Editorial officEs

LONDON www.nature.com/naturematerials
The Macmillan Building, 4 Crinan Street, London N1 9XW
T: +44 207 833 4000 F: +44 207 843 4563
Editor Vincent Dusastre
Senior Editors Victoria Cleave, Joerg Heber, Fabio Pulizzi
Associate Editors Hilary Crichton, Alison Stoddart
Consultant Editor Philip Ball
Senior Production Editor Derna Simpson
Assistant Production Editor Jenny Marsden
Art Editor David Shand
Editorial Assistant Theresa Tuson
ManagEMEnt officEs
nPg london nature@nature.com
The Macmillan Building, 4 Crinan Street, London N1 9XW
T: +44 207 833 4000 F: +44 207 843 4563
Managing Director Steven Inchcoombe
Publishing Director David Swinbanks
Publisher Jason Wilde
Associate Publisher Emma Green
Editor-in-Chief, Nature Publications Philip Campbell
Marketing Director Della Sar
Operations Director John Carroll
Director Of Web Publishing Timo Hannay
Associate Director, UK Production Jenny Henderson
Head Of Marketing, Physical Sciences Jane Macmillan
Marketing Manager, Physical Sciences Gurpreet Gill-Bains
Editorial Production Director James McQuat
Managing Production Editor Donald McDonald
Senior Copy Editor Jane Morris
Web Production Manager, UK Deborah Anthony
Production Director Yvonne Strong
Senior Production Controller Kelly Hopkins
Production Controller Emilia Orviss
nPg nEw York nature@natureny.com
75 Varick Street, 9th Floor, New York, NY 10013-1917
T: +1 212 726 9200 F: +1 212 696 9006
Chief Technology Officer Howard Ratner
Head Of Web Services Anthony Barrera
Executive Editor Linda Miller
nPg asia-Pacific nature@natureasia.com
Chiyoda Building 2-37 Ichigayatamachi, Shinjuku-Ku, Tokyo 162-0843 Japan
T: +81 3 3267 8751 F: +81 3 3267 8746
Associate Director Asia-Pacific Antoine E. Bocquet
Manager Koichi Nakamura
Operations Director Hiroshi Minemura
Asia-Pacific Sales Director Kate Yoneyama
Marketing Manager Masahiro Yamashita
Production Manager Takesh Murakami
Asia-Pacific Sales Manager Ken Mikami
nPg india npgindia@nature.com
3a, 4th Floor, Dlf Corporate Park, GurGaon 122002, India
T: +91 12 4288 1054/55 F: +91 12 4288 1052
Head Of Business Development Debashish Brahmachari
Sales And Marketing Manager Harpal Singh Gill
disPlaY advErtising physicalsciences@nature.com
Global Head of Display Advertising Andrew Douglas T: +44 207 843 4975 F: +44 207 843 4996
Asia-Pacific Sales Director Kate Yoneyama T: +81 3 3267 8765 F: +81 3 3267 8746
Advertising Director George Lui T: +44 207 843 4966 F: +44 207 843 4749
Advertising Manager, Physical Sciences Simon Allardice T: +1 415 403 9034 F: +1 415 781 3805
Asia-Pacific Display Advertising Manager Ken Mikami T: +81 3 3267 8751 F: +81 3 3267 8746
naturEjobs naturejobs@nature.com
European Sales Manager Andrew Douglas T: +44 207 843 4975 F: +44 207 843 4996
US Sales Manager Kenneth Finnegan T: +44 207 843 4975 F: +44 207 843 4996
Asia-Pacific Sales Manager Ayako Watanabe T: +81 3 3267 8765 F: +81 3 3267 8746
rEPrints reprint@nature.com
For commercial reprint orders of 600 or more, please contact:
US/Canada: reprints@natureny.com
Northern Europe/UK/ROW: reprints@nature.com
Southern Europe/Latin America: v.jurado@macmillanmedical.com
Asia-Pacific: m.kurosaki@natureasia.com
India: d.brahmachari@nature.com
sitE licEnsE businEss unit
Americas T: +1 888 331 6288 institutions@natureny.com
Asia/Pacific T: +81 3 3267 8751 institutions@natureasia.com
Australia/New Zealand T: +61 3 9825 1160 nature@macmillan.com.au
Europe/Row T: +44 207 843 4759 institutions@nature.com
India T: +91 124 2881054/55 npgindia@nature.com
custoMEr sErvicE
For all print and online assistance, please visit www.nature.com/help
Senior Global Customer Service Manager Gerald Coppin
ORIGINAL RESEARCH TyPESET By River Valley Technologies, www.river-valley.com
PRINTED By Wyndeham Grange, www.wyndeham.co.uk

© 2009 Macmillan Publishers Limited. All rights reserved


Cover image
Two independent studies
demonstrate how control over
magnetic molecules
on surfaces may lead to new
spintronics applications.
Letters p189 and p194
News & Views p165
oN THe Cover
Biomaterials
Model behaviour
Review Article p175
Nanocatalysis
Catalysis by clusters
Letter p213; News & Views p167
epitaxial graphene
The right atmosphere
Letter p203; News & Views p171
NaTure maTerialS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
march 2009 VOL 8 ISSUE 3
editorial
161 Putting evolution to good use
research highlights
163 our choice from the recent literature
News & Views
165 molecular magnets: How a nightmare turns into a vision
Heiko Wende
166 material witness: Shaping fate
Philip Ball
167 Nanocatalysis: Staying put
Gianfranco Pacchioni
168 multiferroics: a way forward along domain walls
Hélène Béa and Patrycja Paruch
170 Transition metals: Can metals be a liquid glass?
Daniel Errandonea
171 epitaxial graphene: How silicon leaves the scene
Peter Sutter
173 Shaping biological matter
Vadim A. Frolov and Joshua Zimmerberg
reView article
175 Deformation and failure of protein materials in physiologically extreme
conditions and disease
Markus J. Buehler and Yu Ching Yung
letters
189 Supramolecular control of the magnetic anisotropy in two-dimensional
high-spin Fe arrays at a metal interface
Pietro Gambardella, Sebastian Stepanow, Alexandre Dmitriev, Jan Honolka,
Frank M. F. de Groot, Magalí Lingenfelder, Subhra Sen Gupta, D. D. Sarma,
Peter Bencok, Stefan Stanescu, Sylvain Clair, Stéphane Pons, Nian Lin,
Ari P. Seitsonen, Harald Brune, Johannes V. Barth and Klaus Kern
→N&V p165
194 magnetic memory of a single-molecule quantum magnet wired to a
gold surface
Matteo Mannini, Francesco Pineider, Philippe Sainctavit, Chiara Danieli,
Edwige Otero, Corrado Sciancalepore, Anna Maria Talarico, Marie-Anne Arrio,
Andrea Cornia, Dante Gatteschi and Roberta Sessoli
→N&V p165
198 room-temperature defect-engineered spin filter based on a
non-magnetic semiconductor
X. J. Wang, I. A. Buyanova, F. Zhao, D. Lagarde, A. Balocchi, X. Marie, C.W. Tu,
J. C. Harmand and W. M. Chen
© 2009 Macmillan Publishers Limited. All rights reserved
 203 Towards wafer-size graphene layers by atmospheric pressure
graphitization of silicon carbide
Konstantin V. Emtsev, Aaron Bostwick, Karsten Horn, Johannes Jobst,
Gary L. Kellogg, Lothar Ley, Jessica L. McChesney, Taisuke Ohta,
Sergey A. Reshanov, Jonas Röhrl, Eli Rotenberg, Andreas K. Schmid,
Daniel Waldmann, Heiko B.Weber and Thomas Seyller
→N&V p171
208 endohedral fullerenes for organic photovoltaic devices
Russel B. Ross, Claudia M. Cardona, Dirk M. Guldi,
Shankara Gayathri Sankaranarayanan, Matthew O. Reese,
Nikos Kopidakis, Jeff Peet, Bright Walker, Guillermo C. Bazan,
Edward Van Keuren, Brian C. Holloway and Martin Drees
Thermal annealing of SiC
produces graphene layers 213 Subnanometre platinum clusters as highly active and selective catalysts
on an insulating substrate, for the oxidative dehydrogenation of propane
but the material is highly
Stefan Vajda, Michael J. Pellin, Jeffrey P. Greeley, Christopher L. Marshall,
inhomogeneous. it is now shown
that using an argon atmosphere
Larry A. Curtiss, Gregory A. Ballentine, JeffreyW. Elam, Stephanie Catillon-Mucherie,
during annealing improves the Paul C. Redfern, Faisal Mehmood and Peter Zapol
uniformity of the graphene layers →N&V p167
dramatically, and yields better
transport characteristics. This
217 mesoporous germanium-rich chalcogenido frameworks with highly
is a very important result for the polarizable surfaces and relevance to gas separation
development of graphene-based Gerasimos S. Armatas and Mercouri G. Kanatzidis
electronic devices.
Letter p203; News & Views p171
articles
223 Shear-induced anisotropic plastic flow from body-centred-cubic tantalum
before melting
Christine J. Wu, Per Söderlind, James N. Glosli and John E. Klepeis
→N&V p170
229 Conduction at domain walls in oxide multiferroics
J. Seidel, L. W. Martin, Q. He, Q. Zhan, Y.-H. Chu, A. Rother, M. E. Hawkridge,
P. Maksymovych, P. Yu, M. Gajek, N. Balke, S. V. Kalinin, S. Gemming, F. Wang,
G. Catalan, J. F. Scott, N. A. Spaldin, J. Orenstein and R. Ramesh
→N&V p168
235 The influence of edge structure on the electronic properties of graphene
quantum dots and nanoribbons
Catalytic oxidative dehydrogenation Kyle A. Ritter and Joseph W. Lyding
of alkanes is limited by poor
activity and/or selectivity.
efficient conversion of propane to classifieds
propylene is now achieved using See the back pages
subnanometre platinum clusters
stabilized on alumina supports.
The clusters are substantially more
active than conventional catalysts,
and are highly selective towards
propylene formation.
Letter p213; News & Views p167

Nature Materials (ISSN 1476-1122) is published monthly by Nature Publishing Group (Porters South, 4 Crinan Street, London N1 9XW, UK). Editorial Office: Porters South, 4 Crinan Street, London N1 9XW, UK. Telephone:
+44 (0)20 7833 4000. Fax: +44 (0)20 7843 4563. Email: materials@nature.com. North American Advertising: Nature Materials, 75 Varick Street, 9th Floor, New York, NY 10013-1917, USA. Telephone: +1 212 726 9200.
Fax: +1 212 696 9006. European Advertising: Nature Materials, Porters South, 4 Crinan Street, London N1 9XW, UK. Telephone: +44 (0)20 7833 4000. Fax: +44 (0)20 7843 4749. New subscriptions/renewals/changes of
address/back issues and all other customer service questions should be addressed to: North America: Nature Publishing Group, Customer Services Department, 75 Varick Street, 9th Floor, New York, NY 10013-1917, USA.
Telephone: +1 (866) 363 7860. Fax: +1 (212) 334 0879. Outside North America: Nature Publishing Group, Subscriptions Department, Brunel Road, Houndmills, Basingstoke, Hants, RG21 6XS, UK. Telephone: +44 (0)1256
329242. Fax: +44 (0)1256 812358. Annual Subscription Rates: US/Canada US$3060 (Canada add 5% GST) (institutional/corporate), US$199 (Canada add 5% GST) (personal 1 year); UK/RoW (excluding Europe and Japan)
£1570 (institutional/corporate), £106 (personal 1 year); Europe ¤2430 (institutional/corporate), ¤164 (personal 1 year). Japan: contact NPG Nature Asia-Pacific, Chiyoda Bldg. 5F, 2-37 Ichigayatamachi, Shinjuku-Ku, Tokyo,
162-0843, JAPAN. Telephone: +81 3 3267 8751. For single back issue prices contact the publisher. Subscription information is available at the Nature Materials homepage at http://www.nature.com/naturematerials. Application
to mail Periodicals postage is paid at Rahway NJ. Postmaster: send address changes to Nature Materials Subscriptions Department, Brunel Road, Houndmills, Basingstoke, Hants, RG21 6XS, UK or Nature Materials Subscriptions
Department, PO Box 5054, Brentwood, TN 37024-5054, USA. Reprints: Nature Materials Reprints Department, Porters South, 4 Crinan Street, London N1 9XW, UK. © 2009 Macmillan Publishers Limited. All rights reserved.

NaTure maTerialS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

© 2009 Macmillan Publishers Limited. All rights reserved


Putting evolution to good use
Even materials scientists have reason to celebrate the seminal insights of Charles Darwin on
his bicentenary.
This year, Charles Darwin is everywhere.
The man who solved the puzzle of evolution
was born 200 years ago, and it is 150 years
since the publication of On the Origin of
Species. Darwin’s work transformed our
understanding of the natural world and our
place within it, but one might be forgiven
for thinking that this scarcely has much
relevance to the science and engineering
of materials.
Yet Darwin’s explanation for how
complex form evolved in the living world
is very much a matter of engineering.
As biologist Steve Jones puts it in his
new book Darwin’s Island (Little, Brown,
London, 2009), ‘Natural selection is a
factory that makes almost impossible
things. It manufactures what seems like
design with no need for a designer.’ Put
that way, the process of evolution no longer
seems a purely biological phenomenon,
but begins to sound potentially useful to
the engineer.
It requires two things: random variation,
which offers a wide range of trial solutions
to a problem, and natural selection,
which picks the best and discards the
others. Iterated repeatedly, these processes
combine to create artefacts of extraordinary
delicacy and ingenuity, from the human eye
to the gecko’s foot.
The traditional engineering approach is
very different, designing not by scattershot
experimentation and careful sifting but
by rational planning. That, on the whole,
is a better option — imagine trying to
make an automobile by random assembly
of randomly machined components.
Darwinian evolution, lacking the luxury
of foresight, achieves its wonders only
because it has large populations and
geological timescales to work on. But
rational design isn’t always possible: we
might not understand the properties
of the components or the target well
enough, or the combinatorial universe of
potential choices may be too vast. Both
are sometimes true in materials science,
for example in the case of compound
superconductors. That’s why, in this and
other cases, something akin to Darwin’s
approach has been tried.
Combinatorial materials synthesis may
have an evolutionary aspect when iterated:
some selection criterion is used to extract
nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
Trees avoid sharp-cornered notches by growing
wood in a shape that reduces concentrations of
tensile stress.
good candidates that are then reshuffled
in the next round. That works well for
in vitro methods of making biomolecules
with new functions, but is rarely attempted
in materials. It is, however, more or less
the basis of Monte Carlo simulated-
annealing methods in computer simulation,
where some ‘cost function’ stands in
for ‘fitness’ to evaluate the trial-and-
error reorganization of the constituents.
Genetic algorithms for surveying energy
landscapes are particularly useful for the
recalcitrant problem of first-principles
structure prediction1.
Some examples of materials design
stake a stronger claim to Darwinian
methodology. Claus Mattheck of the
Karlsruhe Institute of Technology has
long championed the tree as a self-
adjusting materials system that ‘finds’ a
good structural response to stress2. The
‘evolution’ of this system is guided by cell
responses to stress, rather than randomly
varied, but the resulting shapes can serve as
inspiration for a more strictly evolutionary
route enacted by computation. Mattheck
© 2009 Macmillan Publishers Limited. All rights reserved
editorial
and co-workers have used this idea to make
notch structures with little propensity
to initiate cracks. In a related approach,
Paolo Vannucci, Angela Vincenti and their
colleagues in France have devised a genetic
algorithm called BIANCA for optimizing a
given property of a laminate: for example,
zero or isotropic thermal expansion
coefficients, maximum buckling stiffness,
or a particular piezoelectric deformation3–5.
And Jim Platts of Cambridge University
has used iterative computational evolution
to deduce the optimal placement of
fibres in a composite material6,7. Whereas
Mattheck takes inspiration from wood,
Platts and colleagues look to bone, in
which material is actively removed as well
as added in the adaptive sculpting process.
That the growth and morphology of bone
can be seen as a design process involving
minimization was first recognized by the
Australian engineer A. G. M. Michell in
1904, but his perceptive insight languished
until sufficient computer power became
available to make use of it.
Minimization problems like this are very
hard from any analytical first-principles
perspective. An evolutionary genetic-
algorithm search of the design space is
not truly Darwinian, because the target
property is pre-determined rather than
emerging as a consequence of selection by
efficiency of replication. In that sense, it
is a form of directed evolution, more akin
to the selective breeding by the livestock
managers and pigeon fanciers that proved
so informative to Darwin. But it shares with
true Darwinism the use of a population
spread all over the search space, and the
intelligent exploitation of information
about that space by recombining aspects of
both ‘bad’ and ‘good’ designs to guide the
search towards an optimal solution. The
great man would doubtless approve. ❐
References
1. Woodley, S. M. & Catlow, R. Nature Mater. 7, 937–946 (2008).
2. Mattheck, C. Secret Design Rules of Nature (Verlag
Forschungszentrum Karlsruhe, 2007).
3. Vincenti, A. et al. Proc. 14th Int. Conf. Compos. Mater.
Paper 1341 (2003).
4. Vannucci, P. Int. J. Struct. Mulidisc. Optimis. 31, 378–387 (2006).
5. Vannucci, P. & Vincenti, A. Compos. Struct. 79, 454–466 (2007).
6. Platts, M. J. in Nature and Design (eds Collins, M. W. et al.)
157–167 (WIT Press, Southampton, 2005).
7. Platts, M. J. Paper presented at Opticon ‘99 Optimisation
Software Methods &Applications Conf., Newport Beach,
October 1999.
161
 research highlights
infiltration rules can damage the wire, altering its properties.
Langmuir 25, 1181–1187 (2008) Furthermore, the wire is normally placed
horizontally onto the new substrate and
Template synthesis is one of the simplest and the interaction between the two interferes
most easily scalable methods for making one- with the lasing capabilities. By placing a low
dimensional polymer nanostructures. These density of gold catalytic seeds on the growth
structures, however, are highly dependent substrate, Peidong Yang and colleagues
on the synthesis parameters, which can be produced vertical ZnO nanowires far

an advantage when looking for variety, but
not if exact structures are required. Martin
and Mijangos have investigated the range
of polymer nanostructures available via
template synthesis and produced guidelines
for achieving different results. The authors
used the widely studied porous anodic
aluminium oxide templates. They found
that the structures obtained depended on
infiltration method — wetting-based, vacuum
or spin — and the diameter of the template:
narrower pores resulted in nanofibres rather
than nanotubes being created. Composite
tubes containing magnetic nanoparticles
were created by melting a bulk polystyrene/
iron–platinum composite on the surface of
the template. The small size of the particles
(3–4 nm) compared with the thickness of the
nanotube walls (70 nm) enabled the particles
to infiltrate the template with polymer. The
rules should prove valuable in simplifying the
creation of arrays of polymer nanostructures
of various aspect ratios.
K. S. NovoSelov
enough apart to allow the characterization
of individual wires still attached to their
native substrate. Not only were they able to
uncover lasing properties such as the Fabry–
Pérot lasing modes, they also produced
way that conducting electrons are removed the first height-selective 3D maps of the
and an energy gap is opened. The pristine photoluminescence emission of nanowires
material investigated shows a resistivity by using UV-laser scanning confocal
independent of temperature, which is typical microscopy. Such detailed characterization
of graphene. After exposure to hydrogen of heterogeneous nanostructure optical
plasma, however, the resistivity increases properties should aid research aimed at using
considerably and shows a temperature them in optoelectronic and solar applications.
dependence typical of insulators. Most
importantly, the modification can be perfectly
reversed by annealing the hydrogenated
Crossbar memories
Nano Lett. doi:10.1021/nl8037689 (2009)
sample. Transmission electron microscopy
images indicate that the structure of the
hydrogenated samples is comparable to that of
graphane, a material that had previously only
been studied theoretically. Aside from the
clear potential for electronic applications, the
result demonstrates the possibility of creating
a range of two-dimensional molecules by
chemical modification of graphene.

© 2009 ACS
From graphene to graphane
Science 323, 610–631 (2009) Vertical nanocavities
J. Am. Chem. Soc. doi:10.1021/ja8092339 (2009)
Among the various routes for controlling
the electronic properties of graphene, Single-crystalline ZnO nanowires behave With the end of Moore’s law looming in
chemical modification seems to be the least as lasing nanocavities — affording a gain the not-too-distant future, researchers
investigated. Kostya Novoselov and colleagues medium and a resonant cavity owing to are intensively researching alternatives
have now demonstrated that hydrogenation the reflectivity of the planar end-facets. to silicon electronics. One of the more
enables the electronic structure of this single Usually the wires are produced in closely promising approaches is to use self-assembled
layer graphite to be modified, transforming packed arrays, so characterizing the optical nanostructures to complement existing top-
it into an insulator. The secret is that on properties of a single wire requires the down technology. In particular, crossbar
attaching hydrogen atoms to graphene, transfer of one wire from its initial growth arrays promise very high integration densities
the carbon bond is transformed in such a substrate onto another substrate. This process that can easily exceed those achieved with
the present technology. Researchers from the
University of Michigan have now realized
transparent paper Adv. Mater. doi:10.1002/adma.200803174 (2009) a non-volatile crossbar memory array
capable of storing up to 1 kbit of information.
As well as being optically transparent, glass has a low thermal expansion coefficient The array is based on silver nanowires
making it suitable for use in electronic devices. Conversely, it is fragile; therefore a stronger, crossing sandwiched amorphous/p-type
flexible material that retains the thermal expansion coefficient of glass is desirable. Now silicon nanowires. At their intersections,
Masaya Nogi and colleagues have produced a material satisfying these requirements Ag/a-Si/p-Si structures form. The amorphous
from the main constituent of paper — cellulose. A suspension of wood nanofibres is silicon acts as a memristive switch, so that a
filtered, producing a wet sheet of uniformly positioned fibres. Slowly drying this while it is voltage applied between Ag nanowire and p-Si
compressed between meshwire/filter paper layers gives a highly dense sheet of nanofibres layer is able to switch the material between a
bound by hydrogen bonds with few or no cavities. The lack of interstices between fibres high and low resistance state. This crossbar
suppresses light scattering, and once the surface has been smoothed by polishing, the array has write speeds faster than 10 ns, good
sheet becomes up to 71% transparent at a wavelength of 600 nm. The ‘cellulose nanofibre endurance and long retention rates. With
paper’ can be folded and has good mechanical properties. Alternative ways of smoothing a yield of 92% and the possibility of scaling
the surface can prevent cellulose deterioration, making it a prime candidate for roll-to-roll the size of a single element below 30 nm, the
processing for the production of flexible electronic devices such as solar cells. devices establish the potential of these arrays
for non-volatile memory applications.

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 163

© 2009 Macmillan Publishers Limited. All rights reserved

 news & views
Molecular Magnets

How a nightmare turns into a vision

Two independent studies demonstrate how control over magnetic molecules on surfaces may lead to new
spintronics applications.

Heiko Wende

W
olfgang Pauli once claimed that due to an exchange interaction within issue: in contrast to the work on the single
God created the volume but the molecule. molecular magnets by Mannini et al.,
the surface was invented by Gambardella and colleagues follow Gambardella et al. observe no coupling
the devil. Maybe for this reason surfaces a different route to control magnetism2. of the Fe spins in their supramolecular
have a sinister attraction for scientists At first they assemble a two-dimensional network through, for example,
fascinated by the possibilities of completely supramolecular network in situ on a copper intramolecular exchange interactions. In
new effects and hence functionality single crystal. Then they demonstrate that their case, the Fe spins are ordered only by
at surfaces. But the interactions of the preferred magnetic orientation of the the applied magnetic field. However, for
molecular adsorbates with surfaces can be system, its magnetic anisotropy, can be the single molecular magnets investigated
complicated and the experimental control controlled by the adsorption of oxygen by Mannini et al., the spins within one
of these systems can easily turn into a molecules. Without the adsorbed oxygen molecule show ordering phenomena at
nightmare. Two works published in this molecules, the supramolecular assembly a temperature of 0.5 K. Still, owing to
issue by Matteo Mannini and colleagues1 consisting of Fe and benzenedicarboxylic resonant tunnelling effects in zero field,
(page 194) and Pietro Gambardella and acid can be magnetized easily parallel to the remanent magnetization of their single
colleagues2 (page 189) go through this the surface plane. After oxygen adsorption, molecular magnets remains pretty small.
nightmare of surface physics in order this easy direction turns out of the plane. As both studies are carried out at very
to come closer to realizing the vision of The use of oxygen as an adsorbate is low temperatures in the kelvin and even
molecular spintronics. particularly interesting, as in another recent sub-kelvin regime, and the important
In addition to conventional electronics, work it is demonstrated that the magnetic magnetic effects are determined at sizeable
where electron charge is used to carry coupling of paramagnetic porphyrin applied magnetic fields ranging from
information, the basic idea in spintronics molecules to ferromagnetic surfaces can be 0.25 T up to 6 T, an immediate realization
is to exploit the spin, where the use of tailored through oxygen atoms3. of a molecular spintronic device is not
magnetic systems allows control through, It is helpful to look at the similarities yet possible. Nevertheless, both works
for example, magnetic fields. Magnetic but also the fundamental differences demonstrate the current possibilities and
molecules are widely studied as suitable between the two studies published in this the new frontiers of surface science. An
candidates for spintronics applications.
Their advantage is their broad flexibility,
shown by the fact that complex organic
molecules are also nature’s preferred
systems for realizing highly complex
processes. However, molecules must
always be seen as part of the environment
they are placed in, and to use molecular
spintronics for real, a fundamental
understanding of the basic interactions
of magnetic molecules with surfaces is
essential. It is in this context that these two
studies move a step closer to the molecular
spintronics goal.
In the work by Mannini et al., the focal
point is the identification of a magnetic
memory effect of a monolayer of single-
molecule magnets consisting of Fe4
complexes wired to a gold surface1 (Fig. 1).
© AndreA CorniA

They achieved this by demonstrating

that the single molecular magnets show
a magnetic hysteresis, meaning that the
behaviour of the system depends on
its history. Under certain conditions,
controlled by temperature and the Figure 1 | Controlling magnetic molecules on a surface. In the work by Mannini et al.1 single-molecule
applied magnetic field, the Fe spins in the magnets are bonded to a surface and characterized by the highly sensitive X-ray magnetic circular
individual molecules undergo ordering dichroism technique1.

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 165

© 2009 Macmillan Publishers Limited. All rights reserved

news & views

important conclusion of both studies
is that the interaction of the magnetic
molecules with the surface does not
destroy the magnetic behaviour. In
particular, in the study by Mannini et al.
the single molecular magnets were
connected to the surface by thiolate-
terminated aliphatic chains that allow
direct control of the magnetic interaction1.
This bonding was achieved through a
wet-chemical synthesis making use of
self-assembly effects.
In contrast, the supramolecular network
studied by Gambardella and colleagues
was constructed using self-assembly by
evaporating a molecular precursor layer
of terephthalic acid (TPA) followed by
deposition of Fe atoms from an electron-
beam evaporator under ultrahigh vacuum
conditions. Interestingly, both studies
Material Witness
use X-ray absorption spectroscopy at
modern third-generation synchrotron
radiation facilities to reveal the secrets
of their molecules. The X-ray magnetic
circular dichroism technique provides
the necessary sensitivity to probe the
magnetic properties of the molecules in an
element-specific manner.
With these techniques it is shown
that Fe does indeed connect to the TPA
molecules, and thereby the interaction
with the Cu substrate is weakened2. To
support these findings on the electronic
structure, the experimental results of
Gambardella et al. are accompanied by
solid density functional calculations and
atomic multiplet calculations to model the
experimental X-ray absorption spectra2.
In ref. 1 the experimental spectra reveal
that the single molecular behaviour of the
Fe4 complexes is not destroyed by their
connection to the surface, owing to their
structural stability and redox robustness.
This is a clear advance over well-studied
single molecular systems such as Mn12.
In conclusion, both studies demonstrate
that the hell that is surface science is not
the worst place to be, and that fascinating
science can emerge — as long as the place
is cooled down sufficiently. ❐
Heiko Wende is at the Fachbereich Physik and
Center for Nanointegration Duisburg-Essen
(CeNIDE), Universität Duisburg-Essen, 47048
Duisburg, Germany.
e-mail: heiko.wende@uni-due.de
references
1. Mannini, M. et al. Nature Mater. 8, 194–197 (2009).
2. Gambardella, P. et al. Nature Mater. 8, 189–193 (2009).
3. Bernien, M. et al. Phys. Rev. Lett. 102, 047202 (2009).

Shaping fate
The promise of stem-cell therapies
for tissue regeneration hinges on the
fact that, to a first approximation,
stem cells can become any other
cell type. But therein lies one of
the biggest challenges — for how
does a stem cell decide its fate? This
decision is generally made in the
body through complex biochemical
pathways involving diffusing signalling
molecules. One approach to stem-cell
therapeutics is to manipulate these
routes using either natural signalling
factors or synthetic small molecules
that serve the same role.
But there can be another, perhaps
more surprising determinant of
stem-cell fate. It may be influenced
by purely mechanical means such as
stretching or stressing cells, for example
by altering the stiffness of the matrix
in which they grow. Mesenchymal
stem cells (MSCs), which are potential
progenitors of many cell types
including bone-growing osteoblasts,
muscle-making myoblasts and
tissue-making fibroblasts, will guide
differentiation towards myoblasts in a
soft matrix that resembles brain tissue,
but towards osteoblasts in a hard,
bone-mimicking matrix 1. This suggests
a role for materials engineering in
stem-cell therapy.
It’s been long known that cells can
sense and respond to deformation,
for example via switchable ion
channels in their outer membranes.
But more surprising is the fact that
they seem responsive to texture and
order. MSCs grown on nanopatterned
polymer surfaces have been found to
become more osteoblast-like when the
surfaces are embossed with random
arrays of nanopits, compared with
regular, ordered arrays2. That raises
the prospect of using nanopatterned
matrices to define the distributions
of cell types in new tissue seeded
from stem cells, for example in
bone regeneration.
How does this work? Shu Chien
and co-workers at the University of
California at San Diego now think
they have some clues3. They have
found a new guiding factor for MSC
differentiation that seems to be purely
geometric. They grew human MSCs
on substrates of aligned arrays of
titanium dioxide nanotubes with
varying diameter, from 30 to 100 nm,
made electrochemically from thin
films of titanium. The behaviour of
the cells was strongly dependent on
the nanotube size: for 30-nm tubes,
they adhered well but didn’t really
differentiate at all, whereas for 100-nm
tubes they became long, thin and
osteoblast-like.
Elongation is the key. Chien
and colleagues saw that the smaller
nanotubes became quickly decorated
around their open ends with blobs of
PHILIP BALL
protein: an extracellular matrix
deposited by the cells, through which
they can adhere to the surface. But
these blobs were far less abundant on
the wider tubes, simply because there
is less space to put them. As a result,
cells seeking to anchor themselves have
to stretch further in the latter case,
and the researchers think that this
deformation triggers differentiation to
a bone-forming lineage.
That not only suggests a way to
guide bone growth by controlling
nanostructure; because titanium
nanotubes are themselves good
candidates for a biocompatible bone-
fostering implant material, they can
do two jobs at once, providing both
support and guidance. ❐
references
1. Engler, A. J. et al. Cell 126, 677–689 (2006).
2. Dalby, M. J. et al. Nature Mater. 6, 997–1003 (2007).
3. Oh, S. et al. Proc. Natl Acad. Sci. USA
doi:10.1073/pnas.0813200106 (2009).

166 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

© 2009 Macmillan Publishers Limited. All rights reserved


nanocatalysis
staying put
Preparation of supported subnanometre platinum clusters that are stable provides a new design strategy for
industrial nanocatalysts.
gianfranco Pacchioni
i
t has been suggested that catalysis
by supported particles is one of the
first examples of nanotechnology:
heterogeneous catalysts are often made of
small particles of precious metals, typically
a few nanometres in size, deposited on an
‘inert’ oxide support (silica, alumina) or
other porous materials such as zeolites.
However, methods used to fabricate
industrial catalysts produce metal particles
that cover a wide range of sizes, with a
classical statistical distribution. In technical
terms, the particles are not monodispersed,
which opens the question: do they all have
the same chemical activity? To answer this,
methods to construct collections of well-
defined, nearly monodispersed ensembles of
nanoparticles are essential1–3.
On page 213 of this issue, Vajda and
colleagues report on the preparation of size-
selected Pt8–10 clusters stabilized on a high-
surface-area aluminium oxide film prepared
by atomic layer deposition (Fig. 1)4. The
catalytic activity of the clusters for the
oxidative dehydrogenation of propane is
40–100 times higher than that of more
conventional platinum catalysts.
Studies performed on gas-phase mass-
selected metal clusters reveal that below
a given size, typically a few tens of atoms,
one enters a non-scalable regime where the
properties of the particle change markedly
with size5. For instance, properties may
vary substantially on going from an even
to an odd number of atoms in the cluster.
The study of such phenomena is even more
complex when one deals with supported
clusters, thus approaching materials of
a b
Al2O3 Pt
Figure 1 | The catalytic system. a, Schematic
view of a Pt8–10 cluster deposited on a thin
alumina film covering a porous anodized
aluminium oxide support. b, Structure of a Pt8
cluster on the support.
nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
Cluster source Collimator
Metal rod
Skimmer
He
Continuous
cluster beam
Vaporization
laser Ion
focusing lens
4.5 kHz
Figure 2 | The apparatus used to generate, mass-select and soft-land platinum clusters on an oxide
support to perform catalytic tests on well-defined nanoclusters.
practical interest. It was only towards the
end of the previous century that Ueli Heiz
and co-workers were able to produce and
gently deposit mass-selected metal clusters
on a thin magnesium oxide film grown on a
metal support 6,7. These experiments opened
the way to the preparation of monodispersed
ensembles of nanoparticles8,9. These have
surprising properties; some cluster sizes
are chemically very active, whereas others
are inert. The part that surface defects in
the supporting oxide play in stabilizing
the clusters and in modifying their charge
state was also discovered: a Au8 cluster
grown on an oxygen vacancy in a MgO
film converts CO to CO2 in the presence of
oxygen at low temperature, unlike the same
cluster deposited on a defect-free MgO
film6. Despite these considerable advances,
the possibility of using tiny subnanometre
particles in catalytic processes of industrial
interest has been hampered by the inability
to stabilize the clusters on a substrate and
prevent them from sintering — that is,
forming large particles at the expense of the
smaller ones.
Vajda and colleagues have been able
to overcome this problem. They have
produced gas-phase neutral and positively
charged platinum clusters using a laser
vaporization source; the positively charged
clusters are guided into a quadrupole mass
spectrometer, and selected according to
mass (Fig. 2).
© 2009 Macmillan Publishers Limited. All rights reserved
news & views
Biased substrate holder
Quadrupole
mass selector
Focusing Focusing Focusing
lens lenses lenses
Focusing Detector
lenses Quadrupole
deflector
Clusters of single size /
narrow size distribution
The kinetic energy of the clusters landing
on the support is controlled by biasing
the substrate. In this way low deposition
energies are obtained, thus avoiding
fragmentation of the clusters on their impact
with the surface, an approach known as
soft landing. The kinetic energy per atom
is similar to that used in the deposition
of Ag19 clusters on Pt(111) in an earlier
experiment by Schaub and colleagues who
showed, using atomic-resolution scanning
tunnelling microscope images, that in
these conditions the clusters remain intact
after deposition10. This is vital to guarantee
that one has the same cluster sizes on the
surface as were produced in the gas phase.
Further characterization of the samples is
provided by synchrotron X-ray scattering
measurements that show the stability and
the three-dimensional shape of the platinum
clusters. Critically, because of the porosity of
the support, the clusters remain stable over a
temperature range of 20–400 °C, an essential
aspect in performing catalytic reactions on
these well-defined nanocatalysts.
Despite the great advances in the
characterization of supported nanoclusters,
there are still aspects that are difficult to
evaluate experimentally. For instance, it is
not easy to determine the cluster charge.
This is often an important parameter
because charged gas-phase clusters can be
highly reactive11. To overcome this problem,
the group performed density functional
167
news & views

theory calculations, and these show that
the Pt8 cluster on alumina withdraws
electrons from the surface and becomes
negatively charged.
Once the nanocatalysts had been
characterized, catalysis tests were carried
out under atmospheric pressure, and the
turnover frequencies were measured at a
temperature around 500 °C. The surprising
result is that the activity of the Pt8–10 catalyst
is far higher than that of any other reported
platinum-based catalyst for oxidative
dehydrogenation of propane. Clearly,
the special activity is associated with the
tiny dimension of the catalyst particles.
To rationalize these findings Vajda and
colleagues performed density functional
theory calculations on the reaction
mechanism using a Pt8 cluster on an Al2O3
model. The calculations show the existence
of a relatively small energy barrier for the
breaking of the first C–H bond, which is
likely to be the rate-limiting step of the
reaction because the rest of the pathway to
propylene formation is thermodynamically
downhill. Furthermore, there are no
important differences between the
computed reaction profiles on a gas-phase
or supported Pt8 nanocatalyst, providing
evidence that the support has little effect
on the reaction barrier in this case. The
conclusion is that the undercoordinated
sites in small platinum clusters are much
more active than a platinum surface for
propane dehydrogenation.
The work of Vajda and collaborators
represents an important example of catalytic
activity of mass-selected metal clusters
under realistic conditions of temperature
and pressure. The fact that the clusters
are both active and stable is a sizeable
step towards the design of nanocatalysts
of practical use. Although challenging,
the development of nanometre and
subnanometre clusters stabilized on specific
supports could result in new catalysts for a
variety of industrially relevant processes. ❐
Gianfranco Pacchioni is in the Dipartimento di Scienza
dei Materiali, Università degli Studi Milano-Bicocca,
via R. Cozzi 53, 20125 Milano, Italy.
e-mail: gianfranco.pacchioni@unimib.it
References
1. Bell, A. T. Science 299, 1688–1691 (2003).
2. Heiz, U. & Landman, U. (eds) Nanocatalysis (Springer, 2006).
3. Freund, H. J. & Pacchioni, G. Chem. Soc. Rev.
37, 2224–2242 (2008).
4. Vajda, S. et al. Nature Mater. 8, 213–216 (2009).
5. Socaciu, L. D. et al. J. Chem. Phys. 120, 2078–2081 (2004).
6. Sanchez, A. et al. J. Phys. Chem. A 103, 9573–9578 (1999).
7. Abbet, S. et al. J. Am. Chem. Soc. 122, 3453–3457 (2001).
8. Benz, L. et al. J. Chem. Phys. 122, 081102 (2005).
9. Fan, C. Y., Wu, T. P., Kaden, W. E. & Anderson, S. L.
Surf. Sci. 600, 461–467 (2006).
10. Schaub, R. et al. Phys. Rev. Lett. 86, 3590–3593 (2001).
11. Fallace, W. T. & Whetten, R. L. J. Am. Chem. Soc.
124, 7499–7505 (2002).

Multiferroics

A way forward along domain walls

The discovery that domain walls in insulating thin films of the multiferroic compound BiFeO3 are electrically
conducting opens the door for a number of possible applications.

Hélène Béa and Patrycja Paruch

C
orrelated oxides are an
expanding field of study, with a
richness of fundamental physics
and phenomenal possibilities for
multifunctional applications1. Despite
[001]
their relatively simple structure, these
systems present a great complexity
of different properties, including [100]
superconducting as well as ferroic orders [010]

such as magnetism or ferroelectricity,

sometimes within the same material2.
Multiferroic materials that display
ferroelectricity and magnetism are good Ph
candidates for spintronic applications
relying on magnetoelectric coupling 3. Pv
Interfaces in such correlated oxides,
intrinsically nanoscopic because of
their small thickness, are of particular Potential
interest because specific strain or polar
Bandgap
boundary conditions can lead to additional
functionality absent from the already
multifunctional parent material4–6.
Domain walls in ferroic compounds are
one such type of interface, separating Figure 1 | Schematic representation of a 109° domain wall in BiFeO3. The orientation of polarization
regions with different orientations of the orientation across the domain wall is depicted within the material. Shown below are the horizontal (Ph)
magnetic, ferroelectric or ferroelastic order and vertical (Pv) components of the polarization projected in the (010) plane, along with the resulting
characterizing the material7,8. potential step, and the calculated bandgap.

168 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

© 2009 Macmillan Publishers Limited. All rights reserved.


Many studies of ferroic domain walls
have focused on their behaviour in terms of
the basic properties of the parent compound,
and on their increasingly important (and
often negative) effect on the performance of
ever-smaller electronic devices9,10. However,
identifying their additional functionalities
and incorporating them in the design of
applications could in fact allow domain
walls to be used as active, nanoscale device
components. The report by Jan Seidel and
his colleagues on page 229 of this issue11
is an important advance in this direction,
demonstrating electrical conductivity
across two kinds of ferroelectric domain
wall in the insulating room-temperature
multiferroic BiFeO3.
In BiFeO3 the ferroelectric polarization
is oriented along the diagonals of the
monoclinically distorted tetragonal unit
cell, giving rise to eight different possible
polarization states and three types of domain
wall separating regions with polarization
orientations differing by 180°, 109° and
71° (ref. 12). Seidel et al. report that out
of these three, the first two show a clear
signature of electric conductance in local
probe measurements. Using high-resolution
transmission electron microscopy and
density functional theory calculations, the
authors correlate this unusual conducting
behaviour in an otherwise insulating
material with a local structural distortion.
This distortion reduces the electronic
bandgap in the region of the domain
wall, but also increases the polarization
component normal to the domain wall,
resulting in an electrostatic potential step
(Fig. 1). Either effect may explain the
increase in conductivity. The first decreases
the band edge relative to the local probe
tip used in the conductivity measurements,
whereas the second leads to an accumulation
of free charge carriers at the domain wall
in order to provide screening for the
polar discontinuity caused by the increase
in polarization.
These observations open intriguing
possibilities for potential device
applications. The authors mention that
their theoretical analysis suggests very high
sensitivity of the conductive response to
unit cell size, which can easily be changed
by applying strain to a thin film. Further
detailed investigations looking at the
conductivity of domain walls in BiFeO3
films on different substrates, with different
lattice mismatch and hence strain, as well
as in relaxed versus strained samples would
therefore be of interest.
A possible application could be as a
local strain sensor incorporated on devices
such as an atomic force microscopy (AFM)
probe, as shown in Fig. 2a. As the electrical
nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 169
news & views
a b c
R
BiFeO3
BiFeO3
Magnet
DW Current
Figure 2 | Potential applications for conductive domain walls (DW). a, Variations in the conductivity
of domain walls (purple) as a function of strain could be envisaged for sensing applications (R is
resistance). b, Alternatively, devices could be created in which the conductive domain walls are uniformly
moved along the film (shown in top view), thus changing the conducting state between two electrodes
(yellow). These could be useful as memory devices or as switches. c, One possibility for controlling the
domain wall motion would be to use an underlying ferromagnet in which domain walls are moved by a
spin-polarized current.
resistance measurement is sensitive to the the domain walls of BiFeO3 would follow
strain state, the bending of the cantilever this movement, making it possible to create
to which the probe tip is attached may be a multilevel resistance-state device that is
quantified. Such device geometry could written by an electrical current.
potentially avoid the precision alignment Obviously, such functional application of
and heating that accompanies the use these conductive ferroelectric domain walls
of a standard AFM probe in laser beam requires that practical obstacles be overcome,
deflection mode. for example the motion of domain walls in
An important prerequisite for using BiFeO3 through an underlying ferromagnet.
domain walls for such applications is the Indeed, only the reverse effect — that is,
control of their nucleation and position. the manipulation of a ferromagnet domain
In their prototype device, Seidel et al. use configuration by an electric field applied on
an AFM tip to write domain walls in the BiFeO3 — has so far been demonstrated15.
ferroelectric film. In future devices, such Nonetheless, Seidel and colleagues have
writing could be done using a multi-tip taken an important step, identifying and
probe on an array of device elements. understanding a new property of ferroic
However, a faster approach, requiring domain walls that may eventually lead
no mechanical motion, would be to use to their incorporation into ever-smaller
previously defined domain wall patterns, nanoscale technological devices. ❐
moved uniformly, as in a racetrack
memory 13, so that the desired domain wall Hélène Béa and Patrycja Paruch are in the
pattern defining a fixed conductive state Département de la Matière Condensée, Université
lies between the measurement electrodes de Genève, 1211 Geneva 4, Switzerland.
(Fig. 2b). Although ferroelectric wall e-mail: Helene.Bea@unige.ch,
motion could be induced by the application Patrycja.Paruch@unige.ch
of an electric field, care has to be taken
as such fields could lead to a change of References
domain size, and thus non-uniform motion 1. Dagotto, E. Science 318, 1076–1077 (2007).
of the domain wall patterns, and their 2. Bibes, M. & Barthelemy, A. IEEE Trans. Electron. Devices
54, 1003–1023 (2007).
possible erasure.
3. Spaldin, N. A. & Fiebig, M. Science 309, 391–392 (2005).
Furthermore, as BiFeO3 is a multiferroic 4. Ohtomo, A. & Hwang, H. Y. Nature 427, 423–426 (2004).
material, in which ferroelectric and 5. Reyren, N. et al. Science 317, 1196–1199 (2007).
antiferromagnetic domains are strongly 6. Bousquet, E. et al. Nature 452, 732–737 (2008).
magnetoelectrically coupled, one could 7. Zubko, P. et al. Phys. Rev. Lett. 99, 167601 (2007).
8. Přívratská, J. & Janovec, V. Ferroelectrics 222, 23–32 (1999).
envisage a device geometry based on a 9. Paruch, P., Tybell, T., Giamarchi, T. & Triscone, J.‑M.
BiFeO3 layer adjacent to a ferromagnetic J. Appl. Phys. 100, 051608 (2006).
film (Fig. 2c). Spin-polarized currents 10. Scott, J. F. J. Phys.: Condens. Matter 18, R361–R386 (2006).
in the ferromagnetic layer could then be 11. Seidel, J. et al. Nature Mater. 8, 229–234 (2009).
12. Lebeugle, D. et al. Phys. Rev. B 76, 024116 (2007).
used to drive its magnetic domain walls by
13. Parkin, S. S. P. et al. Science 320, 190–194 (2008).
means of the spin transfer torque effect 13,14. 14. Ralph, D. C. & Stiles, M. D. J. Magn. Magn. Mater.
The exchange coupling between the 320, 1190–1216 (2008).
ferromagnetic layer and BiFeO3 means that 15. Zhao, T. et al. Nature Mater. 5, 823–829 (2006).
© 2009 Macmillan Publishers Limited. All rights reserved.
erratum

A way forward along domain walls

Hélène Béa and Patrycja Paruch

Nature Materials 8, 168–169 (2009); published online: 20 February 2009; corrected after print: 20 February 2009.
In the print version of this News & Views article the reference list should have included the following:
3. Spaldin, N. A. & Fiebig, M. Science 309, 391–392 (2005).
This reference is included in the HTML and PDF versions.

© 2009 Macmillan Publishers Limited. All rights reserved.

news & views

transition Metals

can metals be a liquid glass?

The melting of transition metals on compression is a challenging topic. Computer simulations suggest that
hot-compressed tantalum becomes a one-dimensional, liquid-like glass, with important implications for
understanding planetary interiors.

Daniel errandonea

i
n materials science and geophysics, it is
essential to understand the transition
between solid and liquid phases of
elements at high pressures. An accurate
picture of materials performance under
extreme conditions and detailed models
of planetary interiors both depend on
this knowledge. On page 223 of this issue,
Christine Wu and colleagues1 address the
long-standing issue of tantalum’s melting
curve and show how, before melting, the
studied materials might become a one-
dimensional, liquid-like glass.
Interest in the melting at high pressure
of transition metals such as iron and
tantalum stems from their importance to
understanding the inner properties of the
Earth and other planets, because iron is the
most abundant element in their cores. Thus,
a 298 K bcc b 3,250 K bcc
{111} plane
–
{112} plane
–
{110} plane
3 3,250 K
g(r)
transition metals are among the elements
most extensively studied at high pressures.
Laboratory experiments conducted using
diamond-anvil cells (DACs) — devices that
consist of two opposing diamonds with a
sample compressed between their culets2 —
have covered pressures up to 200 GPa
and temperatures up to 5,000 K, thereby
mimicking Earth’s outer core3–5. Higher
pressures and temperatures have also
been explored by taking advantage of the
propagation of shock-wave fronts through
the studied metal and measurement of the
sound speed6. In theoretical calculations,
transition metals have been studied up to
400 GPa and 10,000 K (refs 7, 8). Despite
such dramatic advances, knowledge about
transition metals at extreme pressure
(P) and temperature (T) is still far from
c 1D glass d Glass
3,250 K 5,000 K
complete, and controversies exist regarding
the interpretation of different results.
In their study, Wu and colleagues
now analyse the role that shear stresses
have on tantalum at extreme conditions
before melting. Under shear stress, the
shape of a material tends to change
without volume change. In particular,
their molecular dynamics simulations —
where atoms are allowed to interact for a
period of time by known physics, giving
a view of the atomic motion — suggest
that on plastic deformation beyond
3,250 K, solid tantalum becomes a one-
dimensional, viscous glass (1D glass),
in analogy to liquid glasses before they
transform into an ideal liquid. Apparently,
the proposed picture is consistent with
DAC experiments4 and might provide an
explanation of the as yet unexplained low
melting slopes of transition metals.
To study the structural behaviour of
tantalum, Wu et al. applied a technique
widely used to identify amorphization
by monitoring changes in the radial
distribution function — a function that
describes how the density of surrounding
matter varies with distance from a
particular point. Their results (Fig. 1)
illustrate how solid tantalum, with a
body-centred cubic (b.c.c.) structure,
becomes a 1D glass at high pressures and
temperatures. The main characteristic
of this glass is the merging of peaks in
the radial distribution function, which
indicates that the third and fourth nearest-
neighbour distances in the {110} planes
become indistinguishable. Therefore,
tantalum is proposed to become a partial-
disordered structure with only long-range
order along the [111] direction, resembling
a smectic liquid (made of well-defined
layers that can slide along one another like

2
soap). The calculations by Wu et al. also
1
suggest that on further heating, the 1D glass
0 transforms into an ideal liquid with only
2 4 6 2 4 6 2 4 6 2 4 6
r (Å) r (Å) r (Å) r (Å) short-range order. Basically, the molecular
dynamics simulations indicate that the
Figure 1 | Computer simulations comparing the distribution of atoms in different phases of tantalum at b.c.c. lattice of tantalum breaks down
30 GPa. Planar projections of tantalum atomic coordinates are shown as well as the corresponding radial before the actual melting. Shear strain
distribution functions g(r). a, Un-sheared b.c.c. crystal. b, B.c.c. crystal heated to 3,250 K without shear. provides the mechanical driving force for
c, 1D glass induced by shear and heating. d, 3D glass. the solid–1D-glass transition. This required

170 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

© 2009 Macmillan Publishers Limited. All rights reserved


stress reflects the existence of an energy
barrier needed to overcome the resistance
of materials to plastic deformation, which
is material-specific and depends strongly
on pressure. This conclusion is supported
by the fact that in soft metals with isotropic
bonding, such as aluminium and copper,
the experimental melting temperatures
agree with those calculated for the solid–
ideal-liquid transition.
The proposed behaviour for tantalum
will not resolve all of the questions about
its behaviour at extreme P–T conditions,
but could imply significant progress. In
particular, it does not necessarily explain
why, in experiments performed under
different stress conditions (with the sample
immersed under soft or stiff pressure
media), the same melting temperature is
detected in DAC experiments4. However,
it clearly provides a new framework for
studying the P–T region constrained
by 200 GPa and 5,000 K. The work of
Wu et al. will, without doubt, stimulate
further laboratory experiments and
computer calculations. In particular, new
shock-wave measurements of the sound
speed between 100 and 200 GPa might be
ePitaxial graPHene
How silicon leaves the scene
Large and homogeneous layers of graphene are obtained by annealing silicon carbide in a dense noble gas
atmosphere that controls the way in which silicon sublimates. Epitaxial graphene thus gets back on track towards
future electronic applications.
Peter sutter
s
ince it was first isolated in 2004,
graphene, a sheet of pure carbon
just one atom thick, has generated
a flurry of research activities. Although
much of the initial ‘gold rush’ has focused
on the fascinating properties of this two-
dimensional crystal1 — which have as
much to do with fundamental quantum
electrodynamics and particle physics as
with solid state physics and materials
science — researchers have recently begun
addressing the more mundane question of
how some of these characteristics might
be harnessed in applications ranging
from post-Moore’s law electronics over
ultra-responsive sensors and actuators to
transparent solar cell contacts. Assuming
a continuation of ‘top-down’ processing
similar to today’s silicon microelectronics,
the bottleneck clearly lies in synthesizing
the required starting material: structurally
nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
beneficial to understanding the structural
properties of tantalum beyond 3,250 K.
Furthermore, recording of second-scale
time-resolved X-ray diffraction patterns
during temperature cycling would
provide valuable information9. These
experiments will not only be appropriate
to test Wu and colleagues’ findings, but
also to check other hypotheses on the
high-pressure behaviour of transition
metals. For example, the creation of local
structures in the liquid, owing to Peierls
and Jahn–Teller distortions10, are expected
to lower the melting point of transition
metals on compression. In particular, X-ray
diffraction experiments can give valuable
information on the radial distribution
function of tantalum and similar metals.
These experiments will be fundamental for
distinguishing between the plastic flow of a
liquid-like 1D glass and the free motion of
a ‘true’ or a ‘clustered’ liquid.
The most important implication of the
proposed shear-induced melting is that
similar 1D-glass formation features are
also found in other transition metals such
as molybdenum and vanadium, implying
that this melting mechanism is expected to
perfect, macroscopically large graphene
sheets with uniform thickness, into which
active device structures can be carved.
On page 203 of this issue, Emtsev et al.
demonstrate an important step towards
this goal for a specific fabrication strategy:
graphene epitaxy on silicon carbide (SiC)2.
Epitaxial growth on SiC (Fig. 1), a
highly resistive material which would not
shunt the current flow in graphene and
which is already available in the form of
large-diameter wafers, is often seen as one
of the most likely avenues to graphene-
based electronics. The graphitization
of SiC by Si sublimation during high-
temperature vacuum annealing was
demonstrated as early as the 1960s3,
and has recently been refined4. Even so,
considerable challenges have remained.
Although graphene grows with a well-
defined orientation and its lattice fits well
© 2009 Macmillan Publishers Limited. All rights reserved
news & views
hold generally for other transition metals,
probably even for iron. Such conclusions
therefore could have important geophysical
implications. It is possible that a new era
in the study of high-pressure melting of
metals has begun. An explanation for
different experimental observations and
theoretical predictions may finally be
within our grasp. ❐
Daniel Errandonea is with the MALTA Consolider
Team, Fundación General Universidad de Valencia,
46100 Burjassot, Spain.
e-mail: daniel.errandonea@uv.es
references
1. Wu, C. J., Söderlind, P., Glosli, J. N. & Kepleis, J. E. Nature Mater.
8, 223–228 (2009).
2. Walker, J. Nature 265, 498–499 (1977).
3. Boehler, R. Nature 363, 534–536 (1993).
4. Errandonea, D. et al. Phys. Rev. B 63, 132104 (2001).
5. Errandonea, D. J. Phys. Chem. Solids 67, 2017–2026 (2006).
6. Hixson, R. S., Boness, D. A., Shaner, J. W. & Moriarty, J. A.
Phys. Rev. Lett. 62, 637–640 (1989).
7. Belonoshko, A. B. et al. Phys. Rev. Lett. 100, 135701 (2008).
8. Taioli, S., Cazorla, C., Gillan, M. J. & Alfe, D. Phys. Rev. B
75, 214103 (2007).
9. Dewaele, A., Mezouar, M., Guignot, N. & Loubeyre, P.
Phys. Rev. B 76, 144106 (2007).
10. Ross, M., Errandonea, D. & Boehler, R. Phys. Rev. B
76, 184118 (2007).
on a reconstructed interfacial layer on
Si-face SiC, graphitization in vacuum has
so far invariably transformed the ordered
substrate surface (Fig. 1a) into a rough,
highly corrugated landscape covered by
small graphene domains with pronounced
thickness variations (Fig. 1c). Such disorder
not only hinders further processing, it can
also limit device performance by scattering
charge carriers flowing in the active
graphene layer.
The solution is quite obvious, at least
in principle. Roughness, excess surface
steps as well as small-grained graphene
are generated if a system cannot attain its
most favourable morphology because the
necessary microscopic processes — the
detachment, diffusion and reattachment of
surface atoms — are suppressed. Increasing
the growth temperature should lift such
constraints. However, for graphitization in
171
news & views
vacuum, higher annealing temperatures
also drive faster sublimation and higher
graphene growth rates, which in turn
promote roughening and the nucleation of
small graphene flakes. Progress can only
be made if mass transport on the surface is
effectively decoupled from Si evaporation
away from the surface. Emtsev et al.
accomplished this feat by resurrecting
an 80-year-old trick originally developed
to extend the lifespan of incandescent
lightbulb filaments5. Instead of processing
their samples in vacuum, they placed
them in a furnace with an unreactive
argon atmosphere at nearly ambient
pressure. The dense cloud of gas molecules
hinders the transport of Si atoms away
from the SiC surface, reducing the overall
sublimation rate and allowing an increase
in graphitization temperature by several
hundred degrees. Samples produced by
this process have smooth surfaces made up
of large, flat substrate terraces that reach
tens of micrometres in length and several
micrometres in width, and are separated by
steps several times the height of the original
a b
<1 µm
c
d G3 G2
G1
~3 µm
Figure 1 | Evolution of the surface during graphitization of Si-face SiC. a, SiC starting surface with a
staircase of flat terraces and atomic steps. b, Unit cell of 6H-SiC. c, Schematic morphology of vacuum-
graphitized SiC. d, Morphology obtained in high-pressure argon. The surface termination is predominantly
monolayer graphene, G1 (e), with narrow stripes of bilayer (G2) and trilayer (G3) graphene near the upper
edge of the substrate steps.
172
substrate steps (Fig. 1d). The carbon
layer grown on this surface now consists
primarily of monolayer graphene, with
narrow bilayer (and occasional trilayer)
stripes at the terrace edges.
The key question, of course, is whether
the improved sample morphology
translates to a reduced scattering of
charge carriers in the epitaxial graphene.
Emtsev et al. begin to address this question
by comparing the electron mobility — a
measure of the charge carrier drift velocity
in an applied electric field — for samples
prepared in vacuum and in ambient-
pressure argon, patterned into simple test
structures. Their results are encouraging.
Samples produced by the new process show
a nearly twofold improvement over the
previous record mobility in Si-face epitaxial
graphene. But the findings also raise a
number of issues. Particularly puzzling is
the enormous difference — about a factor
of 100 — in electron mobility between
epitaxial graphene on SiC and exfoliated
graphene. The fact that the highest carrier
mobilities were achieved in suspended
Si-face
SiC
C
Si
SiC
C-face
e
SiC
nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
© 2009 Macmillan Publishers Limited. All rights reserved
sheets6, whereas both epitaxial graphene
and supported exfoliated graphene
show far lower values, raises questions
about fundamental limitations to carrier
transport due to long-range scatterers in a
substrate or support. The nature of these
scattering centres is generally unknown,
but the work by Emtsev et al. sets the
stage for their identification for epitaxial
graphene on SiC by reducing roughness
and increasing the graphene domain size.
A careful comparison of transport parallel
and perpendicular to substrate steps, for
instance, should reveal whether the steps
themselves, or the two- and three-layer
domains near the steps, cause electron
scattering. The complex buffer layer
sandwiched between the graphene and the
SiC substrate needs to be considered as
another possible scatterer that would be
difficult to eliminate.
For some time now, epitaxial graphene
on SiC has been considered one of
the leading contenders in the race to
find ways to mass-produce graphene,
particularly for electronics. The improved
properties demonstrated by Emtsev et al.
should provide an additional boost
for this technology. But the race is still
far from over. Competing approaches
have also seen rapid progress recently.
Transition-metal-catalysed graphene
growth, in particular, has been shown
to yield macroscopic single-crystalline
graphene domains with very low defect
density and outstanding thickness control7.
Still needed are methods to isolate the
carbon layers from a metal substrate. A
recent successful demonstration of the
transfer of large graphene flakes, using
growth on thin polycrystalline nickel films
that can be etched away to free the carbon
layers, underscores the viability of this
approach8. A true leap forward, however,
would be the development of an atomically
precise ‘bottom-up’ synthesis of graphene
nanostructures. If the tools of materials
science, chemistry and nanotechnology
are combined, such a vision might just
become a reality. ❐
Peter Sutter is at the Center for Functional
Nanomaterials, Brookhaven National Laboratory,
Upton, New York 11973, USA.
e-mail: psutter@bnl.gov
references
1. Geim, A. K. & Novoselov, K. S. Nature Mater. 6, 183–191 (2007).
2. Emtsev, K. V. et al. Nature Mater. 8, 203–207 (2009).
3. Badami, D. V. Nature 193, 570–571 (1962).
4. Berger, C. et al. J. Phys. Chem. B 108, 19912–19916 (2004).
5. Fonda, G. R. Phys. Rev. 31, 260–266 (1928).
6. Bolotin, K. I. et al. Phys. Rev. Lett. 101, 096802 (2008).
7. Sutter, P. W., Flege, J.-I. & Sutter, E. A. Nature Mater.
7, 406–411 (2008).
8. Reina, A. et al. Nano Lett. 9, 30–35 (2009).

