VOLUME 85, NUMBER 4 PHYSICAL REVIEW LETTERS 24 JULY 2000

Energy Dissipation and Path Instabilities in Dynamic Fracture of Silicon Single Crystals
T. Cramer,1 A. Wanner,2 and P. Gumbsch1
1
Max-Planck-Institut f ür Metallforschung, Seestrasse 92, 70174 Stuttgart, Germany
2
Institut f ür Metallkunde, Universität Stuttgart, Seestrasse 71, 70174 Stuttgart, Germany
(Received 10 January 2000)
Brittle fracture usually proceeds at crack driving forces which are larger than those needed to create
the new fracture surfaces. This surplus can lead to faster crack propagation or to the onset of additional
dissipation mechanisms. Dynamic fracture experiments on silicon single crystals reported here show
several distinct transitions between different dissipation mechanisms. Cleavage fracture is followed by
the propagation of a faceted crack front, which is finally followed by a path instability and the propa-
gation of multiple cracks. The fracture surface qualitatively corresponds to the mirror, mist, and hackle
morphology of amorphous materials. However, the corresponding fracture mechanisms, which remain
largely unknown in the amorphous materials, can clearly be identified here.

PACS numbers: 62.20.Mk, 46.50. + a

An external load acting on a precracked body exerts a constant equal to the specific surface energy 2g of the two
driving force on the crack. This driving force is equivalent fracture surfaces created by the advancing crack.
to the mechanical energy release per unit crack advance, Fracture experiments on amorphous materials such as
the energy release rate G [1]. In an ideally brittle material polymethylmetacrylate (PMMA) and glass have revealed
crack extension takes place if the crack is supplied with a that cracks do not propagate faster than about 共0.4 0.5兲cR
driving force larger than the specific energy 2g required [4,5]. These experiments also show that the fracture sur-
to create the fracture surfaces [1,2]. This threshold driv- faces are only smooth and mirrorlike if the driving force is
ing force is called the critical energy release rate Gc . In close to Gc . If the driving force is increased and the crack
general, the driving force is not only a function of the ge- thus exceeds a critical velocity yc , the fracture surfaces be-
ometry of the body and of the external forces but also a come rougher and exhibit a so-called mist and hackle mor-
function of the shape, size, and orientation of the crack phology [6]. Below yc , Eq. (1) has recently been shown to
itself. Hence, the driving force may change and rise to appropriately describe the observed behavior [7]. Above
levels well above Gc as the crack length l increases dur- yc , microscopic bifurcations were proposed as the expla-
ing propagation. Interestingly, such supercritical fracture nation for the lower than expected terminal crack velocity
is observed in many practical cases and, from the experi- [7]. The bifurcations are predicted to result from a dy-
mentalist’s point of view, can be realized more easily than namic crack branching instability [8].
crack extension under an approximately constant driving So far most of the fundamental studies in this field have
force just above the threshold Gc . dealt with amorphous materials, where a propagating crack
How does the crack respond to a driving force that is can easily deviate from its original plane and thus form
significantly larger than required? How is the surplus of a rough fracture surface due to the absence of long-range
supplied energy “spent”? Continuum mechanical consid- order. In contrast, brittle crystalline materials exhibit well-
erations have so far only answered these questions for the defined cleavage planes, which render the formation of a
specific case of a crack propagating along a straight path rough fracture surface more difficult. Hence, the following
[3]. According to these considerations, a straight crack question arises: What is limiting the velocity of cracks in
is expected to attain increasingly higher velocities for in- brittle crystalline materials?
creasing energy release rates. The crack velocity asymp- The dynamics of fast fracture in brittle crystalline ma-
totically reaches an upper bound which is equal to the terials has been addressed experimentally only in very
Rayleigh wave velocity cR , the velocity of acoustic sur- few studies. Interestingly, most studies gave significantly
face waves. For a straight crack propagating at a velocity higher values of 共0.7 0.9兲cR for the terminal crack veloc-
y below cR , the dynamic fracture energy, G共y兲, can be ex- ities [9–13]. However, a correlation between the fracture
pressed as: surface morphology and the crack dynamics has not been
µ ∂ attempted yet and the origin of the terminal velocity in
y crystalline materials remains unclear.
G共y兲 苷 G共l, s兲 1 2 , (1)
cR The purpose of this paper is to provide a detailed time-
resolved analysis of the brittle fracture characteristics of
where the energy release rate G共l, s兲 is the static, time- single crystalline silicon and a correlation with the fracture
independent energy flux into the crack tip which represents surface morphology. This correlation and the quantitative
the geometry of the specimen and the applied stress s. In analysis by continuum theory allows one to clearly identify
the simplest case, the dynamic fracture energy G共y兲 is a specific energy dissipating mechanisms.

