Nanophotonics 2020; 9(8): 2475–2486 Research article Min Zhanga, Zehui Fana, Xixi Jiang, Hao Zhu*, Lin Chen*, Yidong Xia*, Jiang Yin, Xinke Liu, Qingqing Sun and David Wei Zhang MoS2-based Charge-trapping synaptic device with electrical and optical modulated conductance https://doi.org/10.1515/nanoph-2019-0548 by incident light tuning, which further enables pattern Received December 25, 2019; revised January 31, 2020; accepted recognition with accuracy rate reaching 100%. Such January 31, 2020 experimental demonstration paves a robust way toward a multitask neuromorphic system and opens up potential Abstract: The synapse is one of the fundamental elements applications in future artificial intelligence and sensing in human brain performing functions such as learning, technology. memorizing, and visual processing. The implementation of synaptic devices to realize neuromorphic computing Keywords: MoS2; composite dielectric; nonvolatile and sensing tasks is a key step to artificial intelligence, memory; synaptic device; ANN. which, however, has been bottlenecked by the complex circuitry and device integration. We report a high-perfor- The rapidly evolving computing science and technology mance charge-trapping memory synaptic device based on have urged more unexploited data processing modes to be two-dimensional (2D) MoS and high-k Ta O –TiO (TTO) 2 2 5 2 fast tracked for the development and deployment in the composite to build efficient and reliable neuromorphic field of artificial intelligence [1–4]. Traditional computing system, which can be modulated by both electrical and systems based on von Neumann architecture have lost the optical stimuli. Significant and essential synaptic behav- edge and become inefficient when dealing with increas- iors including short-term plasticity, long-term potentia- ingly complex problems such as speech, image, and video tion, and long-term depression have been emulated. Such data processing, which often need huge physical resources excellent synaptic behaviors originated from the good and enormous energy consumption [5–7]. On the other nonvolatile memory performance due to the high den- hand, the human brain, which consists of a highly inter- sity of defect states in the engineered TTO composite. connected and massively parallel network, is an efficient The 2D synaptic device also exhibits effective switching information processing and storage system [8]. The 1011 neurons in the brain are responsible for the cognition and aMin Zhang and Zehui Fan: These authors contributed equally to this calculation, and the 1015 synapses are the critical units in work. signal delivery and processing [9, 10]. Synaptic behaviors *Corresponding author: Hao Zhu and Lin Chen, State Key Laboratory including short-term plasticity (STP), long-term potentia- of ASIC and System, School of Microelectronics, Fudan University, tion (LTP), and long-term depression (LTD) function in the Shanghai 200433, China, e-mail: [email protected] (H. Zhu); [email protected] (L. Chen). https://orcid.org/0000-0002-7145- biological brain for memory and actions, and the human 7564 (L. Chen); and Yidong Xia, College of Engineering and Applied learning behaviors are achieved by modulating the weight Science, Nanjing University, Nanjing, 210093, China, e-mail: xiayd@ of synapse through spike controlling. Therefore, great nju.edu.cn efforts have been made to realize an artificial brain-like Min Zhang: State Key Laboratory of ASIC and System, School of network based on the biological synaptic plasticity to Microelectronics, Fudan University, Shanghai 200433, China; process tremendous data in various forms with faster and School of Mathematics, Physics and Information Engineering, Jiaxing University, Jiaxing 314001, China speed and lower power consumption [11]. Zehui Fan, Xixi Jiang, Qingqing Sun and David Wei Zhang: State Key So far, different prototypes of synaptic devices have Laboratory of ASIC and System, School of Microelectronics, Fudan been developed emulating synaptic dynamics toward the University, Shanghai 200433, China implementation of brain-like computing and data process- College of Engineering and Applied Science, Nanjing Jiang Yin: ing systems, such as conductive bridging (CBRAM) [12–14], University, Nanjing, 210093, China Xinke Liu: College of Materials Science and Engineering, Shenzhen resistive change memory [15, 16], and phase change University, Shenzhen 518060, China. https://orcid.org/0000-0002- memory [17, 18]. Most of the reported synaptic devices are 3472-5945 based on passive and two-terminal structures similar to Open Access. © 2020 Hao Zhu, Lin Chen, Yidong Xia et al., published by De Gruyter. This work is licensed under the Creative Commons Attribu- tion 4.0 Public License. 