MeMBranes
shaping biological matter
Biological membranes form an extremely complex and dynamic network in cells, guided by specialized
protein machinery. A new algorithm analyses membrane shape to extract forces applied by proteins controlling
the membranes.
Vadim a. frolov and Joshua Zimmerberg
t
he space occupied by living
organisms is determined by
membranes. Each cell is delimited
by a membrane, and additional
membranes form the functional hierarchy
of compartments packed within it. The
intracellular compartments, for example
vesicles and organelles, define the spatial
organization of metabolic pathways, thus
providing the topological framework
for cellular life. Recent advances in
structural techniques and super-resolution
microscopy now allow imaging of
intracellular membrane structures with
unprecedented resolution. Paul Wiggins
and co-workers1 are, perhaps, first to
notice that the wealth of membrane
structures obtained awaits detailed
elastic modelling.
The core of each membrane is the lipid
bilayer, the bimolecular film consisting of
two monolayers of amphiphilic molecules,
the lipids. This bilayer is primarily
responsible for the barrier function of
membranes. However, the lipids are also
deeply involved in regulation of membrane
shape and dynamics: both depend on the
material properties of the lipid bilayer 2,3.
One property is the elasticity, the resistance
to stretching and compression that
accompanies most of the deformations of
lipid bilayers imposed by the specialized
proteins that determine the shape of
cellular membranes. The lipid bilayer can
be approximated as a continuous elastic
surface whose behaviour is characterized by
a limited set of bulk material parameters4.
Despite its relative simplicity, the elastic
approximation has been successfully used
to explain a wide variety of membrane
phenomena, from the multitude of shapes
of red blood cells5 to the structure of
metastable intermediates of membrane
fission and fusion6.
How can continuum elastic modelling
help us to understand the action of
specialized proteins creating cellular shape
(and thus how the shape is encoded)
at small, almost molecular scales?
Proteins can be simply considered as a
new bulk component effectively altering
nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
a system in which a thin membrane cylinder
F F
b application (Fig. 1b). Then the force
Boundary
Boundary
F F F that is, by making the virtual work
F required to displace the vertices equal
Figure 1 | Calculation of forces acting on a
membrane boundary from the membrane
shape. a, Illustration of a spherical membrane
vesicle with a cylindrical tether pulled from it
by application of an external force (F). b, The
membrane shape obtained from images is
parameterized by control vertices (red dots); an
effective boundary on which the external forces
act is introduced (blue); small perturbations of
the vertices from their stationary positions (red
arrows) yield local internal forces due to elastic
response of the membrane. The total internal
force (F, blue) projected to the tether axis is
equal to the external force (F, red) acting on
the boundary.
averaged elastic parameters of the lipid
bilayer. Different algorithms linking
protein embedment into a membrane to
elastic deformations have recently been
developed2,7,8, yielding phenomenological
descriptions of membrane morphogenesis
in cells.
The Wiggins group take a different
approach. They assume that proteins act
externally, by applying forces required
to support a membrane shape from
its boundary. The force distribution
can be calculated by analysing the
membrane shape within this boundary,
by the established algorithms of shape
parameterization and localized force
balancing 5,7,9. The researchers tested
this approach on a simple experimental
© 2009 Macmillan Publishers Limited. All rights reserved
news & views
(tether) was pulled from a giant vesicle
(Fig. 1a). The pulling force, controlled
experimentally, was compared with those
obtained by membrane shape analysis. For
that, the membrane was parameterized
by an arbitrary set of control vertices,
excluding the areas of the external force
balance at each vertex was obtained using
a proximal equilibrium approximation6,
to the corresponding energy of elastic
deformations. The minimization was
conducted by conventional procedures5,7,
taking into account elastic deformations
and the work of hydrostatic pressure
and lateral tension. Finally, they
obtained the external force acting at
the membrane boundary to cancel the
internal ones exactly (Fig. 1b). This force
correctly estimated the pulling force, as
expected given the decades of studies
on membrane shapes that have used the
giant vesicle prototype (see, for example,
refs 9, 10). However, Wiggins and
colleagues emphasize that the proximal
equilibrium approximation1 is tractable
for shapes of greater complexity because,
as is common for algorithms based on
local balancing of forces and moments5,7, it
does not explicitly rely on assumptions of
shape symmetries generally used to analyse
simple membrane vesicles.
This notion is yet to be tested.
Conceivably, it is even more important
that the work of Wiggins’s group
reiterates the importance of membrane
boundary in supporting a bulk elastic
stress11. This conceptualization could be
incisive when studying how particular
protein complexes control the shape of
cellular membranes.
Such proteins often have to restrain
large membrane areas and/or create a
substantial membrane stress, examples
being the protein rings supporting the
large structure of the tubular endoplasmic
reticulum. This membrane system makes
up the bulk of the internal membranes
173
news & views

a b proven successful in model systems, their

algorithm may provide unique information
about the forces applied by proteins in
their physiological environment, where
direct mechanical measurements of forces
are not possible. Such measurements can
be particularly useful in the case of cell
and tissue damage associated with the
alterations of lipid metabolism. Perhaps
slight changes in the bulk material
properties of the membrane will be
revealed by organelles that are distorted
from their equilibrium shapes. These could
d then be related to specific lipid species to
c reveal the chemical basis of the changes in
T T membrane material properties. ❐

F Vadim A. Frolov and Joshua Zimmerberg are

F F
Fission part of the Program on Physical Biology, Eunice
Kennedy Shriver National Institute of Child Health
and Human Development, Bethesda, Maryland
20892-1855, USA.
e-mail: zimmerbj@mail.nih.gov

Figure 2 | Hypothetical algorithm of calculation of forces shaping an organelle. a, Cartoon illustrating references
the endoplasmic reticulum shape controlled by rings of specialized proteins (blue). b, The shape is 1. Lee, H. J., Peterson, E. L., Phillips, R., Klug, W. S. &
parameterized by a set of vertices labelling ‘free’ membrane pieces. c, From the shape analysis, the forces Wiggins, P. A. Proc. Natl Acad. Sci. USA
(F) and torques (T) exerted by the rings are calculated1,9. d, Protein rings can also enforce membrane 105, 19253–19257 (2008).
narrowing, leading to shape instability and, ultimately, membrane fission13. 2. Zimmerberg, J. & Kozlov, M. M. Nature Rev. Mol. Cell. Biol.
7, 9–19 (2006).
3. McMahon, H. T. & Gallop, J. L. Nature 438, 590–596 (2005).
4. Seifert, U. Adv. Phys. 46, 13–137 (1997).
of a cell, and a substantial part of it of the processes, and because of the large 5. Jenkins, J. T. J. Math. Biol. 4, 149–169 (1977).

needs to remain freely accessible for the
endoplasmic reticulum to communicate
with other organelles12. The forces applied
by these boundary rings can be revealed
by analysing the shape of the ‘free’ pieces
of the membrane (Fig. 2a–c). Protein
size of the proteins involved, the proteins’
assembly can be seen as controlling
membrane remodelling through a ring-like
boundary 6,13 (Fig. 2d).
The recent reconstitution of endoplasmic
reticulum morphogenesis and membrane
6. Chernomordik, L. V. & Kozlov, M. M. Annu. Rev. Biochem.
72, 175–207 (2003).
7. Agrawal, A. & Steigmann, D. J. Biomech. Model Mechanobiol.
doi:10.1007/s10237-008-0143–0 (2009).
8. Brown, F. L. Annu. Rev. Phys. Chem. 59, 685–712 (2008).
9. Hess, S. T., Gudheti, M. V., Mlodzianoski, M. & Baumgart, T.
Methods Mol. Biol. 400, 367–387 (2007).

assembly into ring-like structures is also
responsible for shaping the intermediates
of membrane fusion and fission6. In these
cases the need for multiple proteins is
stipulated by the high energy requirements
remodelling in simple systems containing
a few proteins and lipid matrix 12,13 is
expected to make the membrane shape
analysis proposed by Wiggins and
co-workers more straightforward. If
10. Heinrich, V., Bozic, B., Svetina, S. & Zeks, B. Biophys. J.
76, 2056–2071 (1999).
11. Capovilla, R., Guven, J. & Santiago, J. A. Phys. Rev. E
66, 021607 (2002).
12. Hu, J. et al. Science 319, 1247–1250 (2008).
13. Bashkirov, P. V. et al. Cell doi:10.1016/j.cell.2008.11.028 (2008).

174 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

© 2009 Macmillan Publishers Limited. All rights reserved

 review article
Published online: 20 february 2009 | doi: 10.1038/nmat2387

Deformation and failure of protein materials in

physiologically extreme conditions and disease
Markus J. Buehler1,2,3* and Yu Ching Yung4
Biological protein materials feature hierarchical structures that make up a diverse range of physiological materials. The analysis
of protein materials is an emerging field that uses the relationships between biological structures, processes and properties to
probe deformation and failure phenomena at the molecular and microscopic level. Here we discuss how advanced experimental,
computational and theoretical methods can be used to assess structure–process–property relations and to monitor and predict
mechanisms associated with failure of protein materials. Case studies are presented to examine failure phenomena in the
progression of disease. From this materials science perspective, a de novo basis for understanding biological processes can
be used to develop new approaches for treating medical disorders. We highlight opportunities to use knowledge gained from
the integration of multiple scales with physical, biological and chemical concepts for potential applications in materials design
and nanotechnology.

P
roteins are essential building blocks of life, forming a diverse
group of biological materials that ranges from spider silk
to bone, and from tendons to the skin, all of which play an
important part in providing key biological functions1–7. These
materials are distinct from the conventional categories of structure
and material, as they represent the merger of these two concepts
through hierarchical formation of structural elements that range
from the nanoscale to the macroscale (Fig. 1a, b).
Protein materials are abundant in biology and play an essential
part in the biological function of all cells and tissues within
organisms. Many such materials with mechanical function form
structural filaments, trusses or fibres, whereas others retain the
globular structure of their protein constituents. Biological protein
materials are commonly characterized by where they reside with
respect to their associated tissue, to serve as either extracellular
or intracellular protein materials1. Intracellular protein materials
provide the architecture of a cell and include vimentin, microtubules,
actin (proteins from a cell’s cytoskeleton) and lamins (forming the
cell’s nuclear envelope)1–3. Extracellular protein materials, secreted
by cells into the surrounding microenvironment, include elastin
and collagen1,4,5. Other protein materials such as fibrin6–9 appear in
biological processes such as the clotting of blood. Distinct protein
materials in organisms cannot be considered in isolation. For
instance, lamins are connected to extracellular protein materials
(such as collagen) through cytoskeletal proteins (for example actin
and nesprin).
Here we focus on defining structural components of biological
protein materials from a materials science perspective, and on their
role in mechanical materials phenomena, specifically materials
failure, in diseased or altered physiological conditions. Advances in
experimental, theoretical and computational materials science have
led to a deeper understanding through the linking of structure to
process and property (Fig. 1c). An overview of failure in biological
systems, broadly defined as the loss of its ability to provide an
intended physiological function, will be specifically presented in the
context of material breakdown due to a range of causes, including
altered chemical or physical boundary conditions (such as extreme
forces), weakening of tissues due to structural flaws because of
genetic defects, or the inability of a tissue to provide its function
because of the interaction with an ectopic or foreign material.
Within the biological sciences, the field of genomics has
advanced our knowledge base through the successful sequencing of
entire genomes. Extensive efforts have been made to move beyond
genomics, where fields such as systems biology provide explanations
of mechanisms of how genes affect phenotypes and biological
function10,11. More recently, a movement to understand the materials
science of proteins, aimed at developing relationships between
structure (of biological protein materials), process (for example self
assembly) and property (for example strength or elasticity), shows
great promise in contributing to a deeper understanding of biological
systems. This study of material properties of hierarchical protein
structures and their effect on molecular and microscopic properties,
by using mechanistic insight based on structure–process–property
relations in the biological context (Fig.  1d), provides a basis for
understanding disease processes. This could help in developing new
approaches to treating genetic and infectious diseases, injury and
trauma, as well as in improving engineered materials by translating
material concepts from biology. Associated materials science
challenges and opportunities are summarized in Table 1.
Materials in biology often involve non-protein components
such as crystal platelets (in bone, dentin and nacre, for example),
which have important effects on their mechanical properties.
They also contain larger-scale structural features, which influence
the overall mechanical properties (for example porosities and
interaction with fluids). Although proteins are an important
element that affects most materials in biology, they alone do not
provide the information necessary to characterize all relevant
material properties. The discussion of these effects is beyond the
scope of this review.
The cascaded arrangements of building blocks at defined
length scales form hierarchical structures (for example molecules,
filaments, mesoscale structures) which control a material’s
properties. Figure  1a,  b shows the structure of two example
materials, showing lamin intermediate filaments (in the following
referred to as ‘lamins’)12–15 and collagen (for example tendon and
cornea)4,5, and illustrates how structure and material converge.

1
Laboratory for Atomistic and Molecular Mechanics, Department of Civil and Environmental Engineering; 2Center for Computational Engineering; 3Center
for Materials Science and Engineering, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, USA.
4
Laboratory for Cell and Tissue Engineering, Harvard University, School of Engineering and Applied Sciences, 40 Oxford Street, 415 ESL, Cambridge,
Massachusetts 02138, USA. *e-mail: mbuehler@MIT.EDU

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 175

review article Nature materials doi: 10.1038/nmat2387

O(10 µm)
a O(1 nm) O(1 µm)

O(20 nm) O(50 nm)

Coiled-coil dimer Filament

α-Helix
Lamina mesh Cell nucleus
network (blue colour)

Universal motifs Functional properties

O(100 µm)
O(10 µm)
b O(1 nm)
O(300 nm) O(1 µm)
O(1 cm)

TC assembly Collagen microfibril

Tropocollagen
triple helix Collagen
Collagen fibril Tendon
fibre

c Process d Process HN

Property Process H1 Property HN

Structure Process H0
(transcription, Property H1
Requirement translation)
Requirement RN
Property H0

Structure H0 Requirement R1

Structure H1 Requirement R0

Structure HN

Gene regulation/ Sensing (e.g. changed

activation requirements)

Figure 1 | Examples of hierarchical multiscale structures of biological protein materials. Here we show lamin intermediate filaments (lamins),
collagenous tissues (tendon) and the materials science paradigm for the analysis of biological protein materials. a, Structure of lamins. α-Helical protein
domains assemble into dimers, which form filaments that define a lattice-like lamin network of the cell’s nucleus. Lamina mesh network snapshot adapted
and reprinted with permission from ref. 16. © 1986 NPG. Cell nucleus image ©TenOfAllTrades. b, Structure of collagenous tissues, from nanoscale
to macroscale. Tropocollagen (TC) molecules assemble to form fibrils, which form fibres that provide the structural basis of collagen tissues such as
tendon. Collagen fibril image reprinted with permission from ref. 76. © 2002 Elsevier. Image of Achilles tendon from SPL. c, Materials science triangle
that links structure, process and property. d, Materials science paradigm applied to the hierarchical structure of protein materials (Hi refers to hierarchy
levels i = 0 to N, Ri refers to material property requirements at hierarchy levels i = 0 to N). The cycle initiates at H0 (process H0, the only level at which
protein expression occurs) to form protein constituents (structure H0). Their properties (property H0) control the association at the next hierarchical level
(process H1 leading to structure H1). This cycle continues through all hierarchical levels i = 0 to N, where process and structure H1 and beyond denote
protein assembly stages. At each stage, the properties of the structure control the assembly at the next level. Overall, properties at different hierarchical
levels (properties Hi) are regulated by corresponding physiological demands (requirements Ri), which are sensed and transduced intracellularly to activate
genetic regulation, resulting in changes to process H0.

Lamins provide structural support to the cell’s nucleus and
form an interface between the cell’s cytoskeleton and chromatin.
Figure  1a shows the structure of lamins12–15, characterized by a
hierarchical assembly of α-helix-based protein domains. The basic
building block of lamins is an α-helical coiled-coil dimeric molecule.
Lamin dimers form filaments that develop into a network with a
lattice-like structure, found primarily at a cell’s nuclear membrane16.
Because the cell’s cytoskeleton itself is coupled to the surrounding
extracellular matrix (for example collagen) by adhesion proteins,
macroscopic deformation of tissues typically leads to deformation
of the cell nucleus17–20. The lamin network thus aids the coupling
of tissue-level mechanical signals to complex biochemical processes
(such as gene regulation) within the cell nucleus16,17,21,22. The study
of strain-regulated mechanisms of cell response falls broadly into
the field of mechanotransduction3,23. Lamins represent a model
material that illustrates the intimate connection between structural
material and biochemical properties. Recent studies have provided
substantial evidence that underlines the role of lamins in cancer and
genetic diseases17,21,22,24.
Collagen, the most abundant protein on the Earth, is a fibrous
structural protein with superior mechanical properties. It is found in
a number of tissues exposed to severe tensile or compressive loading,
including tendon, bone, teeth, cartilage and the cornea4,25–29. The
hierarchical structure of collagen is summarized in Fig. 1b. Collagen
consists of triple helical tropocollagen molecules that appear as
‘nano-ropes’ with lengths of around 300  nm (ref.  30). Staggered

176 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

Nature materials doi: 10.1038/nmat2387
Table 1 | Key materials science challenges associated with the study of biological protein materials, examples and opportunities for
further development of methods.
Materials science challenge Example
Confinement effects Change of material properties at
different length scales and timescales
(for example size effects, pulling rate
effects)
Environmental effects Effects of chemicals and solvent (for
example pH) on formation, function and
material breakdown
Material failure Material degradation in genetic and
infectious diseases, failure due to impact example osteogenesis imperfecta,
and stress in injuries or trauma
Multiscale synthesis or Structure control, from nanoscale
fabrication to macroscale; how to control
implementation of hierarchies
Multiscale characterization Structure measurement at various
timescales and length scales
(for example assembly stages),
characterization of mechanisms of
deformation
Significance
Adaptation of universal or diverse
protein structures, biomaterials
Tissue remodelling, protein
filament adaptation/remodelling,
mechanotransduction
(chemomechanical signalling)
Mechanisms of genetic diseases (for
progeria, muscle dystrophies,
Alzheimer’s disease), nanomedicine
Synthesis of synthetic biofibres (such
as spider silk) and bioinspired or
biomimetic materials (scaffolding
protein materials for tissue engineering,
interfaces between microdevices and
biological systems)
Application of atomic force microscopy, Statistical analysis, high-throughput
optical/magnetic tweezers, microfluidic methods, error/uncertainty analysis
devices, transmission electron
microscopy for material characterization,
integration of measuring and imaging
approaches across multiple scales to aid
model development
review article
Need for model development
Constitutive material models with
explicit consideration of hierarchies
Chemomechanical models of tissues,
atomistic-level descriptions that
integrate mechanics and chemistry
Constitutive strength models with
explicit consideration of hierarchies,
plasticity/failure models (for example
rupture of hydrogen bonds), disease
models (such as role of molecular
defects and misfolds)
Integration of universal and diverse
elements in hierarchical structures,
translation from biology to
nanotechnology (carbon nanotubes,
nanowires, peptides)

arrays of tropocollagen molecules form fibrils, which combine to
form collagen fibres and then to form tissues such as tendon, which
functions as glue between muscle and bone. Tropocollagen molecules
show extreme extensibility, withstanding up to 50% tensile strain
before catastrophic failure. This, along with their ability to sustain
stresses of several gigapascals, is crucial for their physiological
mechanical role in tissues31. The staggered arrangement of molecules
into fibrils provides the basis for its ability to dissipate mechanical
energy through molecular sliding rather than leading to catastrophic
failure32–34. This staggered architecture plays a key part in increasing
the toughness of various collagen materials such as tendon or bone.
The overall hierarchical arrangement is important because, through
structures formed at characteristic geometric length scales, it enables
superior molecular properties to be visible at larger, biologically
important, intermediate mesoscales. A possible concept to explain
the observed length scales at each hierarchical level is that they are
a result of structural adaptation towards maximizing target material
properties (such as strength and dissipation) through geometric
size effects, as discussed in the context of bone35. Several genetic
diseases are related to defects in the collagen structure, leading to
mechanically compromised tissues (an example is the brittle bone
disease osteogenesis imperfecta).
The biological materials introduced above feature properties
that are characteristic for this entire class of materials, including
robustness, adaptability and multifunctionality 4,11. Robustness is
defined as the degree of separation between stability and failure,
and is crucial to understanding the role of protein materials in
biology, for both physiological and pathological conditions (such
as varying pH, forces or structural changes). Adaptability refers
to the ability of a material to cope with environmental changes by
changing its structural arrangement to cope better with changed
conditions. For example, tissue remodelling in bone plays a crucial
part in improving the material’s damage tolerance through its
intrinsic ability to repair itself 36. The underlying mechanism is that
small cracks, formed because of physiological mechanical load, are
detected by ‘bone remodelling units’ and removed from the tissue
before they reach a critical size at which catastrophic failure would
occur. Such mechanisms show the intricate connections between
tissue formation and overall failure properties. In vivo, the structures
of protein materials self-maintain or adapt by means of feedback
loops by translating spontaneous demands in the microenvironment
(through intracellular signalling) to regulate gene activation or
deactivation. Ultimately, this alters the material’s structural makeup
to better suit the local physiological needs (Fig. 1d). Multifunctionality
refers to the material’s ability to provide multiple functions such as
mechanical strength and the ability to control biochemical processes
(as observed in lamins that have structural as well as biochemical
roles). Many protein materials show graceful degradation, reflecting
the material’s ability to induce a controlled breakdown of a system’s
function after damage without leading to catastrophic failure and
without affecting materials in the environment. The hierarchical
structure may be the basis for these unique properties, distinguishing
them from many engineered materials.
The formation of protein materials in biological systems occurs
through self assembly of protein molecules whose structure is
encoded in an organism’s DNA. Protein molecules are composed
of long polymer chains, constructed of amino acid monomers,
and synthesized by a translation process that converts the genetic
information transcribed from DNA into RNA into a polypeptide
sequence of amino acids. Protein synthesis is initiated at precise
sites during mRNA translation and terminated when signalled by
the presence of encoded stop codons (messages). Once a protein
molecule is constituted, the polypeptide chain folds into its unique
three-dimensional (3D) conformation with the aid of chaperone
proteins1. Functional activation of the protein structure often
involves binding with other molecules to form the protein’s intended

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 177

review article Nature materials doi: 10.1038/nmat2387
a Within cell nucleus Intra-/extracellular

DNA nucleotides
define type of AA

Example: Proline -
DNA
nucleotides

Sequence of amino acids Protein molecule

Transcription/ polypeptide (1D structure) (3D structure)
translation

b Radial growth c
Lamin
Axial growth
dimers Banded structure

I II III

Axial growth Radial growth

d e

Spider silk PO42–, K+,H+

filaments

Protein Chemistry (pH, ions)

constituents mechanical shear
and elongation

PO42–, K+,H+

Figure 2 | Formation and assembly of biological protein materials, including cellular expression of proteins and assembly into larger-scale hierarchical
structures. a, Generation of protein molecules through translation from the DNA genetic code, leading to the expression of individual 3D protein
molecules. b, Schematic biological assembly of lamins14 and experimental visualization. The self assembly of protein molecules into large-scale structures
proceeds through a complex process that involves several intermediate structural steps. For lamins, assembly occurs first by association of two dimers
into tetramers, then by a period of growth in the axial direction (I), followed by growth in the radial direction (II and III). Experimental visualizations of
the assembly process of lamins are obtained from electron microscopy studies (rat liver lamin A/C, scale bar 100 nm). Adapted and reprinted with
permission from ref. 14. © 2004 Annual Reviews. c, Experimental visualization of the assembly of tropocollagen into collagen microfibrils based on
contact-mode atomic force microscopy height images. The characteristic regular-spaced banding structure of collagen microfibrils is visible in the marked
region where multiple tropocollagen molecules begin to interact (black, 0 nm height, white, 35 nm height). Adapted and reprinted with permission from
ref. 76. © 2002 Elsevier. d, Schematic diagram of the biological process of formation of spider silk protein filaments in a spider’s spinning duct, involving a
controlled change of pH, solvent chemicals and mechanical shear, leading to the formation of the β-sheet-rich spider silk filaments39. e, Synthetic process
of spider silk formation, realized in a microfluidic device (left), together with a micrograph (right) of the resulting synthetic fibres (scale bar 10 μm).
Adapted and reprinted with permission from ref. 39. © 2008 National Academy of Sciences USA.

functional structure1 (Fig. 2a). In vivo, many protein materials obtain
their specific functionality by post-translational modifications,
which include hydroxylations (for example during formation of
tropocollagen’s hydroxyproline residues), phosphorylations and
glycosylations (for example during formation of cartilage tissues), as
well as enzymatic crosslinking. These modifications are particularly
important for the material properties of tissues, as they control the
interaction between proteins and with other material components
(such as inorganic materials or sugar-based components) as well
as their bioactive properties. These modifications are particularly
difficult to mimic ex  vivo or through synthetic approaches,
posing a challenge in the development of biomimicking and
biocompatible materials.
The resultant individual protein components self-assemble at
different timescales to form a protein structure with hierarchical
geometrical entities (Fig. 1a, b). Assembly mechanisms have been
investigated based on the combination of imaging tools with
controlled assembly conditions (pH, temperature, solvent). This
approach has elucidated the mechanism of lamin assembly 14,37
(Fig. 2b). In this process, lamin dimer association and axial growth
are followed by radial growth to form larger fibrils, resulting
in formation of a network. Figure  2b also shows the direct

178 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

Nature materials doi: 10.1038/nmat2387 review article
a b
Diversity

Diversity: entire protein structure Continuum

Organisms
Genetic RC approaches
sequence Organs (biomechanics)
Networks, tissues Mesoscale

Length
science
AH Protein, enzymes,
of protein
BH filaments (e.g. dimer,
materials
Universality: protein motifs (AH, BS, RC) tetramer)
Single-
O(µm, µs)
molecule
Protein motif
Universality

ACGT DNA
20 AAs methods
chemistry Double helix AH, BS, RC

Ångström Nanometre Micrometre Millimetre/metre Time

Nanoscale Mesoscale Macroscale

Figure 3 | Universality and diversity of biological protein materials. a, Degree of structural diversity and universality in protein materials, plotted against
biological length scales (equivalent to hierarchical levels). Inset, a visualization of the topoisomerase protein (biological role is to cut strands of the DNA
double helix; visualization using visual molecular dynamics140). This inset illustrates how universal motifs define the overall functional properties of this
protein, whereas the entire protein structure represents diversity. See also Table 2. AH, α-helix; BS, β-sheet; RC, random coil; AA, amino acid. b, Placement of
mesoscale science of protein materials, between single-molecule studies and micromechanical and continuum-type biomechanics applications.

Table 2 | Occurrence of protocols in four sample materials, illustrating that some protocols are universal to all protein materials
whereas others are specific.
Protocol Biological protein material
Collagen (tendon) Lamin Spider silk Amyloids
DNA nucleotides* (ACGT) ✓ ✓ ✓ ✓
Vimentin DNA sequence† ✓
Collagen DNA sequence† ✓
Spider silk DNA sequence† ✓
Amyloid DNA sequence† ✓
DNA double-helical structure* ✓ ✓ ✓ ✓
Protein building blocks* (20 AAs) ✓ ✓ ✓ ✓
Motifs* AH, BS, RC ✓ ✓ ✓ ✓
Collagen fibril† ✓
Lamin filaments† ✓
β–Nanocrystal composite† ✓
Cross-β fibril structure† ✓
*Universal; diverse. AH, α-helix; BS, β-sheet; RC, random coil; AA, amino acid.
†

experimental visualization of these assembly processes. A snapshot
of the assembly process for collagen fibrils is shown in Fig.  2c,
revealing the characteristic banding structure of collagen fibrils
when multiple molecules assemble into microfibrils. The biological
process of in  vivo assembly of many structural protein materials
involves a dynamic change of physical and chemical conditions, as
has been shown for the synthesis of spider silk38,39 (Fig. 2d).
In addition to studies of naturally occurring biological protein
materials, recent research led to the development of techniques
that enable one to change their structural makeup, and to design
and synthesize synthetic analogues40–44 through recombinant DNA
techniques, RNA interference knockdowns or sequence insertions.
Two primary routes of development pursued include ex  vivo
assembly (for example self-assembling peptide systems40,41,44) and
in vivo expression of protein materials (for example through bacterial
hosts42,45). The ability to control the DNA sequence information at
a fundamental level provides us with the ability to engineer the
structure of protein materials at the molecular (amino acid) scale.
EAK16, for example, belongs to an interesting class of self-assembled
peptides that constitutes a material platform for a variety of biological,
biomimetic and nanotechnology applications46. Knowledge of the
details of the in vivo processes enables one to mimic these processes
ex vivo, as has been demonstrated for spider silk fibres in microfluidic
devices39 (Fig. 2e).
The ability to control self assembly synthetically and genetically
provides a powerful framework for the study of the links between
structure, property and process. The approach of adding and
deleting molecular domains with distinct chemical functionality
has made possible the detailed study of the mechanisms of self-
assembly processes (for example identification of essential protein
domains)47–50. This allows one to determine how changes to the
protein structure alter biological function and disease properties.
The evolution of protein materials through genetic selection
and structural alterations has resulted in a specific set of protein
building blocks that define their structure. Protein materials exist
in abundant variety, and the need exists to formulate a widely
applicable model to categorize all such materials systematically, in
order to establish a fundamental understanding, and to make use of
hierarchical structural building blocks to develop a new generation
of advanced nanomaterials51–53. A protocol is defined here as a
term that encompasses a general analysis of protein materials that
describes the use of structural building blocks (for example α-helices,
β-sheets, random coils) during their formation and function, and the
process or mechanism of use of this material (for example synthesis,

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 179

review article
breakdown, self assembly). The phenomenon of universality exists
ubiquitously in biology, where certain protocols are commonly
found in all protein materials (such as the use of hierarchical levels
of building blocks: DNA nucleotides, DNA double helical structure,
α-helices, β-sheets and the process of transcription/translation,
protein synthesis). However, other protocols are highly specialized
(such as the use of specific DNA sequences for a particular protein
structure, the resultant protein motifs of tendon fascicles, lattice-like
lamin structure, and so on), thus representing diversity. Protocols
can therefore be classified as either universal or diverse.
Universal and diverse protocols are distributed heterogeneously
across hierarchical levels (Fig.  3a and Table 2). The four DNA
nucleotides (ACGT) represent a universal protocol common to
all protein materials, where their arrangements in diverse patterns
form the immense variety of genetic sequences found in biology.
Genetic sequences are universally encoded in a double-helical DNA
structure, regardless of the specific nucleotide sequence. Through the
universal process of transcription and translation, protein molecules
are synthesized into a 1D sequence of the universal 20 amino-acid
building blocks. Virtually all proteins contain one or more of these
universally found motifs: α-helices, β-sheets and random coils.
These universal motifs arrange into unique, diverse larger-scale
protein structures (for example enzymes, fibres, filaments). A greater
diversity of protocols is generally found at higher levels, suggesting
that biological functionality is associated with structural diversity.
Universality is associated with protocols that can be used to derive
diverse functionality at larger hierarchical levels. A fundamental
difference between engineered materials and naturally formed
biological materials is that functionality in biology can be created by
arranging universal building blocks in different patterns, rather than by
inventing new types of building block. The formation of hierarchical
arrangements provides the structural basis that allows the existence of
universality and diversity within a single material. This combination
of dissimilar concepts may explain how protein materials are capable
of combining disparate material properties, such as high strength and
high robustness, together with multifunctionality.
Biological functionality must be understood at varying scales.
Biochemistry focuses on biological functionality at molecular
scales. The mesoscale, which encompasses length scales ranging
from nanometres to micrometres and timescales of nanoseconds
to microseconds, is a particularly important level necessary to
understand how specific protein materials derive their unique
properties and what role they play (in both physiological and
pathological phenomena; Fig. 3b ). The mesoscale science of protein
materials, through the linking of molecular properties to properties
of protein materials at the microscale, thus represents an important
frontier of materials research with great potential for fundamental
contributions to biology and medicine as well as for the de  novo
synthesis of engineered materials.
The approach of making use of universal building blocks to create
diverse multifunctional hierarchical structures has been successfully
applied in current macroscale engineering paradigms. For example,
in the design of structures such as buildings or bridges, universal
constituents (bricks, cement, steel trusses, glass) are combined to
create multifunctionality (structural support, living space, thermal
properties, light harvesting) at larger length scales. The challenge of
using similar concepts that span to the nanoscale, as exemplified in
biological protein materials by merging structure and material, could
lead to the emergence of new technological concepts. A key obstacle
in the development of new materials lies in our inability to control
the structure formation directly at multiple hierarchical levels. The
concept of universality and diversity and the knowledge gained
from how to characterize these materials at different hierarchical
levels may help in addressing these challenges.
Nature’s use of a limited number of universal building blocks,
arranged in a variety of ways, is a limitation as well as a strength
180 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
Nature materials doi: 10.1038/nmat2387
of biological systems. For example, although the performance of
structural tissues in our body is poor compared with most engineered
materials (such as steel or concrete), their performance is remarkably
good considering their inferior building blocks. Understanding these
material concepts and their translation to the design of synthetic
materials could provide us with new concepts for materials design based
on inexpensive, abundant and environmentally benign constituents.
Multiscale failure mechanisms of protein materials
The stability and failure mechanisms of biological protein materials
play an integral part in defining their functional role in biological
systems. In this context, a bottom-up description must begin with
the study of the properties of individual protein molecules. A
variety of experimental, theoretical and computational tools have
been developed to assess structure–process–property relations and
to monitor and predict materials failure mechanisms.
In biological protein materials, each hierarchical level has its
own defects and failure mechanism. Over multiple hierarchical
levels, there are two primary types of response. First, if a material
is able to tolerate defects and failure at specific scales, it retains
its overall integrity because the presence of multiple hierarchical
levels decreases the overall probability of catastrophic failure.
Second, some defects and failure mechanisms lead to a catastrophic
breakdown of the material (for example in genetic disease, where
proteins are misfolded owing to errors in the DNA code). Many
physiological conditions exist in the realm where the material is
able to tolerate defects and failure mechanisms (such as microcracks
in bone36), and many diseases originate because of the material’s
inability to mitigate material defects and breakdown. In both cases,
the basic mechanisms of failure can be evident through the breaking
of hydrogen bonds, protein unfolding, sliding of molecules against
each other and breaking of crosslinks. Failure occurs because the
intramolecular bonding (for example hydrogen bonds) is crucial to
defining the structure of protein building blocks. Similarly, adhesion
forces due to a charged surface or hydrophobic forces between
molecules, fibrils, fibres and filaments are elementary to providing
function at larger length scales. Often, these failure mechanisms are
competing and depend on loading rate, pH, density of crosslinks
or molecular geometry 34,54–56. These biological failure mechanisms
can be analogous to failure in engineering materials, which is
mediated by defects such as cracks, dislocations or mass transport
along grain boundaries. The quantitative analysis of multiscale
failure mechanisms provides great potential in explaining the role
of materials in their respective biological context.
Advances in experimental techniques have allowed the structural
characterization of protein materials across multiple scales
(Fig.  4). Techniques such as nuclear magnetic resonance (NMR)
spectroscopy and X‑ray crystallography have advanced our ability to
identify 3D protein structures57. Site-specific studies using NMR, a
technique that allows internal molecular motions to be probed while
spanning a range of timescales with precise spatial resolution, have
provided insight into the dynamics of large protein complexes58. The
analysis of molecular-scale deformation mechanisms has also been
achieved through use of X‑ray crystallography, providing structural
and temporal information about mechanisms of deformation
and assembly (for example in intermediate filaments, tendon or
bone)25,32,50,59. The development of the Protein Data Bank in the
1970s enhanced this progression, serving as a public repository for
tens of thousands of 3D atomistic protein structures, identifying
the structure of numerous proteins from varying species sources60.
X‑ray tomography provides insight into the internal structure of
a wide range of length scales encountered in protein materials.
The high penetrating power of X‑ray tomography, coupled with a
near absence of reflection at the interface of dissimilar materials,
makes this an ideal probe for reconstructing the morphology
and composition of proteins at high resolutions (up to 50  nm)61.
Nature materials doi: 10.1038/nmat2387 review article
a
O(300 nm, µs) Nano-
indentation
MEMS testing
>min Micropipette
Collagen fibril
c

Continuum
models
s
O(Å, ns)
Meso-
scale
models Optical/magnetic tweezers
Timescale

µs

Non-
reactive Atomic force microscopy b
MD 10

ns Reactive
450 nm

Width (nm)
MD
280 nm 5
Tomography
500 nm
QM/DFT LIF
ps
Transmission elctron
0
microscopy (e.g. cryo)

Å X-ray nm µm m Length scale

diffraction
NMR
Nanoparticles
DNA (nanowires, Cells
polypeptides carbon
nanotubes) Tissues
Secondary protein
organs
structures (e.g. β-sheets,
organisms
α-helices)

Figure 4 | Tools for characterization and modelling of deformation and failure of biological protein materials. Respective timescale and length-scale
domains of applicability are shown. Experimental methods include X‑ray diffraction, transmission electron microscopy, atomic force microscopy (AFM),
optical/magnetic tweezers; microelectromechanical systems (MEMS) testing and nanoindentation. Theoretical and simulation tools include quantum
mechanics/density functional theory (QM/DFT), molecular dynamics (MD), coarse-grained models and mesoscale atomistically informed continuum
theories, as well as continuum models. The lower part of the figure indicates classes or scales of protein materials that can be studied with these types of
technique. a, Coarse-graining process in the development of mesoscale models, here illustrating the representation of the full-atomistic description of a
tropocollagen molecule by a collection of ‘super-atoms’ or beads, reaching length scales and timescales several orders larger. b, Deformation of a lamin
intermediate filament (LIF) using AFM. Adapted and reprinted with permission from ref. 141. © 2007 NPG. c, Tensile stretching experiment of a collagen
fibril with a MEMS device (scale bar 2 μm). Reprinted with permission from ref. 82. © 2006 Royal Society.

Mass spectroscopy provides information about the chemical
content of protein materials. Fourier-transform infrared (FTIR)
spectroscopy is a related approach that elucidates information about
chemical bonding and is an established tool used for the structural
characterization of proteins62,63. Developments in single-molecule
force spectroscopy have made it possible to study chemical,
mechanical and structural properties using manipulation techniques
such as atomic force microscopy (AFM) and optical/magnetic
tweezers20,64–72. The force-measuring capability of AFM also makes it
possible to elucidate molecular determinants of mechanical stability
and the role of force-induced conformational changes of proteins
in the regulation of physiological function73,74, and it has been used
to image the surface structure and assembly patterns (for example
in bone and collagen72,75–77). AFM has also been effectively used in
combination with fluorescent techniques for imaging, as reported
in studies of fibrin stretching 6. In a recent study, in situ fluorescent
labelling was used to visualize unfolding of cytoskeletal proteins in
cells under different mechanical loading conditions78, providing a
quantitative link between cellular-scale deformation and underlying
molecular-scale rupture mechanisms. The basic concept of this
approach is that protein unfolding due to rupture leads to the
exposure of binding sites, which can be used to bind fluorescent
markers for imaging. This technique has shown that protein rupture
occurs as an underlying mechanism in cells under mechanical
load78. Nanoindentation is another method used to probe the
mechanical response of protein material composition and structure
at multiple length scales79, and has allowed the characterization of
multiscale mechanical behaviour, as demonstrated for bone80,81. The
development of devices based on microelectromechanical systems
(MEMS) provides the ability to carry out tensile tests of micrometre-
sized samples, as demonstrated for individual collagen fibrils82. The
available experimental methods cover a broad range of timescales
and length scales. Remaining challenges include specificity with
respect to the handling of individual molecules and filaments,
their control at very short timescales, and the ability to observe
mechanisms directly at molecular scales.

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 181

review article
a b
Intramolecular H-bonds
Energy
I I–II II
r 20 Molecular
II
I I–II r stretching
3D path
c d
2,500
F F 2,000
Rupture force (pN)
1,500
Fxb
Eb 1,000
500
xb
0
–9
Figure 5 | Chemomechanical behaviour of protein constituents.
a, Schematic diagram of the energy–distance curve of a protein domain
transitioning from the folded to the unfolded state. The inset depicts this
process during unfolding of a convolution in an α-helix protein domain.
State I corresponds to the case when donor–acceptor pairs form hydrogen
bonds (H-bonds) in an α-helix convolution. The dotted line represents a
second state II, corresponding to the unfolded configuration of an α-helix
convolution when hydrogen bonds are broken. The transition between the
two local minima occurs through an activated state I–II in a 3D deformation
path (timescale of rupture is approximately 60 ps, where state I–II is
reached after 20 ps and state II after another 40 ps; the concurrent
breaking of all three to four hydrogen bonds within a convolution of the
α-helix represents the basic mechanism of rupture)55. b, Mechanism of
deformation of a single tropocollagen molecule under tension, showing
the dynamics of hydrogen-bond breaking under stretching of a single
tropocollagen molecule (hydrogen bonds break at a rate of two hydrogen
bonds for every 1% increase in strain)86. c, Resulting effective energy
landscape for such a process, showing how the energy barrier is reduced
because of a laterally applied force F (the inset displays the loading
condition of an α-helix protein domain under tensile loading). d, Strength
properties of several protein domains as a function of deformation speed
(α-helices and β-sheets, data plotted from refs 55,67,70,142). Faster
deformation (that is, short timescales) generally leads to an increased
rupture force. The dotted lines are included to guide the eye; they
approximate regimes of straight lines in the plot of F against ln(v).
Complementing experimental approaches, atomistic simulation
methods provide a bottom-up description of materials based
on the dynamical trajectory of each atom, by considering their
atomic interactions. Molecular dynamics (commonly abbreviated
as MD) solves each atom’s equation of motion83. In classical
molecular dynamics, the 3D structure of an atom is approximated
by a point particle. The numerical integration of Newton’s law
(F  =  ma), by considering appropriate force fields, enables one to
simulate a large ensemble of atoms that represents a larger material
volume. The atomic interactions are described by decomposing
the behaviour of chemical bonds into stretching, bending or
rotation contributions83–85. Because of computational limitations,
atomistic-based simulation approaches are typically limited to
billions of particles (that is, length scales below micrometres) and
timescales of less than a microsecond.
Despite the shortcomings with respect to accessible timescales
and length scales, the uniqueness of atomistic methods is the ability
to directly visualize the dynamics of deformation and failure. For
182 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
Nature materials doi: 10.1038/nmat2387
example, results of molecular dynamics simulations elucidated
50 H-bond mechanisms of hydrogen‑bond rupture in protein domains,
40 as shown in Fig.  5a (inset)55. Similar hydrogen‑bond failure
mechanisms have been observed in other proteins, such as β-sheet
H-bond
30
rupture crystals55. Studies analysing the deformation and failure mechanisms
of tropocollagen molecules found that initial molecular rotation is
rotation followed by hydrogen‑bond breaking and subsequent stretching
10 Backbone
of the protein’s polypeptide backbone, followed by covalent bond
0
0 20 40 rupture, explaining collagen’s characteristic stiffening 31,86 (Fig. 5b).
Tensile strain (%) Molecular dynamics simulations have also been useful in defining
experimental protocols for in situ fluorescent labelling by identifying
key rupture mechanisms and exposing domains78.
Experimental (titin 127) Larger length scales and timescales can be reached through
MD simulation (127)
the use of a hierarchy of simulation techniques, integrated
MD simulation (single AH)
Experimental (titin 127/31) through multiscale methods (Fig.  4). These methods are based
Experimental (single AH) on the concept of informing coarser scales from finer scales,
enabling one to establish direct links between chemical structure
and larger scales. In multiscale approaches, groups of atoms in
protein structures are often represented by ‘super-atoms’ or beads,
–7 –5 –3 –1 1 3 thereby effectively reducing the degrees of freedom (a coarse-
Pulling rate (log (m s–1)) graining approach for tropocollagen is shown in Fig.  4a). The
direct comparison between molecular dynamics simulation (in
particular coarse-graining methods) and experiment has become
more viable owing to the emergence of experimental tools that
reach ever smaller scales.
At molecular and submolecular scales, an integrated chemo­
mechanical approach is mandatory to describe the failure
properties of biological protein materials. Theoretical approaches
that link chemistry and mechanics are often derived from a
phenomenological theory originally postulated by Bell more than
30  years ago87–90. Several sophisticated, advanced models are now
available91–96 and the ‘original’ Bell model is less often used in the
analysis of protein rupture. However, the ideas behind Bell’s model
provide the intellectual foundation for many theories. The basic
concept of the interplay of applied forces and the resulting energy
landscape change is illustrated in Fig. 5c. The Bell model is based
on a statistical approach to describe the likelihood of bond rupture
under the presence of an energy barrier Eb, which is reduced due
to an applied force F at the failure point xb. The probability for a
bond to break is expressed using an Arrhenius expression, which
predicts that the lifetime of a bond decreases exponentially as the
energy barrier decreases under the applied force. Expressions that
relate the failure force F to the pulling rate v applied to a protein
predict a characteristic logarithmic dependence of the rupture
force as a function of timescales or pulling rate v, resulting in
a dependence of the form F(v; xb Eb)  =  a(xb,T)  ln(v)  +  b(xb, Eb,T)
(where a and b are parameters that depend on the energy landscape
and temperature T).
Figure  5d displays a set of experimental and computational
results, illustrating this type of relationship. Each straight line
reflects a particular rupture mechanism, characterized by an energy
barrier Eb and a transition state distance xb. Variations of pulling
rates often induce a change in failure mechanism. The analysis
of the results of the pulling rate dependence of forces allows the
interpretation of experimental and computational results through
the determination of effective energy barriers and transition states,
and provides insight into microscopic failure mechanisms.
The original Bell model does not provide a direct link between the
atomistic chemical structure of a bond (or arrangements of bonds)
and the resulting behaviour. For instance, the effective energy barrier
shown in Fig. 5c may represent a cluster of several weak bonds that
break simultaneously, or a single, stronger bond. The model further
does not capture the effect that the energy landscape itself may
change as the protein deforms. Other models have been reported
that describe the strength of proteins in dependence on the force
rate (the rate of increase of force over time)91–96. Some models also
Nature materials doi: 10.1038/nmat2387 review article
a Loading case
Fij 132
i
Strength properties
60
Experiment
40 132,212
Theory 212
3D protein ∆xapp= 0.32 nm
20
structure
0

Number of events
30 132
25
j 20 3,132
Fij 15
i ∆xapp= 0.125 nm
10
5
FIm
I 3
15
6,221
m 10
∆xapp= 0.09 nm
5
221
j 0
Rupture strength (pN)
6

b 14 c Computational
d
Experiment (type II TC) WLC
12 Experimental
Experiment (type I TC) 600
10 Titin filament
Young’s modulus (GPa)

MD 8 500 Theory
Force (pN)

Force (in pN)

8 7 400 Experiment
WLC Force
6 6 300
5 200
4
4
2 100
3
0 2 0
–2 1
0.5 0.6 0.7 0.8 0.9 1 0 0 50 100 150 200
Reduced extension (x/L) Extension (nm)

Figure 6 | Integration of theory, simulation and experimental techniques. a, Results of an elastic network model calculation, used to describe the
anisotropic failure properties of a larger protein structure. Reprinted with permission from ref. 98. © 2008 APS. b, Entropic elastic behaviour of stretching a
single tropocollagen (TC) molecule, comparing experiment (optical tweezers), coarse-grained mesoscale simulation (molecular dynamics, MD) and theory
(worm-like-chain model, WLC). Data replotted from refs 31 and 103; x/L is the ratio of extension x to contour length L. c, Young’s modulus prediction of a
single tropocollagen molecule. The graph shows a comparison between experimental and full-atomistic molecular dynamics simulation results (based on
averages from literature sources summarized in ref. 86; error bars correspond to standard deviation). d, Comparison between experimental and theoretical
modelling of stretching a long titin filament, obtained through the application of a theoretical approach that involves the Bell model and a worm-like-chain
description for protein elasticity. Reprinted with permission from ref. 104. © 1998 APS.

include a description of the transition between multiple transition A multiscale analysis that explicitly calculates or measures material
states, effectively describing the occurrence of several straight lines properties at multiple scales is compulsory.
in the F–ln(v) domain91–96 (similar to the overall behaviour shown Despite these challenges, several studies have led to direct
in Fig. 5d). comparison between experiment and theory. Protein strength
Challenges in linking simulation and experiment include models have been extended to more complex protein geometries,
differences in accessible timescales and length scales. For example, a vital step to arrive at structure–process–property relationships of
the computational prediction of the 3D folded structure of proteins strength properties98–100. Figure  6a shows an elastic bond network
directly from the amino acid sequence is, despite advances in recent model for protein unfolding mechanics, which combines an elastic
years, still at a relatively early stage and can typically only be applied model of a network of bonds with a bond fracture model based
reliably to simpler protein structures, as the timescales for protein on Bell’s concept 98. The model, here applied to green fluorescent
folding are beyond current computational capabilities97. protein, treats bonds as springs and provides a direct structure–
Because biological protein materials feature properties that property relationship between the geometry and bond rupture
are strongly length-scale and timescale dependent, the explicit (for example the protein’s mechanical anisotropy). It is capable of
consideration of particular scales for the quantitative comparison accurately reproducing experimental results reported earlier for the
of measurements and theory is essential. Measurements at different same protein structure101 (for comparison see Fig. 6a).
timescales and length scales lead to significant variations in material A major success is the use of the worm-like-chain model to
properties. For timescales, this can be observed in the results of describe entropic elasticity 69,102. The use of Bell-type strength
strength properties reviewed in Fig. 5d55,67,70. This plot explains why models in conjunction with worm-like-chain models resulted in
many molecular dynamics simulations predict a vastly different quantitative agreement between theory and experiment. Figure 6b
behaviour from experimental studies, because they are carried out depicts the comparison of an optical tweezers experiment of
at different timescales. An implication of these strong size effects stretching a single tropocollagen molecule with results from
is that one cannot directly use single-molecule measurements to molecular dynamics simulation (based on the mesoscale
infer larger-scale properties, and that homogenization approaches coarse-grained model shown in Fig.  4a) and the theoretical
commonly used for crystalline materials cannot be applied directly. worm-like-chain model31,102,103. All three approaches provide a

