Graphene enhanced charge transfer in ITO optoelectronic synapses for artificial vision systems†

Jiran Liang*^ac, Xuan Yu*^abc, Chuantong Cheng*^bd, Beiju Huang*^bd, Zidong Wang^bd and Liting Huang^bd
aSchool of Microelectronics, Tianjin University, Tianjin 300072, China. E-mail: liang_jiran@tju.edu.cn; yuxuan@aust.edu.cn
^bThe State Key Laboratory on Integrated Optoelectronics, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China. E-mail: chengchuantong@semi.ac.cn; bjhuang@semi.ac.cn
^cTianjin Key Laboratory of Imaging and Sensing Microelectronic Technology, Tianjin University, Tianjin 300072, China
^dCollege of Materials Science and Opto-Electronic Technology, University of Chinese Academy of Sciences, Beijing 100049, People's Republic of China

First published on 23rd August 2024

---

---

1. Introduction

The rapidly developing field of artificial intelligence has put forward higher requirements for machine vision recognition. In traditional artificial vision recognition systems, a single electrically modulated synapse typically needs to be combined with image sensors to convert light signals into electrical signals, and complex interconnections can significantly increase power consumption and reduce bandwidth.^1–5 Compared with traditional artificial visual structures, the human brain shows an excellent visual structure, which can efficiently process complex information at low power consumption, thanks to the synergistic effects of billions of synapses and neurons.^4,6–8 Nickel nanocrystal-embedded titanium dioxide,⁹ Ta₂O₅/WO₃,¹⁰ and tantalum oxide¹¹ based memristors have been used to simulate various functions of human neural synapses. Photosensitive cells in the eyes of people can convert light signals into electrical signals, which are transmitted by synapses and neurons. The retina can preprocess the perceived signals, reduce the complexity of the signals, and enable the brain to obtain simple and accurate information.^12–16 Similarly, optoelectronic synaptic devices can directly preprocess signals and perform neural morphology calculations, which can effectively reduce the latency and power consumption of artificial vision systems.^17–20 For example, researchers have simulated the characteristics of biological synapses using the photoelectric opening behavior of all-oxide memristors, achieving the function of photoreceptors.²¹ Therefore, the construction of artificial optoelectronic synapses is an effective way to achieve an artificial visual system that can integrate perception, transmission, processing, and storage.

The persistent photoconductivity (PPC) phenomenon is an important working mechanism of metal oxide based optoelectronic synaptic devices.^4,22 Under light stimulation, oxygen vacancies in amorphous metal oxides are continuously ionized to produce V⁺_O and V²⁺_O, accompanied by sustained photocurrent generation.²³ Indium tin oxide is a typical metal oxide with PPC properties. Due to its photosensitive properties, ITO was used as a photosensitive material in optoelectronic devices.^24–26 However, the synapse composed solely of an ITO layer can only simulate excitatory postsynaptic currents and has a longer relaxation time due to the PPC effect.²⁷ For artificial optoelectronic synapses, the writing and erasing of light signals are not only the basis for simulating biological synaptic characteristics, but also a necessary factor for improving repeatability. Due to the slow recombination of ionized oxygen vacancies with electrons in semiconductors, the challenge with photosynapses based on the PPC effect is the limited controllability of the relaxation characteristics of PPC. The recovery time of conductivity changes caused by PPC behavior is relatively long, which directly affects the application of photoelectric synapses in simulating artificial vision.^28,29 As a two-dimensional material, graphene has been widely studied due to its high stability, strong modulability, and ease of integration.^30–34 Researchers have utilized graphene nanoribbons,³⁵ single-layer graphene/SiO₂/Si,³⁶ and two-dimensional graphene nanosheets/amorphous indium gallium zinc oxide structures³⁷ to achieve optical information storage. These studies indicate that graphene has high modulability. Compared with some artificial optoelectronic synapses based on organic³⁸ and perovskite³⁹ materials, the preparation of ITO-graphene optoelectronic synapses is compatible with existing integrated circuit technologies and facilitates large-scale integration. Compared with the nA level photoresponsivity synapses composed of some inorganic materials,⁴⁰ ITO graphene synapses have a higher photoresponsivity amplitude (μA level). Therefore, using graphene with easy to modulate conductivity as a conductive channel and ITO with high photoresponsivity as a functional layer to provide photogenerated electrons is an effective way to achieve artificial optoelectronic synapses with high photoresponsivity, process compatibility, and fast modulation.

