Thin Solid Films 571 (2014) 23–31

Contents lists available at ScienceDirect

Thin Solid Films

journal homepage: www.elsevier.com/locate/tsf

Influence of Ti top electrode thickness on the resistive switching

properties of forming free and self-rectified TiO2 − x thin films
P. Bousoulas ⁎, I. Michelakaki, D. Tsoukalas
Department of Applied Physics, National Technical University of Athens, Iroon Polytechniou 9 Zografou, 15780 Athens, Greece

a r t i c l e i n f o a b s t r a c t

Article history: The memory performance of titanium oxide (TiO2 − x)-based resistive memories containing an ultra thin reactive
Received 15 May 2014 Ti top electrode can be greatly enhanced. Very good switching memory characteristics were demonstrated for an
Received in revised form 20 September 2014 Au/Ti/TiO2 − x/Au/SiO2/Si structure with the insertion of a Ti nanolayer at the Au/TiO2 − x interface. Due to the
Accepted 22 September 2014
superb ability of Ti to absorb oxygen atoms from the dielectric matrix, a large amount of oxygen vacancies is
Available online 2 October 2014
created, which are crucial for the stable function of the memory devices. As the Ti thickness increases, a thick
Keywords:
interfacial layer is created, which degrades the resistive switching behavior. The induced interface thickness is
Thin films found also to affect the fluctuation of the ON/OFF processes. The very good switching characteristics which
Resistive memory devices were recorded for the devices containing Ti as top electrode, denote the direct impact that Ti has on the oxygen
Non-volatile memories vacancy density. Oxygen vacancy distribution is also found to be directly associated with the filaments' diameter.
Conduction mechanisms Thus, the resistive switching mechanism is proposed to be associated with the formation/rupture of oxygen
Oxygen vacancies vacancy-based conducting filaments at the Ti/TiO2 − x interface. Self-rectifying characteristics were also recorded
Conducting filaments for all samples in the low resistance states. Conduction mechanism analysis revealed that trap-assisted-tunneling
Electrical characterization
is the dominant conduction mechanism, which also strongly affects the distribution of the current during SET
Sputtering
process.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction in resistive memories [9,10]. The physical mechanisms which are

The continuous scaling of semiconductor technology has pushed the
conventional flash memories towards its limits. Requirements for
higher integration densities, bigger operation speeds and superior de-
vice performance are incessantly increasing. So the research community
has turned into alternative memory concepts. Among them resistive
memories seem to gain significant interest, due to their simple fabrica-
tion process and low cost materials involved. The resistive switching
phenomenon is classified into many categories [1]. The most common
classification is into unipolar or bipolar switching. In unipolar switching,
the switching does not depend on the polarity of the applied bias, but on
the amplitude, whereas in bipolar the switching strongly depends on
the bias polarity. Resistive switching has been reported in many transi-
tional metal oxides, such as HfO2 [2], Cr2O3 [3], Ta2O5 [4], perovskite
oxide thin films, such as SrTiO3 [5], organic materials [6] and on TiO2
nanoparticle assemblies [7]. One of the most common materials which
have been studied for resistive switching is TiO2 [8], which is extensive-
ly used in electronic and optoelectronic applications. However, a fully
understanding of the switching mechanism is essential for the proper
application of the devices. Many models have been proposed for the
elucidation of the underlying physics and the conduction mechanisms
⁎ Corresponding author.
E-mail address: panbous@mail.ntua.gr (P. Bousoulas).
responsible for the transition from a high-resistance-state (HRS) to a
low-resistance-state (LRS) are classified as filament-type [11] and
interface-type mechanisms [12]. In the filamentary resistive switching,
current flows through limited local paths, mainly consisted of oxygen
vacancies (V+20 ) [13], whereas in the interface type resistive switching
the current is defined by the barrier height between the dielectric
matrix and the electrode [14]. For the both types of switching, there is
a general consensus that the migration of oxygen ions under the appli-
cation of an external electric field and thus the intentionally creation of
oxygen vacancies plays a major role in the switching phenomenon [15].
So by adjusting the density of oxygen vacancies into the device active
core, we can obtain excellent switching characteristics.
In this paper we examine how we can improve and stabilize the
switching effect of TiO2 − x based resistive devices by altering the Ti
top electrode (TE) thickness, demonstrating that there is a direct
connection between the switching phenomenon and the electrode
material. It is known that switching between different resistance states
of TiO2 is associated with the oxygen vacancies into the active layer [16].
Yang et al. [17] examined the effect of introduction of an ultrathin Ti
adhesion layer under the Bottom Electrode (BE) on TiO2 based Resistive
Random Access Memory (RRAM) showing improved switching unifor-
mity. Oh et al. [18] showed that the oxidation reaction between various
metal TEs (including Ti) and TiO2 thin film at the interface is a crucial
factor for the resistive switching characteristics. Kim et al. [19] have

http://dx.doi.org/10.1016/j.tsf.2014.09.041
0040-6090/© 2014 Elsevier B.V. All rights reserved.
24 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31

investigated the role of Ti thickness on atomic layer deposited TiO2 thin

films, suggesting that the amount of generated oxygen vacancies can be
controlled by varying the thickness of Ti. Several other research groups
have also studied the switching performance of various metal oxides
using a Ti nanolayer at the TE/metal oxide interface, demonstrating im-
pressive switching properties [20–23]. All the above results indicate
that oxygen ion migration has a direct impact on the resistive switching
properties by deliberately controlling the defects in the film. Thus, is of
major importance to explore the influence of Ti electrode thickness on
the oxygen ion migration and oxygen vacancy creation in metal oxide
based memory devices.
The main focus of our work is the repercussion of a Ti TE nanolayer
thickness on the switching effect, which provides us a way of controlling
the oxygen vacancy density within the film. We have already investigat-
ed one way of manipulating the oxygen vacancy density of TiO2 − x thin
films by increasing the oxygen content during the fabrication proce-
dure [24], obtaining excellent non-volatile memory performance,
and at the same time structural analysis results which highlight the
non-stoichiometric nature of the film (thus giving rise to formation of
oxygen vacancies) in conjunction with columnar growth structure
(which give rise to formation of grain boundaries) were provided. So
here we provide another feasible way for producing stable memory de-
vices by depositing ultra thin Ti layers above the dielectric film. Through
electrical characterization techniques (Section 3.A) we associate the Ti
TE thickness effect on the HRS and LRS of the devices and investigate
its impact during significant functions of memory devices. A chemically
active metal (like Ti) capture oxygen ions from the bulk film leading to
the generation of oxygen vacancies, which are essential for the forma-
tion of conducting filaments. Nevertheless, a relatively thick titanium
layer will form a dense interfacial layer which would deteriorate the
switching performance of the memory cell. At the same time we elabo- Fig. 1. a) Schematic structure of the TiO2 thin film sandwiched between metal electrodes.
rate on the complex conduction mechanisms that rule the switching An ultrathin titanium layer is deposited above the TiO2 layer, b) band diagram of a) for
effect (Section 3.B). By analyzing the electronic transport mechanisms VG = 0 V.

