Invisible Memristor for Brain-Inspired Computing
Item Type Article
Authors Kumar, Dayanand;Keong, Lai Boon;Elatab, Nazek;Tseng, Tseung- Yuen
Citation Kumar, D., Keong, L. B., El-Atab, N., & Tseng, T.-Y. (2022).
Enhanced Synaptic Features of ZnO/TaOx Bilayer Invisible Memristor for Brain-Inspired Computing. IEEE Electron Device Letters, 1–1. https://doi.org/10.1109/led.2022.3217983
Eprint version Post-print
DOI 10.1109/LED.2022.3217983
Publisher IEEE
Journal IEEE Electron Device Letters
Rights This is an accepted manuscript version of a paper before final publisher editing and formatting. Archived with thanks to IEEE.
The version of record is available from IEEE Electron Device Letters.
Download date 2024-01-12 23:07:06
Link to Item http://hdl.handle.net/10754/685262
Abstract—
In this letter, we present a transparent bilayer ZnO/TaOx memristive synapse for brain-inspired computing. The device shows excellent AC endurance (109 cycles) and high-temperature retention (104 s) without any degradation at 1000C. The device exhibits highly stable repetitive 5130 potentiation (P) and depression (D) epochs with 1.026 M pulses. The multilevel characteristics (MLC) of the device are achieved by changing the pulse height from 0.7 to 1.1 V for long-term potentiation (LTP) and from -0.9 to -1.3 V for long-term depression (LTD) having gradual conductance change for both P and D cases. The synaptic features such as paired-pulse facilitation (PPF) and spike time-dependent plasticity (STDP) are measured using consecutive AC pulses. These unique features confirm that the synaptic device has excellent capability for the application in the brain-inspired computing systems.
Index Terms—memristor, synapse, PPF, STDP.
I.INTRODUCTION
For a brain-inspired computing technology, a memristor has an excellent capability due to its exceptional features such as high speed in nanoseconds, high ON/OFF ratio, long retention at high temperature, low power consumption, magnificent endurance, and CMOS compatibility for 3D integration technology [1-7]. Nowadays, invisible electronics have become the premier importance of the current technological industries to make environmentally and human-friendly electronic devices [3, 5, 8, 9]. The fabrication of the transparent device is either on a glass or plastic substrates to replace Si and makes them Si-free low-cost green electronic device [3, 9-11]. Invisible memristor has become extensively attractive due to its imminent applications in transparent devices like diodes, touch panel devices, and flat panel displays [11, 12]. The conduction mechanism in the memristor is based on the formation and rupture of conductive filament (CF) in the metal oxide layer [2, 3]. In recent years, various reports have been published on memristive-based synaptic devices for their applications in brain inspired computing. However, the implementation of those devices does not meet the recent requirements of the industries in terms of the high stability and repeatability of potentiation (P) and depression (D) epochs.
This work was supported by the Ministry of Science and Technology, Taiwan, under Project MOST 109-2221-E-009 -034 -MY3 and the Higher Education Sprout Project of the National Yang Ming Chiao Tung University and Ministry of Education (MOE), Taiwan. This research was also support by the King Abdullah University of Science and Technology (KAUST) baseline fund.
D. Kumar and Nazek El-Atab are with the Smart, Advanced Memory Devices and Applications (SAMA) Laboratory, Electrical and Computer Engineering, King Abdullah University of Science and Technology (KAUST), Thuwal 23955-6900, Saudi Arabia.
D. Kumar L. B. Keong, and T. Y. Tseng are with the Institute of Electronics, National Yang Ming Chiao Tung University, Hsinchu 30010, Taiwan, e-mail: [email protected] & [email protected].
Several memristors have been proposed but their poor endurance (repeatable P and D epochs) restricts the devices for neuromorphic computing application [13-21]. To enhance the endurance of P and D, we fabricate a fully transparent bilayer ZnO/TaOx memristive synaptic device for the synaptic application.
In this work, we present the ZnO/TaOx based synaptic device for brain-inspired computing. The device exhibits superb 5130 repetitive P and D epochs without any deprivation.
The synaptic device exhibits excellent AC durability of 109 cycles, excellent transparency (>85 %), and high-temperature retention of 104 s at 100 0C. These tremendous aspects of the device make it more suitable for neuromorphic computing systems. Table I represents the comparison of previous published single and bilayer-based memristive synaptic devices with current work.
