Acta Materialia 51 (2003) 5801–5821

www.actamat-journals.com

Designing hybrid materials夽

M.F. Ashby a,∗, Y.J.M. Bréchet b
a
Engineering Department, University of Cambridge, Trumpington Street, CB2 1PZ Cambridge, UK
b
L.T.P.C.M., Domaine Universitaire de Grenoble, BP75, 38402 Saint Martin d’Heres Cedex, France

Accepted 31 August 2003

Abstract

The properties of engineering materials can be mapped, displaying the ranges of mechanical, thermal, electrical and
optical behavior they offer. These maps reveal that there are holes: some areas of property-space are occupied and
others are empty. The holes can sometimes be filled and the occupied areas extended by making hybrids of two or
more materials or of material and space. Particulate and fibrous composites are examples of one type of hybrid, but
there are many others: sandwich structures, foams, lattice structures and more. Here we explore ways of designing
hybrid materials, emphasizing the choice of components, their shape and their scale. The new variables expand the
design space, allowing the creation of new “materials” with specific property profiles.
 2003 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Keywords: Designing materials; Hybrid materials; Composites; Sandwich structures; Foams

1. Introduction: hybrid materials
1.1. Extending material-property space
Fig. 1 is an example of a material-property
chart. It shows the thermal conductivities of some
2300 different materials, plotted against their
Young’s moduli. It is one of many, each a slice
through material-property space; the assembly of
all the slices can be thought of as a map of this
space [1,2]. All the charts have one thing in com-
mon: parts of them are populated with materials
∗
Corresponding author. Tel.: +44-01223-332-635; fax: +44-
01223-332-662.
夽
The Golden Jubilee Issue—Selected topics in Materials
Science and Engineering: Past, Present and Future, edited by
S. Suresh.
and parts are not. Some parts are inaccessible for
fundamental reasons that relate to the size of atoms
and the nature of the forces that bind their atoms
together. But other parts are empty even though,
in principle, they are accessible. If they were
accessed, the new materials that are there could
allow novel design possibilities.
One approach to this—the traditional one—is
that of developing new metal alloys, new polymer
chemistries and new compositions of glass and cer-
amics so as to extend the populated areas of the
property charts, but this can be an expensive and
uncertain process. An alternative is to combine two
or more existing materials so as to allow a super-
position of their properties—in short, to create a
hybrid (Fig. 2). The spectacular success of carbon
and glass-fiber reinforced composites at one
extreme, and of foamed materials at another

1359-6454/$30.00  2003 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/S1359-6454(03)00441-5
5802 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821

Fig. 1. A material-property chart of thermal conductivity and Young’s modulus for 2300 materials. Each small circle is a plot of
these properties for a real material. The large ellipses enclose, approximately, the circles for a given family of materials. A large
area of the chart is empty: there are no materials with high conductivity and low modulus. The challenge is to create hybrids that
fill the hole. (This and the other charts were created using the CES 4 software system, Ref. [42].)

(hybrids of material and space) in filling previously

empty areas of the property charts is encourage-
ment enough to explore ways in which such
hybrids might be designed. What is the best way
to go about doing so?

1.2. What might we hope to achieve?

Fig. 3 shows schematically the fields occupied

Fig. 2. Hybrid materials combine the properties of two (or
more) monolithic materials, or of one material and space. They
include fibrous and particulate composites, foams and lattices,
sandwiches and almost all natural materials. One might imagine
two further dimension: those of shape and scale.
by two families of materials, plotted on a chart
with properties P1 and P2 as axes. Within each field
a single member of that family is identified
(materials M1 and M2). What might be achieved
by making a hybrid of the two? The figure shows
four scenarios, each typical of a different class of
hybrid. We consider the case when large values
of P1 and P2 are desirable, low values not. Then,
depending on the shapes of the materials and the
way they are combined, we may find any one of
the following.
앫 “The best of both” scenario (Point A). The ideal,
often, is the creation of a hybrid with the best
properties of both components. There are
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
Fig. 3. The possibilities of hybridization. The properties of the
hybrid reflect those of its component materials, combined in
one of several possible ways.
examples, most commonly when a bulk pro-
perty of one material is combined with the sur-
face properties of another. Zinc coated steel has
the strength and toughness of steel with the cor-
rosion resistance of zinc. Glazed pottery
exploits the formability and low cost of clay
with the impermeability and durability of glass.
앫 “The rule of mixtures” scenario (Point B). When
bulk properties are combined in a hybrid, as in
structural composites, the best that can be
obtained is often the arithmetic average of the
properties of the components, weighted by their
volume fractions. Thus unidirectional fiber com-
posites have an axial modulus (the one parallel
to the fibers) that lies close to the rule of mix-
tures.
앫 “The weaker link dominates” scenario (Point
C). Sometimes we have to live with a lesser
compromise, typified by the stiffness of particu-
late composites, in which the hybrid properties
fall below those of a rule of mixtures, lying
closer to the harmonic than the arithmetic mean
of the properties. Although the gains are less
spectacular, they can still be useful.
앫 The “worst of both” scenario (point D)—not
something we want.
These set certain fixed points, but the list is not
5803
exhaustive. Other combinations are possible, some
relying on the physics of percolation, others on
atomistic effects. These will emerge below.
1.3. When is a hybrid a “material”?
There is a certain duality about the way in which
hybrids are viewed and discussed. Some, like filled
polymers, composites or wood are treated as
materials in their own right, each characterized by
its own set of material properties. Others—like gal-
vanized steel—are seen as one material (steel) to
which a coating of a second (zinc) has been
applied, even though this could be regarded as a
new material with the strength of steel but the sur-
face properties of zinc (“stinc”, perhaps?). Sand-
wich panels illustrate the duality, sometimes
viewed as two sheets of face-material separated by
a core material, and sometimes—to allow compari-
son with bulk materials—as a “material” with their
own density, flexural stiffness and strength. To call
any one of these a “material” and characterize it
as such is a useful shorthand, allowing designers
to use existing methods when designing with them.
But if we are to design the hybrid itself, we must
deconstruct it, and think of it as a combination of
materials (or of material and space) in a defined
geometry.
2. The method: “A + B + shape + scale”
First, a working definition: “a hybrid material is
a combination of two or more materials in a prede-
termined geometry and scale, optimally serving a
specific engineering purpose” [3], which we para-
phrase as “A + B + shape + scale”. Here we allow
for the widest possible choice of A and B, includ-
ing the possibility that one of them is a gas or sim-
ply space. These new variables expand the design
space, allowing an optimization of properties that
is not possible if choice is limited to single, mono-
lithic materials.
The basic idea, illustrated in Fig. 4, is this.
Monolithic materials offer a certain portfolio of
properties on which much engineering design is
based. But if the design requirements are excep-
tionally demanding, no single material may be
5804 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
Fig. 4. The steps in designing a hybrid to meet given design requirements.
found that can meet them: the requirements lie in
a hole in property space. Then the way forward is
to identify and separate the conflicting require-
ments, seeking optimal material solutions for each,
and then combine them in ways that retain the
desirable attributes of both. The best choice is the
one that ranks most highly when measured by the
performance metrics that motivate the design: min-
imizing mass or cost, or maximizing some aspect
of performance (the criteria of excellence). The
alternative combinations are examined and
assessed, using the criteria of excellence to rank
them. The output is a specification of a hybrid in
terms of its component materials and geometry.
Consider as an example the design of a hybrid
material for long-span power cables. The objec-
tives are to minimize the electrical resistance, but
at the same time to maximize the strength since
this allows a greater span. In multi-objective opti-
mization, of which this problem is an example, it
is conventional to express each objective such that
a minimum is sought; we thus seek materials with
the lowest values of resistivity, R, and reciprocal
of tensile strength, 1 / sts. Fig. 5 shows the result:
materials that best meet the design requirements lie
near the bottom left. Those with the lowest resist-
ance—copper, aluminum, and some of their
alloys—are not very strong, and the materials that
are strongest—drawn carbon and low-alloy steel—
do not conduct very well. Now consider a cable
made by interweaving strands of copper and steel
such that each occupies half the cross-section.
Assuming the steel carries no current and the cop-
per no load (the most pessimistic scenario) the per-
formance of the cable will lie at the point shown
on the figure—it has twice the resistivity of the
copper and half the strength of the steel. It occupies
a part of property space that was previously empty,
offering performance that was not previously poss-
ible.
But while some conflicting requirements can be
met in this way, others need a more inventive
approach.1 So the question arises: are there general
ways in which material hybridization can be
explored systematically? It is unrealistic to suppose
that one method and one tool can solve all such
problems. Instead we examine examples of hybrid
design and attempt to extract principles that could
help tackle other, as yet unformulated problems of
this class.
3. A + B: selecting components for composites
Aircraft engineers, automobile makers, and
designers of sports equipment all have one thing in
common: they want materials that are stiff, strong,
tough and light. The single-material choices that
best achieve this are the light alloys: alloys based
on magnesium, aluminum and titanium. Much
research aims at improving their properties. But
they are not all that light—polymers have much
lower densities. Nor are they all that stiff—cer-
amics are much stiffer and, especially in the form
of small particles or thin fibers, much stronger.
These facts are exploited in the subset of hybrids
that we usually refer to as particulate and
fibrous composites.
Any two materials can, in principle, be com-
bined to make a composite, and they can be mixed
in many geometries (Fig. 6). In this section, we
1
An interesting example is that of flexible ferromagnets.
Monolithic ferromagnetic materials are stiff, metallic or cer-
amic, solids. Elastomeric ferromagnetic hybrids offer several
properties that these monolithic solids do not. The hybrids are
made by mixing up to 30% of sub-micron iron particles into an
elastomer resin before polymerising it. The result is a compliant
ferromagnetic material that has the property that it is mag-
netostrictive, and that its stiffness increases when placed in the
magnetic field because the magnetic dipoles that are induced in
the particles attract one another. The material has a fast (1 ms)
response time, making it suitable for vibration damping.
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821 5805