In this article, a high-performance artificial optoelectronic synapse based on ITO/graphene heterojunctions is proposed. By introducing P-type doped graphene as a conductivity modulation layer, an IPSC is generated by combining the photogenerated electrons by ITO under purple light stimulation with the majority of charge carrier holes in graphene. The electron exchange of ITO/graphene heterojunctions under light stimulation results in a higher photoresponse amplitude of the device. ITO/graphene synapses can simulate various biological synaptic plasticity characteristics under purple light stimulation. In addition, under positive gate piezoelectric stimulation, electrons are attracted from the ITO/graphene contact interface to the ITO film, and the electrons returning to the ITO film combine with ionized oxygen vacancies. This electron loss behavior accelerates the conductivity recovery process of graphene conductive channels, solving the problem of difficulty in simulating short-term synaptic plasticity in single-layer ITO devices. Finally, we prepared a 2 × 2-pixel imaging chip that simulates artificial vision functions. After 39 training sessions with purple light stimulation, the chip produced long-term memory of the images. On the one hand, good process compatibility makes it possible to prepare large-scale integrated arrays of ITO graphene artificial synapses. On the other hand, the realization of multiple synaptic plasticity and high photoresponsivity make ITO graphene synaptic devices show great potential in constructing neuromorphic visual systems.

2. Results and discussion

Photoreceptor cells such as rods and cones in the human retina can convert light into electrical signals when stimulated by light. Among them, cone cells can sense strong light and provide high-resolution color vision, while rod cells are mainly responsible for sensing weak light. When the presynaptic membrane receives an action potential, changes in Ca⁺ channels can lead to the release of neurotransmitters, thereby altering the membrane potential of the postsynaptic membrane.⁴¹ Electrical signals rely on the Ca⁺ dynamics of synapses to be transmitted between neurons (Fig. 1a).⁴² This characteristic is similar to the PPC phenomenon of amorphous metal oxides, which generates photogenerated electrons when stimulated by light to change the conductivity of the film. As shown in Fig. 1b, oxygen vacancies in the ITO layer can provide photogenerated electrons to graphene after being excited by light. Photogenerated electrons can alter the conductivity of the graphene layer, i.e. the synaptic weight of artificial synapses, allowing stimulus signals to undergo memory transmission between synapses. According to this approach, Au/ITO-graphene/Au artificial synaptic devices were manufactured for simulating biological synaptic characteristics (Fig. 1c). The manufacturing method of the device is described in the Experimental section. Raman spectroscopy is used to characterize the number of layers and defect characteristics of graphene materials (Fig. S1, ESI†). In order to simulate the function of the human retina in converting light signals into electrical signals for transmission and memory, a 405 nm purple light pulse with a pulse width of 2 s and a light intensity of 198 mW cm⁻² was applied to the artificial synapse, and the device generated an inhibitory postsynaptic current IPSC (Fig. 1d). The Ca⁺ dynamics of biological synapses were simulated by the ITO layer, and the control of artificial synaptic weights was achieved under light stimulation.

The rapid writing and erasing of electrical conductivity changes caused by light signals is one of the basic functions of optoelectronic synapses. Here, we shorten the recovery time of conductivity changes caused by the PPC effect by applying the positive voltage stimulation to the gate. Firstly, 1.5 s of purple light stimulation (405 nm, 198 mW cm⁻²) was applied to the Au/ITO-graphene/Au artificial optoelectronic synapse. Due to the presence of relaxation behavior, the generation of the artificial synapse ΔPSC slowly decreases over time. Subsequently, multiple 10 V positive gate piezoelectric stimuli with a pulse width of 150 ms were applied to the device during the relaxation process, with the time interval between each of the two pulses Δt being 150 ms. As shown in Fig. 2a, the device stimulated by light to produce ΔPSC returned to its initial state after approximately 170 seconds. In the absence of positive gate piezoelectric stimulation, the Au/ITO-graphene/Au artificial photoelectric synapse generated by the PPC effect ΔPSC still exists after 200 seconds (Fig. 2b). Therefore, applying a positive gate voltage to the device as electrical stimulation after writing information under light stimulation can accelerate the relaxation of PPC and achieve rapid information erasure. This is of great significance for ITO graphene artificial synapses in simulating short-term synaptic plasticity.