788 0031-9007兾00兾85(4)兾788(4)$15.00 © 2000 The American Physical Society

VOLUME 85, NUMBER 4 PHYSICAL REVIEW LETTERS 24 JULY 2000

Dynamic fracture experiments were conducted on com-

mercially available (boron-doped) silicon single crystal
plates which were loaded to force a 兵110其 cleavage crack
in a 具11̄0典 direction. This cleavage system is known to give
almost perfectly flat fracture surfaces [14]. The specimen
dimensions were height H 3 length L 3 thickness 苷
150 3 100 3 0.75 mm3 . Crack propagation was initiated
from a sawn notch by the application of a monotonically
increasing tensile stress. Crack extension was monitored
via the change of the electric resistance of a thin metal layer
sputter-deposited on one surface of the specimen. These
crack extension measurements were calibrated in situ by
means of a set of conductor lines deposited on the opposite
surface of the specimen. More experimental details are
given in Ref. [15].
The far-field stress s at which fracture occurred varied
between 4.7 and 15.5 MPa via the geometry of the notch
(length lc and root radius). The energy release rate for our
specific specimen geometry is almost constant for 0.4L #
l # 0.9L [16]. Within this steady-state regime, the energy
release rate is approximately equal to the energy release
rate expected for an infinitely wide strip which is given by
FIG. 1. (A) Measured instantaneous crack velocity y as a func-
s 2 H共1 2 n 2 兲 tion of the crack length. (B) Dependence of the measured crack
Gs 苷 , (2) velocity on crack driving force is compared to the theoretical
2E prediction from Eq. (1) for two specimens fractured at relatively
where n 苷 0.36 and E 苷 169 GPa [17] are the effec- low applied stress.
tive Poisson ratio and Young’s modulus, respectively, for
the given crystallographic orientation of our silicon spec- (Fig. 2B) at high energy release rates. Following Eq. (1)
imens. The energy release rate Gs varies between 7 and the slope of the G共G兲 curve is given by 共cR 2 yt 兲兾cR ,
89 J兾m2 for the far-field stresses given above. The over- which uniquely defines a terminal crack velocity yt 苷
loads Gs 兾Gc therefore range between 2 and 33 (with Gc 苷 3800 m兾s 苷 0.85cR . The G共y兲 curve (Fig. 2C) corre-
2g共110兲 苷 3.4 J兾m2 [18]). spondingly shows a sharp increase as the terminal velocity
The dependence of the crack velocity on the crack ex- is approached.
tension is shown in Fig. 1A for a fracture stress of s 苷 Two important conclusions can be drawn from the ex-
6.8 MPa. A short transient is followed by propagation perimental results so far. The first is that the continuum
at almost constant velocity. This steady-state velocity is mechanical solution for the propagating crack describes
close to the average velocity yav , defined as the ratio of the situation at low loads with good accuracy. The sec-
the crack path, L 2 lc , and the duration of the fracture ond conclusion is that the dynamics of crack propagation
process. The instantaneous crack velocity is analyzed as at higher energy release rates does not seem to follow the
a function of the instantaneous crack driving force via dynamics of a single crack which consumes energy only
the calculated dependence of the energy release rate on for the creation of new fracture surface. Nevertheless, the
crack length G共l兲 for the given specimen geometry [16] continuum solution can be applied to calculate the energies
in Fig. 1B. The crack velocity closely follows the contin- related to the additional dissipation mechanisms.
uum mechanical solution y共G兲 obtained from inversion of The fracture surface morphologies of the specimens
Eq. (1) for G 苷 2g共110兲 . were analyzed on a large range of length scales using
The average crack velocities from all our fracture ex- different microscopical methods. At the lowest fracture
periments are compared to the continuum mechanical so- stress, corresponding to G 苷 7 J兾m2 and marked with the
lution y共G兲 in Fig. 2A. The measured velocities are again open circle in Fig. 2, the fracture surface is smooth and
in quantitative agreement with the continuum theory at the mirrorlike. It corresponds to the (110) cleavage plane and
lowest energy release rates 共Gs 苷 7 J兾m2 兲. At high energy remains mirrorlike and featureless down to length scales
release rates, however, the crack velocity is lower than the well below 5 nm as observed by atomic force microscopy
continuum mechanical prediction which approaches the (AFM). At higher energy release rates G # 14 J兾m2
Rayleigh wave velocity (cR 苷 4.5 km兾s [15]). and above crack velocities of yav 苷 3000 m兾s 苷 共2兾3兲cR
The dynamic fracture energy G calculated from Eq. (1) (triangles in Fig. 2) the fracture surface still appears mir-
as a function of Gs and yav , is displayed in Figs. 2B rorlike when observed by light optical microscopy. How-
and 2C. The fracture energy increases linearly with Gs ever, AFM reveals that the crack partially deviated from the

