**About the Special Issue Editors**

**Byungjin Cho** received his MSc degree and PhD in materials science and engineering from the Gwangju Institute of Science and Technology in 2007 and 2012. He worked as a postdoctoral researcher at UCLA, USA, in 2012. After this, he was a senior researcher at Korea Institute of Materials Science (KIMS), Korea, from 2013 to 2016. Since 2017, he has been working as an assistant professor at Chungbuk National University. He has published dozens of SCIE papers on the synthesis of emerging 2D nanomaterials and nanoelectronics, such as sensor, transistor, and synapse devices.

**Yonghun Kim** received his MSc degree and PhD in materials science and engineering from the Gwangju Institute of Science and Technology in 2011 and 2016, and he was a postdoctoral researcher at the Korea Institute of Materials Science, Korea, in 2017. He has been a senior researcher at the Korea Institute of Materials Science since 2017. He has published dozens of papers on the synthesis of emerging two-dimensional nanomaterials, semiconductor integrated processes, neuromorphic synapse devices and advanced electrical characterization and reliability analysis.

### *Editorial* **Preparation and Properties of 2D Materials**

**Byungjin Cho 1,\* and Yonghun Kim 2,\***


Received: 19 March 2020; Accepted: 14 April 2020; Published: 16 April 2020

Since the great success of graphene, atomically thin layered nanomaterials—called two-dimensional (2D) materials—have attracted tremendous attention due to their extraordinary physical properties. In particular, van der Waals heterostructured architectures based on a few 2D materials, named atomic scale Lego, have been proposed as unprecedented platforms for the implementation of versatile devices with a completely novel function or extremely high performance, shifting the research paradigm in materials science and engineering [1]. Thus, diverse 2D materials beyond existing bulk materials have been widely studied for promising electronic, optoelectronic, mechanical, and thermoelectric applications. In particular, this Special Issue includes the recent advances in unique preparation methods, such as exfoliation-based synthesis and the vacuum-based deposition of diverse 2D materials, as well as their device applications based on their interesting physical properties. This editorial consists of the following two sections: Preparation Methods of 2D Materials and Properties of 2D Materials.

#### **1. Preparation Methods of 2D Materials**

Solution-based exfoliation methods for two-dimensional (2D) materials have been intensively investigated due to the ease of the process. In this regard, Zhang et al. investigated the cost-effective exfoliation method of multilayered 2DMoS2 nanosheets and quantum dots from natural SiO2-containing molybdenite in different solutions under mild ultrasonic conditions [2]. This simple method provides several advantages such as high yields, low cost and large-scale industrial perspectives compared with conventional methods. 2D-MoS2 nanosheets with dimensions of 50–200 nm were prepared. Furthermore, the excellent photoconductivity of the nanosheets under visible light was demonstrated in various solution conditions. Meanwhile, the conventional method to prepare saturable absorber materials uses the Langmuir–Blodgett (LB) technique, the merits of which include its low cost. In this respect, Wang et al. demonstrated a low-cost reflective WS2 saturable absorber (SA) on a silver-coated mirror for the first time [3]. By using the simple LB method, a large-area and highly uniform 2D-WS2-coated SA was successfully shown. Moreover, the optical saturation properties of WS2 SA were thoroughly analyzed, with the duration being around 409 ns and the highest peak power being 5.2 W. Thus, highly reflective WS2 SA, created using the simple LB method, could be used in a diverse optical modulator with a wavelength of 1.3 μm.

MoO3 is a promising material with well-recognized applications such as electronics, photocatalysis, electrocatalysis, batteries, and pseudocapacitors [4]. Among the various crystal structures of MoO3, the orthorhombic α-MoO3 provides unique 2D morphologies with layered structures. α-MoO3 has been conventionally obtained via the hydrothermal method or sputtering. However, such conventional preparation methods have faced some critical challenges related to substantial energy, complex equipment, and expert operational skills. Thus, Ramos et al. report a new preparation method to obtain

highly crystalline α-MoO3 using vapor-phase synthesis [5]. They obtained highly ordered multilayer α-MoO3 from molybdate using carbon nitride (g-C3N4) with a lamellar template. This simple method may be applied to electrocatalytic hydrogen evolution and ultrasensitive plasmonic biosensing.

