ARTICLE IN PRESS

0026-2692/$ - se

doi:10.1016/j.m

�CorrespondE-mail addr

Microelectronics Journal 38 (2007) 842–847

www.elsevier.com/locate/mejo

Iterative optimization of tail breaking force of 1mil wire thermosonicball bonding processes and the influence of plasma cleaning

J. Leea,�, M. Mayera, Y. Zhoua, S.J. Hongb

aMicrojoining Laboratory, Department of Mechanical and Mechatronics Engineering, Center for Advanced Materials Joining,

University of Waterloo, Waterloo, Ontario, CanadaN2L 3G1bMK Electron Co. Ltd., Yongin, South Korea

Received 23 May 2007; accepted 15 July 2007

Available online 29 August 2007

Abstract

An online tail breaking force measurement method is developed with a proximity sensor between wire clamp and horn. The wire under

the tensile load measures about 1.5 cm extending from the bond location to the wire clamp. To increase the sensitivity, the bondhead

speed is reduced to 2mm/s during breaking the tail bond. It takes roughly 10ms to break the tail bond. The force resolution of the

method is estimated to be better than 5.2mN. An automatic wire bonder used to continuously bond up to 80-wire loops while recording

the on-line proximity signals. All wires are directed perpendicular to the ultrasound direction. The tail breaking force for each bond is

evaluated from the signal and shown automatically on the bonder within 2min after bonding.

Results are obtained for a typical Au wire and a typical Cu wire bonding process. Both wires are 25mm in diameter and bonded on Ag

plated diepads of standard leadframes at 220 1C. An average Cu tail breaking force of higher than 50mN is obtained if the leadframe is

plasma cleaned before the bonding with 100% Ar for 5min. This result is comparable to that obtained with Au wire. The standard

deviation of the Cu tail breaking force is about twice that obtained with Au wire. The tail breaking force depends on the bonding

parameters, metallization variation, and cleanliness of the bond pad. The cleanliness of the bonding pad is more important with Cu wire

than with Au wire.

r 2007 Elsevier Ltd. All rights reserved.

Keywords: Wire bonding; Thermosonic; Tail breaking force; Leadframe; Tensile test

1. Introduction

Wire bonding is a widely used technology for the firstlevel interconnection in microelectronic devices. Amongwire bonding technologies, thermosonic wire bonding isone of the most commonly used processes in electronicspackaging due to its stability, flexibility and cost effective-ness. The bonded wire carries current and heat between anintegrated circuit and a leadframe or a substrate [1].

High integration of IC chips and therefore finer pitchinterconnection is the current trend in wire bondingindustry [2]. However, fine and ultra fine pitch wirebonding can show poor wedge bondability [3] and wedgebond tailing [4]. The wedge bond sometimes is called

e front matter r 2007 Elsevier Ltd. All rights reserved.

ejo.2007.07.095

ing author. Tel.: +1519 888 4567; fax: +1 519 888 6197.

ess: j75lee@engmail.uwaterloo.ca (J. Lee).

second bond, crescent bond, or stitch bond. As thecapillary tip diameter decreases, the wedge bond sizedecreases. Therefore, the wedge and tail bondability isusually reduced.Fig. 1(a) and (b) defines the location of the tail bond.

The wire is pinched by the capillary separating the secondbond into tail bond and wedge bond as shown in Fig. 1(a).An example top view of the tail bond area after the wirehas broken from it is shown in Fig. 1(b). Tail bonds shouldbe strong enough to hold the wire before the clamp closes.After wedge bonding, the capillary moves up to the tailheight where it remains until the clamp closes. It then startspulling the wire to break it, resulting in a tail of apredefined length. However, if the tail bond is weak, it canlead to non-uniform tail length and therefore non-uniformfree air ball (FAB) formation. Sometimes, the bonder stopsbefore flaming off the tail because the tail bond was weak

ARTICLE IN PRESS

20µm

Crescent bond

Bonded wire

Tail bond area

Capillaryimprint

DiepadWire

Tail bond

CapillaryWiretail

Fig. 1. Schematic (a) and SEM image (b) of tail bond location.

Proximity sensor

Bonding material

HornCapillary

Clamp

Measured distance

z

Fig. 2. Schematic of online signal measurement using proximity sensor.

0 20 40 60 80 100 120 140-1000

-500

0

500

1000

Time [ms]

Prox

imity

sig

nal [

PSU

]

A

Ball bond Wedge bond Tail bond

Fig. 3. Signal measured with proximity sensor during bonding process.

