⇦Chinese Physics Letters, 2020, Vol. 37, No. 8, Article code 088501 Polymer-Decorated 2D MoS$_{2}$ Synaptic Transistors for Biological Bipolar Metaplasticities Emulation Yuhang Zhao (赵宇航), Biao Liu (刘标), Junliang Yang (阳军亮), Jun He (何军)*, and Jie Jiang (蒋杰)* Affiliations Hunan Key Laboratory of Super Microstructure and Ultrafast Process, School of Physics and Electronics, Central South University, Changsha 410083, China Received 17 May 2020; accepted 12 June 2020; published online 28 July 2020 Supported by the Central South University Research Fund for Innovation-Driven Program (Grant No. 2019CX024), the Natural Science Foundation of Hunan Province (Grant No. 2018JJ3652), the China Postdoctoral Science Foundation (Grant Nos. 2018M632985 and 2018T110839), and the Fundamental Research Funds for the Central Universities of Central South University (Grant No. 2018zzts333).
*Corresponding author. Email: jiangjie@csu.edu.cn; junhe@csu.edu.cn Citation Text: Zhao Y H, Liu B, Yang J L, He J and Jiang J et al. 2020 Chin. Phys. Lett. 37 088501    Abstract Biological bipolar metaplasticities were successfully mimicked in two-dimensional (2D) MoS$_{2}$ transistors via the implementation of two different MoS$_{2}$ surface decorations, poly (vinyl alcohol) (PVA) and chitosan bio-polymers. Interestingly, the depressing metaplasticity was successfully mimicked when the PVA bio-polymer was used as the surface decoration layer, whereas the metaplasticity of long-term potentiation was realized when the chitosan bio-polymer was taken as the surface decoration layer. Furthermore, the electronic band structures of the 2D MoS$_{2}$ devices with different surface decorations were further investigated using first-principles calculations for understanding the underlying mechanisms of such bipolar metaplasticities. These results will deepen our understanding of metaplasticity, and have great potential in neuromorphic computing applications. DOI:10.1088/0256-307X/37/8/088501 PACS:85.30.Tv, 85.30.-z, 85.35.-p © 2020 Chinese Physics Society Article Text Synaptic strength (weight), in the human brain, can be regulated by ionic species and concentrations precisely.[1,2] This phenomenon in neuroscience is considered as the synaptic plasticity, which plays an important role in memory and learning. Specifically, the long-term depression (LTD, i.e., weakening of the connection between neurons) and long-term potentiation (LTP, i.e., strengthening of the connection between neurons) are two different key characteristics of synaptic plasticity observed in CA1 (region of the hippocampus).[3] Most interestingly, synaptic plasticity itself can be dynamically modulated by historical activity in biological systems, which is often regarded to be metaplasticity.[4] Different from the normal synaptic plasticity, this primary feature highlights a prior stimulus that does not cause obvious changes in synaptic transmission, but rather changes the duration or magnitude of subsequent synaptic plasticity.[4,5] Dynamically linked the previous neural activity with the current response, metaplasticity is a particularly fascinating regulatory mechanism in terms of adaptive behaviors (memory and learning) and maladaptive behaviors, such as post-traumatic stress disorder and phobias.[6] Therefore, metaplasticity is thought to be a higher-order form of synaptic plasticity which can also be called “plasticity of synaptic plasticity”.[7] In general, unipolar metaplasticity exists in two main effects: either (i) facilitate subsequent LTP through activating group 1 metabotropic glutamate receptors (mGluRs),[8] or (ii) facilitate subsequent inhibition effect through activating N-methyl-d-aspartate receptors (NMDARs).[9] This intriguing regulatory mechanism in biological neural systems has inspired the use of neuromorphic devices to mimic the metaplasticity function across a broad range of research fields.[10,11] Many devices that mimic this intriguing synaptic metaplasticity have recently been successfully implemented in two-terminal memristors.[10–17] Some groups reported the effect of metaplasticity on the spike-timing-dependent plasticity (STDP) performance of an artificial synapse, and successfully mimicked the intriguing metaplasticity with a memristor by controlling the internal filament geometry configuration. Although these devices are very impressive, the development of metaplasticities based on novel device materials and architectures are still needed. As a promising candidate, field-effect transistors (FETs), having an extra gate terminal, can utilize hysteresis relaxation to achieve linear and dynamic plasticity,[18–21] possibly resulting in a higher-level metaplasticity in neuromorphic devices.[22] In this work, biological bipolar metaplasticities (BBMs) were successfully mimicked in 2D MoS$_{2}$ transistors for the first time using two different MoS$_{2}$ surface decorations, poly (vinyl alcohol) (PVA) and chitosan bio-polymers. Interestingly, the depressing metaplasticity was successfully mimicked when the PVA bio-polymer was used as the surface decoration layer, whereas the metaplasticity of LTP was realized when the chitosan bio-polymer was employed as the surface decoration layer. These BBMs represent a dynamic response coupled to prior activity that has the potential via synergistic reaction to provide sophisticated regulation of plasticity across space and time and promote homeostasis.[23] More importantly, these intriguing results can be theoretically explained via the first-principle calculation owing to ion adsorption effects on the MoS$_{2}$ interface. These findings are of great importance in driving more advanced neuromorphic computations in the future.
cpl-37-8-088501-fig1.png
Fig. 1. (a) A schematic picture of the PVA-decorated 2D MoS$_{2}$ synaptic transistor. (b) A top-view optical image of the device. (c) AFM measurement at the edge of MoS$_{2}$ flake. Inset: the schematic diagram of the layered atomic structure of multilayer MoS$_{2}$. (d) Raman spectroscopy of the exfoliated MoS$_{2}$ flake. (e) FTIR of the PVA film. Inset: the intrinsic molecular structure of the PVA. (f) The transfer characteristic curve of the MoS$_{2}$ transistor with a fixed bias of $V_{\rm DS} = 0.1$ V. (g) Magnified schematic diagrams of neurons (left panel) and two adjacent neurons connected by a synapse. (h) A typical EPSC response of the MoS$_{2}$ synaptic transistor triggered by a presynaptic spike (4 V, 10 ms). (i) EPSCs recorded in response to the stimulus train with different frequencies.