2476 M. Zhang et al.: MoS2-based Charge-trapping synaptic device that of resistive random access memory (RRAM) due to the dielectric/2D material interface, where a large number difficulty in obtaining multiple and distinguished conduc- of defects and traps may exist because of the absence of tive states in transistor-based architecture [19]. In addi- dangling bonds on most layered semiconductors. This tion, conventional semiconductors or functional films can significantly deteriorate the switching capabilities in the synaptic devices are designed and engineered to and the control over the charge-trapping process. Here, perform limited functions. Nevertheless, this has become we report the fabrication and characterization of an arti- increasingly incapable of fulfilling multitask with the ficial synapse based on the flash-like CTM device using development of various portable electronic devices such 2D MoS2, a typical TMD semiconductor as the channel as image/video process and object detection. Therefore, and (Ta2O5)x(TiO2)1−x composite charge-trapping dielec- the integration of functional materials with features such tric. Back-gate FET geometry with engineered gate stack as optoelectronic properties as the active component in ensures sufficient optical contrast for the identification of synaptic device is very attractive because it will leverage MoS2 flakes with least dielectric/MoS2 interface traps. Dif- the advantages afforded by the new materials with the ferent from the limited trapping sites with various energy vast infrastructure of the synaptic electronics. levels in conventional single-material charge-trapping As a stable and repeatable physical mechanism, layer, the high density of defect states formed due to the charge trapping has been widely applied in building non- interdiffusion between the two kinds of high-k oxides volatile memory devices with good data retention. Such of Ta2O5 and TiO2 is responsible for the charge storage. memory is based on the trapping and detrapping of elec- Charge-trapping efficiency, programming/erasing speed, trons/holes in the charge storage medium changing the and retention capability have been greatly improved switching threshold. The charge-trapping memory (CTM) because of the PBCB (potentials at the bottom of con- in field-effect transistor (FET)–like three-terminal archi- duction band) between MoS2 and high-k composite. This tecture has also been used in realizing artificial synapses provides a solid basis for the plasticity study in which the as the amount of trapped charges can be effectively con- STP, LTP, and LTD behaviors have been successfully emu- trolled through the operation voltage and device structure lated. A near-perfect recognition rate has been achieved engineering to avoid the blurring between conductive in the artificial neural network (ANN) built in this work. states. But in terms of programming speed, conventional Furthermore, 2D materials are photosensitive [30–33] and charge-trapping synaptic devices are largely limited by have been explored in many optoelectronic applications the trapping/releasing process of the charges because of [34–36]. Thus, synaptic functions have been effectively the low trapping efficiency and unsuitable band offsets modulated by a series of optical operations owing to the with respect to Si in traditional nitride or oxide films. robust optoelectronic properties in MoS2 and the engi- Exploration focusing on the optimization of trapping neered dielectric stack. Such control and tuning of the mechanism and the engineering of the dielectric stack memory performance and synaptic behaviors pave attrac- is critical to improve the trapping efficiency and opera- tive ways toward the design of new synaptic devices for tion speed while maintaining good reliability. An addi- future neuromorphic computing and advanced sensing tional and effective strategy to achieve miniaturized yet applications. performance-enhanced memory is to utilize atomically The structure of the back-gate MoS2 synaptic FET is thin two-dimensional (2D) semiconductors as the active schematically shown in Figure 1A. The ultrathin MoS2 channel, which have been widely used in various fields channel is mechanically exfoliated from bulk and trans- [20–23]. Two-dimensional materials such as bilayer gra- ferred to the top surface of the back-gate stack. In the most phene, anisotropic black phosphorus (BP), and transition reported experimental work on MoS2 transistors, the exfo- metal dichalcogenides (TMDs) have been integrated in liated 2D flakes are generally transferred onto the relatively FET-based synaptic devices with reinforced gate control thick (~300 nm) SiO2 surface to gain sufficient contrast for and suppressed short-channel effects [24–27]. Tian et al. the optical identification