we calculate the size of conducting filaments that mainly consist of ox-

ygen vacancies and electrons. Moreover, we determine that the density
of oxygen vacancies into the dielectric matrix decisively affects the fila-
ment size. Moreover self-rectifying behavior was recorded for both HRS
and LRS, which is of crucial importance for crossbar structures, thus
eliminating the necessity of integration of an extra diode in the memory
cell. The presence of high vacancy density with the film has also a prom-
inent effect on the SET current distribution, imposing a gradual and not
a steep change of the measured current. In addition the ability of Ti TE to
act as an oxygen reservoir enables multi-level switching applications.
Finally, we present a schematic interpretation of the switching effect
(Section 3.C) in order to highlight the important role of Ti TE on the
oxygen vacancy distribution.
2. Experimental
The film deposition technique is based on a physical vapor deposi-
tion process. This technique is based on the nucleation and growth in
flight of the metallic atoms from the gas phase and the simultaneous
oxidation under oxygen gas flow. Using such a technique room temper-
ature formation of thin film becomes possible under high purity vacuum
conditions. The structure of our samples was the following: Au/Ti/TiO2
− x/Au/SiO2/Si, as it can be noticed in Fig. 1(a), while their energy
band diagram configuration is presented in Fig. 1(b). The SiO2 was
grown thermally on a p-type Si substrate, with 300 nm thickness. After-
wards, a 40 nm Au BE was deposited on the SiO2 by e-gun evaporation.
Subsequently, titanium oxide was deposited through reactive RF mag-
netron sputtering technique from a high purity Ti target (99.9%) at
room temperature. Deposition was performed at a pressure 0.3 Pa
using an r.f. power of 200 W. The deposited films, with 45 nm thickness,
were not subjected to any kind of thermal annealing after the fabrica-
tion process. The flow of Ar was 10 sccm and O2 was 2.5 sccm (oxygen
content 20%). After deposition of the TiO2 film, titanium and gold were
evaporated and patterned to form metal–insulator–metal capacitors,
through a lift-off photolithographic process. For titanium we used vari-
ous thicknesses, namely 0 nm for Sample N1, 2 nm for Sample N2, 4 nm
for Sample N3 and 8 nm for Sample N4, whereas for gold we evaporated
40 nm. The Au capping was used in order to protect the titanium layer
from surrounding oxidation. The side of each top electrode is 100 μm
and their shape is square. In addition we used 3 different size cells
(200 μm, 400 μm and 1000 μm) in order to test the influence of
electrode area on the switching effect. Electrical characterization was
performed by applying all signals to the TE, keeping the BE grounded.
Current–voltage (I–V) measurements were performed with an
HP4140B picoamperometer, capacitance–voltage (C–V) measurements
at 1 MHz using an Agilent 4284A impedance analyzer while ns pulses
were applied with an HP8116A pulse generator.
3. Results and discussion
3.1. Memory performance
3.1.1. DC characterization
Fig. 2 shows the typical DC I–V hysteresis loops for all samples. Four
switching cycles are indicated by arrows 1 to 4 by sweeping the voltage
bias between − 5 V and 5 V (step 200 mV) in forward and backward
direction. The devices were switched from HRS to LRS under the appli-
cation of a positive voltage bias on TE (SET process) and turned back to
HRS at negative bias voltage (RESET process). All devices were initially
found in the HRS and exhibited bipolar resistive switching behavior
without applying any electro-forming voltage bias. The very basic idea
behind this is that the low oxygen content of the deposited film in
conjunction with the Ti TE ability to act as an oxygen reservoir provides
the necessary conditions for the formation of oxygen vacancy-based
 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31 25

Sweep Bias Amplitude (V)

-6 -4 -2 0 2 4 -6
6 -4 -2 0 2 4 6
-2
10 3 3 10-2
RR~400
10-4 RR~5 10-4
4
10-6 10-6
4
1st 1st
10-8 2nd 2nd 10-8
10th 2 10th
Current (A)
2 100th
10-10 10-10

Current (A)
100th
1 1
Sample N1 Sample N2
10-12 10-12
10-2 3 3 10-2
RR~10 RR~300
4
10 -4 4 10-4

10-6 10-6
1st 1st
10-8 2nd 2nd 10-8
10th 10th
2 2
10-10 1 100th 1 100th
10-10
Sample N3 Sample N4
10-12 10-12
-6 -4 -2 0 2 4 -6
6 -4 -2 0 2 4 6
Sweep Bias Amplitude (V)

Fig. 2. I–V forward and backward sweeps, performed for all samples, revealing the continuous transition between the HRS (forward sweep) and LRS (backward sweep). The arrows in the
graphs indicate the switching direction. Sweeps start from −5 V (denoted with arrow 1), proceed to 5 V (denoted with arrow 2) and then return back to −5 V (denoted with arrows 3 and
4), i.e. −5 V → 5 V → −5 V. For each sample four switching curves are presented: the first, the second, the tenth and the one hundredth switching cycle. Also rectification ratios (RR) for
the LRS are presented for all samples for the one hundredth switching cycle curve.

