Implementation of an artificial neuron circuit model based on high speed AlN stacked threshold switching devices

Zhe Fan, Hong Fan, Chang He, Xiaobing Yan

Front. Phys. ›› 2025, Vol. 20 ›› Issue (3) : 034201.

PDF(2452 KB)
Front. Phys. All Journals
PDF(2452 KB)
Front. Phys. ›› 2025, Vol. 20 ›› Issue (3) : 034201. DOI: 10.15302/frontphys.2025.034201
RESEARCH ARTICLE

Implementation of an artificial neuron circuit model based on high speed AlN stacked threshold switching devices

Author information +
History +

Abstract

There has been a notable surge of interest in neuromorphic network computation, particularly concerning both non-volatile and volatile threshold devices. In this research, we have developed a multi-layer thin film architecture consisting of Al/AlN/Ag/AlN/Pt, which functions as a threshold switching (TS) device characterized by rapid switching speeds of 50 ns and minimal leakage current. We have effectively demonstrated biological neuron-like behaviors, such as threshold-driven spikes, all-or-nothing spikes, intensity-modulated frequency response, and frequency-modulated frequency response, through the deployment of a leaky integrate-and-fire (LIF) artificial neuron circuit, which surpasses earlier neuronal models. The resistance switching mechanism of the device is likely due to the migration of nitrogen vacancies in conjunction with silver filaments. This threshold switching device shows significant potential for applications in next-generation artificial neural networks.

Graphical abstract

Keywords

volatile memristor / artificial neurons / AlN / high speed switching

Cite this article

Download citation ▾
Zhe Fan, Hong Fan, Chang He, Xiaobing Yan. Implementation of an artificial neuron circuit model based on high speed AlN stacked threshold switching devices. Front. Phys., 2025, 20(3): 034201 https://doi.org/10.15302/frontphys.2025.034201

1 Introduction

The rapid development of computer science and technology over the past few decades has brought people to come into the era of big data on the Internet. However, the working principles of computers remain nearly unchanged which makes it to be difficulty to meet the rapid increasing of data market. Constrained by a bottleneck of the von Neumann architecture, a traditional computer confronted enormous challenges when it comes to storing and processing even larger amounts of data [1, 2]. By contrast, the mammalian brain has a large number of neurons (1012) and synapses (1015) interconnected each other to form a huge neural network. In fact, people have been trying to mimic a brain by utilizing artificial neural networks, thus requiring a large number of artificial neurons and artificial synaptic interconnections [36]. Unfortunately, it was actually impossible to implement all of them in hardware nowadays, whereas they usually could be only simulated with software codes written in a computer.
To solve this problem, a nanoscale semiconductor device has been created with a very simple structure consisting of two electrodes and a dielectric film sandwiched between two layer. And an internal ion transport process is very similar to the workings of a biological synapse [7]. Currently, synapses as a medium for connecting neurons have been successfully implemented by utilizing memristors [8]. For one hand, many synaptic functions have been successfully modelled by a variety of memristors, such as Spike-Timing Dependent Plasticity (STDP), Paired-Pulse Facilitation/Paired-Pulse Depression (PPF/PPD), and Long-Term potentiation /Long-Term Depression (LTP/LTD) [9, 10]. On the other hand, the neuron is responsible for responding to the input from the synapse and deciding whether to continue to the next neuron or not. However, most artificial neuron models primarily based on CMOS circuits have to use a large number of active devices, therefore leading to lower integration and higher power consumption. In order to apparently discover the working principle of neurons, a number of models have been proposed, including the Hodgkin-Huxley (HH) model, the integrate-and-fire (IF) model, and the leaky integrate-and-fire (LIF) model. Among them, the HH model is the model to be the closest one to the biological neuron [11, 12]. The LIF is the most widely used neuron model [13, 14]. It is mainly used to determine whether a neuron should generate spike firing to the follow-up neuron by comparing the magnitude of the local gradient potential (LGP) with the threshold voltage.For example, Liu et al. [14] realized an integration-and-fire artificial neuron based on a Ag/SiO2/Au threshold switching memristor. This neuron displayed four critical features for action-potential-based computing: the all-or-nothing spiking of an action potential, threshold-driven spiking, a refractory period, and a strength-modulated frequency response For another example, the artificial neurons constructed by Yao et al. [15] not only functioned under biological action potentials (for example, 100 mV, 1 ms), but also showed a time integral close to that of biological neurons. The potential of using memristors to directly process biosensing signals has also been demonstrated [15]. In this study, we prepared an Al/AlN/Ag/AlN/Pt threshold switching device with a fast turn-on speed of 50 ns and a high electric resistance of more than 1011 Ω. The device showed a fast switching speed and a small leakage current. A primary RC circuit was built with a foresaid device to verify the effectiveness of LIF model. In addition, a neuron circuit based on CMOS technology has been built to verify it to obey the HH model as well as the leaky integrate-and-fire function, which is not available in the previously neuron model. This new proposal device seems to be promising for applications of next-generation artificial neural networks.