Figure 1(a) schematically shows the bottom-gate PVA-decorated 2D MoS$_{2}$ transistor (specific fabrication processes are provided in the following). The PVA powder is purchased from Sigma-Aldrich, and the corresponding molecular weight is about 85000–124000. In this work, to mimic an artificial synapse, PVA is chosen as the top surface decoration layer rather than a top dielectric layer. Actually, there are other amorphous polymers such as poly-$\alpha$-methylstyrene, poly-methyl methacrylate, and polystyrene that can be excellent candidates of gate dielectric materials according to the literature.[24–27] Two-dimensional MoS$_{2}$-based devices have high application prospects in many of the most advanced electronic applications due to the fact that the MoS$_{2}$ material has interesting electron properties, as well as special atomically scalable structure advantages.[28,29] Figure 1(b) shows the optical top-view micrograph of the MoS$_{2}$ FET where the two nickel electrodes are connected by an MoS$_{2}$ flake. By using atomic force microscopy (AFM), the thickness of as-fabricated MoS$_{2}$ flakes is estimated to be $\sim $9.7 nm, as shown in Fig. 1(c). The number of MoS$_{2}$ layers in our device should be $\sim $15 layers, since the thickness of monolayer MoS$_{2}$ is about 0.65 nm.[30] The inset of Fig. 1(c) shows the van der Waals layered atomic structure of MoS$_{2}$. The properties of the exfoliated MoS$_{2}$ are determined from their Raman spectra, as shown in Fig. 1(d). Both in-plane mode ($\sim $382 cm$^{-1}$) and out-of-plane $A_{1g}$ mode ($\sim $406 cm$^{-1}$) can be observed.[31] Compared with MoS$_{2}$ single crystal raw materials, good crystal properties are obtained in our exfoliated MoS$_{2}$ flakes.[31,32] Figure 1(e) displays the molecular structure (in the illustration) and Fourier transform infrared spectroscopy (FTIR) of the PVA membrane. Two major peaks, linked to C–H stretching at 2930 cm$^{-1}$ and O–H stretching at 3432 cm$^{-1}$, are clearly observed.[32] Moreover, other two peaks (1096 cm$^{-1}$ and 1637 cm$^{-1}$) are originated by C–O and C=O stretching, respectively.[33] Figure S1(a) in the supplementary material displays the output curve ($I_{\rm DS}$ vs $V_{\rm DS}$) of our neuromorphic device with the $V_{\rm GS}$ from $-2$ V to 2 V in steps of 1 V. The contact resistance is roughly estimated to be about $5\times 10^{4}\,\Omega$ according to the output curves. At present, for the non-saturation behavior of output curves, we tend to believe that it is possibly due to the short-channel effect.[34] More specifically, for a low gate bias in the output curve, this phenomenon is possibly caused by the barrier breakdown in the source/channel interface due to the strong $V_{\rm DS}$ electric field in a short channel. If a large gate bias is applied, the source/channel barrier would be lower due to the increase of electron concentrations in the channel, and the output curve exhibits more like a linear-resistor characteristic in such a case. Figure 1(f) shows the transfer curve ($I_{\rm DS} $–$V_{\rm GS}$), with a fixed $V_{\rm DS}$ of 0.1 V. A clear anticlockwise hysteresis is obtained when $V_{\rm GS}$ steps from $-2.0$ to 1.5 V and then back, which is potentially due to the presence of mobile protons in the PVA bio-polymer.[35] A key characteristic of our neuromorphic device is the ability to realize a higher-order form of synaptic plasticity (metaplasticity). A magnified schematic diagram of the neurons and synapses is shown in Fig. 1(g). When the inhibitory and excitatory signals arrive, the neurotransmitters from the presynaptic neuron are released into the synaptic cleft and then arrive at the postsynaptic membrane via diffusion, resulting in an action potential in the postsynaptic neuron.[1–3] Some special neurotransmitters (glutamate and NMDA) play important roles in various complex learning, emotional, and psychological functions during this process via weight changes in the postsynaptic currents.[1,2,36] Such a biological synapse can be mimicked in our 2D neuromorphic MoS$_{2}$ transistor, where the bottom gate serves as the presynaptic input terminal and the source/drain electrodes act as postsynaptic output terminals. Figure 1(h) shows the excitatory postsynaptic current (EPSC) of the bottom-gate MoS$_{2}$ transistor triggered by a presynaptic spike (4 V, 10 ms). The triggered EPSC attained its maximum value (96.1 nA) at the end of the input spike and then gradually returned to the resting current (3.9 nA), resembling a biological excitatory synapse.[2,3] The time constant could be well fitted through an exponential equation as follows: $$ I = I_{0}+I_{1} \exp (-t/ \tau),~~ \tag {1} $$ where $\tau$ is the retention time constant, $I_{1}$ is the EPSC amplitude, and $I_{0}$ is the ultimate amplitude of decay current. Figure S1(b) shows a good fitting for the decay characteristic. It was obtained that $\tau$ is about $\sim $9 ms, which means that the characteristic time of ion migration under excitatory signals is found to be $\sim $9 ms. At the same time, the paired-pulse facilitation (PPF), in biological systems, is a form of short-term synaptic plasticity, and is necessary for decoding temporal information in visual or auditory signals.[37] It is a phenomenon whereby the EPSC evoked by an impulse increases when the impulse closely follows a prior impulse.[2,3,37] This typical neuron behavior reappears in our neuromorphic device, as shown in Fig. S1(c), with the PPF behavior observed when two successive spikes (4 V, 10 ms) are applied to the bottom gate (presynaptic input). It is clear that the amplitude triggered by the second presynaptic spike (337.4 nA) is obviously larger than that triggered by the first one (220.3 nA), which is possibly due to the inability of the protons to migrate back to their original state when the interval is smaller than the ionic relaxation time.[22] In biological signal processing, dynamic rate/frequency-dependent filters may further extend above the PPF behavior.[38] An observable change in the synaptic efficacy may be present after particular temporal patterns in a neural system activity occur.[2,39] Interestingly, short-term synaptic facilitation may lead to high-pass temporal filtering, which can be mimicked in our device. Figure 1(i) illustrates the EPSC responses of the bottom-gate MoS$_{2}$ transistor to pulse stimuli with different frequencies that span 2–50 Hz. The pulse stimuli at each frequency consist of a series spike train (4 V, 10 ms). It is clear that the peak value increases significantly for the high-frequency stimuli, which illustrates that a high-pass frequency filter is realized in our MoS$_{2}$ neuromorphic device. Here, the frequency-dependent gain can be defined as the ratio between the tenth EPSC ($A_{10}$) and the first EPSC ($A_{1}$), as shown in Fig. S1(d). Interestingly, the gain increases linearly from $\approx $1.05 to $\approx $12.77 when the stimulus frequency changes from 10 to 50 Hz. Therefore, the results presented here indicate that the 2D MoS$_{2}$ synaptic transistor can act as a good dynamic high-pass filter for information transmission. Extracting information concerning fast changes in signal amplitude and rejecting that from slow fluctuations, that is, high-pass temporal filtering, are central to the utility of the jamming avoidance response. Such a high-pass filter function mimicked here is very interesting for computing the direction of motion of an electric image and therefore directional selectivity in artificial nervous systems.[2] In addition to the above-mentioned basic synaptic plasticity functions, the higher-order form of synaptic plasticity (metaplasticity) in biological systems is an important component of the brain learning mechanism.[5,9] This metaplasticity may increase the efficacy of both the memory and processing capacity of neurotransmitters, resulting in a great robustness and competitive characteristic of neuron weights.[4,40] The competitive winner-takes-all and Hebbian learning rules can be promoted via metaplasticity for neuromorphic computing applications (e.g., sound localization and pattern recognition).[39] Interestingly, bipolar metaplasticities, which are inspired by the existence of metaplasticity in biological systems, are successfully mimicked in our MoS$_{2}$ transistors. We use the SiO$_{2}$ as the bottom gate, so that the MoS$_{2}$ channel can be exposed to the outside and well decorated, which is beneficial to the emulation of our biological functions. Actually, the top-contact structure is facile for the fabrication process and that is also our next research direction.
cpl-37-8-088501-fig2.png
Fig. 2. (a) The schematic diagram for two different forms of metaplasticity. (b) Three continuous transient EPSC responses measured with the same spiking stimulus (4 V, 10 ms). (c) The EPSC triggered by the successive gate electric spikes with the same amplitudes (4 V, 10 ms) five times. (d) The EPSC triggered by applying sequential 11 spiking stimuli with an amplitude of 4 V, a width of 10 ms, and an interval of 0.6 s. (e) Typical depressing metaplasticity phenomenon. (f) Apply the priming spikes with different amplitude for examining its influence on the subsequent depressing metaplasticity. (g) Apply priming spikes with different widths for examining its influence on the subsequent depressing metaplasticity.