filamentary path [25]. In other words, the filamentary path is already
opened and is strengthened during electrical measurements. Further-
more, the oxygen gettering ability of Ti provides capabilities for multi-
level switching [26], as we are going to ascertain below. It is also
interesting to remark that all devices exhibit a clear rectifying effect,
as can be confirmed from the highly asymmetric curves of Fig. 2. The
current for both HRS and LRS at the positive bias voltage region is
much higher compared to that at a negative bias voltage, denoting
that the devices behave as a diode [27]. The rectification ratio at ±1 V
is about 1–2 orders of magnitude and is improved as we increase the
Ti thickness. For all measurements a current compliance (CC) was set
at 10 mA in order to avoid the hard breakdown of the dielectric.
Sample N1 exhibits bipolar resistive switching with low levels of
measuring current for both HRS and LRS. The OFF/ON ratio estimated
at V = 1 V is about 2 orders of magnitude, margin as requested for
digital applications. However, the switching performance is deteriorat-
ed by consecutive cycle function, rendering problematic the device op-
eration. The transparency of Au TE to oxygen ions makes possible the
accumulation of them at the interface so that some serious defects, e.g.
voids and anode deformation may eventually take place [28]. If the TE
is chemically inert it is not efficient to capture oxygen ions from the
TiO2 − x dielectric matrix. As a result, it requires high voltages to create
a conductive channel or filament in the TiO2 − x film. The repeated cy-
cling switching operation (Fig. 9) leads to the fast approach between
HRS and LRS due to the fast recombination of oxygen ions and oxygen
vacancies.
In Sample N2 we record an improved switching behavior with
respect to Sample N1. Slightly increased currents for both HRS and LRS
are measured with a slight increased OFF/ON ratio. DC endurance
discloses a more stable behavior, indicating the existence of a different
mechanism at the TE/TiO2 − x interface. Since the standard Gibbs energy
for formation of metal oxides is higher for the Ti with respect to Au,
more oxygen ions are migrated into the Ti to form a TiO2 film. Hence
the TE acts as an oxygen reservoir in this case. The extracted oxygen
ions produce a gradient of oxygen vacancies which generate conductive
filaments with conical shape. The ultrathin Ti TE, however, does not in-
duce a large number of oxygen vacancies within the film therefore cre-
ating filaments with small diameter.
As far as it concerns Sample N3, a beneficial outcome of introducing
a 4 nm Ti as TE is discernible. A remarkable enlargement of the
switching is recorded accompanied with a good cycling endurance.
The increase of the ratio is attributed to the creation of high density of
oxygen vacancies that are necessary components for the formation of
strong percolating filaments. The progressive displacement of HRS at
positive bias to higher current values indicates partial rupture of the
filaments.
Finally for Sample N4, although we have gained in terms of OFF/ON
switching ratios that become larger than six orders of magnitude, the
cycling endurance exhibits severe degradation after 100 cycles. The
thicker Ti TE attracts a great number of oxygen ions leading to creation
of a thick interfacial layer [29]. Given that many oxygen ions are
absorbed in the TE, it is more difficult for them to shift back into the
TiO2 − x, hence increasing the dispersion of the switching characteristics
[49]. An ultra high number of oxygen vacancies is assumed to be created
leading to the formation of filaments with large diameter, which do not
shut down totally. The creation of large filaments can cause Joule
heating when large current passes through the filament, which can
raise the filament temperature, thereby creating variations in the device
performance. All the above results denote a way of controlling the den-
sity of oxygen vacancies into the active layer in order to achieve stable
and repeatable operation of our devices. Recently, it has become widely
accepted that the resistive switching processes in binary oxide
non-volatile memories are associated with oxygen vacancies [30,31].
Therefore, the Ti-induced interface layer plays an important role of
manipulating the oxygen vacancies, thus providing a possibility of
enhancing the switching effect. In addition, we have to notice from
Fig. 2 that at the first switching cycle for all samples in the negative
bias region, the LRS branch does not coincide with the HRS branch, sug-
gesting partial rupture of the filaments. The RESET process is attributed
to the recombination of low electron occupied oxygen vacancies with
26 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31
oxygen ions. The drift velocity of the ion migration is given by the
following expression [32]:
    insertion of the 8 nm TE, makes difficult for oxygen ions to remove
α E qαE
vion ¼ exp − m sinh
t0 kB T kB T
where E is the average electric field, Z is the charge number (2 for
oxygen ions), q is the charge quantity, t0 is the lattice vibration time
(~1013 s), α is the lattice constant (also the mesh size), q the charge quan-
tity and Em the oxygen migration barrier (1.1 eV for Ti TE and 0.1 eV
for Au TE). When the electric field is relatively small (E ~ 1 MV/cm)
the exponential dependence dominates over the hyperbolic sinusoidal
dependence, thus the interface barrier should play a decisive role
during RESET procedure. From Fig. 2, we record a bigger deviation of
the HRS and LRS branches in the negative bias region for Sample N2
with respect to Sample N1, which stems from the difference in the
barrier heights. For Samples N3 and N4 we record and even bigger
deviation of the two branches due to the migration of many oxygen
ions. However, for all Samples the values of current in the negative
bias region coincide after the first switching cycle, but the transition
from HRS to LRS is gradual and not steep, suggesting the existence of
multiple filaments and not a single one. Indeed, only in a device with
many conducting filaments we can observe the gradual transition,
since each filament is ruptured under different voltage bias [33]. More-
over, also for the SET process we do not observe a steep transition. This
behavior is strongly associated with the complex conduction mecha-
nism procedure (Section 3.B) and the dynamical evolution of the barrier
height as a function of the voltage bias that drops at the filament. Hence,
the modulation of filament cross-section during the voltage sweep
determines the transition rate (r) of the SET process. As the barrier
cross section becomes smaller the transition rate increases and
thus the transition becomes more abrupt [34]. Furthermore, the self-
rectifying characteristics are attributed to the generation of high num-
ber of oxygen vacancies into the device active core. The work function
difference between Au and TiO2 − x is only 0.1 eV, so electron
injection at positive bias is limited neither by electron barrier height
nor from barrier width, since accumulation of oxygen vacancies at the
cathode (Au) modulates the barrier width, resulting in high tunneling
current [35]. On the other hand electron injection from the Ti TE at
negative bias is mainly limited by the high barrier interface between
the Ti/TiO2 − x interface (1.1 eV). So the increased Ti TE thickness
improves the self-rectification characteristics due to its enhanced
oxygen gettering ability. In Fig. 3 we present the dependence of both
resistance states and the OFF/ON resistance ratio as a function of the
TE thickness, during the first switching cycle.
1011
Resistance (Ohm)
109 HRS
LRS
7
10
105
103
0 2 4 6 8 -6
Ti Thickness (nm)
Fig. 3. Dependence of the OFF/ON switching ratio as a function of the Ti TE thickness, at a
read out voltage of 1 V.
Although the increase of Ti TE thickness leads to high switching
ratios, at the same time the statistical variations are enlarged. The
reason is that the creation of a dense interfacial layer induced by the
ð1Þ from the interface and recombine with oxygen vacancies. Furthermore,
the ultra high number of oxygen vacancies which is created by 8 nm TE
thickness, seriously compromise the stability of the device, since it can
lead to the aggregation of point defects into extended defects, deterio-
rating the switching performance. The resistance of LRS is slightly
decreasing until 2 nm TE thickness and then sharply reducing and
remaining stable for 4 nm and 8 nm TE thicknesses. This abrupt
decrease between 2 and 4 nm indicates a critical thickness for the inter-
facial redox reaction, as has been reported for SrTiO3 by grazing
incidence small angle x-ray spectroscopy for 5 nm Ti above the film
[36]. This behavior is also consistent with the creation of conducting
filaments with Magnéli phase structure, where the resistivity of the
filaments is independent of the bulk film, while the sharp decrease
between 2 nm and 4 nm TE thicknesses indicates the generation
of high number of oxygen vacancies, which augment the filaments'
diameter. So, for Samples N1 and N2 the filaments should be in nm
scale and grow during electrical measurements, whereas for Samples
N3 and N4 the filaments should be larger in shape. Besides, the low
oxygen content gives rise to the formation of several discrete Magnéli
phases [37] TinO2n − 1 (n = 4–10, mostly n = 4 or 5), which even at
room temperature present metallic behavior. In contrast, the small de-
crease of HRS from Au to Ti TE is explained from the larger barrier height
of Ti with respect to Au and the chemical inert nature of Au. For Ti TE the
HRS is practically invariable, with the only difference of the statistical
variations.
In order to highlight the oxygen ion migration, C–V hysteresis char-
acteristics were measured. Fig. 4 shows the typical C–V curves for all
Samples under sweeping voltages from − 5 V → 5 V → − 5 V. The
capacitances at LRS were very high for Samples N2, N3 and N4 with
respect to Sample N1, under SET operation at read voltage of + 4 V,
due to the movement of oxygen ions through grain boundaries and
storage at the Ti TE reservoir. The charge trapping density (oxygen
vacancies) was calculated from C–V hysteresis characteristics using
the equation below [38]:
ΔV  C LRS
Nch arg e ¼ ð2Þ
qA
where q is the electronic charge, ΔV the hysteresis memory window
(evaluated at 0.5 nF) and A is the area of the resistive switching memory
6.0n
Sample N1
5.0n Sample N2
Sample N3
106
Off/On Resistance Ratio
Sample N4
Capacitance (F)
4.0n
105
3.0n
104 2.0n
103 1.0n
RESET SET
102 0.0
-4 -2 0 2 4 6
Sweep Bias Amplitude (V)
Fig. 4. C–V hysteresis characteristics for all samples. The sweeps begin from −5 V, proceed
to 5 V and return back to −5 V.
 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31 27