789
VOLUME 85, NUMBER 4 PHYSICAL REVIEW LETTERS 24 JULY 2000

FIG. 3. Trench cut into the end of the mirror zone of a speci-
men fractured at Gs 苷 41 J兾m2 using a focussed ion beam
(FIB). The crack propagated from the bottom to the top of
the image on 兵111其 cleavage facets.

relatively smooth planes which extend in the crack prop-

agation direction. Most of the smooth planes are inclined
by an angle of about 35± against the (110) cleavage plane,
which indicates that they result from cleavage on 兵111其
facets. Rough, nonplanar structures always link two cleav-
age facets which appear to have simultaneously propagated
with two independent crack fronts on different height lev-
els (e.g., Fig. 4B). This suggests that the material placed
between the two crack fronts has probably sheared-off far
behind the primary crack front. A very clear example of
such a secondary shear fracture is shown at higher mag-
FIG. 2. (A) Average crack velocity yav as a function of the nification in Fig. 4C, where even subsurface cracks on the
steady state energy release rate Gs . The fracture surface is 兵111其 planes can be resolved. The surface features ob-
smooth and mirrorlike over the entire crack path for the specimen served in the hackle region are described in more detail in
fractured at the lowest Gs (open circle). A faceted fracture Ref. [15].
surface is observed at higher Gs (triangles). At the highest This fracture surface analysis shows that there is no
Gs , the fracture surface is very rough (squares). The full line
corresponds to the continuum mechanical solution obtained from unique fracture mode for one crack velocity. Mirror and
Eq. (1) using G 苷 3 J兾m2 . Equation (1) is used to determine hackle zones coexist on large sections of the fracture sur-
the fracture energy G as a function of (B) the energy release rate faces. There is, however, a unique set of fracture modes
Gs or (C) the average velocity yav . and crack velocities for a given overload. Three modes of
crack propagation can be distinguished for silicon loaded
initial (110) plane and displays hills and valleys extend- along the [110] axis so as to advance the crack in a 关11̄0兴
ing in the crack propagation direction. Similar features are direction. First, brittle crack propagation of one single
observed in the initial transient mirrorlike region on the crack on the (110) cleavage plane is observed. This case
fracture surfaces of the specimens fractured at higher frac-
ture loads which clearly faceted on 兵111其 planes towards
the end of the mirror zone (Fig. 3).
Crack propagation eventually becomes unstable at G .
40 J兾m2 and at velocities close to the terminal velocity
(squares in Fig. 2). The instability manifests itself in the
onset of pronounced surface features. In this “hackle” re-
gion the typical structure size is of the order of 1 200 mm.
The transition from the mirror to the hackle zone is abrupt
and the different zones can clearly be distinguished. The FIG. 4. (A) Scanning electron microscopy (SEM) micrograph
hackle region appears to start from the specimen sides and showing the coexistence of hackle regions at the sides and a
expands gradually into the interior. The fracture surface mirror region in the center of the fracture surface. Crack propa-
shows an extended region where parts of the crack front gation was from left to right. (B) SEM and (C) FIB micrographs
show typical features of the hackle region at higher magnifica-
produced mirror and other parts of the crack front produced tion. The trench in (C) has been cut into the fracture surface
hackle zones (Fig. 4A). The hackle region itself is char- by a FIB. Subsurface 兵111其 cleavage cracks are clearly visible.
acterized by an alternate change of rough structures and The crack propagated from top to bottom in (B) and (C).