Fig. 5. Designing a hybrid—here, one with high strength and high electrical conductivity. The figure shows the resistivity and
reciprocal of tensile strength for 1700 metals and alloys. We seek materials with the lowest values of both. The construction is for
a hybrid of hard-drawn OFHC copper and drawn low alloy steel, but the figure itself allows many hybrids to be investigated [42].

restrict the discussion to fully dense, strongly

Fig. 6. Schematic of hybrids of the composite type: unidirec-
tional fibrous, laminated fiber, chopped fiber and particulate
composites. Bounds and limits, described in the text, bracket
the properties of all of these.
bonded, composites such that there is no tendency
for the components to separate at their interfaces
when the composite is loaded, and to those in
which the scale of the reinforcement is large com-
pared to that of the atom or molecule size and the
dislocation spacing, allowing the use of con-
tinuum methods.
On a macroscopic scale—one which is large
compared to that of the components—a composite
behaves like a homogeneous solid with its own set
of thermo-mechanical properties. Calculating these
precisely can be done, but it is difficult. It is much
easier to bracket them by bounds or limits: upper
and lower values between which the properties lie
[4–7]. The term “bound” will be used to describe
a rigorous boundary, one which the value of the
property cannot—subject to certain assumptions—
exceed or fall below. It is not always possible to
derive bounds; then the best that can be done is to
derive “limits” outside which it is unlikely that the
value of the property will lie. The important point
5806 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821

is that the bounds or limits bracket the properties Table 1

of all arrangements of matrix and reinforcement Criteria of excellence for minimum weight design
shown in Fig. 6; by using them we avoid the need
Mode of loading and Stiffness at minimum weight
to model individual geometries. geometry

3.1. Density Tensile loading of ties E/r

Bending of beams E1 / 2 / r
When a volume fraction f of a reinforcement r Bending of plates E1 / 3 / r
(density rr) is mixed with a volume fraction (1⫺
f) of a matrix m (density rm) to form a composite
with no residual porosity, the composite density is stiffness per unit mass, measured by the indices
given exactly by a rule of mixtures (an arithmetic listed in the table (for derivations see Ref. [2]). If
mean, weighted by volume fraction) a possible hybrid has a value of any one of these
that exceed those of the light alloys, it achieves
r ⫽ frr ⫹ (1⫺f)rm. (1) our goal.
The geometry or shape of the reinforcement does Consider, as an illustration of the method, the
not matter except in determining the maximum design of a composite for a light, stiff beam of
packing-fraction of reinforcement and thus the fixed section-shape, to be loaded in bending. The
upper limit for f. efficiency is measured by the index E1 / 2 / r shown
in Table 1. Imagine, as an example, that the beam
3.2. Modulus is at present made of an aluminum alloy. Beryllium
is both lighter and stiffer than aluminum; ceramics
The modulus of a composite is bracketed by the are stiffer, but not all are lighter. What can these
well-known Voigt and Reuss bounds. The upper hybrids offer?
bound, Eu, is obtained by postulating that on load- Fig. 7 is a small part of the E–r property chart.
ing the two components suffer the same strain; the
stress is then the volume-average of the local
stresses and the composite modulus follows a rule
of mixtures:
Eu ⫽ f Er ⫹ (1⫺f)Em. (2)
Here Er is the Young’s modulus of the reinforce-
ment and Em that of the matrix. The lower bound,
El, is found by postulating instead that the two
components carry the same stress; the strain is the
volume-average of the local strains and the com-
posite modulus is
EmEr
E1 ⫽ (3)
f Em ⫹ (1⫺f)Er
More precise bounds are possible [8,9], but the
simple ones are adequate to illustrate the method.