device. The charge trapping densities of Samples N1 and N4 are 2 × 100

1013 /cm2 and 1 × 1015/cm2, respectively. The Ti nanolayer Samples

Cumulative Probability (%)

had a higher density due to trapping more oxygen ions and hence gen-
erating more oxygen vacancies, which improved the resistive switching 80
performance. We remark that the capacitance at negative voltages
shown in Fig. 6 is much lower than for positive voltages due to less
charge flow as indicated also by I–V curves (Fig. 2). 60
By further increasing the bias polarity we can achieve bigger ON-
state currents and very high switching ratios (~107). Due to the higher
voltage bias stronger conducting filaments are expected to propagate 40
Sample N1
during ON process, influencing the value of the ON-state current. So Sample N2
the sweep bias amplitude is a key parameter which affects the ON- Sample N3
state. Further changing the bias polarity to the negative voltage causes 20 Sample N4

the migration of oxygen ions to reoxidize the conducting filaments

and switch the device back to OFF-state. However, as shown in Fig. 5,
the OFF-state current does not decrease, implying the existence of a
large number of unruptured conducting filaments. Thus, by increasing
the voltage bias, more oxygen ions can be driven from the interface
and reoxidize the remaining conducting filaments. But, if we continue
to increase the voltage bias the oxygen stored in the Ti reservoir will
be consumed continuously and as a result the oxygen ion level will
be too low to reoxidize the conducting filaments, resulting in a field-
induced ON-state [26]. This behavior was observed for Samples N2
and N3 (indicated in Fig. 5 by arrows), whereas for Samples N1 and
N4 the lack of an interfacial reservoir in combination with the low
diameter conducting filaments and the storage of many oxygen ions,
respectively, inhibit the manifestation of field-induced ON process.
The cumulative probability graph in Fig. 6 reveals the distribution of
the two resistance states, for all Samples. As it concerns LRS, although
the variations are noticeable due to the percolation nature of conducting
filaments [39], the uniformity is quite promising and is improving more
as we increase the thickness of Ti TE due to the formation of bigger
conducting filaments. On the other hand, the HRS values exhibit large
fluctuations which are increased from device to device because of the
difficulty of oxygen ions to return back into the dielectric matrix, as
we marked earlier.
In order to examine the behavior of our samples under consecutive
cycling function, we conducted 100 DC switching cycles (Fig. 7).
0
103 105 107 109 1011
Resistance (Ohm)
Fig. 6. Cumulative probability of HRS and LRS for all samples (read voltage 1 V).
Although the Samples N1–N3 sustain at least 50% of their initial
switching ratio, the Sample N4 loses about 80% of its ratio. This deterio-
ration stems from the capture of many oxygen ions into the titanium TE.
The migration of oxygen ions can by described by two mechanisms:
drift and diffusion. The basic motivations for drift and diffusion effects
are the gradients of the electrostatic and electrochemical potentials,
respectively. Oxygen ion diffusion is considered to be activated by the
external input signals, as the DC endurance patterns indicate [24]
or during the deposition of the Ti TEs [36], with result the transforma-
tion of the metallic Ti TEs into a Ti sub-oxide films. As it concerns HRS,
there is a general trend of reducing to lower resistance values which is
attributed to an increased number of traps which enhance the conduc-
tivity of the film. On the other hand, LRS presents smaller fluctuations
with respect to HRS, which are diminished as we increase the Ti thick-
ness, since stronger conducting filaments are created. Nevertheless,
the distribution of oxygen ions and oxygen vacancies in TiO2 − x varies
with time after applying voltage bias through the TiO2 − x electrolyte.

Sweep Bias Amplitude (V)

-6 -3 0 3 6 -6 -3 0 3 6
10-2 10-2
3 3
10-4 4 10-4

10-6 10-6
4
10-8 10-8
Current (A)

2 2
Current (A)

10-10 1 1 10-10
Sample N1 Sample N2
10-2 3 3 10-2
4 4
10-4 10-4

10-6 10-6

10-8 1 10-8
2 1
2
10-10 10-10
Sample N3 Sample N4
10-12 10-12
-6 -3 0 3 6 -6 -3 0 3 6
Sweep Bias Amplitude (V)

Fig. 5. I–V forward and backward sweeps, performed for all samples, between −6 V and 6 V. The colored arrows at the graphs of Sample N2 and N3 indicate the field-induced ON state. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
28 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31

HRS
1010

109

Resistance (Ohm)
Sample N1
109
Resistance (Ohm)

