WORKING AROUND THE FLUORINE
FACTOR IN WIRE BOND: RELIABILITY
Jeanne Pavio, Robert Jung, Craig Doering, Randal Roebuck and Hario Franzone
Texas Instruments
Dallas, Texas
ABSTRACT;
Fivorine contamination on integrated cireuit chips has remained a critical _problen for wire bond
reliability, aggravated by the change from wet chemical etch on the part of the chip manufacturers.
Yet, we have found that the phenomenon can be the result of different conpounds with failure mechanisns
that vary with the particular fluorine contaminant. that happens to be present as well as the quantity of
that contaminant. Our study analyzes aluninun fluoride anc fluorocarbon polymer contaminatfon and their
associated failure modes on various hybrid products. It examines wire bond processing which, in some
eases, can overcone certain levels of contanination to produce strong wire bonds with good reliability.
INTRODUCTION TESTING AND RESULTS OF
WYBRIO A
In the midst of a successful history of wire We becane alerted to the wirebond problem when
bonding reliability on several different hybrids, approxinately thirty percent of electrical test
we first experienced wirebond problens as an failures could be attributed to resistive or
Jsolated instance on one particular hybrid. open wirebonds. Although compression and shepe
These problens were not evident at the assenbly of the bonds were normal, sectioned bonds
operation, and the wirebond schedule was in use exhibited extrenely limited intermetal lic fora
for various prograns being manufactured in the ation (See Fig. 1 and Fig. 2). The failures
seme agsenbly area, vere attributed to contamination of the bond pad
area.
Initial failures surfaced as wires on nybrids Imeciately an ine _
Hately an in-process non-destructive bond
vvich becane resistive during environmental derictestieae uactrtured te elartvieroceay
testing. Failure analysis of these units did Shecrepanctes dariay the! manufacturtnarcyede
not reveal gross intermetallic formation but {teelf ona dally besiss In additions Auger
did "reveal balls, isleh IS esses) analysis was completed to determine the nature
aluminum pad of the integrated circuit. Although aiegien radspal
there was ball bonding in other areas of the dictent of: the’ costal as Efen ubtie tony Ps
circuit, these balls were not resistive end not a iveipulticeestian at ie
ifte,” Further teeing of sanpe units with fovestad eataetrpnie fatTure of rencon bal
destructive bond pulls indicated that. random,
catastrophic failures (below 2.0 grams) of ini- erst Rrtr les ere, the oven Liceysrege pul)
strengths were in the 6 - 9 gram range, Failures
Ga gis sees io on pres ateata cae = prenironse te
Sta far Sri teres scat dlantese rorareyretiee eo C/OMnvUpenenimbiicrin otter arettiet the
with average reported pulls in the 6.0 to 9.0 the bonds. Failure mode for the majority of the
fram ranges fahet Cae ene BAT eiag fron the oslo pod
A test cycle of microsectioning, SEM analysis, tee ne
ESCA, and Auger was mobilized with. two major jer formed on unprocesse:
goalé:' to define the problen {tself, fsolating Eee ese iade tee tie, save weter site and
Tes use, any once defined) to ettingt to ne control sempie, a totally effferent 1ec. on
elresvent He wstfoe surfering in ret febi ity wihieh no failures had been experienced, Levels,
Gr in eireutt performance. of onygen and carbon were. comparable on the both
tte, The only other contaminant, present. was
enc fivarine on the die used to manufecture hybrid A
related problems on other hybrid-circuits with Our control sample registered no fluorine and our
varfous levels of contaminant present. Our test sample had a fluorine level at approximately
stucy examines attempts to achieve. good bonding Gs 'eh by Atonfe vetaht, Although the fluorine
in spite of the contaminant, utilizing an array a
2 wes present on the aluminum pads, none was re-
af thermocompression end thérmosonte bond Hartad on the reneriagttons
Eventually, we began to experience fluorine
The next approach was to perform an ESCA analysis
Through SEM photographs, microsections, bond to determine exactly what specific type of con
pulls and bond shears, we were able to target pound was involved In the contaminant, to attenpt
levels at which bonds were’ both strong (6.0 - ‘tp renove it, and to begin another lot of fabri-
9.0 grams) and rel fable cation of ICs.
