Bipolar resistive switching behavior in MoS2 nanosheets fabricated on ferromagnetic

shape memory alloy

Anuj Kumar, Shuvam Pawar, Shubham Sharma, and Davinder Kaur

Citation: Appl. Phys. Lett. 112, 262106 (2018); doi: 10.1063/1.5037139

View online: https://doi.org/10.1063/1.5037139
View Table of Contents: http://aip.scitation.org/toc/apl/112/26
Published by the American Institute of Physics
 APPLIED PHYSICS LETTERS 112, 262106 (2018)

Bipolar resistive switching behavior in MoS2 nanosheets fabricated on

ferromagnetic shape memory alloy
Anuj Kumar, Shuvam Pawar, Shubham Sharma, and Davinder Kaura)
Functional Nanomaterials Research Lab, Department of Physics and Centre for Nanotechnology, Indian
Institute of Technology Roorkee, Roorkee 247667, Uttarakhand, India
(Received 21 April 2018; accepted 18 June 2018; published online 29 June 2018)
The present study explores the systematic investigation of resistive switching response of magnetron
sputtered MoS2 thin films sandwiched between a Ni-Mn-In ferromagnetic shape memory alloy
(bottom) and copper (top) electrodes. The Cu/MoS2/Ni-Mn-In device exhibits stable and reproducible
bipolar resistive switching behavior. The current-voltage (I-V) analysis suggests that the device shows
ohmic conduction behavior in the low resistance state (LRS) while space charge limited conduction is
the dominating conduction mechanism in the high resistance state (HRS). The compliance current vs
reset current measurements were also performed which reveals that the power consumption of the
device can be suppressed by decreasing the compliance current. To explain the resistive switching
behavior in the Cu/MoS2/Ni-Mn-In device, a conceptual model based on copper ion migration, is
proposed and well explained. The resistive switching parameters such as the set voltage, LRS and
HRS resistances, are also investigated in a temperature range overlapping with the martensite phase
transformation temperatures of the bottom ferromagnetic shape memory alloy (Ni-Mn-In) electrode. It
provides temperature as an extra degree of freedom to modulate the resistive switching characteristics
of the fabricated device. The Cu/MoS2/Ni-Mn-In device shows a decent endurance of 500 and a long
retention of 103. These results demonstrate the feasibility of the MoS2 based devices for futuristic
tunable nonvolatile resistive random access memory applications. Published by AIP Publishing.
https://doi.org/10.1063/1.5037139

In this digital era, there is a constant need for advance- resistive switching characteristics.13,14 However, recent
ment in information processing and miniaturization in nano- advancement and development in technologies require modula-
scale memory circuits. But, it cannot be further accomplished tion of the switching mechanism with some external extra
by conventional silicon charge storage based random access degrees of freedom such as stress, temperature, light, and mag-
memory (RAM) because they have reached their physical and netic field.15,16 Ferromagnetic shape memory alloy (e.g.,
technological limits.1,2 Therefore, a number of alternative Ni-Mn-In) thin films whose characteristics properties can be
memory technologies have been explored to overcome the tailored by stress and temperature, can be a good choice as an
scalability and technological constraints of conventional electrode in MIM structured futuristic resistive switching devi-
charge storage memory.2 In particular, Resistive random ces. These alloys show martensite transformation which is a
access memory (ReRAM) based on the resistive switching first order structural transformation from the high temperature
phenomenon, is a promising candidate for futuristic memory phase (austenite) to other low temperature phases (martens-
devices due to its high write/erase speed, low power consump- ite).17 They will provide extra degrees of freedom (temperature,
tion, large data storage density, complementary metal oxide stress, etc.) to tune the resistive switching characteristics of the
semiconductor (CMOS) compatibility, and simple structure.3–5 concerned device. There are few reports which explore alloy
Resistive switching is the reversible change in the resis- thin films as an electrode in MIM structured switching
tance of the metal-insulator-metal (MIM) structure between the devices.18,19 Furthermore, the dependence of the crystal struc-
high resistance state (HRS) and low resistance state (LRS) stim- ture, work function, and resistivity of the Ni-Mn-In alloy thin
ulated by the applied electric field. The formation and disrup- film on its processing parameters such as elemental composi-
tion of the conducting filament in the active resistive layer of tion, deposition temperature, thickness, etc., provides the oppor-
the MIM structure is responsible for the resistive switching tunity to optimize the resistive switching characteristics of the
behavior.6,7 The resistive switching characteristics of the MIM memory device with Ni-Mn-In as a bottom electrode material.
structured device are influenced by various parameters.8–10 The The resistive switching phenomenon has been observed in
deposition conditions of the active layer play a crucial role on a number of materials such as transition metal oxides, perov-
its resistive switching behavior because grain boundaries, skite oxides, ferroelectrics, metal nitrides, two dimensional
vacancies, phase, and crystallinity of the active layer are signifi- materials etc.20–23 Recently, resistive switching in molybde-
cantly altered by the growth conditions.11,12 Moreover, the num disulphide (MoS2) thin films is also reported.24,25 Wang
MIM structured devices having an active switching layer with et al. observed photo-resistive switching in MoS2 nano-
different metal electrodes, also exhibit variation in their spheres.26 The MoS2 is a two dimensional material having a
stacked layered structure with intralayer strong covalent bond-
a)
Author to whom correspondence should be addressed: [email protected]. ing and interlayer weak van der Waals interactions with unique
Tel.: þ91-1332-285407. Fax: þ91-1332-273560 electrical and optical properties.27 Although considerable

