close

Вход

Забыли?

вход по аккаунту

?

pssa.201700440

код для вставкиСкачать
Metal Nanocrystals
solidi
status
physica
ORIGINAL PAPER
a
www.pss-a.com
Tuning Resistive, Capacitive, and Synaptic Properties of
Forming Free TiO2-x-Based RRAM Devices by Embedded
Pt and Ta Nanocrystals
Panagiotis Bousoulas,* Ismini Karageorgiou, Vaggelis Aslanidis,
Kostas Giannakopoulos, and Dimitris Tsoukalas
Dr. P. Bousoulas, I. Karageorgiou, V. Aslanidis, Prof. D. Tsoukalas
Department of Applied Physics
National Technical University of Athens
Iroon Polytechniou 9 Zografou, 15780 Athens, Greece
E-mail: [email protected]
CFs. CFs may consist of either oxygen
vacancies/metal precipitates[2] or a metallic
chain of electrochemically active electrode
metal.[3] Nevertheless, practical characteristics of the CFs, such as direction of growth,
composition analysis, ruptured region, etc.,
are difficult to be extracted from the above
switching conjectures. Thus, sophisticated
experiments have been performed in order
to shed light on the atomic structure and
ingredient of the CFs in the low resistance
state (LRS) and high resistance state (HRS)[2]
as well as impedance spectroscopy techniques.[4] Despite of the exact nature of the
CFs, the major issue of resistive random
access memory (RRAM) technology is the
inherit variability of the switching characteristics, which is directly connected with the randomness of the CFs
formation/annihilation. While it seems practically impossible
to control their characteristics, we can force them to evolve
into specific locations within the device active core.[5,6] Thus, a
variety of optimization procedures have been suggested in
order to enhance the memory performance,[7,8] while NCs
incorporation arising as one very promising technique to
overcome the switching variations,[9–11] especially when the
issues related with the uniformity of their distribution, in
terms of diameter and surface density, will be efficiently
harnessed.[12]
In this work, we demonstrate that a wide range of non-volatile
memory properties can be affected and improved by embedded
Pt and Ta NCs. The concentrated electric field effect in
combination with the charge trapping effect in the NCs, are
regarded as the driving forces for the recorded switching
patterns. This possibility to tune the resistance levels over several
orders of magnitude is not usual to conventional materials and
indicates the efficiency of NCs to influence the charge transfer
properties.[13] In addition, insights about the origin of an
intriguing phenomenon, such as capacitance switching,[14] are
provided.
Dr. K. Giannakopoulos
Institute of Nanoscience and Nanotechnology
NCSR “Demokritos”
Aghia Paraskevi, 15310 Athens, Greece
2. Experimental
The incorporation of metal nanocrystals (NCs) within TiO2-x thin films
offers great advantages for adjusting a wide range of non-volatile memory
properties, ranging from resistive and capacitive switching to synaptic
capabilities. In this study, it is demonstrated that by inserting very small
NCs (3 nm diameter) of either Pt or Ta, resistance changes over six
orders of magnitude and capacitance changes over two orders of
magnitude can be induced, with promising variability due to the local
enhancement of the electric field effect, while these effects are attributed
to the energy band diagram configuration induced by the presence of
NCs. The gradual switching pattern observed exhibits also attractive
synaptic properties, offering higher design flexibility for neuromorphic
applications.
1. Introduction
Resistive switching memories based on metal oxides are
emerging as a new research field and at the same time are
intensively studied as one of the most promising candidates for
future non-volatile memory applications. However, in many
aspects their development has surpassed their understanding,
arising thus questions about the underlying nature of the
switching effect and the related uniformity issues. While several
phenomenological models have been devised in order to
elucidate on the origins of the switching effect, the conducting
filament (CF)-based model is gaining significant ground.[1]
Following this assumption, the data storage takes place at least
two resistance states provided by the generation/rupture of the
The ORCID identification number(s) for the author(s) of this article
can be found under https://doi.org/10.1002/pssa.201700440.