For two-dimensional transition metal dichalcogenides (TMDCs), a uniform growth technique is required, especially for applications in electronics and optoelectronics. However, several critical challenges such as high growth temperature, limited growth area, and layer controllability still remain. Thus, Zhong et al. reported the simple growth method of 2D-MoS2 using a two-step process, combining radio frequency (RF) magnetron sputtering and the subsequent sulfurization process [6]. The growth temperature of this two-step process is lower than 600 ◦C, and the crystalline qualities are simply controlled by RF sputtering power. As RF plasma power increases from 10 to 150 W, the crystalline quality also increases, which is confirmed by the intrinsic peak intensities of the Raman spectrum. Recently, a new family of 1D nanomaterials with weak van der Waals interactions was also reported. Kim et al. successfully demonstrated the synthesis of a 1D semiconductor V2Se9 crystal using mass production via the simple transport preparation method [7]. The 1D-V2Se9 crystal exhibited weak van der Waals interaction and a nanoribbon structure. Also, scanning Kelvin probe microscopy (SKPM) analysis showed a variation in work function depending on the thickness of the V2Se9 crystal. This mass-production preparation method for 1D nanomaterials such as V2Se9 could be suitably applied to the metal contact of future van der Waals-based nanoelectronic devices.

#### **2. Properties of 2D Materials**

The electrical properties of 2D semiconducting materials are usually validated via the demonstration of field effect transistor (FET) devices. Thus, research themes involving the performance enhancement of FET devices have long attracted great attention, especially with respect to device junction optimization. In this context, Lim et al. proposed a novel FET structure consisting of a 2D MoS2/black phosphorous (BP) heterojunction, which shows a high on/off ratio of over 1 <sup>×</sup> 107, along with an extremely low subthreshold swing value of ~54 mV/dec and very low off current of ~fA level [8]. Interestingly, the low off current was attributed to the depletion region in the BP layer. Meanwhile, a TiO2 interfacial layer inserted between a metal and 2D TMDCs (MoS2 and WS2) can also lead to enhanced FET properties [9]. In addition, a stable electrical performance could be achieved under a gate bias stress condition, since the TiO2 interfacial layer serves as a Fermi level depinning layer, which reduces the density of the interface states.

The synthesis of p-type MoS2 is often essential for the complementary integration process using pand n-type 2D materials. Lee et al. reported that p-type semiconducting characteristics can be obtained via the addition of a dopant precursor of phosphorous pentoxide during the chemical vapor deposition synthesis process of MoS2 [10]. The p-doped monolayer MoS2 showed p-type conduction with a relatively low field effect mobility of 0.023 cm2/V·s and an on/off current ratio of 103, compared with the pristine n-type MoS2. The performance of the p-doped FET should be further improved. Along with neuron devices, artificial synapse devices have been recently considered as one of the most essential components in implementing a neuromorphic hardware system. Finding a physical parameter that precisely modulates synaptic plasticity is particularly required. Following this motivation, Kim et al. proposed a novel two-dimensional transistor architecture consisting of a NbSe2/WSe2/Nb2O5 heterostructure [11]. NbSe2, WSe2, and Nb2O5 function as a metal electrode, an active channel, and a conductance-modulating layer, respectively. Notably, the post-synaptic current was successfully modulated by the thickness of the interlayer Nb2O5, whose introduction facilitated the realization of reliable and controllable synaptic devices.

The unique optical properties of the 2D materials were intensively investigated. For instance, using thin semiconductor MoS2/ferroelectric lead zirconate titanate heterostructure films, reversibly tunable photoluminescence was demonstrated during ferroelectric polarization reversal using nanoscale conductive atomic force microscopy tips [12]. The spontaneous polarization of the ferroelectric thin films affects the optoelectronic behaviors of MoS2 indirectly via reversible electrochemical processes. Meanwhile, the Raman spectrum of BP transferred onto a germanium-coated polydimethylsiloxane flexible substrate was systematically studied [13]. The Raman spectra obtained from several BP layers with different thicknesses showed the clear peak shifting rates for the Ag1, B2g, and Ag2 modes. A study of the strain–Raman spectrum relationship was also conducted, showing a maximum uniaxial strain of 0.89%. The peak shifting of Ag1, B2g, and Ag2 caused by this uniaxial strain was clearly measured. In another optical study, a systematic investigation of photoluminescence (PL) and Raman spectroscopy of the transferred bilayer-stacked MoS2 were conducted, and compared with freestanding monolayer MoS2 [14]. The interlayer difference and spatial inhomogeneity of exciton and phonon performance are attribute to film–substrate coupling-induced strain and doping. Even surface fluctuations with a thickness of less than one atom layer could be easily identified by Raman and PL spectroscopy, offering useful information about the 2D van de Waals homostructure and heterostructures' effects on the optical properties of 2D materials.