2 Result and discussion

Firstly, we prepared a thin film device consisted of Al/AlN/Ag/AlN/Pt multilayer as shown in Fig.1(a). At room temperature, a layer of 3 nm AlN film was deposited firstly on a substrate of Pt bottom electrodes by a means of magnetron sputtering, followed by deposition of a layer of Ag of 25 nm, and finally a second layer of AlN with the same thickness as the first one was grown on the Ag. Finally, the Al top electrode was fabricated using a shadow mask having holes with diameter of 100 μm. Both AlN and Ag were obtained by bombarding the target with RF source 10 W and DC source 150 W, respectively, under 3 Pa argon environment. Fig.1(b) shows the morphology of the grown film by atomic force microscope (AFM), the RMS roughness is about 189 pm in a randomly selected 10 µm×10 µm scanning area. The uniform surface topography is very important for the excellent performance of the devices [16]. Fig.1(c) shows the X-ray photoelectron spectroscopy (XPS) spectrum of the film [17, 18], which shows the film consisting of Al, N, O, C, and Ag, where the carbon element is due to the organic contamination on the surface of the AlN film, and the oxygen element is also inevitably left on the surface due to the exposure to the air. Ag comes from the intermediate layer of the material. Fig.1(d) shows the core XPS spectra of Al2p, O1s and N1s. The main peak of Al2p is located at 73.53 eV, which corresponds to the Al-N bond, while the secondary peak is located at 74.03 eV which can be assigned as the Al-O bond. A shoulder peak on its high binding energy side might be attributed to impurity elements.
Fig.1 (a) Schematic film structure of Al/AlN/Ag/AlN/Pt. (b) AFM image of the Al/AlN/Ag/AlN/Pt film. (c) XPS full spectrum of the device. (d) High-resolution XPS spectra of Al 2p.

Full size|PPT slide

Fig.2(a) and (b) show the I−V characteristic curves of the device for repeating 50 cycles under linear and logarithm scale, respectively. When the voltage reaches a certain threshold volts (Vth), the current starts to increase abruptly to reach a low resistance state [19]. Then as we gradually decrease the voltage to a specific value (Vh), the device returns to a high resistance state before the scan voltage drops to 0 V, showing the volatile resistance transition behavior [20]. This typical unidirectional threshold switching at a positive scan voltage is due to the spontaneous dissipation of the conducting filament which we will discuss more below. Fortunately, this fluctuating TS behavior provides a strategy to implement “leakage integration and fire”, thus enabling the simulation of the basic spiking neuron behavior. It can be seen that the threshold voltage of the device is about 3 V, the off-state current can be as low as 10−11 A, and the switching current ratio can reach 5 orders of magnitude corresponding to the power consumption of 10−11 W. According to statistics of 50 cycles, the high and low resistance are 1011 Ω and 106 Ω as shown in Fig.2(c), respectively. Fig.2(d) is a statistical plot of the cumulative distribution function of turn-on and hold-up voltage amplitudes over 50 I−V tests, again showing the relative stability of turn-on and hold-up voltages for a single device [21].
Fig.2 (a) The I−V curve of the device. (b) The logarithm of (a). (c) High and low resistance statistics. (d) Switching voltage probability distribution.