The generation of photogenerated electrons in ITO films under UV irradiation is the main reason for changing the conductivity of the device. To investigate the role of the ITO layer in the photoelectric regulation process, we measured the IPSC curves of Au/graphene/Au devices and Au/ITO-graphene/Au devices under 500 ms purple light stimulation (Fig. 3a). A schematic diagram of an Au/graphene/Au device is shown in Fig. S2 (ESI†). When a reading voltage of 0.1 V is applied at both the source and drain terminals, the Au/ITO-graphene/Au device produces a significant IPSC (approximately 1.4 μA) under purple light stimulation. However, Au/graphene/Au devices only have a weak current response (nA level). Compared with Au/graphene/Au devices, ITO-graphene synapses exhibit a significant improvement in light response amplitude. In addition, applying a reading voltage of 0.1 V to the source and drain electrodes results in a separate ITO layer resistance of approximately 71 kΩ, while the resistance decreases to 671 Ω after adding graphene as a conductive channel. This indicates that the photogenerated electrons in the ITO layer are the main factor inducing changes in the conductivity of the graphene layer, and the addition of graphene significantly increases the device's conductivity. By combining the two, a significant postsynaptic current response can be generated under light stimulation.

The bottom of the conduction band of graphene intersects with the top of the valence band at the Dirac point in the Brillouin zone.⁴³ When graphene is in an undoped state, i.e. intrinsic graphene, its Fermi level is at the Dirac point (Fig. S3a, ESI†). Fig. 3b displays the transfer and hysteresis curves of ITO-graphene devices at −15 to 15 V. During the process of gate voltage retrace, the Id Vg curve does not overlap, indicating that ITO graphene devices have overcome non-volatile charge movement under gate voltage stimulation. The graphene layer in the device serves as a conductive channel, and it can be seen that the Dirac voltage of graphene is positive, indicating that graphene is in a P-type doping state.³⁰ At this point, the Fermi level is below Dirac, and most of the charge carriers in graphene are holes (Fig. S3b, ESI†). There is a difference in the Fermi level between ITO and graphene before contact, and the Fermi level alignment occurs after contact. When artificial synapses are exposed to 405 nm purple light, oxygen vacancies (V_O) in ITO are stimulated by purple light pulses, leading to ionization and the generation of electrons (V_O → V⁺_O + e⁻ or V_O → V²⁺_O + 2e⁻).²³ As an amorphous metal oxide with PPC properties, ITO films generate photogenerated electrons under purple light stimulation. Changes in the electron concentration can cause changes in the Fermi level. In order to realign with the Fermi level of graphene, the photogenerated electrons in ITO will flow towards graphene and recombine with the majority carriers and holes in graphene. This electron hole recombination effect will reduce the hole concentration in graphene and increase the Fermi level. This is manifested as a decrease in the conductivity of graphene to point channels under purple light stimulation, resulting in the generation of IPSC (Fig. 3c).

Similar to the forgetting behavior of biological synapses, after removing light stimulation, the device's ΔPSC begins to continuously decrease and gradually returns to its initial state. The photogenerated electrons in the ITO layer combine with V⁺_O and V²⁺_O, causing them to return to the V_O state. Due to the fact that some photogenerated electrons have already flowed into the graphene layer and combined with holes under light stimulation, the decrease in ΔPSC is very slow. The method of applying a positive voltage to the gate can accelerate the decrease of ΔPSC. When a positive gate voltage is applied, the ITO Fermi level decreases due to the gate voltage, and electrons are induced from the ITO/graphene interface to the ITO layer. Electrons entering the ITO layer can recombine with V⁺_O and V²⁺_O to return to the V_O state, while the concentration of graphene holes that have lost electrons increases, thus quickly restoring the conductivity level before light stimulation (Fig. 3d).

The comparative experiment of the current response of Au/ITO-graphene/Au devices and Au/graphene/Au devices to purple light stimulation shows that the ionization of oxygen vacancies in ITO and the electron exchange at the ITO/graphene heterojunction interface play a major role in the optoelectronic modulation process of artificial synapses. To characterize the oxygen vacancies in the ITO thin film of the device, we analyzed and tested the ITO layer using X-ray photoelectron spectroscopy (XPS). Fig. S4 (ESI†) shows the full XPS spectrum of ITO thin films. As shown in Fig. 3e, Gaussian fitting was performed on the O 1s peak, resulting in O 1s (a) at 529.8 eV and O 1s (b) at 531.7 eV, respectively. O 1s (a) corresponds to O in the complete In₂O₃ lattice, while O 1s (b) corresponds to O in the lattice region with oxygen defects.⁴⁴ The peak area ratio of O 1s (a) to O 1s (b) is 0.86. Two peaks of In 3d_5/2 (a) and In 3d_5/2 (b) can be obtained by Gaussian fitting of In 3d_5/2 peaks (Fig. 3f). The In 3d_5/2 (a) peak with a binding energy of 444.1 eV corresponds to In₂O_3−X containing oxygen vacancies, and the In 3d_5/2 (b) peak with a binding energy of 444.75 eV corresponds to In₂O₃. The ratio of the peak area of In 3d_5/2 (a) to that of In 3d_5/2 (b) is 0.9. After the Sn 3d_5/2 peak is divided, two peaks can be obtained, namely Sn 3d_5/2 (a) corresponding to SnO (a binding energy of 486.3 eV) and Sn 3d_5/2 (b) corresponding to SnO₂ (a binding energy of 487.1 eV) (Fig. 3g).⁴⁵ The ratio of the peak area of Sn 3d_5/2 (a) to that of Sn 3d_5/2 (b) is 0.67. This indicates the presence of a large number of oxygen vacancies in the ITO layer of artificial synapses, which are the basis for the PPC phenomenon produced by ITO.

