Microelectronic Engineering 104 (2013) 4247

Contents lists available at SciVerse ScienceDirect

Microelectronic Engineering
journal homepage: www.elsevier.com/locate/mee

Variation of switching mechanism in TiO2 thin lm resistive random access memory

with Ag and graphene electrodes
Yushi Hu a, David Perello a, Minhee Yun a,, Deok-Hwang Kwon b, Miyoung Kim b,
a
b

Electrical and Computer Engineering Department, Swanson School of Engineering, University of Pittsburgh, Pittsburgh, PA 15261, USA
Research Institute of Advanced Materials, Department of Materials Science and Engineering, Seoul National University, Seoul 151-744, Korea

a r t i c l e

i n f o

Article history:
Received 16 May 2012
Received in revised form 17 September
2012
Accepted 7 November 2012
Available online 29 November 2012
Keywords:
Resistive memory
ReRAM
Graphene
Bipolar
Unipolar
Filament
Titanium dioxide
Switching mechanism

a b s t r a c t
We report the fabrication of resistive random access memory (ReRAM) on both Si and PET exible substrates using TiO2 as the dielectric spacer between Ag electrodes. Ag/TiO2/Ag ReRAM shows unipolar
switching behavior with a ramping rate of 50 mV. We further examined the switching mechanism for
Ag based ReRAM in the low resistive state (LRS) and high resistive states (HRS). To elucidate the impact
of electrode material on the switching mechanism, we fabricated a graphene/TiO2/graphene ReRAM
device and observed that the switching behavior changed from unipolar to bipolar due to the unique
physical properties of graphene. This study demonstrates ReRAM based on Ag and graphene electrodes
on both Si and PET substrates, and directly demonstrates the strong dependence of electrode materials
on the switching mechanism.
2012 Elsevier B.V. All rights reserved.

1. Introduction
The development of memory devices, including volatile memories such as DRAM (dynamic random access memory), and nonvolatile memories such as NAND and NOR ash memories, has
reached a turning point where cost effectiveness and performance
gains are evolutionary and not revolutionary in nature. Alternative
devices have been proposed, with the most promising being PCM
(Phase Change Memory) and ReRAM (Resistive Random Access
Memory). PCM has been researched widely both in academia and
reached industrial manufacturing mostly because of its nonvolatility and potential in scalability [1]. However, many challenges remain including switching speed and switching-induced heating
degradation of the phase change material. ReRAM, on the other
hand, has potential for simple cross-bar scalability [2], and a fast
transition speed [35]. The simple resistive electrical operation of
ReRAM also suggests improved efciency [6]. Although ReRAM
shows promising performance, the understanding of its switching
mechanism is still lacking, resulting in a difculty of utilizing this
device structure in the memory industry [7,8].
Two commonly accepted switching mechanisms have been
established for ReRAM dielectrics: a lament based mechanism
Corresponding authors. Tel.: +1 412 648 8989.
E-mail addresses: yunmh@engr.pitt.edu (M. Yun), mkim@snu.ac.kr (M. Kim).
0167-9317/$ - see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.mee.2012.11.009

and a mobile ionic mechanism. Filament based ReRAM switches

between low resistive state (LRS) and high resistive state (HRS)
via the formation and disconnection of a local conduction path
through the dielectric spacer. This conduction path is usually composed of a low resistive metal, metal compound, or is simply a result of a crystalline phase change of the spacer [9,10]. The lament
mechanism irreversibly modies the local material composition
such that complete removal of the path is impossible. Conversely,
for ReRAM devices with mobile ion-driven switching, the electric
eld-driven alignment of mobile charge varies the spacer from
the conducting to non conducting state and vice versa. To achieve
this ionic imbalance, often the spacer is deposited with a vacancy
or donor concentration gradient across the thickness, or through
the use of an explicit multi-layered structure with different charge
concentrations [11]. The changes induced during switching are
reversible and nondestructive, therefore these style devices are
favored for their higher reliability.
An in-depth switching mechanism model has been missing due
to the wide variety of devices displaying bipolar and unipolar ReRAM operation. Many transition metal and complex metal oxides
show ionic-driven or lament switching, or even a combination
of both phenomenon [1214]. As highlighted in a review on the topic [8], the switching is a product of each the spacer/contact interface, ionic transport, and joule-heating induced phase changes.
Differentiating the effects is non-trivial, as clear evidence of local