Full size|PPT slide

Switching speed is a very important key parameter for memristor applications [22]. Here, we use 10 V, 1 MHz pulse signal to calibrate the switching speed of Al/AlN/Ag/AlN/Pt. As shown in Fig.3, the turn-on speed of the memristor is about 50 ns, and the turn-off speed is about 33.7 ns by fitting the drop line of electric current. Superior high switching speed within nanosecond range for our devices makes it more suitable to establish a fast and accurate neural network and analogue data processing. We have listed published data to show a comparison with ours in Tab.1. Importantly, we have applied a wide bandgap AlN to achieve extremely low leakage current and power consumption. The leakage current in the off state is below 10−11 A, and the power consumption for switching on is only 4 nW, indicating a great promise to achieve low-power applications of memristors in neural networks.
Tab.1 Comparison of switching speeds of devices with different thresholds.
Material system Delay time Relaxation time Ref.
Ag/Ag-GeTe/Ag <500 ns [23]
Pt/Ag/HfO2/Pt <300 ns <250 ns [24]
Pt/CuS/GeSe/Pt ~200 ns ~3 μs [4]
Pt/HfAlOx/TiN/Ag/Pt ~50 ns ~500 ns [25]
Ag/TiN/HfOx/HfOy/HfOx/Pt ~60 ns ~500 ns [26]
Ag/TaOx/TaOy/TaOx/Ag ~75 ns ~500 ns [27]
Ag/DG/SiO2/Pt <100 ns <1 μs [28]
Al/AlN/Ag/AlN/Pt ~50 ns ~34 ns This work
Fig.3 (a) Test case of device switching time. (b) The turn-on time of this device. (c) The off time of this device.

Full size|PPT slide

The resistance switching mechanism of the Al/AlN/Ag/AlN/Pt memristor can be interpreted following a conductive filament model [23, 29]. Few N vacancies (VN) are randomly located throughout the AlN film in the initial state without applying a bias voltage, as schematic in Fig.4(a). When a positive bias voltage is applied to the top Al electrode, the N3− in the upper layer of AlN film will drift under an gradually increased electric field toward the Al electrode to create a conductive channel path [30]; while in the bottom layer of AlN film, Ag+ will migrate from Ag electrode under the same electric field and move downward to form a second conductive channel path [7]. N3− will also drift in the same bottom layer, but we prefer that the movement of Ag+ plays a dominant role [2, 31], as indicated in Fig.4(b). As the bias voltage is gradually withdrawn in a resetting process, the conductive filament path formed previously might not be strong enough, and the Ag+ returned to the Ag layer and the N3− refilled the empty space (this process is completely spontaneous), and the device returns to the HRS as shown in Fig.4(c) [32, 33].
Fig.4 (a) Initial state inside the device. States after applying (b) and removing (c) positive bias voltage.

Full size|PPT slide

The process of action potential generation in neurons can be explained by a biological model of “leakage integration and excitation”. Based on the TS device, a LIF artificial neuron was demonstrated as shown in Fig.5(a) [13], where a TSM device is connected in series with a resistor Ro and two elements are connected in parallel with a capacitor. Equivalently, the neuron is connected to the input resistor Rin. The input signal Vin is a source of voltage pulses applied to the left side. In addition, the change in voltage across Ro is considered as an output spike. The construction of the LIF model is divided into an input signal integration unit and an output threshold driving unit [34]. In the RC circuit, the charging loop (CL) simulates the integration process of the membrane potential, whereas the threshold behavior caused by the ionic motion in the neuron is simulated by the Ag, N ions dynamics of the TS device in the discharge loop (DL). As shown in Fig.5(b), in LIF neurons, the received input pulse causes the neuronal membrane potential (Vc) to rise. Exceeding the threshold triggers an action potential, followed by a period of leakage, which generates an output spike through the partial voltage of Ro [14]. Fig.5(c) shows specifically the charging and discharging of the capacitor in the neuronal circuit model under the stimulation of the input pulse. And through Fig.5(d), it can be observed that there is a significant period of inactivity after the spikes are generated, which is very much in line with the refractory period of biological neuron.
Fig.5 (a) Circuit diagram for realization of artificial neuron with Al/AlN/Ag/AlN/Pt TSM. (b) The capacitor charging and spike generation under pulsed stimulation. (c) Voltage of the capacitor. (d) Voltage at the output resistor.