Oxygen vacancies are the main factor affecting the PPC effect of ITO; therefore, the oxygen vacancy concentration in the ITO photosensitive layer can greatly affect the photoresponse intensity of the device.⁴⁶ The ITO layer of artificial synapses is grown by magnetron sputtering, and the oxygen vacancy content of the ITO film can be changed by controlling the oxygen flux. Here, we prepared Au/ITO-graphene/Au devices using growth conditions with oxygen fluxes of 10 sccm and 20 sccm, respectively. As shown in Fig. S5 (ESI†), the artificial synaptic light response of the ITO layer with a 10 sccm oxygen flux is much lower than that of the ITO layer with a 20 sccm oxygen flux. This is because in the case of low oxygen flux, the reaction between In, Sn, and O₂ in the ITO layer is not sufficient, and most of them exist in low valence states. Low oxygen flux leads to poor crystallinity of the thin film, and the ITO layer exhibits strong metallicity, resulting in a decrease in the number of oxygen vacancies.⁴⁷ Therefore, a low oxygen flux is not conducive to enhancing the crystallinity and optical response of ITO films. In this study, unless otherwise specified, the magnetron sputtering oxygen flux of the ITO layer in the device was all 20 sccm.

According to memory time, biological synaptic plasticity is generally divided into short-term plasticity (STP) and long-term plasticity (LTP). STP is a temporary synaptic weight change in response to stimuli, and it can help the brain ignore some unimportant information. LTP is a long-term change in synaptic weight, which is the foundation of memory.⁴⁸ Double pulse facilitated paired pulse facility, PPF, is a manifestation of STP. Its specific manifestation is that when a synapse is stimulated by two consecutive light pulses, the second stimulus causes a stronger change in the synaptic weight than the first stimulus.^49,50 In Fig. 4a, paired 405 nm purple light pulses (1 s, 198 mW cm⁻²) were used to simulate the PPF behavior of biological synapses. Similar to the PPF behavior of biological synapses, the a2 pulse generated a larger IPSC amplitude compared to the artificial synapse under the a1 pulse. The calculation method for the PPF is as follows:

PPF = (I2 − I0)/(I1 − I0)

where I₀ is the initial current, I₁ is the instantaneous current value after the end of the first photostimulation, and I₂ is the instantaneous current value after the end of the second photostimulation. The time interval between two stimuli is denoted as Δt. Under the stimulation of paired 405 nm purple light pulses (1 s, 198 mW cm⁻²), the PPF index gradually decreases with the increase of Δt (Fig. 4b). This is due to the generation of photogenerated electrons in the ITO layer of the artificial synapse under the first purple light stimulation, which flow towards graphene and bind with holes, reducing the conductivity of graphene conductive channels. When the second purple light pulse arrived, the conductivity change caused by the first photogenerated electron had not yet returned to its initial state, so the second purple light pulse intensified the magnitude of the conductivity decrease.