Sample N2
Sample N3
Sample N4

107
108

105
107
Sample N1
Sample N2
LRS Sample N3
103 Sample N4

1 10
DC Switching Cycles
Fig. 7. Distribution of the two resistance under DC sweeping mode, at a read voltage of 1 V.
Thus the DC endurance patterns which seem to stabilize after 3, 10 and
30 switching cycles for Samples N2, N1 and N3, N4 respectively, are
rather transient and depend on the oxidation of the titanium layer and
conducting filament processes.
3.1.2. Pulse characterization
In order to examine the switching dynamics of our devices, we
subjected them into different positive (ON) and negative (OFF) pulses,
with various widths and fixed heights (5 V for SET and − 6 V for
RESET). Fig. 8 shows typical program/erase endurance characteristics
of the resistive memory devices, under the applications of 10 ms pulses.
The Ti layer improves and stabilizes the endurance performance due to
the uniformity of the dominant switching mechanism (redox process).
In addition, ultra fast pulses were applied with duration ranging from
10 ns to 1 μs (Fig. 9), producing multilevel switching phenomena,
which are ascribed to the manipulation of the interfacial oxygen reser-
voir that permits the formation/rupture of conducting filaments and
thus creating a variety of OFF/ON states. We have to notice that HRS
values remain quite constant during pulse measurements, which is
attributed to the fast recombination of oxygen ions with oxygen vacan-
cies. Charge retention measurements were also performed by charging
the devices into one of the two resistance states, by applying 5 and
1010
100 106
10 100 1000
Pulse Width (nsec)
Fig. 9. Evolution of the resistance for both HRS and LRS with the pulse width, for all
samples (read voltage: 1 V). The pulse height for the SET process was 5 V, whereas for
RESET process was −6 V.
−6 V and 100 ns pulses, and then monitoring the evolution with time
of the current value for 17 h at least, at a read voltage of 1 V. The mea-
surements have been performed at room temperature on all samples
and the results are shown in Fig. 10. The retention failure of Samples
N1 and N2 HRS is attributed to the diffusion component of ion migration
of the electrochemical reaction involved. The LRS exhibits a more stable
behavior which is connected with the formation of stronger filaments,
as we increase the thickness of titanium. A resistance ratio of ~ 102
was obtained after the extrapolated data retention, which will be
of great advantage for the future nanoscale low power non-volatile
memory applications.
3.2. Conduction mechanisms
In order to get a better insight into the conduction mechanisms that
govern the switching behavior, we have conducted detailed analysis of
the I–V characteristics. Fig. 11 demonstrates the fitting results of HRS
from Poole–Frenkel (P–F) emission model whereas the inset shows
the Ohmic fit of LRS, both at positive bias region. For what in concerns
1010
Measurements Extrapolation
Resistance (Ohm)

109 109
Resistance (Ohm)

10 years

108

108
107
Sample N1
Sample N2
Sample N1 Sample N3
Sample N2 Sample N4
106 Sample N3
Sample N4 107
10-1 101 103 105 107 109
100 101 102 103 104 105 Waiting Time (sec)
Number of Cycles
Fig. 10. Charge retention measurements performed on the four samples with program-
ming pulses of 5 and −6 V and duration of 100 ns (read voltage: 1 V). The measurements
Fig. 8. Endurance plot for the memory element. The height of the pulse was −4 V, 5 V for were performed at room temperature in the dark on fresh devices. Samples N3 and N4
the SET/RESET respectively and the duration was 10 ms. present excellent retention capability with a 10 years extrapolated ratio of ~100.
 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31 29

LRS, we observe that for Samples N3 and N4 Ohmic conduction domi- Sample N1
nates the LRS (the abrupt change in the characteristics at about 2 V 1010 HRS
Sample N2
Sample N3
comes from the diffusion component of the ion migration, however Sample N4
the Ohmic slope is sustained). For Samples N1 and N2 Ohmic behavior

Resistance (Ohm)
describes only a minor part of the LRS, indicating the existence of 108
filaments with sub-nm scale. On the other hand, HRS does not exhibit
any power-like dependence from the applied voltage (ruling out
Space Charge Limited Current), but fits well with P–F model, except
from low bias region, where it deviates considerably. Other conduction
106
mechanisms were examined, including Schottky Emission and Fowler–
Nordheim tunneling giving low quality fitting, while direct tunneling LRS
should be considered negligible for a 45 nm thick film. The P–F effect 104
is a well known conduction mechanism that is often used to explain
the increase in conductivity of a dielectric when a high electric field is
applied. The P–F effect is the thermal emission of charge carriers from 102
Coulombic traps in the bulk of a material enhanced by the application 1 10
of an electric field and relies on the assumption that there are enough Cell Size (10 cm2)-4

electrons in the traps for subsequent emission. Nonetheless, in films

with significant interface potential barriers or with large number of
traps where a large number of electrons are required to fill the traps
than the available, this assumption may not stand, making more likely
the manifestation of a trap-assisted-tunneling mechanism [40]
Poole–Frenkel emission is basically one step of the multifarious trap-
assisted-tunneling process [40]. The other possible electron conduction
mechanisms are tunneling from BE (cathode during the application of
positive bias voltage to TE) to traps, trap to trap tunneling, emission
from trap to conduction band (P–F) and tunneling from traps to TE
(anode). The transition rate (v) of any processes dominates governs
the whole conduction behavior, since the step with the lowest transi-
tion rate will limit the whole conduction sequence [24]. So, we conclude
that not just one but a combination of different conduction mechanisms
cooperate in order to switch our device from the HRS to the LRS,
converting the whole conduction process to trap-assisted-tunneling
mechanism.
To further elucidate the conduction mechanism, we carried out elec-
trical measurements in different cell areas, as shown in Fig. 12. The out-
come shows a weak dependence of the resistance of LRS on the cell's
geometrical size while the HRS is decreasing as the cell size increases,
denoting that the current distribution is locally confined through the
device, which is a distinct feature of filament type resistive switching
[41]. Following the procedure proposed in our previous work [24] we
calculate the diameter of the filaments, assuming a value of 10−4 Ω m
for the resistivity of the Magnéli based filaments [42].
-4
10-2 Ohmic Fit
Fig. 12. Area dependence of the resistance value in HRS and LRS, respectively, for
all Samples.
The results reveal the existence of tiny filaments (sub nm scale) for
Samples N1 and N2 and larger filaments for Samples N3 and N4, due
to the generation of many oxygen vacancies, as we have calculated
from Eq. (2). From Fig. 13 we record a straight relationship of the
oxygen vacancy density and the filaments' diameter. Thus, oxygen
vacancies directly contribute to the formation of conducting filaments
and are crucial factor in determining the switching behavior.
3.3. Switching model
TiO2 − x is commonly considered as an electrolyte material, where
both electronic and ionic conduction contribute to the whole electrical
conduction. Based upon the theory of formation and rupture of
conducting filaments, electrochemical migration of oxygen ions is
regarded as the driving mechanism [43], which relies on the notion of
interfacial oxygen migration during the application of bipolar voltage
sweeps (Fig. 14). Since the oxygen ions (O−2) are negatively charged,
they drift to the anode due to the applied electric field, hence producing
oxygen vacancies (V+2 0 ). It has been found [44] that oxygen vacancies
introduce very shallow energy states from the conduction band edge
(0.2 eV) so that these traps can be easily ionized even at room temper-
ature. Thus oxygen vacancies directly contribute to the conductivity
of the dielectric by capturing and releasing electrons from the cathode
and forming conducting filaments [45]. The formation of oxygen