Figure 2(a) shows the schematic diagram for two different forms of metaplasticity whereby the special neurotransmitters (Ca$^{2+}$, glutamate, and NMDA) play a critical role.[41] The primary state of the synapses is displayed in the left panel, where Ca$^{2+}$, $\alpha$-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid (AMPA), NMDA, and other neurotransmitters are all outside of the synaptic membrane. A typical metaplasticity of LTP occurs when the increase in AMPA receptors leads to modulation of the neurotransmitter (AMPA $>$ NMDA) and more Ca$^{2+}$ influxes into the synaptic membranes.[7,42] However, increasing the NMDARs yields a varied neurotransmitter (NMDA $>$ AMPA) and less Ca$^{2+}$ influxes into the synaptic membranes would result in a depressing metaplasticity,[8] as shown in the right panel of Fig. 2(a). Interestingly, our 2D MoS$_{2}$ transistor with a PVA bio-polymer can successfully realize the depression metaplasticity. Figure 2(b) illustrates three transient EPSC responses that were measured repeatedly under the same amplitudes of the gate electric spikes (4 V, 10 ms). An interesting result can be observed from this figure: the EPSC responses decreased from 143.7 to 32.7 nA with the repeated stimuli, which indicates that our device has the typical characteristic of inhibition.[42] This phenomenon is further confirmed in Fig. 2(c), EPSC of MoS$_{2}$ neuromorphic transistor was reduced from 398.1 to 235.5 nA triggered by stimulus train with 10 pulses (4 V, 10 ms) (continuously repeated the measurements with five times). The EPSC response was then triggered using eleven sequential stimulus spikes (spike amplitude: 4 V, spike width: 10 ms, spike interval: 0.6 s), as shown in Fig. 2(d), to further study these inhibition effects. An interesting result can be observed from this figure, with continuous spiking stimuli, it is clear that the resulted EPSC gradually decrease from 326.1 to 60.6 nA and get saturated, as shown by the red dashed line in Fig. 2(d). This phenomenon indicates that the transient dynamic of our device exhibits an obvious inhibition memory. In biological systems, such a phenomenon is thought to play an important role in subsequent memory formation and learning, reflecting the possible changes in the central nervous system related to lasting memory.[43,44] Based on the unique property of our device, a depressing metaplasticity is shown in Fig. 2(e). The control experiment, where a single-voltage pulse is applied (upper panel), shows the difference between metaplasticity and conventional synaptic plasticity. The EPSC immediately attains its maximum value (162.7 nA) via a presynaptic spike stimulus (4 V, 10 ms). The probes that are connected to the electrodes are then suspended ($\sim $3 min) to enable the device to return to its original equilibrium state. Conversely, a priming weak spike (0.8 V, 10 ms) triggers a 155.4 nA EPSC from the succedent main voltage pulse (4 V, 10 ms). The lower EPSC peak (155.4 versus 162.7 nA) is observed in our device, which indicates that the priming stimulus successfully regulated the synaptic plasticity. Furthermore, the synaptic weights of metaplasticity can be evaluated by the change of EPSC as follows: $$ w = (E_{1 }- E_{0})/ E_{0},~~ \tag {2} $$ where $w$ is the synaptic weight, and $E_{0}$ and $E_{1}$ are the relative EPSC values before and after the weak priming spike are applied, respectively. A weak priming stimulus can also generate an EPSC by itself in our devices, which is analogous to biological systems. It is generally difficult to distinguish between metaplasticity and conventional plasticity since the priming stimulus can sometimes change the ability to produce plasticity.[4,23] A study of the nervous system has reported that inhibition effect can be obtained when postsynaptic depolarization exceeds a critical value,[45] with the NMDAR playing an important role in inducing subsequent inhibition. The threshold of the plasticity relies to the cell's previous history of stimulus, according to Bienenstock, Cooper, and Munro (BCM) theory.[40] In other words, when the priming spike triggers postsynaptic depolarization, the subsequent inhibition will be induced. At the same time, the NMDA receptor activation lowers the threshold for inhibition.[9,44–46] Therefore, we applied priming spikes with different amplitudes to investigate their influence on subsequent inhibition effect, as shown by the empty circular symbols in Fig. 2(f). These results illustrate that the amplitude of the priming spike can effectively modulate the subsequent metaplasticity. A synaptic weight change of up to 9.3% is obtained when the priming spiking amplitude increases to 1 V. These experimental results also fit the following equation: $$ \left| w \right|=w_{1}+A_{1}{\exp}\Big(\frac{u}{u_{1}} \Big),~~ \tag {3} $$ where $w_{1}$ and $A_{1}$ are different weight constants, and $u_{1}$ is the characteristic voltage constant related to the subsequent metaplasticity. Here we obtain a best-fit solution for $u_{1 }\approx 0.28$ V, which indicates that the significant metaplasticity phenomenon can be observed when the priming spike is larger than $u_{1}$. Therefore, we define a threshold stimulus amplitude of $V_{\rm T1} \approx 0.28$ V to distinguish between the generations of typical metaplasticity and conventional plasticity. A larger pulse width of 100 ms is applied to the priming stimulus to further investigate the threshold characteristic of metaplasticity. Figure 2(g) shows the experimental data (empty circular symbols) and fitted results (solid line), with a good fit obtained when the synaptic weight change is $$ \left| w \right|=w_{2}+A_{2}{\exp}\Big(\frac{u}{u_{2}} \Big),~~ \tag {4} $$ where $w_{2}$ and $A_{2}$ are different weight constants, and $u_{2}$ is a characteristic voltage constant (best fit when $u_{2} \approx 0.24$ V), indicating that the metaplasticity threshold of $V_{\rm T2} \approx 0.24$ V is smaller than $V_{\rm T1}$ (0.28 V). This phenomenon reveals that the subsequent metaplasticity threshold decreases with the increasing priming spike width. Therefore, the metaplasticity threshold can be adjusted by changing the width of the priming stimulus, yielding the results that are very similar to the metaplasticity behavior in the biological nervous system. It has been reported that the psychological tactile spatial sensitivity of professional pianists is very different from that of non-musicians,[47] with long-term piano practice leading to a lower spatial discrimination threshold, which indicates a form of metaplasticity in professional pianists. This type of metaplasticity suggests somatosensory information processing that is beyond training-specific constraints, i.e., reduced spatial discrimination threshold through sustained training.[23,47]
cpl-37-8-088501-fig3.png
Fig. 3. (a) A schematic picture of the chitosan-decorated 2D MoS$_{2}$ synaptic transistor. (b) A top-view optical image of the device. (c) AFM measurement at the edge of MoS$_{2}$ flake. Inset: schematic diagram of layered atomic structure of multilayer MoS$_{2}$. (d) Raman spectroscopy of the exfoliated MoS$_{2}$ flake. (e) FTIR of chitosan film. Inset: the intrinsic molecular structure of the chitosan. (f) EPSC triggered by a presynaptic pulse (4 V, 10 ms).
