Filamentary and Interface-Type Memristors Based on Tantalum Oxide for Energy-Efficient Neuromorphic Hardware

Abstract

To implement artificial neural networks (ANNs) based on memristor devices, it is essential to secure the linearity and symmetry in weight update characteristics of the memristor, and reliability in the cycle-to-cycle and device-to-device variations. This study experimentally demonstrated and compared the filamentary and interface-type resistive switching (RS) behaviors of tantalum oxide (Ta2O5 and TaO2)-based devices grown by atomic layer deposition (ALD) to propose a suitable RS type in terms of reliability and weight update characteristics. Although Ta2O5 is a strong candidate for memristor, the filament-type RS behavior of Ta2O5 does not fit well with ANNs demanding analog memory characteristics. Therefore, this study newly designed an interface-type TaO2 memristor and compared it to a filament type of Ta2O5 memristor to secure the weight update characteristics and reliability. The TaO2-based interface-type memristor exhibited gradual RS characteristics and area dependency in both high- and low-resistance states. In addition, compared to the filamentary memristor, the RS behaviors of the TaO2-based interface-type device exhibited higher suitability for the neuromorphic, symmetric, and linear long-term potentiation (LTP) and long-term depression (LTD). These findings suggest better types of memristors for implementing ionic memristor-based ANNs among the two types of RS mechanisms.

Synaptic and Gradual Conductance Switching Behaviors in CeO2/Nb–SrTiO3 Heterojunction Memristors for Electrocardiogram Signal Recognition

“…Kim et al obtained a high recognition rate of 93% for handwritten numerals in the TiN/TaO 2 /Pt device through adjusting the Schottky barrier height by the oxygen migration between electrodes and TaO 2. Nevertheless, the current transmission in the Schottky barrier-based metal–semiconductor memristors has no relaxation process, which is not conducive to simulating long-term plasticity of neural synapses. , In contrast, the separation and recombination of carriers in the Schottky barrier-based oxide-semiconductor heterojunction take some time. , On the other hand, CeO 2 memristors are widely studied because of their excellent characteristics such as a high ON/OFF ratio and high stability. − However, we hardly see studies on the application of CeO 2 devices in artificial synapses, which may be due to their abrupt conductance change. This disadvantage could be overcome by replacing the metal electrode with a doped semiconductor such as Nb–SrTiO 3 because the gradual conductance change could be achieved by charge trapping/detrapping at the CeO 2 film and the Nb–SrTiO 3 heterojunction …”

Molecular Structure Engineering of Polyelectrolyte Bilayer-Based Memristors: Implications for Linear Potentiation and Depression Characteristics

“…High storage efficiency, good fault tolerance, and neuromorphic computing will be critical components in the next generation of digital technology based on the advantages of low power consumption, which is a promising candidate for breaking through the von Neumann bottleneck. − In the nervous system, the nerve signals are transmitted from the presynaptic synapse membrane to the postsynaptic membrane through the synaptic cleft, the primary signal transmission and regulation unit. Therefore, many electronic devices were used for imitating synaptic behaviors, such as transistors and nonvolatile memory (NVM). − Particularly, the resistive random access memory (RRAM, a typical NVM) has attracted increasing attention due to the synapse-like structure. − RRAMs have the desirable functionalities of fast operation, low power consumption, good stability, and the capacity to be stacked in three-dimensional (3D), which are suitable for synaptic weight storage in neural networks. − Generally, the applied materials of the resistive switching (RS) layer, such as metal oxides, polymers, various molecular systems, and two-dimensional (2D) materials, determine the quality of the RRAM devices. Most RS materials operate according to the conducting filament (CF) mechanism as a result of the CF formation and rupture. − ,− However, the field concentration effect for CF formation, which is influenced by the synchronous Joule heating effect, inevitably results in slack potentiation and depression behaviors and the uncontrollable spike-timing-dependent plasticity (STDP) behavior .…”

“…Therefore, many electronic devices were used for imitating synaptic behaviors, such as transistors and nonvolatile memory (NVM). − Particularly, the resistive random access memory (RRAM, a typical NVM) has attracted increasing attention due to the synapse-like structure. − RRAMs have the desirable functionalities of fast operation, low power consumption, good stability, and the capacity to be stacked in three-dimensional (3D), which are suitable for synaptic weight storage in neural networks. − Generally, the applied materials of the resistive switching (RS) layer, such as metal oxides, polymers, various molecular systems, and two-dimensional (2D) materials, determine the quality of the RRAM devices. Most RS materials operate according to the conducting filament (CF) mechanism as a result of the CF formation and rupture. − ,− However, the field concentration effect for CF formation, which is influenced by the synchronous Joule heating effect, inevitably results in slack potentiation and depression behaviors and the uncontrollable spike-timing-dependent plasticity (STDP) behavior . Therefore, another RS mechanism, preferable for neuromorphic applications, should be investigated.…”

Ultrafast and Low-Power 2D Bi2O2Se Memristors for Neuromorphic Computing Applications

“…In general, when the linearity and symmetry of the conductance regulation are at a higher level, it will show a better performance of ANN. So, we introduce the parameter of asymmetric ratio (AR) calculated by the following eq 5: 53,54…”

“…Our result suggests good linearity and symmetry. 53,54 To test the performance of our device conductance regulation, we compare it with the ideal case. A random 1000 training images and 1000 test images were recognized in the simulated neural network in one epoch.…”