Y. Hu et al. / Microelectronic Engineering 104 (2013) 4247

laments is difcult to image. In addition, the interfacial physics of

contacts to the oxide material can be locally inhomogeneous and
similarly difcult to characterize. Further, the impacts of different
electrodes on the switching of transition metal oxides have been
recognized and emphasized by previous studies. Different mechanisms based on the electrode metal work function [15], free energy
of interfacial oxide formation [16], and dependence of lament
composition on the electrode materials [17] have been proposed
to understand the impacts from electrode materials.
To shed light on the contact and interfacial effects in switching
behavior, we fabricated ReRAMs based on simple evaporated TiO2
thin lms with different electrodes including Ag and graphene on
Si substrate and exible PET substrates. We examined the switching of Ag based TiO2 ReRAM and graphene based TiO2 ReRAM and
found that by changing the Ag electrodes to graphene electrodes,
the TiO2 ReRAM switched from unipolar lament-based to bipolar
behavior. This effect clearly demonstrates the importance of the
electrode, even when utilizing the same spacer material. Electrodes
with low diffusivity and weak coupling to the oxide could permit a
reduction in local-lament formation and allow fabrication of
ReRAM with nondestructive charge transport.
2. Experimental
The TiO2 based resistive memories are fabricated into sandwiched structures using standard microelectronic processes. For
Ag/TiO2/Ag devices, a 3 inch p-type (1 0 0) Si wafer with 100 nm
SiO2 was used as the substrate. The Ag bottom electrode (BE) was
fabricated via photolithography, 50 nm Ag deposited by e-beam
evaporation (Thermionics Laboratory VE180), and then a liftoff process was performed. After e-beam evaporating a 50-nm thick TiO2
layer, a 100-nm thick Ag top electrode (TE) was fabricated via a second photolithography, deposition, and liftoff process. For the graphene/TiO2/graphene device, graphene was grown via Atmospheric
Pressure Chemical Vapor Deposition (APCVD) on copper foil and

43

transferred to SiO2, details of which can be found in reference

[18]. After transfer, 300 C vacuum annealing was performed in a
tube furnace while owing 10 sccm of H2 to improve adhesion of
the graphene. Following a photolithography and plasma etching
step to pattern graphene electrodes, 50 nm of TiO2 layer was evaporated onto the BE graphene layer. Finally, a TE graphene layer was
transferred, annealed for one hour, and patterned via photolithography. For the fabrication of ReRAMs on exible substrates (exible
ReRAM devices), a polyester (PET) lm with 125 lm thickness was
used as the substrate. The fabrication process for the devices was
identical to that described above.
Fig. 1 shows the device structure of the Ag/TiO2/Ag ReRAM fabricated in this research. The Ag BE and TE are separated by a dielectric layer of 50 nm TiO2. The overlapping area of the BE and TE is
200 lm by 200 lm and contactable via microprobes. Fig. 1a illustrates a ReRAM on the Si substrate while Fig. 1b shows ReRAM
on a PET thin lm with same device dimensions. Schematic and
optical images of a single bit device are shown in Fig. 1c and d,
respectively. We note that the roughness of the PET devices was
greater than the Si substrate. Measurements of the different devices was performed with a probe station (Micromanipulator)
and a semiconductor analyzer (Agilent B1500A). The IV test was
conducted using a voltage sweep at a rate of 50 mV/s in the range
of 10 to +10 V in a cyclic manner.
3. Results and discussion
3.1. Unipolar switching behaviors for Ag/TiO2/Ag ReRAM on both Si
and PET substrates
The ReRAM devices on both Si substrate and PET substrates
were electrically tested. A representative IV sweep for a Si based
ReRAM with Ag TE and BE and a 40 nm thick TiO2 dielectric layer is
shown in Fig. 2a. The device is in a high resistance state (HRS) asfabricated. The green (red) line is a representative turn on cycle

Fig. 1. Fabricated planar ReRAM devices. Images of actual devices. (a) Ag/TiO2/Ag ReRAM device fabricated on a Si/SiO2 wafer. (b) Flexible Ag/TiO2/Ag ReRAM. Schematic (c)
and optical image (d) of individual devices. Scale bar for (d) is 200 lm.

