Observation of Subcritical Spall Propagation of aThermal Barrier Coating

Valter Sergo*,† and David R. Clarke*

Materials Department, College of Engineering, University of California, Santa Barbara, California 93106-5050

Observations are reported of the room-temperature propa-gation of a spalling failure mode of a thermal barrier coat-ing (TBC) from its bond coat after oxidation. The coating isa Y2O3-stabilized ZrO2 coating formed by electron-beamdeposition on a Ni-Co-Cr-Al-Y bond coat. The spall shapeevolution and stress redistribution as the spall propagatesare reported. The failure propagates primarily as an inter-face crack between the bond coat and the thermally grownaluminum oxide (TGO) formed on the underside of theTBC during oxidation. The observations are consistentwith subcritical propagation of an interface crack betweenthe TGO and bond coat assisted by the presence of mois-ture. An estimate of 9 J/m2 is made of the fracture resis-tance in air of the interface.

I. Introduction

AS WITH the development of other metal/ceramic systems,the structural reliability of thermal barrier coatings (TBCs)

is of utmost concern, and, therefore, extensive research hasfocused on the mechanisms of failure. Under normal circum-stances, if thermal coatings fail, they generally do so by spall-ing on cooling after thermal cycling.1 This is attributed to thedevelopment of thermal expansion mismatch stresses on cool-ing. Occasionally another type of failure occurs, one in whichthe TBC spontaneously spalls entirely from the alloy at roomtemperature within some indefinite period of time, which canvary from hours to months, after cooling. The implication ofthese failures is that they are associated with some form ofsubcritical crack propagation, but they are so infrequent as topreclude systematic study.

Quite serendipitously we observed one of these room-temperature spalling failures that was occurring at a suffi-ciently slow rate to make some measurements. Recognizing theunusual opportunity this particular time-dependent failure pre-sented, we examined its shape evolution and the stress redis-tribution in detail. Measurements were made of the three-dimensional shape of the spall and of the residual stressdistribution in the thermally grown aluminum oxide (TGO)formed on the underside of the TBC. Residual stress measure-ments were made using photostimulated Cr3+ luminescence-based piezospectroscopy. The methodology has been describedin detail elsewhere,2–5as has its application for the nondestruc-tive, noncontact measurement of residual stress in the TGOformed beneath thermal barrier coatings.6

The TBC examined was one deposited by electron-beam

evaporation of ZrO2 onto a plasma-sprayed Ni-Co-Cr-Al-Ybond coat on a PWA 1484 superalloy. After the first cycle ofoxidation for 2 h at 1121°C, the sample was inadvertentlydropped and would, under normal circumstances, not havebeen oxidized further. No damage was apparent, and thesample was subsequently oxidized for another 10 h at 1121°C.Some indeterminate number of days after cooling, the TBCbuckled and spalled from the edge (Fig. 1(a)). This time wassubsequently designated ast 4 0 d. Over the next severalweeks, the failure advanced in several distinct branches acrossthe coating until the spalling pattern shown in Fig. 1(b) devel-oped. (The term buckling is used here to describe the uplift ofthe TBC, a precursor to subsequent cracking and final separa-tion. Cracking and final separation are referred to as spalling.)

In the following sections, optical microscopy observations ofthe buckle and spall propagation are presented, together withmeasurements of its shape evolution and the residual stressredistribution.

II. Observations

The progression of failure over a period of 20 d is shown inthe sequence of photomicrographs in Fig. 1. During the courseof these observations, the relative humidity in the room was∼55%. The change in the size and shape of the spall can be seenby comparing the optical micrographs in Fig. 1. In the mea-surements described below, particular attention was paid to theextension of the buckle shown at higher magnification in Fig.2. With time, the buckle propagated to the left from the pointQ and then deflected by∼90°. The accompanying crack alongthe buckle ridge also bifurcated, with one branch extendingalong the buckle ridge in Fig. 2(b) to the point designated as O.With further exposure, the buckle continued to propagate, asdid the ridge line crack. Interestingly, the width of the buckleperpendicular to its propagation direction appeared to be ap-proximately constant. In Figs. 2(a) and (b), the buckle extend-ing from point Q was uncracked, but at some time between therecording of Figs. 2(b) and (c), the buckle boundary crackedat C.

