This paper is published as part of a PCCP Themed Issue on: Metal oxide nanostructures: synthesis, properties and applications

Guest Editors: Nicola Pinna, Markus Niederberger, John Martin Gregg and Jean-Francois Hochepied

Editorial

Chemistry and physics of metal oxide nanostructures Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b905768d

Papers

Thermally stable ordered mesoporous CeO2/TiO2 visible-light photocatalysts Guisheng Li, Dieqing Zhang and Jimmy C. Yu, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b819167k

Blue nano titania made in diffusion flames Alexandra Teleki and Sotiris E. Pratsinis, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821590a

Shape control of iron oxide nanoparticles Alexey Shavel and Luis M. Liz-Marzán, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b822733k

Colloidal semiconductor/magnetic heterostructures based on iron-oxide-functionalized brookite TiO2 nanorods Raffaella Buonsanti, Etienne Snoeck, Cinzia Giannini, Fabia Gozzo, Mar Garcia-Hernandez, Miguel Angel Garcia, Roberto Cingolani and Pantaleo Davide Cozzoli, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821964h

Low-temperature ZnO atomic layer deposition on biotemplates: flexible photocatalytic ZnO structures from eggshell membranes Seung-Mo Lee, Gregor Grass, Gyeong-Man Kim, Christian Dresbach, Lianbing Zhang, Ulrich Gösele and Mato Knez, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b820436e

A LEEM/ -LEED investigation of phase transformations in TiOx/Pt(111) ultrathin films Stefano Agnoli, T. Onur Mente , Miguel A. Niño, Andrea Locatelli and Gaetano Granozzi, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821339a

Synthesis and characterization of V2O3 nanorods Alexander C. Santulli, Wenqian Xu, John B. Parise, Liusuo Wu, M.C. Aronson, Fen Zhang, Chang-Yong Nam, Charles T. Black, Amanda L. Tiano and Stanislaus S. Wong, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b822902c

Flame spray-pyrolyzed vanadium oxide nanoparticles for lithium battery cathodes See-How Ng, Timothy J. Patey, Robert Büchel, Frank Krumeich, Jia-Zhao Wang, Hua-Kun Liu, Sotiris E. Pratsinis and Petr Novák, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821389p

Mesoporous sandwiches: towards mesoporous multilayer films of crystalline metal oxides Rainer Ostermann, Sébastien Sallard and Bernd M. Smarsly, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b820651c

Surprisingly high, bulk liquid-like mobility of silica-confined ionic liquids Ronald Göbel, Peter Hesemann, Jens Weber, Eléonore Möller, Alwin Friedrich, Sabine Beuermann and Andreas Taubert, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821833a

Fabrication of highly ordered, macroporous Na2W4O13 arrays by spray pyrolysis using polystyrene colloidal crystals as templates SunHyung Lee, Katsuya Teshima, Maki Fujisawa, Syuji Fujii, Morinobu Endo and Shuji Oishi, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821209k

Nanoporous Ni–Ce0.8Gd0.2O1.9-x thin film cermet SOFC anodes prepared by pulsed laser deposition Anna Infortuna, Ashley S. Harvey, Ulrich P. Muecke and Ludwig J. Gauckler, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821473e

Surface chemistry of carbon-templated mesoporous aluminas Thomas Onfroy, Wen-Cui Li, Ferdi Schüth and Helmut Knözinger, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821505g

ZnO@Co hybrid nanotube arrays growth from electrochemical deposition: structural, optical, photocatalytic and magnetic properties Li-Yuan Fan and Shu-Hong Yu, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b823379a

Electrochemistry of LiMn2O4 nanoparticles made by flame spray pyrolysis T. J. Patey, R. Büchel, M. Nakayama and P. Novák, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821572n

Ligand dynamics on the surface of zirconium oxo clusters Philip Walther, Michael Puchberger, F. Rene Kogler, Karlheinz Schwarz and Ulrich Schubert, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b820731c

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online / Journal Homepage / Table of Contents for this issue

Thin-walled Er3+:Y2O3 nanotubes showing up-converted fluorescence Christoph Erk, Sofia Martin Caba, Holger Lange, Stefan Werner, Christian Thomsen, Martin Steinhart, Andreas Berger and Sabine Schlecht, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821304f

Wettability conversion of colloidal TiO2 nanocrystal thin films with UV-switchable hydrophilicity Gianvito Caputo, Roberto Cingolani, Pantaleo Davide Cozzoli and Athanassia Athanassiou, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b823331d

Nucleation and growth of atomic layer deposition of HfO2 gate dielectric layers on silicon oxide: a multiscale modelling investigation A. Dkhissi, G. Mazaleyrat, A. Estève and M. Djafari Rouhani, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821502b

Designing meso- and macropore architectures in hybrid organic–inorganic membranes by combining surfactant and breath figure templating (BFT) Ozlem Sel, Christel Laberty-Robert, Thierry Azais and Clément

Sanchez, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821506e

The controlled deposition of metal oxides onto carbon nanotubes by atomic layer deposition: examples and a case study on the application of V2O4 coated nanotubes in gas sensing Marc-Georg Willinger, Giovanni Neri, Anna Bonavita, Giuseppe Micali, Erwan Rauwel, Tobias Herntrich and Nicola Pinna, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821555c

In situ investigation of molecular kinetics and particle formation of water-dispersible titania nanocrystals G. Garnweitner and C. Grote, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b821973g

Chemoresistive sensing of light alkanes with SnO2 nanocrystals: a DFT-based insight Mauro Epifani, J. Daniel Prades, Elisabetta Comini, Albert Cirera, Pietro Siciliano, Guido Faglia and Joan R. Morante, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b820665a

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

Nucleation and growth of atomic layer deposition of HfO2 gate dielectric

layers on silicon oxide: a multiscale modelling investigation

A. Dkhissi,*ab G. Mazaleyrat,ab A. Esteveab and M. Djafari Rouhaniab

Received 1st December 2008, Accepted 4th March 2009

First published as an Advance Article on the web 26th March 2009

DOI: 10.1039/b821502b

We apply our recently developed approach, combining advanced ab initio density functional

theory (DFT) methods with a probabilistic kinetic Monte Carlo (KMC) scheme, to quantify the

properties of mesoscopic size systems operating in real experimental conditions. The application

concerns the investigation of the atomic layer deposition (ALD) of HfO2 film growth on Si(100)

surface. We show that the proposed models offer guidance in the optimization of the experimental

deposition processes, in terms of OH density on the substrate, optimal growth temperature, pulse

durations, and finally growth kinetics.

