Research Article Cite This: ACS Appl. Mater. Interfaces 2019, 11, 20965−20972
www.acsami.org
Tunable Resistive Switching Enabled by Malleable Redox Reaction in the Nano-Vacuum Gap Xinglong Ji, Chao Wang, Kian Guan Lim, Chun Chia Tan, Tow Chong Chong, and Rong Zhao* Department of Engineering Product Design, Singapore University of Technology and Design, 8 Somapah Road, Singapore 487372, Singapore
Downloaded via BUFFALO STATE on July 24, 2019 at 06:41:28 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
S Supporting Information *
ABSTRACT: Neuromorphic computing has emerged as a highly promising alternative to conventional computing. The key to constructing a large-scale neural network in hardware for neuromorphic computing is to develop artificial neurons with leaky integrate-and-fire behavior and artificial synapses with synaptic plasticity using nanodevices. So far, these two basic computing elements have been built in separate devices using different materials and technologies, which poses a significant challenge to system design and manufacturing. In this work, we designed a resistive device embedded with an innovative nano-vacuum gap between a bottom electrode and a mixed-ionic−electronic-conductor (MIEC) layer. Through redox reaction on the MIEC surface, metallic filaments dynamically grew within the nano-vacuum gap. The nano-vacuum gap provided an additional control factor for controlling the evolution dynamics of metallic filaments by tuning the electron tunneling efficiency, in analogy to a pseudo-three-terminal device, resulting in tunable switching behavior in various forms from volatile to nonvolatile switching in a single device. Our device demonstrated cross-functions, in particular, tunable neuronal firing and synaptic plasticity on demand, providing seamless integration for building large-scale artificial neural networks for neuromorphic computing. KEYWORDS: nano-vacuum gap device, tunable resistive switching, malleable redox reaction, neuromorphic network, artificial synapse, artificial neuron
■
INTRODUCTION Neuromorphic computing has been widely regarded as a promising alternative to conventional computing.1−3 With a highly parallel neural network structure, neuromorphic machines collocate memory cells and processing units, thus promising to have a higher computing efficiency and consume lesser power compared to von Neumann machines.4−6 To construct a neuromorphic system, it is critical to build the basic functional elements: artificial synapses with synaptic plasticity behavior and artificial neurons with firing activity.7−9 Increasing efforts have been devoted in developing the two elements using nanodevices in order to build large-scale neuromorphic systems. Till now, artificial neurons and artificial synapses were developed separately by using different materials and mechanisms, including phase change,10,11 resistive switching, atomic switch,12 insulator-to-metal transition of Mott insulator,13,14 and spin-transfer torque magnetic memory.15−17 Among these approaches, volatile resistive switching has been actively studied to emulate synaptic plasticity or neuronal firing activity by utilizing the formation and spontaneous rupture of metallic filaments (MFs), providing a new approach to mimic the transient process in biological nervous systems, such as short-term plasticity and leaky integrate-and-fire.18−20 Despite significant advances, the use of diverse material systems and © 2019 American Chemical Society
mechanisms has posed a huge challenge to the construction of large-scale artificial neural networks because of the high complexity of system design and manufacturing. For neuromorphic computing, it is most desirable for a single device to exhibit both volatile and nonvolatile behavior, which would potentially have the ability to mimic the basic processing and learning operations of the mammalian brain with a unified material system and device structure. In this work, we designed a resistive device embedded with an innovative nano-vacuum gap structure that achieved both nonvolatile and volatile switching behaviors in a