Surface and Coatings Technology 113 (1999) 120–125

Improving the adhesion of amorphous carbon coatingson cemented carbide through plasma cleaning

Sam Zhang *, Hong XieGintic Institute of Manufacturing Technology, 71 Nanyang Drive, Singapore 638705, Singapore

Received 17 April 1998; accepted 30 November 1998

Abstract

Diamond-like amorphous carbon coatings 1 mm thick were deposited onto cemented carbide substrates by magnetron sputteringof a graphite target in argon under different substrate bias powers and chamber pressures. Scratch testing was used to assess thecoating adhesion. X-ray photoelectron spectroscopy depth profiling was employed to quantify cobalt loss at the substrate surfaceas a function of bias power during plasma cleaning. It was found that under the same deposition conditions, the scratch adhesionstrength increased with the bias power during plasma cleaning and reached a maximum at about 200 W or −210 V in terms ofinduced voltage. After that, further increases in bias power led to a decrease in adhesion. The increase was attributed to bettercleaning of the sample surface and removal of surface cobalt while the decrease in adhesion was linked to an increase in residualstress which resulted in a different failure mechanism. Thus, an increase in the deposition power density, and therefore moresevere ion bombardment, led to higher residual stress and lower adhesion. Under constant bias and deposition power, however,it was established that below a certain minimum chamber pressure spontaneous coating detachment occurs. © 1999 ElsevierScience S.A. All rights reserved.

Keywords: a-C; Adhesion; Chamber pressure; Diamond-like carbon; Plasma cleaning

1. Introduction during deposition on the adhesion strength between thea-C coating and the WC/Co substrate.

In recent years, various diamond-like carbon (DLC)or amorphous carbon (a-C ) coatings have found theirway into various industries. Industrial applications of 2. Experimental procedurethese coatings depend largely on their adhesion tosubstrates aside from other functional properties.

2.1. Plasma cleaning and coating deposition processCemented carbide is widely used as a tool material, thusa study of the bonding between the DLC coating and The deposition setup (Teer Coatings Inc., UK) istungsten carbide is of great industrial importance. shown schematically in Fig. 1. a-C:H coatings were

There are a number of ways to produce amorphous deposited on polished tungsten carbide ( WC–6%Co)carbon coatings [1]. In a PVD sputtering deposition discs 50 mm in diameter by d.c. magnetron sputteringprocess, important process parameters include chamber

of a graphite target in argon plus 20% hydrogen inpressure, deposition power density, bias power, etc. For terms of gas flow rate. The background pressure in thethis study, the unbalanced magnetron sputtering depos- sputtering chamber was below 6.7×10−5 Pa, and theition system [2,3] has been used in the deposition of

working pressure was between 0.1 and 1.5 Pa duringa-C coatings on cemented carbide ( WC–6%Co) with deposition. The substrates were placed in a rotary samplevarying chamber pressure, bias and target power. The holder facing the rectangular graphite target (99.9%,purpose of this paper is to study the effects of the in

GfE Gesellschaft fur Elektrometallurgie MBH,situ sputtering and the subsequent chamber pressure Germany) 330 mm by 133 mm in size, situated 85 mm

above the sample. A standard r.f. (13.56 MHz) biaspower (RFXII 1250, Advance Energy Industries, Inc.,* Corresponding author. Tel: +65 793 8577; Fax: +65 792 2779;

email: szhang@gintic.gov.sg USA) was applied to the substrate to provide power

0257-8972/99/$ – see front matter © 1999 Elsevier Science S.A. All rights reserved.PII S0257-8972 ( 98 ) 00832-9

121S. Zhang, H. Xie / Surface and Coatings Technology 113 (1999) 120–125

bias power, the failure became adhesive, as discussed indetail in the next section. Both ‘critical loads’ weredetermined via optical inspection of the scratch trackand the following calculation: let D be the total scratchdistance, d the distance traveled before the critical point,then the critical load Lc can be evaluated as

Lc=initial load+(maximum load−initial load)×d/D

(1)

where the initial load was the load exerted on the sampleat the beginning of scratching, i.e. at d=0; the maximumload was the final load applied onto the sample justbefore the scratching was stopped. Normally the initial

Fig. 1. Amorphous carbon was deposited on tungsten carbide viaload was 10 N and the maximum load was 120 N.sputtering of graphite target in magnetron sputtering system underHowever, should the critical load be below 10 N, i.e. ifsubstrate bias power.