In addition, Au/ITO-graphene/Au artificial optoelectronic synapses can also simulate the learning rules of partial synaptic plasticity, such as spike duration dependent plasticity (SDDP) and spike rate dependent plasticity (SRDP) can be achieved through Au/ITO-graphene/Au artificial photosynapses. Applying 405 nm purple light stimulation with different pulse widths to the device, the IPSC increases with the increase of the pulse width, and the device can simulate the SDDP rule (Fig. 4c). The reason why the device can simulate the SDDP rule is that under purple light stimulation, the ITO layer will continue to generate photogenerated electrons and transfer them to graphene, continuously forming a composite effect with holes, resulting in a continuous decrease in graphene conductivity. Therefore, the longer the device is exposed to ultraviolet light, the more the photogenerated electrons it generates, and the greater the impact on the device's conductivity. Similarly, SRDP behavior can be simulated by applying light stimuli of different frequencies (Fig. 4d). Here, ΔPSC is defined as the difference between the current value of the device and the initial value after applying 5 405 nm purple light pulses with a pulse width of 300 ms. It can be seen that during the process of increasing the pulse frequency from 0.2 Hz to 2.5 Hz, the ΔPSC increases with the increase of pulse frequency. This is because high-frequency pulse stimulation accelerates the accumulation of photogenerated electrons in artificial synapses, leading to a deeper degree of synaptic weight inhibition.

The human brain has the function of learning experience.⁵¹ Learning–forgetting–relearning is a typical experiential learning process in the human brain. The human brain has an experiential learning function. Through repeated training and learning of information from short-term memory, the human brain can transform short-term memory into long-term memory that lasts for several hours, days, or even a lifetime. Here, we apply 3 cycles of training to the device to simulate the experiential learning behavior of artificial synapses. During the simulation process, a 405 nm purple light pulse with a duration of 2 seconds is applied to the device, and the time interval between the two learning processes is 18 seconds. Between two training sessions, the IPSC amplitude of the device gradually decreases, which corresponds to the forgetting process of information by biological synapses. After three rounds of training and learning, the ITO graphene synapse is able to produce memory for a longer period of time in response to purple light stimulation. A schematic diagram of the human “learning–forgetting–relearning” process is shown in Fig. 5a.

In order to achieve the perception and long-term memory function of artificial vision systems on images, we prepared a 2 × 2-pixel imaging chip consisting of four ITO graphene optoelectronic synapses (Fig. 5b). In order to achieve LTP of ITO-graphene synapses for light signals, we applied 39 consecutive purple light pulses (405 nm, 200 ms, 198 mW cm⁻², Δt = 200 ms) to two diagonal devices. After 1000 s, the ITO-graphene synapses were still able to retain a portion of the IPSC. As the number of training sessions increases, the images are perceived and remembered by artificial synapses in the form of electrical conductivity changes (Fig. 5c). After a forgetting time of 100 seconds, the image formed by light stimulation still exists in the imaging chip (Fig. 5d). From this, it can be seen that the ITO-graphene photoelectric synapse can store light signals in the form of conductivity changes after multiple trainings.

3. Conclusions

In summary, we propose an artificial optoelectronic synapse based on an ITO/graphene heterojunction. The ITO layer with photoinduced sustained photocurrent characteristics is used as the photosensitive layer, and graphene with high carrier mobility is used as the conductivity modulation layer. Thanks to the strong electron exchange ability of the two, the synaptic weights of Au/ITO-graphene/Au artificial synapses can be bi-directionally modulated by light stimulation and electrical stimulation. Among them, purple light stimulation has an inhibitory effect on the synaptic weight, while gate electrical stimulation accelerates the recovery process of conductivity. Various biological synaptic plasticity characteristics, such as short-term memory, long-term memory, PPF, SRDP, SDDP, and experiential learning behavior, can be simulated by artificial synapses. Applying artificial synapses to artificial vision systems, after multiple trainings, a 2 × 2-pixel imaging chip composed of four artificial synapses has achieved the transformation of images from short-term memory to long-term memory. Combining photosensitive materials with low dimensional materials and introducing light signals into artificial synaptic devices are effective ways to achieve artificial vision systems.