Full size|PPT slide

Fig.6(a) undoubtedly demonstrates the effective influence of the magnitude of input pulse voltage on the output spike intensity for the artificial neuron. The magnitudes of input pulse voltages in Fig.6(a) are 6 V, 8 V, and 10 V, respectively. With increasing input from 6 to 10 V, we apparently find that there is a positive proportional relationship between the spike frequency and the intensity of the input pulse. Others parameters used here are the input pulse frequency of 1 kHz, while the input resistance Rin of 240 kΩ, and the output resistance Ro of 20 kΩ and the shunt capacitance of 10 nf. The reason above direct correspondence can be attributed to an enhancement of ion migration under higher electric field, which results in a shorter integration time and lead to more spikes in a specific duration time. Fig.6(b) shows the experimental relation between the frequency of spikes with the pulse voltage amplitudes. These results show that there is a stable positive proportional relationship between the spike pulse frequency and the input pulse voltage intensity.
Fig.6 (a) Artificial neuron spike generation at 6 V, 8 V, and 10 V input pulse. (b) Statistics for (a). (c) Artificial neuron spike generation at 480 kΩ, 240 kΩ, and 120 kΩ input resistance. (d) Statistics for (c).

Full size|PPT slide

Fig.6(c) exams the effect of the input resistance Rin on the output spike intensity of the artificial neuron. Keeping input pulse amplitude as 8 V, the input pulse frequency is given as 1 kHz, while the output resistance Ro of 20 kΩ and the shunt capacitance of 10 nf are kept unchanged. The input resistances in Fig.6(c) are 480 kΩ, 240 kΩ, and 120 kΩ, respectively. Thus, there is a simple inverse proportional relationship between the spike frequency and the input resistance as evidenced in Fig.6(d). Where the increasing resistance slows down the rise of the capacitance potential, and thus weakens the trigger frequency of the neuron output spikes. When the resistance of the input resistance Rin is varied over a continuous period of time, the output spikes will exhibit corresponding dynamics during this period. This means synaptically plastic devices can be used instead of Rin to construct neurons with synaptic plasticity. The behaviour of the tuning of the input pulse and the change of the circuit element parameters affecting the pulse spike response frequency corroborates the programmability of Al/AlN/Ag/AlN/Pt device-based artificial neurons for neural network applications.
By utilizing the volatile of Al/AlN/Ag/AlN/Pt devices, we further designed a CMOS-based LIF neuron model circuit as shown in Fig.7(a) [11]. Within nature biological systems, the transmission of neural signals is achieved by synapses, while whether the input signal is transmitted to the next neuron or not is determined by the input signal strength upon the neuron. In biological neurons, the local gradient potential (LGP) is generated by the difference of ion concentration between the inside and outside of the cell membrane. Followed above nature principle, in our designed CMOS-based LIF neuron circuit, instead we applied the difference of input pulse intensity to produce different switching states in the Al/AlN/Ag/AlN/Pt devices to simulate the LGP.
Fig.7 (a) CMOS-based artificial neuron circuit model. (b) Effect of different LGPs on the output results. (c) Circuit generates spikes by space integration. (d) Circuit generates spikes by time integration.