failure occurred at the initial load, then the initial loadwould be lowered so that failure would not occur uponduring plasma cleaning and to eliminate charge accumu-initial loading. That, however, never happened in thislation during deposition. The r.f.-induced bias voltagestudy. Also, if the coating survived 110 N maximumon the substrate was varied from −36 V to −60 V and

the deposition time was set to 90 min to give a 1 mm load, then a higher maximum load, up to 140 N, wascoating thickness. To study the effect of plasma cleaning applied.on cobalt removal from tungsten carbide surface, a bias The cobalt loss at the surface of the substrate as apower ranging from 100 to 700 W was applied to the function of bias power during plasma cleaning wassamples for 20 min during plasma cleaning. After plasma studied by means of XPS depth profiling. The XPS wascleaning, the samples were retrieved for Co and W carried out on a VG ESCA Lab 220-IXL. Mg Kacomposition profile analysis using X-ray photoelectron emission was used as the X-ray source at 15 kV andspectroscopy ( XPS). The same plasma cleaning pro- 20 mA and argon gas was used as the etching medium.cedure was applied to identical samples before depos- The etching energy of the ions were fixed at 3 keV, andition took place. The power to the magnetron target the sample current obtained was 0.7 mA. The samplingwas supplied from the MDX magnetron drive from area was 1 mm in diameter. The constant analyzerAdvanced Energy Industries Inc., USA. During depos- energy (CAE) analyzing mode was used and the passition, the chamber temperature was around 300°C as energy was set at 100 eV. The binding energy profilesmeasured using an IR thermometer (MR-6015- for Co and W at different depths from the surface were06C-26231 from Ircon Inc., USA) through a viewing shown in Fig. 2. The Co 2p3/2 and W 4f peaks (bothport in the chamber wall. At higher bias power (500 W 4f7/2 and 4f5/2) were used to calculate the area underor more), the samples were allowed to cool in the

the peaks, giving rise to composition profiles with respectchamber for 10 or 20 min before deposition started.

to etching times (or depth) as shown in Fig. 3. Thecobalt profile was fitted to a parabolic curve, and the2.2. Adhesion and composition measurementsarea under the curve was calculated by integrating thecurve with respect to the depth. The area above theA frictional scratch adhesion tester (Teer Computercobalt curve up to the highest cobalt counts would beControlled 2200) was used to assess the adhesionproportional to the total loss of cobalt for the samplestrength. In the scratch test [4,5], an initial load ofin question. The area under the tungsten curve would10 N, up to maximum load of 140 N and loading ratebe proportional to the compositional content of tungstenof 120 N min−1 at a table speed of 10 mm min−1 werein the sample. As can be seen from Fig. 3, the loss inused. The friction force was recorded as a function oftungsten was relatively negligible after the initial surfacenormal load. The load at the first sign of coating failure,effect. Therefore, for simplicity and to a first orderregardless of mode of failure, was recorded as the ‘lowerapproximation, a constant value was taken for tungstencritical load’, and the load at the first sign of thein the area calculation. Comparing the area of ‘cobaltsubstrate surface or complete removal of the coating,loss’ (CL) with that of the ‘tungsten area’ ( WA) there-again, regardless of mode of failure, was recorded asfore gives:the ‘higher critical load’. In fact, the mode of failure

changed with deposition conditions: at lower bias power,the failure was of a cohesive type, while at very high Relative cobalt loss=CL/WA×100% (2)

122 S. Zhang, H. Xie / Surface and Coatings Technology 113 (1999) 120–125

Fig. 2. Binding energy profile of Co 2p3/2 and W 4f (4f7/2 and 4f5/2) electrons as revealed by XPS.

Fig. 3. XPS depth profile for cobalt and tungsten in a tungsten carbidesample plasma cleaned at 100 W bias power for 20 min.

Fig. 4. Plot of experiment conditions in the bias power–pressure space.At constant deposition power, too low a chamber pressure (or argon

3. Results and discussions flow) led to spontaneous peel off (open triangle) on tungsten carbideand silicon wafer substrate. Under the same conditions but higherchamber pressure, good adhesion was achieved.3.1. Effects of chamber pressure and substrate bias power

during depositionwith spontaneous peel-off. It can be seen that over thesputtering power studied (1.5–3.5 W cm−2), there areChamber pressure, bias power on the substrate and

the magnetron target power on the graphite target combinations of a minimum chamber pressure and amaximum bias voltage where peel-off was not observed.(‘target power’ for short) are the most important pro-

cessing parameters in sputtering depositions. Incorrect For instance, at around −40 V of induced bias, theminimum chamber pressure was 0.4 Pa; however, atcombinations of these conditions result in spontaneous

peel-off of the coatings. This can be conveniently repre- around 1.5 Pa the bias could be set around −50 Vwithout peeling off. This is easily understood in termssented in a two-dimensional space of bias power and

pressure at a given target power, as shown in Fig. 4, of kinetic energy (E ) of the activated sputtering ions,which is proportional to the bias voltage (Vb) andwhere each dot represents deposition under the condi-

tions specified and the crossed dot is for depositions inversely proportional to the root square of the chamber