In biological systems, in addition to the inhibition effect, the LTP is also another important mechanism closely related to basic memory and learning occurring in the central nervous system.[3] Our neuromorphic device, which is based on a 2D MoS$_{2}$ transistor with a chitosan bio-polymer, can successfully realize this important metaplasticity of LTP. The schematic diagram of the device structure is shown in Fig. 3(a), with an optical top-view micrograph of the MoS$_{2}$ device shown in Fig. 3(b), where two nickel electrodes (source and drain) are connected by a MoS$_{2}$ flake. The thickness of the MoS$_{2}$ flake is $\sim $11.7 nm via AFM, as shown in Fig. 3(c). Therefore, there should be $\sim $18 layers based on a 0.65 nm thickness per layer value.[23] The inset of Fig. 3(c) displays a schematic diagram of the van der Waals layered atomic structure of the MoS$_{2}$ flake. The properties of the exfoliated MoS$_{2}$ were determined from the Raman spectra, as shown in Fig. 3(d). Both the in-plane (380 cm$^{-1}$) and out-of-plane $A_{1g}$ ($\sim $404 cm$^{-1}$) modes can be observed in the MoS$_{2}$ flake, indicating that our exfoliated MoS$_{2}$ flake has good crystal properties compared to a single MoS$_{2}$ crystal.[31,32] The FTIR spectra of the chitosan membrane reveal its major chemical bonds and molecular structure, as shown in Fig. 3(e). The peak at $\sim $3440 cm$^{-1}$ is attributed to the O–H group,[48] and the characteristic absorption band at 1660 cm$^{-1}$ is due to N–H bending vibrations. Two peaks are also visible at 1380 and 1080 cm$^{-1}$, which are due to the C–N and C–O groups, respectively.[48,49] The corresponding transfer curve ($I_{\rm DS} $–$V_{\rm GS}$) and output curve ($I_{\rm DS}$–$V_{\rm DS}$) are provided in Figs. S2(a) and S2(b), respectively. The field-effect mobility $\mu$ can be extracted from the equation $$ \mu =\frac{L}{C_{i}\times W\times V_{\rm DS}}\times \frac{dI_{\rm DS}}{dV_{\rm GS}},~~ \tag {5} $$ where $L=6$ µm is the channel length, $W = 9$ µm is the channel width, the SiO$_{2}$ dielectric capacitance is $1.2\times 10^{-8}$ F/cm$^{2}$ ($\varepsilon_{0}$ is the dielectric constant of vacuum, $\varepsilon_{\rm r}$ is the relative dielectric constant of SiO$_{2}$ ($\varepsilon_{\rm r} =3.9$), and $d$ is the thickness of the gate insulator (300 nm)). Therefore, the $\mu$ is calculated to be 5.83 cm$^{2}$/Vs for the pristine state, 69.3 cm$^{2}$/Vs for the –OH adsorption state (PVA-decorating device), and 55.1 cm$^{2}$/Vs for the H$^{+}$ adsorption state (chitosan-decorating device), respectively. The EPSC is triggered by applying a presynaptic spike (4 V, 10 ms) on the bottom gate of the MoS$_{2}$ transistor, with a reading voltage of $V_{\rm DS} = 0.1$ V applied between the drain and source electrodes during the test.
cpl-37-8-088501-fig4.png
Fig. 4. (a) The schematic diagram of metaplasticity of LTP. (b) A series of 20 single pluses are applied with an amplitude of 4 V, a width of 10 ms, and an interval of 1 s to trigger EPSC responses. (c) EPSCs changed with the increase of pulse number. (d) Eight transient EPSC responses measured repeatedly under the same amplitudes of electric spikes (4 V, 10 ms) for the gate terminal. (e) Apply the priming spike with different amplitude for examining its influence on the subsequent metaplasticity of LTP. (f) Apply the priming spike with different widths for examining its influence on the subsequent metaplasticity of LTP. (g) A schematic diagram revealing the relationship between the plasticity, metaplasticity of LTD, and metaplasticity of LTP.
Interestingly, the triggered EPSC at the end of the input spike attained a maximum value of 607.9 nA, but the resting EPSC did not return to its initial current state (528.3 nA) during the test, as shown in Fig. 3(f). From this figure, the resting EPSC difference before and after spike stimulus ($\Delta I_{\rm RE}$) is found to be 5.2 nA, indicating that the long-term memory (LTM) is successfully implemented by our neuromorphic bottom-gate MoS$_{2}$ device. LTM is the most important form of plasticity among the many synaptic properties of mammalian brains.[50] The electrical response of the neurons can be enhanced via continuous modification of the synaptic weight, resulting in biological LTM behavior. Our result provides clear evidence of LTM in our neuromorphic MoS$_{2}$ device. A similar LTM effect is also observed in the PPF measurements, as shown in Fig. S3(a). Here the EPSCs are triggered by applying a pair of presynaptic spikes (spike amplitude 4 V, spike width 10 ms, spike interval 10 ms), whereas the resting EPSC does not return to its initial current state, and a larger $\Delta I_{\rm RE}$ is observed (10.4 nA), which further confirms that our neuromorphic MoS$_{2}$ device can successfully mimic this intriguing LTM behavior. Compared with the potentiation effects of NMDARs priming on inhibition in area CA1, the induction of subsequent LTP is activated by group-1 mGluRs,[7] as shown in Fig. 4(a). The mGluR activation inhibits the Ca$^{2+}$-activated $K^{+}$ channel, which mediates the slow afterhyperpolarization current.[51] The phosphorylation state of an unknown regulatory protein (phosphoprotein, Pr) can regulate this process,[52] with the stimulation of phospholipase C and intracellular release of Ca$^{2+}$ activating protein kinase A and adenylyl cyclase, respectively, to induce subsequent LTP.[53] Most importantly, the trafficking of AMPARs to the extrasynaptic membrane can enhance induction of metaplasticity of LTP.[54] This process also depends on the enhancement of AMPAR-subunit mRNA trafficking to the synaptic sites. Here we successfully mimicked this higher-order form of synaptic plasticity (metaplastic facilitation of LTP) in the bottom-gate MoS$_{2}$ transistor using chitosan bio-polymer as the surface decoration layer. The neuromorphic device is triggered by a 20-spike stimulus train (spike amplitude 4 V, spike width 10 ms, spike interval 1 s), as shown in Fig. 4(b), with a gradual increase in EPSC that eventually becomes saturated after the successive stimuli. The resting current does not return to its initial state ($\sim $486.3 nA) when the spike is interrupted, but rather continues to maintain a high ESPC of 572.8 nA ($\Delta I_{\rm RE} \approx 86.5$ nA). Here the EPSC dependence with the pulse number can be summarized with the Fig. 4(c). A significant synaptic weight potentiation is observed with each stimulus that gradually becomes saturated after applying the 20-spike stimulus train, indicating that the chitosan-decorated MoS$_{2}$ device successfully mimics a strong LTP phenomenon. A good fit (solid line) to the experimental data (solid circular symbols) is obtained by $$ I=I_{0}+K_{0}{\exp}\Big(-\frac{N}{N_{0}} \Big),~~ \tag {6} $$ where $I$ is the EPSC increment, $I_{0}$ and $K_{0}$ are the different weight constants, and $N_{0}$ is the characteristic pulse number constant. The best-fit solution is obtained for $N_{0} \sim 7$, which indicates that the LTP phenomenon approaches saturation when $N_{0} > 7$. $I$ will reach its saturated state $I_{0}$ ($\sim $87.7 nA) as $N\to \infty$, which is the maximum LTP value that the device can attain. In mammalian brains, long-term potentiation reflecting the possible synaptic weight changes in the central nervous system is the most important form of plasticity among many synaptic properties. It plays important roles in various complex learning, emotional and psychological functions.