428Binding energy was reported to be that of 3
Hlosropoiyecr, “in sdettion to the lvortdey the
ELEh detected oxides nttride and’ carbon tn the
fore of 2 hytrocason,
sain sah dh
InGefar FYuar ine. val, attempt
fuloydth 2 ater ea
ececamsucces-
ean, operafien. ana anceiugen
Splasea etch hugir analysis; taker after ‘each
‘ghdan-dperation d{d-not show. any reductianin,,.
Hgneversaiv"ar
from fabrication.
Fluorine levels at 3,0 - 4.0% Atmoic weight on
the AT pads.
‘Auger analysis recorded
The next step was to achieve good bonding fn
spite of this contamination level. Since
successful bonding is a tine/temperature
Phenomenon, it can be seen that an increase in
‘the dwell tine should achieve a positive effect
in bonding strength given maintenance of the
same temperature and pressure. However, no
change in internetalic formation was noted until
we increased bond dwell time to 4 seconds per
bond, which was not feasible for the production
volumes and delivery schedule. At the time,
‘our bonding profile.was strictly thermocon-
Pression with 2 240°C stage temperature, 300°C
capiliary and 55 grams of force using 1.0 mil
gold wire. Testing was performed using a
manual thermosonic bonder which was found to be
unsuccessful. All bonds exhibited limi ted
‘intermetallic formation upon microsectionings
_and within the test sample, approximately 72.
‘of the bonds failed destructive pull testing
(5 grams) at the ball/pad interface,
The final step was to develop a. thermocon-
pression bonding test mtrix. See Table 1,
Destructive bong pulls were completed initially
and after 2 300°C bake of four hours with 450
bonds pulled at each schedule. Results are
tabulated in Table II.
for all schedules the greatest faflure mechanism
¥G0:- 80x bonds) initiatly was wire breakage.
Hifter’the bake operation, Tifted ball. bonds
‘ecane the dominant faflure node in al] but
“Schedule D. Although lifted bond faflures
sficreased in (D) after baking, 55% of the bonds
Still failed due to wire breakage. Examining 2
bond lifts, we see that schedule D provided the
‘best performance in ball adhesion,
Bes
{iticrosections showed that only with schedule D
Werachteved good intenetallic formation across
oe of the bond consistently.
itew schedule was inplenented along with
sale non-destructive pull testing during the
fenbly. operation, sanple destryctive pull. =
Hing after 3 hoir bake at 400C, and Auger
Balysis of each new siice lot of integrated
429
circuits. Although the bond pads registered 2
fluorine ‘contamination level at 3.0 - 4.0% atomic
weight, with the new bonding schedule the resis-
tive bond problem was virtually eliminated,
TESTING AND RESULTS OF
HYBRID 8
typrid 8 was a thermosonfcally bonded hybrié with
2 history of randon wirebond failures.“ Inters
nittent all faflures which exhibited Tittle or
no intermetallic formation focused our attent on
to possible die pad contanination. An Auger
analysis was. then perforned on several devices
to identity possible surface contaminants, and
SIM photos vere taken of the bonding pad: andthe
corresponding ball bond (see Fig. 3 and Fig. 4).
She Auger analysis showed very Nigh carbon’ con~
tamination (60 - 80%) and agoroxinately 5%
Fluorine to.a depth of 150 Aon aluminum metal s~
zetion,. The Auger analysis for the pad and ball
Shown in Fig. Sand Fig. 4 respectively can be
found in Tobie TIT.
We imediately began investigating the history
involved with the die that we were using, and
found that they had been improperly stared for
over a year, lot and slice traceability had been
lost, and that these particular die had 2 history
of fluorine contanination similar to hyorid A,
Production bonding was stopped and a sampling
plan to test all of the die prior to releasing
them to production was instituted,
The die testing consisted of Auger and ESCA
analysis on a selected sample which was followed
by assembling and environmentally testing the
renainder of that sanple before the die would be
released to production. This type of initial
testing allowed us to establish threshold limits
for various contaninant levels. Once an adequate
data base was obtained with the above threshold
limits, testing could be reduced without sacri-
Ficing’ reliability,
‘To overcome, the excessive carbon contamination
on the die, an cxygen-plasna, cleaning epersticn
vas. performed Just prior. to bonding, This...
cliitnated post of the slirface contanination
present. atthe surface. ‘The higer and €SCk
analysis for two samples (Box 37 and Box 49)
Showed moderate Tevels of fluorine. contamination
and that the fluorine was bound as a metallic
Fluoride (binding energies for fluorine for Sox
37 and 49 were 685.8 and 685.4. respectively).