II.DEVICE FABRICATION
A 10 nm TaOx thin film was deposited by RF sputter on ITO bottom electrode (BE) with gas mixture of 3:1 Ar/O2 at 10 mTorr pressure. After that, a 40 nm ZnO film was deposited on the TaOx layer by RF sputter with gas mixture of 2:1 Ar/O2 and at 10 mTorr pressure. Subsequently, a 100 nm ITO top electrode (TE) was deposited using shadow mask (100 μm in dia) by RF sputter also with pure Ar ambiance at 10 mTorr pressure to prepare ITO/ZnO/TaOx/ITO synaptic device. The Agilent B1500 semiconductor analyzer was used to characterize the synaptic device.
III. Results and Discussion
Figures 1(a) and (b) represent the cross-sectional TEM images of ITO/ZnO/TaOx/ITO/glass synaptic structure on the scale bars of 50 and 10 nm, respectively. Figures 1(c) depicts energy dispersive spectroscopy (EDS) line profile of the device, which was measured from the TE towards BE. The EDS profile validates the presence of tantalum (Ta), indium (In), tin (Sn), oxygen (O), and zinc (Zn) in the device. Figure 1(d) displays the biological neural system where pre- and post-synaptic neurons relate to each other via biological Dayanand Kumar, Lai Boon Keong, Nazek El-Atab, Senior Member, IEEE, and Tseung-Yuen Tseng, Fellow, IEEE
Fellow, IEEE Fellow, IEEE
Enhanced Synaptic Features of ZnO/TaO x Bilayer
Invisible Memristor for Brain-Inspired Computing
synapse. The biological synapse which transforms the data from one neuron to another neuron, is dominated by the firing mechanism of calcium or sodium ions, consequently, the connection's strength (synaptic weight) between two neurons is modulated. The electronic synapse (memristor) could be used also as a biological synapse which discloses the connection's strength modulation using voltage pulses, indicated in Fig. 1(e) [3, 21].
The I-V curve of the TaOx (10 nm) based memristor is shown in Fig. 2(a). AC endurance of the device was measured using the SET and RESET voltages of 1.4 and -1.8 V, respectively, as indicated in Fig. 2(b). The device exhibits high stable AC endurance (109 cycles) with a speed of 60 ns without any degradation. The retention test of the device was also measured, indicated in Fig. 2(c). The device confirms the stable retention at high temperature (100 0C) for both high resistance state (HRS) and low resistance state (LRS). The transparency
of the devices was measured using UV–Vis spectrophotometer (U-3010, Hitachi), which is illustrated in Fig. 2 (d). The average transmittance of the device is found to be more than 85
% (including glass substrate) in the visible region from 400 to 700 nm.
Figure 3 (a) illustrates the electroforming process and I-V characteristics of the TaOx (0, 10, 15, and 20 nm) devices for the comparison. It confirms that the forming voltage of the device is increased with increasing the TaOx thickness. The DC endurance of TaOx (10 nm) device is shown in Fig. 3 (b), indicating highly stable DC endurance for at least 10000 cycles without any humiliation. We also measured the DC endurances of TaOx (0, 15, and 20 nm) based devices, shown in inset of Fig.
3 (b). These devices depict poor endurances with high instability in both LRS and HRS. The poor performance and high instability in TaOx (0, 15, and 20 nm) based devices are LRS
1µ 10µ 100µ 1m
1 10 40 70 95 99.5
TaOx (15 nm)
Probability (%)
Current (A) TaOx (10 nm) TaOx (20 nm)
D1 D2 D3 D4 D5 D6 D7 D8 D9D10 1µ
10µ 100µ 1m
TaOx (15 nm)
Current (A)
Device Number TaOx (10 nm) TaOx (20 nm)
HRS
HRS LRS
Fig. 3 (a) Cycle to cycle uniformity of TaOx (10 nm, 15 nm, and 20 nm) based ITO/ZnO/TaOx/ITO devices. (b) Device to device stability randomly chosen 10 devices for TaOx (10 nm, 15 nm and 20 nm) based samples.