44

Y. Hu et al. / Microelectronic Engineering 104 (2013) 4247

Fig. 2. IV curves for AgTiO2Ag ReRAM on Silicon (a and b) and PET substrates (c and d). (a) Linear scale IV curves for a AgTiOAg device on SiO2/Si substrate with turnon current compliance of 5 lA for both positive and negative polarities. (b) Log scale of (a). (c) Linear scale IV curves for a AgTiO2Ag device on PET substrate device with
unipolar behavior for both polarities. (d) Log scale version of (c).

with a positive (negative) voltage sweep from 0 to 0.1 V (0.1 V)

with a ramping rate of 50 mV/s (50 mV/s). The device is switched
from HRS to a low resistance state (LRS) in the turn on cycle. The
blue (black) line represents a subsequent turn off cycle to switch
the device from LRS back to HRS. Fig. 2b shows a semi-log graph
for Fig. 2a. In order to prevent breakdown of the TiO2 dielectric, a
maximum current compliance of 5 lA is used in turn on cycles.
It is clearly seen that the device has an on/off ratio on the order
of 102. This switching behavior is the typical unipolar behavior,
where the device turns on and off with the same voltage polarity
[12]. This device shows a unipolar switching behavior for both
polarities (some reports refer to such a phenomenon as non-polar
behavior). However, the switching voltages and currents for negative and positive polarities are slightly different, possibly due to
the non ideal symmetry of either the sandwich structure or the inter-layer properties after the fabrication process. Similarly, a PET
based ReRAM shows a unipolar switching behavior in response
to an IV sweep. Fig. 2c presents a PET substrate ReRAM with Ag
TE and BE and a 50 nm thick TiO2 layer as dielectric. This device
also shows a unipolar switching behavior for both polarities.
Fig. 2d presents a semi-log graph for this PET substrate ReRAM,
showing a similar 102 on/off ratio for both voltage polarities.
3.2. Switching mechanism for Ag/TiO2/Ag ReRAM on both Si and PET
substrates
It is demonstrated that both the Si substrate and PET substrate
ReRAM with TiO2 dielectric and Ag BE/TEs show a unipolar switching behavior. An in-depth study of the conduction mechanism for
HRS and LRS of the ReRAM is needed to obtain a clearer view of
the switching mechanism for these ReRAM devices. It has been reported that in a LRS state a highly conductive path is formed in the
TiO2 dielectric layer for ReRAM with unipolar switching behaviors.
This layer can be nano laments composed of either metal or
highly conductive Magneli phase of TinO2n1. In either case, the
highly conductive paths provide ohmic conductions for the LRS

state. For ohmic conduction, we have I / V and logI /

logV C, where C is a constant related to the resistance of the
conduction path. In Fig. 3a, we plot the logI vs. logV for a Si substrate -based ReRAM and make a best linear t for both LRS and
HRS. The LRS state has a slope of 1.056 and a R2 (square of correlation coefcient) of 0.996, indicating a good linear t for the LRS.
Moreover, the slope of 1.056 matches the expected value of 1 in
the ohmic conduction model, proving a formation of highly conductive paths during the switching from HRS to LRS. HRS, on the
other hand, has a slope of 1.209 and a R2 (square of correlation
coefcient) value of 0.897. Both the larger slope value 1.209 and
weaker linear correlation of the best-t line during HRS indicates
a more complicated conduction mechanism. A similar correlation
was therefore done on ReRAM devices fabricated on a PET substrate for both the negative and positive voltage polarities, as
shown in Fig. 3b and c. It is clear that for both negative polarity
(Fig. 3b) and positive polarity (Fig. 3c), the linear t of the LRS state
provides slopes very close to 1 with high R2 values. This again indicates dominance of ohmic conduction. Conversely, the HRS clearly
shows deviation in slope in different regions for both polarities. In
the low voltage region, the slopes for both polarities are very close
to 1 with high R2 values, indicating an ohmic conduction mechanism. In high voltage region, the slopes are larger and approach
2, suggesting a transition to trap-related space charge limited conduction (SCLC). This deviation in slope conrms that the conduction mechanism in HRS is again not a simple ohmic conduction
resulting from the combined effects of ohmic conduction and
trap-related conduction SCLC.
By analyzing the conduction mechanism for both the Si and PET
substrate devices, we conclude that the conduction of the LRS state
for these ReRAM devices with Ag BE/TE is caused by the formation
of an ohmic conductive lament. Many different models have been
proposed for the formation of ohmic conductive laments, and the
most widely accepted ones are local phase changes of the dielectric
material due to joule heating [19] and the bridging of electrodes
due to the diffusion of ions from the metal electrodes. It is clear

