to h, and we see in ig 7 that the data are consistent with this fnctional dependence We demonstrate the overall fit of the model to the data by sperimposing the model predictions over the experimental vales of the median shear strength of the A/Al interface in ig **2 4 2 (y-axis nits: **-4 grams**-2) No Polyimide Polyimide 25 off pad, polyimide 25 off pad 25 off pad, no polyimide <25 off pad, no polyimide <25 off pad, polyimide <25 off pad Time**2, nits of, hors**2 ig 6 Median shear strength of bond interface as a fnction of time in 35/85 HAST A linear plot corresponds to the fnctional form of Eq () The intercept is b and the slope is a 2 The circlar filled plot symbols correspond to the data of ig 5 6 A 3 nits of E-3 /gm O/P, No Polyimide 25 O/P, Polyimide <25 O/P, No Polyimide <25 O/P, Polyimide 25 off-pad No Polyimide Polyimide < 25 off-pad Relative Hmidity, ig 7 Acceleration factors, a, temperatre-normalized to 56 C, sing activation energy of Q = 5 ev, to isolate the relative hmidity dependence of acceleration Shows that acceleration is proportional to relative hmidity In ig 7 we plot the experimental vales of the acceleration constant a, scaled to take into accont thermal effects, verss hmidity That is, we plot

8 85/85 2/ (Steam) 5 5 Model Data 25 O/P, No Polyimide <25 O/P, No Polyimide Model Data 25 O/P, No Polyimide <25 O/P, No Polyimide 25 O/P, Polyimide <25 O/P, Polyimide 25 O/P, Polyimide <25 O/P, Polyimide E3 2E E3 2E3 35/85 7 hors 56/85 + x hors 56/65 5 Model Data 25 O/P, No Polyimide <25 O/P, No Polyimide 25 O/P, Polyimide <25 O/P, Polyimide 5 Model Data 25 O/P, No Polyimide <25 O/P, No Polyimide 25 O/P, Polyimide <25 O/P, Polyimide E3 2E E3 2E3 56/85 5 Model Data 25 O/P, No Polyimide <25 O/P, No Polyimide ig 8 A/Al interfacial shear strength in grams as a fnction of stress ambient and time for varios percentages of bond mistargeting, and polyimide verss no polyimide Predictions of the model in Table V are sperimposed on the median interfacial shear strengths derived from distribtions sch as shown in ig 5 25 O/P, Polyimide <25 O/P, Polyimide E3 2E3

9 C m Effect of Bonder Parameters on A/Al Intermetallic Degradation LOW MID HIGH Pll orce (gm) ig 9 Effect of bonding parameters on HAST-indced ( hors of 56/85) weakening of worst-case bonds (25 off-pad, with polyimide) Wire pll reslts Open symbols: Lifts Points: Wire breaks Table VI 56/85, high and low bonder settings Polyimide, low bonder parameters Polyimide, high bonder parameters hr hr hr hr OP Interfacial Bonds 3 < 25 grams Sample Size 3 2 As mentioned in Section 2, wire pll and bond shear data were collected following and hors of 56/85 HAST on polyimide lots bonded at the low and high ends of the ltrasonic power/bond force window as well as at the midpoint settings The wire pll force distribtions for each of the bonder settings are shown in ig 9 The slight decrease in median pll strength (3 grams) and large increase in standard deviation for low bonder settings is seen to be associated with a greatly increased incidence of ball lifts Table IIIE, which lists ball shear data for nominal bonder parameters can be compared with Table VI which shows comparable data for low and high bonder parameters Comparison of these tables shows that polyimide interfacial fractre rates are extremely sensitive to bonding conditions; higher bond power/force settings prodce fewer intermetallic failres and higher bond strengths Tables VI and IIIE show that after hors of 56/85, the low bonder settings prodced intermetallic failres (inclsive of all degrees of mis-targeting), indicating inadeqate mixing of the gold ball and alminm pad Increasing the ltrasonic power and bond force to their mid-point settings improved this failre rate slightly to 883 The incidence of intermetallic fractres was frther redced to 46 when the high end of the power/force window was sed As seen in Table IIIE, however, the samples withot polyimide had the least amont of ball/bond separations; 25 overall for hors of HAST 6 Discssion and Conclsions Effect of Polyimide In the first stages of bond degradation (p to abot 3 degradation of bond strength) presence of polyimide is associated with increased freqency of interfacial fractre and generally weaker bonds Polyimide etch resides or hydrolysis prodcts may have been responsible for the higher incidence of Kirkendall voiding that leads to bond degradation At the later stages of bond degradation, when bond strengths have degraded by 3 to 7 grams or more, bonds on polyimide and non-polyimide material degrade at the same rate Effect of Mistargeted Bonds Bond mistargeting has a strong effect on the incidence of intermetallic fractre for both polyimide and non-polyimide passivated materials This increased incidence of intermetallic fractre is related to the fact that bonds with greater mistargeting are weaker and are less likely to be censored by the dctile ball shear mode Bond strength decreased with increasing bond mistargeting for any given stress history, both with and withot polyimide Effect of Processing Bond strengths immediately after bonding increase by abot dring post mold cre de to formation and growth of the A-Al intermetallic phases at the ball/bond pad interface rther growth of the intermetallic phase dring environmental stressing leads to weakening of the interface de to Kirkendall void formation and coalescence Effect of Bonder Power/orce Settings The bond integrity of polyimide-coated samples was strongly inflenced by the ltrasonic power and bond force The higher end of the bond window prodced the strongest ball bonds Time Dependence of Degradation The strength of bonds on polyimide-passivated material degrades with time according to the relationship = ( at) + b 2 2 or non-polyimide coated material, the as-formed bonds have the same strength as on the polyimide coated material At long stress times, the bond strength of polyimide and non polyimide material also has the same asymptotic time dependence (inversely proportional to time) However, the time dependence of strength at early stress times is different between polyimide and non-polyimide material as evidenced by the higher freqency of intermetallic fractre for polyimide Pacity of non-polyimide shear strength data at early times prevented an accrate determination of the fnctional dependence Acceleration Model Thermal effect of bond degradation was described for both polyimide and non-polyimide cases by an activation energy of Q = 5 ev The effect of moistre was described for all cases by a proportionality of the acceleration factor to relative hmidity All the degradations observed in this stdy depended on the presence of moistre, since if hmidity were not present, even the highest temperatre/time stress sed in this stdy wold show virtally no degradation of bond strength Reliability Predictions of Model The reliability model smmarized in Table V can be sed in conjnction with the reslts