-1.0 -0.5 0.0 0.5 1.0 10-6
10-5 10-4 10-3
Current (A)
Voltage(V) RESET
SET
400 500 600 700 800 0
20 40 60 80 100
Transmittance (%)
Wavelength (nm) Visible Region
~94%
ITO/ZnO/TaOx/ITO ITO/Glass
100101102103104105106107108109 10-6
10-5 10-4 10-3
LRS HRS
Current (A)
Number of Cycle ON/OFF Ratio>10
Vread=0.1V
100 101 102 103 104 10-6
10-5 10-4 10-3
LRS HRS
Current (A)
Time (s) Vread=0.1V
ON/OFF Ratio>10
SET: 1.4 V/60 ns RESET: -1.8 V/60 ns
(a) (b)
(c) (d)
Retention @ 100 0C
Fig. 2 (a). I-V curve of the device (b) AC endurance of the device. (c) Retention test at high temperature. (d) Transmittance of the synaptic structure and ITO/Glass substrate. Inset: the transparent device is in the hand and, also placed above yellow car.
50 nm
ITO ITO ZnO
10 nm
ITO
Biological Synapse Electronic Synapse
(a) (b)
(d)
Neural Subsystem Neurons
0 80 160 240
Intensity (a.u)
Position (nm)
Ta In
Sn O
Zn
TaOx
ITO
ITO
(e)
TaOx
10 nm
ZnO
40 nm
ZnO
40 nm10 nm
(c)
Fig. 1. Cross-section TEM images of the ITO/ZnO/TaOx/ITO/glass substrate on the scales of (a) 50 nm and (b) 10 nm. (c) EDS line profile. (d) Biological synapse. (e) Electronic synapse.
528 530 532 534 536 O 1s
TaOx
Vacancy- 20.49%
Raw Data Fitting Sum
Intensity(a.u)
Binding Energy(eV)
526 528 530 532 534
Raw Data Fitting Sum
Intensity(a.u.)
Binding Energy (eV) Vacancy- 35.37%
O 1s ZnO
0 25 50 75 100 0.80
0.85 0.90 0.95 1.00 1.05 1.10
0.7 V 0.8 V 0.9 V 1.0 V 1.1 V
Conductance(mS)
Pulse Number(#) 0.800 25 50 75 100 0.85
0.90 0.95 1.00 1.05
1.10 -0.9 V
-1.0 V -1.1 V -1.2 V -1.3 V
Conductance(mS)
Pulse Number(#)
(a) (b)
(d) (e)
LTP
LTD Potentiation Depression
528.8 eV
530.2 eV
531.7 eV 531.1 eV
ITO
ITO TaOx
10 nm
ZnO
40 nm
ZnO
40 nm10 nm
ITO
ITO TaOx
10 nm
ZnO
40 nm
ZnO
40 nm10 nm
SET (c) RESET
Fig. 4. (a) and (b) XPS spectra of ZnO and TaOx, respectively. (c) CF model for SET and RESET (d) and (e) LTP and LTD with varying pulse height of 0.7 V to 1.1 V and -0.9 V to -1.3 V, respectively.
LRS
1µ 10µ 100µ 1m
1 10 40 70 95 99.5
TaOx (0 nm) TaOx (15 nm)
Probability (%)
Current (A) TaOx (10 nm) TaOx (20 nm)
D1 D2 D3 D4 D5 D6 D7 D8 D9D10 10-6
10-5 10-4 10-3
TaOx (0 nm) TaOx (15 nm)
Current (A)
Device Number TaOx (10 nm)
TaOx (20 nm)
HRS
HRS LRS -3 -2 -1 0 1 2 3 4 5
10-9 10-8 10-7 10-6 10-5 10-4 10-3
TaOx (0 nm) TaOx (10 nm) TaOx (15 nm) TaOx (20 nm)
Current (A)
Voltage (V) 10-9 0 2500 5000 7500 10000 10-8
10-7 10-6 10-5 10-4
LRS HRS
Current (A)
Number of Cycle
0 500 1000 1500 10-6 10-5 10-4 10-3
HRS LRS
Current (A)
Number of Cycle 10-60 500 1000 10-5 10-4 10-3
HRS LRS
Current (A)
Number of Cycle 10-60 700 1400 2100 10-5 10-4 10-3
HRS LRS
Current (A)
Number of Cycle
(a)
(b)
(c) (d)
TaOx(0 nm)
TaOx(10 nm) TaOx(15 nm)TaOx(20 nm)
Forming
Fig. 3(a). Forming process and I-V characteristics of TaOx (0,10, 15 and 20 nm) based devices. (b) DC endurance of the TaOx (10 nm) device. Inset: DC endurances of the TaOx (0, 15 and 20 nm) based devices. (c) Cycle to cycle uniformity TaOx (0, 10, 15, and 20 nm) based devices. Device to device uniformity of 10 devices which were chosen randomly for TaOx (0, 10, 15 and 20 nm) samples.