Y. Hu et al. / Microelectronic Engineering 104 (2013) 4247

45

Fig. 3. Comparison of the IV curves for LRS and HRS for AgTiO2Ag ReRAM. LRS with slope near 1 is indicative of nearly near-ohmic transport, while HRS with biasdependent slope indicates transitions to hopping or space-charge limited transport. This phenomenon is consistent for each of (a) ReRAM on Si substrate, and positive (c) and
negative (d) bias for ReRAM on PET substrate.

that both dielectric material and electrodes can play an important

role in the formation of conduction paths. However, in this work it
is unclear how this lament is formed. TEM images indicated that
the spacer was amorphous, and although we did not encounter a
lament, this is likely due to the extreme local nature of such
events and hence difculty in encountering one. In order to elucidate the impact of the electrode, and clarify the lament nature,
we examined devices with graphene as the electrode.
3.3. Device structure for Graphene/TiO2/Graphene ReRAM on both Si
and PET substrates
Graphene is a two-dimensional sheet of carbon atoms arranged
in a strong sp2 hybridized hexagonal lattice. Electrically, it is a
semi-metal with modulatable Fermi level and unusually large carrier mobilities [20]. Mechanically, graphene has large tensile
strength and is nearly impervious to atomic migration under typically ReRAM bias ranges. Moreover, the single layer nature permits
high transparency, super thin structures, and easy patterning via
low-power oxygen plasma. These advantages suggest a superior
material for electrodes, especially on transparent and/or exible
substrates. In this work, we use graphene as the electrode material
not only to demonstrate the feasibility of ReRAM with graphene
electrodes, but also to study the impact the different electrode
materials on the switching mechanism of the ReRAM devices. Different from other electrode materials, graphene is unable to form a
highly conductive path via carbon diffusing into the TiO2 or a reaction with the dielectric lm. As a result, graphene gives us a way to
isolate the impact of electrode material diffusion and study the
switching behavior of TiO2. The device structure of graphene based
ReRAM is illustrated in Fig. 4. A single layer of 50 nm thick TiO2 is
sandwiched by two layers of graphene. These two layers of graphene are patterned and used as BE and TE similar to Ag electrodes.
The substrate can be both Si with SiO2 and PET lm. The graphene
is grown on hand-polished 100 micron thick Cu foil via thermal

APCVD (T-APCVD) at 1050 C using a mixture of methane, hydrogen and argon. The graphene was single/bi layer according to
Raman data, as expected for the self-limiting nature of the growth
process on Cu foils.
3.4. Switching mechanism of graphene/TiO2/graphene ReRAM devices
Here the switching behavior of graphene/TiO2/graphene ReRAM
devices was carefully studied in this research. A voltage sweep
with a ramping rate of 50 mV/s was applied to graphene TE while
graphene BE was connected to ground. Fig. 5a illustrates the resulting switching behavior of a graphene based ReRAM on a PET
substrate with 50 nm thick TiO2 dielectric layer. It is clear that
the switching shows bipolar behavior instead of the unipolar
switching behavior found in Ag based ReRAM devices under

Fig. 4. Schematic of graphene/TiO2/graphene ReRAM stack.