4. Experimental section

Device manufacturing

Firstly, a negative photoresist was used for photolithography on an N-type Si/SiO₂ substrate (with a resistivity of 20–30 Ω and an oxide layer thickness of 100 nm). Subsequently, an imaged bottom electrode was obtained by evaporating and depositing a 5 nm titanium layer and a 100 nm gold layer, and peeling off the negative adhesive. The presence of a titanium layer enhances the adhesion of the gold electrode. Subsequently, a negative photoresist was spin coated on this basis and the functional layer was patterned. ITO was grown using a 99.99% purity indium tin oxide target material by magnetron sputtering in argon and oxygen atmospheres. The sputtering temperature is 100 °C, the sputtering power is 250 W, the oxygen flux is 20 sccm, the argon flux is 50 sccm, and the vacuum degree is 0.5 mTorr. The thickness of the ITO functional layer is 50 nm. Then, the single-layer graphene sample grown on copper foil at high temperature was spin coated with polyimide (PMMA) at a speed of 3000 rpm and cured at 100 °C for 5 minutes. The copper foil surface was brought into contact with a mixed solution of CuSO₄ + HCL to corrode copper foil, for a duration of approximately 8 hours. Subsequently, a glass slide was used to remove the PMMA film carrying graphene, and the film was washed three times with deionized water. Next, a graphical Si/SiO₂/ITO/Ti/Au sheet was used to pick up the PMMA film and let it dry naturally for 30 minutes. The sample was heated at 150 °C on a heating plate for 15 minutes to vitrify PMMA. Then, the sample was soaked in acetone for 30 minutes to remove PMMA and leave a graphene layer. Finally, a positive photoresist graphene layer was used and the excess of the layer was removed with a plasma debonding machine to obtain a patterned graphene layer. Graphene exists as a conductive channel between two electrodes, with a length of 70 μm and a width of 20 μm. The confocal microscopy images of Au/ITO-graphene/Au artificial synapses are shown in Fig. S6 (ESI†). Applying a reading voltage of 0.1 V to the source and drain electrodes, the resistance of the individual ITO layer becomes approximately 71 kΩ, while the addition of graphene as the conductive channel results in a resistance of 671 Ω.

Testing equipment

The electrical testing system used in this experiment includes a probe stage (with a microscope and other shooting systems), an arbitrary pulse function generator (Agilent 81150A), a source meter (Keithley 2612B), and corresponding programmable testing software. A LabRAM HR Evolution Raman spectrometer, with a laser source wavelength of 532 nm, was used for obtaining Raman spectra. An American Thermo Escalab 250Xi X-ray photoelectron spectrometer with monochromatic Al Kα as the radiation source (hν = 1486.6 eV) and a power of 150 W, was used for charge correction using polluted carbon C 1s = 284.8 eV. The Japanese OLYMPUS OLS4100 3D measurement laser microscope was used to obtain the confocal microscopy images of the device.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This work was supported by the National Key R&D Program of China under Grant No. 2022YFB3603003, 2023YFB3609300, 2021YFB3601200, and 2023YFE0203900, the National Natural Science Foundation of China under Grant No. 62022081, 62104042, and 61974099, the Tianjin Key R&D Cooperation Project of Tianjin and Chinese Academy of Science under Grant No. 22YFYSHZ00130 and the Youth Innovation Promotion Association of Chinese Academy of Sciences under Grant No. 2022109 and No. Y2022045. Fig. 1a was modified from the Servier Medical Art (https://smart.servier.com/), licensed under a Creative Common Attribution 3.0 Generic License (https://creativecommons.org/licenses/by/3.0/).

