I. INTRODUCTION
With the development of information and communication technology such as artificial
intelligence, internet of things, and autonomous driving, data usage has been rapidly
increasing, and faster computing systems have been required. The current computing
system is configured based on a memory hierarchy to allow a central processing unit
(CPU) to access lower-level memories. However, in addition to the memory wall issue
which means that the gap between CPU speed and memory speed increases, limitations
due to scaling of current memory technologies have been discussed [1]. Although dynamic random access memory (DRAM), which is used for the main memory,
exhibits a high speed of several tens of ns, periodic refreshing is required to prevent
data loss. In addition, as DRAM cells are scaled-down, the height of the capacitor
gradually increases to secure the minimum capacitance, which is also a challenging
issue. Furthermore, NAND flash, which is a non-volatile memory for data storage, is
also facing the limitation of scaling down. Charge-trapping flash memory enhanced
stable data storage ability compared to floating-gate flash, and the high-density
has been realized by stacking; however, due to the limitations of the etch technology,
the number of layers cannot be infinitely increased, and the operation speed is also
several hundreds of ${\mu}$s, which is a lot less than that of DRAM. To innovate the
current memory hierarchy, next-generation memories have been actively investigated.
The next-generation memories aim to have both the high speed of DRAM or static RAM
and the non-volatility of flash memory, and include resistive (RRAM) [2-7], phase-change RAM [8-10], spin-transfer torque magnetic RAM [11-14], and ferroelectric field-effect transistor [15-17].
In particular, RRAM is a prospective candidate for next-generation memory with a simple
metal-insulator-metal (MIM) structure. As the insulator material, single-layers such
as HfO$_{2}$ [18,19], Al$_{2}$O$_{3}$ [20,21], TiO$_{x}$ [22,23], or bi-layers such as Al$_{2}$O$_{3}$/TiO$_{x}$ [24-28], HfO$_{2}$/TiO$_{x}$ [29,30] and HfO$_{2}$/TaO$_{x}$ [31-33] are commonly used. Depending on the formation or rupture of the conductive filament
in the insulator layer, the resistance state changes to a low-resistance state (LRS)
or a high resistance state (HRS). RRAM has advantages such as fast read/write speed,
scalability of 4F$^{2}$, ease of 3D integration, and flexibility in material selection.
However, due to a sneak path problem, it is necessary to co-integrate an additional
selecting element such as a transistor or a selector for accurate cell access [34]. The integration of a transistor and an RRAM, so-called a 1T1R structure, has a cell
area of 6F$^{2}$, and its 3D integration is not an easy task. In addition, for a 1S1R
structure in which an MIM selector is directly stacked on an RRAM device, it is also
challenging to improve the reliability of the selector itself and to optimize operating
conditions compatible with an RRAM device. Therefore, it would be ideal to configure
a high-density passive array by allowing the RRAM device to have a selecting effect
on its own without an external device.
Therefore, in this study, an Al$_{2}$O$_{3}$/TiO$_{x}$ bilayer-based RRAM device is
fabricated and a method to enhance the nonlinear effect without the help of an external
device is proposed. We adjust TiO$_{x}$ deposition conditions to utilize the fact
that the electrical conductivity of TiO$_{x}$ dynamically changes depending on the
oxygen concentration. This study aims to quantitatively analyze the atomic components
of the TiO$_{x}$ layer and to investigate the corresponding effect on the nonlinearity
of the current-voltage (I-V) properties. Also, by constructing a passive crossbar
array based on the fitting parameters, we verify the effect of enhanced I-V nonlinearity
on a read margin through SPICE simulation.
II. DEVICE FABRICATION
Al$_{2}$O$_{3}$/TiO$_{x}$-based RRAM devices were fabricated with an MIM structure
as shown in Fig. 1(a). The insulator layer is composed of double layers of Al$_{2}$O$_{3}$ and TiO$_{x}$,
and a Ti/Pt stack was applied to both a bottom electrode (BE) and a top electrode
(TE). The detailed fabrication process is as follows: First, BE was formed with 2.5
${\mu}$m width on a SiO$_{2}$/Si substrate after initial wafer cleaning. After photolithography
with negative photoresist (PR), 10 nm of adhesion layer Ti and 37 nm of Pt were deposited
by e-beam evaporation. Afterward, the lift-off process was performed by removing PR
with acetone solution. Next, Al$_{2}$O$_{3}$/TiO$_{x}$ oxide layers were deposited.