46

Y. Hu et al. / Microelectronic Engineering 104 (2013) 4247

Fig. 5. Graphene/TiO2/graphene switching characteristics. (a) Bipolar switching of the ReRAM induced from switching the contact material from Ag to graphene. (b) log(I)
log(V) plot showing the non-ohmic contact nature of the device in both the HRS and LRS. This effect is likely a result of the interface between the graphene and TiO2 and lack
of lament formation. (c) A high-resolution TEM image of oxide spacer with a fast-Fourier transformed micrograph, which indicates Anatase poly crystalline.

identical dielectric deposition. This transition of switching mechanism is further proved by the log IV plot shown in Fig. 5b. For LRS,
although the slope for low voltage region is very close to 1, for high
voltage region the slope is larger. The nonlinearity between I and V
can also be seen clearly in Fig. 5a. Thus, the conduction mechanism
for LRS can no longer be explained by ohmic conduction. It is also
noticed that the graphene based ReRAM has a much larger switching voltage (35 V) than Ag based ReRAM (<1 V). This value, while
signicantly increased, is still within reported bias regimes for devices exhibiting bipolar behavior [12]. The TiO2 spacer had anatase
crystallinity for the graphene case as shown in Fig. 5c. The layer
was not designed with a strict vacancy population and therefore
the resistance was signicantly higher than the layer having a lament-based mechanism. Thus, the graphene based ReRAM with
bipolar behavior has greater overall resistance and larger turn-on
voltages. It is noted that a similar transition from unipolar switching behavior to bipolar switching behavior was observed for the
graphene/TiO2/graphene device on the SiO2/Si substrate, although
the on/off ratio is not as big as the device on the PET substrate
(data not shown). More study is required to understand the impact
of different substrates on the switching performances.
The primary difference between the Ag based ReRAM and
graphene based ReRAM is the electrode material, and a possible
explanation of the transition of switching mechanism can be proposed: since the Ag electrode can easily diffuse into the dielectric
layer at a very small voltage (as small as 80 mV for a 40 nm thick
TiO2 and less than 1 V for 50 nm thick TiO2 in this work), a conductive path can be easily formed by a Ag lament bridging BE and TE.
Such phenomenon was examined using a water spacer by Guo
et al. with <0.6 V switching voltage [21]. The utilization of graphene as the electrode material prevents the diffusion of highly
conductive electrode materials into the dielectric layer and
changes the switching to solely dependent on the TiO2 dielectric
and vacancy movement, and also the TiO2/graphene interface. For
AgTiO2Ag devices the oxide layer was completely amorphous,
while the grapheneTiO2graphene devices also contained regions

of Anatase crystallinity. This difference is attributed to the 300 C

annealing in the case of graphene. However, this factor cannot be
attributed for the difference in device behavior as laments can
easily be formed in either case. No evidence of the Magneli phase
was found from TEM diffraction patterns in either case.
We further note that although graphene can have large carrier
mobility, at the nano/micro scale graphene has a limited capacity
to ow large amounts of current. It is also possible that local lament events are prevented further since such a rapid increase in
carriers at a point location would cause breakdown of the graphene
lattice and destroy the lament path.
These results suggest that if one wishes to prevent the formation of laments, process permitting, graphene could be a strong
candidate as an interface layer between the spacer and contact.
Such a barrier could reduce the formation of local laments if the
graphene is highly crystalline, although metal diffusion at the domain edges may remain a concern.

4. Conclusion
We rst demonstrated the successful fabrication of a ReRAM
device based on Ag/TiO2/Ag sandwich structure on both SiO2/Si
and exible PET substrates. These devices showed unipolar switching behaviors under a voltage sweep with a ramping rate of 50 mV/
s. Further study demonstrated that the switching behavior of Ag
based ReRAM on both Si and PET substrates displays an ohmic conduction mechanism in the low resistance state and trap-related
space charge limited conduction (SCLC) in the high resistance state.
We proceeded to fabricate a graphene/TiO2/graphene ReRAM device incorporating graphene layers as both the bottom electrode
and top electrode. The switching behavior of the graphene based
ReRAM changed from unipolar to bipolar. This behavior is attributed to either an interfacial contact effect or inability for localized
laments to form in the TiO2 dielectric layer. From our results, it is
clear that the switching mechanism strongly depends on the