To test whether the buckle extended in the absence of mois-ture, the sample was held for 3 d under dry nitrogen at zerohumidity. No propagation was noted during this period, but, assoon as the sample was again exposed to air, the buckle con-tinued to propagate.

(1) Buckling MorphologyTo quantify the shape of the extending buckle, the height of

the TBC surface was measured as a function of distance in thevicinity of the buckle. Because the heights were relativelylarge, they were measured by counting the rotations of theoptical microscope stage required to bring the surface into fo-cus as a function of position around the buckle. Measurementswere also made with a diamond stylus profilometer; they gaveessentially the same profile. Although measurement error wasrelatively large (±20mm), it was substantially smaller than theheight of the buckle. The height variation att 4 10 d of thebuckle propagating toward the lower left in Fig. 2 is shown inthe perspective three-dimensional graph of Fig. 3. The profile

D. B. Marshall—contributing editor

Manuscript No. 190557. Received November 20, 1997; approved March 27, 1998.Supported by Office of Naval Research under Grant No. NOOO14-97-1-0190.

Additional support provided to Dr. Sergo from the ‘‘Research Abroad’’ project fromthe University of Trieste.

*Member, American Ceramic Society.†Now with Materials Engineering Department, University of Trieste, Trieste, Italy.

J. Am. Ceram. Soc., 81 [12] 3237–42 (1998)Journal

3237

in the direction of the crack, OA, and perpendicular to thecrack, OB, are shown in Fig. 4(a). Also shown in Fig. 4(b) isthe profile along the line QC of the first buckle. A strikingfeature of these profiles is that the buckled portions are essen-tially straight rather than concave or convex, indicating that the

buckled material is not subject to any significant net bendingmoment. An estimate of residual strain in the coating prior tobuckling can also be obtained geometrically from the buckleheight and width. For instance, using the profile measuredperpendicular to the buckle propagation direction, i.e., alongthe direction OB, the compressive residual strain relieved bybuckling is ∼2.7%. Similarly, from the profile along the lineQC, residual strain is also calculated to be∼2.7%.

(2) Piezospectroscopic MeasurementsIn addition to the morphological measurements, a series of

piezospectroscopic measurements of residual stress in the TGOwere made in order to ascertain the strain energy release asso-ciated with failure propagation. Measurements were made us-ing an optical microprobe in which an argon-ion laser wasfocused onto the region of interest identified in the opticalmicroscope. Using a 10× objective lens, a spot size of∼50 mmcould be obtained on the sample surface. Luminescence fromCr3+ ions in the TGO excited by the laser beam was collectedby the same objective lens and passed through the attachedspectrometer to a charged couple device (CCD) detector forfrequency analysis.

Piezospectroscopy measurements were made as a function ofdistance from point O in Fig. 2 in two orthogonal directions,OA and OB. The method is shown schematically in Fig. 5.Previous work has indicated that it is possible to probe anattached TGO through an electron-beam-deposited TBC6 butnot through an additional air gap. Thus, luminescence could beobtained from the TGO shown schematically at A but not at B.In addition, measurements were also made from the ridge crackat point Q in the direction QC. The measurements were madeby traversing the sample with respect to the stationary opticalprobe using a computer-controlledX–Ystage fitted to the op-tical microprobe. The shift,Dn, in frequency of theR2 lumi-nescence line from its stress-free frequency was used to evalu-ate the biaxial stress as follows: For a randomly orientedpolycrystalline Al2O3, the general piezospectroscopic relation-ship2 can be written as

Dn =1

3~P11 + P22 + P33!~s11 + s22 + s33!