3.3. Hybrid design for stiffness at minimum

weight
We need a criterion of excellence to assess the
merit of any given hybrid. Here our criterion is
Fig. 7. Part of the E–r property chart, showing aluminium
alloys, beryllium and alumina (Al203). Bounds for the moduli
of hybrids made by mixing them are shown. The diagonal con-
tours plot the criterion of excellence, E1 / 2 /r.
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
Three groups of materials are shown: aluminum
and its alloys, alumina (Al2O3) and beryllium (Be).
Composites made by mixing them have densities
given exactly by Eq. (1) and moduli that are brack-
eted by the bounds of Eqs. (2) and (3). Both of
these moduli depend on volume fraction of
reinforcement, and through this, on density. Upper
and lower bounds for the modulus–density
relationship can thus be plotted onto the E–r chart
using volume fraction f as a parameter, as shown
in Fig. 7. Any composite made by combining
aluminum with alumina will have a modulus con-
tained in the envelope for Al–Al2O3; the same for
Al–Be. Fibrous reinforcement gives a longitudinal
modulus (parallel to the fibers) near the upper
bound; particulate reinforcement or transversely
loaded fibers give moduli near the lower one.
Superimposed on Fig. 7 is a grid showing the
criterion of excellence E1 / 2 / r. The bound-envelope
for Al–beryllium composites extends almost nor-
mal to the grid, while that for Al–Al2O3 lies at a
shallow angle to it. Beryllium fibers improve per-
formance (as measured by E1 / 2 / r) roughly four
times as much as alumina fibers do, for the same
volume fraction. The difference for particulate
reinforcement is even more dramatic. The lower
bound for Al–Be lies normal to the contours: 30%
of particulate beryllium increases E1 / 2 / r by a fac-
tor of 1.5. The lower bound for Al–Al2O3 is,
initially, parallel to the E1 / 2 / r grid: 30% of par-
ticulate Al2O3 gives almost no gain. The underly-
ing reason is clear: both beryllium and Al2O3
increases the modulus, but only beryllium
decreases the density; the criterion of excellence is
more sensitive to density than to modulus.
In Fig. 8 we return to the big picture. It shows
the moduli and densities of metals and polymers,
and, encircled by a broken ellipse, those of high
performance carbon, aramid, PE and glass fibers.
The construction illustrated in Fig. 7 leads to famil-
ies of polymer–matrix composites that lie in the
shaded ellipse with that name, and to families of
metal–matrix composites that lie in the ellipse
above it. Both ellipses occupy areas of property
space that were previously unoccupied by bulk
materials, and it is an important one, enabling the
design of new lightweight mechanical structures.
Similar methods can be used to select materials
5807
optimum strength, and for tailored values of ther-
mal conductivity, expansion coefficient and spe-
cific heat [7]. The properties of specific composites
can, of course, be computed in conventional ways.
The advantage of this graphical approach is the
breadth and freedom of conceptual thinking that it
allows and the ease of comparison of possible new
hybrids with the population of existing materials.
3.4. Percolation: properties that switch on and
off
Fig. 9, a chart of electrical resistivity against
elastic stiffness (here measured by Young’s
modulus), has an enormous hole. Materials that
conduct well are stiff; those that are flexible are
insulators. Consider designing materials to fill the
hole; to be more specific, consider designing one
that has low modulus, can be molded like a poly-
mer, and is a good electrical conductor. Such
materials find application in anti-static clothing and
mats, as pressure sensing elements, even as solder-
less connections.
Metals, carbon and some carbides and intermet-
allics are good conductors, but they are stiff and
cannot be molded. Thermoplastic and thermoset-
ting elastomers can be molded but do not conduct.
How are they to be combined? Metal coating of
polymers is workable if the product is to be used in
a protected environment, but the coating is easily
damaged. If a robust, flexible, product is needed,
bulk rather than surface conduction is essential.
This can be achieved by mixing conducting par-
ticles into the polymer.
To understand how to optimize this we need the
concept of percolation. Percolation problems are
easy to define, but not easy to solve. Research
since 1960 has provided approximate solutions to
most of the percolation problems associated with
the design of hybrids (see Ref. [10] for a review).
Think of mixing conducting and insulating spheres
of the same size to give a large array. If there are
too few conducting spheres for them to touch, the
array is obviously an insulator. If each conducting
sphere contacts just one other, there is still no con-
necting path. If, on average, each touched two,
there is still no path. Adding more spheres gives
larger clusters, but they can be large yet still dis-
5808 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821

Fig. 8. Young’s modulus and density for 1850 polymers, metals and fibers (broken ellipse). Combining these to create polymer and
metal matrix composites fills a previously empty hole in material-property space [42].