0003-6951/2018/112(26)/262106/5/$30.00 112, 262106-1 Published by AIP Publishing.

262106-2 Kumar et al. Appl. Phys. Lett. 112, 262106 (2018)

contributions have been devoted to use MoS2 thin films for formed by depositing the top electrode dots of copper on
various electronic applications, resistive switching in MoS2 MoS2/Ni-Mn-In nanostructure thin films using a shadow
thin films is yet to be completely explored.28–32 It would be mask. The room temperature grazing incident angle X-ray
interesting and fruitful to further examine resistive switching diffraction (GI-XRD) pattern was collected by a Bruker
in MoS2 based memristors and describe the mechanism advanced D-8 diffractometer equipped with Cu Ka radiation
responsible for it. Therefore, we have prepared a device having (1.54 Å) for phase identification. The cross sectional mor-
a Cu/MoS2/Ni-Mn-In configuration and systematically investi- phology and thickness of Ni-Mn-In/MoS2 was examined
gated its resistive switching characteristics. The temperature using a Ziess field emission scanning electron microscope
control of the resistive switching parameters of the device has (FESEM). The electrical measurements of the device were
been examined. A conceptual model is also proposed and dis- performed by a Keithley semiconductor characterization sys-
cussed to elaborate the responsible mechanism for the resistive tem (SCS 4200).
switching behavior of the fabricated device. The operational Figure 1(a) shows the room temperature X-ray diffrac-
reliability of the fabricated device is examined in terms of its tion (XRD) pattern of the sputtered deposited Ni-Mn-In thin
endurance and retention characteristics. film on the silicon substrate. The diffraction peaks at 42.9
The Cu/MoS2/Ni-Mn-In devices were fabricated on the and 50.1 correspond to the (220) and (311) reflections of
p-Si (100) substrates by the D.C. magnetron sputtering tech- the austenite phase of the Ni-Mn-In alloy. The presence of
nique using commercial targets of Ni50Mn35In15, MoS2, and (220) fundamental reflection along with the (311) superlat-
copper. Before deposition, the 2  1 cm2 Si substrates were tice peak confirm the L21 structured austenite phase of the
cleaned by the standard RCA cleaning technique to remove Ni-Mn-In thin film at room temperature.33 To determine the
the organic contaminants and oxide layers from the top structural transition temperatures, the resistance vs tempera-
surface of the substrates. First, the Ni-Mn-In underlayer of ture measurements of the Ni-Mn-In thin film were performed
120 nm is uniformly deposited on the cleaned Si substrates during both heating and cooling modes using four probe
with a sputtering power of 100 W at a temperature of 500  C arrangement [inset Fig. 1(a)]. There is a drastic decrease in
in a pure argon atmosphere with 10 mTorr pressure. The electrical resistance at 292 K during the heating mode and a
MoS2 thin films were fabricated over these pre-deposited Ni- sharp increase in resistance at 290 K during the cooling
Mn-In coated Si substrates using the D.C. magnetron sputter- mode. This thermal hysteresis is recognized due to the first
ing technique. The bottom electrode is made available by order martensite transformation. The sudden decrease in the
partially masking the Ni-Mn-In layer during MoS2 deposi- electrical resistance during the martensite phase transition
tion. For MoS2 thin film fabrication, the substrate tempera- can be ascribed to the super-zone boundary effect triggered
ture, power, and argon working pressure were kept fixed at by structural and magnetic phase transformation.34 It causes
200  C, 50 W, and 10 mTorr, respectively. For electrical alteration in the electronic band structure near the Fermi sur-
characterization of the device, a capacitor like structure is face. The extracted values of martensite start (Ms), martensite