Full size|PPT slide

The circuit described in Fig.7(a) consists of two major modules, e.g., the CMP and the TIMER. The input signal is fed from the left side, while the integrated voltage is applied to the Al/AlN/Ag/AlN/Pt memristor device, and the LGP is adjusted according to the different switching states of the device, and then the CMP module plays a role in comparing the outputs, and finally the TIMER module completes the neural spike response of the whole circuit model.
Among them, the monostable flip-flop consisting of a 555 timer constitutes the complete TIMER module. The monostable flip-flop has two states, stable and transient, which can be switched under the stimulation of an applied pulse. When the input voltage of the 555 timer is less than 1/3Vcc, the timer flips to 1 and enters the transient stable state, and the circuit triggers a high level with an amplitude of 2/3Vcc, and vice versa, the timer flips to 0, and the circuit is in a stable state with an output of a low level of 0.
Set the power supply voltage Vcc to 5 V, Vt to 0.8 V, and the amplitude of the input pulse voltage is in range of 1−4.5 V. Then we start to simulate the output voltage condition of the LGP as well as the neuron functional circuit. As shown in Fig.7(b), when the input pulse voltage is larger than 3 V, the device is triggered so that the LGP is higher than Vt, and the LIF neuron circuit generates a spike pulse in outputs, where the CMP output is 0 V less than 1/3Vcc (1.66 V), and output keeps a high level. When LGP is less than Vt, the CMP outputs high and output is not a high level. In summary, when LGP is greater than Vt, the circuit outputs a spike pulse, achieving the ignition function of the neuron. On the contrary, when LGP is lower than Vt, it cannot output neuron spikes, indicating the circuit is leaking for a short period of time. Thus, the LIF model can be successfully simulated in our CMOS circuit to realize the leakage function in biological neurons.
For a nature biological system, there are two different modes of signaling through synapses, e.g., spatial integration and temporal integration. Spatial integration originates the simultaneous arrival and summation of potentials from different dendrites to the same monitored neuron. Whereas, the sum of signals arriving at a given neuron at different frequencies is called temporal integration. When the LGP is below the threshold, the LGP leaks and the device Al/AlN/Ag/AlN/Pt quickly returns to its highly resistive state. When the LGP is above the threshold, this neuron model is stimulated to produce spikes. To further investigate the artificial neuron model with stable bionic properties, we simulate both the temporal and spatial integration of the LGP by applying pulses with changed frequencies as well as amplitudes through a signal generator.
As indicated in Fig.7(c), to investigate the relationship between the input pulse amplitude and the frequency of the generated spikes, we specifically assessed the difference in the frequency of the generated spikes at four different amplitudes. It is clearly observed that as the voltage amplitude increases, the number of spikes generated increases, i.e., the spike frequency increases. In addition, as shown in Fig.7(d), we also tested the spike frequency for different input pulse frequencies. Again, as the input pulse frequency increases, the output pulse frequency changes. The above results show that under this neuron model, the spike frequency can be modulated by both the input pulse amplitude and frequency. The results can also be analyzed quantitatively by adjusting the parameters of the input pulse to obtain our expected spike frequency. Therefore, the designed LIF model using Al/AlN/Ag/AlN/Pt memristor enables all requirements for simulating the time-integral and spike-impulse response functions of artificial neurons, which may provide a great help to the efficiency and accuracy of neuromorphic computation in future.

3 Experimental

In our experiments, AlN thin films were prepared on Pt substrates by radio frequency (RF) magnetron sputtering at a pressure of 3 Pa (only Ar) and its flow rate is 50 sccm. Ag thin films were prepared by direct current (DC) magnetron sputtering at a pressure of 3 Pa (only Ar) and its flow rate is 25 sccm. Al electrode were prepared by direct current (DC) magnetron sputtering at a pressure of 1 Pa (only Ar) and its flow rate is 25 sccm.

4 Conclusions

In conclusion, we have prepared an Al/AlN/Ag/AlN/Pt device with high switching speed, whose function is mainly realized by diffusion dynamics. We also constructed RC circuits and CMOS-based artificial neuron circuits to realize the threshold-driven spikes, all-or-nothing spikes, intensity-modulated frequency response and frequency-modulated frequency response characteristics of the neurons. These results show that artificial neurons based on diffusive memristors with Ag dynamics have great potential in neuromorphic computation.