Fig. 9. When conducting, particles or fibers are mixed into an insulating elastomer, a hole in material-property space is filled.
Carbon-filled butyl rubbers lie in this part of the space [42].
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
crete. The array first becomes a conductor when a
single trail of contacts links one surface to the
other, that is, when the fraction p of conducting
spheres reaches the percolation threshold, pc. For
simple cubic packing pc = 0.248, for close packing
pc = 0.180. For a random array it is somewhere in
between—approximately 0.2.2
Make the spheres smaller and the transition is
smeared out. The percolation threshold is still 0.2,
but the first connecting path is now thin and
extremely devious—it is the only one, out of the
vast number of almost complete paths, that actually
connects. Increase the volume fraction and the
number of conducting paths increases initially as
(p⫺pc)2, then linearly, reverting to a rule of mix-
tures [13]. If the particles are very small, as much
as 40% may be needed to give good conduction.
But a loading of 40% seriously degrades the mold-
ability and compliance of the polymer.
Shape gives a way out. If the spheres are
replaced by fibers, they touch more easily and the
percolation threshold falls. If their aspect ratio is
f = L / d (where L is the fiber length, d the
diameter) then—to an adequate approximation—
empirically, the percolation threshold falls from fc
to roughly fc / f1 / 2 [14–18]. Fig. 9 shows the area
of the property chart where these hybrids lie. With
sufficient aspect ratio the percolation threshold
falls to a few percentage points.
The concept of percolation is a necessary tool
in designing hybrids. Electrical conductivity works
that way; so too does the passage of liquids
through foams or porous media—no connected
paths, and no fluid flows; just one (out of a million
possibilities) and there is a leak. Add a few more
connections and there is a flood. Percolation ideas
are particularly important in understanding the
transport properties of hybrids: properties that
determine the flow of electricity or heat, of fluid,
or of flow by diffusion, specially when the differ-
ences in properties of the components are extreme.
Most polymers differ from metals in their electrical
conductivity by a factor of about 1020. The dif-
2
These results are for infinite, or at least very large, arrays.
Experiments [11,12] generally give values in the range 0.19–
0.22, with some variability because of the finite size of the
samples.
5809
fusion of water through solids differs from the flow
rate of water through channels by a similar factor.
It is then that single connections really matter.
Percolation influences mechanical properties
too, particularly when mechanical connection is
important, as in arrays of loose powders or fibers.
If there are no bonds between the particles or fib-
ers, the array has no tensile stiffness or strength.
If each particle is bonded to another, or to several
forming discrete clusters, there is still no stiffness
or strength. These only appear when there are con-
nected paths running completely through the array.
The plasticity of 2-phase hybrids, too, can be
viewed as a percolation problem. Plasticity may
start in one phase at a low stress, allowing patches
of slip to form, but full plasticity requires that the
slip patches link to give connected paths through
the entire cross-section of the sample. Mechanical
and electrical percolation can be combined,
exploiting scale. A latex reinforced with a small
volume fraction of cellulose fibers coated with
polypyrrole to make them conducting, gives a
material with a shear modulus two orders of mag-
nitude higher than the rule of mixture predicts
combined with good conduction because of the
high aspect ratio of the fibers [19]. Here we are
escaping the continuum bounds—a topic we return
to later.
3.5. Creating anisotropy
The elastic and plastic properties of bulk mono-
lithic solids are frequently anisotropic, but weakly
so—the properties do not depend strongly on direc-
tion. Hybridization gives a way of creating and
controlling anisotropy, and it can be large. We
have already seen an example in Fig. 7, which
shows the upper and lower bounds for the moduli
of composites. The longitudinal properties of
unidirectional long-fiber composites lie near the
upper bound, the transverse properties near the
lower one. The vertical width of the band in
between them measures the anisotropy.
Consider a second example, that of creating
hybrids with anisotropic thermal conductivity. A
saucepan made from a single material when heated
on an open flame, develops hot spots that can
locally burn its contents. That is because the sauce-
5810 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
pan is thin, and heat is transmitted through the
thickness more quickly than it can be spread trans-
versely to bring the entire pan surface to a uniform
temperature. The metals of which saucepans are
usually made—cast iron, or aluminum or copper—
have a isotropic thermal conductivities whereas
what we clearly want is a thermal conductivity that
is higher in the transverse direction than in the
through-thickness direction. A bi-layer (or multi-
layer) hybrid can achieve this.
Heat transmitted transversely in a bi-layer sheet
has two parallel paths; the total heat transmitted is
a sum of that in each of the paths. If it is made of
a layer of material 1 with thickness t1 and conduc-
tivity l1, bonded to a layer of material 2 with thick-
ness t2 and conductivity l2, the conductivity paral-
lel to the layers is
lII ⫽ fl1 ⫹ (1⫺f)l2 (4)
(a the rule of mixtures), where f = t1 / (t1 + t2). Per-
pendicular to the layers the conductivity is
1 f (1⫺f)
⫽ ⫹ (5)
lⲚ l1 l2
(the harmonic mean). Fig. 10 shows lⲚ and l
plot-
ted against f for a bi-layer of copper (l=390
W/m K) and cast iron (l = 30 W / m K). For single
materials the two are equal; layering them gives
Fig. 10. Creating anisotropy. The thermal conductivities of
copper and cast iron are isotropic. Anisotropy is created by
combining them as a bi-layer.
trajectories for lⲚ and l
that separate. The
maximum separation occurs broadly where each
occupies about half the thickness, where the ratio
of the conductivities (the anisotropy ratio) is 3.8.
Mechanical anisotropy is most easily created
and managed through shape. This is the topic of
the next section.
4. Shape: structures, sandwiches and
segmented assemblies
The shape and configuration of components A
and B of a hybrid play a key role in determining
its properties. Shape can be used to enhance or
diminish stiffness and strength, to impart damage
tolerance, and—as we have already seen—to
manipulate the percolation limit.
4.1. Shape efficiency and shape factors
Beams with hollow-box or I-sections are stiffer
and stronger in bending than solid sections of the
same cross-sectional area; so, too, are panels with
ribs or waffle stiffeners, or those with an expanded
core to create a sandwich (Fig. 11). These are
Fig. 11. Making high-efficiency structures. Shape gives the
sections a greater flexural stiffness and strength per unit mass
than the solid section from which they are made.
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
examples of the use of shape to increase structural
efficiency. To characterize this we need a metric—
a way of measuring the structural efficiency of a
section shape, independent of the material of which
it is made. An obvious one is that given by the
ratio j of the stiffness or strength of the shaped
section to that of a ‘neutral’ reference shape. For
a beam we take the reference shape to be that of
a solid square section with the same cross-sectional
area, and thus the same mass per unit length, as
the shaped section. For a panel, we take it to be a
solid, plain panel with the same mass per unit area
as the shaped section, as shown in the figure. We
call j the shape factor [2,20] and define that for
stiffness je as
Flexural stiffness of shaped section
je ⫽
Flexural stiffness of reference section
and that for strength jf as
Flexural strength of shaped section
jf ⫽
Flexural strength of reference section
Shape can be used to reduce flexural stiffness
and strength as well as increase them. Springs, sus-
pensions, flexible cables and other structures that
must flex yet have high tensile strength, use shape
to give a low bending stiffness. Low shape
efficiency is achieved by forming the material into
strands or leaves, as suggested Fig. 12. Values of
j for the stiffness of structural sections can be as
Fig. 12. Making low-efficiency structures. Shape gives the
sections a lower flexural stiffness and strength per unit mass
than the solid section from which they are made.
5811
high as 50; for multi-strand or multi-leaf structures
as low as 0.01.
Note the origins of efficiency. The flanges of the
I-section or the faces of the sandwich panels lie far
from the neutral axis; they stretch when the section
is loaded in bending. Subdivision, as in Fig. 11,
lowers efficiency because the slender strands or
leaves bend easily, but do not stretch when the sec-
tion is bent: an n-strand cable is less stiff by a fac-
tor of 3/πn than the solid reference section; an n-
leaf panel by a factor 1/n2. There is an underlying
principle here: stretch dominated structures have
high structural efficiency; bending dominated
structures have low.
4.2. Shape on a micro-scale
(6)
The sections of Fig. 11 achieve efficiency
through their macroscopic shape. Structural
efficiency can be manipulated in another way:
through shape on a small scale; microscopic or
(7)
“micro-structural” shape. Wood is an example. The
solid component of wood (a composite of cellu-
lose, lignin and other polymers) is shaped into pris-
matic cells, each cell like the hollow tube of Fig.
11. The effect is to disperse the solid component
further from the axis of bending or twisting of the
branch or trunk of the tree, increasing its flexural
stiffness and strength. This is not the only possi-
bility; low efficiency structures give materials with
low stiffness and strength, desirable in cushioning
and packaging.
4.3. Ultra-light, low stiffness hybrids
The point has been made that stretching is a stiff
mode of loading, bending is a compliant one. A
material that responds, at the micro-structural
level, by bending no matter how it is loaded
remotely is much less stiff than one that responds
by stretching. Material made by foaming have
structures that respond in this way.
Fig. 13 shows an idealized cell of a low-density
foam. It consists of solid cell walls or edges sur-
rounding a void space, each cell with an overall
space-filling shape. Cellular solids are charac-
terized by their relative density, which for the
structure shown here (with t ⬍ ⬍ᐉ) is
5812 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821

Fig. 13. A cell in a low density foam. When the foam is loaded, the cell edges bend, giving a low-modulus structure.