[55] Such biological behavior can be further demonstrated via the successful application of an eight-spike stimulus train (4 V, 10 ms) to the bottom-gate terminal in Fig. 4(d). Interestingly, the resultant EPSC peaks increase from 594.8 to 632.2 nA, and the corresponding resting current also increases from 519.2 to 559.1 nA after these eight successive stimuli. Figure S3(b) further exhibits the EPSC response triggered by seven successive measurements of a ten-spike stimulus train, with a significant synaptic weight potentiation clearly observed during each spike-train measurement and the resultant EPSC peaks increasing from 675.3 to 696.4 nA. Furthermore, the resting EPSCs before and after the seven ten-spike stimulus trains also increase from 509.2 to 538.1 nA. These results strongly indicate that our MoS$_{2}$ neuromorphic transistor exhibits a clear LTP phenomenon. LTP in the human brain reflects the possible synaptic weight changes in the central nervous system, which are related to LTM.[56] The activation of mGluRs is therefore of great importance for mediating LTP-enhanced synaptic responses during this process.[7,56] Based on the unique property of our device, the metaplasticity of LTP is shown in Fig. S3(c). The upper panel illustrates that the EPSC attains its maximum value (570.7 nA) when triggered by a presynaptic spike (4 V, 10 ms). The probes connected to the electrodes are then suspended ($\sim $3 min) to ensure that the device will restore to its original equilibrium state. The lower panel illustrates that a weak priming spike (0.8 V, 10 ms) causes a 573.6 nA EPSC to be triggered by the succedent main voltage pulse (4 V, 10 ms). Our device attains a higher EPSC (573.6 versus 570.7 nA) compared to the synaptic plasticity in the upper panel. Here we employ a priming spike with different amplitudes (0.1–1 V) before the main spike stimulus (4 V, 10 ms) to determine this metaplasticity of LTP, as shown in Fig. 4(e). A good fit (solid line) to these experimental results (empty circular symbols) is obtained when the synaptic weight change is $$ w=w_{3}+A_{3}{\exp}\Big(\frac{u}{u_{3}} \Big),~~ \tag {7} $$ where $w_{3}$ and $A_{3}$ are different weight constants, and $u_{3}$ is the characteristic voltage constant related to the subsequent metaplasticity. We obtain a best-fit solution for $u_{3} \approx 0.25$ V, which indicates that the significant metaplasticity phenomenon can be observed when the priming spike is larger than $u_{3}$. This result also indicates that a threshold stimulus amplitude of $V_{\rm T3} \approx 0.25$ V exists in the generation of metaplasticity, such that the subsequent metaplasticity can be clearly observed once the amplitude of the previous weak spike exceeds this threshold. Previous studies have employed BCM theory to illustrate the activity-dependent regulation of the LTP threshold.[57] Pharmacological disinhibition or intracellular current injection stimuli in the biological nervous system can modify the depolarizing level and then induce LTP production,[23] while mGluR activation can lower the LTP threshold.[45] A larger pulse width of 100 ms is applied to the priming spike to realize this biological behavior. Figure 4(f) shows the experimental data (empty circular symbols) and fitted results (solid line), with a good fit obtained when the synaptic weight change is $$ w=w_{4}+A_{4}{\exp}\Big(\frac{u}{u_{4}} \Big),~~ \tag {8} $$ where $w_{4}$ and $A_{4}$ are different weight constants, and $u_{4}$ is characteristic voltage constant (best fit when $u_{4} \approx 0.23$ V), indicating that the metaplasticity threshold of $V_{\rm T4} \approx 0.23$ V is smaller than $V_{\rm T3}$ (0.25 V). This phenomenon reveals that the subsequent metaplasticity threshold decreases with increasing priming spike width. Therefore, the metaplasticity threshold can be adjusted by changing the width of the priming stimulus, yielding results that are very similar to the metaplasticity behavior in the biological nervous system. The subsequent metaplasticity threshold decreases with increasing priming spike width, indicating that the ability to produce metaplasticity of LTP is effectively enhanced. These results are very similar to the metaplasticity behavior in the nervous system, where the activated mGluRs can lower the LTP threshold.[45] Similar measurements are also performed to investigate the existence of this metaplasticity phenomenon using the bottom-gate MoS$_{2}$ transistor without surface polymer decoration (details are shown in Figs. S4 and S5), with neither the basic EPSC effect nor the metaplasticity phenomenon observed after five successive stimulus-train measurements. Learning in the human brain can be regulated through the change of synaptic weights, such as LTD and LTP.[58] Interestingly, both LTD and LTP can be dynamically modulated by the prior activity in the biological system (metaplasticity). Figure 4(g) shows the relationship between plasticity and metaplasticity. The black (control) curve shows the postsynaptic cellular response at different levels to the excitatory synapses in hippocampal CA1 pyramidal cells,[44] where $\theta_{\rm LTP}$ and $\theta_{\rm LTD}$ represent the postsynaptic firing thresholds required by afferent stimulation to produce LTP or LTD, respectively. The red and blue lines represent the effects of the priming stimuli on the subsequent LTD and LTP, respectively. Specifically, mGluR activation (blue curve) raises the LTD threshold and lowers the LTP threshold, whereas the NMDAR activation (red curve) lowers the LTD threshold.