DOI: 10.1002/pssa.201700440
Phys. Status Solidi A 2017, 1700440
The structure of the RRAM devices was the following: TiN
(40 nm)/Ti (4 nm)/TiO2-x (22.5 nm)/Pt or Ta NCs/TiO2-x
(22.5 nm)/Au (40 nm)/SiO2/Si, while the fabrication details
1700440 (1 of 5)
© 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
solidi
www.pss-a.com
and the impact of the size and density of NCs can be found in our
previous works.[15,16] The TiO2-x (22.5 nm) was deposited
through reactive RF magnetron sputtering of a high purity
TiO2 ceramic target (99.9%) at room temperature under low
oxygen content (20%) and power of 150 W, while Pt or Ta NCs
were deposited by DC magnetron sputtering of a Pt or Ta target
(99.9% purity). Reference sample with 45 nm thickness and
same oxygen content was also fabricated. The top electrodes (TE)
were square in shape with 100 mm lateral dimension. Electrical
characterization was carried out by applying all signals to the TE,
keeping the BE grounded, by using a Keithley 4200 SCS and an
Agilent 4284A LCR meter, while Transmission Electron
Microscopy (TEM) measurements were employed in order to
examine the structural properties of the NCs, as can be seen in
Figure 1(a and b). The two NCs types are created at about the
same conditions, while a slightly different argon flow rate was
used, in order to obtain similar surface density. The diameter of
the NCs, which is mainly affected from the length of the
aggregation zone,[17] is depicted at the histograms of Figure 1(c
and d), where we present the statistical analysis of their lateral
size distributions. The single peaks in the distributions indicate
the dominant cluster population. Due to the complexity of the
reactions that take place during sputter deposition,[18] it is not
trivial to obtain absolute control of the NC distribution, since
local temperature, density of ions, and collision effects can
influence the total NC growth. This may be attributed to the wellknown absence of total control of the sputtering deposition
process mainly due to the large number of parameters that affect
this (e.g., local temperature, density of ions, collision effects) as
well as the complexity of the reactions that take place during
sputter deposition.
3. Results and Discussion
Figure 2(a) illustrates the resistance–voltage (R–V) characteristics for the reference sample as well as for the NCs-embedded
Figure 1. TEM plan view images of (a) Pt NCs and (b) Ta NCs and (c and
d) corresponding size distributions of the deposited NCs. In the TEM
images, the scale bar corresponds to 20 nm.
Phys. Status Solidi A 2017, 1700440
status
physica
www.advancedsciencenews.com
a
Figure 2. (a) R–V and (b) capacitance–voltage (C–V) characteristics of the
reference and the NCs embedded samples, exhibiting bipolar resistive
and capacitive switching behaviors. The arrows in the graph signify the
switching direction.
samples during the first switching cycle, by sweeping the bias
amplitude between 4 V, with a 200-mV step, revealing the
existence of a switching effect with gradual transitions for both
the SET/RESET processes and a different response for the
reference and the NCs embedded samples. It is worth
mentioning that the observed switching does not require any
forming procedure prior to operation and also that the obtained
characteristics exhibit high nonlinearity that shows potential for
device exploitation as an electronic synapse, as will demonstrate
later. However, the most striking feature is the observed
capacitive switching that exhibits ratios of 1–2 orders of
magnitude between a high capacitance state (HCS) and low
capacitance state (LCS) when sweeping the voltage bias
within 4 V range, with 20 mV step at 1 MHz AC frequency
(Figure 2(b)), which is a great improvement of the memory
margin with respect to those reported in literature.[4,19]
As it can be observed from Figure 2(a and b), the resistive and
capacitive switching are anti-correlated, i.e., the resistive
transition from the LRS into the HRS corresponds to the
capacitive transition from the HCS into the LCS and vice versa.
The presence of NCs will significantly enhance the local electric
field, leading to the creation of even larger number of oxygen
vacancies and as a result larger values of LRS and HCS, due to
charge separation effect.[15] Although, we cannot rule out the
possibility of cation contribution to filament formation,
the existence of capacitance hysteresis spectra indicate that
the nature of the filament is semiconducting, taking into
consideration that in previous studies[20] a metallic filament at
LRS was associated with no measured capacitance hysteresis. In
addition, the temperature dependence measurements (not
shown here) corroborate the semiconductor nature of LRS.
Thus, the capacitance changes can be also correlated with
charging effects of oxygen vacancies and NCs during electron
injection into the CF. Concerning the different responses of the
two NCs embedded samples, the oxidation of Ta and the
consequent decrease of the concentrated electric field effect
could justify the lower operating current values. This is in line
with the configuration of the energy levels, depicted in Figure 3,
where it is clear that the energy band alignment between the
energy levels of oxygen vacancies within TiO2-x and Pt NCs favors
the electron transfer,[21] taken for granted that NCs act as seeds
for CF creation[2,9–11] and electrons are the main constituents of
the CFs.[22] On the other hand, neither Ta nor TaOx NCs can
facilitate electron transfer, since large barriers are created.