References

1. S. Park, M. Chu, J. Kim, J. Noh, M. Jeon, B. Hun Lee, H. Hwang, B. Lee and B. G. Lee, Sci. Rep., 2015, 5, 10123 CrossRef CAS PubMed .
2. M. Chu, B. Kim, S. Park, H. Hwang, M. Jeon, B. H. Lee and B.-G. Lee, IEEE Trans. Ind. Electron., 2015, 62, 2410–2419 Search PubMed .
3. P. A. Merolla, J. V. Arthur, R. Alvarez-Icaza, A. S. Cassidy, J. Sawada, F. Akopyan, B. L. Jackson, N. Imam, C. Guo, Y. Nakamura, B. Brezzo, I. Vo, S. K. Esser, R. Appuswamy, B. Taba, A. Amir, M. D. Flickner, W. P. Risk, R. Manohar and D. S. Modha, Science, 2014, 345, 668–673 Search PubMed .
4. X. Han, Z. Xu, W. Wu, X. Liu, P. Yan and C. Pan, Small Struct., 2020, 1, 2000029 CrossRef .
5. Y. Sun, J. Li, S. Li, Y. Jiang, E. Wan, J. Zhang, Y. Shi and L. Pan, Chip, 2023, 2, 100031 Search PubMed .
6. L. Gu, S. Poddar, Y. Lin, Z. Long, D. Zhang, Q. Zhang, L. Shu, X. Qiu, M. Kam, A. Javey and Z. Fan, Nature, 2020, 581, 278–282 CrossRef CAS PubMed .
7. R. Yang, Y. Wang, S. Li, D. Hu, Q. Chen, F. Zhuge, Z. Ye, X. Pi and J. Lu, Adv. Funct. Mater., 2023, 34, 202312444 Search PubMed .
8. F. Liao, Z. Zhou, B. J. Kim, J. Chen, J. Wang, T. Wan, Y. Zhou, A. T. Hoang, C. Wang, J. Kang, J.-H. Ahn and Y. Chai, Nat. Electron., 2022, 5, 84–91 CrossRef .
9. D. Panda, A. Dhar and S. K. Ray, IEEE Trans. Nanotechnol., 2012, 11, 51–55 Search PubMed .
10. S. Rajasekaran, F. M. Simanjuntak, S. Chandrasekaran, D. Panda, A. Saleem and T.-Y. Tseng, IEEE Electron Device Lett., 2022, 43, 9–12 CAS .
11. D. Panda, Y.-F. Hui and T.-Y. Tseng, Mater. Today Electron., 2023, 3, 100031 CrossRef .
12. D. B. Kastner and S. A. Baccus, Neuron, 2013, 79, 541–554 CrossRef CAS PubMed .
13. H. L. Park, H. Kim, D. Lim, H. Zhou, Y. H. Kim, Y. Lee, S. Park and T. W. Lee, Adv. Mater., 2020, 32, e1906899 CrossRef PubMed .
14. J. Wang, N. Ilyas, Y. Ren, Y. Ji, S. Li, C. Li, F. Liu, D. Gu and K. W. Ang, Adv. Mater., 2024, 36, e2307393 CrossRef PubMed .
15. Y.-B. Guo and L.-Q. Zhu, Chin. Phys. B, 2020, 29, 078502 CrossRef CAS .
16. L. Mennel, J. Symonowicz, S. Wachter, D. K. Polyushkin, A. J. Molina-Mendoza and T. Mueller, Nature, 2020, 579, 62–66 CrossRef CAS PubMed .
17. Z. D. Luo, X. Xia, M. M. Yang, N. R. Wilson, A. Gruverman and M. Alexe, ACS Nano, 2020, 14, 746–754 CrossRef CAS PubMed .
18. H. Yu, H. Wei, J. Gong, H. Han, M. Ma, Y. Wang and W. Xu, Small, 2021, 17, e2000041 Search PubMed .
19. S. Ge, F. Huang, J. He, Z. Xu, Z. Sun, X. Han, C. Wang, L. B. Huang and C. Pan, Adv. Opt. Mater., 2022, 10, 2200409 CrossRef CAS .
20. X. Shan, C. Zhao, X. Wang, Z. Wang, S. Fu, Y. Lin, T. Zeng, X. Zhao, H. Xu, X. Zhang and Y. Liu, Adv. Sci., 2022, 9, e2104632 CrossRef PubMed .
21. P. X. Chen, D. Panda and T. Y. Tseng, Sci. Rep., 2023, 13, 1454 CrossRef CAS PubMed .
22. I. Mönch, P. Feng, S. Harazim, G. Huang, Y. Mei and O. G. Schmidt, Nano Lett., 2009, 9, 3453–3459 CrossRef PubMed .
23. F. Oba, A. Togo, I. Tanaka, J. Paier and G. Kresse, Phys. Rev. B: Condens. Matter Mater. Phys., 2008, 77, 245202 CrossRef .
24. L. Q. Guo, H. Han, L. Q. Zhu, Y. B. Guo, F. Yu, Z. Y. Ren, H. Xiao, Z. Y. Ge and J. N. Ding, ACS Appl. Mater. Interfaces, 2019, 11, 28352–28358 CrossRef CAS PubMed .
25. F. Zhou, Z. Zhou, J. Chen, T. H. Choy, J. Wang, N. Zhang, Z. Lin, S. Yu, J. Kang, H. P. Wong and Y. Chai, Nat. Nanotechnol., 2019, 14, 776–782 CrossRef CAS PubMed .