10-4
-8
Abs(I) (A)

100
10-6
Filament Diameter (nm)

1015
ln(I/B) ln(S)

-12 10-8
LRS
Trap Density/cm2

10-10
0.1 1

-16 Abs(Bias) (V) HRS 10

1014
-20 Sample N1
Sample N2
Sample N3
Sample N4
Poole-Frenkel Fit
1
-24
0 1 2 1013
Sqrt(B) (V0.5) 0 2 4 6 8
Ti Thickness (nm)
Fig. 11. Poole–Frenkel plot demonstrating that HRS arises from the emission of carriers
from the Au electrode. The inset shows the Ohmic fit of LRS. Fig. 13. Dependence of filament formation and trap density as function of the Ti thickness.
30 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31

Fig. 14. Schematic representation of the switching mechanism. During the SET process (a) oxygen ions are pulled out of the lattice, stored in the Ti layer and oxygen vacancies are gen-
erated. The Ti top electrode creates an oxygen gradient, creating even more oxygen vacancies as the thickness of Ti is increased. At the same time electrons are injected from the cathode,
resulting in formation of conducting filaments that connect the two electrodes and switch the device to the LRS (b). If the polarity of the applied bias is reversed (c), oxygen ions drift back
and recombine with oxygen vacancies, switching back the HRS. The filament will be dissolved in the weakest part (d), while the main part of the filament is maintained.