The mechanical properties of the 2D materials are also interesting and attractive. In this regard, the tribological performance of two kinds of WS2 nanomaterials as additives in paraffin oil was investigated, showing that the friction and wear performance of paraffin oil can be greatly improved with the addition of WS2 nanomaterials, and that the morphology and content of WS2 nanomaterials have a significant effect on the tribological properties of paraffin oil [15]. For instance, paraffin oil with WS2 nanoflowers exhibited better tribological properties than that with WS2 nanoplates. The superior tribological properties of the WS2 nanoflowers were attributed to their special morphology, which contributes to the formation of a uniform tribofilm during the sliding process.

We hope this Special Issue will help 2D material researchers follow up the latest research trends and progress in the 2D research community.

**Author Contributions:** B.C. and Y.K. wrote and revised the paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT; Ministry of Science and ICT) (No. 2020R1A2C4001739) and Fundamental Research Program (No. PNK6990 and 6670) of the Korea Institute of Materials Science (KIMS). This research was also financially supported by the Ministry of Trade, Industry and Energy (MOTIE) and the Korea Institute for Advancement of Technology (KIAT) through the National Innovation Cluster R&D program (P0006704\_Development of energy saving advanced parts).

**Acknowledgments:** The Guest Editors highly appreciate the effort of all authors for publishing their works in this Special Issue. We are also grateful to the editorial assistants who have made the publication process of the Special Issue very smooth and efficient.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Artificial 2D van der Waals Synapse Devices via Interfacial Engineering for Neuromorphic Systems**

**Woojin Park 1, Hye Yeon Jang 1, Jae Hyeon Nam 1, Jung-Dae Kwon 2, Byungjin Cho 1,\* and Yonghun Kim 2,\***


Received: 22 November 2019; Accepted: 31 December 2019; Published: 2 January 2020

**Abstract:** Despite extensive investigations of a wide variety of artificial synapse devices aimed at realizing a neuromorphic hardware system, the identification of a physical parameter that modulates synaptic plasticity is still required. In this context, a novel two-dimensional architecture consisting of a NbSe2/WSe2/Nb2O5 heterostructure placed on an SiO2/p+ Si substrate was designed to overcome the limitations of the conventional silicon-based complementary metal-oxide semiconductor technology. NbSe2, WSe2, and Nb2O5 were used as the metal electrode, active channel, and conductance-modulating layer, respectively. Interestingly, it was found that the post-synaptic current was successfully modulated by the thickness of the interlayer Nb2O5, with a thicker interlayer inducing a higher synapse spike current and a stronger interaction in the sequential pulse mode. Introduction of the Nb2O5 interlayer can facilitate the realization of reliable and controllable synaptic devices for brain-inspired integrated neuromorphic systems.

**Keywords:** 2D heterostructure; WSe2; NbSe2; Nb2O5 interlayer; synapse device; neuromorphic system

#### **1. Introduction**

Continuous downscaling has stimulated the development of semiconductor technology for the last several decades, offering advantages, such as lower power consumption, higher integration, faster circuit operation, and reduced device cost per function. However, the side effects from continuous downscaling, to a size of less than 10 nm, limit the further development of the silicon semiconductor technology. This has motivated the exploration of novel computation systems beyond the conventional Von Neumann architecture that can overcome the downscaling limitations. Recently, due to the increasing need to implement sophisticated information processing system mimicking the human brain, the neuromorphic computing system has attracted a great deal of attention [1–5]. For the integrated neuromorphic systems, it is important to realize operations of complex and diverse functions implemented using a parallel architecture consisting of ~10<sup>11</sup> neurons and ~10<sup>15</sup> synapses. Additionally, the unit event should be simultaneously conducted using an extremely small amount of energy [6].