=1

3Pii~s11 + s22 + s33! (1)

whereP are the piezospectroscopic coefficients. For a biaxialstress state, i.e., fors 4 s11 4 s22, s33 4 0, the piezospec-troscopic relation of Eq. (1) can be written as

Dn =2

3Piis (2)

Fig. 1. Appearance of buckling and spalling when (a) it was firstnoticed and (b) 20 d later. R is a point of reference with which theextent of spalling can be assessed.

Fig. 2. Higher magnification images of one branch of the buckling when it was first noticed (left) and then at the days indicated.

3238 Journal of the American Ceramic Society—Sergo and Clarke Vol. 81, No. 12

where Pii is the trace of the piezospectroscopic coefficienttensor that has a value of 7.50 cm−1?GPa−1.7

The variation in residual biaxial stress along the length of thebuckle in the direction OA of the crack is shown in Fig. 6 att 4 0, 2, and 7 d. The location of the crack tip along the ridgeof the buckle at the three times is also noted. All three sets ofresidual stress data show similar behavior, namely that, in thebuckled portion, the stress is low (∼400 MPa), whereas, well

ahead of the buckle, the residual stress is very high (∼4 GPa).In these two identifiable regions, luminescence consists of twodistinct lines, corresponding to theR1,R2 doublet, and the fre-quency shift of theR2 line is plotted as the solid symbols in thefigure. In between the two regions, the luminescence can bedeconvoluted into four distinct lines, aR1,R2 doublet with asmall shift and a second doublet with a larger shift. The largershifts are plotted as the open symbols in the figure. Our inter-pretation of this luminescence is that it is recorded from thevicinity of the interfacial crack front with the luminescencebeing recorded simultaneously from both the cracked (buckled)and uncracked portions, i.e., highly stressed and less highlystressed regions. This region is designated as the decohesionfront in Fig. 6. (In this region, the stress is unlikely to be purebiaxial, and, therefore, the nominal biaxial stress is plotted inthe figure.) Examination of the data indicates that the decohe-sion front advances over the period of measurement from aposition 750mm in front of the initial crack tip (point O) to1100mm 2 d later and then to 1400mm 5 d later. Subsequently,no further advance of the decohesion crack occurred in thisarea of the spalling pattern, although buckling and spallingprogressed elsewhere. (Because the TBC is opaque, the deco-hesion front and the buckle boundary cannot be observeddirectly.)

Fig. 3. Three-dimensional plot of the height of the TBC surface inand around the buckle propagating along the direction OA in Fig. 2.Crack tip is point O in Fig. 2.

Fig. 4. Height profile along (a) lines OA and OB and (b) line OC inFig. 2.

Fig. 5. Schematic diagram of the optical probing of the residualstress distribution in the thermally grown Al2O3 beneath the TBC andalong the buckled regions. Spectra can be obtained through the TBCfrom the Al2O3 attached to the TBC, such as at point A. Experimentsindicate that the spectra could not be obtained from the Al2O3 attachedto the bond coat, such as at B, because of the air gap between theAl2O3 and the TBC.

Fig. 6. Variation in residual stress in the thermally grown Al2O3 asa function of position along the buckle in the direction OA. Distanceis measured from the initial position of the tip of the ridge crack, O.Abrupt rise in residual stress corresponds with the position of thedecohesion front, i.e., the buckle boundary.

December 1998 Observation of Subcritical Spall Propagation of a Thermal Barrier Coating 3239

The variation in residual stress in a direction perpendicularto the buckle (and ridge crack) direction is shown in Fig. 7. Aresidual stress distribution similar to that found along thebuckle direction (Fig. 6) is seen. The most important differenceis that the decohesion front advances in this case from a posi-tion 700mm from point O to a position 1600mm from it in thefirst 2 d, and then does not advance further. A similar variationin residual stress is observed along the line QC, but no advanceof the buckle along this direction is noted.