1700440 (2 of 5)
© 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
solidi
status
physica
www.advancedsciencenews.com
a
www.pss-a.com
Indeed, if we consider the following expression for the
transmission probability using the Wentzel–Kramers–Brillouin
approximation[23]:
( pffiffiffiffiffiffiffiffiffi
)
3
4 2m T ¼ exp e Eb qe E x Et 2 ;
3 qe Eh
Figure 3. Energy band diagrams for the proposed stacks.
we can deduce that a relatively large interface barrier (1.05 eV for
the Ta NCs-TiOx and 2.05 eV for the TaOx-TiOx) reduces
dramatically the tunneling events and consequently the
operating current values.[24] The WKB approximation is widely
used to calculate the transmission probability for electron
tunneling, while it provides the possibility for more accurate
prediction by taking into account the phono-assisted processes.
The above solution is valid for trapezoidal-like barrier in the case
of a linear electrical potential is applied and without considering
Figure 4. Cumulative distribution functions (CDF) of (a) temporal and (b) spatial characteristics, (c) retention measurements at room and elevated
temperatures and (d) endurance responses under the application of 3 V/100 ns pulses. Programming with bigger pulses, in terms of amplitude, results
in better retention performance of the 3 nm Pt NCs sample.[16] The filled symbols correspond to the LRS and the open to the HRS, while a read voltage of
1 V was employed in all cases, (e) detailed pulsing scheme for the performed experiments.
Phys. Status Solidi A 2017, 1700440
1700440 (3 of 5)
© 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
solidi
status
physica
www.advancedsciencenews.com
a
www.pss-a.com
Figure 5. Measured conductance changes at (a) reference sample, (b) 3 nm Ta NCs sample, and (c) 3 nm Pt NCs under the application of 5 V pulses
with 10 ms width and repetition intervals, (d) detailed pulsing scheme for the STDP experiments.
the image force. In the above expression, h
is the reduced
Planck’s constant, me is the effective mass, E is the local electric
field, x is the distance among trap centers, qe is the elementary
charge of electrons, Eb the interface barrier, and Et the trap
energy below the conduction band (0.3 eV for oxygen
vacancies).[25] We remark that NCs charging has been also
confirmed experimentally in similar structures.[26,27] This
interpretation could also account for the weak capacitance
switching of 3 nm Ta NCs samples, since a lower injected current
will impose smaller capacitance values. As a result, the amount
of the injection/removal of electrons seems to be able to impact
both resistive and capacitive properties.[28] For that reason a selfcompliance behavior was observed with 3 nm Ta NCs sample, in
contrast with the sample containing Pt NCs, where a compliance
current (Icc) of 10 mA was enforced. In addition, series resistance
effects could justify this behavior, which was also observed for
the reference sample.[29] Alternatively, capacitive changes have
been attributed to gap modulation effect[4,13] which, however,
needs to change substantially (30 times) in order to explain our
experiments, while CFs in multiple configuration is considered
is incompatible with the area dependence characteristics, which
imply the formation of limited number of CFs.[33]
In order to further characterize the devices, the distribution of
cycle-to-cycle (temporal) and device-to-device (spatial) characteristics were investigated. The results are presented in Figure 4(a
and b), while the data have been collected from 100 consecutive
cycles and 150 different cells, respectively. A lower Icc (104 A)
was used for 3 nm Pt NCs sample in order to improve the
distribution of the LRS, while no Icc was enforced for the other
samples. A clear improvement of the coefficient of variance (σ/m)
can be discerned as regards LRS, while for the HRS in the cycling
operation some issues arise for the 3 nm Pt NCs sample
retaining, however, an effective switching ratio in the range 105.
