Sergiu Clima1, Kiroubanand Sankaran1,4, Maarten Mees1,3, Yang Yin Chen1,3,Ludovic Goux1, Bogdan

Govoreanu1, Dirk J.Wouters1,3, Jorge Kittl1, Malgorzata Jurczak1, Geoffrey Pourtois1,2

1imec, B-3001 Leuven, Belgium; 2PLASMANT,University of Antwerp, B-2610 Antwerpen, Belgium;

3Katholieke Universiteit Leuven, B-3001 Leuven, Belgium;

4ETSF and IMCN, Université Catholique de Louvain, B-1348 Louvain-la-Neuve, Belgium

Ea - hafnium oxides have the lowest activation energies and slightly increasing with the O content/density in the oxide (0.57-0.66 eV), a larger barrier is observed in Al2O3 (1.22 eV) and on the electrodes sides much higher barriers were computed: 2.50 eV in TiN and 4.13 eV in Hf. These activation energies match the experimental window, as measured for HfOx and Al2O3.

5,6

The movement of the O atoms is a local rearrangement of O around Hf atoms, upon which the electrically active defects (VO) show jumps larger than the displacement of any O atom.

O diffusion is facilitated by the free volume that is increasingly more available in the sub-stoichiometric Hf oxides but very difficult in the metallic Hf electrode.

Somewhat higher diffusion coefficients are computed in TiN and Al2O3 vs Hf, but they are still blocking layers for O, if compared to HfOx.

Transition metal oxide valence change Resistor

Random Access Memory (RRAM) operation

principles are based on oxygen-related defect

migration. The switching mechanism is believed to

be driven by the Joule heating enhanced drift of O2-

ions under the applied electric field through the

oxide from/ towards the metal electrode.1,2 The

kinetics of the oxygen diffusion is, therefore, a key

factor for oxide stoichiometry change, which in turn

is responsible for the resistivity of the RRAM cell.

1) Liu, L. F. et al. Engineering oxide resistive switching materials for memristive device application.

Applied Physics a-Materials Science & Processing 102, 991-996, doi:10.1007/s00339-011-6331-2 (2011).

2) Waser, R., Dittmann, R., Staikov, G. & Szot, K. Redox-Based Resistive Switching Memories – Nanoionic

Mechanisms, Prospects, and Challenges. Advanced Materials 21, 2632-2663, doi:10.1002/adma.200900375

(2009).

3) Miron, R. A. & Fichthorn, K. A. Accelerated molecular dynamics with the bond-boost method. Journal of

Chemical Physics 119, 6210-6216, doi:10.1063/1.1603722 (2003).

4) Govoreanu , B. et al. Investigation of forming and its controllability in novel HfO2-based 1T1R 40nm-

crossbar RRAM cells. Ext. Abstr. SSDM Conf.,Nagoya, Japan, pp.1005-1006 (2011).

5) Zafar, S., Jagannathan, H., Edge, L. F. & Gupta, D. Measurement of oxygen diffusion in nanometer scale

HfO2 gate dielectric films. Applied Physics Letters 98, 152903, doi:10.1063/1.3579256 (2011).

6) Nabatame, T. et al. Comparative studies on oxygen diffusion coefficients for amorphous and gamma-

Al2O3 films using O-18 isotope. Japanese Journal of Applied Physics Part 1-Regular Papers Short Notes

& Review Papers 42, 7205-7208, doi:10.1143/jjap.42.7205 (2003).

Motivation Modeling approach

Accelerated Ab Initio Molecular Dynamics

(AIMD) technique combining the bond-boosted

technique3 is used to compute the diffusion

kinetics of O in a series of materials of RRAM

interest. In a TiN/Hf/HfO2/TiN stack, it is

expected that the sputtering of the Hf capping

layer leads to the generation of interfacial

amorphous sub-oxides.4 Beside the electronic

barrier function (a feature needed for self-

rectifying function in a RRAM cell) Al2O3 could

work as O barrier, therefore pinning the

switching layer at a specific location. The

knowledge of diffusion data in amorphous phase

HfOx (x=2,1,1/2), Al2O3 and crystalline TiN and

Hf metallic electrodes helps us understand

where the O comes from and goes into during

switching, as the short time/length simulations

can model the fast and local atomic

rearrangements that must be responsible for the

sub-ns switching events.

a(HfO2) b(HfO1) c(HfO0.5)

e(Al2O3) f(TiN) g(Hf)

Investigated materials Diffusion

f,g

f

c

b

a

RRAM cell

f,g

f

c

b

a

e

SRC RRAM

t

rtrN

D

N

ii

t6

)0()(1

lim 1

2

Model Ea (eV) D0 (m2/s)

HfO0.5 0.57 5.07E-8 HfO1.0 0.60 5.11E-8 HfO1.97 0.66 3.88E-8 Zafar 0.46-0.6 10-8-10-12

Arrhenius plots

a b c g

Ab Initio Molecular Dynamics

DFT/LDA (DZP) forces at each time

step of Newtonian motion of atoms

with a “time machine”

DV

Artificial

potential

b

V

b ett DDD

Frontiers in Electronic Materials Nature Conference

Aachen, 2012

MEM 22