The residual stress (∼4 GPa) in the TGO in the regions of theTBC well ahead of the buckle boundary is the same as theintact region elsewhere on the sample, suggesting that the re-gion ahead of the buckle is undamaged. The numerical value ofthe residual stress in the Al2O3 is similar to that measured onother TBCs:6 it is a combination of the thermal expansionmismatch stress, primarily with the underlying superalloy, andgrowth stress in the oxide. The compressive residual stress,s0r,in the Al2O3 attached to the buckled portions of the TBC isalmost constant from one position to another, at∼400 MPa.The elastic modulus of electron-beam-deposited TBCs, in par-ticular their biaxial modulus parallel to the bond-coat/TBCinterface, is not known, but its columnar microstructure with ahigh density of intercolumnar voids and gaps suggests it is verylow, much smaller than the modulus of dense ZrO2 (∼200GPa). (A very high biaxial compliance is, in fact, a designrequirement of TBCs.) This supposition is supported, indi-rectly, by microhardness measurements made perpendicular tothe TBC surface.8 Hardness derived in this way is∼0.75 GPa,as compared to the hardness of dense ZrO2 of ∼10 GPa. Theindentations also exhibit considerable crushing, suggesting thatthe true hardness is even lower.

In fact, the elastic modulus of the TBC,ETBC, can be esti-mated from the value of the residual stress in the Al2O3 TGOphase,s0r, after separation from the bond coat but remainingattached to the TBC. This is the value shown in Fig. 6 for theregions that have buckled. Using the standard thermal stressequation for a composite laminate,

s0r =E0

1 − n0

~a0 − aTBC!

1 + RDT

(3)

whereR is the ratio

R =E0

ETBCS1 − nTBC

1 − n0D h0

hTBC (4)

and whereE0 is the modulus of the Al2O3 (∼380 GPa),n0 thePoisson ratio, andh0 and hTBC the thicknesses of the Al2O3(∼0.5 mm) and the TBC (∼140 mm), respectively.DT is thedifference between the temperature at which the TBC is de-

posited (∼1000°C) and room temperature. Using these values,and assuming that the Poisson ratio of the Al2O3 and the TBCare the same (∼0.25), the biaxial elastic modulus of the TBC isestimated to be∼1 GPa.

(3) Fracture Surface ObservationsAfter a portion of the spalled TBC had completely separated

from the sample, the fracture surface revealed by the separationwas examined by both luminescence and scanning electronmicroscopy. By probing the luminescence on the fracture sur-face on the alloy side, we determined that it was partiallycovered witha-Al2O3 with other regions free of any Al2O3.The fracture surface is shown more clearly in the scanningelectron micrographs of Fig. 8, which reveal portions of barebond coat with the rest of the surface covered with fracturedAl 2O3. (These assignments were confirmed by energy-dispersive X-ray analysis.) There were two other notable fea-tures of the exposed surface. One was the appearance of grainimpressions in the bare bond coat, interpreted to correspond tothe grains in the matching Al2O3 phase. The other surprisingfeature was that the Al2O3 surface exhibited no fracture mark-ings but rather had the appearance of a free oxide surface.Examination of the underside of the TBC spalled from the

Fig. 7. Variation in residual stress in the thermally grown Al2O3 asa function of position along the direction OB, perpendicular to thebuckling direction.

Fig. 8. Surface on the alloy side exposed by the spalling away ofthe TBC: (a) low-magnification image revealing that the surface isprincipally covered by the thermally grown oxide but also has re-gions in which the bond coat is exposed (the bright regions) and (b)higher-magnification image showing grain imprints in the bond coatand the surface of the thermally grown Al2O3. (Scanning electronmicrographs.)