I. Introduction

The reduction of the SiO2 gate oxide thickness in accordance

with Moore’s law has led to unacceptable tunnelling and

leakage current levels.1 So, the conventional SiO2 gate is

reaching its physical and electrical limitations. The search

for suitable candidates to replace SiO2 as the gate dielectric

in complementary metal-oxide-semiconductor (CMOS)

devices has recently received enormous attention. Of the many

different materials with high dielectric constant k, which have

been used as SiO2 replacements,2 hafnium oxide (HfO2) is one

of the most promising candidates. HfO2 is attractive as it

exhibits a bulk permittivity of almost 25, a wide band gap

(5.68 eV) and a good thermodynamic stability in contact with

silicon.3 However, experimentally the EOT (equivalent oxide

thickness) value of the HfO2 gate stack increases, due to the

growth of an interfacial layer between HfO2 and the silicon

substrate. Among various methods for growing high-k

dielectric films, atomic layer deposition (ALD) shows its

unique ability in depositing ultra thin films with excellent

conformity and uniformity over large areas.4 ALD, which is

a vapour deposition technique, is based on the cycling of

self-terminating surface reactions. Indeed, each precursor is

pulsed into the reaction chamber alternately, and the reaction

between the incoming precursors and surface species is self-

terminating. Thus atomic level control of film growth is

supposed to be achieved. Experimentally, ALD has been

actively investigated for depositing HfO25–21 for which HfCl4

and H2O are often used as precursors.22–24 In practice, the

density of metal atoms deposited per cycle depends on the

temperature, on the chemical nature of the precursors used

and on the density of reactive sites available on the substrate

surface. Thus, it is desired to reach full monolayer coverage

at each cycle. However, Ritala et al.22 found that only a

sub-monolayer of the HfO2 film is deposited during each cycle.

About seven deposition cycles would be needed to entirely

cover the surface25 which then shows a non-negligible

roughness. ALD growth resulting in only partial monolayer

growth per cycle is due to the steric hindrance26,27 or to the

lack of reactive sites.28 Therefore, a detailed understanding of

the basic mechanisms that take place during each cycle is

required if one wants to optimize the deposition. This is

especially true for the few first deposited layers which are of

major importance for subsequent electrical properties.

These basic atomistic mechanisms include the decomposition

of HfCl4 on the substrate and the further hydrolysis of the

resulting SiO2–O–HfCl3 surface complex, but many other reac-

tions, called ‘‘densification mechanisms’’ hereafter, are also

involved in the process. To design optimal experimental setups

in future, it is of great importance that a precise atomic level

description of the basic reaction mechanisms responsible for the

overall process of high-k film growth, and also their relation

with the thermodynamic parameters be established. Our final

aim, towards this objective, is to achieve the ambitious task of

elaborating a new generation of tools dedicated to atomic scale

simulation of technological process. Our strategy is to combine

the ab initio DFT with kinetic Monte-Carlo techniques into a

multiscale approach, which will incorporate enough reaction

mechanisms to be adequate for testing against experimental

setups. Kinetic Monte Carlo (KMC) is a suitable procedure

that can be of great help to understand mesoscopic level

processes, such as deposition of high-k materials.

This paper is dedicated to the application of our KMC tool,

enabling the treatment of HfO2 ALD from HfCl4 and H2O

precursors at the atomic scale. In particular, we present the

results concerning the following points (i) the influence of the

substrate preparation on the coverage of the surface, (ii) an

optimal growth temperature, since it is found experimentally

that HfCl4/H2O ALD of HfO2 has a broad process window

with respect to deposition temperature, between 180 and

600 1C, (iii) surface coverage saturation and (iv) growth

kinetics in different regimes.

Section II is devoted to the description of DFT and KMC

theoretical approaches. Different basic mechanisms used in

KMC simulations are also briefly listed in this section, with an

emphasis on densification mechanisms. Section III gathers

a CNRS, LAAS, 7 avenue du colonel Roche, F-31077, Toulouse,France. E-mail: adkissi@laas.fr; Fax: +33 5 61 33 62 08

bUniversite de Toulouse, UPS, INSA, INP, ISAE, LAAS,F-31077, Toulouse, France

This journal is �c the Owner Societies 2009 Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 | 3701

PAPER www.rsc.org/pccp | Physical Chemistry Chemical Physics

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

ab initio results, implemented as parameters in the KMC

package, and KMC simulation results, corresponding to the

first cycle deposition and to the transient and steady-state

growth regimes. Finally, conclusions are drawn in section IV.