single device. The nano-vacuum gap was created by electrically driving a highly mobile metal layer into a solid electrolyte layer with high ion mobility. MFs dynamically grew within the vacuum gap through redox reaction, different from the standard resistive devices. The nano-vacuum gap provides an additional factor for controlling the redox reaction rate by tuning the electron tunneling efficiency to influence the continuous growth or degeneration of MFs. By varying the distance of the nano-vacuum gap, the evolving dynamics of MFs were fine Received: February 8, 2019 Accepted: May 21, 2019 Published: May 22, 2019 20965
DOI: 10.1021/acsami.9b02498 ACS Appl. Mater. Interfaces 2019, 11, 20965−20972
Research Article
ACS Applied Materials & Interfaces
Figure 1. Highly tunable switching behaviors with different gap distance. (a) Cross-point structure for nano-vacuum gap device demonstration. Inset: Optical micrograph of the 6 × 6 cross-point array; the effective switching area is 2 × 2 μm2. (b) Scanning electron micrograph of a nanovacuum gap device. The top image shows the top view; the bottom image shows the enlarged cross-sectional view. (c) Nonvolatile resistive switching I−V characteristic of a Gap-10 device. With different compliance currents, different resistance states were achieved. Inset: Logarithmic plotting. (d) Volatile-to-nonvolatile switching I−V characteristic of a Gap-30 device. For volatile switching with low compliance current, the high resistance state was restored automatically. For nonvolatile switching triggered by larger compliance current, a reset operation is essential to restore the high resistance state. (e) Volatile switching I−V characteristic of a Gap-50 device, showing typical threshold switching for the positive polarity sweep, and cut-off characteristic for the negative polarity sweep. indicating the formation of a nano-vacuum gap. The final nanovacuum gap device has a structure of TiW/nano-vacuum gap/MIEC/ Pt. Electrical Performance Characterization. The I−V characteristic of the nano-vacuum gap devices was measured using a Keithley 4200 semiconductor parameter analyzer (4200-SCS) with a Cascade Microtech Summit 11000 probe station. Considering that the area of the hysteresis loops is frequency-dependent,21 we performed all the direct current (dc) sweeping at the same sweeping rate. The pulse mode measurement of the nano-vacuum gap devices was carried out by using a Keithley 4225-PUM (pulse measurement unit). Microstructure Characterization. The microstructures of the nano-vacuum gap devices with different final states were characterized. We used a dual-beam-focused ion beam system to prepare cross-sectional samples of the nano-vacuum gap devices and conducted a progressive milling with small ion current to identify the areas of interest. The thickness of the cross-sectional samples was finally milled to ∼60 nm for high-resolution analyses purposes. Low beam current was applied during sample preparation to prevent the influence on the physical situation of Ag. A Tecnai F20 transmission electron microscope system was used for the microstructure characterization at an accelerating voltage of 200 kV. The bright field images were collected for the nano-vacuum gap region to identify the conductive filament. Electron-dispersive X-ray spectroscopy (EDXS) line scan was performed for material component and element distribution characterizations.
controlled, resulting in the co-existence of multiple conductive states and complex switching behavior in various forms in a single device. In addition, the nano-vacuum gap device exhibited varied retention times from several microseconds to tens of hours by applying different operation schemes. The abundant switching behaviors from volatile to nonvolatile enabled our nano-vacuum gap device to operate crossfunctionally from neuromorphic to memory applications. In particularly, we successfully demonstrated synaptic plasticity and neuronal firing activity in a single device, paving the way toward the seamless integration of large-scale artificial neural networks for neuromorphic computing.