The artificial synapse device is considered to be an essential fundamental element for the emulation of biological neural networks [7]. The mechanism of operation for transmitting a spike input stimulus through the synapse can strengthen or weaken the synaptic weight, which is known as synaptic plasticity [8]. The synapse provides the functions of information processing and storage based on the spiking neural network. For this system, conventional solid-state electronics technology has been adopted for emulating the biological synapse function, in order to demonstrate a neuromorphic computing system [9]. In previous studies, conventional silicon-based complementary metal-oxide semiconductor (CMOS) technology was employed for demonstrating solid-state synapse devices, and a network consisting of 256 million configurable synapses and 1 million programmable spiking neurons was demonstrated [10]. The use of the 28-nm fully depleted silicon-on-insulator CMOS technology for 64k-synapse and 256-neuron architecture was also reported [11]. However, these CMOS-based devices are still unsuitable for realizing an artificial intelligence chip, because they cannot meet the requirements of higher integration density and lower power consumption. Si CMOS-based synapse device is based on the operation of complex logic circuits. This means that its power dissipation is essentially higher than that of other types, which is not satisfactory for emulating the biological synapse with an ultralow femtojoule energy consumption.

To eliminate the bottlenecks hindering the further development of neuromorphic computing systems, three-terminal artificial synaptic transistors, based on novel semiconductors have been studied to demonstrate synaptic functions. For instance, diverse semiconducting materials including carbon nanotubes, [12] nickelate, [13], and indium gallium zinc oxide (IGZO) [14,15] have been selected for the realization of synapse platforms. Meanwhile, two-dimensional (2D) transition dichalcogenides (TMDCs) are an intriguing nanomaterial layer for key elements of synaptic transistors due to their advantages of excellent intrinsic scalability, transparency, chemical robustness, and low power consumption [16–19]. In fact, several research groups have demonstrated the corresponding synaptic devices [20,21]. Meanwhile, a variety of oxide layers have been used as the conductance-tuning layers for synapse device applications. For example, phase change memory emulating synaptic behavior was demonstrated using a thin HfO2 interface layer [22]. Additionally, Deswal et al. reported an NbOx-based memristor, showing a gradual and continuous conductance change that is a prerequisite of a biological synapse device [23]. Nevertheless, it is still unclear what physical parameters can be used to precisely manipulate the synaptic functions. Thus, the use of a 2D heterostructure, combined with insulating oxide, can be an alternative approach for the development of energy-efficient artificial synapse devices.

In this work, we designed a vertically-stacked 2D metallic electrode NbSe2/semiconductor WSe2/interlayer Nb2O5 heterostructure placed on an Si/SiO2 substrate with the back-gate configuration. Here, WSe2 and Nb2O5 served as the active channel, and the conductance-tuning layer, respectively. Additionally, the NbSe2 electrode can provide excellent transistor switching characteristics due to a sharp 2D interface and the absence of the metal-induced gap states [24,25]. The post-synaptic current behavior can be modulated precisely by adjusting the thickness of the Nb2O5 layer, with a thicker Nb2O5 interlayer providing higher synapse spike current and strong interaction in paired pulse facilitation testing modes. The charge trapping/detrapping mechanism at the Nb2O5 defect states based on an energy band model was proposed. The novel 2D architecture will pave the way toward extreme integration for the development of the massively parallel neuromorphic circuitry system.

#### **2. Materials and Methods**

#### *2.1. CVD Synthesis of WSe2 and NbSe2*

A selenium (Se)-based semiconducting channel based on WSe2 and a metallic electrode based on NbSe2 were synthesized using a simple two-step process. First, WO3 and Nb2O5 thin films were individually deposited on an SiO2/Si wafer. The thicknesses of the WO3 and Nb2O5 thin films were ~3, and ~5 nm, respectively. This pre-deposited oxide layer on the wafer was directly loaded into the center of thermal furnace and vacuumed with a rotary pump system. Then, the thermal furnace was heated to the desired temperature (~1000 ◦C) under the flow of 5% hydrogen-balanced Ar gas (Ar/H2), while a selenium powder source was sublimated by heating to 500 ◦C. After a 1-h selenization process, the furnace was naturally cooled down to room temperature.