3240 Journal of the American Ceramic Society—Sergo and Clarke Vol. 81, No. 12

sample revealed it was principally ZrO2 with patches of Al2O3,as shown in the low-magnification image in Fig. 9(a) takenfrom close to the edge of one of the spalled flakes of TBC soas to also image the ZrO2 columns of the TBC. Althoughquantitative comparisons were not made, the areal density ofthe Al2O3 patches corresponded to that of the bare bond coatseen on the exposed surface on the alloy. Closer examinationrevealed that the majority of the Al2O3 surface had the mor-phology of a free surface with the angular features of individualgrains clearly delineated as shown in Fig. 9(b). In the image, aselsewhere on the underside of the TBC, the morphology of theZrO2 surface suggested that it also had been a free surface andexhibited no indications of being a fracture surface. Takentogether with observations of the Al2O3 surface on the exposedalloy, these observations suggested that, before failure by TBCbuckling, there was actually very little contact between theTBC and the underlying bond coat. Then, when failure oc-curred, it was by sequential fracture, at the bond-coat/TGOinterface, of the fully bonded regions. The process is shownschematically in Fig. 10. The reason for the areas in whichthere was no bonding between the ZrO2 TBC and the TGO onthe bond coat is not known. However, one can surmise that itwas due to the densification of the metastable Al2O3, initiallyformed in the heating stages prior to coating, as it transformedto a-Al2O3 after TBC deposition.

III. Discussion and Implications

Optical microscopy observations and the associated topo-graphic measurements indicate that the TBC failure propagatesfirst by the extension of a buckled region followed by crackingat the ridge of the buckle and then of the coating at the buckleboundaries. Once ridge and boundary cracking of the TBChave occurred, the coating is then free to detach. Failure propa-gation is thus macroscopically similar to that observed in thefailure of highly stressed thin films under compression, such ascoatings9 and diamond films on silicon substrates.10 Examina-tion of the surfaces exposed by spalling reveals that bucklingpropagates by the successive failure of those regions of TGOattached to both the bond coat and the TBC. This is supportedby the matching morphology of the patches of Al2O3 on theTBC and the bare bond coat on the alloy side and, in turn,suggests that the failure occurs at the Al2O3/bond-coat interfaceand does so by moisture-enhanced subcritical crack growth,because the Al2O3 grain features are preserved. The propaga-tion of the buckle by a subcritical crack growth mechanism ismanifest not only in the appearance of the buckle with time butalso in the redistribution of the residual stress in the TGO withtime (Figs. 6 and 7).

Piezospectroscopic measurements of the residual stress andscanning electron microscopy observations provide the quan-titative data needed to estimate the fracture energy of the TGO/bond-coat interface under propagation conditions. Optical mi-croscopy observations indicate that the spall propagates underapproximately steady state and that the width of the buckledregion remains approximately constant. Under such condi-tions, we idealize the propagation as shown in the plan view inFig. 11 and apply the mechanics of buckling worked out inconsiderable detail for different types of buckling.11–14Of theanalyses, that by Thouless14 is most appropriate, because itincludes the propagation of a crack along the buckle ridge aswell as the buckling itself. Using this analysis and recognizingthat, when the buckle extends by a distanceDx under steady-state conditions, the elastic strain energy in the TGO is partiallyrelieved and the interfacial fracture energy can be obtainedfrom the difference in elastic strain energy between an area2b(Dx) ahead of the propagating buckle and the same area afterbuckling,

GE =h~1 − n2!

2E Fs0 −s1

4 G2

f(5)

Fig. 9. (a) Underside of the TBC after spalling away. Surface isprincipally ZrO2 with patches of attached thermally grown Al2O3. (b)Higher-magnification image reveals the ZrO2 surface is porous withno indication of fracture and the alumina surface has the granularappearance of an oxide/metal interface formed by oxidation of a metal.