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 183

review article
consistent description of the protein’s entropic elastic behaviour.
Measurements and theoretical predictions of the stiffness of single
tropocollagen molecules at larger deformation (when hydrogen
bonds within the molecules are being stretched) are compared in
Fig. 6c, revealing good agreement between various experimental
and molecular dynamics simulation approaches. Under large
forces and displacements, the worm-like-chain model fails and
rupture of hydrogen bonds within protein domains dominates
(see Fig. 5a, b)69. This behaviour has been captured in quantitative
models that combine the worm-like-chain model with the Bell
model as reported previously 104,105. Figure 6d shows a comparison
between experimental and theoretical results based on an
elastically coupled protein model that predicts elastic and strength
properties104. The theoretical prediction and experimental results
are in good agreement, suggesting that this model can accurately
reflect the interplay of strength and elasticity.
A recent paper further illustrates the integration of experiment
and theory, where measurements of molecular rupture forces
between single actin filaments and two actin-binding proteins
provide key insight into the competition of rupture mechanisms
under different pulling rates54. It was observed that whereas the
breaking of α‑actinin/actin clusters dominates at slow pulling rates,
the breaking of filamin/actin clusters governs at higher pulling rates.
This quantitative understanding of competing mechanisms is a
crucial element in the development of failure models of soft tissues.
Studies of fibrin provide another example of a protein material
where experiment, theory and simulation have been effectively
combined to improve our understanding of its material properties6,7.
The initial observation of extreme extensibility of fibrin (>300%
strain)6 motivated single-molecule studies8 as well as molecular
dynamics simulations9, which suggested that the particular coiled-
coil structure of this protein provides the structural basis for fibrin’s
great extensibility.
Role of protein materials in disease
Throughout the past century, the focus on addressing diseases
has derived primarily from a biochemical approach. However,
advancements and increased understanding have yielded a greater
appreciation of the role of materials science of protein materials in
various medical disorders. In this section, the role of biologically
relevant material properties in the progression or activation of
diseased states will be discussed.
Alzheimer’s disease, Parkinson’s disease, type  II diabetes and
prion diseases106–109 have been linked to the formation of foreign
material deposits in tissues. These material deposits, referred to
as amyloid plaques, are highly ordered hierarchical assemblies of
β-sheet protein domains that form spontaneously (Fig.  7a)110–113.
Once formation is initiated, amyloid plaques grow uncontrollably
to length scales of micrometres and persist under a wide range of
pH conditions. The mechanical robustness of these plaques has
been attributed to the large number of hydrogen bonds as well as to
steric and hydrophobic interactions between different parts of the
β-sheet structures110,114,115, although the exact structure–process–
property relationships for this material remain to be investigated.
The properties of amyloid fibrils have been probed using AFM
techniques and molecular dynamics simulations55,68,111,116,117. It is
understood that the formation of amyloid plaque deposits leads
to neurotoxicity, which interferes with the biological function
of the native tissue108,109. The mechanical robustness of amyloid
plaques and the body’s failure to eliminate these material deposits
remain the primary reasons for our inability to reverse the
progression of this disease. An improved understanding of how
the hierarchical structure of amyloid plaques contributes to their
extreme mechanical stability could lead to new strategies for
treatment through targeting selective breakdown of these material
deposits in situ.
184 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
Nature materials doi: 10.1038/nmat2387
Hutchinson–Gilford progeria syndrome, a genetic rapid ageing
disease, is caused by a structural defect in the lamin nuclear
membrane22,118–120. The progression of this disease has been
associated with mechanical failure of the cell’s nuclear membrane
in tissues subject to mechanical loading. A recent study based
on live-cell imaging and micropipette aspiration has shown that
progeria nuclear membranes display a reduced deformability and
feature the formation of fractures on application of mechanical
load22. The cause of these fractures is attributed to changes in the
lamin microstructure, where filaments form more ordered domains
that prevent the dissipation of mechanical stress. This structural
alteration leads to a change in the deformation mechanism, from
a dissipative mode (‘ductile’) in healthy cells to a catastrophic,
localized failure mode (‘brittle’) in diseased cells (Fig. 7b, c). These
mechanisms appear predominantly in cells that are subjected to
mechanical deformation, particularly in endothelial and smooth
muscle cells of the vascular system. Perhaps a loss of mechanical
integrity in the cell’s nuclear membrane can influence gene regulation
by triggering a wide range of biochemical processes that lead to the
rapid ageing phenomenon. However, the exact molecular failure
mechanisms remain unknown and their investigation represents an
opportunity for future research where a materials science approach
could make an important contribution. This disease illustrates how
material failure due to structural flaws within a protein material
can lead to the breakdown of critical biological components. Many
other genetic diseases resulting from structural flaws in the lamin
protein network have been identified, generally referred to as
‘laminopathies’118–120.
Other genetic disorders in collagenous tissues have been linked
to the alteration of the material structure owing to mutations in
the genes that encode the tropocollagen molecule. Osteogenesis
imperfecta is a genetic disease that increases the susceptibility of
bones to catastrophic brittle fracture. The origin of this disease
resides in changes to the structure of tropocollagen molecules caused
by the substitution of a single glycine amino acid (Fig.  7d)121,122.
Some collagen mutations prevent the formation of triple helical
molecules (procollagen suicide), whereas other mutations cause
structural changes to tropocollagen molecules, leading to bending
(caused, for example, by kinks induced by amino acid substitutions),
reduced mechanical stiffness (caused, for example, by changes in
the volume and hydrophobicity), or changes in the intermolecular
adhesion (caused, for example, by changes in surface charges)121–123.
At mesoscopic length scales, these molecular-level changes lead to
poor fibril packing 124,125 and a decrease in crosslink density 125,126.
Changes in the size and shape of mineral crystals in bone (for
example less organized, more rounded crystals) have also been
reported127–129 and are possibly due to a change in the ability of
tropocollagen to bind to the mineral phase of bone130,131. At larger
length scales, the effects of osteogenesis imperfecta mutations lead
to inferior mechanical properties of tendon and bone132 (Fig. 7d). A
mechanically inferior collagen matrix in addition to an increased
and less organized mineral content, and an overall reduced bone
volume caused by reduced bone turnover 133, may explain the
phenomenon of brittle bones, an important feature of osteogenesis
imperfecta125. A related disease in which collagen mutations play a
key role is Alport’s syndrome, in which a mutation in the collagen
gene alters the glomerular basement membrane of kidneys134. The
progression of mechanical failure of the glomerular basement
membrane eventually leads to the kidney’s inability to filter blood,
resulting in renal failure.
The change of cell stiffness in cancer cells has been shown
by using a microfluidic optical cell stretcher 135. Studies of AFM
indentation of cells confirmed that cancer cells display a reduced
stiffness compared with that of healthy cells136 (Fig.  7e). This
mechanical signature of cancer cells could perhaps be used
to define new diagnostic approaches in cancer detection. The
Nature materials doi: 10.1038/nmat2387 review article
a O(5 nm) O(20 nm)
O(1 nm) O(1 µm)
O(1 Å)

Chemical structure β-Strand Plaques

(H-bonds)
β-Helix (top)
Cross-β-structure Protofilament Ectopic material
(bottom)

b c Healthy nucleus Progeria nucleus

Healthy nucleus Progeria nucleus
No load

Tensile load

Fractures
Localized failure

d Mutation site e
Tumour cells Healthy cells
30 15

14 100 20 10
Max. stress (MPa)
Max. strain (%)

12
Counts

80
10
8 60
10 5
6 40
4
2 20
0 0 0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
e

I
O

Young's modulus (kPa) Young's modulus (kPa)

nc

nc
re

re
fe

fe
Re

Re

Figure 7 | Role of changes in biological protein material properties in diseases. a, Structure of amyloids, spanning from the chemical structure
of individual hydrogen bonds to the scale of amyloid plaques. Included is a visualization of molecular models proposed for amyloid fibrils (‘cross-
β-structure’ and ‘β-helix’). Protofilament structure reprinted with permission from ref. 143. © NPG. b, Effect of lamin mutations in progeria on the
mechanical properties of the cell nuclear membrane (scale bar 5 μm, fluorescent imaging). The cell nuclear membrane forms localized fractures under
large mechanical loading. c, Change in the lamin microstructure due to genetic mutation, and effect on deformation mechanism (b and c reprinted
with permission from ref. 22. © 2006 National Academy of Sciences USA). d, Geometry of point mutations in osteogenesis imperfecta (OI) in the
tropocollagen molecule (upper part), and effect on mechanical properties of tendon, where the maximum strain and maximum stress of tendon are
severely reduced under mutations (lower part, data plotted from ref. 132). e, Mechanical signature of cancer versus normal cells, revealing that cancer
cells are softer than normal cells. Adapted and reprinted with permission from ref. 136. © 1997 NPG.

approach based on microfluidic stretching 135 may provide an Future directions

effective high-throughput platform for cancer diagnosis. Studies of
cell mechanics using optical tweezers have revealed that malaria-
infected cells become stiffer, preventing normal blood flow
through lung capillaries137. The elastic cellular microenvironment
(for example the stiffness of collagen-based extracellular matrix)
has been shown to control stem cell differentiation into neuron,
muscle or bone cells138.
These case studies illustrate that the change of material properties
is a crucial element in many diseases. Translation of this knowledge
would allow detection of diseases by measuring material properties
rather than by focusing on symptomatic biochemical readings
alone. Altogether, understanding the role of different hierarchical
levels of protein materials in diseases could potentially bring about a
new paradigm of approaches to address medical disorders; however,
further research is needed to elucidate the underlying multiscale
failure mechanisms.
Along with biomedical applications, understanding of the material
concepts in protein materials could also be used to advance
nanomaterials for engineering applications. Material properties of
nanomaterials are often found to be superior to those of conventional
engineering materials. For instance, carbon nanotubes are one of the
strongest materials known but have rarely been used in structural
materials because of our inability to use their properties at larger
length scales. Understanding protein materials could provide
knowledge that would eventually be translated to connecting
disparate material scales, from nanoscales to macroscales. We may
then be able to apply these material concepts to make use of the
full potential (higher strength, higher robustness) of multiscale
engineered materials. An intriguing application may be the
development of a new material platform based on concepts similar
to those found in biological protein materials. However, rather than
using the universal protein constituents (such as α-helices, β-sheets,

nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 185

review article
random coils or tropocollagen), we could instead use building
blocks such as carbon nanotubes, nanowires and self-assembled
peptides. Some steps towards this goal have been taken in recent
studies, where conducting metallic nanowires were grown through
the use of amyloids as templates139.
Biology makes use of hierarchical structures in an intriguing
way to create multifunctional materials. Even though biochemical
and image-based diagnostics will remain important, the integration
of scales, as well as the mixing of physical, biological and chemical
concepts into new engineering designs, could complement the
current practice of disease diagnosis and treatment, as well as the
design of new materials, and thereby unfold many opportunities for
technological innovations. The wide impact of the use of materials
science approaches in biology and biomedical sciences has yet to
be demonstrated. In some specific areas such as bone (for example
osteoporosis), cartilage (for example arthritis), cardiovascular
aspects and, particularly, in the context of tissue engineering and
regenerative medicine, materials science approaches have started to
play an important part in the biomedical literature.
In addition to mechanical deformation and failure mechanisms
of protein materials, the investigation of other material properties
could be of great interest. Optical properties (for example in
the eye’s cornea and lens), photoelectric properties (for example
photosynthesis in plants), electrical (for example synapses, the links
between neuron cells), motility (for example in muscle tissues) or
thermal properties (for example thermal management) are critically
important and could be studied using a materials science approach.
The study of materials failure could provide an interesting platform
to advance our understanding of diseases.
References
1. Alberts, B. et al. Molecular Biology of the Cell. (Taylor & Francis, 2002).
2. Wang, N. & Stamenovic, D. Mechanics of vimentin intermediate filaments.
J. Muscle Res. Cell. Motil. 23, 535–540 (2002).
3. Ingber, D. E. et al. in International Review of Cytology: A Survey of Cell Biology
Vol 150, 173–224 (Academic, 1994).
4. Fratzl, P. & Weinkamer, R. Nature’s hierarchical materials. Prog. Mater. Sci.
52, 1263–1334 (2007).
5. Gelse, K., Poschl, E. & Aigner, T. Collagens — structure, function, and
biosynthesis. Adv. Drug Deliv. Rev. 55, 1531–1546 (2003).
6. Liu, W. et al. Fibrin fibers have extraordinary extensibility and elasticity.
Science 313, 634–634 (2006).
7. Weisel, J. W. Biophysics: Enigmas of blood clot elasticity. Science
320, 456–457 (2008).
8. Brown, A. E. X., Litvinov, R. I., Discher, D. E. & Weisel, J. W. Forced
unfolding of the coiled-coils of fibrinogen by single molecule AFM.
Biophys. J. 92, 39–41 (2007).
9. Lim, B. B. C., Lee, E. H., Sotomayor, M. & Schulten, K. Molecular basis of
fibrin clot elasticity. Structure 16, 449–459 (2008).
10. Ideker, T., Galitski, T. & Hood, L. A new approach to decoding life: Systems
biology. Annu. Rev. Genom. Hum. Genet. 2, 343–372 (2001).
11. Kitano, H. Computational systems biology. Nature 420, 206–210 (2002).
12. Kim, S. & Coulombe, P. A. Intermediate filament scaffolds fulfill mechanical,
organizational, and signaling functions in the cytoplasm. Genes Dev.
21, 1581–1597 (2007).
13. Kreplak, L. & Fudge, D. Biomechanical properties of intermediate filaments:
from tissues to single filaments and back. BioEssays 29, 26–35 (2007).
14. Herrmann, H. & Aebi, U. Intermediate filaments: Molecular structure,
assembly mechanism, and integration into functionally distinct intracellular
scaffolds. Annu. Rev. Biochem. 73, 749–789 (2004).
15. Strelkov, S. V., Herrmann, H. & Aebi, U. Molecular architecture of
intermediate filaments. BioEssays 25, 243–251 (2003).
16. Aebi, U., Cohn, J., Buhle, L. & Gerace, L. The nuclear lamina is a meshwork of
intermediate-type filaments. Nature 323, 560–564 (1986).
17. Rowat, A. C., Lammerding, J., Herrmann, H. & Aebi, U. Towards an integrated
understanding of the structure and mechanics of the cell nucleus. BioEssays
30, 226–236 (2008).
18. Vaziri, A. & Gopinath, A. Cell and biomolecular mechanics in silico.
Nature Mater. (2007).
19. Vaziri, A. & Mofrad, M. R. K. Mechanics and deformation of the
nucleus in micropipette aspiration experiment. J. Biomech.
40, 2053–2062 (2007).
186 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
Nature materials doi: 10.1038/nmat2387
20. Lim, C. T., Zhou, E. H., Li, A., Vedula, S. R. K. & Fu, H. X. Experimental
techniques for single cell and single molecule biomechanics.
Mater. Sci. Eng. C: Biomim. Supramol. Syst. 26, 1278–1288 (2006).
21. Wilson, K. L., Zastrow, M. S. & Lee, K. K. Lamins and disease: Insights into
nuclear infrastructure. Cell 104, 647–650 (2001).
22. Dahl, K. N. et al. Distinct structural and mechanical properties of the nuclear
lamina in Hutchinson–Gilford progeria syndrome. Proc. Natl Acad. Sci. USA
103, 10271–10276 (2006).
23. Ingber, D. E. Cellular mechanotransduction: putting all the pieces together
again. FASEB J. 20, 811–827 (2006).
24. Omary, M. B., Coulombe, P. A. & McLean, W. H. I. Mechanisms of disease:
Intermediate filament proteins and their associated diseases. N. Engl. J. Med.
351, 2087–2100 (2004).
25. Hulmes, D. J. S., Wess, T. J., Prockop, D. J. & Fratzl, P. Radial packing, order,
and disorder in collagen fibrils. Biophys. J. 68, 1661–1670 (1995).
26. Sasaki, N. & Odajima, S. Elongation mechanism of collagen fibrils and force-
strain relations of tendon at each level of structural hierarchy. J. Biomech.
29, 1131–1136 (1996).
27. Orgel, J. P. R. O., Irving, T. C., Miller, A. & Wess, T. J. Microfibrillar
structure of type I collagen in situ. Proc. Natl Acad. Sci. USA
103, 9001–9005 (1995).
28. Weiner, S. & Wagner, H. D. The material bone: Structure mechanical function
relations. Annu. Rev. Mater. Sci. 28, 271–298 (1998).
29. Nalla, R. K., Kinney, J. H. & Ritchie, R. O. Mechanistic fracture criteria for the
failure of human cortical bone. Nature Mater. 2, 164–168 (2003).
30. Ramachandran, G. N. & Kartha, G. Structure of collagen. Nature
176, 593–595 (1955).
31. Buehler, M. J. & Wong, S. Y. Entropic elasticity controls nanomechanics of
single tropocollagen molecules. Biophys. J. 93, 37–43 (2007).
32. Gupta, H. S. et al. Cooperative deformation of mineral and collagen in bone at
the nanoscale. Proc. Natl Acad. Sci. USA 103, 17741–17746 (2006).
33. Gupta, H. S. et al. Evidence for an elementary process in bone plasticity with
an activation enthalpy of 1 eV. J. R. Soc. Interf. 4, 277–282 (2007).
34. Buehler, M. J. Nature designs tough collagen: Explaining the nanostructure of
collagen fibrils. Proc. Natl Acad. Sci. USA 103, 12285–12290 (2006).
35. Gao, H., Ji, B., Jäger, I. L., Arzt, E. & Fratzl, P. Materials become insensitive
to flaws at nanoscale: Lessons from nature. Proc. Natl Acad. Sci. USA
100, 5597–5600 (2003).
36. Taylor, D., Hazenberg, J. G. & Lee, T. C. Living with cracks: Damage and repair
in human bone. Nature Mater. 6, 263–266 (2007).
37. Janmey, P. A., Leterrier, J. F. & Herrmann, H. Assembly and structure of
neurofilaments. Curr. Opin. Colloid Interf. Sci. 8, 40–47 (2003).
38. Bini, E., Knight, D. P. & Kaplan, D. L. Mapping domain structures in
silks from insects and spiders related to protein assembly. J. Mol. Biol.
335, 27–40 (2004).
39. Rammensee, S., Slotta, U., Scheibel, T. & Bausch, A. R. Assembly
mechanism of recombinant spider silk proteins. Proc. Natl Acad. Sci. USA
105, 6590–6595 (2008).
40. Petka, W. A., Harden, J. L., McGrath, K. P., Wirtz, D. & Tirrell, D. A.
Reversible hydrogels from self-assembling artificial proteins. Science
281, 389–392 (1998).
41. Zhao, X. J. & Zhang, S. G. Molecular designer self-assembling peptides.
Chem. Soc. Rev. 35, 1105–1110 (2006).
42. Langer, R. & Tirrell, D. A. Designing materials for biology and medicine.
Nature 428, 487–492 (2004).
43. Smeenk, J. M. et al. Controlled assembly of macromolecular beta-sheet fibrils.
Angew. Chem. Int. Edn 44, 1968–1971 (2005).
44. Mershin, A., Cook, B., Kaiser, L. & Zhang, S. G. A classic assembly of
nanobiomaterials. Nature Biotechnol. 23, 1379–1380 (2005).
45. van Hest, J. C. M. & Tirrell, D. A. Protein-based materials, toward a new level
of structural control. Chem. Commun. 1897–1904 (2001).
46. Zhang, S. G., Lockshin, C., Cook, R. & Rich, A. Unusually stable beta-
sheet formation in an ionic self-complementary oligopeptide. Biopolymers
34, 663–672 (1994).
47. Kreplak, L., Aebi, U. & Herrmann, H. Molecular mechanisms underlying the
assembly of intermediate filaments. Exp. Cell Res. 301, 77–83 (2004).
48. Strelkov, S. V., Schumacher, J., Burkhard, P., Aebi, U. & Herrmann, H. Crystal
structure of the human lamin a coil 2B dimer: Implications for the head‑to‑tail
association of nuclear lamins. J. Mol. Biol. 343, 1067–1080 (2004).
49. Schietke, R. et al. Mutations in vimentin disrupt the cytoskeleton in fibroblasts
and delay execution of apoptosis. Eur. J. Cell Biol. 85, 1–10 (2006).
50. Sokolova, A. V. et al. Monitoring intermediate filament assembly by
small-angle X-ray scattering reveals the molecular architecture of assembly
intermediates. Proc. Natl Acad. Sci. USA 103, 16206–16211 (2006).
51. Alon, U. Simplicity in biology. Nature 446, 497 (2007).
52. Csete, M. E. & Doyle, J. C. Reverse engineering of biological complexity.
Science 295, 1664–1669 (2002).
Nature materials doi: 10.1038/nmat2387
53. Ackbarow, T. & Buehler, M. J. Hierarchical coexistence of universality and
diversity controls robustness and multi-functionality in protein materials
Theor. Comput. Nanosci. 5, 1193–1204 (2008).
54. Ferrer, J. M. et al. Measuring molecular rupture forces between single
actin filaments and actin-binding proteins. Proc. Natl Acad. Sci. USA
105, 9221–9226 (2008).
55. Ackbarow, T., Chen, X., Keten, S. & Buehler, M. J. Hierarchies, multiple energy
barriers and robustness govern the fracture mechanics of alpha-helical and
beta-sheet protein domains. Proc. Natl Acad. Sci. USA 104, 16410–16415 (2007).
56. Buehler, M. J., Keten, S. & Ackbarow, T. Theoretical and computational
hierarchical nanomechanics of protein materials: Deformation and fracture.
Prog. Mater. Sci. 53, 1101–1241 (2008 ).
57. Astbury, W. T. & Street, A. X-ray studies of the structures of hair, wool and
related fibres. I. General. Phil. Trans. R. Soc. A 230, 75–101 (1931).
58. Mittermaier, A. & Kay, L. E. Review: New tools provide new insights in NMR
studies of protein dynamics. Science 312, 224–228 (2006).
59. Gupta, H. S. et al. Synchrotron diffraction study of deformation mechanisms
in mineralized tendon. Phys. Rev. Lett. 93 (2004).
60. Bernstein, F. C. et al. Protein Data Bank: Computer-based archival file for
macromolecular structures. J. Mol. Biol. 112, 535–542 (1977).
61. Le Gros, M. A., McDermott, G. & Larabell, C. A. X-ray tomography of whole
cells. Curr. Opin. Struct. Biol. 15, 593–600 (2005).
62. Jackson, M. & Mantsch, H. H. The use and misuse of FTIR spectroscopy
in the determination of protein-structure. Crit. Rev. Biochem. Molec. Biol.
30, 95–120 (1995).
63. Haris, P. I. & Chapman, D. The conformational analysis of peptides using
Fourier-transform IR spectroscopy. Biopolymers 37, 251–263 (1995).
64. Prater, C. B., Butt, H. J. & Hansma, P. K. Atomic force microscopy. Nature
345, 839–840 (1990).
65. Fisher, T. E., Oberhauser, A. F., Carrion-Vazquez, M., Marszalek, P. E. &
Fernandez, J. M. The study of protein mechanics with the atomic force
microscope. Trends Biochem. Sci. 24, 379–384 (1999).
66. Oberhauser, A. F., Badilla-Fernandez, C., Carrion-Vazquez, M. & Fernandez,
J. M. The mechanical hierarchies of fibronectin observed with single-molecule
AFM. J. Mol. Biol. 319, 433–447 (2002).
67. Rief, M., Gautel, M., Oesterhelt, F., Fernandez, J. M. & Gaub, H. E. Reversible
unfolding of individual titin immunoglobulin domains by AFM. Science
276, 1109–1112 (1997).
68. Mostaert, A. S., Higgins, M. J., Fukuma, T., Rindi, F. & Jarvis, S. P. Nanoscale
mechanical characterisation of amyloid fibrils discovered in a natural
adhesive. J. Biol. Phys. 32, 393–401 (2006).
69. Marszalek, P. E. et al. Mechanical unfolding intermediates in titin modules.
Nature 402, 100–103 (1999).
70. Sotomayor, M. & Schulten, K. Single-molecule experiments in vitro and in
silico. Science 316, 1144–1148 (2007).
71. Lim, C. T., Dao, M., Suresh, S., Sow, C. H. & Chew, K. T. Large deformation of
living cells using laser traps. Acta Mater. 52, 1837–1845 (2004).
72. Hassenkam, T. et al. High-resolution AFM imaging of intact and fractured
trabecular bone. Bone 35, 4–10 (2004).
73. Shao, Z. F., Mou, J., Czajkowsky, D. M., Yang, J. & Yuan, J. Y. Biological
atomic force microscopy: What is achieved and what is needed.
Adv. Phys. 45, 1–86 (1996).
74. Fisher, T. E., Marszalek, P. E. & Fernandez, J. M. Stretching single
molecules into novel conformations using the atomic force microscope.
Nature Struct. Biol. 7, 719–724 (2000).
75. Kadler, K. E., Holmes, D. F., Graham, H. & Starborg, T. Tip-mediated fusion
involving unipolar collagen fibrils accounts for rapid fibril elongation, the
occurrence of fibrillar branched networks in skin and the paucity of collagen
fibril ends in vertebrates. Matrix Biol. 19, 359–365 (2000).
76. Rainey, J. K., Wen, C. K. & Goh, M. C. Hierarchical assembly and the onset of
banding in fibrous long spacing collagen revealed by atomic force microscopy.
Matrix Biol. 21, 647–660 (2002).
77. Fantner, G. E. et al. Sacrificial bonds and hidden length dissipate energy
as mineralized fibrils separate during bone fracture. Nature Mater.
4, 612–616 (2005).
78. Johnson, C. P., Tang, H. Y., Carag, C., Speicher, D. W. & Discher, D. E. Forced
unfolding of proteins within cells. Science 317, 663–666 (2007).
79. Oliver, W. C. & Pharr, G. M. Measurement of hardness and elastic modulus
by instrumented indentation: Advances in understanding and refinements to
methodology. J. Mater. Res. 19, 3–20 (2004).
80. Thompson, J. B. et al. Bone indentation recovery time correlates with bond
reforming time. Nature 414, 773–776 (2001).
81. Tai, K., Dao, M., Suresh, S., Palazoglu, A. & Ortiz, C. Nanoscale heterogeneity
promotes energy dissipation in bone. Nature Mater. 6, 454–462 (2007).
82. Eppell, S. J., Smith, B. N., Kahn, H. & Ballarini, R. Nano measurements
with micro-devices: mechanical properties of hydrated collagen fibrils.
J. R. Soc. Interf. 3, 117–121 (2006).
nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 187
review article
83. Karplus, M. & McCammon, J. A. Molecular dynamics simulations of
biomolecules. Nature Struct. Biol. 9, 646–652 (2002).
84. Wang, W., Donini, O., Reyes, C. M. & Kollman, P. A. Biomolecular
simulations: Recent developments in force fields, simulations of
enzyme catalysis, protein–ligand, protein–protein, and protein–nucleic
acid noncovalent interactions. Annu. Rev. Biophys. Biomol. Struct.
30, 211–243 (2001).
85. Mackerell, A. D. Empirical force fields for biological macromolecules:
Overview and issues. J. Comput. Chem. 25, 1584–1604 (2004).
86. Gautieri, A., Buehler, M. J. & Redaelli, A. Deformation rate controls elasticity
and unfolding pathway of single tropocollagen molecules. J. Mech. Behavior
Biomed. Mater. 2, 130–137 (2009).
87. Bell, G. I. Models for specific adhesion of cells to cells. Science
200, 618–627 (1978).
88. Kramers, H. A. Brownian motion in a field of force and the diffusion model of
chemical reactions. Physica 7, 10 (1940).
89. Hanggi, P., Talkner, P. & Borkovec, M. Reaction-rate theory: fifty years after
Kramers. Rev. Mod. Phys. 62, 251–341 (1990).
90. Zhurkov, S. N. Kinetic concept of the strength of solids. Int. J. Fracture Mech.
1, 311–323 (1965).
91. Evans, E. Probing the relation between force, lifetime, and chemistry in single
molecular bonds. Annu. Rev. Biophys. Biomol. Struct. 30, 105–128 (2001).
92. Merkel, R., Nassoy, P., Leung, A., Ritchie, K. & Evans, E. Energy landscapes
of receptor–ligand bonds explored with dynamic force spectroscopy. Nature
379, 50–53 (1999).
93. Evans, E. & Ritchie, K. Dynamic strength of molecular adhesion bonds.
Biophys. J. 72, 1541–1555 (1997).
94. Dudko, O. K., Hummer, G. & Szabo, A. Intrinsic rates and activation free
energies from single-molecule pulling experiments. Phys. Rev. Lett. 96 (2006).
95. Hummer, G. & Szabo, A. Kinetics from nonequilibrium single-molecule
pulling experiments. Biophys. J. 85, 5–15 (2003).
96. Hummer, G. & Szabo, A. Free energy reconstruction from nonequilibrium
single-molecule pulling experiments. Proc. Natl Acad. Sci. USA
98, 3658–3661 (2001).
97. Snow, C. D., Sorin, E. J., Rhee, Y. M. & Pande, V. S. How well can simulation
predict protein folding kinetics and thermodynamics? Annu. Rev. Biophys.
Biomol. Struct. 34, 43–69 (2005).
98. Dietz, H. & Rief, M. Elastic bond network model for protein unfolding
mechanics. Phys. Rev. Lett. 100, 098101 (2008).
99. Seifert, U. Rupture of multiple parallel molecular bonds under dynamic
loading. Phys. Rev. Lett. 84, 2750–2753 (2000).
100. Erdmann, T. & Schwarz, U. S. Stability of adhesion clusters under constant
force. Phys. Rev. Lett. 92, 4 (2004).
101. Dietz, H., Berkemeier, F., Bertz, M. & Rief, M. Anisotropic deformation
response of single protein molecules. Proc. Natl Acad. Sci. USA
103, 12724–12728 (2006).
102. Bustamante, C., Marko, J. F., Siggia, E. D. & Smith, S. Entropic elasticity of
lambda-phage DNA. Science 265, 1599–1600 (1994).
103. Sun, Y. L., Luo, Z. P., Fertala, A. & An, K. N. Stretching type II collagen with
optical tweezers. J. Biomech. 37, 1665–1669 (2004).
104. Rief, M., Fernandez, J. M. & Gaub, H. E. Elastically coupled two-level systems
as a model for biopolymer extensibility. Phys. Rev. Lett. 81, 4764–4767 (1998).
105. Keten, S. & Buehler, M. J. Strength limit of entropic elasticity in beta-sheet
protein domains. Phys. Rev. E 78, 061913 (2008).
106. Mesquida, P., Riener, C. K., MacPhee, C. E. & McKendry, R. A.
Morphology and mechanical stability of amyloid-like peptide fibrils.
J. Mater. Sci. Mater. Med. 18, 1325–1331 (2007).
107. Iconomidou, V. A. & Hamodrakas, S. J. Natural protective amyloids.
Curr. Protein Pept. Sci. 9, 291–309 (2008).
108. Hardy, J. & Selkoe, D. J. Medicine: The amyloid hypothesis of Alzheimer’s
disease. Progress and problems on the road to therapeutics. Science
297, 353–356 (2002).
109. Selkoe, D. J. Alzheimer’s disease: Genes, proteins, and therapy. Physiol. Rev.
81, 741–766 (2001).
110. Knowles, T. P. et al. Role of intermolecular forces in defining material
properties of protein nanofibrils. Science 318, 1900–1903 (2007).
111. Smith, J. F., Knowles, T. P. J., Dobson, C. M., MacPhee, C. E. & Welland, M. E.
Characterization of the nanoscale properties of individual amyloid fibrils.
Proc. Natl Acad. Sci. USA 103, 15806–15811 (2006).
112. Dutt, A., Drew, M. G. B. & Pramanik, A. Beta-sheet mediated self-assembly of
dipeptides of Omega-amino acids and remarkable fibrillation in the solid state.
Org. Biomol. Chem. 3, 2250 (2005).
113. Burkoth, T. S. et al. Structure of the beta-amyloid((10–35)) fibril.
J. Am. Chem. Soc. 122, 7883–7889 (2000).
114. Chiti, F. & Dobson, C. M. Protein misfolding, functional amyloid, and human
disease. Annu. Rev. Biochem. 75, 333–366 (2006).
115. Dobson, C. M. Protein folding and misfolding. Nature 426, 884–890 (2003).
review article
116. Nguyen, H. D. & Hall, C. K. Molecular dynamics simulations of spontaneous fibril
formation by random-coil peptides. Proc. Natl Acad. Sci. USA 101, 16180 (2004).
117. Hwang, W., Zhang, S. G., Kamm, R. D. & Karplus, M. Kinetic control of dimer
structure formation in amyloid fibrillogenesis. Proc. Natl Acad. Sci. USA
101, 12916–12921 (2004).
118. Gruenbaum, Y., Margalit, A., Goldman, R. D., Shumaker, D. K. & Wilson, K. L.
The nuclear lamina comes of age. Nature Rev. Mol. Cell Biol. 6, 21–31 (2005).
119. Burke, B. & Stewart, C. L. Life at the edge: The nuclear envelope and human
disease. Nature Rev. Mol. Cell Biol. 3, 575–585 (2002).
120. Lammerding, J. et al. Lamin A/C deficiency causes defective nuclear
mechanics and mechanotransduction. J. Clin. Invest. 113, 370–378 (2004).
121. Prockop, D. J. & Kivirikko, K. I. Collagens: molecular biology, diseases, and
potentials for therapy. Annu. Rev. Biochem. 64, 403–434 (1995).
122. Byers, P. H., Wallis, G. A. & Willing, M. C. Osteogenesis imperfecta:
translation of mutation to phenotype. J. Med. Genet. 28, 433–442 (1991).
123. Gautieri, A., Vesentini, S., Redaelli, A. & Buehler, M. J. Single molecule effects
of osteogenesis imperfecta mutations in tropocollagen protein domains.
Protein Sci. 18, 161 - 168 (2009).
124. McBride, D. J., Choe, V., Shapiro, J. R. & Brodsky, B. Altered collagen structure
in mouse tail tendon lacking the α2(I) chain. J. Mol. Biol. 270, 275–284 (1997).
125. Miller, A., Delos, D., Baldini, T., Wright, T. M. & Camacho, N. P. Abnormal
mineral-matrix interactions are a significant contributor to fragility in
oim/oim bone. Calcif. Tissue Int. 81, 206–214 (2007).
126. Sims, T. J., Miles, C. A., Bailey, A. J. & Camacho, N. P. Properties of collagen in
OIM mouse tissues. Connect. Tissue Res. 44, 202–205 (2003).
127. Grabner, B. et al. Age- and genotype-dependence of bone material properties
in the osteogenesis imperfecta murine model (oim). Bone 29, 453–457 (2001).
128. Camacho, N. P. et al. The material basis for reduced mechanical properties in
oim bones. J. Bone Mineral. Res. 14, 264–272 (1999).
129. Fratzl, P., Paris, O., Klaushofer, K. & Landis, W. J. Bone mineralization in an
osteogenesis imperfecta mouse model studied by small-angle X-ray scattering.
J. Clin. Invest. 97, 396–402 (1996).
130. Rauch, F. & Glorieux, F. H. Osteogenesis imperfecta. Lancet
363, 1377–1385 (2004).
131. Miller, E., Delos, D., Baldini, T., Wright, T. M. & Camacho, N. P. Abnormal
mineral–matrix interactions are a significant contributor to fragility in
oim/oim bone. Calcif. Tissue Int. 81, 206–214 (2007).
132. Misof, K., Landis, W. J., Klaushofer, K. & Fratzl, P. Collagen from the
osteogenesis imperfecta mouse model (oim) shows reduced resistance against
tensile stress. J. Clin. Invest. 100, 40–45 (1997).
188 nature materials | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
Nature materials doi: 10.1038/nmat2387
133. Chavassieux, P., Seeman, E. & Delmas, P. D. Insights into material and
structural basis of bone fragility from diseases associated with fractures: How
determinants of the biomechanical properties of bone are compromised by
disease. Endocrine Rev. 28, 151–164 (2007).
134. Hudson, B. G., Tryggvason, K., Sundaramoorthy, M. & Neilson, E. G. Alport’s
syndrome, Goodpasture’s syndrome, and type IV collagen. N. Engl. J. Med.
348, 2543–2556 (2003).
135. Guck, J. et al. Optical deformability as an inherent cell marker for
testing malignant transformation and metastatic competence. Biophys. J.
88, 3689–3698 (2005).
136. Cross, S. E., Jin, Y.‑S., Rao, J. & Gimzewski, J. K. Nanomechanical analysis of
cells from cancer patients. Nature Nanotech. 2, 780–783 (2007).
137. Dao, M., Lim, C. T. & Suresh, S. Mechanics of the human red blood cell
deformed by optical tweezers. J. Mech. Phys. Solids 51, 2259–2280 (2003).
138. Engler, A. J., Sen, S., Sweeney, H. L. & Discher, D. E. Matrix elasticity directs
stem cell lineage specification. Cell 126, 677–689 (2006).
139. Scheibel, T. et al. Conducting nanowires built by controlled self-assembly
of amyloid fibers and selective metal deposition. Proc. Natl Acad. Sci. USA
100, 4527–4532 (2003).
140. Humphrey, W., Dalke, A. & Schulten, K. VMD: Visual molecular dynamics.
J. Mol. Graphics 14, 33 (1996).
141. Herrmann, H., Bar, H., Kreplak, L., Strelkov, S. V. & Aebi, U. Intermediate
filaments: from cell architecture to nanomechanics. Nature Rev. Mol. Cell Biol.
8, 562–573 (2007).
142. Lantz, M. A. et al. Stretching the alpha-helix: a direct measure of the
hydrogen-bond energy of a single-peptide molecule. Chem. Phys. Lett.
315, 61–68 (1999).
143. www.nature.com/horizon/proteinfolding/highlights/figures/
s3_nonspec1_f1.html.
Acknowledgements
This research was supported by the Army Research Office (W911NF‑06‑1-0291),
the National Science Foundation (CAREER Grant CMMI‑0642545 and MRSEC
DMR‑0819762), the Air Force Office of Scientific Research (FA9550‑08‑1-0321),
the Office of Naval Research (N00014‑08‑1-00844) and the Defense Advanced
Research Projects Agency (DARPA) (HR0011‑08‑1-0067). M.J.B. acknowledges
support through the Esther and Harold E. Edgerton Career Development
Professorship. We thank L. Kreplak, U. Aebi, H. Herrmann, P. Fratzl, M. Rief,
H. Gaub, T. Scheibel and K. Dahl for communications.
 LETTERS
PUBLISHED ONLINE: 1 FEBRUARY 2009 DOI: 10.1038/NMAT2376

Supramolecular control of the magnetic anisotropy

in two-dimensional high-spin Fe arrays at a
metal interface
Pietro Gambardella1,2,3 *, Sebastian Stepanow1,4 , Alexandre Dmitriev4,5 , Jan Honolka4 ,
Frank M. F. de Groot6 , Magalí Lingenfelder4 , Subhra Sen Gupta7 , D. D. Sarma7 , Peter Bencok8 ,
Stefan Stanescu8 , Sylvain Clair3 , Stéphane Pons3 , Nian Lin4 , Ari P. Seitsonen9 , Harald Brune3 ,
Johannes V. Barth10 and Klaus Kern3,4

Magnetic atoms at surfaces are a rich model system for
solid-state magnetic bits exhibiting either classical1,2 or
quantum3,4 behaviour. Individual atoms, however, are difficult
to arrange in regular patterns1–5 . Moreover, their magnetic
properties are dominated by interaction with the substrate,
which, as in the case of Kondo systems, often leads to
a decrease or quench of their local magnetic moment6,7 .
Here, we show that the supramolecular assembly of Fe and
1,4-benzenedicarboxylic acid molecules on a Cu surface results
in ordered arrays of high-spin mononuclear Fe centres on a
1.5 nm square grid. Lateral coordination with the molecular
ligands yields unsaturated yet stable coordination bonds,
which enable chemical modification of the electronic and
magnetic properties of the Fe atoms independently from the
substrate. The easy magnetization direction of the Fe centres
can be switched by oxygen adsorption, thus opening a way to
control the magnetic anisotropy in supramolecular layers akin
to that used in metallic thin films8–11 .
The control of magnetic anisotropy is a key issue in the
development of molecule–metal interfaces for magnetic applica-
tions, both at the single-molecule12 and extended-film level13 . In
metallic multilayers used as storage media or spin-valve devices at
present, tuning of the magnetic anisotropy is achieved either by a
careful choice of the overlayer/substrate composition and thickness
or by oxidation of the magnetic elements11,14,15 . Recent studies
showed that the magnetization direction of surface-supported para-
magnetic molecules can be controlled through exchange coupling
with a magnetic film, which provides robust ferromagnetic proper-
ties but does not enable each molecule to be switched independently
from the substrate or its neighbours16,17 . Alternatively, theoret-
ical work suggested that the sign of magnetic anisotropy could
be reversed in metal–organic complexes by exploiting oxidation
processes that affect the hybridization of molecular orbitals with
metal states carrying non-zero orbital magnetization18 .
Here, we investigate supramolecular self-assembly on a
non-magnetic Cu surface as a means to produce two-dimensional
(2D) arrays of regularly spaced Fe spins, the magnitude and
magnetic anisotropy of which are manipulated by lateral and
axial molecular ligands. Although Fe atoms adsorbed on Cu(100)
constitute a metallic system owing to the substantial overlap
of the Fe 3d and 4s states with the Cu bands, and exhibit a
Kondo temperature of about 55 K (ref. 7), X-ray absorption
spectroscopy (XAS) and multiplet calculations shows that Fe–Cu
hybridization is drastically weakened by square-planar coordina-
tion to 1,4-benzenedicarboxylate molecules (hereafter indicated as
terephthalate acid or TPA), leading to a high-spin configuration
with strongly localized Fe2+ character. X-ray magnetic circular
dichroism (XMCD) measurements confirm the presence of sizeable
spin moments at the Fe sites, together with a large orbital moment
and in-plane magnetic anisotropy that originate from spin–orbit-
induced mixing of ground and first excited molecular states. The
Fe centres in the supramolecular arrays act as selective adsorption
sites for O2 , which produces an out-of-plane reorientation of
the easy magnetization axis at the Fe sites. Furthermore, the
saturation of the orbital magnetic moments of TPA-coordinated
Fe is found to be strongly anisotropic, revealing an unusual feature
of surface-supported magnetic systems.
The co-deposition of transition-metal ions and organic ligands
on crystalline surfaces offers the potential to design supramolecular
grids with programmable structural and chemical features19,20 ,
where the interaction with the substrate is used to stabilize a
planar geometry21,22 . Fex (TPA)y complexes on Cu(100) constitute
a prototypical 2D hetero-assembled system forming a variety of
mono- and bi-nuclear network structures, in which the morphology
is determined by the Fe:TPA stoichiometry, substrate symmetry
and annealing temperature22–24 . Figure 1a shows a scanning
tunnelling microscopy image of the square-symmetry array of self-
assembled mononuclear Fe(TPA)4 complexes investigated here.
Each Fe atom is coordinated to four TPA molecules through
Fe–carboxylate bonds, with the supramolecular Fe(TPA)4 units
organized in a (6 × 6) unit cell with respect to the underlying
Cu lattice (Fig. 1c). Weak hydrogen-bonding interactions between

1 Centre d’Investigacions en Nanociència i Nanotecnologia (ICN-CSIC), UAB Campus, E-08193 Barcelona, Spain, 2 Institució Catalana de Recerca i Estudis
Avançats (ICREA), E-08010 Barcelona, Spain, 3 Institut de Physique des Nanostructures, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015
Lausanne, Switzerland, 4 Max-Planck-Institut für Festkörperforschung, D-70569 Stuttgart, Germany, 5 Department of Applied Physics, Chalmers University
of Technology, 41296 Göteborg, Sweden, 6 Department of Chemistry, Utrecht University, 3584 CA Utrecht, The Netherlands, 7 Solid State and Structural
Chemistry Unit, Indian Institute of Science, Bangalore 560 012, India, 8 European Synchrotron Radiation Facility, BP 200, F-38043 Grenoble, France,
9 Institut de Minéralogie et de Physique des Milieux Condensé, Université Pierre et Marie Curie, F-75252 Paris, France, 10 Physik-Department E20,

Technische Universität München, D-85748 Garching, Germany. *e-mail: pietro.gambardella@icrea.es.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 189

LETTERS
a b
c d
Figure 1 | Planar supramolecular layers of Fe–TPA complexes
self-assembled on Cu(100). a, Fe(TPA)4 array; blue dots indicate the
position of Fe atoms. Each Fe centre is coordinated to four deprotonated
oxygen ligands of the surrounding TPA molecules. STM image size
62 × 62 Å2 . b, Selective uptake of O2 by Fe; a change of STM contrast is
observed for O2 –Fe(TPA)4 (red circle) at the Fe location compared with
Fe(TPA)4 (blue circle). Image size 84 × 84 Å2 . c, Top-view model of
Fe(TPA)4 coordination and supporting Cu(100) substrate. d, Relaxed
Fe(TPA)4 and O2 –Fe(TPA)4 supramolecular structures obtained by DFT.
Fe (blue), O (red), C (black) and H (white) atoms are shown together with
the spin density distribution (green).
the complexes favour long-range order extending over entire
terraces of the substrate23 (see Supplementary Information, Fig. S1).
The resultant superlattice of individual Fe centres has perfect
15 × 15 Å2 periodicity. Despite the four-fold lateral coordination
to the carboxylate ligands and the residual interaction with
Cu, Fe centres are chemically active, forming an array of
coordinatively unsaturated sites. On exposure to O2 , scanning
tunnelling microscopy images indicate that selective adsorption of
oxygen takes place close to Fe on-top positions, as demonstrated
by the change of contrast in Fig. 1b. The structural properties
of both Fe(TPA)4 and O2 –Fe(TPA)4 complexes as well as their
interaction with the substrate were further characterized using
density functional theory (DFT) calculations, as described in
the Methods section. The relaxed supramolecular structures are
shown in Fig. 1d together with an isocontour of the spin density
distribution; the optimal O2 –Fe(TPA)4 geometry corresponds to O2
chemisorbed on Fe with the O–O axis parallel to the surface plane,
in agreement with the angular dependence of the oxygen XAS (see
Supplementary Information, Fig. S4).
L-edge X-ray absorption spectra provide a unique probe of
the bonding and magnetic properties of transition-metal ions25 ,
with the extra advantage of extreme sensitivity to surface-dilute
systems26 . This is of crucial importance for the present study, as the
intensity, lineshape and polarization dependence of dipole-allowed
2p → 3d transitions offer information on the spin state, oxidation
state and ligand field of Fe ions in an element-selective way. Figure 2
shows the XAS and XMCD spectra of 0.025 monolayers of Fe
deposited on Cu(100) at room temperature (Fig. 2a), equivalent
to the Fe content in the supramolecular arrays, Fe(TPA)4 (Fig. 2b)
and O2 –Fe(TPA)4 (Fig. 2c). The XAS intensity, shown after Cu
background subtraction, was measured at 8 K with magnetic field
190
NATURE MATERIALS DOI: 10.1038/NMAT2376
B = 6 T applied out-of-plane (θ = 0◦ ) and at grazing incidence
(θ = 70◦ ) for parallel (I + ) and antiparallel (I − ) alignment of the
photon helicity with B. More details about the measurements are
given in Supplementary Information. The comparison of Fig. 2a–c
reveals telltale differences in the XAS lineshape: whereas Fe/Cu(100)
exhibits broad L3 and L2 peaks typical of a metal, pronounced
narrowing of the spectral features and a well-defined multiplet
fine structure is observed for the Fe–TPA complexes, similar to
the spectra of Fe ions in bulk metal–organic compounds27 . These
changes, related to ligand-induced modifications and localization
of the Fe d-orbitals, indicate that true coordination bonds have
been formed between the Fe centres and carboxylate ligands, with
partial decoupling of Fe from the metal substrate. Interestingly,
electron localization effects are found to be more pronounced in
self-assembled Fe(TPA)4 compared with chemically synthesized
Fe–porphyrin molecules evaporated on metal films17 . We find
that the XAS (I + , I − ) and the XMCD (I + − I − ) spectral shape
depend markedly on the orientation of the beam with respect to
the molecular arrays. This is in essence a linear dichroism effect25
related to the anisotropic spatial distribution of the Fe d-electrons in
a low-symmetry environment that cannot be observed in bulk-like
molecular powder studies27 , but is resolved here owing to the planar
alignment of the complexes. Both the large XAS L3 /L2 branching
ratio and the strong XMCD intensity of Fe(TPA)4 and O2 –Fe(TPA)4
point towards a high-spin Fe configuration for the complexes,
against the tendency of the Cu electrons to reduce the Fe moment
through the Kondo interaction7 .
To gain more insight into the electronic and magnetic properties
of the Fe centres, we carried out XAS and XMCD simulations
using ligand-field multiplet theory with the inclusion of charge-
transfer effects25 . The molecular environment was modelled by a
crystal-field potential defined by the parameters Dq, Ds and Dt , as
appropriate for C4v symmetry. Covalent mixing of the metal valence
d-orbitals with the ligand p-orbitals was simulated using a valence
bond configuration interaction scheme, whereby the ground state
of Fe is taken to be a linear combination of 3d n and 3d n+1 L
configurations, with L indicating a ligand hole. We tested different
combinations of d 5 , d 6 and d 7 configurations, corresponding to
Fe3+ , Fe2+ and Fe1+ , respectively. The simulations show that the
ground state of Fe in both Fe(TPA)4 and O2 –Fe(TPA)4 complexes
possesses almost pure d 6 character, with maximum 14% and 15%
d 7 L weight, respectively, depending on the hopping parameters
assumed in the calculations (see Supplementary Information,
Table S1). Within this admixture range, the calculated spectra show
only minor differences and are in very good agreement with the
experimental data, as shown in Fig. 2d,e. This result has important
consequences as (1) it proves that the Fe d-orbitals are interacting
mainly with the ligands and only weakly perturbed by the metallic
substrate (not so for the s-states, see Supplementary Information),
(2) it shows that Fe(TPA)4 and O2 –Fe(TPA)4 constitute high-spin
Fe2+ complexes, which are expected to exhibit zero-field splitting
and anisotropic g -factors, that is, strong magnetic anisotropy
properties in the language of molecular magnetism. In contrast,
high-spin Fe3+ , with a half-filled d-shell, would have only
minor orbital contributions to the magnetization and very small
magnetic anisotropy.
Consistently with the geometry of the O ligands, we find that
the crystal-field splitting patterns (Fig. 3c,d) are close to square
planar and square pyramidal for Fe(TPA)4 and O2 –Fe(TPA)4 ,
respectively, whereas the strong tetragonal distortion that charac-
terizes Fe(TPA)4 reduces significantly in O2 –Fe(TPA)4 owing to
the presence of the apical ligand (see Supplementary Information,
Table S1). DFT calculations further show that the total number
of Fe valence electrons changes from 6.67 to 6.33 following O2
chemisorption, but that the population of the 3d-states remains
essentially stable (see Supplementary Information, Table S2), in
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2376 LETTERS
a b c
I+ 1
1 I¬
1

θ = 0° θ = 0° θ = 0°
XAS intensity (arb. units)

XAS intensity (arb. units)

XAS intensity (arb. units)

0
0 0
I+ ¬ I¬

I+ 1
1 I¬ 1

θ = 70° θ = 70° θ = 70°

0
0 0
I+ ¬ I¬

705 710 715 720 725 730 705 710 715 720 725 730 705 710 715 720 725 730
Photon energy (eV) Photon energy (eV) Photon energy (eV)

d e
Experiment
Calculation

I+ + I¬0°
I+ + I¬0°
XAS intensity (arb. units)

XAS intensity (arb. units)

I+ + I¬70°

I+ + I¬70°
I+ ¬ I¬0°

I+ ¬ I¬0°

705 710 715 720 725 730 705 710 715 720 725 730
Photon energy (eV) Photon energy (eV)

Figure 2 | Circularly polarized X-ray absorption spectra of Fe(TPA)4 , O2 –Fe(TPA)4 and Fe/Cu(100) measured at the Fe L2,3 -edge and calculated
multiplet structure. a–c, XAS intensity of 0.025 monolayer Fe/Cu(100) (a), Fe(TPA)4 (b) and O2 –Fe(TPA)4 (c) for parallel (I+ ) and antiparallel (I− )
orientation of the photon helicity with field-induced magnetization. Spectra were recorded in the electron-yield mode at T = 8 K and B = 6 T at normal
(θ = 0◦ ) and grazing incidence (θ = 70◦ ). The XMCD signal (I+ –I− ) is shown for each orientation. d,e, Calculated XAS and XMCD spectra of Fe(TPA)4 (d)
and O2 –Fe(TPA)4 (e). See text for details.

agreement with the ligand-field simulations. There are 0.8 excess axis out-of-plane. The field dependence of the XMCD signal in
electrons localized on the O2 ligand, the largest fraction of which Fig. 3a,b underlines the strong change in magnetic anisotropy
is supplied by the Cu substrate acting as a charge reservoir, a induced by O2 . This effect is comparable to that reported for
behaviour that is specific to surface-supported metal–organic sys- ultrathin metal films9–11 , but opposite in sign with respect to
tems. Notably, the formation of the supramolecular complexes oxygen-dosed Fe layers on Cu(100) (ref. 11). With respect to bulk
is accompanied by a substantial increase of the Fe–substrate dis- molecular crystals, the planar and open coordination structure
tance, calculated as 2.32 Å for individual Fe atoms on Cu(100), of the self-assembled Fe array makes such a system extremely
2.71 Å in Fe(TPA)4 and 3.32 Å in O2 –Fe(TPA)4 , showing together sensitive to chemisorption, providing straightforward control of
with XAS how the Fe–Cu metal interactions progressively weaken the preferred Fe spin orientation. Through the analysis of the
with increasing number of ligands. A more detailed discussion of XAS and XMCD spectra, we can further identify the cause for
the crystal-field and DFT electronic-structure results is reported in the easy-axis switch in the metal–organic complexes, revealing
Supplementary Information. a different mechanism from the one proposed by Atodiresei
Confirmation of the strong magnetic anisotropy behaviour of Fe and co-workers18 based on the control of the metal oxidation
in the supramolecular complexes was obtained by angle-dependent state. Our calculations show that only the ligand field is affected
XMCD measurements, exploiting the XMCD proportionality to by O2 adsorption, whereas the formal Fe oxidation does not
the projection of the Fe magnetic moment on the X-ray incidence change. As demonstrated by the crystal-field level schemes in
direction. The relative magnitude of the XMCD spectra at θ = 0◦ Fig. 3c,d, the axial O2 ligand induces a change of the Fe ground
and 70◦ in Fig. 2 indicates that, whereas Fe/Cu(100) has weak state from A1g to Eg . This effect can be understood also in
out-of-plane anisotropy, the Fe(TPA)4 easy axis lies in-plane. O2 an intuitive way, as the on-top O2 molecule binds axially to
adsorption at the Fe sites, however, further drives an abrupt the dz 2 Fe orbital, thus effectively pushing up the antibonding
magnetic anisotropy reorientation transition, rotating the Fe easy A1g state (that is, the term with dz 2 -like symmetry) relative to