Al$_{2}$O$_{3}$ was deposited with a thickness of 1.8 nm for 20 cycles by atomic layer
deposition (ALD), using trimethylaluminum (Al(CH$_{3}$)$_{3}$, TMA) as a precursor
and ozone (O$_{3}$) as a reactant, and the deposition temperature was 450$^{\circ}$C.
TiO$_{x}$ was deposited with a thickness of 27~nm by direct current reactive sputtering.
Nonstoichiometric TiO$_{x}$ is a material whose electrical properties such as resistivity
change from metallic to semiconducting depending on oxygen concentration [35]. Therefore, in this experiment, three samples with different oxygen concentrations
of TiO$_{x}$ were fabricated by controlling the O$_{2}$ gas flow rate. They are named
S1, S2, and S3, and discussed in detail later. Thereafter, the Ti/Pt TE stack was
deposited with 2.5 ${\mu}$m width using the same process with BE with a thickness
of 6~nm/45~nm. TE was patterned to cross vertically with BE, and thus crossbar-shaped
cells were fabricated. Finally, the oxide layers were dry-etched by CHF$_{3}$ gas
for electrical contact to electrodes. Electrical measurements of the fabricated RRAM
devices were performed using the Keysight B1500A instrument. The voltage was swept
in the positive or negative direction, and I-V characteristics were analyzed.
Fig. 1(b) shows a cross-sectional transmission electron microscopy image of the fabricated
RRAM. It is confirmed that the Al$_{2}$O$_{3}$ and TiO$_{x}$ layers were deposited
between TE and BE. The thickness of ALD-deposited Al$_{2}$O$_{3}$ was fixed as a control
condition, and X-ray photoelectron spectroscopy (XPS) analysis was performed to confirm
the change in TiO$_{x}$ composition according to the O$_{2}$ flow rate. For the XPS
measurement, monochromatic Al K-Alpha X-ray source (1486.7 eV) was used. Fig. 2 shows XPS results for TiO$_{x}$ layers of three devices, S1, S2, and S3. The binding
energy range for Ti atoms (450-470 eV) was mainly displayed [36], and four main peaks (Ti$^{0}$, Ti$^{2+}$, Ti$^{3+}$, and Ti$^{4+}$) according to
the oxidation number were specified by vertical dotted lines. As a result, it was
confirmed that the Ti peak compositions of the three samples were distinguished. As
the O$_{2}$ flow rate increases, the intensities of Ti$^{0}$ and Ti$^{2+}$ decrease
while those of Ti$^{3+}$ and Ti$^{4+}$ increase. This result implies that as the O$_{2}$
flow rate increases, the metallic component of TiO$_{x}$ decreases and the semiconducting
component increases. As a result of investigating the atomic ratio with XPS, the chemical
composition of TiO$_{x}$ for S1, S2, and S3 were x = 1.31, 1.52, and 1.74, respectively,
confirming that oxygen concentration increased by the increase in O$_{2}$ flow rate.
Note that we also tested the additional samples with Ti (corresponding to x = 0) and
TiO$_{\mathrm{1.85}}$ as buffer layers. For the device with Ti, it was confirmed that
the pristine state was leaky due to the oxygen scavenging effect of Ti. In addition,
for the TiO$_{\mathrm{1.85}}$ sample, abnormal resistive switching was observed after
the forming process. Therefore, we concluded that the samples with TiO$_{x}$ of 1.31
${\leq}$ x ${\leq}$ 1.74 can exhibit the resistive switching properties, and compared
their nonlinearities.
Fig. 1. (a) Schematic view and (b) cross-sectional transmission electron microscopic images of the fabricated RRAM.
Fig. 2. X-ray photoelectron spectroscopy (XPS) data for three devices with different oxygen concentrations of TiO$_{x}$: TiO$_{\mathrm{1.31}}$, TiO$_{\mathrm{1.52}}$, and TiO$_{\mathrm{1.74}}$.
III. NONLINEARITY MEASUREMENT
Fig. 3(a)-(c) show typical I-V characteristics of S1, S2, and S3, respectively. In each figure,
black, cyan, and blue lines correspond to forming, reset, and set operations, respectively.