■
EXPERIMENTAL DETAILS
Device Fabrication. The proposed highly tunable cross-functional device has an initial structure of TiW/Ag/Ge2Sb2Te5 (GST)/ Pt. A silicon wafer with an ∼1 μm thermal oxidized layer was used as the substrate. TiW with 35 nm thickness and Ag with different thicknesses (10, 30, and 50 nm) were deposited on the Si/SiO2 substrate using a magnetron sputtering system, serving the as bottom electrode and the active metal layer, respectively. Subsequently, 60 nm thick SiO2 was deposited on the bottom electrode, followed by photolithography and liftoff to form the insulation layer and contacting vias (2 × 2 μm2). Then, glassy GST and Pt were deposited and patterned together to form the functional layer and top electrode, respectively. After the fabrication, an electrical initialization process was performed to form the Ag−GST MIEC layer and the nano-vacuum gap. During the electrical formation of the vacuum gap, a positive voltage bias was applied from the bottom electrode to the top electrode. The output current was monitored in real-time as shown in Figure S1a,b. Initially, the current was in the microampere level and increased nonlinearly. At a certain voltage, the current dropped by more than 8 orders of magnitude to the picoampere level,
■
RESULTS AND DISCUSSION A cross-point array architecture was adopted for the demonstration of the cross-functional device, as shown in Figure 1a. In an as-fabricated device, a silver (Ag) layer was sandwiched between the GeSbTe (GST) layer and bottom electrode (TiW). To form the nano-vacuum gap, an electrical initialization operation was performed by applying a voltage 20966
DOI: 10.1021/acsami.9b02498 ACS Appl. Mater. Interfaces 2019, 11, 20965−20972
Research Article
ACS Applied Materials & Interfaces
the Gap-50 device also presented nonvolatile switching (Figure S3, Supporting Information). Considering the low power requirement of many electronic applications, our study was limited to a reasonable operation voltage range (i.e., within 2 V). The above study showed that the nano-vacuum gap provides an additional design flexibility for tuning the switching behaviors compared with the standard memristive devices. Besides nonvolatile resistive switching that has been actively studied for a memristor or memory applications, the incorporation of volatile switching into the same device has provided abilities for wider applications. In this part, we investigated the tunability of the nano-vacuum gap device on volatile switching. Here, a Gap-50 device was chosen as a demonstrator because of a larger tuning window of volatile switching. First, pulse I−V measurement was performed to determine the threshold voltages under different pulse widths (Figure S4). The pulse width-dependent threshold voltages in Figure 2a show that switch-on time exponentially decreased
bias to the bottom electrode and grounding the top electrode, which is schematically described in Figure S1a (Supporting Information). It should be noted that the electrical initialization in this work is to drive the active Ag layer into the GST layer to form the vacuum gap, which is different from the electroforming of a conductive path in conventional resistive memories or memristors. The glassy GST with overstoichiometry of chalcogen has been demonstrated to be very beneficial for Ag ionization and migration.22 Therefore, under the driving force of an external electric field, the Ag layer was fully dissolved into the glassy GST, forming the nano-vacuum gap in our device (Figure 1b). The I−V curve during electrical initialization is shown in Figure S1b (Supporting Information). By designing a device with different thicknesses of the Ag layer, the nano-vacuum gap distance and Ag concentration can be precisely controlled. The chemical dissolution of Ag significantly changed the electrical property of the GST layer. A pure amorphous GST material presents a very low conductivity because of the disorder-induced localized electronic state.23 However, with the doping of Ag, the GST in the effective area with local enrichment of Ag+ ions became mixed-ionic− electronic conductor (MIEC) with a much higher conductivity. The resistances of Ag−GST MIEC with different Ag concentrations are shown in Figure S1c (Supporting Information). In this paper, the Ag concentrations of all devices were fixed at 50 at. % to ensure high conductivity of the MIEC. Devices with different nano-vacuum gap distances were fabricated. We specially chose the three groups of devices with gap distance of 10, 30, and 50 nm (referred to as Gap-10, Gap30, and Gap-50) to present as they showed distinct switching behaviors. After electrical initialization of the vacuum gap, the devices were switched repeatedly by applying voltage bias to the top electrode (opposite to electrical initialization polarity). In Figure 1c, the Gap-10 device shows a nonvolatile resistive switching and switched to different resistance states by adjusting the compliance current. For a device with a 30 nm nano-vacuum gap, a two-step switching behavior was observed (Figure 1d). When a low compliance current (10 μA) was applied, the device only displayed the first step volatile switching with resistance automatically returning to the highresistance state. The device was maintained in the OFF state with high resistance during negative sweeping. When a high compliance current (100 μA) was applied, the second step nonvolatile switching was triggered, which needed a reset operation to switch back to the high-resistance state. This indicates that the Gap-30 device exhibited a volatile-tononvolatile transition within 2 V. By further increasing the nano-vacuum gap distance to 50 nm, a pure volatile switching was observed under 2 V with a typical hysteretic I−V during positive sweeping, whereas the current maintained a constant low during negative sweeping, as shown in Figure 1e. The endurance of the Gap-50 device is up to 106 cycles, while still maintaining stable volatile switching and a large ON/OFF ratio (Figure S2, Supporting Information). The above experimental observation revealed that the resistive switching behavior was highly associated with the nano-vacuum gap distance. Under low voltage operation (