#### *2.2. Fabrication of 3-Terminal Synapse Device*

A heavily doped p-type Si substrate with SiO2 was cleaned by sonication in acetone, methanol, and iso-propyl alcohol (IPA) solution. To precisely tune the synaptic weight corresponding to the drain current, the charge trapping layer of the Nb2O5 thin film was deposited with different thicknesses using thermal evaporation. The thickness of Nb2O5 varied from 2.6 to 3.9 nm, as validated by the cross-sectional transmission electron microscopy (TEM) analysis. Then, the synthesized WSe2 semiconducting channel was transferred onto an SiO2/Si wafer using a poly(methyl methacrylate)-assisted transfer method and patterned using conventional photolithography. Finally, the NbSe2 metallic electrode was transferred for the formation of the NbSe2/WSe2 van der Waals heterojunction, in order to minimize the contact resistance [24,25].

#### *2.3. Electrical Characterization*

Basic electrical characterizations were carried out using a Keithley 2636B source meter (Keithley Instruments, Solon, OH, USA). The amplitude of the applied synaptic pulse, used to generate an excitatory post-synaptic current (EPSC), was 20 V and its duration was varied from 2 to 10 s.

#### **3. Results and Discussion**

Figure 1a shows a schematic of a biological neural network consisting of synapses and neurons. The most important trait of brain-inspired devices is their capability for efficient data processing using an extremely small amount of power in the networks with an astronomical number of synapses and neurons. The parallel network means that processing and storage of information occur simultaneously and do not follow the von Neumann computing paradigm. Therefore, a high device integration density and low energy consumption are crucial for a neuromorphic system. The operation of transmitting a spike input stimulus is illustrated in Figure 1b. The interaction of the pre- and post-synaptic activities affects the long-lasting connection strength, and long-lasting plasticity is considered to be the key mechanism of basic neuromorphic computation. Figure 1c shows the back-gate configuration of the WSe2 synapse transistor. The heavily-doped Si layer was used as the back-gate and NbSe2 was used as the source/drain. The Nb2O5 interfacial layer allows the fine-tuning of the conductance of the WSe2 transistor.

Figure 2a shows a schematic of the electrical measurements of the synapse device in the back-gate pulse system. Figure 2b shows the obtained cross-sectional high-resolution transmission electron microscopy images and the results of the energy-dispersive X-ray spectroscopy (EDS) analysis, thereby, clearly demonstrating the distinct film layers and sharp junction interfaces. The different stacking structures of WSe2-NbSe2, 2.6 nm Nb2O5-WSe2-NbSe2, and 3.9 nm Nb2O5-WSe2-NbSe2 were clearly observed and compared. The boundaries of each layer appeared to be atomically sharp and smooth without a significant interfacial gap. Five layers of NbSe2 and three layers of WSe2 were consistently observed for all of the samples, and the additional interfacial Nb2O5 layer was also clearly observed. The distributions of the W, Se, Nb, and O elements were obtained from the EDS elemental mapping images. The left panel of Figure 2b shows the WSe2-NbSe2 stack architecture without the Nb2O5 deposition. Since, both the NbSe2 and Nb2O5 films contain Nb atoms, the two separate Nb layers were observed only in the samples with the Nb2O5 interfacial layer, verifying the existence of Nb2O5. The middle panel of Figure 2b shows the results for the sample with a 2.6 nm Nb2O5 layer. The right panel of Figure 2b shows the sample with a 3.6 nm Nb2O5 layer. Figure 2c shows that the Raman spectra obtained for the as-synthesized 2D films support the presence of 2D materials, such as WSe2 and NbSe2, demonstrating the successful synthesis of the 2D nanomaterials via the chemical vapor deposition (CVD) technique. The Raman spectra of WSe2 and NbSe2 clearly display the in-plane vibrational modes of W-Se and Nb-Se (E1 2g: 250.3 and 243.2 cm−1) and the out-of-plane vibrational modes that arise from the motion the Se atoms (A1g: 258.5 and 230.6 cm−<sup>1</sup> for WSe2 and NbSe2). Furthermore, two distinct Raman peaks of WSe2 and NbSe2 with stacked device structure were also observed even after transfer process in Figure S1 in the Supplementary information.

**Figure 1.** (**a**) Biological neural network consisting of synapses and neurons. (**b**) Operational mechanism of the transmission of an input stimulus from pre-synapse to post-synapse. (**c**) Artificial synapse transistor comprised by vertically stacked NbSe2/WSe2/Nb2O5/SiO2/p+ Si, mimicking the function of bio synapse.