r∗
rs
⫽
t
ᐉ 冉冊 2
(8)
yield when the force exerted on them exceeds their
fully plastic moment

where r∗ is the density of the foam, rs is the den- sst3

sity of the solid of which it is made, ᐉ is the cell Mf ⫽ (11)
4
size, and t is the thickness of the cell edges. A
remote compressive stress s exerts a force F ⬀
where ssis the yield strength of the solid of which
sᐉ2 on the cell edges, causing them to bend as
the foam is made. This moment is related to the
shown in the figure, and leading to a bending
remote stress by M ⬀ FL ⬀ sL3. Assembling these
deflection d. For the open-celled structure shown
results gives the failure strength s∗
in the figure, the bending deflection is given by

d⬀
FL3
EsI
(9)
s∗ r
⬀
s s rs冉冊 3/2
(12)

where Es is the modulus of the solid of which the
foam is made and I = t4 / 12 is the second moment
of area of the cell edge of square cross-section, t
× t. The compressive strain suffered by the cell as
a whole is then e = 2d / ᐉ. Assembling these results
gives the modulus E∗ = s / e of the foam as
E∗
Es
⬀
r
rs 冉冊 2
(10)
(bending ⫺ dominated behavior)
Since E = Es when r = rs, we expect the constant
of proportionality to be close to unity—a specu-
lation confirmed both by experiment and by
numerical simulation.
A similar approach can be used to model non-
linear properties such as strength. The cell walls
(bending ⫺ dominated behavior)
This behavior is not confined to open-cell foams
with the structure shown in Fig. 14. Most closed-
cell foams also follow these scaling laws. At first
sight an unexpected result because the cell faces
must carry membrane stresses when the foam is
loaded, and these should lead to a linear depen-
dence of both stiffness and strength on relative
density. The explanation lies in the fact that the
cell faces are very thin; they buckle or rupture at
stresses so low that their contribution to stiffness
and strength is small, leaving the cell edges to
carry most of the load (for further details, see
Ref. [21]).
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
Fig. 14. A micro-truss structure and its unit cell.
4.4. Ultra-light, high stiffness hybrids
If conventional foams have low stiffness
because other configuration of the cell edges
allows them to bend, might it not be possible to
devise other configurations in which the cell edges
were made to stretch instead? This thinking leads
to the idea of micro-truss structures [22,23]. To
understand these we need the Maxwell stability cri-
terion.
The condition that a pin-jointed frame of b struts
and j frictionless joints to be both statically and
kinematically determined i.e. just rigid [24,25], in
2-dimensions, is:
M ⫽ b⫺2j ⫹ 3 ⫽ 0 (13)
and in 3-dimensions is
M ⫽ b⫺3j ⫹ 6 ⫽ 0 (14)
If M ⬍ 0, the frame is a mechanism. It has no
stiffness or strength, but will collapse if loaded. If
its joints are locked (instead of pin-jointed) the bars
of the frame bend when the structure is loaded. If,
instead, M ⱖ 0 the frame ceases to be a mech-
anism; its members carry tension or compression
when the frame is loaded, and it becomes a stretch-
dominated structure.
These criteria give a basis for the design of
efficient micro-truss structures. For the cellular
structure of Fig. 13 M ⬍ 0, and it is bending domi-
nated. For the structure shown in Fig. 14, however,
M ⬎ 0 and it behaves as an almost isotropic,
5813
stretch-dominated structure. On average one third
of its bars carry tension when the structure is
loaded in simple tension, regardless of its direc-
tion. Thus
冉冊
E∗ 1 r
⬇ (15)
Es 3 rs
for isotropic stretch-dominated behavior and
冉冊
s∗ 1 r
⬇ (16)
ss 3 rs
for isotropic stretch-dominated behavior.3
Prismatic structures do even better, provided
they are loaded parallel to the prism axis. Fig. 15
shows four such structures which are common in
nature. It is helpful to think, as before, of the
expansion of a solid bar, shown in the center, to
give the structures, with no change of mass—here
the solid black represents solid material, the dotted
areas represent low density foam and the open
areas represent space. The expansion has moved
material away from the axis of bending, increasing
the second moment of area about any axis con-
tained in the plane of the cross-section, and
increasing efficiency in the sense of Eqs. (6) and
(7). For these structures the axial modulus and
strength (measured parallel to the prism axis) fol-
3
This assumes that failure occurs by the axial yielding of a
bar. If the bars are slender or have low modulus they may fail
instead by buckling, giving a lower strength.
5814 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
Fig. 15. Four extensive micro-structured materials which are
mechanically efficient: (a) prismatic cells, (b) fibers embedded
in a foamed matrix, (c) concentric cylindrical shells with foam
between, and (d) parallel plates separated by foamed spacers.
lows Eqs. (15) and (16), but with a constant of
proportionality not of 1/3 but of unity:
E∗ r
⬇
Es rs 冉冊 (17)
for prismatic stretch-dominated behavior and
冉冊
s∗ r
⬇ (18)
ss rs
for prismatic stretch-dominated behavior.3
Loaded transversely, however, they are bending
dominated and follow power laws like those of
Eqs. (10) and (12); they are thus exceedingly
anisotropic.
These results are summarized in Fig. 16, in
which the modulus E∗ is plotted against the den-
sity r. Stretch dominated, prismatic microstruc-
tures like wood give moduli that scale as r / rs
(slope 1); bending dominated, cellular, microstruc-
tures like that of polystyrene foam give moduli that
scale as (r / rs)2 (slope 2). Given that the density
can be varied through a wide range, this allows
great scope for material design. Note how the use
of microscopic shape has expanded the occupied
area of E–r space.
4.5. Ultimate efficiency: the sandwich
A sandwich panel epitomizes the concept of a
hybrid. It combines two materials in a specified
geometry and scale—one forming the faces, the
other the core—to give a structure of high stiffness
and strength at low weight (Figs. 11 and 17). The
separation of the faces by the core increases the
moment of inertia I and the section modulus Z of
the panel with little increase in weight, producing
an efficient structure for resisting bending and
buckling loads. Sandwiches are found where
weight-saving is critical: in aircraft, trains, trucks
and cars, in portable structures, and in sports
equipment. Nature, too, makes use of sandwich
designs: sections through the human skull, the
wing of a bird and the stalk and leaves of many
plants show a low-density foam-like core separat-
ing solid faces. The faces carry most of the load,
so they must be stiff and strong; and they form the
exterior surfaces of the panel so they must tolerate
the environment in which they operate. The core
occupies most of the volume, it must be light, and
stiff and strong enough to carry the shear stresses
necessary to make the whole panel behave as a
load bearing unit, but if the core is much thicker
than the faces these stresses are small.
So far we have spoken of the sandwich as a
structure: faces of material A supported on a core
of material B, each with its own density and modu-
lus. But we can also think of it as a material with
its own set of properties, and this is useful because
it allows comparison with more conventional
materials. To do so we must analyze sandwich per-
formance [21,26–29]. We shall use, as a criterion
of excellence, the bending stiffness per unit width,
Sw divided by the mass per unit area, ma.
The bending stiffness of the panel per unit
width, Sw, is given by
Sw ⫽ (EI)sand ⫽ (19)
冋 册冦 冧
1
1 3
(d ⫺ c3)Ef BEftc
12 1⫹
2GcL2
where the dimensions, d, c, t and L are identified
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821 5815

Fig. 16. Foams and micro-truss structures are hybrids of material and space. Their mechanical response depends on their structure.
Foams are usually bending dominated (Eqs. (10) and (12)), and lie along a line of slope 2 on this chart. Micro-truss structures are
stretch-dominated (Eqs. (16) and (17)) and lie on a line of slope 1. Both extend the occupied area of this chart by many decades [42].