References

[1]
H. S. Wong and S. Salahuddin, Memory leads the way to better computing, Nat. Nanotechnol. 10(3), 191 (2015)
CrossRef ADS Google scholar
[2]
X. Yan, J. Zhao, S. Liu, Z. Zhou, Q. Liu, J. Chen, and X. Y. Liu, Memristor with Ag-cluster-doped TiO2 films as artificial synapse for neuroinspired computing, Adv. Funct. Mater. 28(1), 1705320 (2018)
CrossRef ADS Google scholar
[3]
D. Lee, M. Kwak, K. Moon, W. Choi, J. Park, J. Yoo, J. Song, S. Lim, C. Sung, W. Banerjee, and H. Hwang, Various threshold switching devices for integrate and fire neuron applications, Adv. Electron. Mater. 5(9), 1800866 (2019)
CrossRef ADS Google scholar
[4]
K. Wang, Q. Hu, B. Gao, Q. Lin, F. W. Zhuge, D. Y. Zhang, L. Wang, Y. H. He, R. H. Scheicher, H. Tong, and X. S. Miao, Threshold switching memristor-based stochastic neurons for probabilistic computing, Mater. Horiz. 8(2), 619 (2021)
CrossRef ADS Google scholar
[5]
L. Yuan, S. Liu, W. Chen, F. Fan, and G. Liu, Organic memory and memristors: From mechanisms, materials to devices, Adv. Electron. Mater. 7(11), 2100432 (2021)
CrossRef ADS Google scholar
[6]
F. Zhou and Y. Chai, Near-sensor and in-sensor computing, Nat. Electron. 3(11), 664 (2020)
CrossRef ADS Google scholar
[7]
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, and J. J. Yang, Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing, Nat. Mater. 16(1), 101 (2017)
CrossRef ADS Google scholar
[8]
S. Liu, J. Zeng, Q. Chen, and G. Liu, Recent advances in halide perovskite memristors: From materials to applications, Front. Phys. 19(2), 23501 (2023)
CrossRef ADS Google scholar
[9]
S. H. Jo, T. Chang, I. Ebong, B. B. Bhadviya, P. Mazumder, and W. Lu, Nanoscale memristor device as synapse in neuromorphic systems, Nano Lett. 10(4), 1297 (2010)
CrossRef ADS Google scholar
[10]
L. Zhang, Z. Tang, D. Yao, Z. Fan, S. Hu, Q. J. Sun, X. G. Tang, Y. P. Jiang, X. Guo, M. Huang, G. Zhong, and J. Gao, Synaptic behaviors in flexible Au/WO/Pt/mica memristor for neuromorphic computing system, Mater. Today Phys. 23, 100650 (2022)
CrossRef ADS Google scholar
[11]
H. M. Huang, R. Yang, Z. H. Tan, H. K. He, W. Zhou, J. Xiong, and X. Guo, Quasi-hodgkin-huxley neurons with leaky integrate-and-fire functions physically realized with memristive devices, Adv. Mater. 31(3), 1803849 (2019)
CrossRef ADS Google scholar
[12]
M. D. Pickett,G. Medeiros-Ribeiro,R. S. Williams, A scalable neuristor built with Mott memristors, Nat. Mater. 12(2), 114 (2013)
[13]
D. Dev, A. Krishnaprasad, M. S. Shawkat, Z. He, S. Das, D. Fan, H. S. Chung, Y. Jung, and Roy , MoS2-based threshold switching memristor for artificial neuron, IEEE Electron Device Lett. 41(6), 936 (2020)
CrossRef ADS Google scholar
[14]
Z. Wang, S. Joshi, S. Savel’ev, W. Song, R. Midya, Y. Li, M. Rao, P. Yan, S. Asapu, Y. Zhuo, H. Jiang, P. Lin, C. Li, J. H. Yoon, N. K. Upadhyay, J. Zhang, M. Hu, J. P. Strachan, M. Barnell, Q. Wu, H. Wu, R. S. Williams, Q. Xia, and J. J. Yang, Fully memristive neural networks for pattern classification with unsupervised learning, Nat. Electron. 1(2), 137 (2018)
CrossRef ADS Google scholar
[15]
T. Fu, X. Liu, H. Gao, J. E. Ward, X. Liu, B. Yin, Z. Wang, Y. Zhuo, D. J. F. Walker, J. Joshua Yang, J. Chen, D. R. Lovley, and J. Yao, Bioinspired bio-voltage memristors, Nat. Commun. 11(1), 1861 (2020)
CrossRef ADS Google scholar
[16]
X. Meng, C. Qin, X. Liang, G. Zhang, R. Chen, J. Hu, Z. Yang, J. Huo, L. Xiao, and S. Jia, Deep learning in two-dimensional materials: Characterization, prediction, and design, Front. Phys. 19(5), 53601 (2024)
CrossRef ADS Google scholar
[17]
L. Kesper, M. Schmitz, M. G. H. Schulte, U. Berges, and C. Westphal, Revealing the nano-structures of low-dimensional germanium on Ag(1 1 0) using XPS and XPD, Appl. Nanosci. 12(7), 2151 (2022)
CrossRef ADS Google scholar
[18]
A. Agrawal, X-Ray Photoelectron Spectroscopy: Principles, Techniques and Applications, Nova Science Publishers, Inc., 2023, p. 222
[19]