**Figure 2.** (**a**) Configuration scheme for the electrical measurements of the synapse transistor device. (**b**) Cross-sectional high-resolution transmission electron microscopy and energy-dispersive X-ray spectroscopy (EDS) elemental mapping images recorded from WSe2-NbSe2, 2.6 nm Nb2O5-WSe2-NbSe2 and, 3.9 nm Nb2O5-WSe2-NbSe2 (**c**) Raman spectra for WSe2, and NbSe2 that serve as the active channel and metallic electrode, respectively.

To compare the transfer characteristics of the WSe2-NbSe2 van der Waals hetero-junction devices with different Nb2O5 thickness, DC-mode-based double sweep measurements were performed, as shown in Figure 3a. The double sweep curves of the 2D heterojunction devices were obtained under varying values of able VBG in the range from 10 to −20 V at a fixed drain voltage of −5 V. The WSe2-based transistor showed typical p-type unipolar behavior, with a counterclockwise hysteresis loop, that may be ascribed to the confinement of the hole charges in the trap states induced by the Nb2O5 interlayer [26]. Additionally, the repeatability test of DC transfer double sweep curves, with different Nb2O5 thicknesses, were also shown in Figure S2 in Supplementary information. We also investigated the statistical distribution of the hysteresis window voltages, in order to validate the reliability of the data corresponding to the hysteresis behavior (Figure 3b). The average values of the hysteresis voltage for each device were measured to be ~5, 7, and 11 V, respectively. The value of the error bar was almost same for all of the devices. Thus, it is clear that a thicker Nb2O5 interlayer gives rise to a larger hysteresis window. The dependence of DC sweep speed on transfer curves was also depicted in Figure S3 in Supplementary information.

**Figure 3.** (**a**) Hysteresis behaviors of the two-dimensional (2D) WSe2-NbSe2 hetero-structure transistor devices with different Nb2O5 interlayer thickness. (**b**) Hysteresis window voltage as a function of the Nb2O5 interlayer thickness for the 2D WSe2-NbSe2 devices.

To elucidate the origin of the hysteresis of the 2D heterostructure transistors, the corresponding energy band model was proposed (Figure 4). We previously reported the positive effect of the combination of WSe2-NbSe2 with reduced contact barrier [24,25]. The conventional Richardson-Schottky equation was employed to calculate Schottky barrier,

$$I\_{DS} = A A^\* T^2 \exp\left[-\frac{\left(\Phi\_B - \sqrt{q^3 V / 4\pi \varepsilon\_o \varepsilon\_I d}\right)}{k\_b T}\right] \tag{1}$$

where *A* is the contact area, *A*\* is the effective Richardson constant, *T* is the temperature, Φ*<sup>B</sup>* is the Schottky barrier height, *q* is the electron charge, *V* is the applied forward bias, ε<sup>0</sup> and ε*<sup>r</sup>* are the permittivity of the vacuum and the oxide layer, respectively, *d* is the width of the interface barrier, and *Kb* is the Boltzmann constant. It was mentioned in the references that Schottky barrier at WSe2-NbSe2 contact is significantly lower than that at WSe2-metal(Pd) contact due to Fermi-level de-pinning. Therefore, the 2D WSe2-NbSe2 combination can be an excellent candidate for the fabrication of an energy-efficient low-power synaptic transistor, due to its low contact resistance. Recently, the new methodology for universal 2D material was reported to obtain Schottky barrier, suggesting more accurate calculation [27]. Holes are known to be the major carriers in both the semiconductor channel WSe2 and the metallic source/drain electrode NbSe2. Thus, only the hole charge transport was

considered in our proposed switching model. As shown in Figure 4a, the negative voltage applied to the back gate electrode (p +Si) shifts the corresponding Fermi level upward, accumulating hole charge near the Nb2O5-corresponding defect states. Under a negative gate bias, holes can be easily trapped in the defect states within the Nb2O5 interlayer, depleting the carriers in the WSe2 and leading to a decrease in the drain current. Meanwhile, when a positive voltage is applied to the gate, the Fermi level shifts downward, depleting the trapped holes in the Nb2O5 defects (Figure 4b). Simply put, the trapped holes will be released across the Nb2O5-WSe2 interface, leading to an increase in the drain current. Indeed, we experimentally proved that the amount of the trapped hole carriers is controlled by the Nb2O5 thickness.

**Figure 4.** (**a**) Device operation scheme and energy band model of the 2D heterostructure transistor device for describing the trapping behavior of the hole carriers at the negative gate bias condition; (**b**) device operation scheme and energy band model of the 2D heterostructure transistor device, corresponding to the process of the release of the trapped hole carriers at a positive gate bias condition.