II. Theoretical approaches

Recently we have developed an original approach within a

multiscale strategy29 to simulate the growth of HfO2 on

silicon. In particular, we have shown how this method

provides a unique and fundamental understanding of the

growth mechanism and growth kinetics. KMC is a stochastic-

based model aimed at simulating film growth at the atomic

scale, the final objective being to furnish alternative/new tools

for replacement of the macroscopic conventional TCAD

(Technology Computer Aided Design) tools used for years

by engineers in microelectronics. Such a model performs

virtually the explicit experimental-processing procedure,

atomic-scale event by atomic-scale event. The algorithm is

operated through probabilistic rules that make the overall

simulation comparable with an actual experiment.30,31 Indeed,

from the engineering point of view, the main interest is in the

microstructure that is produced under specific processing

conditions. This microstructure involves millions of atoms

processed during periods exceeding the second. Unfortunately,

the most predictive models, i.e. the quantum-based models,

are not tractable at this scale. Two options can be further

considered: molecular dynamics (MD) and kinetic Monte

Carlo (KMC). MD solves the Newton equations of motion

for the system of atoms under investigation. Trajectories are

continuous and necessitate very short integration time steps

that make the simulation cumbersome for durations that are

orders of magnitude higher than nanosecond. Another

drawback of the MD method is the general lack of adequate

inter-atomic potentials for highly disordered and hetero-

geneous systems. This is particularly true for microelectronics

semiconductor/oxide interfaces such as silicon/high-k gate

oxides. KMC appears to be a simplification compared to

MD: continuous trajectories are replaced with discrete atomic

jumps. A prior knowledge of these hoppings is needed and

must be listed from any other source: quantum-based models

or experimental characterization. Thus, KMC methods allow

the simulation of more ambitious systems in terms of size and

duration of the simulated experiment, meeting the require-

ments of the next generation of processes in microelectronics.

Basic ingredients needed to develop what we call a lattice-

based KMC model are: (i) we will define a lattice framework

able to operate the desired transition between the silicon cubic

structure and the oxide crystal, (ii) we detail the writing of

configurations that allow the association of the lattice sites

with the chemical nature of atoms occupying them, (iii) we

introduce the concept of event that must describe correctly the

chemistry of the process, and (iv) we finally indicate how

to deal with the time evolution of the KMC, namely, the

temporal dynamics of the KMC.

II.1 Coincidence lattice model

In order to define atomic positions in our KMC model, we

have to consider a two-dimensional lattice at the interface

between the two materials. Different atoms are placed within

the 2D lattice at different heights corresponding to the

successive layers. We should emphasize that, in our KMC

view, atomic positions are not defined by accurate Cartesian

coordinates, but only specified approximately around

an equilibrium position within a cell. This view allows

displacements of atoms to satisfy elastic strain usually present

in heterogeneous systems. To represent the above 2D lattice,

we superimpose the conventional cubic cell of oxide and the

silicon-based SiO2 surface in order to find a coincidence

between the two crystallographic arrangements. The silicon

(100) surface is taken as the crystal reference, an ultrathin

silicon dioxide, one atomic layer thick, being able to accom-

modate this structure perfectly. We obtain the schematic

configuration represented in Fig. 1. This resulting picture

optimizes the coincidence between the two materials. The

silicon part contains two layers in order to show the orienta-

tion of the bonds at the interface. Siloxanes bridges and

hydroxyl functions are represented as an indication. For sake

of clarity, we do not represent the dimer formation inducing

Si–O–Si species: silicon atoms are schematically located at the

nodes of the crystal lattice of the diamond structure, without

taking into account the displacements observed usually.

According to the crystallographic data,32 the distance separat-

ing two Si surface atoms is 3.84 A. In addition, if one considers

that the cubic lattice parameter of hafnia is approximately 5 A,

the distance separating two Hf atoms represents the half

diagonal of the base of the cube, approximately 3.5 A. It thus

appears possible to match the two structures, with a tension of

the cubic hafnia.

Let us imagine now that we position several HfO2 cubes in

tension on the substrate, looking for an ideal match of the two

2D lattices. We see that a tilt of the cell is required to obtain a

complete match of the atomic positions. It is then possible to

completely cover the surface without defect: we obtain three

perfectly crystalline layers of HfO2 on Si/SiO2. It is, to some

extent what we would like to obtain in real experiments. Fig. 1

is a top view of this ideal configuration where the various

species are represented by symbols for a better legibility. It

represents a (100) silicon surface, covered by single-crystal

HfO2 in its cubic phase. The square base of a conventional

cubic cell is represented in dotted line: it locates five hafnium

atoms: the four corners of the square and the central atom in

round symbols (l = 2). One atomic layer above, we find four

hafnium atoms corresponding to the centres of the four side

faces of the f.c.c. lattice, in square symbols (l = 3). Between

these two layers, the first four oxygen atoms are in dotted line

round symbols (l = 2 + 1/2). By adding two new layers, one

of oxygen atoms and one of hafnium atoms, one obtains the

totality of the conventional cube which is obviously not the

primitive cell, minimal for the construction of the lattice.

Considering the successive layers, by increasing index l, one

meets various species: silicon atoms marked as triangles, for

l = 1, Hf atoms marked as full rounds for all even layers, Hf

atoms marked as squares for all odd layers, except l = 1, and

oxygen atoms marked as dotted rounds, for all half integer

layers. For sake of clarity of Fig. 1, oxygen atoms are only

shown in the conventional cube. The cell described above

makes it possible to represent in an exhaustive way the oxide

3702 | Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 This journal is �c the Owner Societies 2009

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

lattice, on the basis of its cubic phase, on a silicon (100)

surface. It contains preset sites being able to accommodate

the various atoms constitutive of the system.

II.2 KMC temporal dynamics

Kinetic Monte Carlo approaches differ from static Monte

Carlo techniques by the introduction of time. Most of the

usual Monte Carlo procedures in condensed matter physics

are static (Metropolis algorithm for instance) and their interest

concern the equilibrium structure properties. In KMC,

stochastic aspects are introduced through the use of random

numbers, sampled according to an Arrhenius probability law,

rather than using direct sampling, with uniformly distributed

random numbers. The result of this procedure is that chemical

Fig. 1 Hafnia cubic lattice deposited on the Si/SiO2 substrate, functionalized with hydroxyls groups (left side). Top view of the ideal configuration

(right side). HfO2 cubic cells are matched to the substrate through a coincidence lattice and different types of atoms are positioned in the cell.