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 191

LETTERS
a b
0.5
In-plane
0.4
XMCD (arb. units)
0.3
0.2 Out-of-plane
0.1
0 0
0 1 2 3 4 5
B (T)
c d
0.5 eV B1g(dx2 ¬ y2)
B2g(dxy)
5D
2
Eg(dxz, dyz)
A1g(dz2)
O3 Oh C4v
Figure 3 | Element-selective magnetization curves of the Fe centres and crystal-field diagrams. a,b, Fe(TPA)4 (a) and O2 –Fe(TPA)4 (b) magnetization
measured at T = 8 K with the field applied out-of-plane (θ = 0◦ , filled symbols) and close to the in-plane direction (θ = 70◦ , open symbols). The data
points represent the Fe XMCD intensity integrated over the L3 -edge and normalized by the corresponding L3 -integrated XAS intensity. c,d, Crystal-field
diagram of Fe(TPA)4 (c) and O2 –Fe(TPA)4 (d) derived from the XAS multiplet simulations.
the Eg term in O2 –Fe(TPA)4 . The change of ground state has
important consequences for the magnetic properties: before taking
the spin–orbit perturbation on the crystal-field levels into account,
the orbital angular momentum is quenched for an A1g term. In
C4v symmetry, on the other hand, the Eg term is an orbitally
degenerate doublet with non-zero orbital moment pointing along
the principal symmetry direction. The tendency of O2 –Fe(TPA)4
to magnetize out-of-plane, together with its stronger magnetic
anisotropy compared with Fe(TPA)4 , can be traced back to its
unperturbed crystal-field configuration and non-zero first-order
orbital moment perpendicular to the surface plane. In Fe(TPA)4 ,
the orbital moment arises as a second-order perturbation effect
as spin–orbit coupling admixes the nearly degenerate A1g and
Eg states28 . XMCD sum-rule measurements reflect the difference
between the two compounds, showing that O2 –Fe(TPA)4 has
30% higher orbital magnetic moment compared with Fe(TPA)4 ,
namely 0.55 ± 0.07 and 0.42 ± 0.06µB at T = 8 K, when B = 6 T
is applied out-of-plane and in-plane, that is, along the respective
easy magnetization direction. Note that the orbital magnetic
moment measured for Fe/Cu(100) is much smaller, 0.18 ± 0.03µB .
A complete discussion of the sum-rule analysis is reported in
Supplementary Information.
The ligand-field calculations yield zero-field splitting of about
−3 and +5 meV for Fe(TPA)4 and O2 –Fe(TPA)4 , respectively.
However, we caution that these parameters may be overestimated
by as much as a factor of two, as are the ligand-field predictions
for the orbital moments, owing to covalency effects that are
not treated in the ligand-field approach28 . Our data disclose an
extra unusual effect for surface magnetic systems, namely the
192
NATURE MATERIALS DOI: 10.1038/NMAT2376
0.8 Out-of-plane
XMCD (arb. units)
0.6
0.4
In-plane
0.2
6 0 1 2 3 4 5 6
B (T)
B1g(dx2 ¬ y2)
A1g(dz2)
5D
2
B2g(dxy)
Eg(dxz, dyz)
O3 Oh C4v
strong anisotropy of the saturation magnetic moments of Fe. This
effect is borne out both by the XMCD curves in Fig. 3 and the
ligand-field calculations, which reveal a large and highly anisotropic
orbital magnetization due to the predominant influence of the
low-symmetry ligand field over the Fe–metal interaction, and
confirmed by the different XMCD multiplet structure observed at
θ = 70◦ relative to 0◦ (see Supplementary Information, Fig. S2).
Finally, we note that no sign of magnetic coupling was detected
between the Fe centres. Future investigations may address this issue
using different kinds of molecular ligand to adjust the Fe–Fe spacing
and spin-dependent electron correlation.
The spin behaviour of individual atoms in purely metallic and
molecular environments has been intensively studied in recent
years, leading to the discovery of unusual magnetic1–4,12,17 and
electron transport properties29,30 . The capability to fabricate 2D
arrays of monodisperse spin centres with nanometre spacing,
and to understand and control their magnetic properties at the
interface with a metal substrate, constitutes a basic step towards
the exploitation of single-spin phenomena in heterogeneous devices
based on either a mono- or multilayer geometry.
Methods
Experiments were carried out at the ID08 beamline of the European Synchrotron
Radiation Facility in Grenoble. Fe(TPA)4 and O2 –Fe(TPA)4 layers were prepared by
sequential deposition of TPA and Fe on a clean Cu(100) surface and characterized
in situ by STM and low-energy electron diffraction before the XAS measurements.
Ultrahigh-vacuum conditions were maintained throughout the structural and XAS
characterization measurements. For further details about the sample preparation,
multiplet calculations and XMCD analysis, see Supplementary Information. The
DFT calculations were carried out using the generalized gradient approximation
as the exchange–correlation functional of the Kohn–Sham equations using the
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2376
Vienna Ab-initio Simulation Package code (http://cms.mpi.univie.ac.at/vasp/).
The electronic wavefunctions were expanded in a plane-wave basis set, up to a
cutoff energy of 37 Ryd, and the core–valence interaction was modelled with the
projected augmented-wave method. Three substrate layers (first fully relaxed
with adsorbate layer, second only laterally) together with almost 20 Å of vacuum
between the two surfaces of the slab were used. A (2 × 2) grid of Monkhorst–Pack
k-points in the first Brillouin zone was used to approximate the integration
over the reciprocal space during the atomic relaxation, and a (4 × 4) grid was
used for the analysis.
Received 17 July 2008; accepted 22 December 2008;
published online 1 February 2009
References
1. Gambardella, P. et al. Giant magnetic anisotropy of single cobalt atoms and
nanoparticles. Science 300, 1130–1133 (2003).
2. Meier, F., Zhou, L., Wiebe, J. & Wiesendanger, R. Revealing magnetic
interactions from single-atom magnetization curves. Science 320, 82–86 (2008).
3. Heinrich, A. J., Gupta, J. A., Lutz, C. P. & Eigler, D. M. Single-atom spin-flip
spectroscopy. Science 306, 466–469 (2004).
4. Hirjibehedin, C. F. et al. Large magnetic anisotropy of a single atomic spin
embedded in a surface molecular network. Science 317, 1199–1203 (2007).
5. Eigler, D. M. & Schweitzer, E. K. Positioning single atoms with a scanning
electron microscope. Nature 344, 524–526 (1990).
6. Nagaoka, T., Jamneala, T., Grobis, M. & Crommie, M. F. Temperature
dependence of a single Kondo impurity. Phys. Rev. Lett. 88, 077205 (2002).
7. Wahl, P. et al. Kondo effect of molecular complexes at surfaces: Ligand control
of the local spin coupling. Phys. Rev. Lett. 95, 166601 (2005).
8. Monso, S. et al. Crossover from in-plane to perpendicular anisotropy in
Pt/CoFe/AlOx sandwiches as a function of Al oxidation: A very accurate control
of the oxidation of tunnel barriers. Appl. Phys. Lett. 80, 4157–4159 (2002).
9. Hong, J., Wu, R. Q., Lindner, J., Kosubek, E. & Baberschke, K. Manipulation
of spin reorientation transition by oxygen surfactant growth: A combined
theoretical and experimental approach. Phys. Rev. Lett. 92, 147202 (2004).
10. Sander, D. et al. Reversible H-induced switching of the magnetic easy axis in
Ni/Cu(001) thin films. Phys. Rev. Lett. 93, 247203 (2004).
11. Peterka, D., Enders, A., Haas, G. & Kern, K. Adsorbate and thermally
induced spin reorientation transition in low-temperature-grown Fe/Cu(001).
Phys. Rev. B 66, 104411 (2002).
12. Bogani, L. & Wernsdorfer, W. Molecular spintronics using single-molecule
magnets. Nature Mater. 7, 179–186 (2008).
13. Naber, W. J. M., Faez, S. & van der Wiel, W. G. Organic spintronics. J. Phys. D
40, R205–R228 (2007).
14. Parkin, S. S. P. et al. Magnetically engineered spintronic sensors and memory.
Proc. IEEE 91, 661–680 (2003).
15. Fukuzawa, H. et al. Specular spin-valve films with an FeCo nano-oxide layer by
ion-assisted oxidation. J. Appl. Phys. 91, 6684–6690 (2002).
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
LETTERS
16. Scheybal, A. et al. Induced magnetic ordering in a molecular monolayer.
Chem. Phys. Lett. 411, 214–220 (2005).
17. Wende, H. et al. Substrate-induced magnetic ordering and switching of iron
porphyrin molecules. Nature Mater. 6, 516–520 (2007).
18. Atodiresei, N. et al. Controlling the magnetization direction in molecules via
their oxidation state. Phys. Rev. Lett. 100, 117207 (2008).
19. Ruben, M., Rojo, J., Romero-Salguero, F. J., Uppadine, L. H. & Lehn, J.-M.
Grid-type metal ion architectures: Functional metallosupramolecular arrays.
Angew. Chem. Int. Ed. 43, 3644–3662 (2004).
20. Stepanow, S. et al. Steering molecular organization and host–guest interactions
using two-dimensional nanoporous coordination systems. Nature Mater. 3,
229–233 (2004).
21. Semenov, A. et al. Controlled arrangement of supramolecular metal
coordination arrays on surfaces. Angew. Chem. Int. Ed. 38, 2547–2550 (1999).
22. Barth, J. V., Costantini, G. & Kern, K. Engineering atomic and molecular
nanostructures at surfaces. Nature 437, 671–679 (2005).
23. Lingenfelder, M. A. et al. Towards surface-supported supramolecular
architectures: Tailored coordination assembly of 1,4-benzenedicarboxylate and
Fe on Cu(100). Chem. Eur. J. 10, 1913–1919 (2004).
24. Stepanow, S., Lin, N. & Barth, J. V. Modular assembly of low-dimensional
coordination architectures on metal surfaces. J. Phys. Condens. Matter 20,
184002 (2008).
25. de Groot, F. & Kotani, A. Core Level Spectroscopy of Solids (CRC Press, 2008).
26. Gambardella, P. et al. Localized magnetic states of Fe, Co, and Ni impurities on
alkali metal films. Phys. Rev. Lett. 88, 047202 (2002).
27. Hocking, R. K. et al. Fe L-edge X-ray absorption spectroscopy of low-spin
heme relative to non-heme Fe complexes: Delocalization of Fe d-electrons into
the porphyrin ligand. J. Am. Chem. Soc. 129, 113–125 (2007).
28. Figgis, B. N. & Hitchman, M. A. Ligand Field Theory and Its Applications
(Wiley–VCH, 2000).
29. Rocha, A. R. et al. Towards molecular spintronics. Nature Mater. 4,
335–339 (2005).
30. Timm, C. & Elste, F. Spin amplification, reading, and writing in transport
through anisotropic magnetic molecules. Phys. Rev. B 73, 235304 (2006).
Acknowledgements
We acknowledge the ESRF for provision of beam time. Partial financial support was
received through the EUROCORES 05-SONS-FP-009 SANMAG project of the European
Science Foundation. P.G. and S.S. acknowledge financial support from the Spanish
Ministerio de Educación y Ciencia (SYNSPIN—MAT2007-62341).
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to P.G.
193
 LETTERS
PUBLISHED ONLINE: 1 FEBRUARY 2009 DOI: 10.1038/NMAT2374

Magnetic memory of a single-molecule quantum

magnet wired to a gold surface
Matteo Mannini1,2 , Francesco Pineider1 , Philippe Sainctavit3 , Chiara Danieli4 , Edwige Otero5 ,
Corrado Sciancalepore4 , Anna Maria Talarico4 , Marie-Anne Arrio3 , Andrea Cornia4 , Dante Gatteschi1
and Roberta Sessoli1 *

In the field of molecular spintronics1 , the use of magnetic
molecules for information technology is a main target and the
observation of magnetic hysteresis on individual molecules
organized on surfaces is a necessary step to develop molecular
memory arrays. Although simple paramagnetic molecules can
show surface-induced magnetic ordering and hysteresis when
deposited on ferromagnetic surfaces2 , information storage
at the molecular level requires molecules exhibiting an
intrinsic remnant magnetization, like the so-called single-
molecule magnets3 (SMMs). These have been intensively
investigated for their rich quantum behaviour4 but no magnetic
hysteresis has been so far reported for monolayers of SMMs
on various non-magnetic substrates, most probably owing
to the chemical instability of clusters on surfaces5 . Using
X-ray absorption spectroscopy and X-ray magnetic circular
dichroism synchrotron-based techniques, pushed to the limits
in sensitivity and operated at sub-kelvin temperatures, we
have now found that robust, tailor-made Fe4 complexes retain
magnetic hysteresis at gold surfaces. Our results demonstrate
that isolated SMMs can be used for storing information. The
road is now open to address individual molecules wired to
a conducting surface6,7 in their blocked magnetization state,
thereby enabling investigation of the elementary interactions
between electron transport and magnetism degrees of freedom
at the molecular scale8,9 .
The observation of a magnetic memory effect3 in
single-molecule magnets (SMMs) has represented a breakthrough
in nanomagnetism10 . It has opened the exciting perspective
of exploiting the information storage capability of individual
SMMs, made even more appealing owing to their rich quantum
behaviour4,11 . In fact, the combination of a large spin of the
molecule with an easy-axis magnetic anisotropy results in an
energy barrier that hampers the reversal of the magnetization,
although the barrier can be cross-cut by a tunnelling mechanism
for some particular values of the magnetic field4 . Anchoring SMMs
on conducting surfaces is required to make them individually
addressable by scanning probes8 , which enable both high-resolution
morphological studies and local spectroscopic investigations12 .
Moreover, the proposed scheme provides access to electron
transport through the magnetic molecules, the target of a field
that is referred to as molecular spintronics1,9 . Despite the successful
efforts to organize SMMs on various substrates13 and the reported
possibility to detect magnetic features of atoms14–16 or molecules
on surfaces17–20 , evidence that SMMs retain molecular magnetic
hysteresis when wired to metallic surfaces is still lacking. Without
such evidence, one of the pillars of this emerging field would
break down.
Demonstration that SMM behaviour is not destroyed on
interaction with the surface can be achieved by using synchrotron
radiation at very low temperature. In fact, X-ray magnetic circular
dichroism21 (XMCD) is a unique tool to provide a detailed
magnetic characterization down to sub-monolayers of magnetic
atoms22 or molecules2,5 .
Previous investigations have mainly focused on the archetypal
dodecamanganese family of SMMs (Mn12 ; refs 3,4), but adsorbates
of Mn12 complexes on gold undergo systematic reduction, with
concomitant disappearance of magnetic hysteresis5 . Moreover,
even the surface layers of bulk Mn12 samples show no magnetic
hysteresis, probably owing to structural deformations20 . In contrast,
we have found that a different class of SMMs comprising four
high-spin iron(iii) ions, Fe4 , exhibits the sought redox stability and
structural robustness. Studies on bulk samples have revealed that
Fe4 clusters in the layers closer to the surface–vacuum interface
retain both their static and dynamic magnetic features20 and are
therefore good candidates to investigate the existence of magnetic
hysteresis in molecules grafted onto conducting surfaces.
In the propeller-like molecular structure of Fe4 (see Fig. 1a), the
outer individual s = 5/2 spins interact antiferromagnetically with
the inner one, to give a ground S = 5 spin state characterized by an
anisotropy barrier of about 16 K to be overcome for magnetization
reversal23–25 . The ligands lying above and below the plane of
the iron ions are derivatives of 2-hydroxymethyl-1,3-propanediol,
which can be functionalized to promote interactions with different
substrates as well as to enhance magnetic anisotropy23,24 . For
deposition on gold surfaces, we have designed the sulphur-
functionalized derivative [Fe4 (L)2 (dpm)6 ] with H3 L representing
11-(acetylthio)-2,2-bis(hydroxymethyl)undecan-1-ol and Hdpm
representing dipivaloylmethane25 . The compound can be isolated
in pure, crystalline form and is completely stable in solution (see
Supplementary Information).
A monolayer of Fe4 molecules on a Au(111) surface was obtained
by self-assembly from solution. Its scanning tunnelling microscopy
(STM) characterization (Fig. 1b) carried out at room temperature
in air confirms the surface coverage by a homogeneous layer of
objects with size 2.5 ± 0.4 nm (see Fig. 1c and Supplementary
Information), in good agreement with the molecular size

1 Department of Chemistry and INSTM research unit, University of Florence, 50019, Sesto Fiorentino, Italy, 2 ISTM-CNR, URT Firenze, 50019, Sesto

Fiorentino, Italy, 3 Institut de Minéralogie et de Physique des Milieux Condensés, CNRS UMR7590, Université Pierre et Marie Curie, 75252 Paris Cedex 5,
France, 4 Department of Chemistry and INSTM research unit, University of Modena and Reggio Emilia, 41100, Modena, Italy, 5 Synchrotron Soleil, Saint
Aubin BP48, 91192 Gif sur Yvette Cedex, France. *e-mail: roberta.sessoli@unifi.it.

194 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2374 LETTERS
a b

8.6 nm

c 120

90

Counts
60

30

0
1.0 1.5 2.0 2.5 3.0 3.5 4.0
Diameter (nm)

Figure 1 | Monolayer of Fe4 on gold. a, Schematic diagram of the anchoring on a gold surface of the Fe4 derivative studied in this work through its
thiolate-terminated aliphatic chains. A view along the plane of the four metal atoms with the chains fully stretched has been chosen for the sake of clarity.
Iron atoms are drawn as large green spheres, oxygen in red, carbon in grey and sulphur in light yellow. Inset: A view of the magnetic core structure of Fe4 ,
with the ground-state spin arrangement (white arrows). b, Room-temperature constant-current STM image of the Fe4 monolayer obtained on a scan area
of 55 × 55 nm2 (bias voltage Vb = 350 mV, tunnelling current It = 9 pA). c, Statistical distribution of molecular diameters extracted from five images
obtained in the same experimental conditions as in b. The diameters follow a log-normal distribution centred at 2.50 nm with a standard deviation of
0.14 nm (solid curve).
determined by X-ray diffraction. The gold–sulphur chemical coordination environment of the central and peripheral iron
affinity enables the selective formation of a single-layer-thick ions, an antiferromagnetic interaction is expected to afford an
deposit, but no bidimensional order comparable to that observed approximately halved XMCD/XAS ratio as compared with full
in self-assembled monolayers of aliphatic thiols26 can be detected in polarization induced by the magnetic field (see Supplementary
the STM topographic images. In fact, the shape of the molecule and Information). Fragmentation into simple paramagnetic units
the strong steric hindrance of the equatorial ligands are expected or marked alterations of the core structure would then be
to prevent significant interactions between the aliphatic chains accompanied by significant changes in the XMCD/XAS ratio as
of adjacent molecules, a key factor in the formation of organic compared with the bulk, not observed here.
self-assembled monolayers. As the presence of hysteresis is directly related to the magnetic
The electronic structure of the layer has been investigated anisotropy, we have first investigated the field dependence of
through X-ray absorption spectroscopy (XAS) with an endstation the XMCD signal at 709.2 eV at two different temperatures
capable of reaching sub-kelvin temperatures27,28 . In Fig. 2a, we (T = 1.5 ± 0.1 and 4.5 ± 0.1 K). Interestingly the magnetization
show the XAS spectra recorded at the Fe L2,3 -edge (2p → 3d) isotherms so obtained do not rescale on a unique curve when
on the Fe4 monolayer with left- (σ+ ) and right- (σ− ) hand plotted versus H /T (Fig. 2c), in agreement with what is expected
circular polarization. The average spectra, calculated as (σ− +σ+ )/2, for an anisotropic paramagnet. The field dependence of the
coincide over the entire energy range with those recorded on XMCD signal compares well with the magnetization curve
a thick film of the same compound prepared by drop-casting calculated using the spin Hamiltonian that describes the magnetic
(see Supplementary Information)20 , showing that the electronic behaviour of bulk samples.
structure of the molecule is fully preserved on grafting. However, The encouraging results prompted us to explore the
the signal-to-background intensity ratio is about 30 times smaller field-dependent magnetic response of the sample down to the
in the monolayer as compared with the thick deposit and has been very low temperatures required to observe magnetic memory
found reproducible over five different preparations. As the total- effects in SMM materials. We have therefore exploited the assets
electron-yield mode used to detect X-ray absorption probes only of the X11MA-SIM beamline (beam stability and fast switching
the first few layers in bulk samples29 , the observed intensity fully of X-ray polarization) and further optimized the endstation so
confirms that we are actually monitoring a monolayer deposit. as to achieve the lowest possible level of iron contamination in
Element- and surface-selective magnetometry has then been the thermal shields required to reach sub-kelvin temperatures (see
carried out by recording the dichroic component, that is, the Supplementary Information for details). The recorded hysteresis
difference between the XAS signals obtained using right-hand curves are shown in Fig. 3a–c.
and left-hand circularly polarized X-rays, in an applied field Whereas simple paramagnetic behaviour is found at 1.0 K, a
of 3 T at 0.50 ± 0.05 K. The XMCD features (see Fig. 2b) and gradual opening of the hysteresis loop is detected at sub-kelvin
the XMCD/XAS intensity ratio (0.25–0.30) are identical to those temperatures. At T = 0.50±0.05 K, the loop has the typical butterfly
observed in bulk samples. The XMCD/XAS ratio has to be shape resulting from fast magnetization tunnelling in zero field23,24 ,
regarded as a reliable fingerprint of the spin alignment within showing unambiguously that SMM behaviour is observable on
the tetra iron(iii) core. In fact, owing to the very similar gold-wired molecules. Even though XMCD has recently been

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 195

LETTERS NATURE MATERIALS DOI: 10.1038/NMAT2374

a c
L3
1.50 σ+
L2

XAS (a.u.)
1.45 σ¬
X-ray 0.06
1.40
H

XMCD (a.u.)
1.35
0.04
b
XMCD (a.u.)

0
0.02 Exp. (T = 4.5 K)

Monolayer Exp. (T = 1.5 K)

Bulk reference Simulated
0
¬0.05
690 700 710 720 730 740 0 1 2 3
Energy (eV) μ 0H/T (T/K)

Figure 2 | XAS/XMCD of Fe4 monolayer. a, Iron L2,3 -edge XAS spectra recorded on the Fe4 monolayer at T = 0.50 ± 0.05 K using left- (σ+ ) and right-
(σ− ) hand circularly polarized light in a field induction of 3.0 T. b, XMCD spectrum (open circles) obtained as (σ− − σ+ ) and compared to the bulk
reference spectrum (solid line) recorded in similar experimental conditions on a thick film on the same substrate. c, Field and temperature dependence of
the XMCD signal at 709.2 eV (multiplied by −1) recorded on the Fe4 monolayer. The data have been taken by scanning the field up to 5.5 T at two different
temperatures, and the standard deviation (bars) has been evaluated by averaging four data sets. The lines correspond to the calculated magnetization by
using the spin Hamiltonian H = DS2z + gµB µ0 H ·S, where D is the axial anisotropy parameter, g is the Landé factor, µB is the Bohr magneton and H is the
applied magnetic field. Taking into account a random orientation of the molecular z axes with respect to the magnetic field, an acceptable agreement is
obtained assuming D/kB = −0.62 K and g = 2.00, as found in bulk samples25 .

a b
T = 1.0 K T = 0.70 K
0.02 0.02

0.01 0.01
XMCD (a.u.)
XMCD (a.u.)

0 0

¬0.01 ¬0.01

¬0.02 ¬0.02

¬1.5 ¬1.0 ¬0.5 0 0.5 1.0 1.5 ¬1.5 ¬1.0 ¬0.5 0 0.5 1.0 1.5
µ0H (T) µ0H (T)

c d
T = 0.50 K
Experimental data
0.02 ¬0.008 Fit

0.01
XMCD (a.u.)

XMCD (a.u.)

¬0.010
0

¬0.01 ¬0.012

¬0.02
¬0.014 T = 0.50 K

¬1.5 ¬1.0 ¬0.5 0 0.5 1.0 1.5 0 500 1,000 1,500

µ0H (T) Time (s)

Figure 3 | Hysteresis and magnetization dynamics of Fe4 monolayer. a–c, Temperature dependence of the magnetic hysteresis for the Fe4 monolayer,
monitored through the XMCD intensity (multiplied by −1) at 709.2 eV obtained from XAS signals normalized to the pre-edge intensity (704.0 eV). The
error bars represent the standard deviations estimated by averaging over three field cycles. A sweeping rate of 2 mT s−1 has been used. d, Time dependence
of the dichroic signal (multiplied by -1) at 709.2 eV for the Fe4 monolayer at T = 0.50 ± 0.05 K. The sample was first magnetized in a strong positive
magnetic field (+2.0 T), then the field was rapidly ramped to a moderate negative value (−0.25 T) and the time dependence of the dichroic signal was
measured. The procedure was repeated four times and the grey error bars represent the standard deviation. The black line provides the best-fit to
experimental data using a mono-exponential decay.

used to detect the substrate-driven magnetism of paramagnetic Further evidence that the deposited Fe4 complexes retain slow
(octaethylporphyrinato)iron(ii) complexes deposited on a ferro- magnetic relaxation was gained by measuring the time dependence
magnetic surface2 , this is the first observation of a magnetic hystere- of the XMCD signal after magnetizing the sample in a strong
sis of purely molecular origin in monolayers. magnetic field and then rapidly reducing the field to a small

196 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2374
value in the opposite direction. As reported in Fig. 3d, even if
the XMCD signal is significantly noisier than in thick films20 , a
time dependence of the magnetization is distinctly observed at
T = 0.50 ± 0.05 K. Fitting of the data using a mono-exponential
decay law provided a characteristic time τ = 220 ± 15 s, which
is comparable to that observed20 at the surface of thick films
(τ = 285±10 s at T = 0.55±0.05 K).
Despite the very low blocking temperature of Fe4 and its low
remnant magnetization due to resonant tunnelling in zero field,
both consequences of the relatively small spin of this SMM, the
present results give a positive answer to a key question, that is,
there are no fundamental reasons that preclude the observation
of magnetic hysteresis when SMMs are wired to a conducting
substrate. This conclusion is not at all trivial given the peculiar
origin of magnetic bistability in SMMs: geometrical distortions as
well as changes in the environment and in the effectiveness of spin–
phonon coupling are indeed known to have a profound influence
on the spin dynamics of SMMs (ref. 4). Furthermore, SMMs are
intrinsically fragile systems and previous attempts to demonstrate
slow magnetic relaxation in Mn12 adsorbates most probably failed
owing to redox and structural rearrangement phenomena5,20 .
Clearly emerging from our studies is also the potential of XMCD for
the characterization of monolayers of magnetic molecules, which
are even more demanding than inorganic nanostructures.
Although extra efforts to increase the blocking temperature are
required before SMMs can be realistically incorporated into storage
devices, the present results already pave the way to fundamental
developments in the emerging field of molecular spintronics such
as the possibility to control magnetization reversal by conduction
electrons in a metal–molecule–metal nanojunction. This approach
would mimic at the molecular scale the current-induced motion of
domain walls confined to nanostructures, which was revealed to be
particularly useful for memory storage applications30 .
Methods
A flame-annealed polycrystalline Au substrate, showing triangular reconstruction
of Au(111) (see Supplementary Information), was plunged into a 0.5 mM
dichloromethane solution of Fe4 (the synthesis of which is described elsewhere25 )
for 20 h so as to afford monolayer coverage. Overlayers of physisorbed molecules
not directly grafted to the gold surface were removed by repeated washings with
pure dichloromethane. Samples dried under nitrogen atmosphere were investigated
by room-temperature low-current STM in a controlled atmosphere. Identical
samples for XAS/XMCD investigations were transferred to the ultrahigh vacuum
chamber through a portable glove box inflated with nitrogen gas.
The XAS/XMCD investigations were carried out at BESSY (Berlin, Germany)
and SLS (Villigen, Switzerland). For the experiments, we used the same TBT-XMCD
endstation, working under strong applied magnetic fields with both a pumped
4
He cryostat and a 3 He–4 He dilution set-up27,28 , connected to the UE46-PGM
(BESSY) and X11MA-SIM (SLS) beamlines. The absorbance was measured in the
total-electron-yield mode29 and magnetization and hysteresis curves were recorded
by monitoring the field dependence of the dichroic signal at the energy of its
maximum amplitude. Time-dependent XMCD measurements were obtained by
monitoring the evolution of the dichroism after a rapid switching of the magnetic
field from µ0 H = +2.0 to −0.25 T and continuously switching the polarization
of the light. The required light stability was achieved by the fast switching mode
available at SLS, which uses a couple of undulators with opposite polarizations
rapidly tuned and detuned, respectively.
Received 17 October 2008; accepted 22 December 2008;
published online 1 February 2009
References
1. Rocha, A. R. et al. Towards molecular spintronics. Nature Mater.
4, 335–339 (2005).
2. Wende, H. et al. Substrate-induced magnetic ordering and switching of iron
porphyrin molecules. Nature Mater. 6, 516–520 (2007).
3. Sessoli, R., Gatteschi, D., Caneschi, A. & Novak, M. A. Magnetic bistability in a
metal-ion cluster. Nature 365, 141–143 (1993).
4. Gatteschi, D., Sessoli, R. & Villain, J. Molecular Nanomagnets
(Oxford Univ. Press, 2006).
5. Mannini, M. et al. XAS and XMCD investigation of Mn12 monolayers on gold.
Chem. Eur. J. 14, 7530–7535 (2008).
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
LETTERS
6. Heersche, H. B. et al. Electron transport through single Mn12 molecular
magnets. Phys. Rev. Lett. 96, 206801 (2006).
7. Jo, M. H. et al. Signatures of molecular magnetism in single-molecule transport
spectroscopy. Nano Lett. 6, 2014–2020 (2006).
8. Kim, G. H. & Kim, T. S. Electronic transport in single-molecule magnets on
metallic surfaces. Phys. Rev. Lett. 92, 137203 (2004).
9. Bogani, L. & Wernsdorfer, W. Molecular spintronics using single-molecule
magnets. Nature Mater. 7, 179–186 (2008).
10. Ziemelis, K. Nature Milestones Spin, Milestone 22: (1996) Mesoscopic tunnelling
of magnetization (doi:10.1038/nphys877).
11. Leuenberger, M. N. & Loss, D. Quantum computing in molecular magnets.
Nature 410, 789–793 (2001).
12. Wiesendanger, R. Scanning Probe Microscopy and Spectroscopy Methods and
Applications (Cambridge Univ. Press, 1994).
13. Cornia, A. et al. Preparation of novel materials using SMMs. Struct. Bond.
122, 133–161 (2006).
14. Madhavan, V., Chen, W., Jamneala, T., Crommie, M. F. & Wingreen, N. S.
Tunneling into a single magnetic atom: Spectroscopic evidence of the Kondo
resonance. Science 280, 567–569 (1998).
15. Heinze, S. et al. Real-space imaging of two-dimensional antiferromagnetism
on the atomic scale. Science 288, 1805–1808 (2000).
16. Hirjibehedin, C. F., Lutz, C. P. & Heinrich, A. J. Spin coupling in engineered
atomic structures. Science 312, 1021–1024 (2006).
17. Zhao, A. et al. Controlling the Kondo effect of an adsorbed magnetic ion
through its chemical bonding. Science 309, 1542–1544 (2005).
18. Durkan, C. & Welland, M. E. Electronic spin detection in molecules using
scanning-tunneling-microscopy-assisted electron-spin resonance. Appl. Phys.
Lett. 80, 458–460 (2002).
19. Iacovita, C. et al. Visualizing the spin of individual cobalt-phthalocyanine
molecules. Phys. Rev. Lett. 101, 116602 (2008).
20. Mannini, M. et al. X-ray magnetic circular dichroism picks out single-molecule
magnets suitable for nanodevices. Adv. Mater. 21, 167–171 (2009).
21. Stöhr, J. Exploring the microscopic origin of magnetic anisotropies with x-ray
magnetic circular dichroism (XMCD) spectroscopy. J. Magn. Magn. Mater.
200, 470–497 (1999).
22. Gambardella, P. et al. Ferromagnetism in one-dimensional monoatomic metal
chains. Nature 416, 301–304 (2002).
23. Cornia, A. et al. Energy-barrier enhancement by ligand substitution in
tetrairon(iii) single-molecule magnets. Angew. Chem. Int. Ed.
43, 1136–1139 (2004).
24. Accorsi, S. et al. Tuning anisotropy barriers in a family of tetrairon(III)
single-molecule magnets with an S = 5 ground state. J. Am. Chem. Soc.
128, 4742–4755 (2006).
25. Barra, A. L. et al. New single-molecule magnets by site-specific substitution:
Incorporation of ‘alligator clips’ into Fe4 complexes. Eur. J. Inorg. Chem.
4145–4152 (2007).
26. Ulman, A. Formation and structure of self-assembled monolayers. Chem. Rev.
96, 1533–1554 (1996).
27. Sainctavit, Ph. & Kappler, J.-P. in X-ray Magnetic Circular Dichroism at Low
Temperature in Magnetism and Synchrotron Radiation (eds Beaurepaire, E.,
Scheurer, F., Krill, G. & Kappler, J.-P.) 135–153 (Springer, 2001).
28. Letard, I. et al. Remnant magnetization of Fe8 high-spin molecules: X-ray
magnetic circular dichroism at 300 mK. J. Appl. Phys. 101, 113920 (2007).
29. Nakajima, R., Stohr, J. & Idzerda, Y. U. Electron-yield saturation effects in
L-edge x-ray magnetic circular dichroism spectra of Fe, Co, and Ni. Phys. Rev. B
59, 6421–6429 (1999).
30. Parkin, S. S. P., Hayashi, M. & Thomas, L. Magnetic domain-wall racetrack
memory. Science 320, 190–194 (2008).
Acknowledgements
We acknowledge F. Scheurer, J. P. Kappler and B. Muller for their help in the installation
of the endstation, and the staff of the X11MA-SIM (Swiss Light Source, Paul Scherrer
Institut, Villigen, Switzerland) and UE46-PGM (BESSY synchrotron, Berlin, Germany)
beamlines for their support. In particular, we thank A. Fraile-Rodriguez, L. Joly and
F. Nolting for their excellent technical support at the X11MA-SIM beamline. We thank
L. Gorini for his contribution to the development of the ligand synthesis and M. Etienne
for his help in artwork preparation. This research project has been supported by the
EU, within the EU FP6, through the Key Action: Strengthening the European Research
Area, Research Infrastructures, through NoE MAGMANet, through the Integrated
Infrastructure Initiative ‘Integrating Activity on Synchrotron and Free Electron Laser
Science’ and through the ERANET project ‘NanoSci-ERA: NanoScience in the European
Research Area’. It has been partially financially supported by the Italian CNR and MIUR
and by the German DFG.
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to R.S.
197
 LETTERS
PUBLISHED ONLINE: 15 FEBRUARY 2009 DOI: 10.1038/NMAT2385
Room-temperature defect-engineered spin filter
based on a non-magnetic semiconductor
X. J. Wang1 , I. A. Buyanova1 *, F. Zhao2 , D. Lagarde2 , A. Balocchi2 , X. Marie2 , C. W. Tu3 , J. C. Harmand4
and W. M. Chen1 *
Generating, manipulating and detecting electron spin
polarization and coherence at room temperature is at the heart
of future spintronics and spin-based quantum information
technology1–4 . Spin filtering, which is a key issue for spintronic
applications, has been demonstrated by using ferromagnetic
metals5–8 , diluted magnetic semiconductors9,10 , quantum point
contacts11 , quantum dots12 , carbon nanotubes13 , multiferroics14
and so on. This filtering effect was so far restricted to a
limited efficiency and primarily at low temperatures or under
a magnetic field. Here, we provide direct and unambiguous
experimental proof that an electron-spin-polarized defect,
such as a Gai self-interstitial in dilute nitride GaNAs, can
effectively deplete conduction electrons with an opposite spin
orientation and can thus turn the non-magnetic semiconductor
into an efficient spin filter operating at room temperature
and zero magnetic field. This work shows the potential of
such defect-engineered, switchable spin filters as an attractive
alternative to generate, amplify and detect electron spin
polarization at room temperature without a magnetic material
or external magnetic fields.
In Fig. 1a we illustrate the physical principle of the approach. If
an electron localized at a defect is spin polarized, the defect will
capture only a conduction electron with an opposite spin owing
to the Pauli exclusion principle and can subsequently annihilate it
with a hole. If the corresponding process is efficient, spin-polarized
defects will effectively deplete the conduction electrons with the
opposite spin and will leave behind only those with the same
spin as the defect electrons. This process will turn a pool of spin-
unpolarized electrons entering the material into spin-polarized ones
when they exit, making the material a spin filter. The incoming
electrons to be spin filtered as well as the spin polarization of the
electrons at the defects can be provided either by electrical injection
or by optical excitation, making this versatile spin filter relevant to a
variety of applications in future spin electronics or spin photonics.
For a proof-of-concept purpose, we used optical excitation
and detection as illustrated in Fig. 1b,c. We show below that Gai
self-interstitials in GaNAs are identified as suitable defects for
efficient spin filtering. This is because each Gai is occupied by a
single electron in the doubly positively charged state Ga2+ i . When
this electron is spin polarized, the defect can efficiently capture a
conduction electron with an opposite spin and subsequently rapidly
annihilate it with a free hole. The complete carrier recombination
path is identified as:
photo-excitation e capture h capture
Ga2+ 2+
−−−→ Ga1+
i −−−−−−−→ Gai + e + h − i +h −−−−→ Ga2+
i a direct consequence of the strong spin-filtering effect due to
1 Departmentof Physics, Chemistry and Biology, Linköping University, 58183 Linköping, Sweden, 2 Université de Toulouse, LPCNO: INSA, UPS, CNRS, 135
avenue de Rangueil, 31077 Toulouse cedex, France, 3 Department of Electrical and Computer Engineering, University of California, La Jolla, California
92093, USA, 4 LPN, route de Noazay, 91460 Marcoussis, France. *e-mail: irb@ifm.liu.se; wmc@ifm.liu.se.
198
Spin polarization of the electrons at Ga2+ i was provided by
optical orientation15 using circularly polarized light (σ ± ) via
spin-dependent recombination16–21 , as described in the Methods
section. The spin-filtering effect induced by the spin-polarized
defects drives the spins of the conduction and localized elec-
trons at Ga2+ i towards complete alignment when no further elec-
tron capture and recombination can occur via the defects as
shown in Fig. 1b. Consequently, a high degree of conduction
electron spin polarization Pe can be achieved. Pe is defined as
Pe = ((n+ − n− )/(n+ + n− )), where n+ (n− ) denotes the spin-up
(spin-down) conduction electron concentration. For comparison,
linearly polarized optical excitation (σ x ) was also used. Such exci-
tation equally generates electrons of both spins, and will thus lead
to zero electron spin polarization of the defects at zero magnetic
field (B = 0). Now, the defect can capture conduction electrons of
both spins with an equal probability (Fig. 1c); that is, spin filtering
ceases to function. This leads to zero Pe and shorter lifetimes
for the photo-generated carriers as compared with that under σ ±
excitation, which can be monitored by photoluminescence of the
band-to-band transition.
In Fig. 2a we show the band-to-band photoluminescence
spectrum from GaNAs obtained at room temperature and B = 0.
It consists of the strongly overlapping electron–heavy hole (e–hh)
and electron–light hole (e–lh) transitions, which split under the
tensile strain leaving the e–hh emission at the shorter wave-
length (for a review on GaNAs, see, for example, ref. 22). The
photoluminescence polarization of these transitions, defined by
(I (σ + ) − I (σ − ))/(I (σ + ) + I (σ − )), where I (σ + ) and I (σ − ) denote
photoluminescence intensities with the corresponding polariza-
tions, directly reflects Pe . (Owing to strong spectral overlap with
the more intense e–hh emission with opposite photolumines-
cence polarization, the photoluminescence polarization of the
e–lh emission is severely undermined in the measured spectra.)
Pe = −((I (σ + ) − I (σ − ))/(I (σ + ) + I (σ − ))) when only the e–hh
transition is considered, which is used below to obtain Pe . It
should be pointed out that the actual Pe values should be higher
than that deduced in this way when hh–lh mixing and spectral
overlap are taken into account. Under σ x excitation, zero Pe was
observed as expected. Under σ ± excitation, on the other hand,
a sizable Pe can be detected. Its sign can be changed following
a change in the helicity of the excitation light. Pe was found to
critically depend on N composition and growth conditions, which
determine the concentration of Gai . The observed values (∼32%)
are much higher than that observed in N -free GaAs (<6%), being
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2385 LETTERS
a e a e ¬1/2 +1/2

PL intensity (arb. units)

B=0

PL polarization (%)
Exc:σσ + 40 σσ+ σσ+ σσ¬ σσ¬
¬Pe lh
Gai2+ ¬1/2 +1/2
GaNAs hh
h Exc:σσ x ¬3/2 +3/2
0
950 1,000
Wavelength (nm)

c b 1,050
b CB CB Exc:σσ x

(1)
Exc: Gai Exc: Gai
PL PL

Wavelength (nm)
σσ+ (1+/2+) σσx (1+/2+) 950
(2) 50

c 1,050
VB VB
Exc:σσ +
0
Spin filtering No spin filtering

Figure 1 | Principle of defect-engineered spin filtering and the 950

experimental approach at room temperature and B = 0. a, A schematic
illustration of the spin-filtering effect of conduction electrons via
spin-polarized defects in a non-magnetic semiconductor, for example, Gai d

self-interstitials in GaNAs. b, In our approach, spin polarization of the first
electrons at the defects is achieved by circularly polarized photo-excitation
(σ + as an example), leading to the spin-filtering effect and spin blockade of
carrier recombination via the defects when both conduction and defect
electrons are completely polarized to the same spin orientation.
Consequently, higher free carrier concentrations and thus stronger
intensity of the associated band-to-band photoluminescence (PL)
transition are expected in case b as compared with case c. c, As a reference
in our approach, linearly polarized photo-excitation (σ x ) corresponds to the
case when no spin polarization of the first electrons at the defects can be
created and thus no spin-filtering effect is expected. CB and VB denote the
conduction band and valence band, respectively. The charge transition
(1+/2+) of the spin-filtering Gai defects is also indicated, corresponding to
a transition between Ga1+ 2+
i and Gai that have two and one bound electron,
respectively. The numbers in the parentheses associated with the electron
capture and annihilation via the defects shown in c indicate the sequence of
events. The thickness of the photoluminescence arrows symbolizes the
relative intensity of the band-to-band transition.
Gai preferably introduced in GaNAs. The observed strong Pe is
accompanied by a higher photoluminescence intensity under σ ±
excitation than that under σ x excitation (Fig. 2b,c). This arises from
spin blockade of the Gai -mediated carrier recombination under
σ ± excitation, consistent with the slower photoluminescence decay
shown in Fig. 2b–d.
To obtain quantitative information on the concentration of
spin-filtering defects and their effect on Pe , we carried out a
detailed study of Pe dependence on excitation light intensity
(that is, on the number of photo-generated free carriers).
Representative results obtained at room temperature and B = 0
are shown in Fig. 3, and were analysed by the following coupled
nonlinear rate equations19,21 :
dn±
= −γe n± N∓ −
n± − n∓
+ G± −
n±
dt 2τs τd
dN± N± − N∓ 1
= −γe n∓ N± − + γh pN↑↓
dt 2τsc 2
(1)
dp n+ + n−
= −γh pN↑↓ + G+ + G− −
dt τd
Nc = N↑↓ + N+ + N−
PL (arb. units)
Exc:σσ +
Exc:σσ x
B=0
0 100 200
Time (ps)
Figure 2 | Photoluminescence intensity and polarization (corresponding
to the conduction electron spin polarization Pe ) with or without the
spin-filtering effect. a, A typical band-to-band photoluminescence
spectrum under σ x excitation, and photoluminescence polarization
obtained under σ x and σ + excitation. A schematic diagram of the
conduction band and valence band states and the associated e–hh and e–lh
transitions is also shown. The notations ±1/2 and ±3/2 refer to |mcs i and
|mvJ i of the conduction and valence band electrons. b,c, Temporal evolution
of the band-to-band photoluminescence spectra (by detecting the total
photoluminescence intensity) after σ x - and σ + -polarized pulsed laser
excitation, respectively. Time-integrated photoluminescence spectra, under
σ x (black curve) and σ + (red curve) excitation, are shown in the inset.
d, Decay curves of the total intensity of the band-to-band
photoluminescence under σ x and σ + excitation, normalized to their peak
intensity for easy comparison. All results were obtained from the as-grown
GaN0.0076 As0.9924 epilayer grown at 420 ◦ C. The wavelength of the
excitation photons is 840 nm. All results were obtained at room
temperature and B = 0.
Here, G± is the photo-generation rate of free carriers and n±
(N± ) is the density of conduction electrons (the density of the
defects occupied by a single electron), where the ‘±’ signs refer
to the electron spin orientations Sz = ±1/2. N↑↓ corresponds to
the concentration of the defects having two spin-paired electrons
and Nc is the total defect concentration. The density of free
holes is denoted by p. τs (τsc ) is the spin relaxation times of
conduction (localized) electrons. τd denotes the free carrier decay
time, including all radiative and non-radiative recombination
channels except that via the spin-filtering defects. γe (γh ) is the
trapping coefficient of free electrons (free holes) by the defects. In
the analysis, we use τd = 10 ns, τsc = 1.5 ns and τs = 150 ps (refs 19,
21). Here, we assume that τd is governed by the radiative time
of the band-to-band photoluminescence transition. The analysis is
not sensitive to τd as long as it is much longer than the electron
capture and recombination time via the spin-filtering defects, that
is, τd  (γe N± )−1 and τd  (γh N↑↓ )−1 . The analysis is insensitive

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 199

LETTERS NATURE MATERIALS DOI: 10.1038/NMAT2385

to τsc , as long as it is >1.5 ns. This leaves only γe /γh and γe Nc a 20

as fitting parameters. As the absolute values of γe and defect GaNAs MQWs, Tg = 420 °C
concentration Nc cannot be determined independently here, we
used a combined fitting parameter γe Nc that is proportional to the 15
As-grown
capture rate of free electrons by the defects and can also be used

¬Pe (%)
to compare relative defect concentrations in different samples. The RTA γeNc = 0.15 ps¬1
10
fitting curves based on equation (1) are shown by the dashed lines
in Fig. 3, with γe /γh = 4 and γe Nc values given for each sample.
The analysis reveals that the defect concentration increases with 5
γeNc = 0.037 ps¬1
decreasing growth temperature and increasing N concentration
(not shown in Fig. 3). Post-growth rapid thermal annealing (RTA) 0
0 50 100 150 200
is found to significantly reduce the defect concentration by a factor
Continuous-wave laser power (mW)
of ∼2–15. The observed increase of Pe with excitation power clearly
manifests the spin-filtering effect. b 20
0
GaNAs MQWs, Tg = 580 °C
To identify the exact chemical nature of the spin-filtering
defects, we carried out a detailed study by optically detected 15
magnetic resonance (ODMR; for a review on ODMR, see, for γ eNc = 0.077 ps¬1
example, ref. 23). ODMR is ideally suited here because it monitors

both the conduction electron concentration (via the intensity of
the band-to-band photoluminescence) and the spin polarization
of the electrons at the spin-filtering defects (via the electron
spin resonance (ESR) intensity that is proportional to |N+ –N− |),
as well as the link between them (via ODMR as a change in
the intensity of the band-to-band photoluminescence induced
by the ESR transitions). Under σ x excitation, an equal number
of spin-up and spin-down conduction electrons are created. If
the capture and annihilation of the electrons by the defects
dominates over spin relaxation of both conduction and localized
electrons, no population difference between the two spin states
of the defects is expected, that is, |N+ –N− | = 0, rendering a
vanishing ESR (thus ODMR) signal. Under σ ± excitation, on
the other hand, the dynamic spin polarization of the electrons
at the defects leads to a sizable |N+ –N− | and thus ESR. The
ESR-induced spin flips of the spin-polarized defect electrons will
open up an otherwise spin-forbidden recombination channel (see
Fig. 4a). This will lead to a reduction of free carrier concentrations
and thus of the corresponding band-to-band photoluminescence
intensity, giving rise to a so-called ODMR signal. As examples,
we show in Fig. 4b typical ODMR spectra from GaNAs. As
expected, a vanishingly weak ODMR signal was observed under
σ x excitation, whereas rather strong ODMR signals were observed
under σ ± excitation. These findings clearly show that a sizable
spin polarization of the defect electrons was generated under σ ±
excitation, and a slight change of this polarization by the ESR
directly affects the free carrier concentration, proving the role of the
defects in spin filtering.
A Gai self-interstitial is unambiguously identified by ODMR
as the core of the spin-filtering defects. This conclusion is based
on the following experimental facts. First, the observed multiple
ODMR lines arise from a hyperfine structure derived from a strong
interaction between an unpaired localized electron spin (S = 1/2)
and the nuclear spin of an atom that has two isotopes with a nuclear
spin I = 3/2 and a 60/40 ratio of natural abundance. This gives rise
to four magnetic-dipole-allowed ODMR transitions (1mS = ±1
and 1mI = 0) for each Ga isotope with an intensity ratio of 60/40
(see Fig. 4c). Ga is the only atom with such unique properties, that
is, two naturally abundant isotopes, 69 Ga (60.4% abundant) and
71
Ga (39.6% abundant), and I = 3/2 for both isotopes. Second,
the ODMR spectra are isotropic with a rotation of magnetic field
B with respect to the crystallographic axes. This finding reveals
that the electron wavefunction at the defects should be s-like. This
is consistent with the observed strong hyperfine interaction, as
the s-like electron wavefunction results in a strong Fermi contact
term. The involved defects should then be in the Ga2+ i charge state
occupied by a single unpaired electron with an electronic state
of A1 symmetry, as the alternative GaAs antisite was predicted to
¬Pe (%)
10
5 γ eNc = 0.031 ps¬1
0
0 50 100 150 200
Continuous-wave laser power (mW)
c
GaNAs epilayer
Tg = 420 ° C
40
¬Pe (%)
γeNc = 0.074 ps¬1
20
0
0 50 100 150
Pulsed laser power (mW)
Figure 3 | Dependence of conduction electron spin polarization on the
optical excitation power and the concentration of the spin-filtering
defects. a,b, The symbols denote the Pe values determined from the
photoluminescence polarization of the e–hh transition detected at the peak
position of the photoluminescence spectra under σ + continuous-wave
laser excitation at 840 nm, obtained from the GaN0.012 As0.988 MQWs
grown at 420 ◦ C (a) and the GaN0.011 As0.989 MQWs grown at 580 ◦ C (b).
Results from both as-grown (filled circles) and RTA-treated (filled
triangles) samples are shown. c, The time-integrated values of Pe from the
GaN0.0076 As0.9924 epilayer grown at 420 ◦ C (circles), obtained by
detecting the photoluminescence polarization at 950 nm (see Fig. 2a)
under σ + pulsed laser excitation at 840 nm. The dashed lines are the
simulated curves based on the best fit of equation (1) to the experimental
results, yielding the fitting parameter γe Nc for each sample. The results
clearly show that Pe increases with increasing excitation intensity until the
concentration of the photo-generated conduction electrons is comparable
to that of the spin-filtering defects. With a further increase in the excitation
density, Pe saturates, and even slightly decreases under the
continuous-wave laser excitation, because the minority spins of the excess
conduction electrons can no longer be depleted by the defects. The
maximum value of Pe achievable in each sample is determined by its γe Nc
value and thus its defect concentration. Typically, higher Pe values were
obtained under the pulsed laser excitation with a similar average laser
power as compared to that under the continuous-wave laser excitation.
This can be attributed to the much higher excitation density during the
laser pulses and a discontinued supply of minority spins after the laser
pulses, when the pulsed laser excitation was used. All results were
obtained at room temperature and B = 0.

200 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2385 LETTERS
a Exc: σσx Exc: σσ+ b
N+ N+
Gai2+ Gai2+

N¬ N¬ Exc: σσ x
B B

ODMR (arb.units)
Without ESR & ODMR With ESR & ODMR

Exp

Exc: σσ +
(2)

PL
Exc:
σσ+ Sim
(3)
Gai¬C
(1)
Magnetic field (mT)

c
ms mI

69Ga ¬C
+3/2
10 i +1/2 +1/2
¬1/2
¬3/2
d 0
¬3/2
¬10 ¬1/2 ¬1/2

Energy (GHz)
+1/2
ODMR (arb.units)

+3/2

+3/2
71Ga ¬C +1/2
10 i ¬1/2
+1/2 ¬3/2
0 ¬3/2
¬10 ¬1/2 ¬1/2
+1/2
+3/2
0 0.2 0.4 0.6 100 300 500
γ eNc (ps¬1) Magnetic field (mT)

Figure 4 | Identification of the spin-filtering defects by ODMR. a, Schematic diagrams of the expected ESR transitions under σ x and σ + excitation. In the
former case, an equal population of the two electron spin states of the Ga2+i defects results in vanishing ESR and ODMR signals as the ESR and ODMR
intensity is proportional to the population difference |N+ –N− |. Under σ + excitation, on the contrary, a sizable |N+ –N− | induced by dynamical electron spin
polarization leads to a detectable ESR transition between the two spin states. The ESR transition opens the otherwise spin-forbidden carrier capture and
recombination channel, leading to a reduced number of free carriers available for the band-to-band photoluminescence transition—an ODMR signal. The
numbers in the parentheses indicate the sequence of events. b, Typical ODMR spectra obtained by monitoring the total intensity of the band-to-band
photoluminescence from an RTA-treated GaN0.021 As0.979 epilayer, obtained at 3 K under σ x and σ + excitation at 850 nm. The microwave frequency used
is 9.2823 GHz. A simulated ODMR spectrum of the identified Gai defect (denoted by Gai –C) is also shown. c, Calculated energy levels associated with the
electronic and nuclear spin states of the Ga2+i defect. The allowed ESR transitions (1mS = ±1 and 1mI = 0) occur when the electron spin splitting matches
the microwave photon energy, and are marked by the vertical lines. The calculations and simulation were done with the aid of equation (2) by using the
spin Hamiltonian parameters given in the text. d, ODMR intensities from the identified Gai defects as a function of γe Nc . The latter are determined from the
rate equation analyses (equation (1)). The line is a guide to the eye.

possess a T2 -symmetry state24,25 . To quantitatively confirm this
identification, we have carried out a detailed analysis of the ODMR
results by a spin Hamiltonian
H = µB g B ·S + AS ·I
Here, µB is the Bohr magneton, g is the electronic g -factor
and A is the hyperfine parameter. g = 2.000 (2.005),
A(69 Ga) = 620 (770) × 10−4 cm−1 and A(71 Ga) = 788
(1,000)×10−4 cm−1 are obtained from a best fit to the experimental
data for the Gai defects in the RTA-treated (as-grown) samples.
The ratio A(71 Ga)/A(69 Ga) is in excellent agreement with that
of their nuclear magnetic moments µ(71 Ga)/µ(69 Ga) = 1.27. An
ODMR spectrum simulated by equation (2) using the obtained
parameters is shown as Gai –C in Fig. 4b, showing excellent
agreement with the experimental data. The difference between
the Gai defects in the RTA and as-grown samples lies at their
neighbouring atoms and geometric locations. By using a one-
electron linear combination of atomic orbital scheme26 and the
charge density of the 4s electron |ψ4s (0)|2 = 72.7 × 1024 cm−3
for a free neutral Ga atom25,27 , the localization of the electron
wavefunctions at the Ga2+ i defects is estimated to be 20%
(as-grown) and 16% (RTA). The rather strong localization
shows that they are deep-level defects, a prerequisite for efficient
(2) carrier recombination.
To further quantitatively confirm that the identified Gai
defects are indeed responsible for the spin filtering, we examined
their ODMR intensity as a function of the spin-filtering defect
concentrations deduced by equation (1). The excellent correlation,
shown in Fig. 4d, provides further support for the predominant role
of these Gai defects in spin filtering that has led to the observed
strong Pe at room temperature.
Further improvements in the efficiency of the defect-engineered
spin-filtering can be predicted by (1) increasing the ratio between
the capture rate of conduction electrons by the Gai defects
(∝γe Nc ) and the spin relaxation rate of conduction electrons
(1/τs ) and (2) increasing the ratio between γe Nc and the spin-
independent carrier recombination rate (1/τd ) via other defects.
Indeed, the spin relaxation of conduction electrons tends to
drive Pe towards zero value at thermal equilibrium (B = 0).