J. Lee et al. / Microelectronics Journal 38 (2007) 842–847 843

enough to loose before the clamp could close, resultingin the wire being blown out from the capillary. This failureis sometimes called ‘‘Short tail’’ or ‘‘Tail lift-off’’. Thefrequent occurrence of such process stops reduces thethroughput of the wire bond production.

To improve the tail bond, modified capillaries withlarger inner radius [3] and with a new finishing process onthe tip surface which make deep surface morphology [5]have been developed. In Ref. [4], measurements of the tailbreaking force (TBF) or tail pull force with Au and Cuwires on the Ag and Au metallization are reported. Thesestudies conducted so far are important steps to understandand improve the tail bonding. The present study isundertaken to further contribute to the understanding ofthe tail bond with a method to measure the TBF duringwire bonding process. Initial measurements of the TBF arecarried out and the TBF maximization by optimizing theprocess parameters is demonstrated with a Au wire. Theeffect of metallization and cleanliness on the TBF areinvestigated for both Au and Cu wire bonding processes.

2. Tail breaking force measurement

An ESEC WB 3100 wire bonder and PLCC 44leadframes with Ag metallization are used for the bondingprocess. The thickness of Ag layer is 8mm. A 99.99% purity25mm diameter Au wire available from MK Electron Co.Ltd., Yongin, South Korea, is used to measure the TBF.The capillary used has a tip, chamfer and hole diameter of100, 51 and 35mm, respectively. A proximity sensor is

attached on the clamp to measure real-time signalscorresponding to the gap between clamp and horn asshown in Fig. 2. This sensor is a standard component of thewire bonder and mainly used to detect touchdown duringbonding.The parameters that control the breaking of the wire

(‘‘tailing’’) after the second bond are set to 200 mm and2 mm/ms of Z tear distance and speed, respectively. Theselected speed is slow enough to time-resolve the tailbreaking signal. Fig. 3 shows a signal measured with theproximity sensor during one wire cycle. The processparameters impact force (IF) and bonding force (BF) forthe ball bond are 600 and 200mN, respectively, and forwedge bond are 1000 and 800mN, respectively. The IFparameter is similar to the contact velocity (CV) parameteravailable on other wire bonder types. The auxiliary unit‘‘PSU’’ is introduced for the proximity sensor signal. Thisunit corresponds to mN when the capillary presses on asurface causing the horn to deform slightly. This isdifferent during the tailing when the clamp deforms slightlyby a three to four times smaller amount per N compared tothe horn. The TBF values are obtained in mN from theproximity sensor signal by multiplying a calibration factor,fTBF ¼ 3.5mN.The tail breaking portion of the signal is indicated by the

letter A in Fig. 3. It is enlarged and shown in Fig. 4. Thesignal increases before the tail breaks as the wire tensionincreases. After it reaches a maximum value, the signalsuddenly drops to zero indicating tail breaking. If thesignal is measured without the wire, no change is observedin this period. In the measurement with the wire, the timepoint of the tail break indicated by (c) in Fig. 4 is found

ARTICLE IN PRESSJ. Lee et al. / Microelectronics Journal 38 (2007) 842–847844

automatically by using an evaluation software that locatesthe time of the strongest decay (minimum derivative) of thesignal. Two portions of the signals are averaged, the firstbefore the tail break, the second after, as indicated by(a) and (b) in Fig. 4. The portions (a) and (b) last 3 and8ms, respectively, and are taken 1ms prior and 8ms afterthe time of (c), respectively. The average difference betweenthem is defined to be the TBF signal (STBF), 17 PSU in theexample. An estimation of the TBF signal resolution sS isobtained by firstly determining standard deviations ofsa and sb from the signal regions (a) and (b), respectively,and then using

sS ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffis2a þ s2b

q. (1)

The value of sS is found to be 1.5 PSU whichcorresponds to an estimated TBF resolution ofsS ¼ 5.2mN.

0 5 10 15 20 25 30 35-10

0

10

20

Time after wedge bond (ms)

Prox

imity

sig

nal [

PSU

]

STBF(a)

(b)Signal without wire

(c)

Fig. 4. Example measurement of TBF (a) with and (b) without wire.

Fig. 5. Wire bond used for iterative optimization of tail breaking force.