Fig. 2 Substrate initialization with randomly distributed siloxanes.

This journal is �c the Owner Societies 2009 Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 | 3703

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

aspects are respected, but the most probable events do not

necessarily occur first. Different algorithms do exist for KMC.

In the following, we suggest a procedure that becomes effi-

cient, in comparison with the more conventional BKL (Bortz,

Kalos and Lebowitz) algorithm,33 when dealing with many

different types of events having different probabilities. This is

the case of systems where local deformation energy is taken

into account for the evaluation of the occurrence probabilities,

or where the system exhibits a complex chemistry. It should be

emphasized here that, in our KMC procedure, an event is

defined as elementary mechanism occurring at a specific site.

When the software is launched, after preliminary initializa-

tions (parameters, mechanisms, neighbourhood, configura-

tion), it makes a global scan to determine which events are

authorized and which are forbidden. Forbidden ones get a

maximum time so that they can never occur. Authorized ones

get a ‘‘specific occurrence time’’: this is the time that the

considered event typically takes to occur as soon as it becomes

possible. The ‘‘specific occurrence time’’ of the event m on site

(i,j,k) is given by

Ti;j;k;m ¼� logðZÞ

lm

with

lm ¼ n exp �DEm

kBT

� �

where Z is a random number uniformly distributed between 0

and 1, lm is the probability of occurrence of the mechanism

per unit of time, n is the attempt frequency, in the order of the

typical lattice vibration frequency, DEm the activation energy

of the mechanism, kB the Boltzmann constant and T the

temperature. The two arrival mechanisms (1-precursor arrival

and 2-water arrival) obey Maxwell–Boltzmann statistics:

l1;2 ¼cPSffiffiffiffiffiffiffiffiffiffiffiffiffiM1;2T

p

where c is a constant, P the pressure, S the elementary 2D-cell

area, M1,2 the considered species molar mass and T the

temperature. Going back to the definition of events, it becomes

obvious that different ‘‘specific occurrence times’’ are attribu-

ted to a given elementary mechanism occurring at different sites

if the deposited layer. Following this procedure, we determine a

huge list of ‘‘specific times’’, a kind of ‘‘calendar’’, just as if all

events that will occur were already foreseen. Not exactly

in fact, because this calendar will often be updated as the

configuration evolves. As soon as the calendar is updated,

the software finds the minimum ‘‘specific occurrence time’’ and

the corresponding event occurs. As this minimum time has

passed, it is withdrawn from the other times contained in the

‘‘calendar’’; a new configuration is edited; pertinent events that

are likely to become possible are added to the list and the

forbidden ones withdrawn from the list. We then go back to the

first KMC stage and search again the minimum time.

II.3 KMC application to ALD

The KMC software discussed here is built in order to simulate

the ALD experimental process. ALD consists of several cycles,

each one composed of four phases: precursor pulse, precursor

purge, hydrolysis and water purge. Each phase has its own

thermodynamic parameters and duration: some mechanisms

are always possible whereas others can only occur in a typical

phase. For instance, ‘‘Precursor arrival’’ will obviously exist

during the first phase only.

The basic chemical reactions involved in ALD process are

the incorporation of the precursors, followed by the formation

of the HCl by product, and the hydrolysis of the chlorine

terminated bonds of the incorporated precursor. Each of the

above global reactions can be decomposed into several steps,

called ‘‘elementary mechanisms’’ in our KLC approach.

The thermodynamic characteristics of these ‘‘elementary

mechanisms’’: reaction enthalpy and activation barrier, have

been determined by ab initio calculations and presented in

section III. Obviously, all implemented mechanisms are

assumed to be reversible, with probabilities defined by the

reverse activation barriers, via the Arrhenius law.

In addition to the above basic mechanisms of ALD, we have

also implemented, in our KMC model, less straightforward

mechanisms, involved in the ALD of high-k materials, which

we call ‘‘Densification mechanisms’’. Densification is defined

here as the necessary phase transition of the high-kmetal oxide

materials from their molecular structure in the gas-phase

precursors, to their solid-state structure in the deposited film.

This is particularly true in the ALD deposition of HfO2 where

the metal has a covalent bonding structure in the gas phase,

with a small coordination number in its precursor form,

whereas the metal oxide has mainly an ionic structure, with

a large coordination number, in the deposited film. This

transition is mediated by the presence of oxygen atoms:

densification could occur in the gas-phase in the presence of

pre-oxidized precursors. We have also shown that densifica-

tion can be seen as a local phenomenon where, during bridging

of two metallic centres, a local re-allocation of the oxygen

atoms occurs.

Therefore, densification mechanisms in the kinetic Monte

Carlo account for all transitions from molecular state to ionic

or crystalline states, i.e. nodes of the lattice. These include

reaction between grafted molecular ligand and the surface,

such as the bridging process of eqn (1), but also during

bridging between two grafted molecules in tree-like positions.

Si–O–HfCl3 + Si–OH - Si–O–HfCl2–O–Si + HCl (1)

The by-product is usually HCl (when bridging a Cl-terminated

tree to an OH-terminated tree) or H2O (when bridging two

OH-terminated trees).

III. Results and discussion

III.1 Ab initio results

The aim of ab initio investigations is to understand the

chemistry of possible reactions, during the first cycles of

ALD of HfO2. The related predictive method, cluster- or

periodic-based DFT, is particularly suitable for chemical

mechanisms that are difficult to reach by experiments or other

modelling strategies.