123S. Zhang, H. Xie / Surface and Coatings Technology 113 (1999) 120–125

pressure ( p) [6 ]:

E3Vb/p1/2 (3)

Eq. (3) serves well in understanding the influences oncoating adhesion of bias voltage and deposition chamberpressure: all converge to the effect of ion energy.Increasing the chamber pressure decreases the mean freepath of the ions; therefore decreasing of chamber pres-sure promoted the collision of ions at the substratewhich leads to higher residual stress and spontaneouspeeling. The same reasoning applies when the bias powerincreases at fixed chamber pressure: increasing biasimparts more kinetic energy to the ions, thus promotingpeeling off.

3.2. Adhesion as a function of plasma cleaning power

Fig. 5 summarizes the ‘lower critical load’ of theamorphous carbon coatings on WC samples as a func-tion of r.f. bias power during plasma cleaning. Exceptfor the difference in bias power, all the samples weredeposited under the same conditions: 20% hydrogen gasflow; target power density, 1.30 W cm−2; chamber pres-sure, 1.3 Pa; induced bias voltage, −47 V; coating thick-ness, 1 mm. It can been seen that, under identicaldeposition conditions, adhesion of the coatings on tung-sten carbide varied greatly with the bias power used inplasma cleaning. Another important observation fromFig. 5 is that, under the plasma cleaning and depositionconditions used, the adhesion of all the coatings wasvery good: the lowest critical load is over 40 N, and thehighest is over 85 N, bearing in mind that these values

(a)

(b)

Fig. 6. Scratch tracks from the sample plasma cleaned at 200 W.are the so-called ‘lower critical load’, meaning the loadExcellent adhesion of the amorphous carbon coating is achieved onat the first sign of scratch failure, as indicated inpolished WC surface: (a) at 85 N, the coating starts to fail, (b) afterFig. 6(a). Also, for samples of the highest adhesion, 140 N, the coating is only partially failed.

even after testing up to 140 N, the coating was notseverely damaged as can be seen in Fig. 6(b). The failuremorphology and thus the failure mode seem quite different from those of TiN [7]. It can be seen that the

failure was cohesive instead of adhesive, i.e. the failurewas within the coating itself rather than in between thecoating and the substrate, another indication of goodadhesion.

3.3. Effect of the plasma cleaning

The effect of the plasma cleaning was understood byXPS depth profiling for cobalt loss during the cleaningstage. As discussed in Section 2.2, the ‘relative loss ofcobalt’ can be evaluated using Eq. (2). Plotting theresults against the substrate bias power during plasmacleaning gives Fig. 7. Therefore, an increase in theplasma cleaning power leads to an increased loss ofcobalt until an equilibrium is reached. This is becausecobalt is much easier to sputter than tungsten. ForFig. 5. Adhesion of a-C coating on WC/Co samples plasma cleaned atexample, under argon ion bombardment at 600 eV, theincreasing bias power followed by deposition under identical depos-

ition conditions. sputtering yields for W is 0.6 atoms/ion while that for

124 S. Zhang, H. Xie / Surface and Coatings Technology 113 (1999) 120–125

Fig. 7. Accumulated loss of cobalt at a surface of 10 nm depth incomparison with the tungsten content.

Co is 1.4. (p. 179 in Ref. [8]). However, in cementedFig. 8. Morphology of the start of scratch failure for a sample having

carbide ( WC–Co sintered body), the sputtering rate of undergone severe bombardment: the coating was peeled off piece byCo will decrease with time because the relative composi- piece, revealing substrate surface, typical of adhesive failure.tion of cobalt decreases with respect to tungsten, thusthe relative cobalt loss decreases to reach a dynamicequilibrium with regard to time or power. Therefore, the drop in adhesion strength for severely bombarded

samples could be attributed to increased interfacialthe increase in adhesion is the result of better ‘cobaltcleaning’. In fact, recently Ronkainen and coworkers stress. The proof is in the way the coating fails upon

scratching: as shown in the scratch morphology at the[9] demonstrated that, under constant bias power, anadhesion response similar to that in Fig. 5 was observed start of failure for samples which had undergone severe

bombardment (Fig. 8), the coating was peeled piece bywith plasma cleaning time. Plasma cleaning beforedeposition is a standard deposition procedure to remove piece (the white patches), typical of adhesive failure.