FIG. 1. (a) X-ray diffraction (XRD)

pattern of the Ni-Mn-In thin film. The
inset shows the resistance vs tempera-
ture measurement of the Ni-Mn-In thin
film upon heating and cooling. (b) The
XRD pattern of the MoS2 thin film
deposited over the Ni-Mn-In layer. (c)
Raman Spectra of the MoS2 thin film.
The inset depicts the cross-sectional
FESEM image of the device. (d)
Atomic displacements in Raman vibra-
tion modes of the MoS2 thin film.
262106-3 Kumar et al. Appl. Phys. Lett. 112, 262106 (2018)

finish (MF), austenite start (As), and austenite finish (AF), 383 cm1 and 409 cm1, respectively. The atomic displace-
temperatures are nearly 290.5 K, 271.6 K, 292.4 K, and ments in the E12g and A1g Raman active modes are shown in
312.6 K, respectively. Figure 1(b) displays the X-ray diffrac- Fig. 1(d). The E12g mode involves the in-plane displacement of
tion pattern after depositing the MoS2 thin film over the pre- molybdenum and sulphur atoms while the A1g mode includes
deposited Ni-Mn-In layer on the silicon substrate. The out-of-plane motion of sulphur atoms only. The wavenumber
appeared reflections in the XRD pattern of MoS2/Ni-Mn-In spacing between the E12g and A1g Raman vibration modes of
can be well classified into two categories: (i) the (002), (101), the deposited MoS2 thin film comes out to be 26 cm1 which
and (112) reflections corresponding to the MoS2 hexagonal is consistent with earlier reports on the multilayered MoS2
phase and (ii) the (220) reflection of the L21 structured aus- Raman spectra.35
tenite phase of Ni-Mn-In. No reflections corresponding to The experimental arrangement shaped for collecting the
impurity or secondary phases are detected in the XRD pattern resistive switching response of the Cu/MoS2/Ni-Mn-In
of the MoS2/Ni-Mn-In device. The cross-sectional FESEM device is demonstrated in Fig. 2(a). To study the resistive
image of the deposited sample shows a nearly clear interface switching characteristics, the current-voltage (I-V) measure-
between MoS2 and Ni-Mn-In layers [inset Fig. 1(c)]. The ments of the Cu/MoS2/Ni-Mn-In device were performed. A
thickness of the MoS2 and Ni-Mn-In layers in the bi-layered sweeping voltage in a sequence of 0 V ! 2 V ! 0 V !
nanostructure is 180 nm and 120 nm, respectively. To further 2 V! 0 V is applied on the top copper electrode while the
confirm the MoS2 phase formation, the Raman spectra of the Ni-Mn-In bottom electrode is always grounded. To avoid the
sample were recorded. Figure 1(c) shows the Raman spectra of permanent electrical breakdown of the device, the compli-
the fabricated MoS2 thin film at room temperature. It is ance current was set to 10 mA. The I-V curve of the fabri-
observed that the strong signals corresponding to in-plane E12g cated device shown in Fig. 2(b) exhibits the bipolar resistive
and out-of-plane A1g Raman vibration modes are present at switching behavior. The arrows describe the sweeping