Y. Hu et al. / Microelectronic Engineering 104 (2013) 4247

electrode materials and interfacial properties in ReRAM devices. It

may be further possible to vary the switching mechanism via use
of the proper electrode materials, simplifying fabrication efforts
with use of different electrodes on the same chip. This could potentially broaden the application of ReRAM for different performance
requirements.
Acknowledgments
This paper was studied with the support of the MEST (Ministry
of Education Science and Technology, Grant No. 20120005637) and
Brainpool. The US authors are grateful for nancial support by the
National Science Foundation (NSF, Grant ECCS 1039543) and
AFOSR-MEST.
References
[1] A. Pirovano, A.L. Lacaita, A. Benvenuti, F. Pellizzer, R. Bez, IEEE Trans. Electron
Devices 51 (2004) 452459.
[2] S.H. Jo, K.H. Kim, W. Lu, Nano Lett. 9 (2009) 870874.
[3] J.J. Yang, M.D. Pickett, X. Li, A.A. OhlbergDouglas, D.R. Stewart, R.S. Williams,
Nat. Nanotechnol. 3 (2008) 429433.
[4] M. Fujimoto, H. Koyama, M. Konagai, Y. Hosoi, K. Ishihara, S. Ohnishi, N. Awaya,
Appl. Phys. Lett. 89 (2006). 223509-3.
[5] 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, Electron Device Lett. IEEE 31 (2010) 4446.

[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]

[14]
[15]

[16]
[17]
[18]
[19]
[20]
[21]

47

L. Chua, IEEE Trans. Circuit Theory 18 (1971) 507519.

R. Waser, M. Aono, Nat. Mater. 6 (2007) 833840.
S. Akihito, Mater. Today 11 (2008) 2836.
Q. Liu, S. Long, H. Lv, W. Wang, J. Niu, Z. Huo, J. Chen, M. Liu, ACS Nano 4 (2010)
61626168.
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, Nat. Nanotechnol. 5 (2010) 148153.
D.B. Strukov, G.S. Snider, D.R. Stewart, R.S. Williams, Nature 453 (2008) 8083.
A. Beck, J.G. Bednorz, C. Gerber, C. Rossel, D. Widmer, Appl. Phys. Lett. 77
(2000) 139141.
I.G. Baek, M.S. Lee, S. Seo, M.J. Lee, D.H. Seo, D.S. Suh, J.C. Park, S.O. Park, H.S.
Kim, I.K. Yoo, U.I. Chung, and J.T. Moon, Highly scalable nonvolatile resistive
memory using simple binary oxide driven by asymmetric unipolar voltage
pulses, in: Electron Devices Meeting, 2004, IEDM Technical Digest, IEEE
International, 2004, pp. 587590.
Y. Watanabe, J.G. Bednorz, A. Bietsch, C. Gerber, D. Widmer, A. Beck, S.J. Wind,
Appl. Phys. Lett. 78 (2001) 37383740.
S. Seo, M.J. Lee, D.C. Kim, S.E. Ahn, B.-H. Park, Y.S. Kim, I.K. Yoo, I.S. Byun, I.R.
Hwang, S.H. Kim, J.-S. Kim, J.S. Choi, J.H. Lee, S.H. Jeon, S.H. Hong, B.H. Park,
Appl. Phys. Lett. 87 (2005) 263507.
C.B. Lee, B.S. Kang, A. Benayad, M.J. Lee, S.-E. Ahn, K.H. Kim, G. Stefanovic, Y.
Park, I.K. Yoo, Appl. Phys. Lett. 93 (2008) 042115.
K. Lin, T. Hou, J. Shieh, J. Lin, C. Chou, Y. Lee, J. Appl. Phys. 109 (2011) 084104.
G.H. Han, F. Gunes, J.J. Bae, E.S. Kim, S.J. Chae, H.J. Shin, J.Y. Choi, D. Pribat, Y.H.
Lee, Nano Lett. 11 (2011) 41444148.
F.A. Chudnovskii, L.L. Odynets, A.L. Pergament, G.B. Stefanovich, J. Solid State
Chem. 122 (1996) 9599.
K.I. Bolotin, K.J. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone, P. Kim, H.L.
Stormer, Solid State Commun. 146 (2008) 351355.
X. Guo, C. Schindler, S. Menzel, R. Waser, Appl. Phys. Lett. 91 (2007) 133513.