not favorable for neuromorphic computing. To confirm the high stability of the TaOx (10 nm) based ITO/ZnO/TaOx/ITO device, the cycle to cycle (C2C) and device to device (D2D) uniformity were measured and are displayed in Figs. 3 (c) and (d)), respectively. The C2C (100 continuous cycles are used) and D2D uniformities of the 10 nm TaOx (10 nm) based device indicate high consistency in both LRS and HRS. Also, the resistive switching (RS) of the devices with different thicknesses of TaOx (15 and 20 nm) was measured. These
devices show a wider variation in C2C and D2D distributions, as presented in Figs. 3(c) and 3(d), respectively. This wider distribution is caused by the region for growth and breach of the CF being large in the thicker TaOx (15 and 20 nm) devices leading to the random creation and break of the CF [6, 22, 24, 25]. These poor features confirm that the uniformity of the device is sensitive to the thickness TaOx layer [22]. The TaOx
(0 nm) based device shows the abrupt switching and wider fluctuation in the SET and RESET voltage during the continuous switching cycles. This poor performance of the TaOx (0 nm) device makes it incompatible for neuromorphic computing [24, 25]
To explain the conduction mechanism in the device, the O 1s spectra are analyzed for ZnO and TaOx layers (Figs. 4(a) and (b)). The used X-ray beam size was about 0.25 mm2 during the XPS measurement. The O 1s spectra are fitted using Gaussian functions. The spectra can be deconvoluted into two peaks. The higher intense peaks with the binding energies of 528.8 and 531.1 eV represent the oxygen ions while minor intense peaks with the binding energies of 530.2 and 531.7 eV denote the oxygen vacancies in ZnO and TaOx layers, respectively. The concentration of oxygen vacancy in ZnO and TaOx are 35.37 and 20.49 %, respectively. The concentration of oxygen vacancies performs a vital part to construct the filaments in both layers [16, 23, 24]. We validate the CF model in the device with the basis of oxygen vacancies, as illustrated in Fig. 4(c).
Researchers have described those different quantities of oxygen vacancies create the hourglass shape CF in the bilayer memristors [3, 16, 23, 24]. This hourglass-shaped CF in the device boosts the device's performance. To achieve MLC in long-term (LT) P and D of the memristor, 100 AC pulses were used with a pulse width of 10 μs. The MLC for LTP, pulse height was used from 0.7 to 1.1 V (step size 0.1 V [Fig. 4(d)]),
while MLC for LTD, from −0.9 to −1.3 V (step size −0.1V) [Fig. 4(e)]. The memristor shows stable and gradual conductance changes for LTP and LTD using different pulse heights and reveals excellent ability of MLC.
To emulate the LTP and LTD in the memristive synapse, the conductance changes of the memristor were measured by using 100 (0.8 V/10 µs) P pulses and subsequent 100 (−1.0 V/10µs) D pulses, as shown in Figs. 5(a) and 5(b), respectively.
The device shows the high stable 5130 P and D epochs without any deterioration. Paired-pulse facilitation (PPF) is the important biological feature of synaptic devices [17]. When a pair of identical pulses is applied to the presynaptic terminal, the device current A2 increases significantly, this phenomenon is called PPF. A pair of spikes (pulse amplitude: 0.8 V, Pulse width: 10us) was applied to the device, the device current A2 is higher after second pulse than the device current A1 after first pulse, that is closely related to Δt. The PPF could be achieve by calculating (A2-A1/A1) x100, which shows a stable and gradual decrease in PPF index mimicking mammalian synaptic function [Fig. 5(c)]. The increase in current for the second pulse is the same as for the post-synaptic pulse in neurons which is termed excitatory post-synaptic current (EPSC). STDP is a learning rule which suggests that if two neurons are repeatedly activated at the same time, the strength of their connection is increased. Given a synaptic stimulus, LTP happens if post-spike comes after pre-spike (Δt>0). The combined pulse is mostly positive, resulting in the increment in conductance of the device due to the formation of CF, whereas LTD happens if pre-spike comes after post-spike (Δt<0). In this case, the combined pulse is mostly negative, resulting in a decrement in the conductance of the device due to the rupture of CF.