Z. Zhou, F. Yang, S. Wang, L. Wang, X. Wang, C. Wang, Y. Xie, and Q. Liu, Emerging of two-dimensional materials in novel memristor, Front. Phys. 17, 23204 (2022)
CrossRef ADS Google scholar
[20]
X. Yan, X. Han, Z. Fang, Z. Zhao, Z. Zhang, J. Sun, Y. Shao, Y. Zhang, L. Wang, S. Sun, Z. Guo, X. Jia, Y. Zhang, Z. Guan, and T. Shi, Reconfigurable memristor based on SrTiO3 thin-film for neuromorphic computing, Front. Phys. 18(6), 63301 (2023)
CrossRef ADS Google scholar
[21]
J. Yang, Z. Jian, Z. Wang, J. Zhao, Z. Zhou, Y. Sun, M. Hao, L. Wang, P. Liu, J. Wang, Y. Pei, Z. Zhao, W. Wang, and X. Yan, HfAlO-based ferroelectric memristors for artificial synaptic plasticity, Front. Phys. 18(6), 63603 (2023)
CrossRef ADS Google scholar
[22]
X. Yan,Z. Zhang,Z. Guan,Z. Fang,Y. Zhang, J. Zhao,J. Sun,X. Han,J. Niu,L. Wang, X. Jia,Y. Shao,Z. Zhao,Z. Guo,B. Bai, A high-speed true random number generator based on Ag/SiNx/n-Si memristor, Front. Phys. 19(1), 13202 (2023)
[23]
C. Caidie, T. P. Jun, C. Yimao, Y. Xiaoqin, Y. Yuchao, and H. Ru, In-memory computing with emerging nonvolatile memory devices, Sci. China Inf. Sci. 64(12), 221402 (2021)
CrossRef ADS Google scholar
[24]
S. A. Chekol,F. Cuppers,R. Waser, S. Hoffmann-Eifert, An Ag/HfO2/Pt threshold switching device with an ultra-low leakage (< 10 fA), high on/off ratio (> 1011), and low threshold voltage (< 0.2 V) for energy-efficient neuromorphic computing, in: 2021 IEEE International Memory Workshop (IMW), 2021, pp 1–4
[25]
Y. F. Lu, Y. Li, H. Li, T. Q. Wan, X. Huang, Y. H. He, and X. Miao, Low-power artificial neurons based on Ag/TiN/HfAlOx/Pt threshold switching memristor for neuromorphic computing, IEEE Electron Device Lett. 41(8), 1245 (2020)
CrossRef ADS Google scholar
[26]
Y. F. Lu,H. Y. Li,Y. Li,L. H. Li,T. Q. Wan, L. Yang,W. B. Zuo,K. H. Xue,X. S. Miao, A high-performance Ag/TiN/HfOx/HfOy/HfOx/Pt diffusive memristor for calibration-free true random number generator, Adv. Electron. Mater. 8(9), 2200202 (2022)
[27]
Y. Sun,X. Zhao,C. Song,K. Xu,Y. Xi, J. Yin,Z. Wang,X. Zhou,X. Chen,G. Shi, H. Lv,Q. Liu,F. Zeng,X. Zhong,H. Wu, M. Liu,F. Pan, Performance-enhancing selector via symmetrical multilayer design, Adv. Funct. Mater. 29(13) (2019)
[28]
X. Zhao, J. Ma, X. Xiao, Q. Liu, L. Shao, D. Chen, S. Liu, J. Niu, X. Zhang, Y. Wang, R. Cao, W. Wang, Z. Di, H. Lv, S. Long, and M. Liu, Breaking the current-retention dilemma in cation-based resistive switching devices utilizing graphene with controlled defects, Adv. Mater. 30(14), 1705193 (2018)
CrossRef ADS Google scholar
[29]
F. Zahoor, F. A. Hussin, U. B. Isyaku, S. Gupta, F. A. Khanday, A. Chattopadhyay, and H. Abbas, Resistive random access memory: Introduction to device mechanism, materials and application to neuromorphic computing, Nanoscale Res. Lett. 18(1), 36 (2023)
CrossRef ADS Google scholar
[30]
B. J. Choi, A. C. Torrezan, J. P. Strachan, P. G. Kotula, A. J. Lohn, M. J. Marinella, Z. Li, R. S. Williams, and J. J. Yang, High‐speed and low-energy nitride memristors, Adv. Funct. Mater. 26(29), 5290 (2016)
CrossRef ADS Google scholar
[31]
S. A. Chekol, S. Menzel, R. W. Ahmad, R. Waser, and S. Hoffmann‐Eifert, Effect of the threshold kinetics on the filament relaxation behavior of Ag‐based diffusive memristors, Adv. Funct. Mater. 32(15), 2111242 (2022)
CrossRef ADS Google scholar
[32]
H. Sun, Q. Liu, C. Li, S. Long, H. Lv, C. Bi, Z. Huo, L. Li, and M. Liu, Direct observation of conversion between threshold switching and memory switching induced by conductive filament morphology, Adv. Funct. Mater. 24(36), 5679 (2014)
CrossRef ADS Google scholar
[33]
T. Guo, K. Pan, Y. Jiao, B. Sun, C. Du, J. P. Mills, Z. Chen, X. Zhao, L. Wei, Y. N. Zhou, and Y. A. Wu, Versatile memristor for memory and neuromorphic computing, Nanoscale Horiz. 7(3), 299 (2022)
CrossRef ADS Google scholar
[34]
H. Chen, X. G. Tang, Z. Shen, W. T. Guo, Q. J. Sun, Z. Tang, and Y. P. Jiang, Emerging memristors and applications in reservoir computing, Front. Phys. 19(1), 13401 (2023)
CrossRef ADS Google scholar