To characterize the pulse response of the 2D heterostructure devices, we monitored the spike current response to the gate voltage pulses with the amplitude and duration time of 20 V, and 2 s, respectively (Figure 5a). In neuroscience, it is important to transfer electrical or chemical signal from pre-synapse to post-synapse. This is usually caused by the flow of positively charged ions. EPSC can be generated by the action of ions or electron flow in the neuromorphic system. The gate voltage for the EPSC was fixed at −20 V to give a fair comparison for each case. EPSC reaches the maximum value and then decays back to the initial current state. Interestingly, the spike was generated, even in the reference device without Nb2O5 layer. This might be because of the unintentional charge trap sites, which exist at the diverse interfaces (WSe2-NbSe2 and SiO2-WSe2). Our result showed that the peak values increase with the increasing thickness of the inserted Nb2O5 interlayer. Higher voltage pulses required long decay time to restore the synapse device to the initial current state, leading to stronger nonvolatile properties. The duration time of the pulse voltage, that is applied to the devices also affected the peak EPSC (Figure 5b). A longer pulse duration resulted in a higher peak EPSC. In a biological neural network, paired pulse facilitation is an important synapse parameter for determining synaptic plasticity, that is responsible for learning and memory processes [28]. As shown in Figure 5c, paired pulse facilitation is the phenomenon where the EPSC stimulated by the second spike is enhanced when the first spike is closely followed by the second spike [29,30]. Such essential

synapse behavior can be emulated using our 2D heterostructure transistor. Figure 5d shows the interaction of two sequential spikes for all of the devices. The interval time between the applied pulses was 2 s. The interaction between the output spike current of the control device without Nb2O5 was not observed, indicating the negligible synaptic weight modulation property. Meanwhile, the introduction of the Nb2O5 layer strengthened the interaction of the two spikes; more specifically, a thicker Nb2O5 interlayer induced a much greater current change at the second pulse mode. Such a strong tuning ability of the synaptic weight enhances the electrical plasticity of the artificial synapse device, and may improve the intelligence of the integrated neuromorphic system [12].

**Figure 5.** (**a**) Comparison of the excitatory post-synaptic current (EPSC) behavior of the 2D heterostructure devices with different Nb2O5 interlayer thicknesses. (**b**) Comparison of the EPSC data as a function of the spike duration time for the different 2D heterostructure devices. (**c**) Operation scheme of the synapse circuit for describing paired pulse facilitation that is stimulated by the application of two sequential pulses. (**d**) Comparison of paired pulse facilitation behavior for the different 2D heterostructure devices.

#### **4. Conclusions**

We demonstrated controllable synaptic plasticity with the WSe2/Nb2O5 heterostructure in the WSe2 back-gate device. The Nb2O5 layer served as the conductance-modifying layer and enabled precise modulation of the conductive states and their dynamic change. Essential synaptic functions (EPSC and paired pulse facilitation) were investigated in the WSe2/Nb2O5 heterostructure devices. In particular, the optimized thickness of the Nb2O5 layer strengthened the interaction in the synaptic weight, showing the largest post-synapse current. Thus, the facile one-step Nb2O5 layer deposition process, demonstrated in this work, is an effective approach for the realization of controllable synaptic devices.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2079-4991/10/1/88/s1, Figure S1: Raman spectrum of WSe2 and NbSe2 after transfer process, Figure S2: Repeatability test of DC transfer double sweep curves with different Nb2O5 thickness, Figure S3: The dependence of DC sweep speed on transfer curves.

**Author Contributions:** B.C. and Y.K. designed and conducted the experiments. H.Y.J. and J.H.N. and J.-D.K. supported the execution of the experiments and the data analysis. W.P., B.C., and Y.K. wrote the paper. B.C., and Y.K. supported and guided the experiments and analysis of results. Y.K. conceived the project and supervised the writing of the paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT; Ministry of Science and ICT) (No. 2017R1C1B1005076) and Fundamental Research Program (No. PNK6990) of the Korea Institute of Materials Science (KIMS). This research was also financially supported by the Ministry of Trade, Industry and Energy (MOTIE) and Korea Institute for Advancement of Technology (KIAT) through the National Innovation Cluster R & D program (P0006704\_Development of energy saving advanced parts).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