Table 1

Example of three runs during the first iteration

Run US (%) IF

Wire 1 Step Wire 20 W

First iteration 1 31 3 88 10

2 50 0 50 3

3 50 0 50 6

Optimum parameter 50

3. TBF maximization by iteration

Bonding is performed on the diepad of the leadframe asshown in Fig. 5. Each wire has different bondingparameters. The three bonding parameters, ultrasound(US), IF, and BF, are varied. The parameters, bondingtemperature (T) and bonding time (BT), are fixed to 220 1Cand 15ms, respectively. The unit ‘‘%’’ is used for the USparameter, where 1% is equivalent to a peak-to-peakvibration amplitude of 26.6 nm measured at the center ofthe transducer tip.A series of iterations is carried out each consisting of

optimization runs for US, IF, and BF. Data from oneexample iteration are shown in Table 1. In this example,IF(0)

¼ 1000mN and BF(0)¼ 400mN are the starting

parameters obtained by a few initial trial and errorbonding tests, and US(1), IF(1), and BF(1) are the optimizedparameters that result from this iteration. In the 1stoptimization run of the iteration, the parameters,IF ¼ IF(0) and BF ¼ BF(0) are fixed and US is varied in3% steps from 31% to 88%, which is the bondabilityrange. For US values lower than 31%, wedge lift-offoccurs. For US values higher than 88%, short tail occurs.TBF is measured for each of US parameters and that withthe highest TBF is selected. This optimized value isUS(1) ¼ 50%. It is used for the next run which varies IFand fixes US ¼ US(1) and BF ¼ BF(0). An optimized valueIF(1) is obtained and is used for the third run for the BFoptimization. Consequently, a first optimized set ofparameters is found, US(1) ¼ 72%, IF(1)

¼ 600mN, andBF(1)

¼ 400mN, concluding the 1st iteration. The seconditeration uses this optimized set as starting parameters andproceeds in the same way, resulting in an optimized set ofparameters, US(2), IF(2), and BF(2). Subsequent iterationsare carried out until TBF values do not increase anymore.The final optimized set of parameters is called centerparameters.A total of 20 bonds are made with each parameter

combination. Table 2 summarizes the optimized parameters,average TBF ðTBFÞ, TBF standard deviation (s), and TBFprocess capability index (cpk) after each iteration. Tocalculate the cpk value, a lower specification limit (LSL) of10mN is chosen. The equation for cpk value calculation is

cpk ¼TBF� LSL

3s. (2)

(mN) BF (mN)

ire 1 Step Wire 20 Wire 1 Step Wire 20

00 0 1000 400 0 400

00 50 1250 400 0 400

00 0 600 300 50 1250

600 400

ARTICLE IN PRESS

Table 2

Optimized parameters, maximized TBF, evaluated s and cpk for each

iteration

Optimum parameters TBF (mN) s (mN) cpk

US (%) IF (mN) BF (mN)

First iteration 50 600 40 50.9 5.95 2.35

Second iteration 36 800 600 53.75 5.09 2.85

Third iteration 72 550 1150 55.34 4.73 3.2

Fourth iteration 72 450 950 62.22 2.18 7.97

TB

F (m

N)

TB

F (m

N)

TB

F (m

N)

20 40 60 80 10050

52

54

56

58

US (%)

200 400 600 800 1000 1200 140055

56

57

58

59

60

61

IF (mN)

200 400 600 800 1000 1200 140056

57

58

59

60

61

62

63

BF (mN)

Fig. 6. The fourth iteration results of Au wire TBF: (a) US, (b) IF, and (c)

BF. Gray lines are polynomial fits. Error bars are e ¼ (s/n�1)1/2, n ¼ 20.

J. Lee et al. / Microelectronics Journal 38 (2007) 842–847 845

From the first iteration to the fourth iteration, TBFincreases while s decreases. With the fourth iteration,US(4), IF(4), and BF(4) are found to be 72%, 450mN, and950mN, respectively. TBF, standard deviation (s), and cpkvalue at the optimized parameters are 62.22mN, 2.18mN,and 7.97, respectively.