Aside from the residual stress contribution, too high asurface contamination, oxides, etc. for improved adhe-sion especially for thick coatings. In the case of tungsten sputtering power may induce impurities from the depos-

ition chamber walls to land on the substrate surface [9],carbide, however, plasma cleaning means more than justremoval of the surface contaminants, oxides, etc. and that, in turn, may also contribute to a decrease in

effective adhesion (or the lower critical load value inDepleting the cobalt on sample surface should be takenas an important task in pursuit of good adhesion of a-C scratch testing).on WC–Co.

However, when the cleaning power was increasedabove 200 W in our system, a drop in the lower critical 4. Conclusionsload was observed (see Fig. 5). As discussed inSection 3.1. on the effect of chamber pressure and power Experiments were carried out using magnetron sput-

tering deposition from a graphite target onto an r.f.-density, it is believed the same is true for the bias powerduring plasma cleaning: the higher the power the more biased substrate with the following results.

(1) At constant substrate bias power, too low a chambersevere is the ion bombardment and thus the greater isthe residual stress on the surface of the substrate on pressure (or argon flow) results in spontaneous peel-

off of a-C coatings on WC–Co substrates. Underwhich the coatings are to adhere. Direct measurementof stress in diamond-like carbon coatings on WC–Co is the same power condition, a higher chamber pres-

sure gives rise to better adhesion as a result ofextremely difficult owing to the amorphous nature ofthe coating ( XRD stress measurement obviously reduced residual stresses.

(2) At fixed chamber pressure, increasing substrate biasrequires the diffraction peak position) and the rigidnature of tungsten carbide (it is difficult to bend the power results in a decrease in adhesion of a-C on

WC–Co as a result of an increase in the interfacialsubstrate so the curvature method becomes inefficienthere). However, the results obtained for amorphous stresses.

(3) Proper plasma cleaning effectively removes cobaltcarbon on a silicon wafer using the curvature or bendingmethod [10] should apply for a-C on WC–Co: an from the WC–Co substrate surface and thus greatly

improves the adhesion strength of a-C. Too high aincrease in bias power results in an increase in residualstress [11,12]. It is therefore reasonable to believe that sputter power, however, gives rise to too much

125S. Zhang, H. Xie / Surface and Coatings Technology 113 (1999) 120–125

[3] D.P. Monaghan, D.G. Teer, P.A. Logan, I. Efeoglu, R.D. Arnell,residual stress that results in a decrease in adhesionSurf. Coat. Technol. 60 (1993) 525.of the coating.

[4] A.J. Perry, Surf. Eng. 2 (3) (1986) 183.(4) The influences of the chamber pressure ( p) and bias[5] S.J. Bull, D.S. Rickerby, Surf. Coat. Technol. 42 (1990) 149.

voltage (Vb) on the adhesion strength are readily [6 ] Y. Catherine, in: R.E. Clausing et al. (Eds.), Diamond and Dia-explained in terms of sputtered ion energy (E ) via mond-Like Films and Coatings, Plenum, New York, 1991,E3Vb/p1/2. pp. 193–227.

[7] S.J. Bull, Surf. Coat. Technol. 50 (1991) 25.[8] R.F. Bunshah et al., in: Deposition Technologies for Films andAcknowledgements

Coatings: Development and Applications, Noyes, Park Ridge,NJ, 1982.

This work was funded by Singapore National Science [9] H. Ronkainen, J. Vihersalo, S. Varjus, R. Zilliacus, U. Ehrnsten,and Technology Board via Gintic upstream project P. Nenonen, Surf. Coat. Technol. 90 (1997) 190.U96-P-059 in collaboration with the School of Applied [10] A. Grill, V. Patel, Diamond Related Mater. 2 (1993) 1519.

[11] A.J. Perry, M. Jagner, P.F. Woerner, W.D. Sproul, P.J. Rudnik,Science, Nanyang Technological University.Surf. Coat. Technol. 4344 (1990) 234.

[12] D.C. Yin, N.K. Xu, Z.T. Liu, Y. Han, X.L. Zheng, Surf. Coat.Technol. 78 (1996) 31.References

[1] S. Zhang, B. Wang, J.Y. Tang, Surf. Eng. 13 (4) (1997) 303.[2] D.P. Monaghan, D.G. Teer, K.C. Laing, I. Efeoglu, R.D. Arnell,

Surf. Coat. Technol. 59 (1993) 21.