3704 | Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 This journal is �c the Owner Societies 2009

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

The reactions between the gaseous precursors HfCl4 and

H2O with the hydroxylated SiO2 surface are predicted to

proceed via an exchange mechanism34,35 and can be separated

into two half-reactions: the first half-cycle reaction [eqn (2)]

takes place during the initial step of atomic layer deposition of

HFO2, i.e. the decomposition of HfCl4 precursor molecules:

8SiO2–OH + HfCl4 8SiO2–O–HfCl3 + HCl, (2)

This first half-cycle reaction is itself decomposed into three

elementary reversible mechanisms: (i) chemisorption of the

precursor molecule and its possible desorption, (ii) incorpora-

tion of the chemisorbed molecule and the formation of HCl by

product, and (iii) desorption of HCl which leads to the definite

incorporation of the metal atom. The decomposition of a

global reaction into several elementary mechanisms allows

the detailed examination of the incorporation process rather

than the use of global sticking coefficients.

The second half cycle reaction [eqn (3)] is water exposure

leading to the hydrolysis of Cl-terminated bonds:

8SiO2–O–Hf(OH)3�x–Clx + H2O

8SiO2–O–Hf(OH)4�x–Clx�l + HCl (3)

where x has values of 1–3.

This second half cycle reaction is also decomposed into two

elementary mechanisms: (i) chemisorption of H2O and

(ii) definite hydrolysis. We have shown that the characteristics

of both mechanisms depend on the value of x.

We have studied these reactions in detail using DFT cluster

models. Results are published in our recent papers,36 reported

here in Table 1 and implemented in our KMC package. Values

reported in Table 1 concern direct reaction characteristics:

reaction enthalpies and activation barriers. The characteristics

of reverse reactions can be easily deduced from the reported

values.

III.2 Kinetic Monte Carlo (KMC) results

A First cycle results. The starting surface is known to play

a crucial role at all stages of the device development and

operation. The first ALD cycle, i.e. precursor–substrate inter-

action, is governed by the specificity of the surface cleaning

processes and subsequent presence of nucleation sites. There-

fore, we start by investigating the dependence of the surface

coverage on the density of OH nucleation sites on the

substrate. Surfaces functionalized with hydroxyl groups can

be prepared in various ways37,38 and allow a variable concen-

tration of sites for incorporating the precursors. We thus

simulated a first ALD phase with a long duration: 100 ms

against 50 ms generally used in experiments, at 300 1C with

different starting substrates covered with 0%, 25%, 75% and

100% OH, distributed arbitrarily (Fig. 2). Such a long time is

required to overcome the poor sticking of the precursors and

to saturate efficiently the reactive sites. For a fifth case, we

used the option of initialization ‘‘no more than one OH by

dimer’’ which leads to 47% hydroxyl coverage.

The resulting coverage rates are represented in Table 2. An

increase in the total coverage is observed when one increases

the number of reactive sites of the substrate. In the case 4, it

can be noted that the total coverage amounts to approximately

55%, indicating that a majority of hafnium atoms are thus

bridged consuming each one two OH: one for the incorpora-

tion and a second for the bridging process.

Experimentally, at the end of the first ALD cycle at 300 1C,

a total coverage of 33% was measured,37 which is in good

agreement with our simulations with 50% and 47% OH

coverages. On the other hand, Zhuravlev’s results39 confirm

the 50% OH coverage of SiO2 surfaces at 300 1C (see Table 3).

The excellent agreement between the simulated total coverages

and those obtained experimentally confirms the great relia-

bility of DFT results for the first mechanisms of growth and

the success of their implementation at the mesoscopic scale.

An important issue in the ALD process is the growth

temperature. The ALD of HfO2 is often performed using

metal chloride and water successively as precursors.22–24

Experimentally, it is found that HfCl4/H2O ALD of SiO2

has a broad process window with deposition temperature

between 180 1C and 600 1C. Further the metal chloride–water

ALD process suffers from some disadvantages such as chlorine

contaminations and slow growth rates due to submonolayer

growth per cycle.40–42 As a result, high temperatures are

required to minimize the contamination. Unfortunately, the

film quality degrades at high temperatures. Our aim here is to

optimize this temperature as a compromise between these two

opposite effects, by examining KMC simulations at different

temperatures. We use the Zhuravlev results39 for the distribu-

tion of OH groups on the substrate surface as a function of

temperature (Table 3).

Fig. 3 shows the total coverage of the substrate with respect

to temperature, after the first ALD cycle. The total coverage

increases from 200 1C to 300 1C, then decreases until 400 1C.

This confirms the existence of an optimal growth temperature

at T = 300 1C, at least for the first ALD cycle. This predicted

temperature is in excellent agreement with the optimal experi-

mental temperature largely used for ALD of HfO2 on

SiO2/Si.43–44 This result confirms again the credibility of

our model.

Another important issue in the ALD process is the duration

of precursor pulses. Experimentally, in the case of HfO2

Table 1 Reaction enthalpies and activation barriers for precursor incorporation and hydrolysis processes

Precursor incorporation

Si–OH–HfCl4 complex TSa Si–O–HfCl3–HCl complex HCl desorptionEnergy/eV 0.48 0.88 0.22 0.12

Hydrolysis processes

M–Cl3–H2O complex TSa M–OHCl2–HCl complex HCl desorptionEnergy/eV 0.68 0.97 0.63 0.20

a TS: transition state.

This journal is �c the Owner Societies 2009 Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 | 3705

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

deposition, the pulse duration of HfCl4 precursor is in the

order of 50 ms. The question here is to know whether this time

is sufficient or overestimated? Fig. 4 represents the evolution of

the total coverage as a function of time. Simulations are

preformed under constant flux of HfCl4, at a pressure of

1.33 mbar, at the optimized growth temperature of 300 1C.

The results indicate a saturation of the coverage at 48%. At

the standard duration of 50 ms, the total coverage is only of

40%. The standard experimental duration cannot thus be

reduced without a strong reduction of the total coverage.