The origin for these fluctuations seem to be the generation of
oxygen vacancies within the broken region of the CF or the
partial rupture of an already existing one.[30] Application of even
lower Icc improves this effect.[16] Retention measurements were
also performed at room and elevated temperatures (Figure 4(c)),
revealing the distinguishing nature of the two resistance states
Phys. Status Solidi A 2017, 1700440
for all samples. The fluctuations that are recorded for the HRS of
Pt NCs sample can be ascribed to trapping/detrapping of
electrons within the oxygen vacancies in the broken region of the
CF. On the other hand the LRS perturbations, which are
recorded for all samples, are due to the movement of oxygen
vacancies out of the CF, mainly caused by diffusion effect.[31] The
retention measurement were carried out by applying 3 V/
100 ns pulses, in order to impose either the SET or RESET
transition respectively, and then the resistance value was
monitored for about 105 s. As regards endurance performance
(Figure 4(d)), Pt NCs sample shows negligible degradation
following the application of a sequence of 107 SET/RESET
pulses, while for the reference sample the switching window
disappears after the same amount of pulses. The Ta NCs sample
exhibit some fluctuations, however, a switching ratio of 10 is
maintained in every case. The gradual switching pattern permits
also the demonstration of spike-timing dependent plasticity
(STDP) properties.[32] In order to assess the applicability of our
devices as synaptic learning modules, we measured the
conductance change under the application of a train of pulses,
with amplitude 5 V and width 10 ms, in order to induce the
potentiation/depreciation characteristics. The results, depicted
in Figure 5, divulge that all samples respond to the application of
each positive/negative pulses and more importantly, that
synaptic plasticity properties have been measured within six
orders of magnitude, while modulation of the CF diameter is
considered the driving force for this effect.[33] The small size of
the NCs allows their facile integration into smaller devices, in
terms of oxide thickness and electrode area, anticipating even
lower operating voltages and enhanced uniformity characteristics, since smaller electrode areas will enclose less NCs.
4. Conclusions
In conclusion, the incorporation of Pt and Ta NCs within TiO2-x
thin films results in a significant improvement of a variety of
switching parameters, such as the resistive, capacitive, and
synaptic properties. The large enhancement of the switching
1700440 (4 of 5)
© 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
solidi
www.pss-a.com
performance and in particular the enhancement of the measured
variability stems from the enforcement of the postulated
percolative networks, where the switching effect takes place.
The different degree of reactivity of the two materials against
oxidation as well as the formation of large energy barriers in the
case of Ta NCs, could interpret the measured data pattern.
Moreover, the small size of the NCs allows their facile integration
into smaller devices, in terms of oxide thickness and electrode
area, anticipating even lower operating voltages.
Acknowledgment
Two of the authors (P. Bousoulas and D. Tsoukalas) would like to
acknowledge financial support from Research Projects for Excellence IKY
(State Scholarship Foundation)/SIEMENS.
Conflict of Interest
The authors declare no conflict of interest.
Keywords
capacitance switching, nanocrystals, oxygen ion reservoir, oxygen
vacancies
Received: June 30, 2017
Revised: September 1, 2017
Published online:
[1] Y. Yang, W. D. Lu, IEEE Trans. Electron Dev. 2016, 15, 465.
[2] J.-Y. Chen, C.-W. Huang, C.-H. Chiu, Y.-T. Huang, W.-W. Wu, Adv.
Mater. 2015, 27, 5028.
[3] S. Menzel, R. Waser, Nanoscale 2013, 5, 11003.
[4] L. Qingjiang, A. Khiat, I. Salaoru, C. Papavassiliou, X. Hui,
T. Prodromakis, Sci. Rep. 2014, 4, 4522-1.
[5] M. Uenuma, B. Zheng, K. Kawano, M. Horita, Y. Ishikawa,
I. Yamashita, Y. Uraoka, Appl. Phys. Lett. 2012, 100, 083105-1.
[6] W. Banerjee, S. Maikap, C.-S. Lai, Y.-Y. Chen, T.-C. Tien, H.-Y. Lee,
W.-S. Chen, F. T. Chen, M.-J. Kao, M.-J. Tsai, J.-R. Yang, Nanoscale
Res. Lett. 2012, 7, 194-1.
[7] S. Kim, S. H. Choi, J. Lee, W. D. Lu, ACS Nano 2014, 8, 10262.
[8] K. M. Kim, J. Zhang, C. Graves, J. J. Yang, B. J. Choi, C. S. Hwang,
Z. Li, R. S. Williams, ACS Nano Lett. 2016, 16, 6724.
[9] J. H. Yoon, J. H. Han, J. S. Jung, W. Jeon, G. H. Kim, S. J. Song,
J. Y. Seok, K. J. Yoon, M. H. Lee, C. S. Hwang, Adv. Mater. 2013, 25,
1987.
Phys. Status Solidi A 2017, 1700440
status
physica
www.advancedsciencenews.com
a
[10] B. K. You, W. I. Park, J. M. Kim, K.-I. Park, H. K. Seo, J. Y. Lee,
Y. S. Jung, K. J. Lee, ACS Nano 2013, 8, 9492.
[11] S.-C. Qin, R.-X. Dong, X.-L. Yan, Appl. Phys. A 2014, 118, 605.