Declarations

The authors declare that they have no competing interests and there are no conflicts.

Data availability

The data that support the findings of this study are available within the article.

Acknowledgements

This work was financially supported by the National Natural Science Foundation Joint Regional Innovation Development Project (Grant No. U23A20365), the National R&D Plan “Nano Frontier” Key Special Project (Grant No. 2021YFA1200502), the Cultivation Projects of National Major R&D Project (Grant No. 92164109), the National Natural Science Foundation of China (Grant Nos. 61874158, 62004056, and 62104058), the Special Project of Strategic Leading Science and Technology of Chinese Academy of Sciences (Grant No. XDB44000000-7), Hebei Basic Research Special Key Project (Grant No. F2021201045), the Support Program for the Top Young Talents of Hebei Province (Grant No. 70280011807), the Supporting Plan for 100 Excellent Innovative Talents in Colleges and Universities of Hebei Province (Grant No. SLRC2019018), the Interdisciplinary Research Program of Natural Science of Hebei University (No. DXK202101), the Institute of Life Sciences and Green Development (No. 521100311), the Natural Science Foundation of Hebei Province (Nos. F2022201054 and F2021201022), the Outstanding Young Scientific Research and Innovation Team of Hebei University (Grant No. 605020521001), the Special Support Funds for National High Level Talents (Grant No. 041500120001), the Advanced Talents Incubation Program of the Hebei University (Grant Nos. 521000981426, 521100221071, and 521000981363), High-level Talent Funding Program of Hebei Province(Grant No. B20231003), Yanzhao Young Science Project (Grant No. F2023201076), Science and Technology Project of Hebei Education Department (Grant Nos. QN2020178 and QN2021026), and Baoding Science and Technology Plan Project (Grant Nos. 2172P011 and 2272P014).

RIGHTS & PERMISSIONS

2025 Higher Education Press
AI Summary AI Mindmap
PDF(2452 KB)

221

Accesses

0

Citations

Detail

Sections
Recommended

/