Therefore, the synaptic weight change is the function of the Δt between pre-and post-spike, described as STDP [17]. To mimic the STDP in the memristor, two simultaneous spikes [(pre-spikes (0.8 V, 10 μs) and post-spikes (−0.1 V, 10 μs)]
with different interval time Δt are applied to the TE and BE individually. The change of synaptic weight (ΔW) can be calculated by (W1−W0/W0) ×100, where W1 and W0 are the synaptic weight of synapse after and before application of spike pair, respectively. When Δt>0, the ΔW is enhanced, and the ΔW is reduced with the increase of Δt. On the other side, when Δt<0, the ΔW is reduced, and the ΔW enhances with the decrease of Δt. These synaptic characteristics of the memristor confirm it to be suitable for brain inspired computing.
IV.CONCLUSIONS
A transparent bilayer ZnO/TaOx based memristor is presented for artificial synaptic device. The device demonstrated superb AC endurance (109 cycles), and high-temperature retention (104 s) at 100 °C without any humiliation. With 1.026M pulses, the device showed highly steady repeated 5130 P and D epochs. Consecutive AC pulses were used to measure synaptic characteristics including PPF and STDP. The device's promising synaptic characteristics demonstrate how well-suited it is for brain inspired computing.
REFERENCES
[1] J. J. Yang, D. B. Strukov, and D. R. Stewart, “Memristive devices for computing,” Nature Nanotechnol., vol. 8, no. 1, pp. 13–24, Jan. 2013,
-100 -75 -50 -25 0 25 50 75 100 -80
-60 -40 -20 0 20 40 60 80
Exp. Data Fitting Curve
DW (%)
Dt (ms)
Model ExpDec1
Equation y = A1*exp(-x/t1) + y0 Reduced Chi-Sqr
2.29793
Adj. R-Square 0.99203
B y0 A1 t1
Model ExpDec1
Equation y = A1*exp(-x/t1) + y0 Reduced Chi-Sqr
1.7392
Adj. R-Square 0.99664
D y0 A1 t1
50 60 70 80 90 100
10 20 30 40 50 60 70
80 Experiment data
Fitting curve
PPF Index(%)
Interval time(ms)
(c) (d) Δt>0
LTP
Δt<0 LTD
Pre-spike Δt<0 Post-spike Pre-spike
Δt>0 Post-spike 0 200 400 600 800 1000
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
1.6 Potentiation
Depression
Normalized Conductance
Pulse Number (#)
1st 2nd 3rd 4th 5th
Potentiation Pulse (0.8 V, 10 µs) Depression Pulse
(-1.0 V, 10 µs ) Read @ 0.1 V
(a)
1.024M 1.025M 1.026M 0.0
0.2 0.4 0.6 0.8 1.0 1.2 1.4
1.6 Potentiation
Depression
Normalized Conductance
Pulse Number (#)
5126th 5127th 5128th 5129th 5130th
Potentiation Pulse (0.8 V, 10 µs) Depression Pulse
(-1.0 V, 10 µs ) Read @ 0.1 V
(b)
Δt
A1 A2
10 µs
0.8 V
Fig. 5. (a) First 1-5 P and D epochs. (b) Last 5126-5130P and D epochs. (c) PPF and (d) STDP of the synaptic device.
doi: 10.1038/nnano.2012.240.
[2] L. Goux A. Fantini, G. Kar, Y.-Y. Chen, N. Jossart, R. Degraeve, S.
Clima, B. Govoreanu, G. Lorenzo, G. Pourtois, D.J. Wouters, J.A. Kittl, L.
Altimime, M. Jurczak, “Ultralow sub-500nA operating current high-performance TiN/Al2O3/HfO2/Hf/TiN bipolar RRAM achieved through understanding-based stack-engineering,” presented at the Symp.
VLSI Technol. (VLSIT), 2012, doi: 10.1109/VLSIT.2012.6242510.
[3] D. Kumar, A. Saleem, L. B. Keong, A. Singh, Y. H. Wang, and T. Y.
Tseng, “ZnSnOy/ZnSnOx Bilayer Transparent Memristive Synaptic Device for Neuromorphic Computing,” IEEE Electron Devices Lett., vol.
3, pp 1211-1214, Aug. 2022, doi:10.1109/LED.2022.3186055.