[45,46] Therefore, the ability of synaptic plasticity to undergo persistent potentiation or depression can be profoundly altered by previous neuronal activities, i.e., metaplasticity of LTP or depressing.[59] Compared to the unipolar metaplasticities, the main advantages of bipolar metaplasticities may include, but are not restricted to, synchronization control of neural network excitability, such that synaptic efficacy is maintained within the range of operation[23] Some diseases, such as depression, attention deficit hyperactivity disorder, post-traumatic stress disorder, and schizophrenia, can even be controlled due to the interesting regulatory mechanism.[60] Therefore, it is promising that the use of neuromorphic devices to mimic these fascinating biological behaviors could greatly expand artificial neural network applications and provide new bio-realistic neuromorphic computing approaches. In order to further investigate the underlying mechanism of BBMs in our neuromorphic devices, the electronic band structure of 2D MoS$_{2}$ with different surface polymer decorations (PVA or chitosan) can be investigated using the first principle calculations.[61] The current method is realized by the Vienna ab initio simulation package (VASP) code, depending on density functional theory (DFT).[62] In the calculation, a plane-wave cut-off of 450 eV is used. The Brillouin zone was sampled by a Monkhorst–Pack $6\times 6\times 1$ $k$-point grid. All of the atoms are allowed to relax until the Hellmann–Feynman force was below 0.01 eV/Å. In the self-consistent process, the energy convergence criterion is approximately $1\times 10^{-4}$ eV.[62,63] Figure 5(a) shows the charge density difference, the yellow and red represent electron depletion and accumulation. It is clearly observed that the surface charge distribution of MoS$_{2}$ is strongly affected by the surface functional groups (PVA or chitosan bio-polymer). More specifically, in the left panel of Fig. 5(a), many charged holes are trapped on the surface of layered MoS$_{2}$ crystal due to the adsorption of –OH in PVA bio-polymer. However, MoS$_{2}$ is a well-known n-type semiconductor and its electrons are majority carriers.[64] Therefore, the adsorbed holes indicate electron depletion on the MoS$_{2}$ surface, resulting in a continuous decrease in the channel current. Conversely, the H$^{+}$ in chitosan bio-polymer would induce many electrons in the surface of MoS$_{2}$, indicating that larger channel currents can be obtained as shown in the right panel of Fig. 5(a). Furthermore, the electronic band structures of the pristine state on the MoS$_{2}$ surface, the –OH, and H$^{+}$ adsorption states, can be systematically compared, as shown in Fig. 5(b). The typical semiconductor property of pristine MoS$_{2}$, with a bandgap of $\sim $1.68 eV, can be found in the left panel, which is in good agreement with a previous study.[65] It should be noted that, in the middle panel, the Fermi level in the band structure would shift toward the valence band when the MoS$_{2}$ surface is decorated with a PVA bio-polymer. Therefore, the low conductance state can be obtained in our neuromorphic device, which is in good agreement with the depressing metaplasticity. Interestingly, the Fermi level in the band structure will shift toward the conduction band when the MoS$_{2}$ surface is decorated with the chitosan bio-polymer, resulting in a high conductance state, as shown in the right panel of Fig. 5(b). This phenomenon is highly consistent with the metaplasticity of LTP which exhibits continuous enhancement after successive stimuli. A cross-sectional schematic diagram is proposed to explain such depression and potentiation effects via surface polymer decoration to intuitively understand the above-mentioned BBMs in an MoS$_{2}$ synaptic transistor, as shown in Fig. 5(c). Compared with the pristine state, the –OH group in the PVA bio-polymer can be regarded as a negatively charged group, which would induce electron reduction on the MoS$_{2}$ surface. However, the H$^{+}$ in the chitosan bio-polymer would result in electron accumulation on the MoS$_{2}$ surface, such that a higher conductance state is observed. The bottom-gate MoS$_{2}$ device without any surface polymer decoration is also fabricated to further confirm the rationality of our conjecture, with no clear metaplasticity observed (Fig. S5). Such functional-group induction effects from the opinion of first-principles calculations can be well utilized to understand the variation of channel conductance, i.e., the implementation of surface polymer decoration in the above-mentioned BBMs. Some special neurotransmitters (glutamate, and NMDA) play important roles in the generation of bipolar metaplasticities in biological nervous systems.[41] This intriguing phenomenon can allow the modified synaptic network to remain in a useful dynamic range and promote homeostasis.[23,66] Therefore, the demonstration of BBMs in our device is a key step in the development of next-generation neuromorphic electronics.
cpl-37-8-088501-fig5.png
Fig. 5. (a) The charge density difference of MoS$_{2}$ adsorbed OH ion and H ion. Left panel: MoS$_{2}$ surface functionalized by –OH group from PVA bio-polymer. Right panel: MoS$_{2}$ surface functionalized by H$^{+}$ from chitosan bio-polymer. (b) The electronic band structures of MoS$_{2}$ surface under the pristine state, –OH adsorption state, and H$^{+}$ adsorption state, respectively. (c) Cross-sectional schematic diagram of charge distribution of the MoS$_{2}$ transistor under the pristine state, –OH adsorption state, and H$^{+}$ adsorption state, respectively.