Fig. 6(a)–(c) shows the fourth TBF iteration results ofAu wire. The gray solid lines are polynomial fits obtainedwith US, IF, and BF, respectively. Polynomial fitting isapplied in the response surface method widely used inoptimizing the wire bonding process [6,7]. The responsesurface method is typically conducted with data collectedfrom processes with parameters being varied among two orthree parameter values (2n or 3n design of experiment). Toofew parameter values can oversimplify the effect ofbonding parameters on TBF. The iterative optimizationmethod applied in this study reveals the degrees of thepolynomials required to suitably fit the experimentalvalues. In this example, the minimum degrees of poly-nomial suggested fits for the US, IF, and BF parametersare 3, 1, and 3, respectively. The variation of TBF with USis shown in Fig. 6(a). The TBF increases gradually from45.8 to 52mN and then decreases as US is increased. TheTBF decreases when IF is increased as shown in Fig. 6(b).In case of rising BF as shown in Fig. 6(c), the TBFincreases from 400 to 900mN and then decreases. Insummary, the changes of US, IF, and BF parametersinduce 10%, 7%, and 8% of TBF change, respectively.

In a similar way, the TBF of a Cu wire bonding processis maximized, resulting in an optimized set of parameters ofUS ¼ 80%, IF ¼ 1100mN, BF ¼ 500mN, and TBF, s,and cpk of 58.54mN, 7.1mN, and 2.28, respectively. TheCu wire used is 25 mm in diameter available from MKElectron, Co. Ltd., Yongin, South Korea. A shielding gasof 5%H2+95%N2 with a flow rate of 0.5 l/min is used forCu wire bonding process to eliminate the oxidation of freeair balls during EFO.

4. Bond surface cleanliness and TBF

Besides the bonding parameters, other factors such aspad metallization and cleanliness affect the TBF. Sincemany variables are to be considered in metal plating [8,9],the thickness and surface quality can vary over the surfaceand from sample to sample [10]. Therefore, the diepad

induced TBF variation and the diepad cleanliness areinvestigated.TBF measurements are carried out on five diepads fresh

from the box but not plasma cleaned, and on five diepadsplasma cleaned with 100% Ar for 5min. Before and after

ARTICLE IN PRESSJ. Lee et al. / Microelectronics Journal 38 (2007) 842–847846

plasma cleaning, the diepad to diepad variation of the TBFobtained with Au and Cu wires is compared. The resultsare shown in Fig. 7(a) and (b), respectively. Before plasmacleaning, the Cu wire TBF is between 48.5 and 60.12mN, arange four times larger than that of Au wire. After plasmacleaning, the diepad induced TBF variation of both Au andCu wires become as small as 1.4 and 4.2mN, respectively.

To further study the influence of plasma cleaning and theTBF process is carried out with the US parameter variedfrom 46% to 84% in steps of 4%. The BF, IF, BT, and Tparameters are fixed to 500mN, 1000mN, 25ms, and220 1C, respectively. They are the optimized parameters forthe Cu wire TBF process. The TBF comparison results areshown in Fig. 8. Each point is the average of 40measurements from bonds distributed over eight diepadsamples. Similar to the wedge bondability of Au wire onthe bonding pads [11,12], the TBF is increased after plasma

no plasma plasma

46

48

50

52

54

56

58

60

62

TB

F (m

N)

1a 2a 3a 4a 5a57

58

59

60

61

62

63

Number of diepad

TB

F (m

N)

1b 2b 3b 4b 5b

no plasma

1a 2a 3a 4a 5a 1b 2b 3b 4b 5b

plasma

Number of diepad

Fig. 7. Comparison of diepad induced TBF variation with and without

plasma cleaning: (a) Au wire and (b) Cu wire. Error bars are e ¼ (s/n�1)1/2,n ¼ 80.

cleaning on the diepad. This increase is larger with Cu wirethan with Au wire. The errors are in the range of0.79–1.11mN for the Cu TBFs obtained without plasmacleaning and they decreases to a range of 0.48–0.95mNwith plasma cleaning.These higher TBF and lower error values obtained after

plasma cleaning result in higher cpk values, which meansthat the process is more stable after plasma cleaning asshown for Au wire in Figs. 9 and 10. The error of cpk asexpressed with the error bars in Figs. 9 and 10 is calculatedwith the equation [13]:

scpk ¼ Z1�a

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

9nþ

cpk2

2ðn� 1Þ

s, (3)

where Z1�a is 1.96 at a 95% confidence level, and n ¼ 40 isthe sample size. With Au wire, a cpk above 9 is obtained

40 50 60 70 80 9040

45

50

55

60

65

US (%)

TB

F (m

N)

Cu wire without plasmaCu wire with plasma

Au wire with plasmaAu wire without plasma

Fig. 8. Effect of diepad cleanliness on Au and Cu wire TBF. Error bars

are, n ¼ 40.

40 50 60 70 80 900

5

10

15

US (%)

cpk

Cu, plasmaCu, no plasma

Fig. 9. Comparison of Au wire cpk values before and after plasma

cleaning.