[4] D. Kumar, U. Chand, L. W. Siang and T. Y. Tseng, “Flexible ZrN Based Resistive Switching Memory,” IEEE Electron Devices Lett., vol. 41, pp.
705-708, May 2020, doi: 10.1109/LED.2020.2981529.
[5] D. Kumar, A. Saleem, L. B. Keong, A. Singh, Y. H. Wang, and T. Y.
Tseng, “Light Induced RESET Phenomenon in Invisible Memristor for Photo Sensing,” IEEE Electron Devices Lett., vol. 43, pp. 1069-1072, July 2022, doi: 10.1109/LED.2022.3172866.
[6] D. Kumar, U. Chand, L. W. Siang and T. Y. Tseng, “High-Performance TiN/Al2O3/ZnO/Al2O3/TiN Flexible RRAM Device for Memory Applications with High Bending Condition” IEEE Trans. Electron Devices, vol. 67, pp 493-498, Feb. 2020, doi 10.1109/TED.2019.2959883.
[7] J. T. Qiu, S. Samanta, M. Dutta, S. Ginnaram, and S. Maikap,
“Controlling resistive switching by using an optimized MoS2 interfacial layer and the role of top electrodes on ascorbic acid sensing in TaOx -based RRAM,” Langmuir, vol. 35, no. 11, pp. 3897–3906, Feb. 2019, doi:
10.1021/acs.langmuir.8b04090.
[8] S. Bae, H. Kim, Y. Lee, X. Xu, J. S. Park, Y. Zheng, J. Balakrishnan, T.
Lei, H. R. Kim, Y. I. Song, Y. J. Kim, K. S. Kim, B. Ozyilmaz, J. H. Ahn, B. H. Hong, and S. Iijima, “Roll-to-roll production of 30-inch graphene films for transparent electrodes,” Nature Nanotech.,vol. 5, pp. 574-578, Aug. 2010, doi:10.1038/NNANO.2010.132.
[9] K. Qian, R. Y. Tay, M. F. Lin, J. Chen, H. Li, J. Lin, J. Wang, G. Cai, V. C.
Nguyen, E. H. T. Teo, T. Chen, and P. S. Lee, “Direct Observation of Indium Conductive Filaments in Transparent, Flexible, and Transferable Resistive Switching Memory,” ACS NANO, vol. 11, pp. 1712-1718, Jan.
2017, doi: 10.1021/acsnano.6b07577.
[10] T. Zhang, X. Ou, W. Zhang, J. Yin, Y. Xia, and Z. Liu, “High-k -rare-earth-oxide Eu2O3 films for transparent resistive random-access memory (RRAM) devices,” J. Phys. D: Appl. Phys, vol. 47, 2014, Art. no.
065302, doi:10.1088/0022-3727/47/6/065302.
[11] D. H. Kim, N. Lu, R. Ma, Y. S. Kim, R. H. Kim, S. Wang, J. Wu, S. M.
Won, H. Tao, A. Islam, K. J. Yu, T. Kim, R. Chowdhury, M. Ying, L. Xu, M. Li, H. J. Chung, H. Keum, M. M. Cormick, P. Liu, Y. W. Zhang, F. G.
Omenetto, Y. Huang, T. Coleman, J. A. Rogers, “Epidermal Electronics,”
Science, vol. 333, pp. 838-843, Aug. 2011, doi:10.1126/science.1206157.
[12] S. H. Chae, W. J. Yu, J. J. Bae, D. L. Duong, D. Perello, H. Y. Jeong, Q. H.
Ta, T. H. Ly, Q. A. Vu, M. Yun, X. Duan, and Y. H. Lee,” Transferred wrinkled Al2O3 for highly stretchable and transparent graphene–carbon nanotube transistors,” Nat. Materials, vol. 12, pp. 403-409, Mar. 2013, doi: 10.1038/NMAT3572.
[13] J. Woo, K. Moon, J. Song, S. Lee, M. Kwak, J. Park, and H. Hwang,
“Improved synaptic behavior under identical pulses using AlOx /HfO2
bilayer RRAM array for neuromorphic systems,” IEEE Electron Device Lett., vol. 37, no. 8, pp. 994–997, Aug. 2016, doi:
10.1109/LED.2016.2582859.