(d3/12 is the second moment of area of a homo-

geneous panel of thickness d, and we assume
td). Thus
6t
Ẽ ⫽ E (21)
d f
The panel has a density, r, that is the weighted
Fig. 17. The ultimate light, stiff hybrid: the sandwich panel. mean of its faces and core

in Fig. 16, Ef is Young’s modulus of the face sheets

and Gc is the shear modulus of the core. The
r⫽
2t
d f 冉 冊
2t
r ⫹ 1⫺ rc
d
(22)

numerical constant B depends only on the way the (just Eq. (1) with f replaced by 2t/d). The core den-
panel is loaded. Eq. (19) has two terms. The first, sity is chosen to be low, making the second term
in square brackets, is the stiffness if bending were small compared with the first. Neglecting it, and
the only mode of deformation. The second, in curly combining the last two equations gives

冉冊
brackets, is the knock-down in stiffness caused by
shear in the core. If the core adequately resists r
Ẽ⬇3 E ⫽ 3fEf (23)
shear (and this is easily achieved) the second term rf f
can be neglected. To think of the sandwich as a
material, we define an apparent modulus Ẽ—the Comparing this with Eq. (17) we find a gain in
modulus of a homogeneous material with the same stiffness by a factor of 3 (Es there equates to Ef
bending stiffness as the sandwich, requiring that here). The sandwich in flexure is approximately
three times more efficient than the most efficient
1 3 3 1 d3 fibrous composite, even when the fibers are all
(EI)sand ⫽ (d ⫺c )Ef ⫽ td2Ef ⫽ Ẽ (20)
12 2 12 aligned normal to the axis of bending.
5816 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
This, of course, is an idealization. The core
always shears somewhat, and it does have some
mass; a more precise analysis copes with this [29].
It leads to the performance shown in Fig. 18, which
has been constructed in the same way as Fig. 7. It
shows the modulus and density of A + B hybrids.
The shaded band is bounded by the upper and
lower bounds of Eqs. (1)–(3), describing particu-
late and fibrous composites. The flexural perform-
ance of the sandwich is shown as a dashed line; at
its mid-section it lies a factor of 3 above the upper
bound rule of mixtures of Eq. (2). The criterion of
excellence for minimum weight design with pre-
scribed bending stiffness, listed in Table 1, is that
of maximizing E1 / 3 / r. Contours of this criterion
are plotted as diagonal lines on the figure, increas-
ing towards the top left. The sandwich out-per-
forms all alternative hybrids of A + B.
4.6. Subdivision as a design variable
We have already seen how subdivision can
reduce stiffness (Fig. 12). Here we examine
another way in which it can be used: to impart
damage tolerance. A glass window, hit by a pro-
Fig. 18. Sandwich panels (broken line) extend the range of
flexural modulus per unit mass into areas not occupied by
monolithic materials. Their flexural moduli lie above that pre-
dicted by a rule of mixtures by a factor of approximately 3.
jectile, will shatter. One made of small glass
bricks, laid as bricks usually are, will lose a brick
or two but not shatter totally; it is damage-tolerant.
By sub-dividing and separating the material, a
crack in one segment does not penetrate into its
neighbors, allowing local but not global failure.
That is the principle of “topological toughening”.
Builders in stone and brick have exploited the idea
for thousands of years: both materials are almost
as brittle as glass, but buildings made of them—
even those made without cement (“dry-stone
building”)—survive ground movement, even earth-
quakes, through their ability to deform with some
local failure, but without total collapse.
Taking the simplest view, two things are neces-
sary for topological damage tolerance: discreteness
of the structural units, and an interlocking of the
units in such a way that the array as a whole can
carry load. Brick-like arrangements (Fig. 19a) are
damage tolerant in compression and shear, but dis-
integrate under tension. Strand and layer-like struc-
tures (shown earlier as Fig. 12) are damage-tolerant
in tension because if one strand fails the crack does
not penetrate its neighbors—the principle of multi-
strand ropes and cables. The jigsaw puzzle con-
figuration (Fig. 19b) carries in-plane tension, com-
pression, and shear, but at the cost of introducing
a stress concentration factor of about √R / r, where
R is the approximate radius of a unit and r that
of the interlock. Dyskin et al. [30–32] explore a
particular set of topologies that rely on compress-
ive or rigid boundary conditions to create continu-
ous layers that tolerate out-of-plane forces and
bending moments, illustrated in Fig. 19c. This is
done by creating interlocking units with non-planar
surfaces that have curvature both in the plane of
the surface and normal to it. Provided the array is
constrained at its periphery, the nesting shapes lim-
its the relative motion of the units, locking them
together. The bending stiffness of the array is pro-
portional to the stiffness of the boundary con-
straint, falling to zero as the constraint is relaxed.
Topological interlocking of this sort allows the for-
mation of continuous layers that can be used for
ceramic claddings or linings to give surface protec-
tion.
The damage tolerance can be understood in the
following way. We suppose that the units of the
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821 5817

Fig. 19. Examples of topological interlocking: discrete, unbonded structures that carry load. (a) Brick-like assemblies of rectangular
blocks carry axial compression (syy), but not tension or shear. (b) The 2-dimensional interlocking of a jig-saw puzzle carries in-plane
loads ( ± sxx, ± syy, sxy). (c) The units suggested by Dyskin et al., when assembled into a continuous layer and clamped within a
rigid boundary around its edge, can carry out-of-plane loads and bending moments ( ± sxz, ± syz, Mxz, Myz) (Fig. 15(c) derived from
Ref. [30]).

structure are all identical, each with a volume Vs,
and that they are assembled into a body of volume
Vt; there are therefore n = Vt / Vs segments. We
describe the probability of failure of a segment
under a uniform tensile stress s by a Weibull prob-
ability function:
stress, if sufficiently large, causes some segments
to fail. We refer to the fraction that has failed as
the damage, D. If loaded such that each segment
carries a uniform stress s, the damage is simply
Pf(Vs,s). If some segments fail, the body as a
whole remains intact; global failure requires that a

再 冎
fraction D∗ , the critical damage, (say, 10%) must
Vsm fail. Inverting Eq. (24) with V = Vs gives the global
Pf(V,s) ⫽ 1⫺exp⫺ (24)
Vosmo failure stress s∗s of the segmented body:
where m, Vo and so are constants [33,34]. If the
body were made of a single monolithic piece of
the brittle solid, this equation, with V = Vt, would
ss∗ ⫽ so 再 Vo
Vs
ln(1⫺D∗) 冎 1/m
(26)

describe the failure probability. To calculate the Thus segmentation increases the allowable design
design stress s∗t we set an acceptable value for P, stress from s∗t to s∗s , factor of