All samples were electroformed through a voltage sweep up to 2.5 V at a compliance
current (I$_{\mathrm{c}}$) of 200 ${\mu}$A and switched from the initial state to
LRS. I$_{\mathrm{c}}$ was set appropriately to prevent device destruction during the
abrupt forming operations. Afterward, the devices were reset with a negative voltage
sweep to -1.3 V, resulting in the resistive switching from LRS to HRS. For set operations,
positive voltage sweeps were applied with I$_{\mathrm{c}}$ of 1 mA, and devices were
switched from HRS to LRS. All samples show bipolar resistive switching characteristics,
and it is confirmed that I-V nonlinearity becomes larger as the oxygen concentration
of TiO$_{x}$ increases. Note that the displayed data were measured after two weeks
after the fabrication, and there was no significant difference from the measured results
about two years after fabrication. This result suggests that the oxygen composition
of the TiO$_{x}$ layer is in a chemical equilibrium, and the stoichiometry is not
affected by oxygen in the air. The semi-log I-V curves of S1, S2, and S3 are also
displayed in Fig. 4 to specify rectification properties. Fig. 5 shows the forming voltage distributions of S1, S2, and S3, and the average values
are specified by the black dotted line. The average forming voltages of S1, S2, and
S3 were 1.980 V, 2.046~V, and 2.089 V, respectively. This is because the voltage across
TiO$_{x}$ increases, as the resistivity of TiO$_{x}$ increases. As suggested in the
previous work, the conductive filament is formed inside the Al$_{2}$O$_{3}$ layer,
and TiO$_{x}$ acts as an oxygen reservoir [35]. When the forming voltage is applied, the voltage is divided between the Al$_{2}$O$_{3}$
and TiO$_{x}$ layers. Thus, it can be inferred that the forming voltage is increased
because more voltage is divided to TiO$_{x}$ with higher oxygen concentration.
To compare the nonlinearity, the LRS I-V characteristics of S1, S2, and S3 were normalized
and fitted by the empirical model of the device [37]. Fig. 6 shows normalized curves for three devices, and fitted lines with the following equation
to quantify nonlinearity: $I=\alpha \times \sinh \left(\beta \times V\right),\,\,\left(\alpha
>0,\beta >0\right),$where ${\alpha}$ is a scaling factor and ${\beta}$ determines
I-V nonlinearity; the more linear the curve, the smaller ${\beta}$. As shown in Fig. 6, it is confirmed that ${\beta}$ values are 1.410, 2.375, and 2.875 in the order of
S1, S2, and S3. This result suggests the electrical properties of TiO$_{x}$, which
change from metallic to semiconducting conduction. As discussed earlier, the semiconducting
properties of the TiO$_{x}$ layer become dominant as the oxygen concentration increases.
In other words, Poole-Frenkel conduction becomes more dominant than Ohmic conduction,
which can increase the nonlinearity [38,39]. For comparison of nonlinearity, an almost linear curve (${\beta}$ = 0.01) is also
inserted in Fig. 6 as a black dotted line.
Fig. 3. Typical I-V characteristics for (a) S1; (b) S2; (c) S3. For each graph, forming, reset, and set switching curves are shown.
Fig. 4. Semi-log I-V switching characteristics for S1, S2, and S3.
Fig. 5. Forming voltage distributions of S1, S2, and S3.
Fig. 6. I-V nonlinearity fitting results for S1, S2, and S3. Normalized current and voltage for LRS were fitted using a hyperbolic sine function.
IV. READ MARGIN SIMULATION
To simulate the effect of nonlinearity on read margin, a passive crossbar array was
constructed as shown in Fig. 7. To analyze the minimum read margin in the worst case, the cell furthest from the
voltage source is selected (marked in blue), read voltage, V$_{\mathrm{read}}$, of
0.5 V is applied to the selected TE, and 0 V is applied to the selected BE. The rest
unselected cells were displayed in navy and set to LRS, and a grounding read scheme
was adopted to unselected TEs and BEs to improve read margin [40]. For realistic simulation, the cell-to-cell line resistance, R$_{\mathrm{line}}$,
was assumed to be 2 ${\Omega}$, which is a reasonable value because it can be realized
with a 100-nm-thick Pt electrode. After configuring the crossbar array circuit, the
read margin was defined as follows: $\frac{\left(I_{\mathrm{LRS}}-I_{\mathrm{HRS}}\right)}{I_{\mathrm{LRS}}}\times
100\left[\mathrm{\% }\right],$ where I$_{\mathrm{LRS}}$ and I$_{\mathrm{HRS}}$ are
the total read current in the selected BE when the selected cell is LRS and HRS, respectively.