790
VOLUME 85, NUMBER 4 PHYSICAL REVIEW LETTERS 24 JULY 2000

is fully described by the continuum mechanical solution, material with only one type of cleavage system like GaAs,
Eq. (1), under the assumption of a constant fracture energy where the 兵110其 planes are the only cleavage planes, one
which corresponds to the surface energy of the two (110) would expect to reach very high terminal velocities before
surfaces. Second, at intermediate overloads the crack the crack can make use of alternative paths.
propagates on two sets of 兵111其 planes, which results In general, the dynamic fracture process can be sepa-
in a faceted fracture surface and leads to the increased rated into two different regimes: a first regime at low
fracture energy (triangles in Fig. 2). The size of the overloads, where the behavior of the crack is material-
facets increases with increasing G. Because the faceting independent in the sense that it is fully described by lin-
does not change the total surface energy significantly ear elastic continuum mechanics once the surface energy
g共111兲
[ cos共35 ± 艐 g共110兲 [19]], it is concluded that the crack must and the elastic properties of the material are known, and
兲
spend the additional energy by emitting phonons. Whether a second regime, where material-specific dissipation pro-
this radiation is connected with the formation of the rims cesses dominate. Future investigation must focus on the
between the two 兵111其 fracture surfaces or whether it is mechanisms which govern the latter to further advance our
connected with the mixed mode I-III loading on the 兵111其 understanding of catastrophic failure in materials.
facets is unclear. The onset of the faceting can not be Financial support from Deutsche Forschungsgemein-
identified on the fracture surfaces, which indicates that the schaft through grant Gu 367/8 is gratefully acknowledged.
faceting may grow continuously out of small disturbances
of a planar crack.
The third mode is only observed at the highest over-
loads, where the specimens exhibit rough hackle zones.
[1] R. Thomson, in Solid State Physics, edited by H. Ehren-
The abrupt onset of the hackle zone is clearly connected
reich and F. Seitz (Academic Press, New York, 1986),
with the onset of crack path instabilities. The path instabil- Vol. 39, p. 2.
ity is apparently connected with the ability of the cracks on [2] A. A. Griffith, Philos. Trans. R. Soc. London A 221, 163
the 兵111其 facets to overgrow each other. This locally results (1920).
in the simultaneous propagation of two crack fronts and in [3] L. B. Freund, Dynamic Fracture Mechanics (Cambridge
secondary shear fracture of the material placed between. University Press, New York, 1990).
This fracture mechanism exhibits several ways of energy [4] J. Fineberg et al., Phys. Rev. Lett. 67, 457 (1991).
dissipation. Not only is the total surface area increased [5] S. P. Gross et al., Phys. Rev. Lett. 71, 3162 (1993).
significantly, but unsteady propagation of the crack front [6] J. W. Johnson and D. G. Holloway, Philos. Mag. 14, 731
must also be accompanied by significant “noise” and radi- (1966).
ation. Hence, the observation that the crack velocity does [7] E. Sharon and J. Fineberg, Nature (London) 397, 333
(1999).
not increase beyond 0.85cR even at energy release rates as
[8] E. H. Yoffe, Philos. Mag. 42, 739 (1951).
high as 30Gc clearly displays the ability of this fracture [9] J. J. Gilman, J. Appl. Phys. 27, 1262 (1956).
mechanism to dissipate enormous amounts of energy. [10] D. Hull and P. Beardmore, Int. J. Fract. Mech. 2, 468
Coming back to the initial question, how does a crack (1966).
respond to large driving forces? The results presented here [11] J. E. Field, Contemp. Phys. 12, 1 (1971).
show that the crack makes use of all possible ways to spend [12] J. E. Hauch et al., Phys. Rev. Lett. 82, 3823 (1999).
energy. Initially the easiest way to spend energy is to run [13] T. Cramer, A. Wanner, and P. Gumbsch, Phys. Status Solidi
faster. Faceting then apparently allows to spend some more (a) 164, R5 (1997).
energy and finally a very high driving force leads to a path [14] G. Michot, Cryst. Prop. Prep. 17 – 18, 55 (1988).
instability which splits the initially continuous crack front [15] T. Cramer, A. Wanner, and P. Gumbsch, Z. Metallkd. 90,
and results in the propagation of multiple cracks. This 675 (1999).
[16] T. Cramer, P. Gumbsch, and A. Wanner, in Werkstoffwoche
path instability bears some similarity with the bifurcation
’98, Bd. X, edited by W. Muster, J. Zieps, and R. Link
instability found in PMMA [7]. The terminal velocity (Wiley-VCH, Weinheim, 1998), p. 23.
at which the instability occurs is of course determined [17] W. A. Brantley, J. Appl. Phys. 44, 534 (1973).
by the material and its crystallographic orientation. One [18] R. Pérez and P. Gumbsch, Phys. Rev. Letters 84, 5347
could argue that a low terminal velocity is expected if (2000).
alternative fracture paths are easily accessible as they are [19] S. B. Bhaduri and F. F. Y. Wang, J. Mater. Sci. 21, 2489
in isotropic amorphous glass or PMMA. Inversely, for a (1986).

791