再 冎 再 冎
which we call P∗ (say 10⫺6, meaning that it is
ss∗ ln(1⫺D∗) 1/m
D∗ 1/m
acceptable if one in a million fail) and invert the ⫽ n ⬇ n ∗ (27)
equation to give s∗t ln(1⫺P∗) P

s∗t ⫽ so 再 Vo
Vt
ln(1⫺P∗) 冎
1/m
(25)
(expanding the logarithm as a series and retaining
the first term—an acceptable approximation for
small P∗ and D∗). Both n and D∗/P∗ are con-
Now consider the segmented body. A remote siderably greater than 1, so the equation suggests
5818 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
that segmentation always increases the design
stress.
The merit of this argument is its simplicity, illus-
trating how damage tolerance can arise. But it is a
little too simple. What are the underlying assump-
tions? First, that the number of segments is large;
only then can the damage D be reliably equated to
the failure probability Pf. Second, that the stress s
is uninfluenced by damage; in reality its average
value increases by the factor 1 / (1⫺D). Third, that
the stress is uniform; in reality both the inter-
locking shapes and the load shed by a failed seg-
ment onto its immediate neighbors increases the
stresses locally by a effective stress concentration
factor Kc. And fourth, that segment failure is
uncorrelated (“non-associated cracking”), not
linked (“associated cracking”). The first of these is
simply a pre-condition that must be met for seg-
mentation to be successful. The second and third
of these can be included approximately by mod-
ifying Eq. (27) to give:
冉 冊再 冎
s∗s 1⫺D∗ D∗ 1/m
⬇ n ∗
s∗t Kc P Continuum methods, on which many of the
The last depends on the values of the Weibull con-
stants, and requires numerical modeling of the kind
used by Curtin [35] and Shaw [36] to explore the
conditions for non-associated cracking. Eq. (28)
however is still useful—it allows for both a gain
and a loss in allowable design stress.
Fig. 20 gives an idea of the messages that it car-
Fig. 20. The ratio ss / st for the listed values of nD∗/P∗, with
Kc = 2 and D∗ = 0.1.
ries. It is constructed using Kc = 2, and values of
nD∗/P∗ between 104 and 1010. The Weibull modu-
lus m of brittle solids typically lies in the range 5
(typical of a porous brittle ceramic) and 30 (a value
that might describe the variability of a high techni-
cal ceramic). When the Weibull modulus is large—
more than 25—segmentation offers little gain,
indeed there may be a loss of allowable stress. But
when it is low (as it is in building materials like
brick and stone, and as it can be for certain cer-
amics and glasses), there is a gain. Segmentation
gives a new design variable. Since the only
requirement is that of interlocking shape, the seg-
ments can be made of different materials. Just as
builder constructing a wall can insert porous bricks
for ventilation and transparent bricks to admit
light, the designer of a segmented hybrid can add
functionality through the choice of material for
the units.
5. Scale: escaping the continuum bounds
(28)
arguments of Sections 3 and 4 are based, lead to
results that depend on shape, but are independent
of scale. Materials, however, are not continua; the
discreetness at the atomistic level introduces a
length scale that, as it is approached (even
remotely), causes properties to behave in ways that
differ, sometimes dramatically, from the simple
continuum estimates. The general result is that
mechanical and transport properties become scale-
dependent when the hybrid microstructure is sub-
micron.
This, the bread and butter of the materials scien-
tist, extends beyond the scope of this paper. Here
we simply list some of the origins of scale-depen-
dent properties. We have already seen how seg-
mentation can lead to scale-dependent damage tol-
erance. Small particles and fine fibers are generally
stronger than bulk samples of the same materials
because the probability that they contain a large
defect scales with their volume. Precipitation hard-
ened materials, aged to peak hardness, can
approach a strength of s∗⬇sp√f (where sp is the
strength of the particles and f their volume
fraction), compared with upper bound s∗⬇spf +
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
sm(1⫺f) of the continuum approximation. Fine-
grained materials, particularly those with grains of
a few nanometers, exhibit strengths that exceed
those of large-grained bulk samples. Scale, then,
can have a profound influence on mechanical
properties.
Transport properties, too, can be scale depen-
dent. The mean-free paths of electrons, phonons
and of diffusing atoms and molecules are limited
by the scale of the microstructure when this is
small—a fact exploited in micro-cellular foams to
give exceptional thermal insulation. Convection,
acoustic absorption and light scattering, too, are
directly linked to aspects of structural scale.
6. Summary and conclusions
The properties of engineering materials can be
thought of as defining the axes of a multi-dimen-
sional space with each property as a dimension.
Sections through this space can be mapped. These
maps reveal that some areas of property-space are
occupied, others are empty—there are holes. The
holes can sometimes be filled by making hybrids:
combinations of two (or more) materials, or of
material in space, in chosen configuration and
scale. Here we have surveyed conceptual tools for
suggesting and assessing hybrids to fill specified
needs. A useful starting point is the concept of a
hybrid as “A + B + shape + scale”. Successful
hybrids, as a rule, exploit the first three of these;
with micron and nanometer scale fabrication tech-
nologies now a reality, it becomes possible to add
the last, opening up wider horizons.
Continuum bounding methods give tools for
scanning the possibilities offered by a set of
hybrids, provided the scale is such that the con-
tinuum approximation applies. Shape—the way A
and B are configured—can extend the populated
areas of property-space in ways that complement
efforts to create new monolithic materials. Dis-
criminating choice of shape can enhance or dimin-
ish physical, mechanical, thermal and electrical
properties. Scale introduces a new variable. In
hybrids with structural units that are sub-micron, a
new length scale (basically that of the atom) makes
itself evident. Here the bounds break down, and
5819
continuum methods must be replaced by statistical
or dislocation mechanics.
Much, in a short paper such as this, has been
ignored. Fabricating successful hybrids can be dif-
ficult and expensive (but so, too, is the alternative
of seeking to develop a new monolithic material).
Part of the difficulty stems from the multitude of
possible choices: choice of materials, choice of
process to combine them, and choice of the internal
geometry and topology of the constitutive
materials. Part derives from the need to make these
choices in such a way as to optimally meet a set
of design requirements. The hybrid must be both