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 201

LETTERS
Spin-independent carrier recombination via other defects equally
depletes conduction electrons of both spins, competing with spin
filtering. The maximum Pe value is expected to approach 100%
when γe Nc  1/τs and 1/τd , independent of initial spin polarization
of the incoming electrons before being spin filtered. Future efforts
should therefore focus on prolonging the spin relaxation time
of conduction electrons28,29 and increasing the concentration of
the spin-filtering defects without introducing other competing,
spin-independent carrier recombination centres.
So far there is a lack of an efficient room-temperature spin
filter—a key element for spintronics—despite intense efforts. This
work demonstrates a significant development of our ability in
drastically increasing spin-filtering efficiency at room temperature
by purposely introducing spin-filtering defects. Our approach offers
another attractive feature—the freedom and ease in switching
the spin direction of the spin filters by reversing electron
spins of spin-filtering defects either optically or electrically. The
proposed spin filtering not only can generate Pe or amplify
weak Pe , but can also detect spin polarization of injected
electrons by monitoring the ratio between the numbers of
electrons passing and being blocked by the spin filter. This
work has thus demonstrated the potential of defect-engineered,
switchable spin filters for creating, amplifying and detecting
electron spin polarization at room temperature in a non-magnetic
semiconductor without requiring external magnetic fields, desirable
for practical device applications.
Methods
Samples. The dilute nitride GaNAs samples were grown by gas-source as well
as solid-source molecular beam epitaxy on (100)-oriented semi-insulating
GaAs substrates with a 2,500-Å-thick GaAs buffer. Two types of sample were
studied: (1) 7-period GaAs/GaNAs (200/70 Å) multiple-quantum-wells (MQWs)
structures, which were grown either at Tg = 420 ◦ C with N composition [N ] = 1.2%
and 2% or at 580 ◦ C with [N ] = 1.1%; and (2) GaNAs epilayers (about 0.1 µm
thickness) with [N ] = 0.76–2.1%, grown at 420 ◦ C. Post-growth RTA was carried
out at 850 ◦ C for 10 s with a halogen lamp in a flowing N2 ambient.
Dynamical polarization of the spin-filtering Gai defects. In our experiments,
the initial spin polarization of conduction electrons was generated by
circularly polarized light excitation at energy above the e–hh and e–lh
band-to-band transition in GaNAs, taking advantage of the selection rules
and the threefold higher oscillator strength of the e–hh optical transition as
compared with the e–lh transition15 . A spin-polarized conduction electron
can be captured by the Ga2+ defect only if the first localized electron has a
i
spin orientation opposite to that of the conduction electron, owing to the
Pauli exclusion principle. After the subsequent recombination between one
of two localized electrons and an unpolarized free hole (due to much faster
spin relaxation), a half number of the Ga2+ are left with a localized electron
i
with its spin orientation parallel to that of the conduction electrons. Such
a continuous spin-dependent recombination process16–21 will dynamically
polarize the spin of the first localized electrons at the Ga2+ towards that of
i
conduction electrons.
Experimental techniques. Both continuous-wave and time-resolved
photoluminescence experiments were carried out at room temperature and
at zero magnetic field. Photo-excitation above the bandgap energy of GaNAs
was provided by circularly polarized light (and linearly polarized light) from a
Ti:sapphire laser at a wavelength of 840–855 nm, propagating along the direction
normal to the sample surface. Laser power of up to 200 mW was used with a
typical spot size of about 0.1 mm. The resulting polarized photoluminescence
was measured in a backscattering geometry. In time-resolved photoluminescence
experiments, the samples were excited by 1.5 ps pulses generated by a mode-locked
Ti:sapphire laser with a repetition frequency of 80 MHz. The time-resolved
photoluminescence was then recorded using a streak camera with an overall
time resolution of 8 ps.
The ODMR experiments were done at 2.5–300 K with a modified ESR
spectrometer working at the X-band (∼9.3 GHz). Photoluminescence was
excited by a Ti:sapphire laser at a wavelength of 850 nm. The ODMR signals
were detected as spin-resonance-induced changes of the photoluminescence
intensity monitored by a cooled Ge detector with a proper selection
of optical filters.
202
NATURE MATERIALS DOI: 10.1038/NMAT2385
Received 8 September 2008; accepted 13 January 2009;
published online 15 February 2009
References
1. Wolf, S. A. et al. Spintronics: A spin-based electronics vision for the future.
Science 294, 1488–1495 (2001).
2. Žutić, I., Fabian, J. & Das Sarma, S. Spintronics: Fundamentals and applications.
Rev. Mod. Phys. 76, 323–410 (2004).
3. Awschalom, D. D. & Flatté, M. E. Challenges for semiconductor spintronics.
Nature Phys. 3, 153–159 (2007).
4. Chappert, C., Fert, A. & Nguyen Van Dau, F. The emergence of spin electronics
in data storage. Nature Mater. 6, 813–823 (2007).
5. Zhu, H. J. et al. Room-temperature spin injection from Fe into GaAs.
Phys. Rev. Lett. 87, 016601 (2001).
6. Hammar, P. R. & Johnson, M. Detection of spin-polarized electrons injected
into a two-dimensional electron gas. Phys. Rev. Lett. 88, 066806 (2002).
7. Jiang, X. et al. Highly spin-polarized room-temperature tunnel injector for
semiconductor spintronics using MgO(100). Phys. Rev. Lett. 94, 056601 (2005).
8. Jonker, B. T. et al. Electrical spin-injection into silicon from a ferromagnetic
metal/tunnel barrier contact. Nature Phys. 3, 542–546 (2007).
9. Fiederling, R. et al. Injection and detection of a spin-polarized current in a
light-emitting diode. Nature 402, 787–790 (1999).
10. Ohno, Y. et al. Electrical spin injection in a ferromagnetic semiconductor
heterostructure. Nature 402, 790–792 (1999).
11. Potok, R. M. et al. Detecting spin-polarized currents in ballistic nanostructures.
Phys. Rev. Lett. 89, 266602 (2002).
12. Folk, J. A. et al. A gate-controlled bidirectional spin filter using quantum
coherence. Science 299, 679–682 (2003).
13. Hauptmann, J. R. et al. Electric-field-controlled spin reversal in a quantum dot
with ferromagnetic contacts. Nature Phys. 4, 373–376 (2008).
14. Gajek, M. et al. Tunnel junctions with multiferroic barriers. Nature Mater. 6,
296–302 (2007).
15. Meier, F. & Zakharchenya, B. P. Optical Orientation (North-Holland, 1984).
16. Weisbuch, C. & Lampel, G. Spin-dependent recombination and optical spin
orientation in semiconductors. Solid. State. Commun. 14, 141 (1974).
17. Miller, R. C., Tsang, W. T. & Nordland, W. A. Spin-dependent recombination
in GaAs. Phys. Rev. B 21, 1569–1575 (1980).
18. Paget, D. Optical-pumping study of spin-dependent recombination in GaAs.
Phys. Rev. B 30, 931–946 (1984).
19. Kalevich, V. K. et al. Spin-dependent recombination in GaAsN solid solutions.
JETP Lett. 82, 455–458 (2005).
20. Lombez, L. et al. Spin dynamics in dilute nitride semiconductors at room
temperature. Appl. Phys. Lett. 87, 252115 (2005).
21. Lagarde, D. et al. Electron spin dynamics in GaAsN and InGaAsN structures.
Phys. Status Solidi A 204, 208–220 (2007).
22. Buyanova, I. A. & Chen, W. M. Physics and Applications of Dilute Nitrides
(Taylor & Francis Books, 2004).
23. Chen, W. M. Applications of optically detected magnetic resonance in
semiconductor layered structures. Thin Solid Films 364, 45–52 (2000).
24. Baraff, G. A. & Schluter, M. Electronic structure, total energies, and abundances
of the elementary point defects in GaAs. Phys. Rev. Lett. 55, 1327–1330 (1985).
25. Thinh, N. Q. et al. Properties of Ga-interstitial defects in AlGaNP. Phys. Rev. B
71, 125209 (2005).
26. Watkins, G. D. & Corbett, J. W. Defects in irradiated silicon. I. Electron spin
resonance of the Si–A center. Phys. Rev. 121, 1001–1014 (1961).
27. Koh, A. K. & Miller, D. J. Hyperfine coupling constants and atomic parameters
for electron paramagnetic resonance data. Atom. Data Nucl. Data Tables 33,
235–253 (1985).
28. Ohno, Y. et al. Spin relaxation in GaAs(110) quantum wells. Phys. Rev. Lett. 83,
4196–4199 (1999).
29. D’yakonov, M. I. & Kachorovskii, V. Yu. Spin relaxation of two-dimensional
electrons in noncentrosymmetric semiconductors. Sov. Phys. Semicond. 20,
110–112 (1986).
Acknowledgements
W.M.C. and I.A.B. gratefully acknowledge the support from Linköping University
through the Professor Contracts, the Swedish Research council (VR), the Swedish
Energy Agency, the Knut and Alice Wallenberg Foundation, the Wenner-Gren
Foundations and the Swedish Foundation for International Cooperation in Research
and Higher Education (STINT). The work at UCSD is partially supported by NSF
Grant No. DMR- 0606389.
Additional information
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to I.A.B. or W.M.C
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
 LETTERS
PUBLISHED ONLINE: 8 FEBRUARY 2009 DOI: 10.1038/NMAT2382

Towards wafer-size graphene layers by atmospheric

pressure graphitization of silicon carbide
Konstantin V. Emtsev1 , Aaron Bostwick2 , Karsten Horn3 , Johannes Jobst4 , Gary L. Kellogg5 ,
Lothar Ley1 , Jessica L. McChesney2 , Taisuke Ohta5 , Sergey A. Reshanov4 , Jonas Röhrl1 ,
Eli Rotenberg2 , Andreas K. Schmid6 , Daniel Waldmann4 , Heiko B. Weber4 and Thomas Seyller1 *

Graphene, a single monolayer of graphite, has recently
attracted considerable interest owing to its novel magneto-
transport properties1–3 , high carrier mobility and ballistic
transport up to room temperature4 . It has the potential for
technological applications as a successor of silicon in the
post Moore’s law era5–7 , as a single-molecule gas sensor8 ,
in spintronics9–11 , in quantum computing12 or as a terahertz
oscillator13 . For such applications, uniform ordered growth of
graphene on an insulating substrate is necessary. The growth of
graphene on insulating silicon carbide (SiC) surfaces by high-
temperature annealing in vacuum was previously proposed
to open a route for large-scale production of graphene-based
devices5,6 . However, vacuum decomposition of SiC yields
graphene layers with small grains (30–200 nm; refs 14–16).
Here, we show that the ex situ graphitization of Si-terminated
SiC(0001) in an argon atmosphere of about 1 bar produces
monolayer graphene films with much larger domain sizes
than previously attainable. Raman spectroscopy and Hall
measurements confirm the improved quality of the films thus
obtained. High electronic mobilities were found, which reach
µ = 2,000 cm 2 V−1 s−1 at T = 27 K. The new growth process
introduced here establishes a method for the synthesis of
graphene films on a technologically viable basis.
The successful development of graphene-based electronic
devices depends on a large-scale availability of the material.
Several methods for graphene production have been proposed.
Mechanical exfoliation17 leads to isolated, high-quality crystals
with dimensions only in the 10 µm range, which questions
the practicality of this method. Large, high-quality graphene
islands were grown on the surface of transition metals such
as Ru(0001) (ref. 18), but this technique requires transfer to
an insulating substrate, with methods that have yet to be
developed. Liquid-phase exfoliation19 yields high quantities of
monolayer and few-layer graphene. Although this method—in
contrast to graphene oxide reduction20 —prevents the formation
of defects, the electrical properties are similar to graphene oxide
reduction, indicating poor transport at contacts between the
individual graphene sheets.
The preparation of single-layer graphene by the thermal
decomposition of silicon carbide (SiC) has been proposed as a
viable route for the synthesis of uniform, wafer-size graphene layers
for technological applications5–7 . A considerable advantage of this
method is that insulating SiC substrates can be used so that transfer
to another insulator is not required. However, the large-scale
structural quality is limited at present by the lack of continuity
and uniformity of the grown film15,16 . On the Si-terminated (0001)
basal plane, vacuum annealing leads to small graphene domains
typically 30–100 nm in diameter, whereas on the C-terminated
(0001̄) face, larger domains (∼200 nm) of multilayered, rotationally
disordered graphene have been produced14 . The small-grain
structure is due to morphological changes of the surface in the
course of high-temperature annealing. Moreover, decomposition
of SiC is not a self-limiting process and, as a result, regions of
different film thicknesses coexist, as shown by low-energy electron
microscopy15,16 (LEEM). Such inhomogeneous films do not meet
the demands of large-scale device production, which requires
larger domains and tighter thickness control. Homogeneous film
thickness is particularly important because the electronic structure
of the film depends strongly on the number of layers. For
example, although monolayer graphene is a gapless semiconductor,
a forbidden gap can be induced in bilayer graphene and tuned by
an external electrostatic potential21,22 .
Here we demonstrate a method of preparing graphene on
SiC(0001) that results in a significantly improved film quality. In
Fig. 1, we compare samples prepared by vacuum annealing with
samples produced by ex situ annealing under an argon atmosphere.
Figure 1a–c shows the morphology of the 6H–SiC(0001) surface
before and after the formation of a graphene monolayer by
annealing in ultrahigh vacuum (UHV) as determined by atomic
force microscopy (AFM) and LEEM. The initial 6H–SiC(0001)
surface in Fig. 1a, obtained after hydrogen etching, is characterized
by wide, highly uniform, atomically flat terraces. The step direction
and terrace width (of the order of 300–700 nm) are determined by
the incidental misorientation of the substrate surface with respect
to the crystallographic (0001) plane. The step height is 1.5 nm,
which corresponds to the dimension of the 6H–SiC unit cell in the
direction perpendicular to the surface (c axis). On defect-free areas
of the sample, the terraces typically extend undisturbed over 50 µm
in length. The morphology of the surface covered with a monolayer
of graphene prepared by vacuum annealing is shown in Fig. 1b.
The surface obviously undergoes significant modifications; it is now
covered with small pits up to 10 nm in depth, and the original

1 Lehrstuhlfür Technische Physik, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erwin-Rommel-Str. 1, 91058 Erlangen, Germany, 2 Advanced Light
Source, Lawrence Berkeley National Laboratory, One Cyclotron Road, Berkeley, California 94720, USA, 3 Department of Molecular Physics,
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, 14195 Berlin, Germany, 4 Lehrstuhl für Angewandte Physik,
Friedrich-Alexander-Universität Erlangen-Nürnberg, Staudtstr. 7, 91058 Erlangen, Germany, 5 Sandia National Laboratories, Surface & Interface Sciences
Department, PO Box 5800, Albuquerque, New Mexico 87185, USA, 6 National Center for Electron Microscopy, Lawrence Berkeley National Laboratory,
One Cyclotron Road, Berkeley, California 94720, USA. *e-mail: thomas.seyller@physik.uni-erlangen.de.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 203

LETTERS NATURE MATERIALS DOI: 10.1038/NMAT2382

a b c

720 nm 1.0 µm
1 µm E = 2.9 eV

0 ML 1 ML 2 ML

d e

8.0 µm
10 µm
E = 5.2 eV

f g i A A
1 1

2 2
Vertical position

2.0 µm 2.0 µm
B B

1 1 A B
1 2 1 32 1 2 1
2 2 40
3 3
Height (nm)

30
4 µm 1 1
E = 5.2 eV
20
0 5 10 15
h Energy (eV)
0 10
Energy (eV)

5 0
0 2 4 6 8 10
10 Distance (µm)

15
Horizontal position

Figure 1 | Morphological changes of 6H–SiC(0001) during graphene growth. a, Initial surface after H-etching imaged by AFM. The step height is 15 Å.
b, AFM image of graphene on 6H–SiC(0001) with a nominal thickness of 1 ML formed by annealing in UHV at a temperature of about 1,280 ◦ C. c, LEEM
image of a UHV-grown graphene film on SiC(0001) with a nominal thickness of 1.2 monolayers. The image contrast is due to the locally different layer
thickness. Light, medium and dark grey correspond to a local thickness of 0, 1 and 2 ML, respectively. d, AFM image of graphene on 6H–SiC(0001) with a
nominal thickness of 1.2 ML formed by annealing in Ar (p = 900 mbar, T = 1,650 ◦ C). e, LEEM image of a sample equivalent to that of d revealing
macro-terraces covered with graphene up to 50 µm long and at least 1 µm wide. f, Close-up LEEM image revealing monolayer coverage on the terraces and
bilayer/trilayer growth at the step edges. g,h, Electron reflectivity spectra (grey-scale images) taken at the positions indicated by the blue lines in f.
Monolayer, bilayer and trilayer graphene are readily identified by the presence of 1, 2 or 3 reflectivity minima, respectively. i, Close-up AFM images of the
film shown in d. In the right-hand-side image, the z scale was adjusted such that the terraces appear at the same height. The profile shows that small
depressions 4 and 8 Å in height exist at the step edges due to second and third layer nucleation.

steps are hardly discernible any longer. This indicates that graphene
growth is accompanied by substantial changes in the morphology
of the substrate itself, leading to a considerable roughening. As a
consequence of this roughening, the graphene layer acquires an
inhomogeneous thickness distribution as can be seen in the LEEM
image shown in Fig. 1c. The irregularly shaped graphene islands
are at most a few hundred nanometres in size, in agreement with
X-ray diffraction14 . Moreover, monolayer graphene areas coexist
with graphene
√ √ bilayer islands as well as with uncovered regions of
the (6 3×6 3) buffer layer23 .

204 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2382 LETTERS
a b

Intensity (arb. units)

G

SiC

S2

S1

288 287 286 285 284 283 282

Binding energy (eV)

c d
G-peak
(1,592 ± 5) cm¬1 2D-peak
0 (2,706 ± 5) cm¬1

D-peak
(1,356 ± 5) cm¬1
Binding energy (eV)

Intensity (arb. units)

54 cm¬1
ED
0.5

37 cm¬1
1.0

(2,717 ± 5) cm¬1
(1,596 ± 5) cm¬1
1.5
1.5 1.6 1.7 1.8 1.9 1,400 1,600 1,800 2,600 2,800
Electron momentum Kll (Ŭ1) Raman shift (cm¬1)

Figure 2 | Atomic and electronic structure of ex-situ-grown monolayer graphene. a, LEED pattern at 74 eV showing the diffraction spots due
√ √
to the SiC(0001) substrate (blue arrows) and the graphene lattice (red arrows). The extra spots are due to the (6 3 × 6 3) interface layer. b, C1s
√ √
core-level spectrum measured at a photon energy of 700 eV. The spectrum contains contributions from the SiC substrate (marked SiC), the (6 3 × 6 3)
interface layer (marked S1 and S2) and from the graphene layer (G) residing on top of the interface layer. c, π-bands probed by ARPES in the vicinity of the
K-point of the hexagonal Brillouin zone measured along the 0 K-direction. The position of the Dirac energy (ED ) at 0.45 eV below the Fermi energy is
consistent with previous reports on UHV-grown graphene on SiC(0001). Faint features marked by yellow arrows signal the presence of small regions of
bilayer graphene in agreement with the LEEM results. d, Comparison of Raman spectra of Ar-grown (red) and UHV-grown (blue) epitaxial graphene on
6H–SiC(0001). The spectra of the D- and G-line shown here are corrected for the emission of the substrate by subtraction of a reference spectrum26
(see Supplementary Information).

In stark contrast to the low quality resulting from vacuum
graphitization (Fig. 1b), films grown under 900 mbar of argon have
a greatly improved surface morphology, as demonstrated by the
AFM image in Fig. 1d. Large continuous terraces are found, extend-
ing over large distances parallel to the step edges. Step bunching is
manifested by the formation of macro-terraces that are a factor of
5–8 times wider than the original terraces. The macro-steps are par-
allel to the original steps, increase in step height by the same factor
and reach average heights of 8–15 nm. Parallel to the steps, uninter-
rupted macro-terraces more than 50 µm long have been observed.
The thickness distribution of the graphene film grown under an
argon atmosphere is determined by LEEM as shown in Fig. 1e,f.
A series of spatially resolved LEEM I –V spectra taken along a
vertical and a horizontal line in Fig. 1f is shown in Fig. 1g,h. The
layer thickness is easily determined from the number of minima
in the individual spectra; the LEEM image taken at a particular
energy shows stripes that follow in width and orientation the
macro-terraces with a contrast that is determined by the graphene
layer thickness15,16 . Hence, we can unambiguously conclude that
except for narrow stripes at the edges, the large atomically
flat macro-terraces are homogeneously covered with a graphene
monolayer. The domain size of monolayer graphene is significantly
larger than that of the vacuum-annealed samples as a comparison
between Fig. 1c,f shows. In fact, the domain size seems to be
limited only by the length and width of the SiC terraces. Narrower,
darker regions at the downward edges of the terraces correspond
to bilayer and in some cases trilayer graphene (see region 3 in
Fig. 1f). In the AFM image, these regions (see Fig. 1i) appear
as small depressions of around 0.5 nm and 1 nm amplitude
located at the very edge of the macro-step. This indicates that
the nucleation of new graphene layers starts at step edges of
the substrate surface.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 205

LETTERS
The structural and electronic properties of Ar-grown graphene
layers were probed by low-energy electron diffraction (LEED),
photoelectron spectroscopy and Raman spectroscopy as shown
in Fig. 2. Although these methods alone cannot assess the
morphological quality of the sample surface on a large scale, they
provide extra information as detailed below. The LEED pattern
(Fig. 2a) demonstrates that the graphene layer is well ordered and
aligned with respect to the substrate, such that the basal plane
unit vectors of graphene and SiC subtend an angle of 30◦ . The
C1s core-level spectrum (Fig. 2b) shows√the characteristic signals
√ Van der Pauw
of the SiC substrate, the non-metallic (6 3 × 6 3) interface layer
(buffer layer) that decouples the electronic structure of graphene
from the substrate and the graphene monolayer that lies on top of
the buffer layer, in agreement with vacuum-grown films23 . Angle-
resolved photoelectron spectra (Fig. 2c) reveal the characteristic
band structure of monolayer graphene24 . Note that, as for vacuum-
grown layers24 , the Dirac point (ED ) is shifted below the Fermi
level (EF ) owing to electron doping (n ≈ 1.1 × 1013 cm−2 ) from the
substrate. Figure 2d compares Raman spectra of epitaxial graphene
grown in UHV and in Ar. For the Ar-grown sample, the G- and 2D-
lines are observed at 1,596 and 2,717 cm−1 , respectively. The 2D-line
consists of a narrow peak (full-width at half-maximum: 37 cm−1 ),
which is characteristic of monolayer graphene25,26 . Its blueshift
(38 cm−1 ) with respect to exfoliated graphene25 was previously
explained by compressive strain26 . The UHV-grown sample shows a
broader and less symmetric 2D-peak, which clearly indicates that it
contains more than one component in agreement with the broader
thicknesses distribution determined by LEEM. Its smaller blueshift
(29 cm−1 ) is probably due to a partial strain relief of the smaller
domains. A strong D-peak at 1,356 cm−1 signals the presence of
many defects and domain boundaries. Therefore, although our
epitaxial growth process results in a marked improvement in surface
morphology, all other properties such as orientation with respect to
the substrate, electronic structure and charge carrier density remain
unaltered as compared to vacuum-grown layers.
What is the reason for the observed improvement of the
surface morphology of the Ar-annealed samples compared with the
samples annealed in UHV? From the data in Fig. 1, it is clear that
the surface undergoes considerable morphological changes at the
temperature where graphitization occurs. The large roughness of
the UHV-annealed samples suggests that the surface is far from
equilibrium, such that a transformation to a smooth morphology
cannot be achieved under these conditions. The key factor in
achieving an improved growth is the significantly higher annealing
temperature of 1,650 ◦ C that is attainable for graphene formation
under argon at a pressure of 900 mbar as compared with 1,280 ◦ C in
UHV. Graphene formation is the result of Si evaporation from the
substrate. For a given temperature, the presence of a high pressure
of argon leads to a reduced Si evaporation rate because the silicon
atoms desorbing from the surface have a finite probability of being
reflected back to the surface by collision with Ar atoms, as originally
pointed out by Langmuir27,28 . Indeed, in the presence of the Ar
atmosphere, no sublimation of Si from the surface is observed at
temperatures up to 1,500 ◦ C, whereas Si desorption commences
at 1, 150 ◦ C in vacuum (see Supplementary Information). The
significantly higher growth temperature thus attained results in an
enhancement of surface diffusion such that the restructuring of the
surface is completed before graphene is formed. Ultimately, this
leads to the markedly improved surface morphology that we observe
here. The macro-step structure is also responsible for the tighter
thickness control. New graphene layers start to grow from the step
edges; hence having fewer steps along well-defined crystallographic
directions reduces the nucleation density of multilayer graphene.
To evaluate the electronic quality of our graphene layers, we
determined the carrier mobility of monolayer epitaxial graphene
on SiC(0001) using Hall effect measurements. Table 1 compares
206
NATURE MATERIALS DOI: 10.1038/NMAT2382
Table 1 | Hall mobilities (in cm2 V−1 s−1 ) for Hall bars and van
der Pauw structures on UHV- and Ar-grown graphene
measured at T = 300 and 27 K.
Method Structure 300 K 27 K
Hall bar 900 1,850
Ar
Van der Pauw 930 2,000
Hall bar 470 —
UHV
550 710
carrier mobilities for Ar-grown and UHV-grown samples measured
on Hall bars and in van der Pauw geometry. The mobilities obtained
by Ar annealing are higher by about a factor of two at 300 K and
by three at 27 K. This suggests that the large number of domain
boundaries in vacuum-grown graphene has a major negative impact
on carrier mobility. On the other hand, no significant difference
in electron mobility was observed between the two geometries,
indicating that the step edges have a minor role, similar to graphene
field-effect transistors in multilayer graphene29 . Note that scanning
tunnelling microscopy has provided evidence that graphene layers
are continuous over step edges5,30 and graphene field-effect
transistors with multilayer graphene29 show no dependence on
the alignment with respect to substrate steps. However, extra
experiments with Hall bars aligned parallel and perpendicular
to steps will be carried out to unambiguously clarify the effect
of steps on carrier mobility. The electron density determined
by the Hall measurements is n ≈ 1 × 1013 cm−2 , in agreement
with the angle-resolved photoelectron spectroscopy (ARPES)
measurements. Hence, we compare the measured mobilities with
exfoliated monolayer graphene on SiO2 in the high doping limit,
where values of the order of 10,000 cm2 V−1 s−1 are reported
for n ≈ 5 × 1012 cm−2 (refs 1,2). For epitaxial graphene on
SiC(0001), other groups reported values of 1,200 cm2 V−1 s−1 for
multilayered graphene5,29 .
In summary, we have shown that the growth of epitaxial
graphene on SiC(0001) in an Ar atmosphere close to atmospheric
pressure provides morphologically superior graphene layers in
comparison with vacuum graphitization. Extensive step bunch-
ing taking place during processing yields arrays of parallel ter-
races up to 3 µm wide and more than 50 µm long. The terraces
are essentially completely and homogeneously covered with a
monolayer of graphene. At present, downward step edges, where
the initiation of second- and third-layer graphene growth is de-
tected, are prohibiting an even larger extension of the mono-
layer graphene domains. Because the substrate step direction
and step width are determined by the magnitude and azimuthal
orientation of the surface misorientation with respect to major
crystallographic directions, a proper choice of these parameters
controls terrace width and length and hence the ultimate size of
the graphene domains. An improved substrate quality in terms
of crystallographic orientation is therefore expected to lead to
further improvements. In comparison to UHV treatment, the
technique presented here is much closer to standard prepara-
tion conditions in semiconductor manufacturing, permitting the
use of standard chemical vapour deposition equipment for the
fabrication of graphene layers. All necessary processing steps
(hydrogen etching and graphene synthesis) can be carried out
in a single reactor. Electrical measurements confirm the im-
proved film quality.
Methods
Graphene layers were synthesized on commercial, nominally on-axis oriented
wafers of 6H–SiC(0001) purchased from SiCrystal AG. For experiments concerning
the morphology of the surface and for the surface science experiments, we used
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2382
nitrogen-doped substrates with a doping concentration of 1–2 × 1018 cm−3 .
For electrical measurements (see below), semi-insulating 6H–SiC(0001) was
used. Before graphene epitaxy, the samples were etched in hydrogen (grade 5.0,
p = 1 bar, T = 1,550 ◦ C, t = 15 min) to remove surface polishing damage. Graphene
growth was carried out in a vertical cold-wall reactor comprising a double-walled,
water-cooled quartz tube and a graphite susceptor in a slow flow of argon (purity
5.0). Heating and cooling rates were 2–3 ◦ C per second. Typical annealing time
was 15 min. A wide range of annealing temperatures from 1,500 to 2,000 ◦ C and
reactor gas pressures from 10 to 900 mbar were tested. However, except for very
low pressures studied, the morphology of the surface after graphene formation in
an Ar atmosphere is generally much smoother and the graphene domain size much
larger compared with vacuum annealing.
Surface composition and graphene thicknesses were determined from
core-level photoelectron spectroscopy by means of a Specs PHOIBOS150
analyser in combination with a monochromatized Al Kα source with an energy
resolution of ∼350 meV. Owing to the chemical inertness of graphene, the
samples can be easily transported through air. As-prepared graphene samples
showed no detectable oxygen on the surface (below 1% of a monolayer) even
after air exposure for about 1 h. Prolonged air exposure, however, leads to a
fractional layer of physisorbed hydrocarbons and water, which can be removed
by annealing in vacuum at around 600 ◦ C. ARPES measurements were carried
out at the Advanced Light Source using a Scienta R4000 analyser with an overall
resolution of ∼25 meV at a photon energy of 94 eV and at a sample temperature
of 20 K. Core-level measurements were carried out at BESSY-II with a Specs
PHOIBOS150 analyser with a resolution of ∼125 meV at a photon energy
of 700 eV and at room temperature. The surface morphology was probed by
AFM in non-contact mode. LEEM measurements at room temperature with
a spatial resolution better than 10 nm were carried out at Sandia National
Laboratory and at the National Center for Electron Microscopy, Lawrence Berkeley
National Laboratory. The laterally averaged graphene thickness determined
by LEEM is in perfect agreement with the average layer thickness obtained
by X-ray photoelectron spectroscopy. The crystal structure of the films was
monitored by LEED.
Raman spectra were measured at room temperature by means of a triple
spectrometer equipped with a liquid-nitrogen-cooled CCD (charge-coupled
device) detector. A frequency-doubled NdYVO4 laser with a wavelength of
532 nm was used for excitation. A confocal optical microscope was used to record
micro-Raman spectra with a spatial resolution of 2 µm.
For the electrical characterization, the samples were patterned by two electron
beam lithography steps. The first step defined the graphene film (undesired areas
were etched with oxygen plasma). A second step defined the contact pads, which
consist of a thermally evaporated Ti/Au double layer, patterned by a standard
lift-off technique. Two different geometries were investigated: square graphene
films (100 µm × 100 µm) with contact pads at the four corners for van der Pauw
measurements as well as Hall bars (4 µm×50 µm) placed along macro-terraces. The
measurements were carried out in a continuous-flow cryostat (sample in vacuum),
using magnetic fields of ±0.66 T at temperatures between 300 and 27 K.
Received 14 July 2008; accepted 7 January 2009;
published online 8 February 2009
References
1. Novoselov, K. S. et al. Two-dimensional gas of massless Dirac fermions in
graphene. Nature 438, 197–200 (2005).
2. Zhang, Y. B., Tan, Y. W., Stormer, H. L. & Kim, P. Experimental observation
of the quantum Hall effect and Berry’s phase in graphene. Nature 438,
201–2004 (2005).
3. Novoselov, K. S. et al. Unconventional quantum Hall effect and Berry’s phase
of 2π in bilayer graphene. Nature Phys. 2, 177–180 (2006).
4. Novoselov, K. S. et al. Room-temperature quantum Hall effect in graphene.
Science 315, 1379 (2007).
5. Berger, C. et al. Ultrathin epitaxial graphite: 2D electron gas properties
and a route toward graphene-based nanoelectronics. J. Phys. Chem. B 108,
19912–19916 (2004).
6. Berger, C. et al. Electronic confinement and coherence in patterned epitaxial
graphene. Science 312, 1191–1196 (2006).
7. Geim, A. K. & Novoselov, K. S. The rise of graphene. Nature Mater. 6,
183–191 (2007).
8. Schedin, F. et al. Detection of individual gas molecules adsorbed on graphene.
Nature Mater. 6, 652–655 (2007).
9. Son, Y.-W., Cohen, M. L. & Louie, S. G. Half-metallic graphene nanoribbons.
Nature 444, 347–349 (2006).
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
LETTERS
10. Trauzettel, B., Bulaev, D. V., Loss, D. & Burkard, G. Spin qubits in graphene
quantum dots. Nature Phys. 3, 192–196 (2007).
11. Yokoyama, T. Controllable spin transport in ferromagnetic graphene junctions.
Phys. Rev. B 77, 073413 (2008).
12. Fal’ko, V. I. Graphene: Quantum information on chicken wire. Nature Phys. 3,
151–152 (2007).
13. Rana, F. Graphene terahertz plasmon oscillators. IEEE Trans. Nanotech. 7,
91–99 (2008).
14. Hass, J. et al. Highly ordered graphene for two dimensional electronics.
Appl. Phys. Lett. 89, 143106 (2006).
15. Hibino, H. et al. Microscopic thickness determination of thin graphite films
formed on SiC from quantized oscillation in reflectivity of low-energy electrons.
Phys. Rev. B 77, 075413 (2008).
16. Ohta, T. et al. Morphology of graphene thin film growth on SiC(0001).
New J. Phys. 10, 023034 (2008).
17. Novoselov, K. S. et al. Two-dimensional atomic crystals.
Proc. Natl Acad. Sci. USA 102, 10451–10453 (2005).
18. Sutter, P. W., Flege, J.-I. & Sutter, E. A. Epitaxial graphene on ruthenium.
Nature Mater. 7, 406–411 (2008).
19. Hernandez, Y. et al. High-yield production of graphene by liquid-phase
exfoliation of graphite. Nature Nanotech. 3, 563–568 (2008).
20. Gomez-Navarro, C. et al. Electronic transport properties of individual
chemically reduced graphene oxide sheets. Nano Lett. 7, 3499–3503 (2007).
21. McCann, E. & Fal’ko, V. I. Landau-level degeneracy and quantum Hall effect
in a graphite bilayer. Phys. Rev. Lett. 96, 086805 (2006).
22. Ohta, T. et al. Controlling the electronic structure of bilayer graphene. Science
313, 951–954 (2006).
23. Emtsev, K. V. et al. Interaction, growth, and ordering of epitaxial graphene
on SiC{0001} surfaces: A comparative photoelectron spectroscopy study.
Phys. Rev. B 77, 155303 (2008).
24. Bostwick, A. et al. Quasiparticle dynamics in graphene. Nature Phys. 3,
36–40 (2007).
25. Ferrari, A. C. et al. Raman spectrum of graphene and graphene layers.
Phys. Rev. Lett. 97, 187401 (2006).
26. Röhrl, J. et al. Raman spectra of epitaxial graphene on SiC(0001).
Appl. Phys. Lett. 92, 201918 (2008).
27. Langmuir, I. Convection and conduction of heat in gases. Phys. Rev. (Series I)
34, 401–422 (1912).
28. Fonda, G. R. Evaporation of tungsten under various pressures of argon.
Phys. Rev. (Series II) 31, 260–266 (1928).
29. Kedzierski, J. et al. Epitaxial graphene transistors on SiC substrates. IEEE Trans.
Electron Devices 55, 2078–2085 (2008).
30. Seyller, Th. et al. Structural and electronic properties of graphite layers grown
on SiC(0001). Surf. Sci. 600, 3906–3911 (2006).
Acknowledgements
We thank F. El Gabaly for assistance with the LEEM measurements and M. Gick
for help with the sample preparation. We gratefully acknowledge support by the
DFG under contract SE 1087/5-1, contract WE 4542-5-1, and within the Cluster
of Excellence ‘Engineering of Advanced Materials’ (www.eam.uni-erlangen.de) at
the Friedrich-Alexander-Universität Erlangen-Nürnberg, by BaCaTeC, and by the
BMBF under contract 05 ES3XBA/5. A part of the work was carried out at Sandia
National Laboratories, a multiprogram laboratory operated by Sandia Corporation, a
Lockheed Martin company, for the United States Department of Energy, Office of Basic
Energy Sciences, Division of Materials Science and Engineering, under Contract No.
DE-AC04-94AL85000. The work carried out at the Advanced Light Source was supported
by the Director, Office of Science, Office of Basic Energy Sciences, of the US Department
of Energy under Contract No. DE-AC03-76SF00098.
Author contributions
K.V.E. developed the growth process of large-area graphene with the help of S.A.R.
AFM measurements were done by K.V.E. and T.O. T.O. and A.S. carried out the LEEM
measurements with the help of G.K. Photoelectron spectroscopy measurements were
carried out by K.V.E., T.S., A.B., J.L.M., E.R. and K.H. J.J., D.W. and H.B.W carried out
lithography and electrical measurements. J.R. carried out Raman measurements. T.S.,
K.V.E. and L.L. wrote the manuscript with revision and input from all other co-authors.
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to T.S.
207
 LETTERS
PUBLISHED ONLINE: 8 FEBRUARY 2009 DOI: 10.1038/NMAT2379

Endohedral fullerenes for organic photovoltaic

devices
Russel B. Ross1 , Claudia M. Cardona2 , Dirk M. Guldi3 , Shankara Gayathri Sankaranarayanan3 ,
Matthew O. Reese4 , Nikos Kopidakis4 , Jeff Peet5 , Bright Walker6 , Guillermo C. Bazan5 ,
Edward Van Keuren1 , Brian C. Holloway2 and Martin Drees2 *

So far, one of the fundamental limitations of organic
photovoltaic (OPV) device power conversion efficiencies
(PCEs) has been the low voltage output caused by a molecular
orbital mismatch between the donor polymer and acceptor
molecules. Here, we present a means of addressing the
low voltage output by introducing novel trimetallic nitride
endohedral fullerenes (TNEFs) as acceptor materials for use
in photovoltaic devices. TNEFs were discovered in 1999 by
Stevenson et al.1 ; for the first time derivatives of the TNEF
acceptor, Lu3 N@C80 , are synthesized and integrated into OPV
devices. The reduced energy offset of the molecular orbitals
of Lu3 N@C80 to the donor, poly(3-hexyl)thiophene (P3HT),
reduces energy losses in the charge transfer process and
increases the open circuit voltage (Voc ) to 260 mV above
reference devices made with [6,6]-phenyl-C61 -butyric methyl
ester (C60 -PCBM) acceptor. PCEs >4% have been observed
using P3HT as the donor material. This work clears a path
towards higher PCEs in OPV devices by demonstrating that
high-yield charge separation can occur with OPV systems that
have a reduced donor/acceptor lowest unoccupied molecular
orbital energy offset.
The trimetallic nitride endohedral fullerenes (TNEFs) used in
this study are Ih -C80 fullerenes incarcerating rare-earth metals
bound in a trimetallic nitride cluster. The spherically symmetric
C80 and trimetallic nitride cluster do not independently exist, but
when combined together form a stable molecule. Owing to the
large variety of metals that can be used to make the endohedral
cluster, these TNEF nanomaterials are multifunctional, and have
been sought for their unique properties in applications such as
magnetic resonance imaging contrast agents2 and optoelectronics3 .
Recent theoretical4 and experimental data5 suggest that the lowest
unoccupied molecular orbital (LUMO) energies for this type of
fullerene are much higher than the LUMO energy of empty-cage
species with respect to the Fermi level. The higher LUMO provides
a path towards higher Voc , and therefore, higher efficiencies in
organic photovoltaic (OPV) devices. Here, Lu3 N@C80 was chosen
because its LUMO energy is closer to the commonly used OPV
donor polymer poly(3-hexyl)thiophene’s (P3HT) LUMO level than
are those of other available fullerenes such as C60 (refs 6,7).
Photophysics and OPV device measurements demonstrate that
the minimum LUMO offset needed for exciton dissociation is
satisfied by Lu3 N@C80 ’s LUMO energy7 . Cyclic voltammetry and
Osteryoung square-wave voltammetry (OSWV), shown in Fig. 1
and Table 1, confirm that the reduction potential and therefore
the LUMO levels of the Lu3 N@C80 and its methano derivatives
are 204 and 280 mV, respectively, more negative than that of
[6,6]-phenyl-C61 -butyric methyl ester (C60 -PCBM), the most
common fullerene acceptor used in OPV devices so far. The open
circuit voltage in OPVs is correlated to the difference between
the donor’s highest occupied molecular orbital (HOMO) and the
acceptor’s LUMO (refs 7–10). Therefore, the increase in reduction
potential is representative of the expected Voc gain from Lu3 N@C80 -
based OPV devices9,11 .
Similar to empty-cage fullerenes, the solubility of pristine
Lu3 N@C80 is not high enough for solution processing of
OPVs. Thus, an exohedral functionalization of Lu3 N@C80 was
required to allow its incorporation into solution-processed bulk-
heterojunction OPV devices. Here, we report the first methano
derivatives of Lu3 N@C80 that have been isolated and fully char-
acterized. It is important to emphasize that the chemical reactivity
of Lu3 N@C80 differs from empty-cage fullerenes. Owing to these
differences in reactivity5,12–14 , the standard protocol used for the
synthesis of C60 -PCBM (ref. 15) had to be significantly modified
to obtain the Lu3 N@C80 -PCBX (X = M,B,H,O) analogues. For
example, the reagents had to be increased to 25 times the amounts
used for empty-cage fullerenes, the reaction was run at 120 ◦ C
instead of 70 ◦ C and the highest yield was obtained after 25 min
rather than 22 h, see Fig. 2 and Supplementary Methods. The
Lu3 N@C80 -PCBX family of derivatives—PCBM (methyl), PCBB
(butyl), PCBH (hexyl) and PCBO (octyl)—have proven to have
thermal stability up to 300 ◦ C in a thermal gravimetric analyser, see
Supplementary Fig. S1. The absorption spectra of bare Lu3 N@C80
(Fig. 3) resembles closely its PCBX derivatives, which suggests
that this class of TNEF derivatives have a fulleroid character16 .
Electrochemical characterization of these Lu3 N@C80 -PCBX deriva-
tives also showed the kinetically reductive irreversible behaviour
of the pristine Lu3 N@C80 and they retained the same reduction
potential advantage compared to C60 -PCBM, but their solubility in
organic solvents and miscibility in P3HT varied significantly. The
ability to vary the solubility and miscibility by exchanging the X
portion of the PCBX functional group on TNEF molecules and not
affect the molecular orbitals can serve as a vital tool for designing

1 Georgetown University, 37th and O st. NW, Washington, District of Columbia 20057, USA, 2 Luna Innovations Incorporated, 521 Bridge Street, Danville,
Virginia 24541, USA, 3 Department of Chemistry and Pharmacy & Interdisciplinary Center for Molecular Materials (ICMM), Friedrich-Alexander-
Universität Erlangen-Nürnberg, Egerlandstr. 3, 91058 Erlangen, Germany, 4 National Renewable Energy Laboratory, 1617 Cole Blvd, Golden, Colorado
80401, USA, 5 Center for Polymer and Organic Solids, University of California, Santa Barbara, California 93106, USA, 6 Department of Chemistry and
Biochemistry (or Center for Polymers and Organic Solids) University of California at Santa Barbara, Santa Barbara, California 93117, USA.
*e-mail: dreesm@lunainnovations.com.

208 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2379 LETTERS
a
Table 1 | Redox potentials (V versus Fc/Fc+ ) of the first
oxidation and first reduction processes measured by OSWV
O in 0.05 M n-Bu4 NPF6 /o-DCB.
O
(1) Ep,ox(1) Ep,red(1)
C60
C60 -PCBM +1.132 −1.220
Lu3 N@C80 +0.635 −1.424
Lu3 N@C80 -PCBM +0.556 −1.510
Lu3 N@C80 -PCBH +0.564 −1.500

(2) Lu3N
OH H+ O
O O Hexanol O O
O p-toluenesulphonyl hydrazide
CH3CN
O
O
(3) Lu3N NO
NH
O S O

O
CH3
O NaOCH3
(4) Pyridine
Lu3N
o-DCB O
N2 O
Lu3N O
+
0 ¬0.5 ¬1.0 ¬1.5 ¬2.0 ¬2.5 ¬3.0 N⊕ O ∼120 ° C
Lu3N
V versus Fc/Fc+ N

b Figure 2 | Synthetic protocol for Lu3 N@C80 -PCBH.

1 µA
[C60]PCBM
1 µA
[Lu3N@C80]PCBH
0 ¬0.5 ¬1.0 ¬1.5 ¬2.0 ¬2.5 ¬3.0
V versus Fc/Fc+
Figure 1 | TNEF electrochemical properties to establish LUMO levels.
a, Cyclic voltammograms of the reductive behaviour of (1) C60 -PCBM,
(2) Lu3 N@C80 , (3) Lu3 N@C80 -PCBM, (4) Lu3 N@C80 -PCBH in 0.5 M
n-Bu4 NPF6 /o-DCB (o − DCB : 1,2 − dichlorobenzene) with ferrocene as the
internal standard, 100 mV s−1 scan rate. b, OSWV measurements of
C60 -PCBM and Lu3 N@C80 in 0.05 M n-Bu4 NPF6 /o-DCB.
the optimal composite morphology for high-performance OPV
devices. Lu3 N@C80 -PCBH was chosen among the synthesized
Lu3 N@C80 -PCBX derivatives for this study because of its similarities
in solubility and miscibility to C60 -PCBM in the processing
of P3HT-based OPV devices. Characterization of space-charge-
limited devices further confirmed that Lu3 N@C80 -PCBH possesses
similar charge-carrier mobility to C60 -PCBM (4.0 × 10−4 for
Lu3 N@C80 -PCBH versus 1.4 × 10−3 cm2 V−1 s−1 for C60 -PCBM),
see Supplementary Fig. S2.
Photophysical studies were carried out to understand the
excited-state interaction of the Lu3 N@C80 derivatives with the
P3HT donor polymer. Charge transfer between photoexcited
TNEF and P3HT was confirmed for the first time by means of
subnanosecond photolytic experiments. Specifically, we contrasted
the photophysical behaviour in spin-coated films with that in
solution by exciting either at 387 nm (fullerenes) or at 540 nm
(P3HT) and comparing the resulting photospectra with those of
the corresponding radical cation and radical anion produced in
radiolytic experiments, see Supplementary Fig. S3. Initially the
P3HT singlet excited-state features are discernible in photoexcited
films of P3HT/Lu3 N@C80 -PCBH or P3HT/C60 -PCBM—see Fig. 4
and Supplementary Fig. S4. However, they are in both cases
ultrashort: they decay in less than 550 fs (>2.0 × 1012 M−1 ) to
form a new photoproduct. There is a decisive difference between
the photoproducts of the two acceptors: in the near-infrared
region, the photoproducts absorb with maxima at 665, 890
and 1,020 nm for P3HT/C60 -PCBM and 665, 890 and 1,025 nm
for P3HT/Lu3 N@C80 -PCBH. The first two maxima (665 and
890 nm) are in excellent agreement with those noted for the
one-electron oxidized radical cation of P3HT in dichloromethane
(600 and 855 nm—Supplementary Fig. S3a); the last ones (1,020
or 1,025 nm) correspond to the one-electron reduced C60 -PCBM
and Lu3 N@C80 -PCBH radical anions generated in toluene, acetone
and 2-propanol (see Supplementary Fig. S3b). The visible portion
of the transient absorption spectra, on the other hand, is in
both cases dominated by strong bleaching of the ground state,
revealing minima at 500 nm. These fingerprints again corroborate
the attributes seen during the P3HT oxidation. In other words,
both Lu3 N@C80 -PCBH and C60 -PCBM give rise to the rapid
formation of the radical ion pair states, under excitation of
either the donor or the acceptor. These radical ion pair states
were found to be stable on the timescale of our investigation
(up to 3,000 ps). Complementary nanosecond experiments (see
Supplementary Fig. S5) confirm the remarkable stability of the
radical ion pair states in both films with lifetimes exceeding
the experimental time window of our apparatus (1.0 ms). These
transient absorption measurements with P3HT/Lu3 N@C80 -PCBH

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 209

LETTERS
0.08
0.06
Absorbance
0.04
0.02
0
500 600 700 800
Wavelength (nm)
Lu3N@C80-PCBO
Lu3N@C80-PCBH
Lu3N@C80-PCBB
Lu3N@C80-PCBM
Lu3N@C80
300 400 500 600 700 800
Wavelength (nm)
Figure 3 | Absorption spectra of Lu3 N@C80 and its methano derivatives.
Inset: Spectra for all four methano derivatives with their parent molecule
Lu3 N@C80 overlaid and magnified in the absorption onset region.
and P3HT/C60 -PCBM films on quartz, photoexciting P3HT at
540 nm, revealed the same amplitudes for the P3HT radical cation
absorptions at 855 nm, which indicates that the charge transfer
process for the Lu3 N@C80 -PCBH-based film is as efficient as that
seen for C60 -PCBM. The presented efficiency and stability of the
charge-separated states of P3HT/Lu3 N@C80 -PCBH should yield
photocurrent densities in OPV devices that are equal to or greater
than the P3HT/C60 -PCBM reference system.
OPV devices using a P3HT/Lu3 N@C80 -PCBH active layer
show a significant increase in open circuit voltage compared
with P3HT/C60 -PCBM reference devices. Figure 5a illustrates the
illuminated and dark current density (J ) versus voltage (V )
curves for an optimized P3HT/Lu3 N@C80 -PCBH device with a
PCE of 4.2%. The measured P3HT/Lu3 N@C80 -PCBH device has a
similar photocurrent and fill factor (FF) to the P3HT/C60 -PCBM
reference device, thereby demonstrating that the higher Voc of
the TNEF-based OPV device leads to a higher PCE. The short-
circuit current densities (Jsc ) of both devices were cross-referenced
with an integration of the external quantum efficiency (EQE)
measurements. Both of the devices’ EQE spectra shown in Fig. 5b
integrated within 2% of the measured Jsc under solar simulation
tests. It is important to note here that there were differences in
the fabrication conditions to achieve optimum performance for
each of the two devices shown in Fig. 5. The P3HT/C60 -PCBM
reference film was annealed in an inert atmosphere for 10 min
at 150 ◦ C after deposition of a lithium fluoride/aluminium top
electrode. The optimal processing for the P3HT/Lu3 N@C80 -PCBH
film was achieved by annealing the film at 110 ◦ C for 10 min
before the LiF/Al top electrode was deposited with an extra 30 s
anneal at 140 ◦ C after production. The optimal processing of
these two devices differs slightly; however, the absorption spectra
and grazing incident X-ray diffraction data shown in Fig. 5c,d
demonstrate that the blend films are similar in overall absorption
and polymer order17–20 . A 12% lower crystallinity of the P3HT phase
has been interpreted from the grazing incident X-ray diffraction
data, and is one example of how future devices made with
P3HT/Lu3 N@C80 -PCBH active layers may be improved in Jsc and
FF. The similar absorption of these two films combined with
the magnitude of the extracted photocurrent further validates the
charge transfer efficiency of the P3HT/Lu3 N@C80 -PCBH system
measured during photolytic experimentation.
210
NATURE MATERIALS DOI: 10.1038/NMAT2379
a P3HT.+ 1∗P3HT
0.0016
0.0008
P3HT.+
ΔOD (a.u.)
0
P3HT.+
¬0.0008 Lu3N@C80.¬
900
¬0.0016
400 600 800 1,000 1,200
Wavelength (nm)
b 0.0008
0.0004
900
516 nm
0 1,000 nm
ΔOD (a.u.)
¬0.0004
¬0.0008
¬0.0012
¬0.0016
0 500 1,000 1,500 2,000 2,500 3,000
Time (ps)
Figure 4 | Time-resolved TNEF photophysics. a, Differential absorption
spectra (visible and near-infrared) obtained on femtosecond flash
photolysis (387 nm) of a P3HT/Lu3 N@C80 -PCBH film in air with several
time delays between 0 and 50 ps at room temperature—arrows indicate
the main characteristics of the P3HT singlet excited state (asterisk) and the
radical ion pair state (dot). b, Time-absorption profiles of the spectra
shown in a at 516 nm (filled circles) and 1,000 nm (open circles),
monitoring the radical ion pair state.
Open circuit voltages as high as 890 mV (Jsc = 5.4 mA cm−2 and
FF = 0.52) have been observed for P3HT/Lu3 N@C80 -PCBH devices,
which is to our knowledge, the highest Voc in P3HT/fullerene OPV
devices reported so far. This observed 260 mV Voc enhancement
over the P3HT/C60 -PCBM reference device is close to the
280 mV increase predicted by electrochemistry. Fluctuations in the
P3HT/Lu3 N@C80 -PCBH devices’ Voc from 890 to 810 mV have
been attributed to processing-induced surface states at the active
layer/cathode interface. Evidence of this surface-state interaction is
witnessed in Lu3 N@C80 -PCBH-based films with variations in film
dry times and choice of cathode materials. Films dried in less than
two minutes and capped with a LiF/Al cathode exhibit Voc values
that approach the maximum predicted Voc , as is demonstrated
by the 890 mV device. In contrast, films prepared under similar
conditions with a 20 nm calcium/80 nm Al cathode exhibit this
increase in Voc with longer dry times (data not shown). In an
effort to optimize our PCE and to uncover all of the variables
involved in achieving the paramount bulk heterojunction with
Lu3 N@C80 -PCBH, we are further investigating these fluctuations
in the P3HT/Lu3 N@C80 -PCBH device performance that are
encountered under different processing techniques. Our results of
this work will be addressed in a future publication.
In summary, here we report the first synthesis of Lu3 N@C80
methano derivatives and their use as novel acceptor materials in
state-of-the-art OPV devices that have an 890 mV open circuit volt-
age, the highest reported open circuit voltage for a P3HT/fullerene
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2379 LETTERS
a 10 b 70

60
5
50

J (mA cm¬2) 40

EQE (%)
0
30

20
¬5
10

¬10 0
¬0.5 0 0.5 1.0 400 500 600 700
V (V) Wavelength (nm)

c d 7

0.5 6

5
0.4

Intensity (a.u.)
4
Absorbance

0.3
3

0.2 2

1
0.1
0
0
300 400 500 600 700 800 900 4.0 4.5 5.0 5.5 6.0 6.5 7.0
Wavelength (nm) 2θ (°)

Figure 5 | Blend-film electronic and optical properties. a, J–V curves of P3HT/Lu3 N@C80 -PCBH (triangles) PCE = 4.2%, Voc = 810 mV,
Jsc = 8.64 mA cm−2 and FF = 0.61 and P3HT/C60 -PCBM (squares and dashed lines) PCE = 3.4%, Voc = 630 mV, Jsc = 8.9 mA cm−2 and FF = 0.61 blend
devices. Filled symbols show the dark curves and open symbols show devices under simulated Air Mass 1.5 (100 mW cm−2 ). b, EQE P3HT/Lu3 N@C80 -
PCBH (dashed) and P3HT/C60 -PCBM (solid) blend devices. c, Absorption of P3HT/Lu3 N@C80 -PCBH (solid) and P3HT/C60 -PCBM (dashed) blend films.
d, X-ray diffraction of the P3HT (100) peak normalized to thickness for a 1:1 blend film of P3HT/Lu3 N@C80 -PCBH (solid) and a 1:0.8 reference film of
P3HT/C60 -PCBM (dashed) blend.

device. Photophysical experiments confirm the efficient charge
transfer between P3HT and the Lu3 N@C80 -PCBH. Optimization of
the P3HT/Lu3 N@C80 -PCBH active-layer morphology has resulted
in OPV devices with higher Voc values and a similar photocurrent
and fill factor compared to P3HT/C60 -PCBM reference devices.
The higher overall PCE of the P3HT/Lu3 N@C80 -PCBH devices
is attributed to a better positioned LUMO level that captures
more of the energy associated with each absorbed photon. This
work demonstrates that reducing the donor/acceptor LUMO offset
by using TNEF acceptor materials can lead to enhanced OPV
performance through the Voc . Using the improved LUMO level
offset of our Lu3 N@C80 -PCBH acceptor materials and combining
it with previously reported low-bandgap donor polymers21,22 , OPV
efficiencies greater than 10% may now be feasible6,7 . With this
we have introduced an entirely new class of acceptor materials
for organic solar cells that show a straightforward path towards
commercially viable devices.
Received 11 July 2008; accepted 6 January 2009;
published online 8 February 2009
References
1. Stevenson, S. et al. Small-bandgap endohedral metallofullerenes in high yield
and purity. Nature 401, 55–57 (1999).
2. MacFarland, D. K. et al. Hydrochalarones: A novel endohedral metallofullerene
platform for enhancing magnetic resonance imaging contrast. J. Med. Chem.
51, 3681–3863 (2008).
3. Jones, M. A. G., Taylor, R. A., Ardavan, A., Porfyrakis, K. & Briggs, G. A. D.
Direct optical excitation of a fullerene-incarcerated metal ion. Chem. Phys. Lett.
428, 303–306 (2006).
4. Campanera, J. M., Bo, C., Olmstead, M. M., Balch, A. L. & Poblet, J. M.
Bonding within the endohedral fullerenes Sc3 N@C78 and Sc3 N@C80 as
determined by density functional calculations and reexamination of the crystal
structure of {Sc3 N@C78 }·Co(OEP)·1.5(C6 H6 )·0.3(CHCl3 ). J. Phys. Chem. A
106, 12356–12364 (2002).
5. Cardona, C. M., Elliott, B. & Echegoyen, L. Unexpected chemical and
electrochemical properties of M3 N@C80 (M = Sc, Y, Er). J. Am. Chem. Soc. 128,
6480–6485 (2006).
6. Koster, L. J. A., Mihailetchi, V. D. & Blom, P. W. M. Ultimate efficiency
of polymer/fullerene bulk heterojunction solar cells. Appl. Phys. Lett. 88,
0935111 (2006).
7. Scharber, M. C. et al. Design rules for donors in bulk-heterojunction
solar cells—Towards 10% energy-conversion efficiency. Adv. Mater. 18,
789–794 (2006).
8. Brabec, C. J. et al. Origin of the open circuit voltage of plastic solar cells.
Adv. Funct. Mater. 11, 374–380 (2001).
9. Kooistra, F. B. et al. Increasing the open circuit voltage of bulk-heterojunction
solar cells by raising the LUMO level of the acceptor. Org. Lett. 9,
551–554 (2007).
10. Brabec, C. J. et al. The influence of materials work function on the open circuit
voltage of plastic solar cells. Thin Solid Films 403/404, 368–372 (2002).
11. Lenes, M. et al. Fullerene bisadducts for enhanced open-circuit voltages and
efficiencies in polymer solar cells. Adv. Mater. 20, 2116–2119 (2008).
12. Campanera, J. M., Bo, C. & Poblet, J. M. Exohedral reactivity of trimetallic
nitride template (TNT) endohedral metallofullerenes. J. Org. Chem. 71,
46–54 (2006).
13. Cardona, C. M., Kitaygorodskiy, A. & Echegoyen, L. Trimetallic nitride
endohedral metallofullerenes: Reactivity dictated by the encapsulated metal
cluster. J. Am. Chem. Soc. 127, 10448–10453 (2005).
14. Rodriguez-Fortea, A., Campanera, J. M., Cardona, C. M., Echegoyen, L.
& Poblet, J. M. Dancing on a fullerene surface: Isomerization of
Y3 N@(N-ethylpyrrolidino-C80 ) from the 6,6 to the 5,6 regioisomer.
Angew. Chem. Int. Ed. 45, 8176–8180 (2006).