Moreover, it could be, according to our simulations, slightly

increased, at least for the very first phase of ALD. We should

mention that the standard time of 50 ms, corresponding to the

precursor pulse phase, is an experimental value, very difficult

to measure in industrial ALD equipments, since their

measurement threshold level is in this order of 50 ms. There-

fore, this experimental value must be regarded as indicative.

Examining the results of simulation of this first cycle, we can

assert the excellent validity of DFT calculations reported in

ref. 29 and 36a concerning HfCl4 decomposition and metal

incorporation, since these mechanisms are massively used in

this first cycle. A correlation is obtained between the total

coverage and the preparation of the surface: distributions of

siloxanes and hydroxyl groups. Further, using the Zhuravlev

model to initialize the starting substrate, according to

the temperature, we highlighted the existence of an optimal

temperature for the incorporation of the metal precursors

during the first phase of ALD. In our simulations, this

optimum was a consequence of a competition between two

contrary effects. However, the experimental results do not

reveal this expected behaviour at low temperatures. We can

therefore assume that some basic mechanisms are missing in

our simulator, in order to reproduce reliably the experimental

results at temperatures lower than 300 1C. This does not

constitute a serious drawback since the growth is generally

carried out at 300 1C, even higher. We finally highlighted the

saturation of the substrate by the precursors, the light over-

estimate of the pulse phase duration and coverage limit are

probably due to an overestimate of hydroxyl density obtained

by the model of Zhuravlev. This model has to be adapted to the

substrates used in the cases investigated here. In all these cases,

the tendencies were in agreement with the experimental results.

B Kinetics of growth. In experiments, several growth

modes are successively observed.37 After the first cycle

consuming a large number of metal precursors, a slow rate

transient mode allows reaching full coverage of the substrate.

Then, a faster steady-state mode, characteristics of the deposi-

tion of HfO2 on HfOx(OH)y, is established. Our aim here is to

reproduce these experimental observations. In the following,

we will investigate the kinetics of growth of HfO2 films, for

both transient and steady state regimes, up to 10 ALD cycles.

Simulation results are presented, and compared with experi-

mental data, in Fig. 5a and b, for transient and steady state

regimes, respectively. Growth rates are expressed in number of

metal atoms incorporated in the film rather than coverage, since

some precursors are incorporated on top of already deposited

metal atoms. This is a consequence of the full coverage being

only reached after a large number of ALD cycles.

As we demonstrated above, Fig. 5a shows that the first cycle

surface coverage is well reproduced by our simulations. This is

no more the case for the second cycle where simulated growth

Table 2 Coverage and crystallinity for different initial substrates

25% random OH 50% random OH 75% random OH 100% random OHRandom OHon dimers (47%) Exp.

Coverage of HfO2 18% 32% 43% 55% 30% 33%

Table 3 Zhuravlev model for substrate initialization. Fromthe Monte Carlo point of view, OH density is the percentage ofhydroxylated sites

T/1C % OH

200 68250 59300 52350 43400 34

Fig. 3 Evolution of coverage rates and crystallinity for different initial substrates.

3706 | Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 This journal is �c the Owner Societies 2009

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

rate increases slightly, while the experimental rate shows a

severe decrease. At this point, we should mention the main

difference between the first and second cycles of ALD.

During the first cycle, incoming precursors react only with

hydroxylated SiO2. Reactions with already deposited pre-

cursors are forbidden since the Cl-terminations have not yet

been hydrolyzed. During the second cycle, some precursors

react with still uncovered SiO2 surface while others react with

the already deposited and hydrolyzed Hf(OH)x molecules.

Obviously, energetic parameters are different for these two

reaction pathways. This hopefully leads to lower sticking of

the incoming precursors to the HfO2 surface. In the absence of

reliable data, based on DFT calculations, this difference has

not been taken into account in the KMC simulations. The

lower sticking of precursors to HfO2 explains the decrease of

the growth rate, observed experimentally, but not in the KMC

simulations. The slight increase of the theoretical growth rate

in the second ALD cycle is due to the increase of hydroxyl

concentration on the hydrolyzed HfO2 surface, with respect to

the initial SiO2 substrate.

Further, a large number of incorporations participate to the

constitution of long life-time ‘‘trees’’, where two precursors,

on top of each other, are covalently bonded. The transition to

an ionic structure, close to the bulk HfO2 structure, can only

be achieved via densification mechanisms which are expected

to play an important role during ALD process. Continuing

further, from cycle 3 to cycle 6, our simulations are in good

agreement with experimental data, about 1014 hafnium atoms

per cm2 per ALD cycle. Beyond cycle 7, the simulated growth

rate decreases and the growth stops at cycle 9. This corres-

ponds to a saturation resulting from a lack of elementary

mechanisms in our KMC, particularly diffusion mechanisms

which allow the smoothing of the growing surface, and the

inability of the densification mechanisms to be fully efficient.

An examination of the final structure indicates specific sites

where further densification mechanisms should occur but

cannot be performed by the mechanisms already implemented

in our software package. To overcome this problem, we are

presently considering a revision of the algorithm in order to

implement new densification mechanisms.

To complete our investigation, we have also examined the

case of 100% OH concentration on the initial substrate, as an

insight into the steady state regime represented by the growth

of HfO2 on fully hydrolyzed HfO2 surface. In practice, the

Fig. 4 Evolution of coverage rates as a function of growth temperature after the first ALD phase in standard conditions.

Fig. 5 Evolution of growth rate as a function of ALD Cycle: (a) transient regime, (b) steady-state regime.

This journal is �c the Owner Societies 2009 Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 | 3707

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

simulation is performed by artificially depositing a full mono-

layer of HfO2 on the silicon substrate. We apply the Monte

Carlo procedure only after this first step. Results are very

similar to the previous case (see Fig. 5b). However, the rise of

growth rate observed previously during the second is no more

present, since the 100% coverage is artificially reached from

the beginning. In experiments, after a slow growth to cover

the substrate, a faster steady-state growth regime, around

1.2 � 1014 hafnium atoms per cm2 and ALD cycle, is observed.