FIG. 2. Schematic configuration of the

Cu/MoS2/Ni-Mn-In device. (b) The
current-voltage (I-V) characteristics of
the Cu/MoS2/Ni-Mn-In device. The
inset shows the electroforming curve.
(c) The ln I vs ln V plots analysis of
the fabricated device in HRS and LRS.
(d) Temperature dependence of resis-
tive switching parameters. (e) LRS and
HRS resistances as a function of cell
size area. (f) The reset current (Ireset)
as a function of compliance current
(Icomp).
262106-4 Kumar et al. Appl. Phys. Lett. 112, 262106 (2018)

direction of the applied voltage. Initially, the as fabricated slight dependency of the LRS resistance on the cell size vali-
pristine device was in the high resistance state (HRS). dates the existence of a localized metallic filament.
During the electroforming process, an abrupt increase in the The dependence of the maximum reset current (Ireset,
current is observed at 3.4 V and the device has been toggled reset current of LRS at reset voltage) on the compliance cur-
from the high resistance state (HRS) to the low resistance rent (Icomp) is also examined [Fig. 2(f)]. For each compliance
state (LRS) [inset Fig. 2(b)]. Subsequent to the complete current value, the Ireset is measured in 4 SET-RESET cycles.
electroforming process, the device switched from HRS to It depicts that the Ireset current approximately linearly
LRS during the positive biasing at the transition voltage of increases with the increase in Icomp current. It can be ascribed
1.36 V known as the SET voltage which is lower than the to the fact that higher compliance current results in the for-
electroforming voltage. In the voltage regime from 2 V to mation of a stronger conducting filament across the MoS2
0 V, the device remained in the LRS. During the negative layer. Therefore, more energy or larger reset current is
voltage sweep, the current suddenly decreased at 0.98 V required to rupture this strong filament.23 In this way, the
and resulting in the LRS to HRS transition, named as the magnitude of reset current can be controlled by the compli-
RESET process. These two resistance states (LRS and HRS) ance current. It can be concluded that the power consump-
with set and reset voltages of 1.36 V and 0.98 V, respec- tion by the device can be suppressed by lowering the
tively, act as the two logic states (ON and OFF) for the mem- compliance current.
ory application. The temperature and cell area dependence of HRS and
In order to clarify the electrical conduction properties LRS resistances clearly demonstrates that the widely
and switching mechanism in the Cu/MoS2/Ni-Mn-In device, accepted localized copper conductive filamentary mecha-
the current-voltage (I-V) curves have been replotted in dou- nism is responsible for the resistive switching behavior in
ble logarithm scales for HRS as well as LRS in the positive the Cu/MoS2/Ni-Mn-In device.14,23 When the electrochemi-
voltage region [Fig. 2(c)]. The double logarithm plot for cally active copper (Cu) top electrode of the freshly prepared
LRS shows the ohmic behavior with a slope of 1.08 due to device is positively biased [Fig. 3(a)], then Cu metal atoms
conductive filament formation. On the other hand, the I-V got oxidized to Cu ions (Cu ! Cunþ þ n e) and started
characteristics in HRS are relatively more complicated and drifting towards the bottom electrode along the electric field
direction via grain boundaries or crystal defect in the MoS2
can be classified into two regions. The I-V curve of HRS in
nanosheets. These Cu ions reduced to Cu atoms (Cunþ þ n
the low voltage region (0 < V < 0.5) obeys ohmic conduc-
e ! Cu) when they reached at the bottom electrode.23 This
tion (V-I) and in the high voltage region (V > 0.5) exhibits