Fig. 10. Schematic representation of the TBC/Al2O3/bond-coat sys-tem before and after fracture (not to scale). Postspalling microscopyobservations (Figs. 8 and 9) suggest that, before failure, the TBC wasattached only to the Al2O3 over a small fraction of the entire coatedarea. Failure then occurred by the successive fracture of the intactregions at the bond-coat/Al2O3 interface.

December 1998 Observation of Subcritical Spall Propagation of a Thermal Barrier Coating 3241

where

s1 =p2E

12~1 − n2!Sh

bD2

(6)

and wheres0 is the residual stress in the film prior to buckling,s1 the critical buckling stress,E the elastic modulus,h the filmthickness, andb the half-width of the buckle. The additionalterm f is the fraction of the total buckle area that fractures andis included to account for the observation that only a portion,f,of the TBC/TGO/bond-coat interface is apparently intact be-fore failure. From microscopy observations, this factor is de-termined to be 0.22 using a commercial image analyzer. Fromthe piezospectroscopic measurements and calculations of thecritical buckling,s0 >> s1, Eq (5) can be approximated as

GE 4s0

2h~1 − n2!

2Ef

(7)

This expression can be evaluated by assuming that the elasticstrain energy in the ZrO2 TBC is negligible. This is reasonabledespite its significant thickness because of its exceptionallylow elastic modulus. There are, unfortunately, no values in theopen literature for the biaxial elastic modulus of electron-beam-deposited TBCs. (In fact, this remains an unansweredchallenge, because the in-plane biaxial modulus determines thecompliance of the TBC when subjected to the very large ther-mal mismatch strains.) However, as described in Section II(2),the value of the residual stress in the Al2O3 remaining attachedto the TBC is consistent with a biaxial modulus of the TBC of∼1 GPa, which is considerably smaller than the modulus ofdense Y2O3-stabilized ZrO2. This low value for the biaxialmodulus of the TBC is also consistent with the buckle profiles(Fig. 4), which show that the buckled material is straight and,hence, not under appreciable bending moment. The piezospec-troscopic measurements indicate that the residual biaxial stressin the TGO in the intact region is∼4 GPa, whereas after thebuckle front passes, the residual biaxial stress falls to∼0.4 GPa.Assuming that the elastic modulus of the TGO is∼380 GPa andits thickness is 0.5mm, and using the expression above, theinterfacial fracture energy is estimated to be∼9 J?m−2. Thiscompares quite fortuitously with the fracture energy of thenickel/sapphire interface of 10 J?m−2.15 Although there are sev-eral other reports in the literature of the fracture resistance ofmetal/ceramic interfaces,16–17 including observations of sub-critical crack growth in the presence of moisture,17 the authors

are unaware of any other work on the fracture resistance of thenickel/Al2O3 interface, the interface pertinent to the oxide/bond-coat interface.

For some reason, presumably associated with its depositionas discussed above, the TBC in this particular instance is notwell-attached to the bond coat and, as a result, may not betypical of high-performance TBCs. This may be why it is pos-sible to observe the slow propagation of the TBC in this case.However, the relatively low fracture energy of the Al2O3/bond-coat interface and its susceptibility to moisture-enhanced crackgrowth suggest that it is of utmost importance to shield thisinterface from moisture. In retrospect, it is possible that thefailure observed was initiated when the sample was dropped,thereby damaging the TBC at the edge and exposing the Al2O3/bond-coat interface. This suggests, in turn, that systematicstudies of subcritical growth in TBC systems at different hu-midities be undertaken, together with the use of deliberatedlyintroduced controlled defects to initiate failure under pre-scribed conditions.

IV. Conclusions

Room-temperature time-dependent failure of an electron-beam-deposited ZrO2 TBC occurs over a period of severalweeks. By combining optical and scanning electron micros-copy observations with piezospectroscopic measurements, it isconcluded that the observed macroscopic failure by bucklingand subsequent spalling of the TBC is mediated by subcriticalcrack growth along the TGO/bond-coat interface, presumablyenhanced by the presence of moisture. Fracture energy of theinterface is estimated to be∼9 J?m−2.