10 of package mechanics calclations of shear stress applied to the ball bonds in a plastic package or example, if it were known that the shear stress applied to a bond wold never exceed, say, 25 grams in service, and we have a reliability reqirement of, say, bond per million with interfacial fractre, then it is possible to se the model in Table V to calclate the time ntil this reqirement is violated Loci of constant times to failre are plotted as contors in the hmidity-temperatre plane in ig The parameters for the 25 off-pad polyimide case were sed It can be seen in ig that the time to failing the reliability reqirements we have set in the example is greater than 7 hors (more than years!) for typical service ambients Ths, while the moistre-indced bond failre mechanism discssed in this paper is mch more accelerated than the well-known prely thermal failre mechanism (accelerated by baking), nonetheless the moistre-indced mechanism does not pose any serios reliability jeopardy 6 RH 2 E8 hr E7 hr E-2 atm E6 hr E5 hr atm E4 hr E3 hr 2 atm atm hr 5 atm 4 atm 3 atm Temperatre (deg C) Iso-time to Bond per million failing Vapor pressre isobar Limits of typical service ambients Average city climates ig Model predictions for time to bond per million with shear strength < 25 grams Worst-case model parameters were assmed (polyimide, 25 off-pad) Also shown is the range of typical service ambients, and climates of varios world cities Water vapor pressre isobars are also sperimposed Acknowledgments The athors wold like to express their deep gratitde to Jeanie Vierra and Tim Boyer for the incredible amont of time and work they pt into rnning the factorial experiments Thanks are also expressed to Bertha Jones, Jlio Tribiana and Deepak Goyal for their analytic services, and to Gay Samelson for her technical spport in Epoxy Encapslated Devices," in 28th Annal Proc, Reliability Physics, March, pp [ 3 ] E Philofsky, "Intermetallic ormation in Gold- Alminm Systems," Solid State Electronics, Vol 3, pp39-3, 97 [ 4 ] W Gerling, "Electrical and Physical Characterization of Gold-Ball Bonds on Alminm Layers," in IEEE Electronic Components Conference, 984, pp3-2 [ 5 ] T Ramsey, C Alfaro, and H Dowell, "Metallrgy's Part in Gold Ball Bonding," Semicondctor International, pp98-2, April [ 6 ] R C Blish and L Parobeck, "Wire Bond Integrity Test Chip," in Proc 2st Annal Proceedings International Reliability Physics Symposim, 983, pp [ 7 ] M Shell-De Gzman and M Mahaney, "The Bond Shear Test: An Application for the Redction of Common Cases of Gold Ball Bond Process Variation," in 3th Annal Proc, Reliability Physics, March 2, pp [ 8 ] The more severe bond degradation seen after hors in 56/85 HAST as compared with that shown in Table III below is de to the fact that the material was assembled at a different time with different bonding parameters We will see in Section 5 that degradation in stress is very sensitive to bonder force settings [ 9 ] S Groothis, W Schroen and M Mrtza, "Compter Aided Stress Modeling for Optimizing Plastic Package Reliability," in Proc 23rd Annal Proceedings International Reliability Physics Symposim, 985, pp 84-9 [ ] W Nelson, Applied Life Data Analysis New York: Wiley, 982, pp73-92 [ ] D S Peck, "Comprehensive Model for Hmidity Testing Correlation," in Proc 24th Ann Int'l Reliability Physics Symposim, 986, pp44- [ 2 ] A two-dimensional least-sqares regression fit of log a verss log h and /T was done sing the transformation of Eq (3): log a = log a + p log h Q k ( / T) log a, p and Q are the coefficients which minimize the sm of sqares of deviation of the model from the data References [ ] T Hnd, "Thermosonic Gold-Ball Bond Accelerated Life Test," in Proc th Electronic Components & Technology Conference, May, pp [ 2 ] A Gallo, "Effect of Mold Compond Components on Moistre-Indced Degradation of Gold-Alminm Bonds