ARTICLE IN PRESS

40 50 60 70 80 900

5

US (%)

cpk

Cu, plasmaCu, no plasma1

2

3

4

6

Fig. 10. Comparison of Cu wire cpk values before and after plasma

cleaning.

J. Lee et al. / Microelectronics Journal 38 (2007) 842–847 847

with 46–62% of US. It decreases to about 6 as US increasesabove 62%. Not much difference is shown in the cpk valuesbefore and after plasma cleaning. With Cu wire withoutplasma cleaning, the highest cpk values are 2.45 and 2.36and are obtained with 58% and 70% of US, respectively.After plasma cleaning, the highest cpk of 4.61 with Cu wireis obtained with 62% of US. Therefore, it is concluded thatplasma cleaning of an Ag plated bonding surface increasesthe tail bonding quality of a Cu wire bonding processsignificantly.

5. Conclusion

With a sensor measuring the distance between wireclamp and horn, the TBF can be measured with aresolution better than 5.2mN.

The TBF depends on the bonding parameter combina-tion. The highest TBF of 62.2272.18mN with Au wire isobtained at 72%, 450mN, and 950mN of US, IF, and BF,respectively. The cpk is 7.97.

Fitting polynomials to TBF values obtained with variousparameters, as it is used for surface response optimizationmethods, shows that the minimum degrees of a polynomialto fit well is 3, 1, and 3, for the parameters, US, IF and BF,respectively.

Using plasma cleaning prior to bonding, the tail break-ing stability of the Cu wire process increases significantly.

Acknowledgments

This work is supported by MK Electron. Co. Ltd.(Yonging, South Korea), Microbonds Inc. (Markham,Canada), NSERC, AuTEK, and OCE (all from Canada).The technical help of Oerlikon A&P is gratefully acknowl-edged (Swizerland, USA, and Singapore).

References

[1] G. Harman, Wire Bonding in Microelectronics: Materials, Processes,

Reliability, and Yield, McGraw-Hill, New York, 1997.

[2] S. Murali, N. Srikanth, C.J. Vath III, Effect of wire diameter on the

thermosonic bond reliability, Microelectron. Reliab. 46 (2006)

467–475.

[3] K.S. Goh, Z.W. Zhong, Two capillary solutions for ultra-fine-pitch

wire bonding and insulated wire bonding, Microelectron. Eng. 84

(2007) 362–367.

[4] J. Beleran, A. Turiano, D.M. Calpito, D. Stephan, Saraswati,

F. Wulff, C. Breach, Tail pull strength of Cu wire on gold and

silver-plated bonding leads, in: Proceedings of SEMICON, 2005,

pp. 1–8.

[5] D. Calpito, D. Alcala, Tail lift-off solution for fine pitch application,

in: Proceedings of SEMICON, 2004, pp. 1–5.

[6] M. Sheaffer, L. Levine, B. Schlain, Optimizing the wire bonding

process for copper ball bonding, using classic experimental design,

IEEE Trans. Compon. Hybrids, Manuf. Technol. CHMT-10 (1987)

321–326.

[7] J. Antony, Improving the wire bonding process quality using

statistically designed experiments, Microelectron. J. 30 (1999)

161–168.

[8] S. Lambert, The influence of temperature on the efficiency of

electroplating from various ionic liquids, Circuit world 32 (2006)

36–41.

[9] R.D. Srivastava, A.K. Agarwal, R. Sharma, Studies on electrolytic

Cu–Ni alloy plating, Surf. Technol. 9 (1979) 171–178.

[10] I. Kenny, J.C. Mitchell, G. Walsh, Gold plating on spectacle frames,

Phathal. Physiol. Opt. 17 (1997) 263–266.

[11] Y.H. Chan, J.K. Kim, D. Liu, P.C.K. Liu, Y.M. Cheung, M.W. Ng,

Effect of plasma treatment of Au–Ni–Cu bond pads on process

window of Au wire bonding, IEEE Trans. Adv. Packag. 28 (2005)

674–684.

[12] Y.S. Lin, A surface analysis on oxygen plasma-cleaned gold pattern-

plated substrates for wire bondability, Surf. Coat. Technol. 173

(2003) 47–57.

[13] N.E. Zhang, G.A. Stenback, D.M. Wardrop, Interval Estimation of

the process capability index, Commun. Stat.: Theory Methods 19

(1990) 4455–4470.