Acknowledgments: The authors are grateful to Dr. Tolpygo for invalu-able discussions, to Dr. C. Schmid, (University of Trieste) for the image analy-sis, and to Professor M. Thouless (University of Michigan) for his suggestionsafter reviewing the original manuscript.

References1Coatings for High Temperature Structural Materials.National Academy

Press, Washington, DC, 1996.2Q. Ma and D. R. Clarke, ‘‘Stress Measurement in Single Crystal and Poly-

crystalline Ceramics Using Their Optical Fluorescence,’’J. Am. Ceram. Soc.,76 [6] 1433–40 (1993).

3Q. Ma and D. R. Clarke, ‘‘Piezo-Spectroscopic Determination of ResidualStresses in Polycrystalline Alumina,’’J. Am. Ceram. Soc., 77 [2] 298–302(1994).

4D. M. Lipkin and D. R. Clarke, ‘‘Sample–Probe Interactions: Sampling Mi-croscopic Property Gradients,’’J. Appl. Phys., 77 [5] 1855–63 (1995).

5D. M. Lipkin and D. R. Clarke, ‘‘Measurement of the Stress in Oxide ScalesFormed by Oxidation of Aluminum-Containing Alloys,’’Oxid. Met., 45 [3/4]267–80 (1996).

6R. Christensen, D. M. Lipkin, D. R. Clarke, and K. Murphy, ‘‘Non-Destructive Evaluation of Oxidation Stresses through Thermal Barrier CoatingsUsing Cr3+ Piezospectroscopy,’’Appl. Phys. Lett., 69 [24] 3754–56 (1996).

7J. He and D. R. Clarke, ‘‘Determination of the Piezo-Spectroscopic Coeffi-cients for Chromium-Doped Sapphire,’’J. Am. Ceram. Soc., 78 [5] 1347–53(1995).

8P. Heydt, M.S. Thesis, University of California, Santa Barbara, CA, to bepublished.

9G. Gille, ‘‘Strength of Thin Films and Coatings’’; pp. 420–72 inCurrentTopics in Materials Science, Vol. 12. Edited by E. Kaldis. North Holland,Amsterdam, The Netherlands, 1985.

10D. Nir, ‘‘Stress Relief Forms of Diamond-like Carbon Thin Films on Sili-con,’’ Thin Solid Films, 112, 41–49 (1984).

11A. G. Evans and J. W. Hutchinson, ‘‘On the Mechanics of Delaminationand Spalling in Compressed Films,’’Int. J. Solids Struct., 20, 455–66 (1984).

12J. W. Hutchinson and Z. Suo, ‘‘Mixed Mode Cracking in Layered Materi-als,’’ Adv. Appl. Mech., 29, 63–91 (1992).

13M. D. Thouless, J. W. Hutchinson, and E. G. Liniger, ‘‘Plane Strain, Buck-ling-Driven Delamination of Thin Films,’’Acta Metall. Mater., 40, 2639–49(1992).

14M. D. Thouless, ‘‘Combined Buckling and Cracking of Films,’’J. Am.Ceram. Soc., 76 [11] 2936–38 (1993).

15J. Stolken and A. G. Evans, to be published.16‘‘Proceedings of the International Symposium on Metal/Ceramic Inter-

faces’’ (special issue),Acta Metall. Mater., 40, S1–S368 (1992).17T. Oh, R. M. Cannon, and R. M. Ritchie, ‘‘Sub-critical Crack Growth

along Ceramic-Metal Interfaces,’’J. Am. Ceram. Soc., 70 [12] C-352–C-355(1987). h

Fig. 11. Schematic representation of a buckle propagation used toestimate the crack driving force and, hence, the interfacial fractureresistance.

3242 Journal of the American Ceramic Society—Sergo and Clarke Vol. 81, No. 12