accumulation of the Cu atoms continues until a conductive
nonlinearity with a slope of 2.1. Therefore, the space
path is established between the two electrodes and causes a
charge limited conduction (SCLC) is the dominating conduc-
transition from HRS to LRS [Fig. 3(b)]. This phenomenon is
tion mechanism in the HRS region. Furthermore, the temper-
known as electroforming and the corresponding voltage is
ature dependent resistances in LRS and HRS are also
called the forming voltage. After completion of electroform-
investigated in the temperature range of 220 K to 320 K [Fig.
ing operation, the voltage required to form the filament is
2(d)]. The resistance in HRS exhibits negative dependence
called the SET voltage and it is lower than the electroform-
on temperature which is a typical semiconductor behavior.
ing voltage because the conduction filament is not entirely
In contrast, the resistance in LRS increases with increase in dissolved after RESET operation. As the negative voltage is
temperature which confirms the presence of the metallic fila- applied to the top electrode, this conductive Cu filament rup-
ment in the LRS of the device.14 In addition to this, the SET tured from its weakest part by a combined effect of the elec-
voltage shows an increasing trend with the decreasing tem- tric field driven electrochemical dissolution and local Joule
perature [inset Fig. 2(d)]. It could be ascribed to the first heating process. In this way, the device returned back to the
order martensite transformation from the high temperature HRS again [Fig. 3(c)].
austenite phase to the low temperature martensite phase. The The reliability of the Cu/MoS2/Ni-Mn-In device is
martensite transformation generates a temperature dependent investigated in terms of its endurance and retention charac-
strain in the Ni-Mn-In electrode which transfers to the upper teristics which are crucial parameters for practical applica-
MoS2 thin film and produces some straggling in the bottom tion of resistive switching memory devices [Figs. 3(d) and
few layers of the MoS2 thin film. This straggling may cause 3(e)]. The device was attributed to a pulsed voltage of 2 V
spreading of the metal filament from its bottom part and for 100 ns during SET operation and 2 V for 100 ns during
enhancing the depletion gap between the disconnected por- RESET operation to examine its endurance and retention
tions of the filament. It leads to an increase in the SET volt- properties at room temperature. The resistances correspond-
age of the device with the decrease in the temperature. ing to HRS and LRS were measured at a readout voltage of
Therefore, the conducting filament through the MoS2 thin 0.1 V. The resistances in HRS and LRS are about 5.6 kX and
film can be controlled by varying its temperature. In addi- 15.4 X, respectively, which results in a good ON/OFF resis-
tion, the resistances in LRS and HRS are also analyzed as a tance ratio of 3  102. The endurance characteristics of the
function of cell size [Fig. 2(e)]. As the cell area decreases device show consistent switching performance for 500 cycles
from 105 lm2 to 103 lm2, the resistance of HRS increases without significant degradation. The retention measurements
from 103 to 104 X whereas the LRS resistance shows a little reveal that each storage state is highly stable up to 103 s with
variation. The stronger dependence of HRS resistance on the a little bit deviation.
top electrode area indicates that current in the device flow In conclusion, the Cu/MoS2/Ni-Mn-In device was
through the whole cell area in HRS. On the other hand, the fabricated using the D.C. sputtering technique for
262106-5 Kumar et al. Appl. Phys. Lett. 112, 262106 (2018)