[14] W. Wu, H. Wu, B. Gao, P. Yao, X. Zhang, X., Peng, S. Yu, and H. Qian,
“A methodology to improve linearity of analog RRAM for neuromorphic computing,” in Proc. IEEE Symp. VLSI Technol., Jun. 2018, pp. 103–104, doi: 10.1109/VLSIT.2018.8510690.
[15] I.-T. Wang, C.-C. Chang, L.-W. Chiu, T. Chou, and T.-H. Hou, “3D Ta/TaOx /TiO2/Ti synaptic array and linearity tuning of weight update for hardware neural network applications,” Nanotechnology, vol. 27, no.
36, Aug. 2016, Art. no. 365204, doi: 10.1088/0957- 4484/27/36/365204.
[16] D. Kumar, S. Shrivastava, A. Saleem, A. Singh, H. Lee, Y. H. Wang, and T. Y. Tseng, “Highly Efficient Invisible TaOx/ZTO Bilayer Memristor for Neuromorphic Computing and Image Sensing,” ACS Appl. Electron.
Mater., vol. 4, pp. 2180-2190, Aprl. 2022, doi: 10 .1021/acsaelm.1c01152.
[17] X. Zhang, S. Liu, X. Zhao, F. Wu, Q. Wu, W. Wang, R. Cao, Y. Fang, H.
Lv, S. Long, Q. Liu, and M. Liu, “Emulating Short-Term and Long-Term Plasticity of Bio-Synapse Based on Cu/a-Si/Pt Memristor,” IEEE
Electron Devices Lett., vol. 38, pp 1208-1211, Sept. 2022, doi:
10.1109/LED.2017.2722463.
[18] A. Saleem D. Kumar, A. Singh, S. Rajasekaran, and T. Y. Tseng,
“Oxygen Vacancy Transition in HfOx-Based Flexible, Robust, and Synaptic Bi-Layer Memristor for Neuromorphic and Wearable Applications,” Adv.Mater. Technol., vol. 7, 2022, Art. no. 2101208. doi:
0.1002/admt.202101208.
[19] S. H. Choi, S. O. Park, S. Seo, S. Choi, “Reliable multilevel memristive neuromorphic devices based on amorphous matrix via quasi-1D filament confinement and buffer layer,” Sci. Adv., vol. 8, 2022, Art. no. eabj7866, doi: 10.1126/sciadv.abj7866.
[20] J. Park, H. Ryu, and S. Kim, “Nonideal resistive and synaptic characteristics in Ag/ZnO/TiN device for neuromorphic system,” Sci.
Reports, vol. 11, 2021, Art. no. 16601, doi:
10.1038/s41598-021-96197-8.
[21] C. L. Hsu, A. Saleem, A. Singh, D. Kumar, and T. Y. Tseng, “Enhanced Linearity in CBRAM Synapse by Post Oxide Deposition Annealing for Neuromorphic Computing Applications,” IEEE Trans. Electron Devices, vol. 68, no. 11, pp. 1296 –1301, Nov. 2021, doi:
10.1109/TED.2021.3112109.
[22] U. Chand, C.Y. Huang, and T. Y. Tseng, “Mechanism of High Temperature Retention Property (up to 200 °C) in ZrO2-Based Memory Device With Inserting a ZnO Thin Layer,” IEEE Electron Devices Lett., vol. 35, pp 1019-1021, Oct. 2024, doi: 10.1109/LED.2014.2345782.
[23] R. Zhang, H. Huang, Q. Xia, C. Ye, X. Wei, J. Wang, L. Zhang, and L. Q.
Zhu, “Role of oxygen vacancies at the TiO2/HfO2 interface in flexible oxide-based resistive switching memory,” Adv. Electron. Mater., vol. 5, no. 5, Apr. 2019, Art. no. 1800833, doi: 10.1002/aelm.201800833.
[24] D. Kumar, P. S. Kalaga, and D. S. Ang, “Visible light detection and memory capabilities in MgO/HfO2 bilayer-based transparent structure for photograph sensing,” IEEE Trans. Electron Devices, vol. 67, no. 10, pp.
4274–4280, Oct. 2020, doi: 10.1109/TED.2020.3014271.
[25] T.-L. Tsai, H.-Y. Chang, J. J.-C. Lou, and T.-Y. Tseng, “A high performance transparent resistive switching memory made from ZrO2/AlON bilayer structure,” Appl. Phys. Lett., vol. 108, no. 15, Apr.
2016, Art. no. 153505, doi: 10.1063/1.4946006.