Therefore, I$_{\mathrm{LRS}}$ and I$_{\mathrm{HRS}}$ include the current of the selected
cell and leakage current from half-selected and unselected cells.
Fig. 8 shows the simulation results of the read margin depending on the array size, N. Because
of the line resistance, the read margin decreases for all cases as the array size
increases. It can be seen that as the nonlinearity increases, it is more tolerant
to the reduction of the read margin. When N = 2$^{3}$ ${\times}$ 2$^{3}$, since the
array size is relatively small, high read margins are obtained for all samples. However,
as array size increases exponentially, the read margin decreases sharply. For example,
for N = 2$^{7}$ ${\times}$ 2$^{7}$, when ${\beta}$ = 0.01 (linear) and ${\beta}$ =
1.410 (S1), the read margin is 2.32% and 5.09%, respectively. On the other hand, for
${\beta}$ = 2.375 (S2) and ${\beta}$ = 2.875 (S3), the read margin is 14.02% and 22.97%,
respectively, achieving more than the minimum requirement of 10% [41]. In conclusion, by increasing TiO$_{x}$, nonlinearity in current conduction can be
increased and a high read margin for a large-scale memory array could be induced.
Fig. 7. Schematic circuit diagram of RRAM passive crossbar array for read margin simulation. It was configured considering cell-to-cell line resistance between cells, and the worst case is assumed. A grounding read scheme where unselected TEs and BEs are grounded is adopted.
Fig. 8. SPICE simulation results of the read margin depending on array size N.
V. CONCLUSIONS
In this study, we fabricated samples S1, S2, and S3 with different resistivity of
TiO$_{x}$ in Al$_{2}$O$_{3}$/TiO$_{x}$ bilayer-based RRAM devices. During the TiO$_{x}$
sputtering process, the oxygen concentration was changed by controlling the O$_{2}$
flow rate. The atomic compositions of TiO$_{x}$ for S1, S2, and S3 were found to be
TiO$_{\mathrm{1.31}}$, TiO$_{\mathrm{1.52}}$, and TiO$_{\mathrm{1.74}}$, respectively,
through XPS analysis. As the oxygen atomic ratio increases, Ti$^{3+}$ and Ti$^{4+}$
peaks become dominant, confirming that the semiconducting property of TiO$_{x}$ is
strengthened. As a result of fitting the I-V measurement data, the nonlinearity factor
${\beta}$ values for S1, S2, and S3 were investigated as 1.410, 2.375, and 2.875,
respectively. Finally, the effect of I-V nonlinearity in the passive crossbar array
was confirmed through SPICE simulation. As a result, at an array size of 2$^{7}$ ${\times}$
2$^{7}$, S1 device showed a read margin of 5.09%, which is insufficient for an accurate
read operation. On the other hand, 22.97% of the read margin for S3 device was achieved
despite assuming the worst case, suggesting that the high nonlinearity can effectively
suppress the sneak path current caused by unselected cells. If nonlinearity is optimized
by adjusting the thickness of Al$_{2}$O$_{3}$ and TiO$_{x}$ as future work, it is
expected that the RRAM passive array can be utilized as a next-generation storage
class memory.
ACKNOWLEDGMENTS
This work was supported by INHA UNIVERSITY Research Grant. The authors would like
to thank the BK21 FOUR Program. The EDA tool was provided by the IC Design Education
Center (IDEC), Korea.