FIG. 3. (a)–(c) Schematic representa-

tion of the proposed analytic copper
filament model. (d) Endurance and (e)
retention characteristics of the Cu/
MoS2/Ni-Mn-In device with a read
voltage of 0.1 V, respectively.

10
non-volatile resistive switching memory applications. The J. Hou, S. S. Nonnenmann, W. Qin, and D. A. Bonnell, Adv. Funct. Mater.
Cu/MoS2/Ni-Mn-In device shows bipolar resistive switch- 24, 4113 (2014).
11
N. Ghenzi, M. J. Rozenberg, R. Llopis, P. Levy, L. E. Hueso, and P.
ing behavior with an On/OFF resistance ratio of 366, Stoliar, Appl. Phys. Lett. 106, 123509 (2015).
endurance of 500, and retention of 103 s. The device shows 12
R. Zazpe, M. Ungureanu, F. Golmar, P. Stoliar, R. Llopis, F. Casanova, D.
ohmic conduction behavior in LRS due to metallic fila- 13
F. Pickup, C. Rogero, and L. E. Hueso, J. Mater. Chem. C 2, 3204 (2014).
ment formation while space charge limited current S. C. Oh, H. Y. Jung, and H. Lee, J. Appl. Phys. 109, 124511 (2011).
14
R. Prakash, B. P. S. Rathore, and D. Kaur, J. Alloys Compd. 726, 693
(SCLC) is the dominating conduction mechanism in the (2017).
HRS region. The resistive switching behavior of the fabri- 15
D. P. Sahu and S. N. Jammalamadaka, Sci. Rep. 7, 17224 (2017).
16
cated device is well explained by the formation/rupture of 17
B. Sun, Y. Liu, W. Zhao, and P. Chen, RSC Adv. 5, 13513 (2015).
the Cu filament across the MIM structure. The introduc- E. Faran and D. Shilo, Exp. Tech. 40, 1005 (2016).
18
J. C. Wang, D. Y. Jian, Y. R. Ye, L. C. Chang, and C. S. Lai, J. Phys. D:
tion of Ni-Mn-In alloys as a bottom electrode provides the Appl. Phys. 46, 275103 (2013).
opportunity to tune the MoS2 resistive switching charac- 19
M. Kawai, K. Ito, and Y. Shimakaw, Appl. Phys. Lett. 95, 012109
teristics by the temperature parameters because the strain 20
(2009).
generated during martensite transformation affects the M. Li, F. Zhuge, X. Zhu, K. Yin, J. Wang, Y. Liu, C. He, B. Chen, and R.
W. Li, Nanotechnology 21, 425202 (2010).
copper filament present in the MoS2 active layer. These 21
B. P. S. Rathore, R. Prakash, and D. Kaur, Curr. Appl. Phys. 18, 102
results demonstrate the potential of the Cu/MoS2/Ni-Mn- (2018).
22
In device for futuristic tunable non-volatile memory P. Han, B. Sun, J. Li, T. Li, Q. Shi, B. Jiao, Q. Wua, and X. Zhang,
device applications. J. Colloid Interface Sci. 505, 148 (2017).
23
C. Chen, Y. C. Yang, F. Zeng, and F. Pan, Appl. Phys. Lett. 97, 083502
The financial support provided by the Department of (2010).
Science and Technology (DST), India, under the Nano 24
X. Y. Xu, Z. Y. Yin, C. X. Xu, J. Dai, and J. G. Hu, Appl. Phys. Lett. 104,
Mission with Reference No. SR/NM/NT-1065/2015 (G) is 033504 (2014).
25
highly acknowledged. The author, Anuj Kumar, is highly P. Cheng, K. Sun, and Y. H. Hu, Nano Lett. 16, 572 (2016).
26
W. Wang, G. N. Panin, X. Fu, L. Zhang, P. Ilanchezhiyan, V. O.
thankful to the University Grants Commission (UGC), India, Pelenovich, D. Fu, and T. W. Kang, Sci. Rep. 6, 31224 (2016).
for the award of the Senior Research Fellowship (SRF). 27
G. Zhang, H. Liu, J. Qua, and J. Li, Energy Environ. Sci. 9, 1190 (2016).
28
M. Acerce, D. Voiry, and M. Chhowalla, Nat. Nanotechnol. 10, 313
1 (2015).
G. I. Meijer, Science 319, 1625 (2008).
2 29
D. S. Jeong, R. Thomas, R. S. Katiyar, J. F. Scott, H. Kohlstedt, A. Y. Huang, J. Guo, Y. Kang, Y. Ai, and C. M. Li, Nanoscale 7, 19358
Petraru, and C. S. Hwang, Rep. Prog. Phys. 75, 076502 (2012). (2015).
3 30
R. Waser and M. Aono, Nat. Mater. 6, 833 (2007). D. Bartolomeo, L. Genovese, T. Foller, F. Giubileo, G. Luongo, L. Croin,
4
D. Ielmini, Semicond. Sci. Technol. 31, 063002 (2016). S. J. Liang, L. K. Ang, and M. Schleberger, Nanotechnology 28, 214002
5
N. Singh, K. Singh, and D. Kaur, Ceram. Int. 43, 8970 (2017). (2017).
6 31
L. Yang, C. Kuegeler, K. Szot, A. Ruediger, and R. Waser, Appl. Phys. J. Kang, W. Liu, and K. Banerjee, Appl. Phys. Lett. 104, 093106 (2014).
32
Lett. 95, 013109 (2009). A. D. Bartolomeo, L. Genovese, F. Giubileo, L. Iemmo, G. Luongo, T.
7
Y. B. Zhu, K. Zheng, X. Wu, and L. K. Ang, Sci. Rep. 7, 43664 (2017). Foller, and M. Schleberger, 2D Mater. 5, 015014 (2018).
8 33
Y. C. Junga, S. Seonga, T. Leea, S. Y. Kima, I. S. Park, and J. Ahn, Appl. K. Singh and D. Kaur, Appl. Phys. Lett. 107, 262901 (2015).
34
Surf. Sci. 435, 117 (2018). H. S. Akkera, I. Singh, and D. Kaur, J. Alloys Compd. 642, 53 (2015).
9 35
C. Hu, Q. Wang, S. Bai, M. Xu, D. He, D. Lyu, and J. Qi, Appl. Phys. C. Jiang, S. L. Rumyantsev, R. Samnakay, M. S. Shur, and A. A.
Lett. 110, 073501 (2017). Balandin, J. Appl. Phys. 117, 064301 (2015).