[12] A. J. M. Mackus, M. J. Weber, N. F. W. Thissen, D. G. Alonso,
R. H. J. Vervuurt, S. Assali, A. A. Bol, M. A. Verheijen,
W. M. M. Kessels, Nanotechnology 2015, 27, 034001-1.
[13] B. J. Choi, A. B. K. Chen, X. Yang, I-W. Chen, Adv. Mater. 2011, 23,
3847.
[14] A. A. Bessonov, M. N. Kirikova, D. I. Petukhov, M. Allen, T. Ryhänen,
M. J. A. Bailey, Nature Mater. 2015, 14, 199.
[15] P. Bousoulas, D. Sakellaropoulos, J. Giannopoulos, D. Tsoukalas,
“Improving the resistive switching uniformity of forming free TiO2-x
based devices by embedded Pt nanocrystals”, in Proc. of the 45th
European Solid State Device Research Conference (ESSDERC), 2015,
pp. 274–277.
[16] P. Bousoulas, S. Stathopoulos, D. Tsialoukis, D. Tsoukalas, IEEE
Electron Dev. Lett. 2016, 37, 874.
[17] E. Quesnel, E. P. Vaujour, V. Muffato, J. Appl. Phys. 2010, 107,
054309-1.
[18] I. Kolev, A. Bogaerts, IEEE Trans. Plasma Sci. 2006, 11, 886.
[19] I. Salaoru, A. Khiat, Q. Li, R. Berdan, T. Prodromakis, Appl. Phys. Lett.
2013, 103, 233513-1.
[20] X. L. Jiang, Y. G. Zhao, Y. S. Chen, D. Li, Y. X. Luo, D. Y. Zhao, Z. Sun,
J. R. Sun, H. W. Zhao, Appl. Phys. Lett. 2013, 102, 253507-1.
[21] V. Y. -Q. Zhuo, Y. Jiang, M. H. Li, E. K. Chua, Z. Zhang, J. S. Pan,
R. Zhao, L. P. Shi, T. C. Chong, J. Robertson, Appl. Phys. Lett. 2013,
102, 062106-1.
[22] K. Kamiya, M. Y. Yang, B. M. Köpe, M. Niwa, Y. Nishi, K. Shiraishi,
IEEE Trans. Electron Dev. 2013, 60, 3400.
[23] N. F. Mott, E. A. Davis, Electronic Processes in Non-Crystalline
Materials, Clarendon, Oxford, UK 1979.
[24] M. J. Yun, H.-D. Kim, S. M. Hong, J. H. Park, D. S. Jeon, T. G. Kim,
J. Appl. Phys. 2014, 115, 094305-1.
[25] M. Ramamoorthy, R. D. King-Smith, D. Vanderbilt, Phys. Rev. B 1949,
49, 7709.
[26] A. Gambardella, M. Prezioso, M. Cavallini, Sci. Rep. 2014, 4,
4196.
[27] G. Khurana, P. Misra, N. Kumar, S. Kooriyattil, J. F. Scott, R. S. Katiya,
Nanotechnology 2015, 27, 015702-1.
[28] M. Park, S. Park, K.-H. Yoo, ACS Appl. Mater. Interfaces 2016, 8,
14046.
[29] B. Gao, B. Chen, F. Zhang, L. Liu, X. Liu, J. Kang, H. Yu, B. Yu, IEEE
Trans. Electron Dev. 2013, 60, 1379-1.
[30] X. Guan, S. Yu, H.-S. P. Wong, IEEE Trans. Electron Dev. 2012, 59,
1172.
[31] D. Ielmini, F. Nardi, C. Cagli, A. L. Lacaita, IEEE Electron Dev. Lett.
2010, 31, 353.
[32] Z. Wang, S. Joshi, S. E. Savel’ev, H. Jiang, R. Midya, P. Lin, M. Hu,
N. Ge, J. P. Strachan, Z. Li, Q. Wu, M. Barnell, G.-L. Li, H. L. Xin,
R. S. Williams, Q. Xia, J. J. Yang, Nature Mater. 2016, 16, 101.
[33] P. Bousoulas, P. Asenov, I. Karageorgiou, D. Sakellaropoulos,
S. Stathopoulos, D. Tsoukalas, J. Appl. Phys. 2016, 120, 154501-1.
1700440 (5 of 5)
© 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Документ
Категория
Без категории
Просмотров
2
Размер файла
1 185 Кб
Теги
pssa, 201700440
1/--страниц
Пожаловаться на содержимое документа