References
McKee S. A., 2004., Reflections on the memory wall, in Proceedings of the 1st conference
on Computing frontiers, pp. 162
Pan F., Gao S., Chen C., Song C., Zeng F., 2014, Recent progress in resistive random
access memories: Materials, switching mechanisms, performance, Mater. Sci. Eng. R-Rep.,
Vol. 83, pp. 1-59
Wong H.-S. P., Lee H.-Y., Yu S., Chen Y.-S., Wu Y., Chen P.-S., Lee B., Chen F. T.,
Tsai M.-J., 2012, Metal-oxide RRAM, Proc. IEEE, Vol. 100, No. 6, pp. 1951-1970
Jeon K., Kim J., Ryu J. J., Yoo S.-J., Song C., Yang M. K., Jeong D. S., Kim G. H.,
2021, Self-rectifycing resistive memroy in passive crossbar arrays, Nat. Commun.,
Vol. 12, pp. 2968
Duan Y., Gao H., Guo J., Yang M., Yu Z., Shen X., Wu S., Sun Y., Ma X., Yang Y., 2021,
Effect of nitrogen capture ability of quantum dots on resistive switching characteristics
of AlN-based RRAM, Appl. Phys. Lett., Vol. 118, pp. 013501
Le B. Q., Levy A., Wu T. F., Radway R. M., Hsieh E. R., Zheng X., Nelson M., Raina
P., Wong H.-S. P., Wong S., Mitra S., 2021, RADAR: A fast and energy-efficient programming
technique for multiple bits-per-cell RRAM arrays, IEEE Trans. Electron Devices, Vol.
68, No. 9, pp. 4397-4403
hsieh E. R., Zheng X., Le B. Q., Shih Y. C., Radway R. M., Nelson M., Mitra S., Wong
S., 2021, Four-bits-per-memory one-transistor-and-eight-resistive-random-access-memory
(1T8R) array, IEEE Electron Devices Lett., Vol. 42, No. 3, pp. 335-338
Choi Y., Song I., Park M.-H., Chung H., Chang S., Cho B., Kim J., Oh Y., Kwon D.,
Sunwoo J., 2012, A 20nm 1.8 V 8Gb PRAM with 40MB/s program bandwidth, in 2012 IEEE
International Solid-State Circuits Conference, pp. 46-48
Sarwat S. G., 2017, Materials science and engineering of phase change random access
memory, Mater. Sci. Technol., Vol. 33, No. 16, pp. 1890-1906
Chen X., Ding F., Huang X., Lin X., Wang R., Chan M., Zhang L., Huang R., 2021, A
robust and efficient compact mode lfor phase-change memory circuit simulations, IEEE
Trans. Electron Devices, Vol. 68, No. 9, pp. 4404-4410
Khvalkovskiy A., Apalkov D., Watts S., Chepulskii R., Beach R., Ong A., Tang X., Driskill-Smith
A., Butler W., Visscher P., 2013., Basic principles of STT-MRAM cell operation in
memory arrays, J. Phys. D: Appl. Phys., Vol. 46, No. 7, pp. 074001
Apalkov D., Khvalkovskiy A., Watts S., Nikitin V., Tang X., Lottis D., Moon K., Luo
X., Chen E., Ong A., 2013, Spin-transfer torque magnetic random access memory (STT-MRAM),
ACM J. Emerg. Technol. Comput. Syst. (JETC), Vol. 9, No. 2, pp. 1-35
Pourmeidani H., Demara R. F., 2021, High-accuracy deep belief network: Fuzzy neural
networks using MRAM-based stochastic neurons, IEEE J. Explor. Solid-State Comput.