vacancies in TiO2 − x dielectric matrix may be described using the
Kröger–Vink defect notation:
x :: − 1
Oo →V 0 þ 2e þ O2 :
2 For very small thicknesses of Ti (0–2 nm) poor resistive switching
The above reaction can be considered to be a self-doping process,
due to the release of electrons, thus causing a huge increase in the
electrical conductivity. Furthermore, titanium ions (Ti+ 4
) in oxygen
i
vacancy region might be formed in order to accomplish the local charge
neutrality, by capturing electrons from the cathode. The reduced Ti ions
and oxygen vacancies can form a metallic phase, TiO2− n/2 where n is
higher than 1.5, therefore creating conduction paths (filaments) in the
TiO2 − x insulating matrix. Density Functional Theory (DFT) studies
[46] have shown that electron injection favors the vacancy stability
and stabilizes the filament, whereas removal of the electrons causes
the oxygen vacancy isolation that ruptures the filament. The underlying
mechanism for the filament stabilization process is the formation
of bonding electron states whose occupation can be controlled by appli-
cation of electric fields. Under the application of a negative bias on the
TE, the oxygen ions are repelled from the interface and reoxidize the
conducting filaments, as it has been shown in experimental [47–49]
and theoretical works [50]. These results indicate the disruption of an
ordered oxygen vacancy chain, thus changing the device from the LRS
into the HRS. We have finally highlighted the use of the Au/Ti/TiO2 − x
structure as one of the most promising methods to improve the perfor-
mance of this kind of memory [51], since TiO2 − x layer acts as an oxygen
vacancy and oxygen ion source, Ti TE acts as an oxygen reservoir for the
oxygen ions that are extracted from the dielectric matrix, and Au pro-
vides a barrier for oxygen ions, due to its chemically inert nature. Stack-
ing of these layers produces variations in the oxygen vacancy
concentration in the memory cell, leading to effective ON/OFF switching
behavior. This three-layer configuration provides opportunities for
multilevel resistive switching applications by controlling the interfacial
oxygen ion migration and thus manipulates the degree of filament
formation and rupture.
4. Conclusions
In summary, the dependence of various thicknesses of Ti which
was deposited as TE, on TiO2 − x RRAM devices, was systematically
investigated. Promising switching performance was obtained, in terms
of switching ratio, DC cycling and pulse endurance, switching dynamics
and retention capability. Since electrochemical migration of oxygen
ions is considered the driving mechanism, the extent of oxygen ions
migration into the TE is essential for enhanced switching performance.
ð3Þ
characteristics were recorded, since few oxygen ions are absorbed. By
depositing thicker Ti layers (4–8 nm), significant amount of oxygen
ions is diffused into the TE, thus promoting the development of oxygen
vacancies based filaments. The migration of oxygen ions gives also rise
to self-rectifying effects while the oxygen vacancy density is found to
be directly connected with filament size, implying their crucial role on
the formation of conducting filaments. However, a thicker Ti TE displays
considerable deterioration during consecutive cycling switching due to
the formation of a dense interfacial oxide layer. The existence of high
number of oxygen vacancies which is attributed to the low oxygen con-
tent of our samples in combination with the oxygen vacancies induced
by the Ti TE, inhibits the full manifestation of P–F emission, indicating
that trap-assisted-tunneling is the main conduction mechanism of
HRS. At the same time, traps set out limitations for the measured cur-
rent, permitting only the gradual and not steep transition of the current.
Calculation of filaments' diameter disclosed their nm scale, indicating
potential for aggressive device scaling.
Acknowledgments
The authors would like to acknowledge financial support from
research program “Aristeia II” (Grant No. 4543). This research has
been co-financed by the European Union (European Social Fund ESF)
and Greek national funds through the Operational Program “Education
and Lifelong Learning” of the National Strategic Reference Framework
(NSRF).
References
[1] R. Waser, R. Dittmann, M. Salinga, M. Wuttig, Function by defects at the atomic scale —
new concepts for non-volatile memories, Solid State Electron. 54 (2010) 830.
[2] V. Sriraman, Z. Chen, X. Li, X. Wang, N. Singh, G.-Q. Lo, HfO2 based resistive switching
non-volatile memory (RRAM) and its potential for embedded applications, Int. Conf.
Solid State Integr. Circ. 32 (2012) 101.
[3] S.C. Chen, T.C. Chang, S.Y. Chen, C.W. Chen, S.C. Chen, S.M. Sze, M.J. Tsai, M.J. Kao,
F.S.Y. Huang, Bipolar resistive switching of chromium oxide for resistive random
access memory, Solid State Electron. 62 (2011) 40.
[4] Y. Yang, P. Sheridan, W. Lu, Complementary resistive switching in tantalum oxide-
based resistive memory devices, Appl. Phys. Lett. 100 (2012) 203112.
 P. Bousoulas et al. / Thin Solid Films 571 (2014) 23–31
[5] R. Muenstermann, R. Dittmann, K. Szot, S. Mi, C.-L. Jia, P. Meuffels, R. Waser, Realiza-
tion of regular arrays of nanoscale resistive switching blocks in thin films of
Nb-doped SrTiO3, Appl. Phys. Lett. 93 (2012) 023110.
[6] B. Cho, S. Song, Y. Ji, T.-W. Kim, T. Lee, Organic resistive memory devices: perfor-
mance enhancement, integration, and advanced architectures, Adv. Funct. Mater.
21 (2011) 2806.
[7] E. Verrelli, D. Tsoukalas, P. Normand, A.H. Kean, N. Boukos, Forming-free resistive
switching memories based on titanium-oxide nanoparticles fabricated at room
temperature, Appl. Phys. Lett. 102 (2013) 022909.
[8] K.M. Kim, B.J. Choi, M.H. Lee, G.H. Kim, S.J. Song, J.Y. Seok, J.H. Yoon, S. Han, C.S. Hwang,
A detailed understanding of the electronic bipolar resistance switching behavior in
Pt/TiO2/Pt structure, Nanotechnology 22 (2011) 254010.
[9] D.C. Kim, S. Seo, S.E. Ahn, D.-S. Suh, M.J. Lee, B.-H. Park, I.K. Yoo, I.G. Baek, H.-J. Kim,
E.K. Yim, J.E. Lee, S.O. Park, H.S. Kim, U.-I. Chung, J.T. Moon, B.I. Ryu, Electrical obser-
vations of filamentary conductions for the resistive memory switching in NiO films,
Appl. Phys. Lett. 88 (2006) 202102.
[10] Y.B. Nian, J. Strozier, N.J. Wu, X. Chen, A. Ignatiev, Evidence for an oxygen diffusion
model for the electric pulse induced resistance change effect in transition-metal
oxides, Phys. Rev. Lett. 98 (2007) 146403.
[11] D.-H. Kwon, K.M. Kim, J.H. Jang, J.M. Jeon, M.H. Lee, G.H. Kim, X.-S. Li, G.-S. Park, B. Lee,
S. Han, M. Kim, C.S. Hwang, Atomic structure of conducting nanofilaments in TiO2
resistive switching memory, Nat. Nanotechnol. 5 (2010) 148.
[12] S.X. Wu, L.M. Xu, X.J. Xing, S.M. Chen, Y.B. Yuan, Y.J. Liu, Y.P. Yu, X.Y. Li, S.W. Li
Reverse-bias-induced, Bipolar resistance switching in Pt/TiO2/SrTi0.99Nb0.01O3/Pt
devices, Appl. Phys. Lett. 93 (2008) 043502.
[13] D. Li, M. Li, F. Zahid, J. Wang, H. Guo, Oxygen vacancy filament formation in TiO2: a
kinetic Monte Carlo study, J. Appl. Phys. 112 (2012) 073512.
[14] R. Dong, D.S. Lee, M.B. Pyun, M. Hasan, H.J. Choi, M.S. Jo, D.J. Seong, M. Chang, S.H. Heo,