feasible and optimal.
To explore this design space efficiently optimiz-
ation tools are needed. The starting point is a sim-
ple “screening” of options—the obvious way for-
ward when the selection space is discrete. Thus it
is possible to create a database of, say, composites
by computing the properties of virtual materials
with 5%, 10%, 15%… of reinforcing fibers and
discrete choices of lay-up; selection proceeds by
rejecting all the combinations that fail to meet the
design requirements. This method becomes
impractical when many material choices and lay-
ups are allowed. When the optimization variables
are continuous (such as the volume fraction of
reinforcement in a composite with pre-chosen
constituents), linear programming or steepest
descent methods can be efficient. When the “poten-
tial landscape” is very rugged, simulated annealing
can offer a practical alternative. When the variables
are both discrete (such as the choice of materials)
and continuous (the thicknesses of face sheets and
core of a sandwich, for instance), a genetic algor-
ithm, which allows an unevenly populated space to
be explored, can be efficient. Refs. [37–41]
examples of their use in hybrid design.
These tools allow promising candidates to be
identified. They need the back-up of “expert tools”
advising on the compatibility of the constituents,
the practicality of processes to assemble them into
a hybrid, and the loss of properties (the “knock-
down factor” relating real and ideal hybrids) for
a given combination of the materials, architecture
and process.
5820 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821
Acknowledgements
The ideas, methods and tools described here
have evolved over the past 15 years. Numerous
colleagues in many countries have (sometimes
unknowingly) stimulated or contributed to this
evolution. Among these we would particularly like
to recognize Profs. Mick Brown, Chris Calladine,
Norman Fleck and David Cebon (all of Cambridge
University), Dave Embury (McMaster University,
Canada), Tony Evans (UCSB), John Hutchinson
(Harvard University) and Haydn Wadley (UVA),
and Dr. Luc Salvo (University of Grenoble).
References
[1] Ashby MF. On the engineering properties of materials.
Acta Metall 1989;37:1273–93.
[2] Ashby MF. Materials selection in mechanical design, 2nd
ed. Oxford: Butterworth Heinemann, 1999.
[3] Kromm FX, Quenisset JM, Harry R, Lorriot T. An
example of multimaterial design. Adv Eng Mater
2002;4:371–4.
[4] Watt JP, Davies GF, O’Connell RJ. Reviews of geophys-
ics and space physics 1976;14:541.
[5] Schoutens JE, Zarate DA. Composites 1986;17:188.
[6] Chamis CC. In: Engineers guide to composite materials.
Materials Park, Ohio, USA: American Society of Metals,
1987:3-8–3-24.
[7] Ashby MF. Criteria for selecting the components of com-
posites. Acta Mater 1993;41:1313–35.
[8] Eshelby JD. Proc R Soc Lond 1957;A241:376.
[9] Hashin Z, Strikman S. J Appl Phys 1962;33:3125.
[10] Stauffer D, Aharony A. Introduction to percolation theory,
2nd ed. London, UK: Taylor and Francis, 1994 revised.
[11] Fitzpatrick JP, Malt RB, Spaepen F. Phys Lett
1974;A47:207.
[12] Brown LM. A simple model of a metal non-metal tran-
sition. Physics Education 1977;July issue:318–20.
[13] Last BJ, Thouless DJ. Phys Rev Lett 1971;27:1719.
[14] Nielsen LE. Thermal conductivity of particulate-filled
polymers. J Appl Polym Sci 1973;17:3819.
[15] Nielsen LE. The thermal and electrical conductivity of
two-phase systems. Ind Eng Chem Fund 1974;13:17.
[16] Bigg DM. Conductive polymeric compositions. Polym
Eng Sci 1977;17:892.
[17] Bigg DM. Mechanical and conductive properties of metal
fibre-filled polymer composites. Composites 1979;April
issue:95–100.
[18] Yi Y-B, Sastry AM. Analytical approximation of the 2-
dimensional percolation threshold for fiels of over lapping
ellipses. Phys Rev 2002;E66 066130-1–066130-8.
[19] Bréchet Y, Cavaille JY, Chabert E, Chazeau L, Dendievel
R, Flandin L, Gautier C. Polymer-based nanocompoosite:
effect of filler–filler and filler–matrix interactions. Adv
Eng Mater 2001;3(8):571.
[20] Ashby MF. On material and shape. Acta Mater
1991;39:1025.
[21] Gibson LJ, Ashby MF. Cellular solids, structure and
properties, 2nd ed. Cambridge, UK: Cambridge University
Press, 1997.
[22] Deshpande VS, Ashby MF, Fleck NA. Foam topology:
bending versus stretching dominated architectures. Acta
Mater 2001;49:1035–40.
[23] Deshpande VS, Fleck NA, Ashby MF. Effective properties
of the octet-truss lattice material. J Mech Phys Sol
2001;49:1747–69.
[24] Maxwell JC. On the calculation of the equilibrium and
stiffness for frames. Phil Mag 1864;27:294.
[25] Calladine CR. Theory of shell structures. Cambridge, UK:
Cambridge University Press, 1983.
[26] Allen HG. Analysis and design of structural sandwich
panels. Oxford, UK: Pergamon Press, 1969.
[27] Cordon J. Honeycomb structure. In: Dostal CA, editor.
Engineered materials handbook. Metals Park, Ohio, USA:
ASM International; 1990:721–8.
[28] Zenkert D. In: An introduction to sandwich construction.
Solihull, London, UK: Engineering Advisory Services
Ltd., Chameleon Press Ltd, 1995, ISBN 0 947817778.
[29] Ashby MF, Evans AG, Fleck NA, Gibson LJ, Hutchinson
JW, Wadley HNG. Metal foams, a design guide. Oxford,
UK: Butterworth Heinemann, 2000.
[30] Dyskin AV, Estrin Y, Kanel-Belov AJ, Pasternak E.
Toughening by fragmentation: how topology helps. Adv
Eng Mater 2001;3:885–8.
[31] Dyskin AV, Estrin Y, Kanel-Belov AJ, Pasternak E. Topo-
logical interlocking of platonic solids: a way to new
materials and structures. Phil Mag 2003;83:197–203.
[32] Dyskin AV, Estrin Y, Pasternak E, Kohr HC, Kanel-Belov
AJ. Fracture resistant structures based on topological inter-
locking with non-planar contacts. Adv Eng Mater
2003;5(3):116–9.
[33] Weibull W. J Appl Mech 1951;18:293.
[34] Davidge RW. Mechanical behavior of ceramics. Cam-
bridge, UK: Cambridge University Press, 1986.
[35] Curtin WA. Acta Metall et Mater 1993;41:1369.
[36] Shaw MC. The fracture mode of ceramic/metal multilay-
ers: role of the interface. Key Eng Mater 1996;116-
117:261–78.
[37] Salvo L, Bréchet Y, Pechambert P, Bassetti D, Jantsen A.
Logiciels de sélection des composites. Matériaux et Tech-
nique 1998;5:31.
[38] Landru D, Bréchet Y. New design tools for materials and
process selection. Materiaux et Techniques 2002;4:6.
[39] Bréchet Y, Bassetti D, Landru D, Salvo L. Challenges in
material and process selection. Prog Mater Sci
2001;46:407.
[40] Bréchet Y, Ashby MF, Salvo L. Selection des materiaux
et des procedes de mis en oeuvre. Switzerland: Les Presses
 M.F. Ashby, Y.J.M. Bréchet / Acta Materialia 51 (2003) 5801–5821 5821

Polytechniques et Universitaires Romandes de Lausanne, [42] CES 4. The Cambridge Material Selector, Granta Design,
2002. Cambridge. 2003;(www.Grantadesign.com)
[41] Ashby M, Bréchet Y, Cebon D, Salvo L. Materials and
process selection strategies. To appear in Advanced
Engineering Materials, 2003.