Devices Circuits, Vol. 7, No. 2, pp. 125-131
Na T., Kang S. H., Jung S.-O., 2020, Distribution analysis and multiple-point tail
fitting yield estimation method for STT-MRAM, J. Semicond. Technol. Sci., Vol. 20,
No. 3, pp. 271-280
Tan A. J., Chatterjee K., Zhou J., Kwon D., Liao Y.-H., Cheema S., Hu C., Salahuddin
S., 2019, Experimental demonstration of a ferroelectric HfO$_{2}$-based content addressable
memory cell, IEEE Electron Device Lett., Vol. 41, No. 2, pp. 240-243
Bae J.-H., Kwon D., Jeon N., Cheema S., Tan A. J., Hu C., Salahuddin S., 2020, Highly
scaled, high endurance, ${\Omega}$-gate, nanowire ferroelectric FET memory transistors,
IEEE Electron Device Lett., Vol. 41, No. 11, pp. 1637-1640
Nguyen M.-C., Lee K., Kim S., Youn S., Hwang Y., Kim H., Choi R., Kwon D., 2022, Incremental
drain-voltage-ramping training method for ferroelectric field-effect transistor synaptic
devices, IEEE Electron Device Lett., Vol. 43, No. 1, pp. 17-20
Lee D. K., Kim M.-H., Kim T.-H., Bang S., Choi Y.-J., Kim S., Cho S., Park B.-G.,
2019, Synaptic behaviors of HfO$_{2}$ ReRAM by pulse frequency modulation, Solid-State
Electron., Vol. 154, pp. 31-35
Shim W., Luo Y., Seo J.-S., Yu S., 2020, Investigation of read disturb and bipolar
read scheme on multilevel RRAM-based deep learning inference engine, IEEE Trans. Electron
Devices, Vol. 67, No. 6, pp. 2318-2323
Sarkar B., Lee B., Misra V., 2015., Understanding the gradual reset in Pt/Al$_{2}$O$_{3}$/Ni
RRAM for synaptic applications, Semicond. Sci. Technol., Vol. 30, No. 10, pp. 105014
Mahata C., Kim M.-H., Bang S., Kim T.-H., Lee D. K., Choi Y.-J., Kim S., Park B.-G.,
2019., SiO$_{2}$ layer effect on atomic layer deposition Al$_{2}$O$_{3}$-based resistive
switching memory, Appl. Phys. Lett., Vol. 114, No. 18, pp. 182102
Kim T.-H., Kim S., Park B.-G., 2021, Improved rectification characteristics by engineering
energy barrier height in TiO$_{x}$-based RRAM, Microelectron. Eng., Vol. 237, pp.
111498
Kim T.-H., Kim M.-H., Bang S., Lee D. K., Kim S., Cho S., Park B.-G., 2020, Fabrication
and characterization of TiO$_{x}$ memristor for synaptic device application, IEEE
Trans. Nanotechnol.
Park J., Kim T.-H., Kim S., Lee G. H., Nili H., Kim H., 2021, Conduction mechanism
effect on physical unclonable function using Al$_{2}$O$_{3}$/TiO$_{\mathrm{X}}$ memristors,
Chaos Solitons Fractals, Vol. 152, pp. 111388
Kim T.-H., Lee J., Kim S., Park J., Park B.-G., Kim H., 2021., 3-bit multilevel operation
with accurate programming scheme in TiO$_{x}$/Al$_{2}$O$_{3}$ memristor crossbar array
for quantized neuromorphic system, Nanotechnol., Vol. 32, No. 29, pp. 295201
Kim T.-H., Kim S., Hong K., Park J., Hwang Y., Park B.-G., Kim H., 2021, Multilevel
switching memristor by compliance current adjustment for off-chip training of neuromorphic
system, Chaos Solitons Fractals, Vol. 153, pp. 111587
Kim S., Kim T.-H., Kim H., Park B.-G., 2020., Current suppressed self-compliance characteristics
of oxygen rich TiO$_{y}$ inserted Al$_{2}$O$_{3}$/TiO$_{x}$ based RRAM, Appl. Phys.
Lett., Vol. 117, No. 20, pp. 202106
Kim H., Mahmoodi M., Nili H., Strukov D. B., 2021, 4K-memristor analog-grade passive
crossbar circuit, Nat. Commun., Vol. 12, pp. 5198
Zhang K., Sun K., Wang F., Han Y., Jiang Z., Zhao J., Wang B., Zhang H., Jian X.,
Wong H. P., 2015, Ultra-low power Ni/HfO$_{2}$/TiO$_{x}$/TiN resistive random access
memory with sub-30-nA reset current, IEEE Electron Device Lett., Vol. 36, No. 10,
pp. 1018-1020
Ding X., Feng Y., Huang P., Liu L., Kang J., 2019, Low-power resistive switching characteristic
in HfO$_{2}$/TiO$_{x}$ Bi-layer resistive random-access memory, Nanoscale Res. Lett.,
Vol. 14, No. 1, pp. 1-7
Ryu H., Kim S., 2020., Self-rectifying resistive switching and short-term memory characteristics
in Pt/HfO$_{2}$/TaO$_{x}$/TiN artificial synaptic device, Nanomaterials, Vol. 10,
No. 11, pp. 2159
Kim S., Abbas Y., Jeon Y.-R., Sokolov A. S., Ku B., Choi C., 2018., Engineering synaptic
characteristics of TaO$_{x}$/HfO$_{2}$ bi-layered resistive switching device, Nanotechnol.,
Vol. 29, No. 41, pp. 415204
Ryu H., Kim S., 2021, Implementation of a reservoir computing system using the short-term
effects of Pt/HfO$_{2}$/TaO$_{x}$/TiN memristors with self-rectification, Chaos Solitons
Fractals, Vol. 150, pp. 111223
Deng Y., Huang P., Chen B., Yang X., Gao B., Wang J., Zeng L., Du G., Kang J., Liu
X., 2012, RRAM crossbar array with cell selection device: A device and circuit interaction
study, IEEE Trans. Electron Devices, Vol. 60, No. 2, pp. 719-726
Kim T.-H., Nili H., Kim M.-H., Min K. K., Park B.-G., Kim H., 2020., Reset-voltage-dependent
precise tuning operation of TiO$_{x}$/Al$_{2}$O$_{3}$ memristive crossbar array, Appl.