‘nce’ the binding energies for all of the elenents
were anelyzed and compared 4t was apparent that
the fluorine was bonded with aluninum. The
results of ‘the Auger and ESCA analysts can be
found in Table TVs
Eighteen hybrids were then built using the
standard assembly processes and sequence with
Box 37 and Box 43 die. These units would
establish the bondability of relatively low
contaminated metal fluorine die. After the 18
devices were bonded, 4 units (2"from Box 37 and
Zifrom Box 49)y were destructively bond pulled
“after they received the naraal nil-std environ~
ental conditioning. These devices were Used as
B control sample,The sequence of assenbly and environmental
conditioning can be found in Table V.
Six units were then baked for one hour at 300°;
delidded, and destructively wire pulled. This
conditioning would accelerate any die pad con-
tamination to. 2 failure point, and would allow
tus to study faflure mechanisms. The Fenaining
B units woulg then be baked an” additional. three
hours at 300°C, delidded, and destrudtively
pulled to further cull out any margirfal bonds.
After all of the devices were pull tested and
‘the data analyzed, it was now evident that
moderate metal. fluoride contamination could be
bonded and produce reliable bonds. The wire pull
analysis for the 18 units can be found in Table
ML.
Additional testing was then performed identically
‘to the previous units using die that had various
fluorine contamination levels to determine, if
possible, a maximum fluorine contamination’ level.
The three groups of die used in these tests were
Box X, Box 42, and Box $3, The Auger data for
these’ boxes’ can be found in Table VII.
Witn each increase in fluorine level, the wire
feilure mode 1, ball lifts, increased until the
device failgd the .5 om pull after the 4 hour
bake at 300°C, Microsections of the bonds
adjacent to ball failures, and pictures fron a
SEH (see Figs. 5, 6 & 7) showed the failure point
to be the first intermetallic layer between the
gold ball and the Au/Al intermetallic of the
bonded aluminun pad.
concLusroN
Comprehensive testing of Hybrids A and B defini-
tely show that the presence of fluorine on a
sinum bond pads fs detrimental to the reliability
of gold wire bonds. Both the bonding state of
the fluorine and the quantity of fluorine
present will affect the ball bond failure
mechanism.
Fluorine that is present asa fluorocarbon
polymer foras a barrier to intermetallic forma-
‘tion and inhibits bonding performed on standard
«bond sthedules, In spite of the barrier, a
thermoconpression schedule was developed’ that
would produce reliable bonds on pads containing
less than 4% fluorine.
Fluorine that is present as 2 metallic fluoride
increases the rate of intermetallic formation
rather than acting as a barrier to bonding.
‘Thus, initial bond strengths are high but
degrade severley over the life of the part. The
raxinum allovable anount of fluorine present as
a metallic fluoride was found to be 6. While
reliable bonding can be accomplished when sma?)
quantities of fluorine are present, ideally, the
bond pads should be free of fluorine. No
practical method of cleaning was found that
would renove either type of fluorine. It is
Clear that a team effort between both the hybrid
manufacturers and the integrated circuit manu
facturers will be needed to effect a long-term
solution to this problem.
tet THRE 01, AUGER ANALYSIS OF AGL FILIRE
poe waa pax snr | arome aearive «
4] ase nee foo mim 9 lowe es
£ [are Hee ee eh Set Se
rover tt see wpe ie 9 fess pst]
SHMLE IV, AUGER AND ESCA AWLISIS CF BOHES 97 6 69
sete 1.
Sone some eae An 30% ae
soo some [sao tirs Seow vires | mow Cine” ee LFTs
aa re aa we
: ne
ox| oasis | seers ‘srote agUaTIVE §
no| “tee feet} se UE at
ay | mae | 0 2a as
3 | kick | ako ER ee
» | Sol . 83 8 so)
| wr | oo ze me - |
G| Mat | fo 33 Be 2)
3 | Soe ss 83 BS wo]