J.M. Lee, H.K. Park, H. Hwang, Mechanism of current hysteresis in reduced rutile TiO2
crystals for resistive memory, Appl. Phys. A 93 (2008) 409.
[15] Joshua Yang, M.D. Pickett, X. LI, D.A.A. Ohlberg, D.R. Stewart, R.S. Williams,
Memristive switching mechanism for metal/oxide/metal nanodevices, Nat.
Nanotechnol. 3 (2008) 429.
[16] S.G. Park, B. Magyari-Kope, Y. Nishi, Impact of oxygen vacancy ordering on the
formation of a conductive filament in TiO2 for resistive switching memory, IEEE
Electron Device Lett. 2 (2011) 197.
[17] J.J. Yang, J.P. Strachan, Q. Xia, D.A.A. Ohlberg, P.J. Kuekes, R.D. Kelley, W.F. Stickle,
D.R. Stewart, G.M. Ribeiro, R.S. Williams, Diffusion of adhesion layer metals controls
nanoscale memristive switching, Adv. Mater. 22 (2010) 4034.
[18] S.C. Oh, H.Y. Jung, H. Lee, Effect of the top electrode materials on the resistive
switching characteristics of TiO2 thin film, J. Appl. Phys. 109 (2011) 124511.
[19] S.J. Kim, M.G. Sung, M.S. Joo, W.G. Kim, J.Y. Kim, J.H. Yoo, J.N. Kim, B.G. Gyun, J.Y. Byun,
J.S. Roh, S.K. Park, Y.S. Kim, Characteristics and the model of resistive random access
memory switching of the Ti/TiO2 resistive material depending on the thickness of Ti,
Jpn. J. Appl. Phys. 50 (2011) 04DD14.
[20] S.-Y. Wang, D.-Y. Lee, T.-Y. Tseng, C.-Y. Lin, Effects of Ti top electrode thickness on
the resistive switching behaviors of rf-sputtered ZrO2 memory films, Appl. Phys.
Lett. 95 (2009) 112904.
[21] A. Prakash, S. Maikap, H.-C. Chiu, T.-C. Tien, C.-S. Lai, Enhanced resistive switching
memory characteristics and mechanism using a Ti nanolayer at the W/TaOx inter-
face, Nanoscale Res. Lett. 8 (2013) 288.
[22] H.Y. Lee, Y.S. Chen, P.S. Chen, T.Y. Wu, F. Chen, C.C. Wang, P.J. Tzeng, M.-J. Tsai, C. Lien,
Low-power and nanosecond switching in robust hafnium oxide resistive memory
with a thin Ti cap, IEEE Elec. Dev. Lett. 1 (2010) 44.
[23] S.Z. Rahaman, S. Maikap, T.-C. Tien, H.-Y. Lee, W.-S. Chen, F.T. Chen, M.-J. Kao, M.-J.
Tsai, Excellent resistive memory characteristics and switching mechanism using a
Ti nanolayer at the Cu/TaOx interface, Nanoscale Res. Lett. 7 (2012) 345.
[24] P. Bousoulas, I. Michelakaki, D. Tsoukalas, Influence of oxygen content of room
temperature TiO2− x deposited films for enhanced resistive switching memory
performance, J. Appl. Phys. 115 (2014) 034516.
[25] T. Gu, Role of oxygen vacancies in TiO2-based resistive switches, J. Appl. Phys. 113
(2013) 033707.
[26] S.-Y. Wang, C.-W. Huang, D.-Y. Lee, T.-Y. Tseng, T.-C. Chang, Multilevel resistive
switching in Ti/CuxO/Pt memory devices, J. Appl. Phys. 108 (2010) 114110.
[27] C.-W. Hsu, C.-C. Wan, I.-T. Wang, M.-C. Chen, C.-L. Lo, Y.-J. Lee, W.-Y. Jang, C.-H. Lin,
T.-H. Hou, 3D vertical TaOx/TiO2 RRAM with over 103 self-rectifying ratio and sub-μA
operating current, IEDM 264 (2013) 10.4.1.
31
[28] J.J. Yang, F. Miao, M.D. Pickett, D.A.A. Ohlberg, D.R. Stewart, C.N. Lau, R.S. Williams,
The mechanism of electroforming of metal oxide memristive switches, Nanotech-
nology 20 (2009) 215201.
[29] H.Y. Jeong, J.Y. Lee, S.-Y. Choi, Interface-engineered amorphous TiO2-based resistive
memory devices, Adv. Funct. Mater. 20 (2010) 3912.
[30] F. Nardi, S. Larentis, S. Balatti, D.C. Gilmer, D. Ielmini, Resistive switching by voltage-
driven ion migration in bipolar RRAM — Part I: experimental study, IEEE Trans.
Electron Devices 9 (2012) 2461.
[31] B. Magyari-Köpe, S.G. Park, H.-D. Lee, Y. Nishi, First principles calculations of oxygen
vacancy-ordering effects in resistance change memory materials incorporating
binary transition metal oxides, J. Mater. Sci. 47 (2012) 7498.
[32] N.F. Mott, R.W. Gurney, Electronic Processes in Ionic Crystals, Clarendon, Oxford,
1948.
[33] B. Gao, B. Sun, H. Zhang, L. Liu, X. Liu, R. Han, J. Kang, B. Yu, Unified physical model of
bipolar oxide-based resistive switching memory, IEEE Electron Device Lett. 12
(2009) 1326.
[34] E. Miranda, D. Jiménez, J. Suñé, From post-breakdown conduction to resistive
switching effect in thin dielectric films, reliability physics symposium (IRPS), IEEE
Int. (2012) GD.5.1.
[35] C.-W. Hsu, I.-T. Wang, C.-L. Lo, M.-C. Chiang, W.-Y. Jang, C.-H. Lin, T.-H. Hou,
Self-rectifying bipolar TaOx/TiO2 RRAM with superior endurance over 1012
cycles for 3D high-density storage-class memory, Symp. VLSI Technol. Dig. (2013)
T166.
[36] S. Stille, Ch. Lenser, R. Dittmann, A. Koehl, I. Krug, R. Muenstermann, J. Perlich,
C.M. Schneider, U. Klemradt, R. Waser, Detection of filament formation in forming-
free resistive switching SrTiO3 devices with Ti top electrodes, Appl. Phys. Lett. 100
(2012) 223503.
[37] G.V. Samsonov, The Oxide Handbook, IFI/Plenum Press, New York, 1982.
[38] S. Maikap, A. Das, T.Y. Wang, T.C. Tien, L.B. Chang, High-κ HfO2 nanocrystal memory
capacitors prepared by phase separation of atomic-layer deposited HfO2/Al2O3
nanomixtures, J. Electr. Soc. 156 (2009) K28.
[39] S. Blonkowski, Filamentary model of dielectric breakdown, J. Appl. Phys. 107 (2010)
084109.
[40] S. Yu, X. Guan, H.-S. Philip Wong, Conduction mechanism of TiN–HfOx–Pt resistive
switching memory: a trap-assisted-tunneling model, Appl. Phys. Lett. 99 (2011)
063507.
[41] S. Zhang, B.S. Long, W. Guan, Q. Liu, Q. Wang, M. Liu, Resistive switching character-
istics of MnOx-based ReRAM, J. Phys. D 42 (2009) 055112.
[42] C. Acha, M. Monteverde, M. Nunez-Regueiro, A. Kuhn, M.A.A. Franco, Electrical resis-
tivity of the Ti4O7 Magnéli phase under high pressure, Eur. Phys. J. B 34 (2003) 421.
[43] Y.C. Yang, F. Pan, Q. Liu, M. Liu, F. Zeng, Fully room-temperature-fabricated nonvol-
atile resistive memory for ultrafast and high-density memory application, Nano Lett.
9 (2009) 1636.
[44] E. Cho, S. Han, H.S. Ahn, K.R. Lee, S.K. Kim, C.S. Hwang, First-principles study of point
defects in rutile TiO2− x, Phys. Rev. B 73 (2006) 193202.
[45] Y. Yang, P. Gao, S. Gaba, T. Chang, X. Pan, W. Lu, Observation of conducting filament
growth in nanoscale resistive memories, Nat. Commun. 3 (2012) 732.
[46] K. Kamiya, M.Y. Yang, B.M. Köpe, M. Niwa, Y. Nishi, K. Shiraishi, Vacancy cohesion-
isolation phase transition upon charge injection and removal in binary oxide-
based RRAM filamentary-type switching, IEEE Trans. Electron Devices 10 (2013)
3400.
[47] C. Chen, C. Song, J. Yang, F. Zeng, F. Pan, Oxygen migration induced resistive
switching effect and its thermal stability in W/TaOx/Pt structure, Appl. Phys. Lett.
100 (2012) 253509.
[48] Y.H. Do, J.S. Kwak, Y.C. Bae, K. Jung, H. Im, J.P. Hong, Oxygen ion drifted bipolar
resistive switching behaviors in TiO2–Al electrode interfaces, Thin Solid Films 518
(2010) 4408.
[49] C. Chen, S. Gao, F. Zeng, G.S. Tang, S.Z. Li, C. Song, H.D. Fu, F. Pan, Migration of inter-
facial oxygen ions modulated resistive switching in oxide based memory devices, J.
Appl. Phys. 114 (2013) 014502.
[50] B. Gao, J. Kang, L. Liu, X. Liu, B. Yu, A physical model for bipolar oxide-based resistive
switching memory based on ion-transport-recombination effect, Appl. Phys. Lett. 98
(2011) 232108.
[51] S.R. Lee, S.R. Lee, Y.-B. Kim, M. Chang, K.M. Kim, C.B. Lee, J.H. Hur, G.-S. Park, D. Lee,
M.-J. Lee, C.J. Kim, U.-I. Chung, I.-K. Yoo, K. Kim, Multi-level switching of triple-
layered TaOx RRAM with excellent reliability for storage class memory, Proc.
Symp. VLSIT (2012) 71.