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 211

LETTERS
15. Hummelen, J. C. et al. Preparation and characterization of fulleroid and
methanofullerene derivatives. J. Org. Chem. 60, 532–538 (1995).
16. Lukoyanova, O. et al. ‘Open rather than closed’ malonate methano-fullerene
derivatives. The formation of methanofulleroid adducts of Y3 N@C80 . J. Am.
Chem. Soc. 129, 10423–10430 (2007).
17. Li, G. et al. ‘Solvent annealing’ effect in polymer solar cells based on
poly(3-hexylthiophene) and methanofullerenes. Adv. Funct. Mater. 17,
1636–1644 (2007).
18. Erb, T. et al. Correlation between structural and optical properties of
composite polymer/fullerene films for organic solar cells. Adv. Funct. Mater.
15, 1193–1196 (2005).
19. Peet, J. et al. Method for increasing the photoconductive response
in conjugated polymer/fullerene composites. Appl. Phys. Lett. 89,
252105 (2006).
20. Zhokhavets, U., Erb, T., Gobsch, G., Al-Ibrahim, M. & Ambacher, O. Relation
between absorption and crystallinity of poly(3-hexylthiophene)/fullerene films
for plastic solar cells. Chem. Phys. Lett. 418, 347–350 (2006).
21. Muhlbacher, D. et al. High photovoltaic performance of a low-bandgap
polymer. Adv. Mater. 18, 2931–2931 (2006).
212
NATURE MATERIALS DOI: 10.1038/NMAT2379
22. Peet, J. et al. Efficiency enhancement in low-bandgap polymer solar cells by
processing with alkane dithiols. Nature Mater. 6, 497–500 (2007).
Acknowledgements
This material is based on work supported by the Air Force Office of Scientific Research
under Contract No. FA9550-06-C-0010. Any opinion, findings and conclusions or
recommendations expressed in this material are those of the authors and do not
necessarily reflect the views of the Air Force Office of Scientific Research. This material is
also based on work supported by the National Science Foundation under Grant
No. 0348955 and Grant No. IIP-0740454. Additional thanks to The Cluster of Excellence
‘Engineering of Advanced Materials’ and the Alexander von Humboldt Foundation for
generous support (S.G.S.).
Additional information
The authors declare competing financial interests: details accompany the full-text HTML
version of the paper at www.nature.com/naturematerials. Supplementary Information
accompanies this paper on www.nature.com/naturematerials. Reprints and permissions
information is available online at http://npg.nature.com/reprintsandpermissions.
Correspondence and requests for materials should be addressed to M.D.
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
 LETTERS
PUBLISHED ONLINE: 8 FEBRUARY 2009 DOI: 10.1038/NMAT2384

Subnanometre platinum clusters as highly

active and selective catalysts for the oxidative
dehydrogenation of propane
Stefan Vajda1,2,3 *, Michael J. Pellin4 , Jeffrey P. Greeley2 , Christopher L. Marshall1 ,
Larry A. Curtiss1,2,4 *, Gregory A. Ballentine1† , Jeffrey W. Elam5 , Stephanie Catillon-Mucherie1 ,
Paul C. Redfern1 , Faisal Mehmood4 and Peter Zapol1,2,4

Small clusters are known to possess reactivity not observed a b c

in their bulk analogues, which can make them attractive
for catalysis1–6 . Their distinct catalytic properties are often
hypothesized to result from the large fraction of under-
coordinated surface atoms7–9 . Here, we show that size-
preselected Pt8−10 clusters stabilized on high-surface-area
supports are 40–100 times more active for the oxidative
dehydrogenation of propane than previously studied platinum
and vanadia catalysts, while at the same time maintaining
high selectivity towards formation of propylene over by-
products. Quantum chemical calculations indicate that under-
coordination of the Pt atoms in the clusters is responsible for
the surprisingly high reactivity compared with extended sur-
faces. We anticipate that these results will form the basis for
development of a new class of catalysts by providing a route to
bond-specific chemistry, ranging from energy-efficient and en-
vironmentally friendly synthesis strategies to the replacement
of petrochemical feedstocks by abundant small alkanes10,11 .
The oxidative dehydrogenation (ODH) of alkanes is a reaction
that is exothermic overall and is, thus, an attractive alternative
to dehydrogenation of alkanes, which is an endothermic process
requiring significant energy input. However, current ODH catalysts
have limited activity and/or poor selectivity resulting from an
inability to prevent complete oxidation12 . The reaction scheme
for propane ODH is shown in formula (1) and includes other
competing pathways that lead to COx species.
A B
C3H 8 C 3H 6 C 3H 4
D
(1)
C E
COx
In formula (1), channel A corresponds to the pathway for
propylene production and channels B, C, D and E result in
less desirable products.
Al2O3 Pt Al2O3 Pt SnO
Figure 1 | Depiction of the catalytic system. a,b, Illustration of Pt clusters
deposited in AAO membrane with ALD coating of Al2 O3 (a), and with
added SnO (b). c, Structure of Pt8 cluster on an Al2 O3 surface from density
functional calculations. In this structure there are four Pt–O distances
between 2.2 and 2.4 Å and the closest Pt–Al distance is 2.55 Å. See
Supplementary Information for more details.
Methods for producing size-selected clusters and soft-landing
them on catalytic supports are now well established13–16 . However,
testing these catalytic clusters under realistic reaction conditions
requires supports on which the clusters resist sintering. Both
alumina and tin oxide surfaces are known to stabilize platinum
clusters15,17 . We used atomic layer deposition (ALD) to coat porous
anodized aluminium oxide (AAO, Anopore) membranes with
alumina (Al2 O3 /AAO) before Pt-cluster deposition as illustrated
in Fig. 1. The ALD process ensures a uniform surface chemistry18
for the attachment of the clusters. Membranes were used in this
investigation because they provide a high surface area so that high
dispersions of size-selected clusters can be achieved, and catalytic
tests under well-defined conditions can be carried out. The Pt8−10
clusters were mass-selected, and a total of 900 (±135) ng Pt was
soft-landed on the large pore side (∼200 nm diameter) of the
membrane. A pair of membranes with identical Pt8−10 and alumina
loadings were synthesized; one was left as-is and one was treated
with an equivalent of two-monolayer-thick tin oxide ALD, resulting
in an overcoat of the alumina support around the platinum clusters
with tin oxide (SnO/Al2 O3 /AAO, Fig. 1).
Synchrotron grazing-incidence small-angle X-ray scattering
studies of alumina-supported size-selected Pt clusters carried out
in our laboratory15 have provided evidence for the stability and

1 Chemical Sciences and Engineering Division, Argonne National Laboratory, 9700 South Cass Avenue, Argonne, Illinois 60439, USA, 2 Center for
Nanoscale Materials, Argonne National Laboratory, 9700 South Cass Avenue, Argonne, Illinois 60439, USA, 3 Department of Chemical Engineering,
School of Engineering & Applied Science, Yale University, 9 Hillhouse Avenue, New Haven, Connecticut 06520, USA, 4 Materials Science Division,
Argonne National Laboratory, 9700 South Cass Avenue, Argonne, Illinois 60439, USA, 5 Energy Systems Division, Argonne National Laboratory,
9700 South Cass Avenue, Argonne, Illinois 60439, USA. † Present Address: Max-Planck-Institut für Metallforschung, Stuttgart, Germany.
*e-mail: vajda@anl.gov; curtiss@anl.gov.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 213

LETTERS NATURE MATERIALS DOI: 10.1038/NMAT2384

Selectivity Activity
a CO2 d
16.3% Pt8¬10 /SnO/Al2O3 /AAO, 400 °C
CO Pt8¬10 /SnO/Al2O3 /AAO, 500 °C
Propylene
83.7% 0% Pt8¬10 /Al2O3 /AAO, 550 °C

b CO2
15.7%
VOx /Al2O3, 1.4 V nm¬1, 390 °C

Catalyst
Propylene CO VOx /Al2O3, 8.0 V nm¬1, 390 °C
64% 20.3% VOx /Al2O3, 16.6 V nm¬1, 390 °C
VOx /Al2O3, 34.2 V nm¬1, 390 °C

c CO2
32% Pt /monolith, 400 °C

Propylene Pt /monolith, 500 °C

68% CO
0% 1E-3 0.01 0.1 1 10
TOF (s¬1)

Figure 2 | Catalyst activity and selectivity. a–c, Selectivity of the Pt8−10 -based catalysts at various temperatures and support compositions: SnO/Al2 O3
at 400 ◦ C (a), SnO/Al2 O3 at 500 ◦ C (b) and Al2 O3 at 550 ◦ C (c). d, TOFs of propylene produced on the Pt8−10 catalysts (green) and reference ODH
catalysts (grey) expressed as number of propylene molecules formed per metal atom. Pt monolith and vanadia data from refs 29 and 22, respectively. See
Supplementary Information for more details.

Pt4 + C3H8(g)

0.24 eV 0.42
0.54
Pt4---C3H8
0.37
0.95
Pt4-(H,C3H7)

Pt4-(2H,C3H6)

Figure 3 | Reaction path. Diagram of primary reaction steps in channel A (see formula (1)) from DFT calculations for the dehydrogenation of propane on a
Pt4 cluster leading to formation of propylene adsorbed on the cluster. Energies (in eV) of the equilibrium structures are relative to the reactants. Energy
barriers for the transition state structures are relative to the preceding equilibrium structure (‘true’ barriers). The first barrier corresponds to breaking of
the first C–H bond (on the CH2 group) and the second barrier corresponds to breaking of the second C–H bond (on a CH3 group). A number of other
reaction steps involving hydrogen migration are not included in this diagram. The dotted lines in the structures indicate partial bonds. See Supplementary
Information for more details of structures and energies and other reaction pathways.
shape of the Pt clusters. The supported clusters, similar to those below). In addition, we carried out calculations to investigate the
used in this study, showed no evidence of agglomeration over a stability of a positively charged Pt8 cluster on alumina and found
temperature range of 20–400 ◦ C, and they maintained a three- that it withdraws electrons from the surface and becomes negatively
dimensional structure. At the higher temperatures used in the charged as in the case of the supported neutral cluster (see below).
catalytic testing in this study, there was also no evidence of any In a sense, these highly uniform clusters can be considered an ideal
change in cluster size, as there was no change in selectivity or activity model of a single-site catalyst on technologically relevant supports,
after 30 h of testing. We have also carried out density functional in which all active sites closely resemble each other2,19,20 .
theory (DFT) calculations14 to investigate the structure of a Pt8 The catalyst tests were carried out under atmospheric pressure
cluster on a θ -alumina surface. The structure (Fig. 1) indicates that in a flow reactor at temperatures from 400 to 550 ◦ C by using
the cluster maintains its three-dimensional structure, as is found 10 s.c.c.m. total flow of reactants in argon carrier gas with 2.63 and
in the X-ray studies. The Pt cluster forms Pt–O bonds with the 2.73 mol% of oxygen and propane, respectively. The temperature
surface, resulting in significant charge transfer to the cluster. The range was chosen to give a direct comparison with the performance
binding of the cluster to the surface (∼3 eV) is consistent with of reported Pt-and VOx -based ODH catalysts. The measured
the stability of the subnanometre Pt8 clusters on alumina, but turnover frequencies (TOFs) are shown in Fig. 2, along with the
does not significantly affect the cluster’s chemical reactivity (see highest reported values for platinum and vanadia. The TOFs were

214 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2384
calculated as the number of propylene molecules produced per Pt
atom per second (see Supplementary Information for details of the
method of calculation). The activity and selectivity at 500 ◦ C was
unchanged over 30 h of testing.
The main finding of this study is that the activity of the
Pt8−10 catalyst is markedly higher than any reported platinum
or vanadia-based ODH catalysts. In fact, at 400 ◦ C, the observed
TOFs are 40–100 times higher than those previously reported.
The product distribution (Fig. 2) indicates that selectivity for
propylene over the formation of carbon oxide species is also
achieved. The total conversion approaches 25% ± 4% at high
temperatures. The uncertainty in the reported TOF is around
15%, determined primarily by the uncertainty in the amount of
Pt loading. The uncertainty in conversion rates is given by the
precision at determining the diameter of the Pt spots on the
membrane using grazing-incidence small-angle X-ray scattering on
poreless flat-alumina supported Pt clusters prepared under identical
deposition conditions.
To understand the observed high activity of small Pt clusters
for propane ODH, we carried out DFT calculations on key steps
for channel A in the reaction diagram in formula (1). Channel
A is the predominant channel because up to 84% of the product
formed is propylene (Fig. 2). A tetrahedral Pt4 cluster was used
as a model for the Pt8−10 clusters because the three-coordinated
Pt in this cluster is representative of the under-coordination of Pt
found in the larger Pt8−10 clusters. The use of a free cluster model
(without a substrate) is justified by the fact that the observed activity
results for alumina and alumina/tin oxide substrates are similar
(Fig. 2) and also by some model calculations done to investigate the
substrate effect (see below).
The calculated transition states and intermediates in reaction
channel A (formula (1)), leading to formation of propylene on
the Pt4 cluster, are shown in Fig. 3. The ‘true’ barrier to breaking
of the first C–H bond is only 0.42 eV. The corresponding barrier
referenced to gas-phase propane (the ‘apparent’ barrier) is slightly
smaller, at 0.18 eV; this barrier was found to be similarly small
(0.05 eV) when recalculated on a Pt8 cluster. We note that, in
spite of the small magnitude of the energetic barrier, this step will
probably be rate-limiting because of the very large entropic loss,
and consequentially lower pre-exponential factor, associated with
propane adsorption at high temperatures on the clusters. The rest
of the pathway is thermodynamically downhill to formation of
propylene, which binds to Pt4 by its π bond. Interestingly, our
calculations also indicate that dissociated oxygen atoms adsorbed
on the cluster21 do not significantly affect the calculated C–H bond
activities in this pathway. In the overall reaction scheme, oxygen
serves as a means for removal of hydrogen as water on the basis
of these calculations.
The experimental barrier for propylene formation from propane
on the subnanometre Pt clusters can be estimated to pro-
vide comparison with theory. The barrier determined from the
experimental data for the Pt8−10 /SnO/Al2 O3 /AAO catalyst at 400
and 500 ◦ C is 0.2 eV. This is consistent with the barrier found in
our DFT investigation for channel A. Similarly good agreement
is found when we include models for the Al2 O3 support in our
calculations. First, we considered a negatively (−1) charged Pt4
cluster, which represents the charge transfer that occurs to the
cluster when it is supported on alumina. The propane C–H reaction
apparent barrier is 0.21 eV, an insignificant change from the neutral
cluster. Second, we calculated the barrier for a Pt8 cluster on a
θ-alumina surface. The apparent barrier is 0.19 eV, also a small
change from the result for the gas-phase Pt8 cluster. The lack
of a significant support effect is consistent with a larger binding
energy per atom (∼3 eV per atom) calculated for Pt8 and Pt4
clusters compared with the energy of each of the three Pt–alumina
surface bonds (∼1 eV). Although more detailed studies are needed
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
LETTERS
H
H3CHCCH3
A
H
H2CCCH3
H-CH2CHCH2 B
H-CHCHCH3
H3C-CH2CH3 C
H2CHC-CH3
D
H2C=CHCH3
0 0.5 1.0 1.5 2.0 2.5 3.0
Barrier (eV)
Figure 4 | Energy barriers of bond breaking. ‘True’ barriers for breaking
C–C and C–H bonds in propane and propylene reactions on a Pt4 cluster.
Bonds that are broken are shown in red. Letters correspond to channels in
formula (1). All of the barriers plotted here correspond to energies relative
to the reactants (propane or propylene). In propane, the C–H bond
breaking on the centre carbon (A) is favoured over a terminal carbon. In
propylene C–H bond breaking all three sites (B) have similar barriers. See
Supplementary Information for more details of structures and energies.
to fully elucidate the effect of the substrate, these results are
strong evidence that the support has little effect on the reaction
barrier in this case. In contrast to the low C–H bond scission
barrier found on Pt4 , DFT calculations for C–H bond scission in
propane on a Pt(111) surface give an apparent activation barrier
1 eV larger than the corresponding barrier on Pt4 . In addition, the
C–H bond cleavage barriers on a supported V2 O5 dimer are close
to 1.2 eV (ref. 22) from experiment and about 2 eV from theory
for a V2 O5 surface23 .
Thus, both theory and experiment lead to the conclusion that
under-coordinated Pt sites in small Ptn clusters are much more
active than a Pt surface for propane ODH. This can be explained
by the attractive interaction between the under-coordinated Pt and
propane. The DFT calculations show that the initial adsorption
complex between propane and the Pt4 cluster (Fig. 3) results in sig-
nificant charge transfer from a propane C–H bonding orbital to the
cluster. This weakens the C–H bond as demonstrated by its length-
ening and a concomitant lowering of the C–H vibrational frequency
by ∼500 cm−1 . In contrast, propane is very weakly adsorbed on a
Pt(111) surface with essentially no C–H bond lengthening or charge
transfer owing to higher coordination of the surface Pt atoms.
The observed product distribution, favouring propylene
formation over COx (84% versus 16% at 400 ◦ C—Fig. 2), suggests
that C–C or C=C cleavage on the Pt clusters is less favourable
than is C–H cleavage. Calculated activation barriers for various
such reactions (Fig. 4) confirm this conclusion. For example, the
barrier for breaking a C–C bond in propane is greater than 1 eV.
In addition, propylene itself is relatively unreactive; the barriers
for C–C and C=C breaking range from 2.2–2.7 eV. In addition,
the barrier to break a C–H bond in the methyl group of adsorbed
propylene is quite large (1.27 eV). The relatively large magnitude
of this barrier is due to the considerable structural rearrangement
required for the methyl group in propylene to come in contact with
the Pt4 cluster. These results, taken together, indicate that propylene
will be easily formed, but other by-products, such as allene and COx ,
will not form easily.
The calculated selectivity trends can be understood from the
electronic structures of the C–C and C–H bond-breaking transition
states. The larger barriers observed for C–C versus C–H bond
breaking are probably due to the sp3 directionality of the orbitals on
C compared with the spherical nature of the orbital on hydrogen,
215
LETTERS
which results in poorer overlap between the adsorbate and the
reaction site orbitals in the transition state for breaking the C–C
bond compared with that for the C–H bond. This argument has
been put forward by Blomberg et al. for other surface reactions24 .
Computations indicate that CH4 (ref. 25) and H2 (ref. 26) have a
small barrier and no barrier, respectively, for dissociation on a Pt4
cluster, which supports the overlap explanation.
To our knowledge, the work reported here is the first
investigation of size-preselected Pt clusters under realistic high-
temperature catalytic conditions. It has revealed a very high activity
of subnanometre Pt-cluster-based catalysts for the ODH of propane
to propylene. Combined with quantum chemical studies, this work
has shown that the high activity is due to the under-coordination
of the Pt in the clusters and that the clusters favour the scission of
C–H bonds relative to C–C or C=C bonds. Some recent work in our
laboratory demonstrates that small gold clusters (Au6−10 ) are highly
active for propylene epoxidation, thus providing further evidence
for the unique catalytic properties of subnanometre clusters27 . In
the future, size-selected clusters stabilized on appropriate supports
with uniform surface chemistry hold great promise for design of
new catalytic materials. It will be a challenging task to scale up the
production of size-selected clusters by more conventional chemical
methods, but there are very encouraging efforts suggesting that this
will ultimately be possible2,3 .
Methods
Production of narrow cluster size distributions. The clusters emerging from the
continuous beam cluster source possess identical velocity. Owing to the variation
in their kinetic energy with size, in addition to the single mass selection on a
quadrupole mass filter, narrow distributions of clusters with one to four sizes can be
isolated using a quadrupole deflector operated in energy filter mode.
Determination of the fraction of clusters in the pores acting as active catalysts.
As only 10.5% of the facing area of the membrane was exposed to clusters, and to
the propane feed, a multiplication factor of 9.5 is applied to obtain the corrected
propane conversion rate on the Pt-cluster-coated fraction of the AAO membrane.
By calculating the relative pore opening area to the total surface area of the
membrane, 28% of the Pt8−10 particles, corresponding to 252 ng of Pt metal, enter
the pores. The rest forms metallic platinum on the face of the AAO membrane as
confirmed by X-ray photoemission spectroscopy. Metallic Pt is known to possess
orders of magnitude lower TOFs for ODH of propane28 . Thus, 252 ng of Pt was
used for the calculation of the TOFs.
Activation barrier calculation from the experimental data. Using Arrhenius law
and assuming that the conversion rate is roughly proportional to the fraction
of activated C–H bonds, an estimate was obtained of the activation barrier
from two available experimental values for Pt8−10 /SnO/Al2 O3 /AAO catalyst at
different temperatures.
Theoretical methods. The calculations were done using the B3LYP functional
with the LANL2DZ basis set for Pt and 6-31G∗ for Al, C, H and O. The Pt(111)
results were obtained with plane-wave calculations using the RPBE functional. See
Supplementary Information for more details.
Received 21 July 2008; accepted 12 January 2009;
published online 8 February 2009
References
1. Xu, Z. et al. Size-dependent catalytic activity of supported metal clusters.
Nature 372, 346–348 (1994).
2. Gates, B. C. Supported metal clusters: Synthesis, structure, and catalysis.
Chem. Rev. 95, 511–522 (1995).
3. Argo, A. M., Odzak, J. F., Lai, F. S. & Gates, B. C. Observation of ligand effects
during alkene hydrogenation catalysed by supported metal clusters. Nature
415, 623–623 (2002).
4. Fu, Q., Saltsburg, H. & Flytzani-Stephanopoulos, M. Active Nonmetallic
Au and Pt species on ceria-based water-gas shift catalysts. Science 301,
935–938 (2003).
5. Campbell, C. T. The active site in nanoparticle gold catalysis. Science 306,
234–235 (2004).
6. Chen, M. S. & Goodman, D. W. The structure of catalytically active gold on
titania. Science 306, 252–255 (2004).
216
NATURE MATERIALS DOI: 10.1038/NMAT2384
7. Lemire, C., Meyer, R., Shaikhutdinov, S. & Freund, H.-J. Do quantum size
effects control CO adsorption on gold nanoparticles? Angew. Chem. Int. Ed. 43,
118–121 (2004).
8. Wei, J. & Iglesia, E. Mechanism and site requirements for activation and
chemical conversion of methane on supported Pt clusters and turnover rate
comparisons among noble metals. J. Phys. Chem. B 108, 4094–4103 (2004).
9. Hvolbaek, B. et al. Catalytic activity of Au nanoparticles. Nano Today 2,
14–18 (2007).
10. Hutchings, G. J., Scurrell, M. S. & Woodhouse, J. R. Oxidative coupling of
methane using oxide catalysts. Chem. Soc. Rev. 18, 251–283 (1989).
11. Labinger, J. A. & Bercaw, J. E. Understanding and exploiting C–H bond
activation. Nature 417, 507–509 (2002).
12. Cavani, F., Ballarini, N. & Cericola, A. Oxidative dehydrogenation of ethane
and propane: How far from commercial implementation? Catal. Today 127,
113–131 (2007).
13. Benz, L. et al. Landing of size-selected Ag+ n clusters on single crystal TiO2
(110)-(1 × 1) surfaces at room temperature. J. Chem. Phys. 122, 081102 (2005).
14. Lee, S., Fan, C., Wu, T. & Anderson, S. L. CO oxidation on Aun /TiO2
catalysts produced by size-selected cluster deposition. J. Am. Chem. Soc. 126,
5682–5683 (2004).
15. Winans, R. E. et al. Reactivity of supported platinum nanoclusters studied by in
situ GISAXS: Clusters stability under hydrogen. Top. Catal. 39, 145–149 (2006).
16. Yoon, B. et al. Charging effects on bonding and catalyzed oxidation of CO on
Au8 clusters on MgO. Science 307, 403–407 (2005).
17. Sadykov, V. A. et al. Oxidative dehydrogenation of propane over monoliths at
short contact times. Catal. Today 61, 93–99 (2000).
18. Pellin, M. J. et al. Mesoporous catalytic membranes: Synthetic control of pore
size and wall composition. Catal. Lett. 102, 127–130 (2005).
19. Bell, A. T. The impact of nanoscience on heterogeneous catalysis. Science 299,
1688–1691 (2003).
20. Somorjai, G. A., Contreras, A. M., Montano, M. & Rioux, R. M. Cluster
chemistry and dynamics special feature: Clusters, surfaces, and catalysis.
Proc. Natl Acad. Sci. 103, 10577–10583 (2006).
21. Xu, Y., Shelton, W. A. & Schneider, W. F. Thermodynamic equilibrium
compositions, structures, and reaction energies of Ptx Oy (x = 1–3) clusters
predicted from first principles. J. Phys. Chem. B 110, 16591–16599 (2006).
22. Argyle, M. D., Chen, K., Bell, A. T. & Iglesia, E. Effect of catalyst structure
on oxidative dehydrogenation of ethane and propane on alumina-supported
vanadia. J. Catal. 208, 139–149 (2002).
23. Redfern, P. C. et al. Quantum chemical study of mechanisms for oxidative
dehydrogenation of propane on vanadium oxide. J. Phys. Chem. B 110,
8363–8371 (2006).
24. Blomberg, M. R. A., Siegbahn, P. E. M., Nagashima, U. & Wennerberg, J.
Theoretical study of the activation of alkane C–H and C–C bonds by different
transition metals. J. Am. Chem. Soc. 113, 424–433 (1991).
25. Xiao, L. & Wang, L. Methane activation on Pt and Pt4 : A density functional
theory study. J. Phys. Chem. B 111, 1657–1663 (2007).
26. Cruz, A., Bertin, V., Poulain, E., Benitez, J. I. & Castillo, S. Theoretical study of
the H2 reaction with a Pt4 (111) cluster. J. Chem. Phys. 120, 6222–6225 (2004).
27. Lee, S. et al. Selective propene epoxidation on immobilized Au6−10 clusters: The
effect of hydrogen and water on selectivity and activity. Angew. Chem. Int. Ed.
(in the press, 2008) <http://dx.doi.org/10.1002/anie.200804154>.
28. Yu, C., Xu, H., Ge, Q. & Li, W. Properties of the metallic phase of zinc-doped
platinum catalysts for propane dehydrogenation. J. Mol. Catal. A 266,
80–87 (2007).
29. Silberova, B., Fathi, M. & Holmen, A. Oxidative dehydrogenation of ethane
and propane at short contact time. Appl. Catal. A 276, 17–28 (2004).
Acknowledgements
The work at Argonne National Laboratory was supported by the US Department
of Energy, BES-Chemical Sciences, BES-Materials Sciences, and BES-Scientific User
Facilities under Contract DE-AC-02-06CH11357 with UChicago Argonne, LLC,
Operator of Argonne National Laboratory. S.V. gratefully acknowledges the support by
the Air Force Office of Scientific Research. We acknowledge grants of computer time at
the Laboratory Computing Resource Center (LCRC) at Argonne National Laboratory,
the National Energy Research Scientific Computing Center (NERSC) at Lawrence
Berkeley National Laboratory and the Molecular Science Computing Facility (MSCF) at
Pacific Northwest National Laboratory. The authors are indebted to E. Iglesia and P. Stair
for valuable discussions, A. Holmen for providing the exact dimensions of the monolith
used in their studies of Pt-based catalysts and thank J. Moore for carrying out X-ray
photoemission spectroscopy analysis of the Pt/AAO sample.
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to S.V. or L.A.C.
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

Mesoporous germanium-rich chalcogenido
frameworks with highly polarizable surfaces
and relevance to gas separation
Gerasimos S. Armatas1 * and Mercouri G. Kanatzidis1,2†
Mesoporous materials with tunable non-oxidic framework
compositions can exhibit new kinds of functionality including
internal surfaces with high polarizability. As the chemical
and physical characteristics of the framework components can
induce useful catalytic, absorption and optoelectronic features,
the mesoporous structure can promote fast mass diffusion ki-
netics and size-selective transport of guest molecules1 . So far,
synthetic efforts have resulted in mesoporous metal chalco-
genides on using structure-directing moulds of soft or hard
templates. These include ordered mesoporous II–VI semicon-
ductors (such as CdS (refs 2,3), ZnS (ref. 4) and CdTe (ref. 5)).
Recently, template-free synthetic routes for high-surface-area
chalcogenide aerogels have been reported6,7 . Here, we describe
a novel kind of porous materials based on germanium-rich
chalcogenide networks and ‘soft’ highly polarizable surfaces.
We demonstrate that these materials can exhibit excellent
selectivity for separating hydrogen from carbon dioxide and
methane. These highly polarizable mesoporous structures have
important implications for membrane-based gas separation
process technologies including hydrogen purification.
Amphiphilic surfactants with polar head and nonpolar tail
groups can template the metal-ion-mediated crosslinking poly-
merization of Zintl clusters (Ge4 S4− 10 ) into mesostructured metal
chalcogenides8–11 . These materials possess nanoscale pore peri-
odicity of hexagonal or cubic morphology. However, the pore
structure of these solids is not accessible because of the inability to
remove the surfactant.
We have shown that Ge4− and (Ge9 )4− Zintl ions can result in
ordered mesostructured germanium and binary intermetallic alloy
semiconductors12–15 . Unlike the linking polymerization reactions
of the (Ge9 )4− ions with metal ions, the new reaction described
here involves polymerizing the (Ge9 )-clusters with chalcogens
(S, Se and Te) in what seems to be a redox process. The
resulting materials, NU–GeQ-N (NU: Northwestern University),
possess ‘Ge9 Qx ’ (Q = S, Se, Te) Ge-rich framework compositions
with periodically ordered pores in hexagonal as well as cubic
symmetry. This pore symmetry is preserved after template removal
to give mesoporous germanium-linked chalcogenide materials with
accessible pore channels and high surface areas. Because of the ‘soft’
atoms (S, Se and Te), the polarizability of the internal surface of
these materials is anticipated to be higher than those of metal oxides,
porous carbons and organic polymers and therefore provides an
entirely new medium through which to study the diffusion and
separation of gasses.
1 Department of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, Illinois 60208-3113, USA, 2 Materials Science Division, Argonne
National Laboratory, Chicago, Illinois 60439, USA. *Present address: Department of Materials Science and Technology, University of Crete, 71003
Heraklion, Greece. † e-mail: m-kanatzidis@northwestern.edu.
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
LETTERS
PUBLISHED ONLINE: 15 FEBRUARY 2009 DOI: 10.1038/NMAT2381
To produce the germanium-based chalcogenides, we used
surfactant-templated coupling of (Ge9 )4− ions with the elemental
chalcogens S, Se and Te. The self-assembly of amphiphilic surfac-
tant N -eicosane-N ,N -dimethyl-N -(2-hydroxyethyl) ammonium
bromide (EDMHEAB) in formamide/ethylenediamine mixture so-
lution provides a suitable template for the growth of the mesostruc-
tured frameworks. The strong reducing character of (Ge9 )4− favours
the reduction of S, Se or Te atoms, producing mesostructured
germanium-linked chalcogenide frameworks (equation (1)).
EDMHEAB + Ge4−
9 + 2N Q → (EDMHEA)x Ge9 Qn
1
−→ NU–GeQ-N (1)
NH4 NO3
The new family of Ge-rich chalcogenides NU–GeQ-1 and
NU–GeQ-2 (Q = S, Se and Te) was obtained after removal of
the surfactant from the pores, through exchange with ammonium
nitrate (NH4 NO3 ) following mild heating at 80 ◦ C.
Energy-dispersive X-ray spectroscopy (EDS) of the samples
indicates an atomic ratio of Q:Ge9 ∼ 2 for NU–GeQ-1 and close to 4
for NU–GeQ-2 materials. These ratios are very close to the nominal
composition. The EDS spectra did not show any signal from other
elements such as potassium or halides, suggesting the complete
polymerization reaction of (Ge9 )4− clusters. Thermogravimetric
analysis measurements indicated a ∼14–25 wt% of the surfactant
content in as-prepared samples, whereas after the template removal
only a ∼6–7 wt% of the surfactant remaining in the pores (see
Supplementary Figs S1–S3). The inorganic residue at 600 ◦ C was
found to be crystalline Ge and GeTe (in the case of NU–GeTe-1,
Supplementary Fig. S4). The analytical data and the estimated
chemical formulae of as-prepared and mesoporous samples are
given in Table 1.
The X-ray diffraction (XRD) patterns of the as-prepared
NU–GeQ-N materials show an intense broad diffraction peak
at low 2θ scattering angles of 1.5◦ –3◦ and a broader feature at
∼3.5◦ –5.5◦ , which according to transmission electron microscopy
(TEM) could be attributed respectively to the (100) diffraction
peak and the overlapped pair of (110) and (200) peaks of two-
dimensional hexagonal p6mm symmetry (Fig. 1a). Interestingly,
the XRD pattern of as-prepared NU–GeSe-2 and NU–GeTe-2
shows a main diffraction peak associated with a weak shoulder
(2θ ∼ 2.3◦ ) and a broad feature, respectively. These peaks according
to TEM could be attributed respectively to the distinct and
the overlapped pair of (211) and (220) diffraction peaks of
217
LETTERS NATURE MATERIALS DOI: 10.1038/NMAT2381

Table 1 | Analytical data of as-prepared and mesoporous NU–GeQ-1 and NU–GeQ-2 materials.
†
Sample Atomic ratio* Surfactant Chemical formula Colour Energy gap
(Q:Ge9 ) (wt%) (eV)
NU–GeS-1 1.9 7.2 (25.2) (EDMHEA)0.15 Ge9 S1.9 Red–orange 1.87
NU–GeS-2 4.0 6.9 (23.7) (EDMHEA)0.16 Ge9 S4.0 Brown 1.65
NU–GeSe-1 2.2 6.3 (14.1) (EDMHEA)0.15 Ge9 Se2.2 Red–brown 1.63
NU–GeSe-2 4.1 5.1 (22.6) (EDMHEA)0.14 Ge9 Se4.1 Brown 1.52
NU–GeTe-1 1.8 6.6 (16.7) (EDMHEA)0.16 Ge9 Te1.8 Light black 1.48
NU–GeTe-2 3.9 6.3 (16.5) (EDMHEA)0.21 Ge9 Te3.9 Black 1.20

*EDS elemental data normalized to the nine-atom Ge cluster.

†
Per cent of surfactant content determined from thermogravimetric analysis data. In parenthesis: the per cent of surfactant content of as-prepared materials.

a b patterns can be assigned to the [110] and [100] direction of the

Absorbance log (α /s) (a.u.)

body-centre-cubic Ia3̄d symmetry (Fig. 2d,f). On the basis of pore

(220) symmetry and XRD data, the lattice parameter (ao ) of as-prepared
and porous samples was estimated from the d-spacing of the main
diffraction peak (see Table 2).
vi
The permanent porosity of template-removed NU–GeQ-N
was probed with nitrogen physisorption measurements. Figure 3
v
Intensity (a.u.)

(220) shows the N2 adsorption–desorption isotherms, which feature

1 2
iv
c condensation in narrow size distribution mesopores16 . The small
iii
ii
i
2 4 6 8 10
2 θ ( °)
Figure 1 | XRD and optical absorption spectra of mesostructured and
mesoporous NU–GeQ-N. a, Low-angle XRD patterns of as-prepared
samples containing surfactant mesostructured (red lines) and
template-free mesoporous (blue lines): (i) NU–GeS-1, (ii) NU–GeS-2,
(iii) NU–GeSe-1, (iv) NU–GeSe-2, (v) NU–GeTe-1 and (vi) NU–GeTe-2
materials. b, Optical absorption spectra of mesoporous NU–GeQ-1 (solid
lines) and NU–GeQ-2 (dashed lines) semiconductors: Q = S (black lines),
Se (red lines) and Te (blue lines). c, Schematic model of hexagonal
mesoporous NU–GeQ-1 material (black: Ge and yellow: chalcogen atom).
cubic Ia3̄d symmetry. The broadness in the XRD peaks of these
materials probably arises from the presence of disordered porous
domains. The XRD patterns of template-removed materials show
the main diffraction peak slightly shifted to higher scattering
angles, implying a small contraction of the lattice periodicity (see
Fig. 1a). The mesoporous frameworks do not possess atomic-scale
periodicity judging from the absence of Bragg diffraction peaks at
high diffraction angles.
The mesoporous germanium chalcogenides exhibit a
well-defined electronic structure as revealed by the sharp optical
absorption onsets in the energy range from ∼1.2 to ∼1.9 eV
(Fig. 1b). The energy bandgap strongly depends on the chemical
composition and continuously decreases from the sulphur to
selenium to the tellurium analogues. This variation of energy
gap is consistent with the integration of chalcogenide atoms
into the framework.
Figure 2 shows typical TEM images and the corresponding
fast-Fourier transform (FFT) patterns of NU–GeQ-1 and
NU–GeS-2, revealing a hexagonal arrangement of parallel pore
channels. The TEM images of NU–GeSe-2 and NU–GeTe-2
show regular arrays of mesopores, which after indexing the FFT
3 4 5 typical type-IV curves associated with a distinct step at a
Energy (eV)
relative pressure (P/P0 ) of ∼0.2 that is attributed to nitrogen
increase in the adsorption isotherms at high relative pressure
(P/P0 > 0.9) implies the presence of a small fraction of large
interparticle voids. The analysis of the adsorption isotherms with
the Brunauer–Emmett–Teller method gives surface areas in the
range 350–516 m2 g−1 , which are very large if we account for
the heavier inorganic frameworks. The porous structure of these
materials was characterized using non-local density function theory
analysis, which indicates a quite narrow size distribution of pores
with mesopore size at ∼2.9–3.1 nm. Therefore, the estimated
framework wall thickness of NU–GeQ-N materials from the
d-spacing and the pore size is ∼1.5–∼2.0 nm. The wall thickness
increases in going from sulphur to tellurium analogues, which is
consistent with the increasing atomic radius. All textural parameters
of the NU–GeQ-N family are listed in Table 2.
The chemical state of germanium and chalcogen atoms in the
framework was investigated with X-ray photoelectron spectroscopy
(XPS), which is a very sensitive surface technique. The spectra
were obtained from NU–GeQ-1 samples after surface sputtering
with a low-energy Ar+ ion beam. The XPS spectra show a
Ge 3d core-level signal in the range of 30.2–30.8 eV and S 2p,
Se 3d and Te 3d5/2 core-level signals at 162.5 eV, 54.4 eV and
571.6 eV, respectively (see Supplementary Fig. S5). These binding
energies imply the frameworks have partially oxidized Ge and
reduced chalcogenide atoms17–19 . Detailed analysis of the Ge 3d
XPS peak in all samples reveals a small deconvoluted peak
close to 29 eV, suggesting some zero-valence germanium atoms20 .
This is consistent with the Ge-rich nature of the frameworks
and implies that a small portion of Ge atoms engages in Ge–
Ge bonds. A quantitative analysis from the XPS data indicates
atomic ratios between the S, Se and Te atoms and nine-atom Ge
cluster close to 2, 2.3 and 1.9, respectively (see Supplementary
Information). For the NU–GeTe-2 sample, the atomic ratio
was close to 4. These results are in good agreement with
those of EDS analysis.
A most interesting feature of the porous chalcogenide networks
is the potentially high polarizable internal pore surface. The high
polarizability arises from the heavier atoms (Ge and chalcogens)
that comprise the framework. The specific gas interactions with
these surfaces is a key factor determining the behaviour of gas
mixtures through porous structures21 . The NU–GeQ-N materials

218 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2381 LETTERS
a b

20 nm 20 nm 20 nm 20 nm

c d 022 022
224

224

20 nm 20 nm 20 nm 20 nm

e f 022
022

224
224
50 nm 50 nm 20 nm 20 nm

Figure 2 | TEM images of mesoporous NU–GeQ-N. a–f, TEM images of hexagonal mesoporous NU–GeS-1 (a), NU–GeS-2 (b), NU–GeSe-1 (c) and
NU–GeTe-1 (e) viewed along the pore channel axis (left image in each part) and perpendicular to the pore channel axis (right image in each part) and TEM
images of cubic mesoporous NU–GeSe-2 (d) and NU–GeTe-2 (f) taken along the [110] direction (left image in each part) and along the [100] direction
(right image in each part). Insets in each panel show the corresponding FFT images.

Table 2 | Textural properties of mesoporous NU–GeQ-1 and NU–GeQ-2.

‡ ‡
Sample Surface area Pore volume* Pore size (Dp ) d-spacing Unit cell (ao ) WT
† †
(m2 g−1 ) (cm3 g−1 ) (nm) (nm (nm) ) (nm (nm) ) (nm)
NU–GeS-1 516 0.30 3.0 3.9 (4.3) 4.5 (5.0) 1.5
NU–GeS-2 433 0.30 3.0 4.0 (4.4) 4.7 (5.1) 1.7
NU–GeSe-1 486 0.32 3.0 4.3 (4.6) 4.9 (5.3) 1.9
§ §
NU–GeSe-2 397 0.23 2.9 4.0 (4.6) 9.8 (11.3) 1.7
NU–GeTe-1 442 0.26 3.0 4.3 (4.7) 5.0 (5.5) 2.0
§ §
NU–GeTe-2 350 0.27 3.1 4.5 (4.6) 10.9 (11.3) 2.0

*Cumulative pore volume at P/Po = 0.95.

†
In parenthesis: the d-spacing and unit cell parameter of as-prepared samples containing surfactant materials is given.
‡ √
For hexagonal p6mm pore symmetry, the unit cell parameter is given by a0 = 2/ 3d100 and the framework wall thickness is given by WT = a0 − Dp .
§ √
Assuming cubic Ia3̄d pore symmetry, we obtained the unit cell parameter by a0 = d211 6 and the framework wall thickness by WT = a0 /ξ0 −Dp /2, where ξ0 (= 3.092) is the reduced area of the unit cell.

provide a unique opportunity to study the gas adsorption
and gas separation processes in this system as a function of
surface polarizability (for example, in going from the S to
Se to Te analogue).
We examined the ability of NU–GeQ-1 materials to selectively
absorb carbon dioxide (CO2 ) and methane (CH4 ) from a gas
mixture with hydrogen (H2 ), which is relevant to hydrogen
purification processes. Generally, hydrogen is produced by steam
reforming of natural gas or methanol cracking22 and its separation
from gas mixtures containing CO2 , CO, N2 and CH4 is a
technologically important procedure in industrial feedstock and
next-generation fuel sources23 . The adsorption isotherms for CO2 ,
CH4 and H2 of the NU–GeQ-1 chalcogenides indicate a significantly
higher affinity for CO2 and CH4 compared with hydrogen at 0 ◦ C
(Fig. 4a and Supplementary Fig. S11). Analysis of the adsorption
data using the ideal adsorption solution theory24 reveals that
NU–GeS-1 and NU–GeSe-1 have CO2 /H2 separation factors of
∼15–16 in the low-pressure limit, whereas the NU–GeTe-1 has
a separation factor of ∼39 in the low-pressure limit (Fig. 4b).
Similarly, a CH4 /H2 separation factor of ∼4–7 was determined for
NU–GeS-1 and NU–GeSe-1 and a separation factor of ∼28 was
determined for NU–GeTe-1 (see Supplementary Fig. S12).
The ‘soft’ character of the chalcogenide atoms in the framework
is expected to favour strong dispersion/induced-dipole interactions

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 219

LETTERS NATURE MATERIALS DOI: 10.1038/NMAT2381

a 350 b 350 1.0 c 350

dV/dD (cm3 g¬1Ŭ1)

1.0 0.8

dV/dD (cm3 g¬1Ŭ1)

dV/dD (cm3 g¬1Ŭ1)

Adsorbed volume (cm3 g¬1) 0.8

Adsorbed volume (cm3 g¬1)

0.8

Adsorbed volume (cm3 g¬1)

300 300 300 0.6
0.6 0.6
0.4
250 0.4 250 0.4 250
0.2 0.2 0.2
200 0 200 0 200 0
0 25 50 100 150 0 25 50 100 150 0 25 50 100 150
150 Pore width (Å) 150 Pore width (Å) 150 Pore width (Å)

100 100 100

50 50 50
0 0 0
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
P/Po P/Po P/Po

Figure 3 | Nitrogen adsorption–desorption isotherms of mesoporous NU–GeQ-1 and NU–GeQ-2 materials. a, NU–GeS-N. b, NU–GeSe-N. c, NU–GeTe-N.
NU–GeQ-1: black lines; NU–GeQ-2: red lines. The filled and open symbols are the adsorption and desorption data, respectively. Inset: Non-local density
function theory pore size distribution.

a 1.4 b 100
NU¬GeS-1
Adsorbed amount (mmol g¬1)

1.2 - 1 NU¬GeSe-1
Te 80 NU¬GeTe-1
Ge -1

Selectivity CO2/H2
1.0 ¬ e NU¬GePbTe-1
NU eS
0.8 ¬G CO2 60
NU
0.6 1
eS- 40
¬G
NU
0.4
e-1
ePbT 20
0.2 NU¬G
H2
0 0
0 200 400 600 800 1,000 0 100 200 300 400 500 600
p (mm Hg) p (mm Hg)

c 30 d 1
NU¬GeS-1 NU¬GeTe-1 NU¬GePbTe-1
NU¬GeSe-1
25
NU¬GeTe-1
Relative intensity (a.u.)

0
NU¬GePbTe-1
20
qst (KJ mol¬1)

¬1
15 10
qst (KJ mol¬1)

8
10 6 ¬2 CH4
4 CO2
2
5 CO
0
0 0.05 0.10 0.15 0.20 ¬3 H2
Amount adsorbed (mmol g¬1)
0
0 0.2 0.4 0.6 0.8 1.0 0 2 4 6 8 10 0 2 4 6 8 10
Amount adsorbed (mmol g¬1) Time (min) Time (min)

Figure 4 | Adsorption isotherms, heat of adsorption and gas mixture separation curves of mesoporous NU–GeQ-N. a, Adsorption isotherms for CO2
(filled symbols) and H2 (open symbols) of mesoporous NU–GeS-1 (black), NU–GeSe-1 (red), NU–GeTe-1 (blue) and NU–GePbTe-1 (green symbols) at 0 ◦ C.
The corresponding red lines are fits to the data. b, The CO2 /H2 selectivities predicted by applying the ideal adsorption solution theory to the CO2 and H2
single-component isotherms at 0 ◦ C of NU–GeQ-1 and NU–GePbTe-1 samples. c, The loading dependence of isosteric heat of the CO2 adsorption that is
determined by fitting the isotherms at −10 and 0 ◦ C to appropriate virial equations (see Supplementary Information). Inset: The loading dependence of
isosteric heat of adsorption for H2 of the NU–GePbTe-1 sample. d, The gas mixture separation curves (H2 /CH4 /CO2 /CO 20:22:18:5 v/v in N2 ) from
NU–GeTe-1 and NU–GePbTe-1 solids at 0 ◦ C. The relative intensities for each gas were obtained using a mass spectrometer gas analyser that could detect
the ion peaks at m/z+ = 16 (CH4 ), 44 (CO2 ), 12 (CO) and 2 (H2 ).

with guest molecules. In this case, the separation process relies heat of adsorption for probe molecules with different polariz-
on the differences between the adsorption strengths of adsorbates ability. The isosteric heat of adsorption (qst ) for CO2 , CH4 and
and the surface and is more selective for the strongly adsorbing CH3 CH3 (α(CH3 CH3 ) = 4.23 cm−3 ) of NU–GeQ-1 was deter-
molecule. Indeed the strength of these interactions is mostly mined to be ∼17–20, ∼17–29 and ∼23–41 kJ mol−1 at the limit
determined by the polarizability (α) of the adsorbate/adsorbent of zero coverage, respectively (see Fig. 4c and Supplementary In-
system. The NU–GeTe-1, which contains the more polarizable formation). These values of qst indicate favourable adsorption
telluride compared with the selenide NU–GeSe-1 and sulphide interactions between these molecules and the chalcogenide sur-
NU–GeS-1, exhibits higher solubility separation of the more face, which depend strongly on the polarizability of adsorbed
polarizable CO2 and CH4 molecules over H2 (polarizability (α): molecules as well as the porous framework. As the polariz-
α(CO2 ) = 2.51 > α(CH4 ) = 2.45 > α(H2 ) = 0.79 cm−3 ). ability of probe molecules increases and the soft character of
To examine the role of dispersion/induced-dipole interactions chalcogenide composition increases, the heat of adsorption between
on the adsorption selectivity process, we also determined the the two also increases.