26. F. Yu, L. Q. Zhu, H. Xiao, W. T. Gao and Y. B. Guo, Adv. Funct. Mater., 2018, 28, 1804025 CrossRef .
27. K. Liang, H. Ren, Y. Wang, D. Li, Y. Tang, C. Song, Y. Chen, F. Li, H. Wang and B. Zhu, IEEE Electron Device Lett., 2022, 43, 882–885 CAS .
28. P. K. Ivanoff Reyes, C.-J. Duan, Z. Xu,Y. Garfunkel, E. Lu and Y. Cheng, Appl. Phys. Lett., 2012, 101, 031118 CrossRef .
29. M. Lee, M. Kim, J.-W. Jo, S. K. Park and Y.-H. Kim, Appl. Phys. Lett., 2018, 112, 052103 CrossRef .
30. Y. X. Hou, Y. Li, Z. C. Zhang, J. Q. Li, D. H. Qi, X. D. Chen, J. J. Wang, B. W. Yao, M. X. Yu, T. B. Lu and J. Zhang, ACS Nano, 2021, 15, 1497–1508 CrossRef CAS PubMed .
31. M. T. Sharbati, Y. Du, J. Torres, N. D. Ardolino, M. Yun and F. Xiong, Adv. Mater., 2018, 30, 1802353 CrossRef PubMed .
32. J. Liang, X. Yu, J. Qiu, M. Wang, C. Cheng, B. Huang, H. Zhang, R. Chen, W. Pei and H. Chen, ACS Appl. Mater. Interfaces, 2023, 15, 9584–9592 CrossRef CAS PubMed .
33. X. Yu, C. Cheng, J. Liang, M. Wang, B. Huang, Z. Wang and L. Li, Adv. Funct. Mater., 2024, 34, 2312481 CrossRef CAS .
34. J. C. Meyer, A. K. Geim, M. I. Katsnelson, K. S. Novoselov, T. J. Booth and S. Roth, Nature, 2007, 446, 60–63 CrossRef CAS PubMed .
35. H. Suzuki, N. Ogura, T. Kaneko and T. Kato, Sci. Rep., 2018, 8, 11819 CrossRef PubMed .
36. C. Biswas, F. Gunes, D. L. Duong, S. C. Lim, M. S. Jeong, D. Pribat and Y. H. Lee, Nano Lett., 2011, 11, 4682–4687 CrossRef CAS PubMed .
37. M.-K. Dai, Y.-R. Liou, J.-T. Lian, T.-Y. Lin and Y.-F. Chen, ACS Photonics, 2015, 2, 1057–1064 CrossRef CAS .
38. P. Zhao, R. Ji, J. Lao, W. Xu, C. Jiang, C. Luo, H. Lin, H. Peng and C.-G. Duan, Org. Electron., 2022, 100, 106390 CrossRef CAS .
39. Y. T. Li, J. Z. Li, L. Ren, K. Xu, S. Chen, L. Han, H. Liu, X. L. Guo, D. L. Yu, D. H. Li, L. Ding, L. M. Peng and T. L. Ren, ACS Appl. Mater. Interfaces, 2022, 14, 28221–28229 CrossRef CAS PubMed .
40. Z. Li, G. Zou, Y. Xiao, B. Feng, J. Huo, J. Peng, T. Sun and L. Liu, Nano Energy, 2024, 127, 109733 CrossRef CAS .
41. G. Voglis and N. Tavernarakis, EMBO Rep., 2006, 7, 1104–1110 CrossRef CAS PubMed .
42. G. Zhang, J. B. Liu, H. L. Yuan, S. Y. Chen, J. H. Singer and J. B. Ke, J. Neurosci., 2022, 42, 6487–6505 CrossRef CAS PubMed .
43. G. N. Dash, S. R. Pattanaik and S. Behera, IEEE J. Electron Devices Soc., 2014, 2, 77–104 Search PubMed .
44. J. C. C. Fan and J. B. Goodenough, J. Appl. Phys., 1977, 48, 3524–3531 CrossRef CAS .
45. H. Han, Y. Zoo, S. K. Bhagat, J. S. Lewis and T. L. Alford, J. Appl. Phys., 2007, 102, 083705 CrossRef .
46. T. C. Gorjanc, D. Leong and C. Py, Thin Solid Films, 2002, 413, 181–185 CrossRef CAS .
47. S. Li, X. Qiao and J. Chen, Mater. Chem. Phys., 2006, 98, 144–147 CrossRef CAS .
48. A. H. Jaafar, R. J. Gray, E. Verrelli, M. O’Neill, S. M. Kelly and N. T. Kemp, Nanoscale, 2017, 9, 17091–17098 RSC .
49. S. Ham, S. Choi, H. Cho, S. I. Na and G. Wang, Adv. Funct. Mater., 2018, 29, 1806646 CrossRef .
50. G. Liu, C. Wang, W. Zhang, L. Pan, C. Zhang, X. Yang, F. Fan, Y. Chen and R. W. Li, Adv. Electron. Mater., 2015, 2, 1500298 CrossRef .
51. W. Wang, S. Gao, Y. Li, W. Yue, H. Kan, C. Zhang, Z. Lou, L. Wang and G. Shen, Adv. Funct. Mater., 2021, 31, 2101201 CrossRef CAS .

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d4tc01913j

This journal is © The Royal Society of Chemistry 2024