In simulations, the growth rate is again too high at the

beginning, because of the inadequate parameterization model.

From cycle 4, the calculations are in good agreement with

experiments, but the growth stops later as it lacks mechanisms.

Our simulations correctly reproduce the two regimes of the

growth observed experimentally,37 at least for some ALD

cycles. Results are schematically presented in Fig. 6. In the

transient regime, after the first cycle leading to relatively high

substrate coverage, the growth rate decreases to about 0.1

layer per cycle. Our simulator suffers from a saturation effect

inherent to the method but simulations are very realistic

during approximately 5 ALD cycles. The opening of the

siloxanes bridges seems to be at the origin of the slow rate

of this mode. Siloxane openings delay the full coverage of the

surface and a more effective functionalization. The duration of

this stage is not accessible to our simulator without new

optimizations.

The steady-state regime then follows (see Fig. 6). The

growth rate is higher, because of the higher hydroxyl concen-

tration on the surface. The growth rate is in agreement with

experimental data:25,44 about 1.2 � 1014 Hf per cm2 per cycle,

i.e. 0.17 monolayer per cycle. It still remains very low

compared to what could be expected in an ‘‘atomic layer

deposition’’ technique where full monolayer coverage should

normally be reached at each cycle.

IV. Conclusions

The purpose of the present paper is to describe a new genera-

tion of tools dedicated to the modelling of high-k material

deposition on silicon substrates, providing a growth-modelling

environment for atomic layer chemical vapour deposition

(ALD) of the materials as HfO2. A multi-scale strategy has

been chosen to achieve this investigation. The corresponding

‘‘bottom-up’’ approach, from quantum level calculations to

macroscopic rate theory, is necessary to understand and

implement the most pertinent atomic scale mechanisms in

the growth models. We point out that this atomistic descrip-

tion is crucial to match the requirements of microelectronics

and emerging nanotechnology scales. Indeed, our approaches

provided interesting results concerning the substrate reactivity,

growth mechanisms and the optimization of the growth

temperature for HfO2 deposition. Moreover, some studies

are still running to make progress towards the associated

technology, in particular the calibration of activation barriers

for densification mechanisms and siloxane bridge opening.

These calibrations will allow us to draw a clear picture of

the growth mechanisms and growth kinetics, for several

deposition cycles, until a full monolayer is achieved.

Acknowledgements

This work was founded by the French National Research

Agency (ANR) through LN3M project No. ANR-05-CIGC-

0003-02.

References

1 P. A. Packan, Science., 1999, 285, 2079.2 G. D. Wilk, R. M. Wallace and J. M. Anthony, J. App. Phys.,2001, 89, 5243.

3 G. D. Wilk and D. A. Muller, App. Phys. Lett., 2003, 83, 3984.4 (a) S. M. George, A. W. OTT and J. W. Klaus, J. Phys. Chem.,1996, 100, 13121; (b) Z. Xu, M. Houssa, S. D. Gendt andM. Heyns, App. Phys. Lett., 2002, 80, 1975.

5 S. Ferrari, G. Scarel, C. Wiemer and M. Fanciulli, J. App. Phys.,2002, 92, 7675.

6 H. B. Park, M. J. Cho, J. Park, S. W. Lee, C. S. Hwang, J. P. Kim,J. H. Lee, N. I. Lee, H. K. Kang, J. C. Lee and S. J. Oh, J. App.Phys., 2003, 94, 3641.

7 T. Kawahara and K. Torii, IEICE Transactions on Electronics.,2004, E87C, 2.

8 J. J. Ganem, I. Trimaille, I. C. Vickridge, D. Blin and F. Martin,Nuclear Instruments & Methods in Physics Research SectionB-Beam Interactions with Materials and Atoms, 2004, 219, 856.

9 W. Chen, Q. Q. Sun, M. Xu, S. J. Ding, D. W. Zhang andL. K. Wang, J. Phys. Chem. C, 2007, 111, 6495.

10 D. H. Triyoso, R. I. Hegde, J. Grant, P. Fejes, R. Liu, D. Roan,M. Ramon, D. Werho, R. Rai, L. B. La, J. Baker, C. Garza,T. Guenther, B. E. White and P. J. Tobin, Journal of VacuumScience & Technology B, 2003, 22, 2121.

11 Z. M. Rittersma, F. Roozeboom, M. A. Verheijen, J. G. M. vanBerkum, T. Dao, J. H. M. Snijders, E. Vainonen-Ahlgren, E. Tois,M. Tuominen and S. Haukka, Journal of the ElectrochemicalSociety, 2004, 151, 716.

12 H. S. Chang, H. Hwang, M. H Cho and D. W. Moon, AppliedPhysics Letters, 2005, 86, 031906.

13 D. H. Triyoso, M. Ramon, R. I. Hegde, D. Roan, R. Garcia,J. Baker, X. D. Wang, P. Fejes, B. E. White and P. J. Tobin,Journal of the Electrochemical Society, 2005, 152, 203.

14 M. Cho, H. B. Park, J. Park, S. W. Lee, C. S. Hwang, J. Jeong,H. S. Kang and Y. W. Kim, Journal of the Electrochemical Society,2005, 152, 49.

15 K. Kukli, T. Aaltonen, J. Aarik, J. Lu, M. Ritala, S. Ferrari,A. Harsta and M. Leskela, Journal of the Electrochemical Society,2005, 152, F75.

16 D. H Triyoso, R. I. Hegde, B. E White and P. J. Tobin, Journal ofApplied Physics, 2005, 97, 124107.

17 D. Hellin, A. Delabie, R. L. Puurunen, P. Beaven, T. Conard,B. Brijs, S. De Gendt and C. Vinckier, Analytical Sciences, 2005,21, 845.