In summary, we have experimentally demonstrated the BBMs of 2D bottom-gate MoS$_{2}$ neuromorphic transistors using different surface decorations, PVA and chitosan bio-polymers. It is found that the amplitude and width of the priming spikes can significantly affect the EPSC response, thereby regulating the generation of metaplasticity in the MoS$_{2}$ neuromorphic transistors. Such features are very similar to those in biological nervous systems, where the degree and polarity of synaptic metaplasticity can be effectively modulated. Furthermore, to better understand the underlying physics of BBMs in our neuromorphic devices, the electronic band structure of 2D MoS$_{2}$ with different surface polymer decorations have been systematically investigated by the first-principle calculation. The realization of bionic metaplasticity may alleviate many network hierarchy problems associated with statistical changes between the neuronal and synaptic elements. This work will not only deepen our understanding of the potential mechanisms of nanoscale ionic devices but also provide new insight into neuromorphic engineering. Experimental Details. Multilayer MoS$_{2}$ flakes were first exfoliated from a bulk crystal using Scotch tape and then transferred onto a heavily doped silicon substrate with 300-nm-thick thermally-grown SiO$_{2}$. Ni ($\sim $30 nm thickness) was subsequently deposited to form the drain and source electrodes via photolithography (photoresist: PR1-2000A1), thermal evaporation (vacuum pressure: 10$^{-4}$ Pa), and lift-off process ($\sim $30 min). PVA powder (Sigma-Aldrich) was then dissolved in pure water (10 wt%) at 70℃ to prepare the PVA solution. Finally, the PVA solution was drop-casted onto the MoS$_{2}$ channel as the surface decoration layer. The fabrication process was duplicated for the second device, with the only difference so that a chitosan solution (2 wt% in acetic acid) was drop-casted onto the 2D MoS$_{2}$ channel as the surface decoration layer. PVA and chitosan are ideal material candidate gate dielectrics due to their low cost, non-toxicity, biodegradability, and excellent film-forming properties. FTIR of PVA and chitosan film were tested through an equipment of NICOLET 6700 FT-IR. The morphology and thickness of MoS$_{2}$ were measured by AFM equipment (Agilent 5500 SPM). The Raman spectroscopy was obtained by LABRAM ARAMIS. Using a semiconductor parameter characterization system (Keithley4200SCS), the electrical properties and biological synaptic functions of neuromorphic devices were tested at room temperature with the relative humidity of 50%. The presynaptic stimulus was applied on the bottom-gate electrode and the postsynaptic current was monitored using a small read voltage between the drain electrode and source electrode. Our biomimetic tests were all carried out after the solution was dried. References A Sub‐10 nm Vertical Organic/Inorganic Hybrid Transistor for Pain‐Perceptual and Sensitization‐Regulated Nociceptor Emulation2D MoS 2 Neuromorphic Devices for Brain-Like Computational SystemsG protein-coupled receptors control NMDARs and metaplasticity in the hippocampusMetaplasticity: the plasticity of synaptic plasticityThe neurobiology of learning and memoryLearning to learn – Intrinsic plasticity as a metaplasticity mechanism for memory formationMechanisms of heterosynaptic metaplasticityFacilitation of long-term potentiation by prior activation of metabotropic glutamate receptorsPriming of associative long-term depression in the dentate gyrus by θ frequency synaptic activityEmulation of synaptic metaplasticity in memristorsSynaptic Plasticity and Metaplasticity of Biological Synapse Realized in a KNbO 3 Memristor for Application to Artificial SynapseProgrammable Synaptic Metaplasticity and below Femtojoule Spiking Energy Realized in Graphene-Based Neuromorphic MemristorNanogenerator-induced synaptic plasticity and metaplasticity of bio-realistic artificial synapsesSynaptic Metaplasticity Realized in Oxide Memristive DevicesFull imitation of synaptic metaplasticity based on memristor devicesBipolar to unipolar mode transition and imitation of metaplasticity in oxide based memristors with enhanced ionic conductivitySynaptic plasticity, metaplasticity and memory effects in hybrid organic–inorganic bismuth-based materials2D electric-double-layer phototransistor for photoelectronic and spatiotemporal hybrid neuromorphic integrationVertical organic-inorganic hybrid transparent oxide TFTs gated by biodegradable electric-double-layer biopolymerTransient security transistors self-supported on biodegradable natural-polymer membranes for brain-inspired neuromorphic applicationsEnhanced performance of multilayer MoS2 transistor employing a polymer capping layerSynergistic Gating of Electro‐Iono‐Photoactive 2D Chalcogenide Neuristors: Coexistence of Hebbian and Homeostatic Synaptic MetaplasticityMetaplasticity: tuning synapses and networks for plasticityTemperature gradient controlled crystal growth from TIPS pentacene-poly(α-methyl styrene) blends for improving performance of organic thin film transistorsAir-stable solution-processed n -channel organic thin film transistors with polymer-enhanced morphologyPolymer additive controlled morphology for high performance organic thin film transistorsPoly(α-methylstyrene) polymer and small-molecule semiconductor blend with reduced crystal misorientation for organic thin film transistorsAnalysis of non-uniform hetero-gate-dielectric dual-material control gate TFET for suppressing ambipolar nature and improving radio-frequency performanceTotal Ionizing Dose Radiation Effects in the P-Type Polycrystalline Silicon Thin Film TransistorsSingle-layer MoS2 transistorsFrom Bulk to Monolayer MoS2: Evolution of Raman ScatteringSynaptic electronics: materials, devices and applicationsDSC and FTIR studies on Gellan and Polyvinyl alcohol (PVA) blend filmsHigh-performance short channel organic transistors using densely aligned carbon nanotube array electrodesBidirectionally-trigged 2D MoS2 synapse through coplanar-gate electric-double-layer polymer coupling for neuromorphic complementary spatiotemporal learningDeep-ultraviolet-triggered neuromorphic functions in In-Zn-O phototransistorsLightweight flexible indium-free oxide TFTs with AND logic function employing chitosan biopolymer as self-supporting layerFlexible Ionic-Electronic Hybrid Oxide Synaptic TFTs with Programmable Dynamic Plasticity for Brain-Inspired Neuromorphic ComputingThe Mechanisms and Functions of Synaptic FacilitationSparse coding with memristor networksTheory for the development of neuron selectivity: orientation specificity and binocular interaction in visual cortexLTP and LTDPoly(vinyl alcohol)-gated junctionless Al-Zn-O phototransistor for photonic and electric hybrid neuromorphic computationHomosynaptic long-term depression in area CA1 of hippocampus and effects of N-methyl-D-aspartate receptor blockade.Different voltage-dependent thresholds for inducing long-term depression and long-term potentiation in slices of rat visual cortexCalcium-Dependent But Action Potential-Independent BCM-Like Metaplasticity in the HippocampusSuperior tactile performance and learning in professional pianists: evidence for meta-plasticity in musiciansPreparation and inductive heating property of Fe3O4–chitosan composite nanoparticles in an AC magnetic field for localized hyperthermiaStructure–process–property relationship of the polar graphene oxide-mediated cellular response