Phys. Lett., Vol. 117, No. 15, pp. 152103
Chastain J., King Jr R. C., 1992, Handbook of X-ray photoelectron spectroscopy, Perkin-Elmer
Corporation, Vol. 40, pp. 221
Kim T., Kim H., Kim J., Kim J.-J., 2017, Input Voltage Mapping Optimized for Resistive
Memory-Based Deep Neural Network Hardware, IEEE Electron Device Lett., Vol. 38, No.
9, pp. 1228-1231
Xu B., Sohn H. Y., Mohassab Y., Lan Y., 2016, Structures, preparation and applications
of titanium suboxides, RSC Adv., Vol. 6, No. 83, pp. 79706-79722
Chong L. H., Mallik K., de Groot C H , Kersting R., 2006, The structural and electrical
properties of thermally grown TiO$_{2}$ thin films, J. Phys.-Condes. Matter, Vol.
18, No. 2, pp. 645-657
Gao Y., Kavehei O., Al-Sarawi S. F., Ranasinghe D. C., Abbott D., 2016, Read operation
performance of large selectorless cross-point array with self-rectifying memristive
device, Integration, Vol. 54, pp. 56-64
Zhou J., Kim K.-H., Lu W., 2014, Crossbar RRAM arrays: Selector device requirements
during read operation, IEEE Trans. Electron Devices, Vol. 61, No. 5, pp. 1369-1376
Tae-Hyeon Kim received the B.S. and Ph.D. degrees in electrical and computer engineering
from Seoul National University (SNU), Seoul, Republic of Korea, in 2015 and 2022,
respectively. He is currently a Postdoctoral Researcher in SNU. His current research
interests include the resistive switching devices, synaptic devices, and neuromorphic
systems.
Sungjoon Kim received the B.S. degree in electronics from Kyung Hee University
(KHU), YongIn, Korea. And received the M.S. degree in electrical engineering from
Seoul National University, Seoul, Korea. During 2010~2020, he worked for LED chip
development group at Samsung Electronics Co. Ltd. He is currently working toward the
Ph.D. degree in electrical engineering.
Kyungho Hong was born in Daejeon, Korea, in 1992. He received the B.S degree in
electrical engineering from Seoul National University (SNU) in 2018. He is currently
working toward the Ph.D. degree in electrical engineering, Seoul National University
(SNU), Seoul, Korea.
Hyungjin Kim received the B.S., M.S., and Ph.D. degrees in electrical engineering
from Seoul National University, Seoul, South Korea, in 2010, 2012, and 2017, respectively.
He is currently an Assistant Professor with the Department of Electronic Engineering,
Inha University, Incheon, Korea. His current research interests include emerging memory
device, and its computing applications.
Byung-Gook Park received his B.S. and M.S. degrees in electronics engineering from
Seoul National University (SNU) in 1982 and 1984, respectively, and his Ph.D. degree
in electrical engineering from Stanford University in 1990. From 1990 to 1993, he
worked at the AT&T Bell Laboratories, where he contributed to the development of 0.1
micron CMOS and its characterization. In 1994, he joined SNU as an assistant professor
in the School of Electrical Engineering, where he is currently a professor.