220 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2381
It is anticipated that the CO2 /H2 and CH4 /H2 selectivity can
be further improved by increasing the soft character of these
chalcogenide frameworks. To test this prediction, we prepared
germanium lead telluride (NU–GePbTe-1) using the isostructural
[Ge6 Pb3 ]4− instead of the [Ge9 ]4− cluster. The expectation was
that the more polarizable Pb would further enhance the selectivity.
The NU–GePbTe-1 possesses a hexagonal mesoporous framework
with a large internal surface area (264 m2 g−1 ) and uniform pores
(Dp ∼ 3.3 nm) (see Supplementary Information). Interestingly, this
material is characterized by a heat of adsorption for CO2 , CH4
and CH3 CH3 that is higher than those observed in NU–GeQ-1,
probably owing to the stronger interactions causing the more
polarizable Pb atoms (see Fig. 4c and Supplementary Information).
For comparison, we determined the isosteric heat of H2 adsorption
for the NU–GePbTe-1 sample, finding a qst ∼ 6 kJ mol−1 at
the limit of zero coverage. These results further support non-
preferential adsorption interactions between this high polarizable
mesoporous surface and hydrogen molecules (Fig. 4c, inset). The
NU–GePbTe-1 exhibited an excellent separation performance of
CO2 and CH4 over H2 with respective separation factors of
∼88 and ∼70 that greatly exceed those of organic polymeric
membranes (sCO2 /H2 ∼ 31; ref. 25) and are comparable to those of
high-performance ceramic membranes (sCO2 /H2 ∼ 60–110; refs 26,
27). Membranes that exhibit CO2 /H2 selectivity greater than
100 are rare and usually possess a complex structure of micro-
and mesoporous composite solids and combine diffusivity and
solubility selectivity27,28 .
Gas mixture separation experiments by passing H2 /CH4 /
CO2 /CO gases through mesoporous NU–GeTe-1 and
NU–GePbTe-1 further support solubility separation of carbon
dioxide and methane over hydrogen. Figure 4d shows the intensity
of each gas component versus time when the gaseous steam
passes through the sample-packed column at 0 ◦ C. These plots
imply very good CO2 /H2 and CH4 /H2 mixture separation for
the examined samples by rapidly absorbing CO2 and CH4 and
allowing H2 through. The separation behaviour of these porous
chalcogenides seems to be strongly related to the polarizability value
of probe molecules and following the range CO2 > CH4 > CO > H2
(α(CO) = 1.95 cm−3 ). Note that this behaviour is repeatable and the
adsorbents can be completely recovered after cleaning the surface
by nitrogen flowing at 30 ◦ C (see Supplementary Figs S22,S23).
Small-molecule diffusion through cavities (free-volume elements)
in porous materials is an inherently nanoscopic phenomenon. The
NU–GeQ-N materials we report here undergo this phenomenon in
a manner that is consistent with their radically new type of heavy
atom surfaces, namely ‘soft’ and highly polarizable by virtue of the
heavy elements that comprise them. This is opposite to the ‘hard’,
less polarizable, surfaces of most known systems such as carbon-,
silica- and metal-oxide-based materials. The new ‘soft’ surfaces have
important implications for membrane separation processes in gas
separation including hydrogen purification. Finally, we remark that
because the NU–GeQ-N are chalcogenide frameworks, they have
the potential to resist sulphur poisoning that can occur in other
separation materials29 .
Methods
Formamide and anhydrous ethanol were degassed under nitrogen. Ethylenediamine
was dried over CaH2 and then distilled from sodium/benzophenone under nitrogen
and stored under nitrogen.
In a typical preparation of NU–GeQ-N (N = 1, 2; Q = S, Se and Te), 193 mg
K4 Ge9 (ref. 15) (0.24 mmol) was dissolved in 3 ml anhydrous ethylenediamine
under stirring at room temperature. This solution was added drop-wise to a clear
solution of 0.9 g N -eicosane-N ,N -dimethyl-N -(2-hydroxyethyl) ammonium
bromide30 surfactant in 12 ml formamide at ∼75 ◦ C. After stirring for ∼5 min,
a clear solution of S or Se in 0.8 ml anhydrous ethylenediamine or Te in 1 ml
anhydrous hydrazine was added and the resulting mixture was aged overnight at
∼80 ◦ C under stirring. The product (EDMHEA)x (GeQ) was isolated by filtration,
washed with warm (80 ◦ C) formamide several times and dried under vacuum
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
LETTERS
at room temperature for 24 h. Mesoporous solids were prepared by reacting
230 mg of (EDMHEA)x (GeQ) with 130 mg of ammonium nitrate (NH4 NO3 )
in 20 ml of anhydrous ethanol at room temperature. The product was isolated
by filtration, washed with anhydrous ethanol, dried under vacuum at room
temperature for 24 h and heated at 80 ◦ C under vacuum for 12 h. NU–GeQ-1
and NU–GeQ-2 structures were obtained on using 15 mg of S, 38 mg of Se or
62 mg of Te (0.48 mmol, Q:Ge4− 9 molar ration of 2) and 30 mg of S, 76 mg of Se or
124 mg of Te (0.96 mmol, Q:Ge4− 9 molar ration of 4), respectively. The Ge6 Pb3 Te2
(NU–GePbTe-1) was prepared according to the above procedure using 291 mg of
K4 Ge6 Pb3 (0.24 mmol) and 62 mg of Te (0.48 mmol, Te:Ge6 Pb4− 3 molar ration of
2). The germanium chalcogenide structures are very delicate on the removal of
the surfactant and can easily collapse if this is not done properly. The optimum
condition requires repeating the NH+ 4 -exchange process three times for short
periods (∼5 min). All manipulations were carried out inside a glovebox under
nitrogen atmosphere or under vacuum.
Received 7 August 2008; accepted 13 January 2009;
published online 15 February 2009
References
1. Antonietti, M. & Ozin, G. A. Promises and problems of mesoscale materials
chemistry or why meso? Chem. Eur. J. 10, 28–41 (2004).
2. Braun, P. V., Osenar, P. & Stupp, S. I. Semiconducting superlattices templated
by molecular assemblies. Nature 380, 325–328 (1996).
3. Gao, F., Lu, Q. & Zhao, D. Synthesis of crystalline mesoporous CdS
semiconductor nanoarrays through a mesoporous SBA-15 silica template
technique. Adv. Mater. 15, 739–742 (2003).
4. Li, J., Kessler, H., Soulard, M., Khouchaf, L. & Tuiler, M.-H. Nanosized
zinc sulfide obtained in the presence of cationic surfactants. Adv. Mater. 10,
946–949 (1998).
5. Jiang, T. & Ozin, G. A. New directions in tin sulfide materials chemistry.
J. Mater. Chem. 8, 1099–1108 (1998).
6. Mohanan, J. L., Arachchige, I. U. & Brock, S. L. Porous semiconductor
chalcogenide aerogels. Science 307, 397–400 (2005).
7. Bag, S., Trikalitis, P. N., Chupas, P. J., Armatas, G. S. & Kanatzidis, M. G.
Porous semiconducting gels and aerogels from chalcogenide clusters. Science
317, 490–493 (2007).
8. MacLachlan, M. J., Coombs, N. & Ozin, G. A. Non-aqueous supramolecular
assembly of mesostructured metal germanium sulphides from (Ge4 S10 )4−
clusters. Nature 397, 681–684 (1999).
9. Trikalitis, P. N., Rangan, K. K., Bakas, T. & Kanatzidis, M. G. Varied pore
organization in mesostructured semiconductors based on the [SnSe4 ]4− anion.
Nature 410, 671–675 (2001).
10. Rangan, K. K. et al. Hexagonal pore organization in mesostructured metal tin
sulfides built with [Sn2 S6 ]4− cluster. Nano Lett. 2, 513–517 (2002).
11. Fröba, M. & Oberender, N. First synthesis of mesostructured thiogermanates.
Chem. Commun. 1729–1730 (1997).
12. Armatas, G. S. & Kanatzidis, M. G. Mesostructured germanium with cubic
pore symmetry. Nature 441, 1122–1125 (2006).
13. Armatas, G. S. & Kanatzidis, M. G. Hexagonal mesoporous germanium.
Science 313, 817–820 (2006).
14. Sun, D. et al. Hexagonal mesoporous germanium through surfactant-driven
self-assembly of Zintl clusters. Nature 441, 1126–1130 (2006).
15. Armatas, G. S. & Kanatzidis, M. G. Mesoporous compound semiconductors
from metal-linked deltahedral Ge9 clusters. J. Am. Chem. Soc. 130,
11430–11436 (2008).
16. Rouquerol, F., Rouquerol, J. & Sing, K. S. W. Adsorption by Powders and Porous
Solids. Principles Methodology and Applications (Academic, 1999).
17. Foix, D., Gonbeau, D., Granier, D., Pradel, A. & Ribes, M. Electronic
structure of thiogermanate and thioarseniate glasses: Experimental (XPS)
and theoretical (ab initio) characterizations. Solid State Ion. 154-155,
161–173 (2002).
18. Ueno, T. X-ray photoelectron and Anger electron spectroscopic studies
of chemical shifts in amorphous Ge–Se system. Japan. J. Appl. Phys. 22,
1349–1352 (1983).
19. Yashina, L. V., Kobeleva, S. P., Shatalova, T. B., Zlomanov, V. P. &
Shtanov, V. I. XPS study of fresh and oxidized GeTe and (Ge,Sn)Te surface.
Solid State Ion. 141/142, 513–522 (2001).
20. Oh, J. & Campbell, J. C. Thermal desorption of Ge native oxides and the loss
of Ge from the surface. J. Electron. Mater. 33, 364–367 (2004).
21. Yang, R. T. Adsorbents: Fundamentals and Applications Ch. 2 (Wiley, 2003).
22. Garcia, L., French, R., Czernik, S. & Chornet, E. Catalytic steam reforming
of bio-oils for the production of hydrogen: Effects of catalyst composition.
Appl. Catal. A 201, 225–239 (2000).
23. Schlapbach, L. & Züttel, A. Hydrogen-storage materials for mobile applications.
Nature 414, 353–358 (2001).
24. Myers, A. L. & Prausnitz, J. M. Thermodynamics of mixed-gas adsorption.
AIChEJ. 11, 121–127 (1965).
221
LETTERS
25. Lin, H., Wagner, E. V., Freeman, B. D., Toy, L. G. & Gupta, R. P.
Plasticization-enhanced hydrogen purification using polymeric membranes.
Science 311, 639–642 (2006).
26. Verweij, H., Lin, Y. S. & Dong, J. Microporous silica and zeolite membranes
for hydrogen purification. MRS Bull. 31, 756–764 (2006).
27. Ockwig, N. W. & Nenoff, T. M. Membranes for hydrogen separation.
Chem. Rev. 107, 4078–4110 (2007).
28. Verweij, H., Lin, Y. S. & Dong, J. Microporous silica and zeolite membranes
for hydrogen purification. MRS Bull. 31, 756–764 (2006).
29. Wang, K. & Stiefel, E. I. Toward separation and purification of olefins
using dithiolene complexes: An electrochemical approach. Science 291,
106–109 (2001).
30. Broxon, T. J. & Chung, R.P.-T. Micellar bound Meisenheimer complexes.
J. Org. Chem. 55, 3886–3890 (1990).
222
NATURE MATERIALS DOI: 10.1038/NMAT2381
Acknowledgements
These studies were supported primarily by the Nanoscale Science and Engineering
Initiative of the National Science Foundation under NSF Award Number EEC-0647560.
We thank Peter C. Stair for the use of a mass-spectrometer gas analyser. This work made
use of the J.B. Cohen X-ray Diffraction facility and the Electron Probe Instrumentation
Center (EPIC) and Keck Interdisciplinary Surface Science (Keck-II) facility of NUANCE
Center at Northwestern University.
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to M.G.K.
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
 ARTICLES
PUBLISHED ONLINE: 25 JANUARY 2009 DOI: 10.1038/NMAT2375

Shear-induced anisotropic plastic flow from

body-centred-cubic tantalum before melting
Christine J. Wu *, Per Söderlind, James N. Glosli and John E. Klepeis

There are many structural and optical similarities between a liquid and a plastic flow. Thus, it is non-trivial to distinguish
between them at high pressures and temperatures, and a detailed description of the transformation between these phenomena
is crucial to our understanding of the melting of metals at high pressures. Here we report a shear-induced, partially disordered
viscous plastic flow from body-centred-cubic tantalum under heating before it melts into a liquid. This thermally activated
structural transformation produces a unique, one-dimensional structure analogous to a liquid crystal with the rheological
characteristics of Bingham plastics. This mechanism is not specific to Ta and is expected to hold more generally for other
metals. Remarkably, this transition is fully consistent with the previously reported anomalously low-temperature melting curve
and thus offers a plausible resolution to a long-standing controversy about melting of metals under high pressures.

H
ow and at what temperature metals melt at high pressures
are fundamental questions that have wide scientific impli-
cations, including our understanding of the Earth’s interior.
Despite dramatic advancements in experimental and theoretical
methods, a remarkable controversy remains for the melting of
metals at high pressures1–21 . Often1–6 , but not always7–9 , the melting
curve of a metal obtained from laser-heated diamond-anvil cell
(DAC) experiments1–6 is much flatter than those obtained by shock-
wave experiments10 , piston–cylinder-apparatus measurements11
and calculations12–17 . For example, extrapolating the Ta DAC melt-
ing curve reported in ref. 1 to 300 GPa (crosses in Fig. 1a) yields
a melting temperature that is approximately 5,000 K lower than
the shock-melting point10 (solid circle). Substantial variations also
exist among calculations12–17 , but the majority of them disagree
qualitatively with the DAC melting slope. The underlying source
of this discrepancy is unlikely to be material specific, as it exists
for a majority of the transition metals that have been studied
experimentally1–6 . In addition, the flat DAC melting curve contrasts
sharply with the classical Lindemann melting law12 (red dashed
curve). In contrast, a DAC melting curve consistent with shock
melting was recently reported for body-centred-cubic (b.c.c.) lead
in ref. 9. In this study, X-ray signal scattered by the liquid was used
as the detection criterion. It suggested that the previous melting
criteria using onset-of-speckle detection and disappearance of solid
X-ray signal alone are not always reliable, even though the previous
DAC experiments reproduced the correct melting temperature at
ambient pressure1,4 . In comparison, the shock-melting criterion10
based on sound-speed measurements works reasonably well for
identifying a solid-to-liquid boundary. However, both theory and
DAC are often needed to provide more detailed identifications.
Several hypotheses18–21 have been proposed to explain the
melting discrepancy of metals. For instance, Belonoshko et al.
have advocated solid-to-solid transition as the explanation for
the observed DAC melting18,20 . In particular, they have recently
presented ab inito evidence of a b.c.c.-to-face-centred-cubic (f.c.c.)
transition in molybdenum (Mo) and suggested it as a candidate18 .
Their study shed important light on the stability of Mo at
elevated temperatures and pressures, which might be connected
to the observed shock transition at 210 GPa (ref. 19). Intriguingly,
founded on intuitive phenomenological modelling and shear
modulus analysis, Ross et al.21 have proposed a visco-elastic
phase (similar to a supercooled liquid22 ) at higher pressures (near
150 GPa) to bridge the low- and high-temperature melting curves,
while maintaining that the DAC melting is the true solid–liquid
boundary. Whether or not melting is true, it strongly signals a
transition from a b.c.c. crystal to a less ordered structure whose
identity remains a mystery.
In searching for a thermally activated disordered structure,
we investigated the role of shear in producing viscous plastic
flow before melting. This is particularly relevant because we
have evidence from ab inito energy landscapes that significant
thermal softening is present at temperatures near DAC melting.
Coincidentally, a hot viscous plastic flow could also have the
appearance of liquid-like motion. Two key distinctions between
viscous plastic flow and liquid are (1) the absence or presence
of long-range thermally diffusive atoms23 and (2) whether
or not it supports a shear at rest. It is notable that DAC
experiments do not involve measuring atomic mobility1–6 .
Interestingly, Ross et al.21 found that the shear modulus has a
discontinuous change to a finite value from b.c.c. to the proposed
visco-elastic phase.
Previously, non-hydrostatic stress and thermal pressure in DAC
had been studied primarily for insulators, which are poor thermal
conductors2,24–26 . In particular, dynamic recrystallization has been
proposed as an explanation for DAC low-temperature melting of
MgO (ref. 20). In addition, the possibility of plastic deformation was
also hypothesized24 . However, previous searches27–29 for mechanical
instability of b.c.c. Ta by experiments (at low temperatures) and
calculations (with b.c.c. symmetry) were not successful. It is also
known that shear plays an important role in disordering transition
of colloidal crystals30 , viscosity of non-Newtonian fluids and the
rheological behaviour of Bingham plastics31,32 . Later, we shall
show that this thermally activated transition results in a novel
structure, different from that of the usual plastically deformed b.c.c.
metals. Instead, it matches the rheological description of a Bingham
plastic31,32 (for example a toothpaste or a tar), which behaves like
a rigid solid under low shear stresses but flows like a viscous fluid
under high shear stresses31,32 .

Lawrence Livermore National Laboratory, Livermore, California 94551, USA. *e-mail: wu5@llnl.gov.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 223

ARTICLES
a 12,000
10,000 Lindemann (ref. 12)
8,000
Temperature (K)
DAC (ref. 1)
6,000
4,000
2,000
0
0 100
Pressure (GPa)
Figure 1 | Predicted MD boundary between b.c.c. and plastic flow, as well as previously reported melting curves of Ta. a, The green dotted curve shows
the boundary between b.c.c. (blue triangles) and plastic flow (green diamonds). b, The atoms shown in a simulation cell are those identified by a central
symmetry parameter as disordered after transforming into a plastic flow. c, Averaged atomic displacements h1R[111] i along [111] as a function of
temperature near 30 GPa.
Unfortunately, the exact magnitude of shear in the DAC is
unknown2,25,26 . However, its presence is indisputable, even with
pressure media, because compression by two diamonds is uniaxial.
So far, the role of shear on melting of metals has not been seriously
considered, perhaps for several reasons. First, DAC experiments
have demonstrated that pressure media have no significant effect
on melting temperatures, although they do influence X-ray patterns
near melting6 . Second, shear is expected to dissipate quickly in
metals near melting. Indeed, we find that the timescale of shear
relaxation falls in between those of melt and plastic flow. Thus, shear
is not important for determining the solid-to-liquid boundary, but
it is critical in inducing plastic flow, particularly when significant
thermal softening is present. Later, we shall show that the transition
temperature is approximately independent of external stresses after
it exceeds a threshold value. We suggest that a combination of
deviatoric and thermal stresses in the DAC is likely to exceed the
transformation threshold, even with the use of pressure media.
Investigation of structural transformations at high pressure
requires a very accurate interatomic potential, making Ta an ideal
prototype for this purpose. In particular, Ta is one of only a few
metals for which a highly accurate quantum-based, many-body
interatomic model-generalized pseudopotential theory (MGPT)
potential has been developed for applications across a wide range
of temperatures, pressures and strains33 . Specifically, the melting
curve (solid red curve in Fig. 1a), established in ref. 34 using
the MGPT potential33 , is in excellent agreement with the high-
temperature shock-melting10 point (black circle) as well as with the
ambient-pressure melting point. Using the same MGPT potential,
we conducted a series of molecular dynamics (MD) simulations
on heated b.c.c. Ta crystals with and without shear. Details of the
simulations are provided in the Methods section.
Here we report our finding of a thermally activated and
shear-induced structural transformation to an anisotropic
Bingham-like plastic flow (green dotted curve) that coincides with
the DAC melting curve (see Fig. 1a). In contrast, under hydrostatic
compression, the same MGPT potential33 gave the solid–liquid
boundary (solid red line in Fig. 1a) at much higher temperatures34 .
For each fixed volume, initial strain was applied along a b.c.c.
slip35 system of {1̄10} [111] . After heating the system to the
targeted temperature and equilibrating it for 40 ps, we analysed the
structure to determine whether it remained b.c.c. (blue triangles) or
transformed into a plastic flow (green diamonds). We found that
224
NATURE MATERIALS DOI: 10.1038/NMAT2375
b ε yz
Shock (ref. 10)
¬
y[110] ¬
x[112]
z[111]
MGPT (ref. 34)
c
0.30
〈ΔR[111]〉
0.25
Plastic flow (MGPT)
B.c.c. (MGPT)
0.20
200 300
2,500 3,000 3,500
T (K)
the transition temperature was independent of the external stress
once it was large enough to overcome the deformation barrier. We
shall show later that the low deformation barrier of Ta is linked to
the flat energy landscape along the Burgers path36 .
The technique we used to identify the disordering transition is
based on the widely used method of identifying amorphization37
by monitoring changes in the radial distribution function g (r).
The main characteristic feature of the disordered structure is the
appearance of the shoulder feature in g (r), shown in Fig. 2c.
The merging of peaks reflects that the third- and fourth-nearest-
neighbour distances in the {1̄10} planes become indistinguishable.
Note that this structure is different from an amorphous or a
supercooled liquid, which possesses only short-range order21,22 .
Instead, it has a partial disordered–partial crystalline structure
with a long-range order in the [111] direction. In fact, it shares
structural similarity with the smectic phase of a liquid crystal, or
one-dimensional polymer. However, the interchain interactions
of Ta are expected to be much stronger than those of liquid
crystals and polymers. We have chosen to call it a one-dimensional
plastic flow because, as shown in Fig. 2, it looks like a standard
high-temperature b.c.c. crystal (blue panels, b) in the {111} and
{112̄} planes, whereas it is similar to a glass (red panels, d) in
the {1̄10} planes. In addition, analysis of the angular distribution
function during melting shows that the [111] chains are the last
crystalline structure to survive before melting (see Supplementary
Information, Fig. S1). This is consistent with {1̄10} [111] being
the primary Ta slip system35 in deformation and [111] chains
having the strongest atomic bonding. Below, we shall show that
the origin of this transition is linked to the observed anisotropy
of the energy landscape, which leads to thermal softening of {1̄10}
planes under shear.
Shear provides the critical driving force of this transition, the
signature of a Bingham plastic31,32 . To make a Bingham plastic flow,
the driving shear stress has to exceed the yield stress. Without shear,
we found no structural transformation before melting. For instance,
Fig. 2b shows that unsheared b.c.c. Ta remains as b.c.c. at 3,250 K.
Under sufficient shear, Ta becomes partially disordered at the same
temperature (Fig. 2c). This result explains why the DAC produces
the correct melting temperature at ambient pressure, a stress-free
state. The threshold stress reflects the materials’ resistance to flow,
which is material specific and depends strongly on pressure. Rough
estimates of threshold strain of Ta fall between 10% and 13% for a
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2375 ARTICLES
a 298 K b.c.c. b 3,250 K b.c.c. c 1D plastic flow d Glass
[110]

{111} plane
–

–
[112]

–
[111]

{112} plane

–
[110]
[111]

–
{110} plane

–
[112]

3 3,250 K 3,250 K 5,000 K

g(r)

0
2 4 6 2 4 6 2 4 6 2 4 6
r (Å) r (Å) r (Å) r (Å)

Figure 2 | Structural properties compared between b.c.c., one-dimensional plastic flow and glass. Planar projections of Ta atomic coordinates are shown
in 8 Å × 8 Å regions, and the corresponding radial distribution functions g(r) near 30 GPa. a, Different symbols indicate atoms from different layers of an
unsheared b.c.c. crystal at 298 K. b (blue panels), A b.c.c. crystal heated to 3,250 K without shear. c (green panels), A one-dimensional (1D) plastic flow
induced by shear and heating. d (red panels), A supercooled isotropic glass. The arrow in c indicates the shoulder feature in g(r) of a plastic flow. The third
and fourth peaks of g(r) of a b.c.c. crystal correspond to the distances inside a {1̄10} plane, shown by arrows in a.

pressure range of 15–65 GPa. In addition, we found that the flowing
transition occurs substantially faster than shear dissipation. For
example, the blue curve in Fig. 3a shows that it takes ∼1.4 ps to
transform from b.c.c. Ta to the plastic flow at 3,250 K and 17 GPa.
In comparison, the stress of the simulation cell relaxed at a much
slower rate (green curve) before reaching a steady state of 1.75 GPa,
and was roughly unchanged up to 375 ps. The finite stress value
is consistent with viscous plastic flow rather than liquid. Before
exceeding the threshold value, the external stress influences the
transition temperature, as Fig. 3b shows. For instance, no transition
takes place before melting if the external stress is less than 5 GPa at
pressures near 17 GPa. However, once the external stress is larger
than the threshold stress (∼9 GPa at a pressure of 17 GPa), the
transition temperature no longer changes with stress.
Unfortunately, the actual range of the initial stress in a DAC
Ta sample is unknown. In fact, the previously reported values
of thermal stress in DAC are fairly scattered2,25,26 . For instance,
a significant thermal pressure (>10 GPa) has been reported for
MgO (ref. 25). For NaCl, if we extrapolated a reported temperature
dependence of thermal pressures (∼0.003 GPa K−1 ; ref. 26) from
800 K to 3,250 K, we would obtain a value of 8.9 GPa. However,
a much smaller value for NaCl was also reported to be 1.2 GPa
at 3,250 K and a pressure of 15.1 GPa (ref. 2). Interestingly, the
higher (8.9 GPa) and the lower values (1.2 GPa) coincide closely
with the MD stresses before and after thermal relaxation (9 GPa
and 1.75 GPa, respectively). However, we should exercise extreme
caution when making such quantitative comparisons between
stresses obtained using different methods, materials and conditions.
As MD simulation is known to overestimate the yield strength of a
material by as much as an order of magnitude20 , we expect an MD
stress to be an upper bound. Qualitatively, we suggest that the initial
stress in a DAC is likely to be larger than the threshold stress, even
with soft pressure media, thereby explaining why pressure media do
not influence the observed transition temperatures.
On transition, disordered atoms, identified by the central
symmetry parameter, increase significantly in population, and
appear throughout the system (Fig. 1b). In contrast, the plastic
deformation of a b.c.c. metal is usually propagated by well-localized
screw dislocations, which are accompanied by possible debris38 .
Also, a decrease of the viscosity is observed along the slip direction.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 225

ARTICLES
a a
8
6
Shear stress (GPa)
4
2
0
0 40
b θ ( 11
4,200
Transition temperature (K)
3,800
3,400
3,000
2 4
Figure 3 | Time history of thermal shear relaxation and transition
temperature as a function of external stress. a, MD stress as a function of
time on heating at temperature T = 3,250 K and pressure P = 17 GPa. Blue
or green indicate whether Ta is a b.c.c. crystal or a plastic flow, respectively.
b, Predicted transition temperatures from b.c.c. to a plastic flow as a
function of external stress at P = 17 GPa. The red dashed line marks the
melting temperature.
For instance, the mean atomic displacement along [111] shows a
discontinuous increase at the transition (see Fig. 1c). In contrast
to the visco-elastic phase proposed in ref. 21, we do not find
polytetrahedra and icosahedral ordering of a supercooled liquid22
or a glass. It is notable that a metallic glass of a pure metal
has yet to be made39,40 . Perhaps this is because [111] chains
are hard to break. Unlike a liquid, long-range atomic diffusions
are not present in this anisotropic plastic flow. However, large
local atomic displacements up to 2 Å are observed along the slip
direction, indicating that it is viscous, likely to flow. Furthermore,
when the pressure exceeds ∼100 GPa, the transition temperature
makes a dramatic increase, consistent with the observation in DAC
experiments that laser-speckle motion becomes undetectable in
Ta around this pressure range1,4 . Notice that the lower boundary
of the proposed b.c.c.-to-f.c.c. transition of Mo also resides18 at
similar pressure (150 GPa). Preliminary analysis of this transition
shows signs of twinning and twisting of domains on {1̄10} planes.
In addition, the equation of state and mechanical property of a
plastic flow are noticeably different from those of b.c.c. and liquid
Ta. However, detailed studies of its transition mechanism and
properties are beyond the scope of this study. Intriguingly, holes
and beads were observed on recovered Ta samples heated above the
DAC melting curve1,4 , which is consistent with onset of plastic flow.
226
NATURE MATERIALS DOI: 10.1038/NMAT2375
δ
b.c.c. → h.c.p.
θ : 109.74° → 120.0° θ
δ : 0 → 0.0833
b
Ta
h.c.p. 5,000
ΔE (K)
3,250 K 3,500
1,500
80 120 160 200 b.c.c.
Time (ps)
108 8
2 0.0
4
°) 0.0
116 δ
0
Melting Figure 4 | Significant anisotropy observed in the ab initio energy
landscape along the b.c.c.-to-h.c.p. Burgers transformation. a, Schematic
drawing of the volume-conserving Burgers transformation. b, Energy
landscape of Ta relative to the b.c.c. structure (1E), near a pressure of
30 GPa. Colours correspond to magnitudes in units of Kelvin. The black
curve is the isoline of 1E equal to the DAC melting temperature of 3,250 K.
Threshold
As we have mentioned above, our MD simulations were
motivated by our ab initio density-functional total-energy results,
which show significant anisotropy when b.c.c. structure is distorted
under an orthorhombic symmetry. In fact, we concluded that the
6 8 10 12 flat energy landscape along the Burgers36 surface is the origin
Stress (GPa) for this transition. Lowering the system symmetry from cubic
to orthorhombic is critical for finding thermal softening at high
temperatures; this was not accounted for in previous studies27–29 of
Ta thermal elasticity. Figure 4a shows the Burgers36 surface, which
is a two-parameter (θ,δ) orthorhombic phase space connecting
b.c.c. and hexagonal close-packed (h.c.p.) structures. As shown in
Fig. 4b, b.c.c. Ta occupies a flat region associated with a geometric
distortion of the {1̄10} planes (variation of θ ) near 30 GPa, which
coincides with the thermal softening of the {1̄10} planes observed
in the MD simulations at the DAC melting temperature of 3,250 K.
Distortions with respect to δ (a fractional displacement between
two adjacent {1̄10} layers) correspond to breaking of bonds along
the [111] chains, thus costing substantially higher energy, which
is consistent with melting near 5,000 K. We found similar flat
features in other b.c.c. metals, such as Mo and V (not shown),
suggesting that this mechanism is likely to hold for other metals
in addition to Ta. Indeed, preliminary MD simulations of V yield
a similar transition. This anisotropic thermal-softening mechanism
is consistent with the observation that the melting discrepancy is
worse for non-close-packed metals (for example b.c.c. and simple
cubic), which are more prone to increasing anisotropy under
pressure. Two metals that do not show the melting discrepancy,
aluminium7 (a simple metal) and copper8 (a noble metal), are both
soft metals with rather isotropic bonding.
We show here that, under shear, b.c.c. Ta transforms into a
viscous plastic flow before becoming a liquid. Remarkably, this
plasticity-mediated shear-induced structural transformation has
considerable similarities to the reported low-temperature DAC
melting. Most of all, the transition boundary is in close proximity
to the low-temperature DAC melt curve1–5 . Its boundary is also
shifted significantly near the pressure condition at which the
onset of speckle motion disappears1,4 . Importantly, this agrees with
X-ray observations of DAC Ta experiments2 and the complicated
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2375
X-ray characteristics observed in melting experiments of other
b.c.c. metals6,9,39,40 . For instance, the strongest X-ray diffraction
peak of a b.c.c. crystal is associated with the {1̄10} planes. This
peak is expected to disappear as b.c.c. Ta transforms into an
anisotropic plastic flow, which is consistent with the disappearance
of X-ray signals in Ta DAC melting2 . At the same time, there
are remaining/new two-dimensional crystalline peaks that associate
with packing of {111} chains. This partially disordered–partially
crystalline feature is consistent with the coexistence of crystalline
and diffusive X-ray signals observed by DAC for iron6 , lead9 ,
titanium39 and zirconium40 . In addition, the viscous plastic flow
is consistent with the DAC observations of optical reflectivity,
holed surface textures and the onset of speckle motion1,4 . Another
possible contributor to solid peaks is dynamic recrystallization20 ,
under which a hot viscous plastic flow crystallizes back to
b.c.c. when cooled by its surroundings. Finally, this transition
is also consistent with the observations of pressure media. The
transition temperature is independent of external stress once it
is large enough to overcome the yield stress. We suggest that
the main difference between the pressure media is the resulting
shear relaxation time. Namely, the initial stress dissipates much
faster using argon medium in comparison to using a high-
strength material such as Al2 O3 medium, thus the pressure
medium may influence the coexistence ratio of disordered versus
crystalline structures.
On the basis of our calculations, we suggest that the above
transition provides a highly probable resolution to the long-
standing melting controversy. Namely, the low-temperature melt-
ing curve of Ta represents a structural transformation from b.c.c. to
a viscous plastic flow under shear. We expect this prediction to be
relevant for all metals showing the melting anomaly. The validation
of our finding depends on future experiments that could shed light
on atomic mobility, anisotropy, shear stress and the possible exis-
tence of a premelting transition. We suggest implanting a soluble
impurity on a localized area of the metal sample, heating it to the
DAC melting temperature and then analysing the mixing between
the impurity and the surrounding metal. A complete mixing would
prove it is a liquid; otherwise, it is more like a viscous plastic
flow. Given the presence of shear in shock compression, we expect
shock-wave experiments to go through a similar transformation as
well (for example the reported transition at 210 GPa in Mo; ref.19).
However, it may be harder to detect a plastic flow than a liquid,
because the transformation to a plastic flow is likely to yield a
smaller change in sound speed than melting. Our finding demon-
strates the importance of looking beyond thermodynamically stable
phases when explaining unconventional phenomena observed by
high-pressure experiments. Furthermore, it has implications for
the possible existence of a viscous iron Earth core, which could be
responsible for the observed inner-core anisotropy.
Methods
The MD simulations were performed using the Gordon Bell Prize-winning MD
code developed by one of the authors, J. Glosli, and his co-workers41 . This code
was designed specifically to handle the many-body (up to four) interactions
of the MGPT (ref. 33) potential on the platform of the massively parallel
teraflop computer Blue Gene/L at Lawrence Livermore National Laboratory.
We conducted our MD simulations in a constant-number (N = 28,800 atoms),
volume and temperature ensemble at four volumes (V = 17.64, 17.16, 16.38
and 14.90 Å3 per atom) and a series of temperatures between 2,000 K and
6,250 K. Figure 1a shows the corresponding pressures at each volume and
temperature. An unsheared simulation cell assumes ideal b.c.c. symmetry and
has 10, 16 and 30 repeating units in the x[112̄], y[1̄10] and z[111] directions,
respectively. This is the most widely used orientation among MD simulations
applied to study plasticity and dislocation mobility38,42 of b.c.c. metals. It gives
a dimension of 80.3 Å × 74.2 Å × 85.2 Å at a volume of 17.64 Å3 per atom.
Before the heating, we sheared the system uniformly along the well-known
b.c.c. slip system of {1̄10} [111] to the targeted strains by displacing atoms
from their ideal b.c.c. lattice positions into the sheared geometry according
to the displacement field gradient. This mimics the potential stress in a DAC
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
ARTICLES
experiment before laser heating. We then applied periodic boundary conditions
in all three dimensions. However, the simulation cell remained in the shape of
a rectangle, independent of the applied stress, because of an implementation
constraint in the MD code. Owing to this restriction, the system is strained by
an amount that would yield increments of lattice spacing at the boundary to
avoid lattice mismatch. However, convergence tests show that the effect of the
boundary, different methods of shearing and insertion of a dislocation have
little influence on the transition temperatures obtained. We heat the system
to the targeted temperature by using a Langevin thermostat43 . After checking
the convergence on time step, we use a 2 fs time step in the time integration
of the Verlet algorithm43 . We heated the system from 298 K to the desired
temperature either immediately or during a short period of time, such as
0.5–4 ps. We found that the rate of heating does not affect the final simulation
result. The system is then allowed to evolve independently for 40 ps, which has
been shown to be a few times longer than the time needed for the structural
transformation (see Fig. 3a). Simulations to 375 ps have also been conducted
to explore the thermal relaxation behaviour at a longer time. We have also
tested the system-size effect by using a larger simulation cell of 54,720 atoms
(x = 12 × [112̄], y = 20 × [1̄10] and z = 38 × [111]), which gives similar results to
those using 28,800-atom cells.
To map out the energy surface of Ta, we carried out ab initio plane-wave
pseudopotential density functional theory calculations using the ABINIT package44 .
Density functional theory forces on ions were calculated using the Perdew, Burke
and Ernzerhof exchange–correlation functional45 . Core electrons were treated
with Troullier–Martins46 pseudopotentials. We used a plane-wave basis function
with a kinetic-energy cutoff of 120 Ryd. Ions were held in positions defined by the
Burgers path, and the electron temperature was set to be the temperature of the
DAC melting at the corresponding pressure imposed by Fermi–Dirac smearing.
The k-point integration was done using the special k-point scheme47 with a mesh
of 30 × 30 × 30 in the full Brillouin zone. All necessary convergence tests were
performed. We achieved a convergence of 1 K per atom in mapping the energy
landscape. As a comparison, we also performed all-electron (no pseudopotential)
calculations in a similar fashion to the pseudopotential ones described above,
which confirmed the plane-wave results.
Received 27 August 2008; accepted 23 December 2008;
published online 25 January 2009
References
1. Errandonea, D. et al. Systematics of transition-metal melting. Phys. Rev. B. 63,
132104 (2001).
2. Errandonea, D., Somayazulu, M., Hausermann, D. & Mao, H. K. Melting of
tantalum at high pressure determined by angle dispersive x-ray diffraction in
a double-sided laser-heated diamond-anvil cell. J. Phys. Condens. Matter 15,
7635–7649 (2003).
3. Errandonea, D. Improving the understanding of the melting behaviour of Mo,
Ta and W at extreme pressures. Physica B 357, 356–364 (2005).
4. Errandonea, D. Phase behavior of metals at very high P–T conditions:
A review of recent experimental studies. J. Phys. Chem. Solids 67,
2017–2027 (2006).
5. Errandonea, D., Boehler, R. & Ross, M. Melting of the alkaline-earth metals to
80 GPa. Phys. Rev. B 65, 012108 (2001).
6. Boehler, R., Santamaría-Pérez, D., Errandonea, D. & Mezouar, M. Melting,
density, and anisotropy of iron at core conditions: New X-ray measurements
to 150 GPa. J. Phys. Conf. Ser. 121, 022018 (2008).
7. Boehler, R. & Ross, M. Melting curve of aluminum in a diamond cell to 0.8
Mbar: Implications for iron. Earth Planet. Sci. Lett. 153, 223–227 (1997).
8. Japel, S., Schwager, B., Boehler, R. & Ross, M. Melting of copper and nickel at
high pressure: The role of d electrons. Phys. Rev. Lett. 95, 167801 (2005).
9. Dewaele, A., Mezouar, M., Guignot, N. & Loubeyre, P. Melting of lead under
high pressure studied using second-scale time-resolved x-ray diffraction.
Phys. Rev. B. 76, 144106 (2007).
10. Brown, J. M. & Shaner, J. W. in Shock Waves in Condensed Matter—1983
(eds Asay, J. A., Graham, R. A. & Straub, G. K.) (North-Holland, 1984).
11. Fateeva, N. S. & Vereshchagin, L. F. Melting curve of tantalum up to 60 kbar.
Sov. Phys. 16, 322–323 (1971).
12. Wang, Y., Ahuja, R. & Johnsson, B. Melting of iron and other metals at
earth’s core conditions: A simplified computational approach. Phys. Rev. B 65,
014104 (2001).
13. Verma, A. K., Rao, R. & Godwal, B. K. Theoretical solid and liquid state shock
hugoniots of Al, Ta, Mo and W. J. Phys. Condens. Matter 16, 4799–4809 (2002).
14. Taioli, S., Cazorla, C., Gillan, M. J & Alfè, D. Melting curve of tantalum from
first principles. Phys. Rev. B 75, 214103 (2007).
15. Luo, S. & Swift, D. C. On high-pressure melting of tantalum. Physica B 388,
139–144 (2007).
16. Liu, Z. L., Cai, L. C., Chen, X. R. & Jing, F. Q. Molecular dynamics simulations
of the melting curve of tantalum under pressure. Phys. Rev. B 77, 024103 (2008).
17. Xi, F. & Cai, L. Theoretical study of melting curves on Ta, Mo and W at high
pressures. Physica B 403, 2065–2070 (2008).
227
ARTICLES
18. Belonoshko, A. B. et al. Molybdenum at high pressure and temperature:
Melting from another solid phase. Phys. Rev. Lett. 100, 135701 (2008).
19. Hixson, R.A., Boness, D.A., Shaner, J.W. & Moriarty, J. A. Acoustic velocities
and phase transitions in molybdenum under strong shock compression.
Phys. Rev. Lett. 62, 637–640 (1989).
20. Belonoshko, A. B. & Dubrovinsky, L. S. A simulation study of induced
failure and recrystallization of a perfect MgO crystal under non-hydrostatic
compression: Application to melting in the diamond-anvil cell. Am. Mineral.
82, 441–451 (1997).
21. Ross, M., Errandonea, D. & Boehler, R. Melting of transition metals at high
pressure and the influence of liquid frustration: The early metals Ta and Mo.
Phys. Rev. B 76, 184118 (2007).
22. Jakse, N., Le Bacq, O. & Pasturel, A. Predication of the local structure of liquid
and supercooled tantalum. Phys. Rev. B 70, 174203 (2004).
23. Wolf, D., Okamoto, P. R., Yip, S., Lutsko, J. F. & Kluge, M. Thermodynamic
parallels between solid-state amorphization and melting. J. Mater. Res. 5,
286–301 (1990).
24. Belonoshko, A. B., Ahuja, R. & Johansson, B. Molecular dynamics of LiF
melting. Phys. Rev. B 61, 11928–11935 (2000).
25. Fiquet, G, Andrault, D., Itié, J. P., Gillet, P. & Richet, P. X-ray diffraction of
periclase in a laser-heated diamond-anvil cell. Phys. Earth Planet. Inter. 95,
1–17 (1996).
26. Vinod, K., Malik, V. S., Sharma, S. K. & Srivastava, S. K. Temperature
dependence of thermal pressure of NaCl and KCl crystals. J. Phys. Chem. Solids
68, 32–35 (2007).
27. Foata-Prestavoine, M., Robert, G., Nadal, M. H. & Bernard, S. First-principles
study of the relations between the elastic constants, phonon dispersion curves,
and melting temperatures of b.c.c. Ta at pressures up to 1000 GPa. Phys. Rev. B
76, 104104 (2007).
28. Gulseren, O. & Cohen, R. E. High-pressure thermoelasticity of
body-centred-cubic tantalum. Phys. Rev. B 65, 064103 (2002).
29. Orlikowski, D., Soderlind, P. & Moriarty, J. A. First-principles thermoelasticity
of transition metals at high pressure: Tantalum prototype in the quasiharmonic
limit. Phys. Rev. B 74, 054109 (2006).
30. Butler, S. & Harrowell, P. The shear-induced disordering transition
in a colloidal crystal: Nonequilibrium Brownian dynamic simulations.
J. Chem. Phys. 103, 4653–4671 (1995).
31. King, R. P. Introduction to Practical Fluid Flow 117
(Butterworth-Heinemann, 2002).
32. Bingham, E. C. An investigation of the laws of plastic flow. US Bureau of
Standards Bulletin 13, 309–353 (1916).
33. Moriarty, J. A. et al. Quantum-based atomistic simulation of materials
properties in transition metals. J. Phys. Condens. Matter 14, 2825–2857 (2002).
34. Moriarty, J. A. et al. Supplemental Proceedings: Volume 1: Minerals, Metals
and Materials under Pressure for the 2008 TMS Annual Meeting, New Orleans,
LA, 313.
35. Hirth, J. P & Lothe, J. Theory of Dislocation 275 (Krieger, 1992).
228
NATURE MATERIALS DOI: 10.1038/NMAT2375
36. Burgers, W. G. On the process of transition of the cubic-body-centered
modification into the hexagonal-close-packed modification of zirconium.
Physica 1, 561–586 (1934).
37. Szczgielska, A., Burian, A., Duber, S., Dore, J. C. & Honkimaki, V. Radial
distribution function analysis of the graphitization process in carbon materials.
J. Alloys Compounds 328, 231–236 (2001).
38. Jaime, M., Cai, W. & Bulatov, V. V. Dynamic transitions from smooth to rough
to twinning in dislocation motion. Nature Mater. 3, 158–163 (2004).
39. Wang, Y. et al. Amorphouslike diffraction pattern in solid metallic titanium.
Phys. Rev. Lett. 95, 155501 (2005).
40. Hattori, T. et al. Does bulk metallic glass of elemental Zr and Ti exist?
Phys. Rev. Lett. 96, 255504 (2006).
41. Streitz, F. H., Glosli, J. N. & Patel, M. V. Beyond finite-size scaling in
solidification simulations. Phys. Rev. Lett. 96, 225701 (2006).
42. Rodney, D. & Martin, G. Dislocation pinning by small interstitial loops: A
molecular dynamics study. Phys. Rev. Lett. 82, 3272–3275 (1999).
43. Allen, M. P. & Tildesley, D. J. Computer Simulations of Liquids
(Clarendon, 1987.).
44. Gonze, X. et al. First-principles computation of material properties: The
ABINIT software project. Comput. Mater. Sci. 25, 478–492 (2002).
45. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation
made simple. Phys. Rev. Lett. 77, 3865–3868 (1996).
46. Troullier, N. & Martins, J. L. Efficient pseudopotentials for plane-wave
calculations. Phys. Rev. B 43, 1993–2006 (1991).
47. Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations.
Phys. Rev. B 13, 5188–5192 (1976).
Acknowledgements
The authors would like to thank J. A. Moriarty for providing the MGPT Ta potential,
M. Ross, N. C. Holmes, W. J. Evans, M. J. Lipp, M. Tang, R. Gee and D. A. Orlikowski
for useful discussions and K. Kline and J. McInnis for their contributions in preparation
of the manuscript and figures. This work was carried out under the auspices of the US
Department of Energy by Lawrence Livermore National Laboratory under Contract
DE-AC52-07NA27344.
Author contributions
C.J.W. originated the central idea, performed and analysed the Ta MD simulations
and the Ta ab initio electronic structure calculations, carried out code development
(central symmetry) and wrote the paper; P.S. contributed in ab initio electronic structure
calculations and manuscript editing; J.N.G. carried out MD code development; J.E.K.
contributed in manuscript editing; all contributed to discussions.
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to C.J.W.
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
 ARTICLES
PUBLISHED ONLINE: 25 JANUARY 2009 DOI: 10.1038/NMAT2373

Conduction at domain walls in oxide multiferroics

J. Seidel1,2 *† , L. W. Martin2,3 *, Q. He1 , Q. Zhan2 , Y.-H. Chu2,3,4 , A. Rother5 , M. E. Hawkridge2 ,
P. Maksymovych6 , P. Yu1 , M. Gajek1 , N. Balke1 , S. V. Kalinin6 , S. Gemming7 , F. Wang1 , G. Catalan8 ,
J. F. Scott8 , N. A. Spaldin9 , J. Orenstein1,2 and R. Ramesh1,2,3

Domain walls may play an important role in future electronic devices, given their small size as well as the fact that their location
can be controlled. Here, we report the observation of room-temperature electronic conductivity at ferroelectric domain walls
in the insulating multiferroic BiFeO3 . The origin and nature of the observed conductivity are probed using a combination of
conductive atomic force microscopy, high-resolution transmission electron microscopy and first-principles density functional
computations. Our analyses indicate that the conductivity correlates with structurally driven changes in both the electrostatic
potential and the local electronic structure, which shows a decrease in the bandgap at the domain wall. Additionally, we
demonstrate the potential for device applications of such conducting nanoscale features.

on scanning probe microscopy14 , are invaluable in observing such

C
orrelated oxide systems are an exciting and challenging
area of condensed-matter research, with their interacting
and competing charge, spin, orbital and lattice degrees
of freedom forming new electronic and magnetic phases1,2 .
These phases can be controlled through stress, optical excitation
and electric and magnetic fields and have great potential for
applications in the fields of spintronics, information storage and
communications. Among the correlated oxides, the multiferroics,
which show more than one type of ferroic order in the same phase,
are attracting particular interest3–6 . The defining characteristic of
a ferroic material is an order parameter (electric polarization
in ferroelectrics, magnetization in ferromagnets or spontaneous
strain in ferroelastics) that has different, energetically equivalent
orientations, the orientation of which can be selected using an
applied field. This often leads to the appearance of domains
of differently oriented regions, separated by domain walls,
coexisting in a sample7 . Such domain walls will become more
technologically important as the dimensions of individual elements
in devices continue to shrink.
Although the morphology and properties of domains and their
walls have been studied for more than 50 years, in recent times there
has been increasing focus on novel functionality at domain walls8–12 .
For example, it has been predicted theoretically that the ferroelectric
walls in magnetoelectric multiferroics can be ferromagnetic even
if the domains themselves are antiferromagnetic9–11 . Conversely,
spin rotation across ferromagnetic domain walls in insulating
ferromagnets can induce a local polarization in the walls of
an otherwise non-polar material5,12 . Experimentally, unusual
functional properties of domain walls have also been observed:
preferential doping along domain walls has been reported to
induce two-dimensional superconductivity in WO3−x (ref. 13)
and enhanced resistivity in phosphates14 , whereas in paraelectric,
non-polar SrTiO3 the ferroelastic domain walls seem to be
ferroelectrically polarized15 . Local probes, particularly those based
effects at the nanoscale.
Here we describe the observation of a new behaviour—
room-temperature electrical conductivity—at ferroelectric domain
walls in the prototypical multiferroic BiFeO3 (BFO). BFO
is a room-temperature G-type antiferromagnet (TN ∼ 650 K)
and a rhombohedral ferroelectric (TC ∼ 1,103 K), with a large
spontaneous ferroelectric polarization (∼90 µC cm−2 ) along the
pseudocubic 111 directions16,17 . Such rhombohedral ferroelectrics
possess 71◦ , 109◦ and 180◦ domain walls forming on {101},{100}
and planes that satisfy the requirement that ±h ± k + l = 0,
respectively18 . All three wall orientations have been observed
in BFO (refs 19,20).
Epitaxial BFO films (∼100 nm thick) were grown using laser
molecular-beam epitaxy in (111), (110) and (100) orientations,
using carefully controlled single-crystal SrTiO3 substrates. A thin
50 nm layer of epitaxial SrRuO3 was used as a bottom electrode
for electrical-contact purposes. Ferroelectric domains were im-
aged using piezoresponse force microscopy (PFM) as described
previously19 . Controlled ferroelectric domain patterns were written
using PFM by applying a d.c. voltage to the probe tip. Local electrical
conductivity was measured using high-resolution conductive
atomic force microscopy (c-AFM) (details of measurements are
provided in the Methods section) by applying a bias voltage (below
the polarization switching voltage) between the conductive AFM
tip and the bottom electrode of the sample.
100-nm-thick epitaxial films grown on (110) surfaces show a
two-variant ferroelectric domain structure in the as-grown states
with average domain sizes between 5 and 10 µm. On electrical
switching at high fields, all three variations of domain walls can
be created21 ; therefore, we begin our study on these (110)-oriented
films. The typical morphology is shown in Fig. 1a. We measured
the RMS roughness of the films to be ∼0.5 nm and there are
no observable surface features, either before or after switching,

1 Department of Physics, University of California, Berkeley, 94720 California, USA, 2 Materials Sciences Division, Lawrence Berkeley National Laboratory,

94720 California, USA, 3 Department of Materials Science and Engineering, University of California, Berkeley, 94720 California, USA, 4 Department of
Materials Science and Engineering, National Chiao Tung University, HsinChu, 30010, Taiwan, 5 Institute of Structure Physics, Triebenberg Laboratory,
Technische Universität Dresden, DE-01062, Germany, 6 Center for Nanophase Materials Science, Oak Ridge National Laboratory, 37831 Tennessee, USA,
7 Forschungszentrum Dresden-Rossendorf, 01314 Dresden, Germany, 8 Department of Earth Sciences, University of Cambridge, Cambridge CB2 3EQ, UK,
9 Materials Department, University of California, Santa Barbara, 93106 California, USA. *These authors contributed equally to this work.
† e-mail: jseidel@berkeley.edu.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 229

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT2373

a b
D

1 µm

c d 0.3 pA
D D
0.15

0
180°

U 71°
U

109°

Figure 1 | Conductivity at different wall types. a, Topography of a BFO(110) thin film, r.m.s. roughness of 0.5 nm. b, Out-of-plane PFM image of a written
domain pattern in a monodomain BFO(110) film showing the out-of-plane polarization component of the domains to be either down, labelled as ‘D’
(white), or up, labelled as ‘U’ (black). c, In-plane PFM image of a written domain pattern in a monodomain BFO(110) film showing all three types of domain
wall, that is, 71◦ (blue), 109◦ (red) and 180◦ (green), as inferred from the combination of both out-of-plane and in-plane PFM images. In these images, both
the out-of-plane (U or D) component and the in-plane projection of the polarization direction (shown as an arrow) are also labelled. d, Corresponding
c-AFM image showing conduction at both 109◦ and 180◦ domain walls; note the absence of conduction at the 71◦ domain walls.

a b
+V 1 µm

BFO ¬2 V

Ground
SRO
STO ¬1.5 V

¬1 V

c 500 d 500
On wall
Off wall
400 400

300 300
Current (fA)

Current (fA)

200 200

100 100

0 0

¬2.5 ¬2.0 ¬1.5 ¬1.0 ¬05 0 0.5 1.0 1.5 2.0 2.5 0 50 100 150 200
Applied field (V) Time (s)

Figure 2 | Conduction at domain walls. a, Schematic illustration of the experimental c-AFM set-up. b, (Top) Out-of plane PFM image of a written 180◦
domain in a monodomain BFO(110) sample and (lower) corresponding c-AFM current maps for −1, −1.5 and −2 V sample bias made with a Pt-coated tip.
c, I–V curves taken both on the domain wall (black) and off the domain wall (red) reveal Schottky-like behaviour. d, Time dependence of the current both
on the wall (black) and off the wall (red) at an applied sample bias of −2 V. Results are qualitatively similar for N-doped diamond tips.