and stimulated growth of osteoblasts on hybrid chitosan network structure nanocomposite scaffoldsArtificial Synapses with Short- and Long-Term Memory for Spiking Neural Networks Based on Renewable MaterialsLong-Lasting Increase in Cellular Excitability Associated With the Priming of LTP Induction in Rat HippocampusMetabotropic Glutamate Receptor-Mediated Depression of the Slow Afterhyperpolarization Is Gated by Tyrosine Phosphatases in Hippocampal CA1 Pyramidal NeuronsGroup I mGluRs Increase Excitability of Hippocampal CA1 Pyramidal Neurons by a PLC-Independent MechanismExtrasynaptic Membrane Trafficking Regulated by GluR1 Serine 845 Phosphorylation Primes AMPA Receptors for Long-term PotentiationMetaplasticity contributes to memory formation in the hippocampusA synaptic model of memory: long-term potentiation in the hippocampusMetaplasticity: A new vista across the field of synaptic plasticitySynaptic plasticity: taming the beastMetaplastic Regulation of Long-Term Potentiation/Long-Term Depression Threshold by Activity-Dependent Changes of NR2A/NR2B RatioMolecular modulation of prefrontal cortex: Rational development of treatments for psychiatric disorders.First-principles calculations for defects and impurities: Applications to III-nitridesInterface engineering of CsPbI 3 -black phosphorus van der Waals heterostructureInfluence of the number of layers on ultrathin CsSnI 3 perovskite: from electronic structure to carrier mobilityTwo-dimensional atomic crystalsSingle-Layer MoS 2 Phototransistors
[1] Feng G, Jiang J, Zhao Y, Wang S, Liu B, He J and Wan Q 2020 Adv. Mater. 32 1906171
[2] Jiang J, Guo J, Wan X, Yang Y, Xie H, Niu D, Yang J, He J, Gao Y and Wan Q 2017 Small 13 1700933
[3] MacDonald J F, Jackson M F and Beazely M A 2007 Biochim. Biophys. Acta - Biomembr. 1768 941
[4] Abraham W C and Bear M F 1996 Trends Neurosci. 19 126
[5] Thompson R F 1986 Science 233 941
[6] Sehgal M, Song C, Ehlers V L and Moyer J R 2013 Neurobiol. Learn. Mem. 105 186
[7] Hulme S R, Jones O D, Raymond C R, Sah P and Abraham W C 2014 Philos. Trans. R. Soc. B. Biol. Sci. 369 20130148
[8] Cohen A and Abraham W C 1996 J. Neurophysiol. 76 953
[9] Christie B R and Abraham W C 1992 Neuron 9 79
[10] Zhu X, Du C, Jeong Y and Lu W D 2017 Nanoscale 9 45
[11] Lee T H, Hwang H G, Woo J U, Kim D H, Kim T W and Nahm S 2018 ACS Appl. Mater. & Interfaces 10 25673
[12] Liu B, Liu Z, Chiu I S, Di M, Wu Y, Wang J C and Lai C S 2018 ACS Appl. Mater. & Interfaces 10 20237
[13] Kim B Y, Hwang H G, Woo J U, Lee W H, Lee T H, Kang C Y, Nahm S 2017 NPG Asia Mater. 9 e381
[14] Tan Z, Yang R, Terabe K, Yin X, Zhang X and Guo X 2016 Adv. Mater. 28 377
[15] Wu Q, Wang H, Luo Q, Banerjee W, Cao J, Zhang X and Liu M 2018 Nanoscale 10 5875
[16] Cheng C, Li Y, Zhang T, Fang Y, Zhu J, Liu K and Huang R 2018 J. Appl. Phys. 124 152103
[17] Mazur T, Zawal P and Szaciłowski K 2019 Nanoscale 11 1080
[18] Jiang J, Hu W, Xie D, Yang J, He J, Gao Y and Wan Q 2019 Nanoscale 11 1360
[19] Hu W, Zheng Z and Jiang J 2017 Org. Electron. 44 1
[20] Hu W, Jiang J, Xie D, Wang S, Bi K, Duan H and He J 2018 Nanoscale 10 14893
[21] Guo J, Jiang J, Zheng Z and Yang B 2017 Org. Electron. 40 75
[22] John R A, Liu F, Chien N A, Kulkarni M R, Zhu C, Fu Q and Mathews N 2018 Adv. Mater. 30 1800220
[23] Abraham W C 2008 Nat. Rev. Neurosci. 9 387
[24] Asare-Yeboah K, Bi S, He Z and Li D 2016 Org. Electron. 32 195
[25] He Z, Shaik S, Bi S, Chen J and Li D 2015 Appl. Phys. Lett. 106 183301
[26] He Z, Chen J and Li D 2019 Soft Matter 15 5790
[27] He Z, Zhang Z, Asare-Yeboah K and Bi S 2019 J. Mater. Sci.: Mater. Electron. 30 14335
[28] Xu H F, Cui J, Sun W and Guo X F 2019 Chin. Phys. B 28 108501
[29] Liu Y, Liu K, Chen R S, Liu Y R, En Y F and Li B 2017 Chin. Phys. Lett. 34 018501
[30] Radisavljevic B, Radenovic A, Brivio J, Giacometti V and Kis A 2011 Nat. Nanotechnol. 6 147
[31] Li H, Zhang Q, Yap C C R, Tay B K, Edwin T H T, Baillargeat A and Olivier D 2012 Adv. Funct. Mater. 22 1385
[32] Kuzum D, Yu S and Wong H S 2013 Nanotechnology 24 382001
[33] Sudhamani S R, Prasad M S and Sankar K U 2003 Food Hydrocolloids DSC FTIR 17 245
[34] Sarker B K and Khondaker S I 2012 Appl. Phys. Lett. 100 023301
[35] Xie D, Hu W and Jiang J 2018 Org. Electron. 63 120
[36] Wang J, Chen Y, Kong L, Fu Y, Gao Y and Sun J 2018 Appl. Phys. Lett. 113 151101
[37] Feng G, Zhao Y and Jiang J 2019 Solid-State Electron. 153 16
[38] John R A, Ko J, Kulkarni M R, Tiwari N, Chien N A, Geok N, Leong W L and Mathews N 2017 Small 13 1701193
[39] Jackman S L and Regehr W G 2017 Neuron 94 447
[40] Sheridan P M, Cai F, Du C, Ma W, Zhang Z and Lu W D 2017 Nat. Nanotechnol. 12 784
[41] Bienenstock E L, Cooper L N and Munro P W 1982 J. Neurosci. 2 32
[42] Malenka R C and Bear M F 2004 Neuron 44 5
[43] Zhao Y, Feng G and Jiang J 2020 Solid-State Electron. 165 107767
[44] Dudek S M and Bear M F 1992 Proc. Natl. Acad. Sci. USA 89 4363
[45] Artola A, Bröcher S and Singer W 1990 Nature 347 69
[46] Hulme S R, Jones O D, Ireland D R and Abraham W C 2012 J. Neurosci. 32 6785
[47] Ragert P, Schmidt A, Altenmüller E and Dinse H R 2004 Eur. J. Neurosci. 19 473
[48] Zhao D L, Wang X X, Zeng X W, Xia Q S and Tang J T 2009 J. Alloys Compd. 477 739
[49] Depan D, Girase B, Shah J S and Misra R D K 2011 Acta Biomater. 7 3432
[50] Park Y and Lee J S 2017 ACS Nano 11 8962
[51] Cohen A S, Coussens C M, Raymond C R and Abraham W C 1999 J. Neurophysiol. 82 3139
[52] Ireland D R, Guevremont D, Williams J M and Abraham W C 2004 J. Neurophysiol. 92 2811
[53] Ireland D R and Abraham W C 2002 J. Neurophysiol. 88 107
[54] Oh M C, Derkach V A, Guire E S and Soderling T R 2006 J. Biol. Chem. 281 752
[55] Crestani A P, Krueger J N and Barragan E V 2019 Neuropsychopharmacology 44 408
[56] Bliss T V and Collingridge G L 1993 Nature 361 31
[57] Abraham W C and Tate W P 1997 Prog. Neurobiol. 52 303
[58] Abbott L F and Nelson S B 2000 Nat. Neurosci. 3 1178
[59] Xu Z, Chen R Q, Gu Q H, Yan J Z, Wang S H, Liu S Y and Lu W 2009 J. Neurosci. 29 8764
[60] Gamo N J and Arnsten A F 2011 Behav. Neurosci. 125 282
[61] Van de Walle C G and Neugebauer J 2004 J. Appl. Phys. 95 3851
[62] Liu B, Long M, Yang J and Cai M Q 2018 Appl. Phys. Lett. 112 043901
[63] Liu B, Long M, Cai M Q and Yang J 2018 J. Phys. D 51 105101
[64] Novoselov K S, Jiang D, Schedin F, Booth T J, Khotkevich V V, Morozov S V and Geim A K 2005 Proc. Natl. Acad. Sci. USA 102 10451
[65] Yin Z, Li H, Li H, Jiang L, Shi Y, Sun Y and Zhang H 2011 ACS Nano 6 74
[66]Artola A, Kamal A, Ramakers G M J, Biessels G J and Gispen W H 2005 J. Neurosci. 22 169