18 P. D. Kirsch, M. A. Quevedo-Lopez, Y. Li, H. J. Senzaki,J. J. Peterson, S. C. Song, S. A. Krishnan, N. Moumen,

Fig. 6 The three phases of growth kinetics.

3708 | Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 This journal is �c the Owner Societies 2009

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online

J. Barnett, G. Bersuker, P. Y. Hung, B. H. Lee, T. Lafford,Q. Wang, D. Gay and J. G. Ekerdt, Journal of Applied Physics,2006, 99, 023508.

19 I. S. Park, T. Lee, D. K. Choi and J. Ahn, Journal of the KoreanPhysical Society, 2006, 49, S544.

20 M. J. Cho, R. Degraeve, Pourtois, A. Delabie, L. A. Ragnarsson,T. Kauerauf, G. Groeseneken, S. De Gendt, M. Heyns andC. S. Hwang, IEEE Transactions on Electron Devices, 2007, 54, 752.

21 L. Nyns, L. Hall, T. Conard, A. Delabie, W. Deweerd, M. Heyns,S. Van Elshocht, N. Van Hoornick, C. Vinckier and S. De Gendt,Journal of the Electrochemical Society, 2006, 153, F205.

22 M. Ritala, M. Leskela, L. Niinisto, T. Prohaska, G. Friedbacherand M. Grasserbauer, Thin Solid Film, 1994, 250, 72.

23 J. Aarik, A. Aidla, A. A. Kiisler, T. Uustare and V. Sammelselg,Thin. Solid Film, 1999, 340, 110.

24 K. Kukli, J. Ihanus, M. Ritala and M. Leskela, Appl. Phys. Lett.,1996, 68, 3737.

25 M. L. Green, M. Y. Ho, B. Busch, G. D. Wilk, T. Sorsch,T. Conard, B. Brijs, W. Vandervorst, P. I. Raisanen, D. Muller,M. Bude and J. Grazul, J. App. Phys., 2002, 92, 7168.

26 M. Ylilammi, Thin Solid Films., 1996, 279, 124.27 E. L. Lakomaa, Appl. Surf. Sci., 1994, 75, 185.28 S. M. George, A. W. Ott and J. W. Klaus, J. Phys. Chem., 1996,

100, 121.29 A. Dkhissi, A. Esteve, C. Mastail, S. Olivier, G. Mazaleyrat,

L. Jeloaica and M. Djafari Rouhani, J. Chem. Theo. Comp.,2008, 4, 1915–1927.

30 M. Kotrla, Comp. Phys. Comm., 1996, 97, 82.31 P. Smilauer and D. D. Vvedensky, Phys. Rev. B, 1995, 52, 14263.32 Crystal Structure, ed. R. Wyckoff, John Wiley & Sons, New York,

1965, vol. 1.

33 A. B. Bortz, M. H. Kalos and J. L. Lebowitz, J. Comput. Phys.,1975, 17, 10.

34 M. Ritala and N. Leskela, in Handbook of Thin Film Materials,ed. H. S. Nalwa, Academic Press, New York, 2001, vol. 1.

35 A. Rahtu and M. Ritala, J. Mater. Chem., 2002, 12, 1484.36 (a) L. Jeloaica, A. Esteve, M. Djafari Rouhani and D. Esteve, App.

Phys. Lett., 2003, 83, 542; (b) A. Esteve, L. Jeloiaca,G. Mazaleyrat, A. Dkhissi, M. Djafari Rouhani, S. Ali Messaoudand N. Fazouan, MRS., 2003, 786, 35; (c) L. Jeloaica, A. Esteve,A. Dkhissi and M. Djafari Rouhani, Comp. Mater. Science., 2005,33, 59; (d) A. Dkhissi, A. Esteve, L. Jeloaica, D. Esteve andM. Djafari Rouhani, J. Am. Chem. Soc., 2005, 127, 9776;(e) A. Dkhissi, A. Esteve, L. Jeloaica, M. Djafari Rouhani andG. Landa, Chem. Phys., 2006, 323, 179.

37 D. Blin, PhD thesis, University of Montpellier, 2003.38 O. Renault, D. Samour, D. Rouchon, P. H. Holliger, A. M. Papon,

D. Blin and S. Marthon, Thin Solid Films, 2003, 428, 190.39 L. T. Zhuravlev, Colloids and Surfaces A, 2000, 173, 1.40 D. Riihela, M. Ritala, R. Matero and M. Leskela, Thin Solid

Films., 1996, 289, 250.41 J. Aarik, A. Aidla, A. Jaek, A. A. Kiisler and A. A. Tammik, Acta

Polytech. Scand., Chem. Technol. Metall. Ser., 1990, 195, 201.42 L. Hiltunen, H. Kattelus, M. Leskela, M. Makela, L. Niinisto,

E. Nykanen, P. Soininen and M. Tiitta, Mater. Chem. Phys., 1991,28, 379.

43 G. D. Wilk and D. A. Muller, App. Phys. Lett., 2003, 83, 3984.44 P. D. Kirsch, M. A. Quevedo-Lopez, H. J. Li, Y. Senzaki,

J. J. Peterson, S. C. Song, S. A. Krishnan, N. Moumen,J. Barnett, G. Bersuker, P. Y. Hung, B. H. Lee, T. Lafford,Q. Wang, D. Gay and J. G. Ekerdt, J. App. Phys., 2006, 99,023508.

This journal is �c the Owner Societies 2009 Phys. Chem. Chem. Phys., 2009, 11, 3701–3709 | 3709

Dow

nloa

ded

by U

nive

rsite

Pau

l Sab

atie

r on

22

May

201

2Pu

blis

hed

on 2

6 M

arch

200

9 on

http

://pu

bs.r

sc.o

rg |

doi:1

0.10

39/B

8215

02B

View Online