230 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2373 ARTICLES
a Domain I Domain II b 4.1 Domain I Domain II

Lattice parameter (Å)

4.0

3.9

109° domain wall

c ([001])
a ([100])
3.8
Pl 0 2 4 6 8 10 12 14 16 18 20 22 24
Unit cell

Pll
c 0.45

0.40

0.35

Displacement of Fe ion from

centre of unit cell (Å)
0.30
0.45

0.30 Displacement from

0.15 centre of unit cell:
[001]
[100] [001] [100]
0
[010] ¬0.15

–0.30 Domain I Domain II

1 3 5 7 9 11 13 15 17 19 21 23 0 2 4 6 8 10 12 14 16 18 20 22 24
Unit cell

Figure 3 | Structural analysis of domain walls. a, Schematic diagram of 109◦ domain wall and exit-wave-reconstructed HRTEM image of a 109◦ domain
wall imaged along the [010] zone axis. b, Extracted a and c lattice parameters for each unit cell across the domain wall. c, Extracted Fe-ion displacement
relative to the Bi lattice for each unit cell across the domain wall. A close-up (upper panel) reveals an increase in the component of polarization
perpendicular to the domain wall.

corresponding to the conducting features. In Fig. 1c, we show the

in-plane PFM image of a written domain pattern controlled to Table 1 | Electronic structure at ferroelectric domain walls.
have all three domain-wall types. The complicated domain shapes
Electrostatic
occur only when the high voltages required to stabilize all three
Domain potential step Change in
domain-wall variants are applied21 . The various domain-wall types
wall type (◦ ) (eV) bandgap (eV)
are determined using both in-plane (Fig. 1c) and out-of-plane
(Fig. 1b) PFM images and are labelled accordingly. Our main 71 0.02 0.05
result is shown in the corresponding c-AFM trace (Fig. 1d): the 109 0.15 0.10
occurrence of electrical conduction at 109◦ and 180◦ domain walls, 180 0.18 0.20
and the absence of conduction at 71◦ domain walls. BFO films
Calculated values of the potential step and reduction in bandgap at all three domain-wall types.
grown on (001)- and (111)-oriented substrates also consistently
show conduction at 109◦ and 180◦ domain walls (Supplementary
Information, Fig. S1); in no cases do 71◦ domain walls show over a timescale of at least 3 min, which is limited by the drift in
conduction within the resolution of our measurements. our scanning system (Fig. 2d). These time-dependent data indicate
Next we discuss in detail our c-AFM measurements on the that the origin of this current is not displacement of domain walls.
(110)-oriented BFO films. A schematic diagram of the experimental Additional details of conduction at domain walls are provided
set-up is shown in Fig. 2a. The spatial resolution of the technique in the Supplementary Information (Supplementary Information,
is limited by the tip radius of ∼20 nm. Figure 2b (top panel) Fig. S2). Ultrahigh-vacuum-based c-AFM measurements were used
shows a PFM image of two domains separated by a 180◦ domain to further probe the nature of conduction and I –V characteristics
wall. The corresponding c-AFM images (lower panels) show of the conducting domain walls—including the observation of
enhanced conduction at the domain wall for applied bias voltages enhanced current values.
of −1 to −2 V. Current–voltage (I –V ) curves (Fig. 2c) show To understand the observed electrical conductivity, we
resistive behaviour within the domain (red data) and Schottky-like performed a combined transmission electron microscopy (TEM)
behaviour suggesting activated conduction at the domain wall and density functional theory (DFT) study of the domain-wall
(black data). We repeated our I –V measurements with a number structure and properties. The structure of domain walls in
of different c-AFM tip materials—including Pt and N-doped ferroelectric perovskites and related materials has been the focus
diamond—and found similar Schottky-like behaviour with slightly of several recent DFT (refs 18,22) and TEM (refs 23–25) studies,
shifted conduction onsets. Furthermore, the current is persistent and the structural distortions associated with ferroelectric domain

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 231

ARTICLES
walls are now reasonably well characterized. However, studies of
the associated electronic properties are in their infancy26 . We chose
the 109◦ domain wall (shown schematically in Fig. 3a) for our
study because we do not obtain conduction at 71◦ domain walls
and because imaging of 180◦ domain walls with high-resolution
TEM (HRTEM) presents practical problems in terms of locating
the wall. (For completeness, HRTEM images of 71◦ domain
walls are shown in Supplementary Information, Fig. S3.) We use
(001)-oriented samples for the TEM analysis, because we can
easily control the density of 109◦ domain walls during growth for
this orientation. TEM images were acquired using the exit-wave
reconstruction approach to eliminate the effects of objective-lens
spherical aberrations; such images can be directly interpreted in
terms of the projection of the atomic columns27 . A representative
image for such a 109◦ domain wall taken along the [010] zone
axis of a cross-sectioned BFO/SRO/STO(001) film, where SRO is
SrRuO3 and STO is SrTiO3 , is shown in Fig. 3a. Analysis of this
image enables us to determine the lattice parameter in the plane
of the film (a) ([100]) and the lattice parameter out of the plane
of the film (c) ([001]) (Fig. 3b). As expected, the in-plane lattice
parameter is slightly smaller and the out-of-plane lattice parameter
larger than the values in bulk BFO (3.96 Å; ref. 16) owing to the
strain inherent in the epitaxial films. In addition, we find that
both the in-plane and out-of-plane film lattice parameters are
unchanged in the vicinity of the domain wall. We also extract the
relative displacement of the Fe ion with respect to the Bi sublattice,
and resolve it into components parallel ([001]) and perpendicular
([100]) to the domain wall (which lies in a (100) plane running
vertically across the image and labelled with a dashed orange line)
by quantitative analysis of the HRTEM data (Fig. 3c); this distance
is representative of the local polarization. The component of the
displacement parallel to the domain wall (along [001]) decreases in
magnitude to zero at the centre of the domain wall before changing
to the same magnitude (but opposite sign) on the other side of the
wall, reflecting the change in polarization orientation of the domain.
Interestingly, the perpendicular displacement component (along
[100]) shows a small increase at the domain wall; we return to the
implications of this later. Similar studies were completed on a 71◦
domain wall as well (Supplementary Information, Fig. S2). Again,
only minor variation in lattice parameters was observed across the
domain wall. In this case a similar step in Fe-ion displacement is
observed parallel to the domain wall, but we are unable to resolve a
step in the perpendicular component across the wall.
To investigate the influence of these structural changes on the
electronic properties, we next performed a density functional study
of the structure and electronic properties for all three ferroelectric
domain-wall variants (details in the Methods section). First we
performed full structural optimizations of the ionic positions with
the lattice parameters fixed to their experimental bulk values; in
particular, the oxygen polyhedral rotations around the polar axis,
which are known to have a profound effect on both the magnetic28
and electronic29 properties and cannot be easily extracted from the
HRTEM data, were accurately calculated. As the sense of the oxygen
rotations around the polar axis is independent of the direction
of polarization along the axis28 we studied two scenarios: first
we initialized the sense of rotation to be continuous across the
domain boundary and second we changed the rotation sense when
the polarization direction changed. As expected, we found that
domain walls with continuous oxygen rotations are considerably
lower in energy, because this avoids formation of an antiphase
boundary associated with the octahedral rotations. In addition, we
investigated domain-wall configurations centred at both the Bi–O
and Fe–O planes and found the Bi–O walls to be slightly lower
in energy, confirming the findings of the HRTEM analysis. Our
lowest-energy calculated configuration for the 109◦ domain wall has
a domain-wall energy of 206 mJ m−2 .
232
NATURE MATERIALS DOI: 10.1038/NMAT2373
a AFM
BiFeO3
1.2 µm
100 nm
PFM OOP
30 nm
SrRuO3 6 µm
1.2 µm
SrTiO3
b
1.5 Number of R
written features
1.0 0 1
2 3
0.5
2 (after erase)
I (nA)
0
¬0.5
¬1.0
¬1.5
¬10 ¬5 0 5 10
Bias (V)
Figure 4 | Proof of concept for device application. a, Schematic illustration
(left) of in-plane electrode structure (green) and how scanning probe tips
can be used to controllably create conductive domain-wall features
between electrodes. The images on the right show AFM (top) and
out-of-plane PFM (bottom) contrast for this written domain area on a
BFO(110) sample. b, I–V characteristics of the devices measured between
the two in-plane electrodes show that the current can be incrementally
controlled through creating or erasing the conducting domain walls.
To confirm that our calculated structure is consistent with our
TEM data, we analysed the layer-by-layer polarization, defined as
the sum over the bulk Born effective charges29 multiplied by the
displacements of the ions from their centrosymmetric reference
positions in each layer. As expected, the local polarization in the
middle of the domain is close to the value calculated for bulk BFO
using the same computationally and experimentally determined
lattice parameters (∼93 µC cm−2 ; refs 29–32), confirming that our
supercell is large enough to capture the essential physics. Consistent
with our TEM analysis, we find an abrupt change in the parallel
polarization component across the domain wall and a small change
in the normal component at the domain wall.
Our calculations indicate that this small change in the normal
component of the polarization across the 109◦ domain wall leads
to a step in the electrostatic potential (planar and macroscopically
averaged) of 0.15 eV across the domain wall (Table 1); a similar
step was computed and explained previously across 90◦ domain
walls in PbTiO3 (ref. 22). Such a potential step should enhance
the electrical conductivity by causing carriers in the material
to accumulate at the domain wall to screen the polarization
discontinuity. Our calculations for the 180◦ domain wall also
yield a variation in the normal component of the polarization,
and a corresponding potential step of 0.18 eV (Table 1). The
normal component results from the polarization rotating towards
successive adjacent corners of the perovskite unit cell, through a 71◦
and than a 109◦ change in the polarization direction before reaching
the reversed polarization16 . This behaviour is in striking contrast to
the 180◦ polarization reversal in tetragonal ferroelectrics22 , where
the polarization changes in only one direction within the wall
plane and no normal component occurs. The 71◦ wall, however,
exhibits no anomaly in the perpendicular component of the ion
displacement at the centre of the wall, again consistent with our
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
NATURE MATERIALS DOI: 10.1038/NMAT2373
TEM data, and therefore exhibits a negligible electrostatic potential
step at the domain wall (Table 1).
Finally, we calculate the electronic properties of our structurally
optimized domain walls, in particular by comparing the
layer-by-layer densities of states in the domain wall and
mid-domain regions. Within the central region of the domain, we
find, as expected, that the local density of states resembles that
of bulk BFO, and the local Kohn–Sham bandgap is equal to the
value of 1.3 eV obtained for bulk BFO with the same choice of
Hubbard parameter U and exchange interaction J , 3 and 1 eV,
respectively29 . (We emphasize that, although the DFT Kohn–Sham
bandgaps do not correspond to experimental bandgaps, changes
in DFT gaps caused by changes in bandwidth as a consequence of
small changes in structure for the same DFT implementation are
qualitatively meaningful.) As the domain wall is approached, we
find that the changes in the structure do indeed cause changes in
the band width and the positions of the band edges. This leads in
the 109◦ (180◦ ) case to a 0.1 eV (0.2 eV) reduction in the bandgap
in the domain-wall layer from the mid-domain calculated value of
1.3 eV (Table 1). For activated conduction at room temperature,
such a change in bandgap, or in band-edge offset relative to
the Fermi energy of the tip, should lead to considerable changes
in conductivity. Consistent with its absence of conduction, the
reduction in bandgap in the 71◦ case is smaller (0.05 eV) (Table 1).
Interestingly, the magnitude of the bandgap reduction is sensitive
to the details of the lattice parameters used in the calculation; if we
constrain the lattice parameters to the bulk values, the changes in
the bandgap are larger than those for the relaxed structures given in
Table 1. This suggests that band-structure changes at domain walls
might be tunable by epitaxial strain.
Our conductivity measurements, TEM analysis and DFT
calculations therefore suggest two mechanisms, which probably
combine to yield the observed conductivity at the 109◦ and 180◦
domain walls: (1) an increased carrier density as a consequence
of the electrostatic potential step at the wall and (2) a decrease
in the bandgap within the wall and corresponding reduction in
band offset with the c-AFM tip. Both factors are the result of
structural changes at the wall.
The potential of these conducting domain walls for possible
device applications is illustrated in Fig. 4. By creating a simple device
structure consisting of in-plane electrodes of SRO separated by
a 6 µm spacing (Fig. 4a), we can measure the I –V characteristics
of BFO films and domain walls macroscopically. Additionally,
the SRO provides nearly Ohmic contacts with the BFO films
and thus we can gain further insight into the conduction of the
walls in the gap, without any interference from the AFM tip
during the measurement process. Monodomain (110)-oriented
BFO films were grown on top of the SRO in-plane electrode
device structures on STO(110) substrates. Conducting domain-wall
features (here are shown 180◦ domain walls, Fig. 4a, right) that
connect the two in-plane electrodes were written using PFM. Again,
no morphological surface features were observed that correspond
to the written domain pattern. I –V measurements (Fig. 4b) reveal a
step-like increase in the measured current between the two in-plane
electrodes on addition of a controlled number of conducting
domain walls. The steps in conduction are essentially equidistant,
increase proportionally to the total number of domain walls written
and show completely reversible behaviour on erasing a given
feature. I –V curves for zero, one, two and three domain features
are shown in Fig. 4b. Such material functionality has potential
application in both logic and memory applications, as the wall
location (and hence electronic conduction) can be precisely con-
trolled on the nanoscale. This demonstrates a possible rewritable,
multiconfiguration device set-up that uses nanoscale conductive
channels (that is, conducting domain walls). Finally, on the basis of
a simple sheet-resistance model, we can estimate the resistivity of a
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
ARTICLES
single domain wall in BFO to be of the order of 1–10  m, which is
between five and six orders of magnitude lower than for bulk BFO.
In summary, our results show that ferroelectric domain walls in
multiferroic BFO show unusual local electronic transport behaviour
that is quite different from that in the bulk of the material or
in conventional ferroelectrics. The conductivity is consistent with
our observed changes in structure at the domain wall and can
be activated and controlled on the scale of the domain wall
width—about 2 nm in BFO. The present work represents a ‘proof
of concept’ that domain walls are discrete functional entities, which
may be addressed and sensed, suggesting potential utility in novel
nanoelectronic applications.
Methods
Nanoscale conductivity measurements were performed on a Digital Instruments
Nanoscope-IV Multimode atomic force microscope equipped with a c-AFM
application module (TUNA). The investigations were performed with commercially
available N-doped diamond-coated Si tips (NT-MDT) and Ti/Pt-coated Si tips
(MikroMasch). Current-amplification settings of the c-AFM equipment of 1 and
10 V pA−1 at an applicable voltage range of ±12 V were used. For a typical scan rate
of 0.5–1.0 µm s−1 , the noise level was of the order of 50 fA at a bandwidth of 250 Hz.
All data were acquired under ambient conditions and at room temperature and all
such c-AFM measurements were made within a few minutes after the domain wall
was created by electrical switching.
Structural investigations have been carried out using a Philips CM300 field
emission gun microscope with a point-to-point resolution of 1.7 Å and information
resolution of 0.8 Å. The domain structure was investigated in cross-section [010]
orientation. The representative phase of the complex electron exit wave, as shown
in Fig. 3a, was reconstructed from a series of 20 high-resolution images of the BFO
109◦ domain wall using the TrueImage software package33,34 . According to the
weak phase-object approximation for a thin object, the heavy Bi columns projected
along the [010] direction correspond to the brightest dots and the less bright ones
represent Fe in the phase.
DFT calculations were performed using the projector augmented wave
implementation of the VASP code35 with standard projector augmented wave
potentials (Bi, 6s2 5d 10 6p3 ; Fe, 3p6 4s2 3d 6 ; O, 2s2 2p4 ) and the local spin density
approximation (LSDA) +U method to describe the exchange–correlation
potential36 . We used a cut-off energy of 550 eV and U /J values of 3/1 eV; these
values were shown in previous work29 to give a bandgap of 1.3 eV for bulk
rhombohedral BFO and to yield a small bandgap even in the high-symmetry cubic
perovskite phase. We constructed supercells containing two oppositely polarized
domains and two domain walls; each domain contained six Bi2 Fe2 O6 units initially
set to the bulk BiFeO3 structure. Note that the systems in our calculations are in all
cases electrically neutral with uncharged domain walls. We used the experimental
bulk lattice parameters, with the rhombohedral angle set to 60◦ so that the supercell
could accommodate the broken symmetry of the domain wall. The Brillouin
zone integration was performed on a 5 × 5 × 1 Γ -centred Monkhorst–Pack
grid incorporating a Gaussian averaging, with an energy width of 0.05 eV. Ionic
positions were relaxed within the local minimum of maintaining two domains
and domain walls per supercell until the total energy of the supercell converged to
within 0.01 eV and the forces acting on individual ions were below 0.03 eV Å−1 .
The magnetism was set to the G-type antiferromagnetic ordering observed in bulk
BFO and remained unchanged throughout the relaxation of the domain boundary.
All parameters chosen for the supercell calculations were tested in calculations
for single unit cells and shown to yield well-converged total energies, with atom
positions and band structures in good agreement with previous work.
Received 12 May 2008; accepted 22 December 2008;
published online 25 January 2009
References
1. Tsuda, N., Nasu, K., Yanase, A. & Siratori, K. Electronic Conduction in Oxides
(Springer, 2000).
2. Imada, M., Fujimori, A. & Tokura, Y. Metal–insulator transitions.
Rev. Mod. Phys. 70, 1039–1263 (1998).
3. Spaldin, N. A. & Fiebig, M. The renaissance of magnetoelectric multiferroics.
Science 309, 391–392 (2005).
4. Eerenstein, W., Mathur, N. D. & Scott, J. F. Multiferroic and magnetoelectric
materials. Nature 442, 759–765 (2006).
5. Cheong, S.-W. & Mostovoy, M. Multiferroics: A magnetic twist for
ferroelectricity. Nature Mater. 6, 13–20 (2007).
6. Ramesh, R. & Spaldin, N. A. Multiferroics: Progress and prospects in thin films.
Nature Mater. 6, 21–29 (2007).
7. Roytburd, A. Elastic domains and polydomain phases in solids. Phase Transit.
45, 1–34 (1993).
8. Thomas, L. et al. Resonant amplification of magnetic domain-wall motion by a
train of current pulses. Science 315, 1553–1556 (2007).
233
ARTICLES
9. Přívratská, J. & Janovec, V. Pyromagnetic domain walls connecting
antiferromagnetic non-ferroelastic magnetoelectric domains. Ferroelectrics 204,
321–331 (1997).
10. Přívratská, J. & Janovec, V. Spontaneous polarization and/or magnetization
in non-ferroelastic domain walls: Symmetry predictions. Ferroelectrics 222,
23–32 (1999).
11. Goltsev, V. et al. Structure and interaction of antiferromagnetic domain walls
in hexagonal YMnO3 . Phys. Rev. Lett. 90, 177204 (2003).
12. Mostovoy, M. Ferroelectricity in spiral magnets. Phys. Rev. Lett. 96,
067601 (2006).
13. Aird, A. & Salje, E. K. H. Sheet superconductivity in twin walls: Experimental
evidence of WO3−x . J. Phys. Condens. Matter 10, L377–L380 (1998).
14. Bartels, M. et al. Impurity-induced resistivity of ferroelastic domain walls in
doped lead phosphate. J. Phys. Condens. Matter 15, 957–962 (2003).
15. Zubko, P. et al. Strain-gradient-induced polarization in SrTiO3 single crystals.
Phys. Rev. Lett. 99, 167601 (2007).
16. Kubel, F. & Schmid, H. Structure of a ferroelectric and ferroelastic monodomain
crystal of the perovskite BiFeO3 . Acta Crystallogr. B 46, 698–702 (1990).
17. Lebeugle, D. et al. Very large spontaneous electric polarization in BiFeO3
single crystals at room temperature and its evolution under cycling fields.
Appl. Phys. Lett. 91, 022907 (2007).
18. Streiffer, S. K. et al. Domain patterns in epitaxial rhombohedral ferroelectric
films. I. Geometry and experiments. J. Appl. Phys. 83, 2742–2753 (1998).
19. Zavaliche, F. et al. Multiferroic BiFeO3 films: Domain structure and
polarization dynamics. Phase Transit. 79, 991–1017 (2006).
20. Béa, H. et al. Influence of parasitic phases on the properties of BiFeO3 epitaxial
thin films. Appl. Phys. Lett. 87, 072508 (2005).
21. Cruz, M. P. et al. Strain control of domain-wall stability in epitaxial BiFeO3
(110) films. Phys. Rev. Lett. 99, 217601 (2007).
22. Meyer, B. & Vanderbilt, D. Ab initio study of ferroelectric domain walls in
PbTiO3 . Phys. Rev. B 65, 104111 (2002).
23. Stemmer, S., Streiffer, S. K., Ernst, F. & Rühle, M. Atomistic structure
of 90◦ domain walls in ferroelectric PbTiO3 thin-films. Phil. Mag. A 71,
713–724 (1995).
24. Floquet, N. & Valot, C. Ferroelectric domain walls in BaTiO3 : Structural
walls model interpreting fingerprints in XRPD diagrams. Ferroelectrics 234,
107–122 (1999).
25. Jia, C.-L. et al. Atomic-scale study of electric dipoles near charged and
uncharged domain walls in ferroelectric films. Nature Mater. 7, 57–61 (2008).
26. Wu, X. & Vanderbilt, D. Theory of hypothetical ferroelectric superlattices
incorporating head-to-head and tail-to-tail 180 degree domain walls.
Phys. Rev. B 73, 020103(R) (2006).
27. Allen, L. J., McBride, W., O’Leary, N. L. & Oxley, M. P. Exit wave reconstruction
at atomic resolution. Ultramicroscopy 100, 91–104 (2004).
234
NATURE MATERIALS DOI: 10.1038/NMAT2373
28. Ederer, C. & Spaldin, N. A. Weak ferromagnetism and magnetoelectric
coupling in bismuth ferrite. Phys. Rev. B 71, 060401(R) (2005).
29. Neaton, J. B. et al. First-principles study of spontaneous polarization in
multiferroic BiFeO3 . Phys. Rev. B 71, 014113 (2005).
30. Li, J. F. et al. Dramatically enhanced polarization in (001), (101), and (111)
BiFeO3 thin films due to epitiaxial-induced transitions. Appl. Phys. Lett. 84,
5261 (2004).
31. Ravindran, P. et al. Theoretical investigation of magnetoelectric behavior in
BiFeO3 . Phys. Rev. B 74, 224412 (2006).
32. Lebeugle, D. et al. Electric-field induced spin flop in BiFeO3 single crystals at
room temperature. Phys. Rev. Lett. 100, 227602 (2008).
33. Coene, W. M. J., Thust, A., Op de Beeck, M. & Van Dyck, D.
Maximum-likelihood method for focus-variation image reconstruction
in high resolution transmission electron microscopy. Ultramicroscopy 64,
109–135 (1996).
34. Thust, A., Coene, W. M. J., Op de Beeck, M. & Van Dyck, D. Focal-series
reconstruction in HRTEM: Simulation studies on non-periodic objects.
Ultramicroscopy 64, 211–230 (1996).
35. Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio
total-energy calculations using a plane-wave basis set. Phys. Rev. B 54,
11169–11186 (1996).
36. Anisimov, V. I., Aryasetiawan, F. & Liechtenstein, A. I. First-principles
calculations of the electronic structure and spectra of strongly correlated
systems: The LDA + U method. J. Phys. Condens. Matter 9, 767–808 (1997).
Acknowledgements
The work at Berkeley is supported by the Director, Office of Science, Office of Basic
Energy Sciences, Materials Sciences Division of the US Department of Energy under
contract No DE-AC02-05CH1123. The authors from Berkeley would like to acknowledge
the support of the National Center for Electron Microscopy, Lawrence Berkeley National
Laboratory. J.S. acknowledges support from the Alexander von Humboldt Foundation.
Y.H.C. would also like to acknowledge the support of the National Science Council,
R.O.C., under contract No NSC 97-3114-M-009-001. A.R. and S.G. acknowledge
support from Deutsche Forschungsgemeinschaft through FOR 520 and Deutsche
Akademische Austauschdienst through GE 1202/5-1, and N.A.S. acknowledges support
from NSF DMR Award No DMR-0605852 and the Miller Institute for Basic Research
in Science, UC Berkeley.
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to J.S.
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
 ARTICLES
PUBLISHED ONLINE: 15 FEBRUARY 2009 DOI: 10.1038/NMAT2378

The influence of edge structure on the electronic

properties of graphene quantum dots and
nanoribbons
Kyle A. Ritter1,2 * and Joseph W. Lyding1,3

Graphene shows promise as a future material for nanoelectronics owing to its compatibility with industry-standard lithographic
processing, electron mobilities up to 150 times greater than Si and a thermal conductivity twice that of diamond. The electronic
structure of graphene nanoribbons (GNRs) and quantum dots (GQDs) has been predicted to depend sensitively on the
crystallographic orientation of their edges; however, the influence of edge structure has not been verified experimentally. Here,
we use tunnelling spectroscopy to show that the electronic structure of GNRs and GQDs with 2–20 nm lateral dimensions varies
on the basis of the graphene edge lattice symmetry. Predominantly zigzag-edge GQDs with 7–8 nm average dimensions are
metallic owing to the presence of zigzag edge states. GNRs with a higher fraction of zigzag edges exhibit a smaller energy gap
than a predominantly armchair-edge ribbon of similar width, and the magnitudes of the measured GNR energy gaps agree with
recent theoretical calculations.

T
he surface structure of bulk, crystalline semiconductors has
profound consequences on the development and manufac-
turing of electronic devices. Bardeen1 explained that the pres-
ence of surface states results in binding of free carriers and induces
the formation of Schottky barriers at semiconductor–metal inter-
faces. A deeper understanding of surface states has enabled scientists
and engineers to optimize the performance of integrated circuits
for 50 years2 . Analogous to the surface states that exist in bulk
crystals, the edge structure of nanometre-sized, two-dimensional
materials such as graphene, a one-atom-thick layer of carbon, can
significantly influence their electronic structure.
Researchers have reported experimental transport
measurements3–8 and theoretical studies9–11 of graphene quan-
tum dots (GQDs) and nanoribbons (GNRs) elucidating their
remarkable promise for future nanoelectronic applications. In
spite of theoretical calculations that predict a localized metal-
lic state for the zigzag edge12 , all transport measurements of
GQDs (ref. 3) and GNRs (refs 4–7) reveal only semiconducting
behaviour. Furthermore, the electronic properties of the graphene
are independent of crystallographic orientation4 in contrast to
theoretical predictions9–12 . Recent theoretical studies show that
transport effects such as Coulomb blockade13 or a mobility
gap induced by edge disorder14,15 may affect the accuracy of
bandgaps measured under transport conditions and explain the
independence of energy gap and crystallographic orientation. By
probing the local electronic structure of GQDs and GNRs using
ultrahigh-vacuum scanning tunnelling microscopy (UHV-STM),
we detect that the crystallographic orientation of the graphene edges
significantly influences the electronic properties of nanometre-sized
graphene. Enabled by the development of an atomically clean,
in situ deposition method16 , we experimentally determine the
energy gap (Eg )–size (L) relation for GQDs with 2–20 nm lateral
dimensions and correlate the Eg measurements with the GQD
edge structure. Predominantly zigzag-edge GQDs with 7–8 nm
average dimensions are metallic and diverge from the Eg –L scaling
law owing to the presence of metallic zigzag edge states, which
spatially decay into the graphene interior with a 1.0–1.2 nm decay
length. In addition to GQDs, we study the electronic structure of
GNRs with 2–3 nm widths and 20–30 nm lengths. GNRs with a
higher fraction of zigzag edges exhibit a smaller energy gap than
a predominantly armchair-edge ribbon of similar width and the
magnitudes of the measured GNR energy gaps agree with recent
theoretical calculations.
Unlike previous studies of micrometre-sized, mechanically
exfoliated graphene monolayers on SiO2 (refs 17,18), nanometre-
sized graphene monolayer samples do not typically exhibit the
2.5 Å spaced, hexagonal graphene lattice in high-resolution STM
topographs. As shown in Fig. 1, we typically observe either a
4.1 Å hexagonal lattice (Fig. 1a–c) or 2.5 Å triangular lattice patterns
(Fig. 1d,e) in the STM topograph of the 3-Å-tall monolayer samples.
Owing to the close proximity of the graphene edges in our samples,
we believe that the electron wavefunction scatters off the edges
and the resultant interference pattern generates the observed STM
topographic contrasts for the GQDs and GNRs. Our √interpretation
√
is supported by previous STM studies where 3 × 3 R30◦
hexagonal superstructures were observed near graphite terrace
edges19–21 and graphene lattice defects22 , as well as the topographs
of 10–15-nm-wide GNRs defined on a highly oriented pyrolytic
graphite (HOPG) substrate through ambient STM lithography23 .
Figure 1a and b show the STM topograph and topographic
derivative, respectively, of a 2.3-nm-wide, 20-nm-long GNR.
Figure 1b delineates the hexagonal pattern along the GNR and
Fig. 1c shows a line contour taken from Fig. 1a, the location
of which is designated by the grey line. From Fig. 1c, the
nearest-neighbour spacing between the hexagons is 4.1 ± 0.3 Å.
The hexagonal pattern with a 4.1 Å average lattice spacing is

1 Beckman Institute for Advanced Science and Technology, University of Illinois, 405 North Mathews Avenue, Urbana, Illinois 61801-2325, USA,
2 Department of Materials Science and Engineering, University of Illinois, 1304 West Green Street, Urbana, Illinois 61801-2920, USA, 3 Department of
Electrical and Computer Engineering, University of Illinois, 1406 West Green Street, Urbana, Illinois 61801-2918, USA. *e-mail: ritter@engineering.uiuc.edu.

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 235

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT2378

a d

e
1.5

Height (Å)
c 1.0
Height (Å)

4.1 Å
0.5 0.5 2.5 Å

0 0
0 5 10 15 20 25
0 10 20 30 40 50
Position (Å)
Position (Å)

Figure 1 | Atomic-resolution imaging √ of GQDs and GNRs. a,b, STM topograph (a) and topographic derivative image (b) of a 2.3-nm-wide GNR exhibiting
a hexagonal superstructure with a 3ao (4.3 Å) lattice constant. The white hexagonal lattice is used to show the hexagonal superstructure. c, Topographic
cross-section taken from a along nearest-neighbour hexagons as indicated by the grey line. The distance between the centre of adjacent hexagons is
4.1 ± 0.3 Å, consistent with the lattice constant of the hexagonal superstructure. d, STM topograph and inset topographic derivative images of a GQD with
≈7 nm lateral dimensions exhibiting triangular lattice symmetry. A white hexagonal lattice is overlaid on the derivative image to highlight the orientation of
the C lattice. e, Line contour taken across the grey line in d showing the 2.5 Å spacing between nearest-neighbour hexagons. Imaging parameters: −0.5 V,
100 pA (a,b), −2 V, 50 pA (d). The scale bars in a and d represent 2 nm.

superstructure was independent of the biases between −0.5 and

−2 V. We also note that the hexagonal superlattice has been
0.8 Armchair observed for multiple GQDs and the superlattice is not limited
Zigzag
to the high-aspect-ratio geometry of graphene nanoribbons. In
Unresolved
contrast to the spatially uniform 4.1 Å periodicity observed in
Fit (without metals)
0.6 ambient STM studies of a 10-nm-wide GNR (ref. 23), we observe
Energy gap (eV)

that the periodicity of the hexagonal superstructure varies along

0.4
0.2
0 the triangular pattern observed for the bilayer samples16 , which is
2 4 6 8
Minimum lateral dimension (nm)
Figure 2 | Energy gap (Eg )–size (L) relation for GQDs. The GQDs are
labelled on the basis of whether the zigzag- or armchair-edge
crystallographic orientation is present in a higher fraction (>0.5). The 10
semiconducting samples are fitted to a power law, which is represented by
the solid black line of the form Eg (eV) = 1.57 ± 0.21 eV nm/L1.19±0.15 . The
two zigzag GQDs with 7–8 nm lateral dimensions are metallic and diverge
from the expected 0.15 eV energy gap. The error associated with the energy
gap is 0.1 eV, which is determined by thermal broadening, and the standard
deviation of the lateral size is calculated from averaging the minimum
lateral dimensions of the GQD samples.
√ hypothesize that the asymmetry in the local density of states of the C
consistent with the 4.3 Å ( 3ao ) hexagonal superstructure and
the 4.1 Å spacing measured for GNRs under ambient conditions.
Although the image in Fig. 1a is acquired at −0.5 V, the hexagonal
the axis of the 2.3-nm-wide GNR in Fig. 1a,b. As this phenomenon
was not observed under ambient conditions for graphite-supported
GNRs (ref. 23), we believe the fluctuations in the periodicity of the
superstructure could be due to underlying Si dangling bonds or
unpassivated C edge atoms from the in situ graphene deposition.
In addition to the hexagonal superstructure, we frequently
observe a triangular lattice for graphene monolayers, identical to
10 12
caused by the asymmetric stacking of the graphene layers. Figure 1d
shows an STM topograph of a 3.3-Å-tall GQD and the inset includes
a topographic derivative of the graphene to highlight the triangular
lattice symmetry. The grey line represents the location of the line
contour taken in Fig. 1e along the zigzag direction. The contour
exhibits a 2.4 ± 0.5 Å periodicity, which agrees with the expected
2.5 Å spacing between nearest-neighbour hexagons. It is unlikely
that the ≈3-Å-tall samples with a triangular lattice are bilayer pieces.
As shown in our previous work16 , monolayers are ≈3 Å tall at −2 V
whereas true bilayer pieces have heights of ≈6 Å under identical
scanning parameters on the same n-type Si(100)-2×1:H surface.
Ishigami et al.17 have observed a triangular lattice for
mechanically exfoliated graphene monolayers on SiO2 . They
atoms that generates the triangular pattern is caused by local charges
on the underlying substrate or curvature of the graphite film17 .
These theories do not explain the triangular patterns observed in

236 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2378 ARTICLES
a b c

d e g

h i

j k

l m n

Figure 3 | STM topographs of the GQDs included in the Eg –L plot in Fig. 2. The samples are labelled in order from the smallest (a) to the largest (n). For
the GQDs where the graphene lattice was resolved, the topographic spatial derivative image is included to highlight the atomic-scale contrast. On the basis
of the atomic-scale resolution images, we label the crystallographic orientation at the graphene edges as follows: armchair (red), zigzag (blue) and
unassigned (black). All of the samples were exfoliated from the graphite crystal under ambient conditions except d, j and k, which were exfoliated in UHV
and H-passivated. The scale bars represent 2 nm. Data from a, e and f were taken from ref. 16.

our experiments. We observe the triangular symmetry for GQDs
that conform to the atomically flat contours of the surface, unlike
the 1 nm height variations and 30-nm-long corrugations for rippled
monolayer films deposited on SiO2 substrates17,18 . Furthermore, the
triangular pattern has been observed on samples with heights of
≈3 Å that do not cross step edges, unlike the sample in Fig. 1d. In
addition, the pattern is extended across the entire graphene piece,
which suggests it is not caused by a local charge on the underlying
Si due to charged dangling bonds24 or underlying dopant atoms25 .
The experimental observations suggest that the source of the
triangular lattice symmetry is intrinsic to the electronic structure
of the graphene because the atomic lattice of the graphene and
the chemically inert Si(100)-2×1:H substrate are incommensurate.
Although we are uncertain of the origin of the triangular symmetry
in nanometre-sized monolayer samples, one potential source of the
triangular pattern is the edges, which could induce an interference
pattern. We believe detailed simulations of graphene monolayers
with different edge structures supported on the H–Si(100) surface

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 237

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT2378

a b
Zigzag Mixed

(1)
(1) Zigzag

Armchair (2)
(2)
Unassigned (3)
(3)

rzig = 0.65 rzig = 0.44

dl/dV (pA V¬1)
c 0.5 90 d 0.5

Sample bias (V)

Sample bias (V)

0 0

¬0.5 6 ¬0.5
0 10 20 30 40 50 60 0 10 20 30 40
Position (Å) Position (Å)

e 80 f 80

40 (1) 40 (1)
dl/dV at V = 0 V (pA V¬1)

dl/dV at V = 0 V (pA V¬1)

0 0
80 80

40 40 (2)
(2)
0 0
80 80

40 (3) 40 (3)
0 0
0 10 20 30 40 50 0 10 20 30 40
Position (Å) Position (Å)

Figure 4 | Comparison of a zigzag- and mixed-edge GQD using spatially resolved tunnelling spectroscopy. a, Predominantly zigzag-edge GQD from the
Eg –L plot in Fig. 2, which exhibits metallic character. The fraction of zigzag edges (rzig ) is 0.65. b, STM topograph of a 5 nm GQD with a mixture of both
zigzag and armchair edges (rzig = 0.44). Although zigzag edges are present, they are shorter than the sample in a. c, dI/dV − V spectra, obtained with
0.42 Å spatial resolution, plotted as a function of position across the green line in a. d, dI/dV − V spectra, obtained with 0.60 Å spatial resolution, plotted
as a function of position across the green line in b. e, Constant voltage, dI/dV versus position contours corresponding with the three numbered lines in a. In
general, we observe an increase in the differential conductance at the edges oriented along the zigzag direction. The spatial decay of the zigzag edge states
into the graphene interior prevents the observation of the expected 0.15 eV energy gap for this ≈8 nm sample. Line (2) is plotted along the solid green line
in a and does not include the dotted green line, which delineates the low conductance region at the left edge of the spectra map in c. f, Constant voltage,
dI/dV versus position contours recorded along the three lines in b. In contrast to the zigzag GQD, the differential conductivity does not increase near the
edges of the mixed-edge GQD and the magnitude of the differential conductivity is substantially lower than the zigzag GQD. The scale bars in a and b
represent 2 nm. STS setpoint: −2 V, 0.1 nA.

may provide insight into the origin of the triangular lattice in STM and black distinguishes edges that were unassigned. The samples
topographs of nanographene. shown in Fig. 3d,j,k were exfoliated in UHV and H-passivated (see
By using scanning tunnelling spectroscopy (STS), we measured the Methods section).
the size-dependent energy gap for 14 GQD samples. Each For the ten GQDs exhibiting an energy gap, the experimental
dI /dV (V /I ) − V spectrum used for the energy gap determination data were modelled with a power law, which resulted in
represents the average of 5–50 (depending on the particular sample) Eg (eV) = 1.57 ± 0.21 eV nm/L1.19±0.15 for the least-squares fit. The
individual I –V spectra from the interior of the GQDs. The length metallic samples were not included when modelling the exper-
scale for each sample was calculated by averaging the minimum imental Eg –L relation because the fit would be skewed owing
lateral dimension of the GQD and the results were compiled into to the limitations of room-temperature STS. The experimen-
the energy gap versus size plot shown in Fig. 2 (see Supplementary tal data are in close proximity to the predicted scaling trend
Information for more details). For the nine samples where the Eg (eV) = 1.68 eV nm/L (refs 26,27) resulting from quantum con-
graphene lattice was resolved, the samples are labelled armchair or finement and the linear dispersion of a graphene monolayer. We be-
zigzag on the basis of the edge symmetry present in the highest lieve that the reasonable correlation between simple theory and our
fraction (>0.5). Although the graphene was not atomically resolved experiment results from the nature of the STS measurement, which
at the edges, the crystallographic orientation of the edges was is not influenced by transport effects such as Coulomb blockade13 or
classified by determining the crystallographic orientation from the mobility gaps14,15 , and the enhanced spatial resolution of the STM.
atomic-scale contrast within the interior of the GQD (ref. 21). The Of the metallic GQDs in Fig. 2, two are predominantly zigzag
STM topographic images of each GQD composing the Eg –L plot edged (Fig. 3l,m) and the largest (Fig. 3n) has a 10 nm minimum
in Fig. 2 are shown in Fig. 3. The edge structure is represented by lateral dimension. The metallic GQD with 7 nm lateral dimensions
the coloured lines, where blue represents zigzag, red is armchair (Fig. 3j) and unresolved edge structure will be discussed below.

238 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2378 ARTICLES
a d
Zigzag edge Armchair edge

b e 100
1,000 Zigzag edge
0Å
0Å 80 2.5 Å
800 4.3 Å
dl/dV (pA V¬1)

dl/dV (pA V¬1)

5.0 Å
8.5 Å 60 7.6 Å
600 12.8 Å
40
400
20
200
0
0 ¬0.50 ¬0.25 0 0.25 0.50
¬0.50 ¬0.25 0 0.25 0.50 Sample bias (V)
Sample bias (V)

c f 103
1,000
Armchair edge
0Å
800

dl/dV (pA V¬1)

dl/dV (pA V¬1)

4.3 Å
Zigzag edge (b)
8.5 Å ξ = 1.17 ± 0.14 nm
600
12.8 Å
Zigzag edge (e)
102 ξ = 0.95 ± 0.11 nm
400

200

0 0 0.3 0.6 0.9 1.2 1.5

¬0.50 ¬0.25 0 0.25 0.50
Distance from edge (nm)
Sample bias (V)

Figure 5 | Detection of zigzag-edge state for GQDs. a, STM topograph of a ≈15 nm GQD with the edge lattice symmetry labelled. b,c, Comparison of
dI/dV spectra collected near the zigzag edge (b) and armchair edge (c) of the GQD sample. Whereas the zigzag edge in b exhibits a localized energy state
at −0.33 eV, a peak is not detected near the armchair-terminated edge in c. d,e, STM topograph (d) and dI/dV spectra (e) from the right zigzag edge of
the green line in Fig. 4a, which shows the spatial decay of the zigzag edge state at 0.29 eV into the interior of the graphene. The zigzag edge state is located
at 0.29 eV compared with the −0.33 eV position for the zigzag edge state of the sample in b. The energetic shift in the zigzag edge state peak is probably
due to charge transfer from the different chemical functional groups bound to the graphene edges of the respective samples. f, dI/dV zigzag-state peak
height plotted as a function of position on a semi-logarithmic scale for the samples in b and e. The decay length (ξ ) for the edge states is extracted from
the slope of the line. For f, the error of the length was determined by the spatial resolution of the measurement. The dI/dV standard deviation was
calculated from individual spectra used to generate the spatially averaged spectra in b and e. The scale bars in a and d represent 2 nm.

For the largest GQD (Fig. 3n), we believe that the reduced energy
gap induced by quantum confinement is not resolved owing to the
60.1 eV constraints of our room-temperature STS measurements.
The two zigzag-edged GQDs diverge from the 0.15 eV expected
energy gaps based on the experimental fit in Fig. 2. The metallic
character of the zigzag-edged, 7–8 nm samples suggests that the
fraction of zigzag edges influences the measured energy gap.
From Fig. 4, we gain more insight into the effect of
zigzag-terminated edges and the measured energy gap by compar-
ing the spatially resolved, differential conductivity of a metallic,
zigzag-edge GQD (rzig = 0.65) and a semiconducting GQD with
a lower fraction of zigzag edges (rzig = 0.44). To quantify the
relative concentration of zigzag edges, rzig (ref. 28) represents the
ratio of zigzag edges to the total length of resolved zigzag and
armchair edges, excluding edges that were unassigned. Figure 4a
and b show the STM topographs of a metallic, zigzag GQD and
a mixed-edge GQD, respectively. The longest, continuous zigzag
edge for the monolayer in Fig. 4a is 4.7 nm, whereas the maximum
zigzag edge length for the mixed-edge sample is 1.8 nm. Figure 4c,d
shows spatially resolved maps of the local graphene electronic
structure where the dI /dV − V tunnelling spectra are plotted
as a function of distance across the green lines of the respective
samples. The colour scale represents the value of the differential
conductance for Fig. 4c,d.
The dI /dV spectra maps for the GQDs in Fig. 4a,b exhibit
pronounced differences. In Fig. 4c, the right end of the spectra map
shows higher conductance with a localized state at 0.29 V near the
pristine edge terminated along the zigzag direction. The left end is
also terminated along the zigzag direction, but there is a reduced
conductance near the 3 Å topographic protrusion. Although the
identity of the topographic protrusion cannot be determined with
certainty, we speculate the protrusion may be a cluster of C atoms
from the exfoliation process on the basis of its height, reduced local
density of states and spatial location 2 nm from the edge. However,
we cannot eliminate the possibility that localized protrusions at
the edge result from ambient functionalization of the edges during
the ex situ exfoliation of graphite. Although the sample crosses a
monatomic Si step edge, multiple 0.42 Å spaced, spatially resolved
tunnelling spectra maps across the GQD reveal that the graphene
electronic structure does not vary as the sheet smoothly traverses
the Si terrace edge.
In Fig. 4d, the differential conductance spatial map for the
mixed-edge GQD exhibits a uniform 0.3–0.4 eV energy gap across
the entire piece even at the edges. The general features of the

NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials 239

ARTICLES NATURE MATERIALS DOI: 10.1038/NMAT2378

a
Armchair
Zigzag b
Unassigned

b c 2
Armchair (rzig = 0.15)
6

dl/dV (V/l) (arb. units)

dl/dV (V/l) (arb. units)

Eg = 0.38 eV

4
Wavg = 2.9 ± 0.3 nm 1

0 0
¬1.0 ¬0.5 0 0.5 1.0 ¬0.4 ¬0.2 0 0.2 0.4
Sample bias (V) Sample bias (V)

d 10 e 2
Zigzag (rzig = 0.95)
dl/dV (V/l) (arb. units)
dl/dV (V/l) (arb. units)

8 Eg = 0.14 eV

6
Wavg = 2.3 ± 0.3 nm
4

2
0 0
¬1.0 ¬0.5 0 0.5 1.0 ¬0.2 ¬0.1 0 0.1 0.2
Sample bias (V) Sample bias (V)

f g
10 Zigzag (rzig = 0.67)
dl/dV (V/l) (arb. units)

dl/dV (V/l) (arb. units)

8 Eg = 0.12 eV
Wavg = 3.3 ± 0.3 nm 1
6

2
0 0
¬1.0 ¬0.5 0 0.5 1.0 ¬0.2 ¬0.1 0 0.1 0.2
Sample bias (V) Sample bias (V)

Figure 6 | Tunnelling spectroscopy of three 20–30-nm-long, 2–3-nm-wide GNRs. a, STM topograph of the three GNRs with the edge symmetries
labelled. b–g, Normalized dI/dV spectra for a predominantly armchair (b–c) and two zigzag (d,e;f,g) GNRs presented in a. The insets of b, d and f show the
STM topograph of the nanoribbon that corresponds with the tunnelling spectroscopy measurements. The dotted regions in b, d and f correspond with the
higher resolution dI/dV(V/I) − V spectra in c, e and g. A higher concentration of zigzag edges tends to decrease the energy gap because the predominantly
zigzag-edged, 2.3-nm-wide nanoribbon exhibits a smaller energy gap (0.14 eV) than the 2.9-nm-wide armchair nanoribbon (0.38 eV). The scale bar in a
represents 2 nm.

spectra maps in Fig. 4c and d were reproducible across the surface
of the respective sheets as shown in Fig. 4e and f, which correspond
with the lines labelled (1)–(3) in Fig. 4a and b, respectively.
Figure 4e,f shows the value of the differential conductance at
0 V as a function of position across both GQDs. Line (2) in
Fig. 4e is recorded along the solid green line in Fig. 4a and does
not include the region labelled by the dotted green line, which
represents the low conductance region at the left end of the
spectra map shown in Fig. 4c. For the zigzag GQD in Fig. 4e,
we observe an increase in the differential conductance within
1–2 nm from the edge of the graphene. For the mixed-edge
GQD, however, the differential conductance at 0 V does not
increase near the graphene edges and the zero-bias, differential
conductance is significantly lower compared with the zigzag
GQD. The increased conductance near the edges in Fig. 4c,e is
suggestive of the localized zigzag edge state, which has been

240 NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials

NATURE MATERIALS DOI: 10.1038/NMAT2378
theoretically predicted12,28 and experimentally measured at graphite
terrace edges using tunnelling spectroscopy20,21,29 . To confirm the
reproducibility of the localized zigzag edge state, we compare the
STS recorded for a spatially separated zigzag and armchair edge for
an extra GQD in Fig. 5.
In Fig. 5b, we detect a localized state at −0.33 eV at the zigzag
edge of the ≈15 nm GQD shown in Fig. 5a. We do not observe
a localized state near the well-defined armchair edge on the same
GQD sample (Fig. 5c), which confirms that the localized state
observed near the zigzag edge is the predicted edge state due to
the termination of π bonding12,28 . Figure 5e shows the averaged
dI /dV − V spectra near the zigzag edge of the sample studied in
Fig. 4a. The amplitude of the localized state at 0.29 eV is maximized
at the zigzag edge and spatially decays into the interior of the
GQD. The position of the zigzag energy state is shifted to 0.29 eV
compared with the −0.33 eV peak position for the zigzag edge state
for the sample in Fig. 5b. We attribute the energetic shift to the
particular functional groups present at the edges of the respective
samples that could shift the peak position from its expected position
of ≈0 eV (refs 12,28). The zigzag state in previous STS studies
of graphite ranged from 20 to 100 meV below the Fermi level20,21
or 90–250 meV above the Fermi level29 . The graphite in previous
studies was prepared using different methods including UHV
H-passivation20 , ambient cleaving21 and hydrogen-etched graphite
studied with an ambient STM (ref. 29), which further support our
interpretation that the position of the zigzag energy state fluctuates
depending on the specific functional groups at the edges.
By plotting the differential conductivity as a function of position
in Fig. 5f, we can extract the decay length (ξ ) for the localized
edge state measured for the respective GQD samples. The 1.17
and 0.95 nm values for the GQDs in Fig. 5b,e, respectively, are
in agreement with the 1.0–1.6 nm average range of decay lengths
measured for zigzag edges on graphite21 . The centre of the graphene
in Fig. 4a is ≈2.5 decay lengths from the zigzag edge. Consequently,
the metallic behaviour at the interior of the zigzag GQD (evident
in Fig. 2) can be explained by the decay of the zigzag edge state.
We believe that the metallic behaviour of the sample in Fig. 3j is
most likely caused by the presence of zigzag edge states. However,
we are unable to confirm this hypothesis as we did not obtain
atomic-resolution imaging or 62 nm spatially resolved STS for this
sample. Interestingly, the 2-nm-wide zigzag GQD in Figs 2 and
3b exhibits an energy gap in agreement with recent theoretical
calculations of GQDs that show that an energy gap exists for
hexagonally shaped, zigzag GQDs if the length of the hexagon
edge is 63 nm (ref. 30).
In addition to GQDs where the geometry consists of an aspect
ratio of ≈1, we have also measured tunnelling spectra for GNRs
with 2–3 nm widths and lengths of 20–30 nm. The fraction of
zigzag edges (rzig ) is calculated for each GNR and the results are
shown in Fig. 6c,e,g. The 0.38 eV energy gap measured for the
armchair nanoribbon in Fig. 6c lies within the 0.1–0.7 eV range of
energy gaps predicted by theoretical calculations of 2.9-nm-wide,
armchair GNRs (refs 9,10). Analogous to our results with the GQDs,
we observe that a higher concentration of zigzag edges tends to
decrease the energy gap for the GNRs—the predominantly zigzag-
edged, 2.3-nm-wide nanoribbon exhibits a smaller energy gap
(0.14 eV) compared with the 2.9-nm-wide armchair nanoribbon
(0.38 eV). However, unlike the 7–8 nm, metallic-zigzag GQDs,
the predominantly zigzag-edge GNRs exhibit a finite energy gap.
Although tight-binding calculations predict zigzag GNRs to be
metallic12,28 , density functional theory predicts the opening of
a small energy gap for zigzag GNRs owing to the staggered
sublattice potential that develops along the zigzag-terminated
edges9 . We find agreement between the measured energy gaps
for the two predominantly zigzag GNRs in our experiments and
the first-principles calculations of zigzag GNRs that predict the
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials
ARTICLES
Eg − W relation to be Eg = (0.933/(W (nm) + 1.5)) (ref. 9). On
the basis of the experimentally determined widths, the 2.3-nm-
(Fig. 6e) and 3.3-nm-wide (Fig. 6g), zigzag GNRs are expected
to have energy gaps of 0.25 and 0.19 eV, respectively, which
compare favourably with the slightly smaller 0.14 ± 0.1 and
0.12 ± 0.1 eV measured values.
Our experimental results suggest that the crystallographic
orientation in nanographene significantly alters the overall elec-
tronic properties of the graphene. By using tunnelling spectroscopy,
we have circumvented the lateral transport effects that have pre-
vented the observation of the influence of crystallographic ori-
entation in transport measurements of GQDs and GNRs. On
the basis of our results, controlled engineering of the graphene
edge structure23,31 will probably be required for obtaining uniform
performance among graphene-based nanoelectronic devices.
Methods
Our experiments were conducted using a homebuilt, room-temperature UHV
scanning tunnelling microscope at a base pressure of 3 × 10−11 torr (ref. 32). In
our experimental set-up, the bias voltage is applied to the sample and the tip is
grounded through a current preamplifier. Topographic images were collected using
current setpoint values ranging from 50 to 100 pA and the sample was typically
biased at −1.5 to −2 V. In addition to STM topographic images, STS can be used
to probe the local density of states of the sample. STS consists of acquiring a
tunnelling current–voltage (I –V ) spectrum by momentarily disabling the feedback
at predefined points in the topographic image and sweeping the voltage over a
specified range while recording the current. Both constant- and variable-spacing33
tunnelling spectroscopy were used during the experiments (see Supplementary
Information). All of the data were obtained using electrochemically etched
tungsten tips. Experiments were carried out on degenerately As-doped n-type
and B-doped p-type Si(100)-2 × 1:H surfaces16 with a resistivity <0.005  cm.
Graphene monolayers were isolated from the HOPG source substrate (Advanced
Ceramics and Mikromasch, ZYA Grade) ex situ by mechanical exfoliation16
and subsequently deposited onto the Si substrate in situ using the dry contact
transfer (DCT) technique34 .
To control the chemical functionalization of the graphene edges, we
have also modified the DCT method for exfoliating nanometre graphene in
UHV and H-passivating the graphene edges. For the in situ exfoliation, a
ZYA-grade HOPG sample was cleaved immediately before transferring it to
the UHV system and the sample was degassed for 24 h at 600 ◦ C to remove
physisorbed molecules. After degassing the HOPG sample, we used STM to
characterize the HOPG surface before in situ exfoliation. The average terrace
size of the HOPG was 100 nm, which suggests that any functionalized edges
present on the HOPG would be negligible for the mechanically exfoliated,
nanometre graphene sheets.
To generate samples of H-passivated nanographene, the HOPG sample
was pressed into contact with a clean, degassed fibre-glass applicator with the
chamber pressure at 1 × 10−10 torr. The HOPG and fibre glass were manipulated
into contact until a visible grey powder was generated on the surface of the
fibre-glass DCT applicator. After mechanically exfoliating the graphene from
the graphite crystal, the DCT applicator was immediately transferred to the
preparation chamber and aligned with a W filament. The chamber was backfilled to
2×10−6 torr H2 and atomic H was generated by heating the W filament to 1,400 ◦ C.
The graphene-loaded DCT applicator was exposed to atomic H for 30 min to ensure
that the dangling C bonds generated in situ were passivated with H.
The STM topographs of the H-passivated monolayer sheets are shown in
Fig. 3d,j,k. The H-passivation of the GQDs results in a lower density of protrusions
along the graphene edges in STM topographs. The decreased density of protrusions
for the UHV, H-passivated GQDs suggests that the topographic protrusions
are primarily due to ambient molecules chemisorbing to the graphene edges.
From our limited data set of three H-passivated GQDs, we do not observe
significant deviations from the Eg –L scaling trend of the samples exfoliated
under ambient conditions.
Received 18 August 2008; accepted 6 January 2009;
published online 15 February 2009
References
1. Bardeen, J. Surface states and rectification at a metal–semiconductor contact.
Phys. Rev. 71, 717–727 (1947).
2. Kilby, J. S. Miniaturized electronic circuit. US Patent No. 3,138,743 (Issued
June 23, 1964; Filed Feb. 6, 1959).
3. Ponomarenko, L. A. et al. Chaotic Dirac billiard in graphene quantum dots.
Science 320, 356–358 (2008).
4. Han, M. Y., Özyilmaz, B., Zhang, Y. & Kim, P. Energy band gap engineering of
graphene nanoribbons. Phys. Rev. Lett. 98, 206805 (2007).
241
ARTICLES
5. Chen, Z., Lin, Y., Rooks, M. J. & Avouris, P. Graphene-nanoribbon electronics.
Physica E 40, 228–232 (2007).
6. Li, X., Wang, X., Zhang, L., Lee, S. & Dai, H. Chemically derived, ultrasmooth
graphene nanoribbon semiconductors. Science 319, 1229–1232 (2008).
7. Lin, Y.-M., Perebeinos, V., Chen, Z. & Avouris, P. Conductance quantization in
graphene nanoribbons. Preprint at <http://arxiv.org/abs/0805.0035> (2008).
8. Adam, S., Cho, S., Fuhrer, M. S. & Das Sarma, S. Density inhomogeneity driven
percolation metal–insulator transition and dimensional crossover in graphene
nanoribbons. Phys. Rev. Lett. 101, 046404 (2008).
9. Son, Y.-W., Cohen, M. L. & Louie, S. G. Energy gaps in graphene nanoribbons.
Phys. Rev. Lett. 97, 216803 (2006).
10. Barone, V., Hod, O. & Scuseria, G. E. Electronic structure and stability of
semiconducting graphene nanoribbons. Nano Lett. 6, 2748–2754 (2006).
11. Son, Y.-W., Cohen, M. M. & Louie, S. G. Half-metallic graphene nanoribbons.
Nature 444, 347–349 (2006).
12. Fujita, M., Wakabayashi, K., Nakada, K. & Kusakabe, K. Peculiar localized state
at zigzag graphite edge. J. Phys. Soc. Jpn. 65, 1920–1923 (1996).
13. Sols, F., Guinea, F. & Castro Neto, A. H. Coulomb blockade in graphene
nanoribbons. Phys. Rev. Lett. 99, 166803 (2007).
14. Querlioz, D. et al. Suppression of the orientation effects on bandgap in
graphene nanoribbons in the presence of edge disorder. Appl. Phys. Lett. 92,
42108 (2008).
15. Gunlycke, D., Areshkin, D. A. & White, C. T. Semiconducting graphene
nanostrips with edge disorder. Appl. Phys. Lett. 90, 142104 (2007).
16. Ritter, K. A. & Lyding, J. W. Characterization of nanometer-sized, mechanically
exfoliated graphene on the H-passivated Si(100) surface using scanning
tunneling microscopy. Nanotechnology 19, 015704 (2008).
17. Ishigami, M., Chen, J. H., Cullen, W. G., Fuhrer, M. S. & Williams, E. D.
Atomic structure of graphene on SiO2 . Nano Lett. 7, 1643–1648 (2007).
18. Stolyarova, E. et al. High-resolution scanning tunneling microscopy imaging
of mesoscopic graphene sheets on an insulating surface. Proc. Natl Acad. Sci.
104, 9209–9212 (2007).
19. Kobayashi, Y., Fukui, K., Enoki, T. & Kusakabe, K. Edge state on
hydrogen-terminated graphite edges investigated by scanning tunneling
microscopy. Phys. Rev. B 73, 125415 (2006).
20. Kobayashi, Y., Fukui, K., Enoki, T., Kusakabe, K. & Kaburagi, Y. Observation
of zigzag and armchair edges of graphite using scanning tunneling microscopy
and spectroscopy. Phys. Rev. B 71, 193406 (2005).
21. Nimi, Y. et al. Scanning tunneling microscopy and spectroscopy of the
electronic local density of states of graphite surfaces near monoatomic step
edges. Phys. Rev. B 73, 085421 (2006).
22. Rutter, G. M. et al. Scattering and interference in epitaxial graphene. Science
317, 219–222 (2007).
23. Tapasztó, L., Dobrik, G., Lambin, P. & Biró, L. P. Tailoring the atomic structure
of graphene nanoribbons using scanning tunneling microscope lithography.
Nature Nanotech. 3, 397–401 (2008).
242
NATURE MATERIALS DOI: 10.1038/NMAT2378
24. Piva, P. G. et al. Field regulation of single-molecule conductivity by a charged
surface atom. Nature 435, 658–661 (2005).
25. Liu, L., Yu, J. & Lyding, J. W. Atom-resolved three-dimensional mapping of
boron dopants in Si(100) by scanning tunneling microscopy. Appl. Phys. Lett.
78, 386–388 (2001).
26. Berger, C. et al. Electronic confinement and coherence in patterned epitaxial
graphene. Science 312, 1191–1196 (2006).
27. Geim, A. K. & Novoselov, K. S. The rise of graphene. Nature Mater. 6,
183–191 (2007).
28. Nakada, K., Fujita, M., Dresselhaus, G. & Dresselhaus, M. S. Edge state
in graphene ribbons: Nanometer size effect and edge shape dependence.
Phys. Rev. B. 54, 17954–17960 (1996).
29. Klusek, Z. et al. Observation of local electron states on the edges of circular
pits on hydrogen-etched graphite surface by scanning tunneling spectroscopy.
Appl. Surf. Sci. 161, 508–514 (2000).
30. Zhang, Z. Z., Chang, K. & Peeters, F. M. Tuning of energy levels and optical
properties of graphene quantum dots. Phys. Rev. B 77, 235411 (2008).
31. Datta, S. S., Strachan, D. R., Khamis, S. M. & Johnson, A. T. C. Crystallographic
etching of few-layer graphene. Nano Lett. 8, 1912–1915 (2008).
32. Lyding, J. W., Skala, S., Hubacek, J. S., Brockenbrough, R. & Gammie, G.
Variable-temperature scanning tunneling microscope. Rev. Sci. Instrum. 59,
1897–1902 (1988).
33. Feenstra, R. M. Tunneling spectroscopy of the (110) surface of direct-gap III–V
semiconductors. Phys. Rev. B 50, 4561–4570 (1994).
34. Albrecht, P. M. & Lyding, J. W. Ultrahigh-vacuum scanning tunneling
microscopy and spectroscopy of single-walled carbon nanotubes on
hydrogen-passivated Si(100) surfaces. Appl. Phys. Lett. 83, 5029–5031 (2003).
Acknowledgements
This work was supported by the Office of Naval Research under grant number
N000140610120 and by the National Science Foundation grant number NSF ECS
04-03489. K.A.R. acknowledges support from a NDSEG fellowship. We thank J. Koepke
for assistance with a portion of the data collection, L. Ruppalt for providing the code for
the normalized dI /dV calculations and P. Albrecht, P. Dollfus, D. Querlioz, A. Rockett,
M. Sztelle and J. Weaver for helpful discussions.
Author contributions
K.A.R and J.W.L. conceived the experiments. K.A.R. carried out the experiments,
analysed the data and wrote the manuscript. J.W.L. provided technical support for the
instrumentation, discussed the data and commented on the manuscript.
Additional information
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions. Correspondence and requests for materials should be
addressed to K.A.R.
NATURE MATERIALS | VOL 8 | MARCH 2009 | www.nature.com/naturematerials