Neuromorphic Synapses with High Switching Uniformity and Multilevel Memory Storage Enabled through a Hf-Al-O Alloy for Artificial Intelligence

Abstract

Due to their high data-storage capability, oxidebased memristors with controllable conductance properties have attracted great interest in electronic devices for high integration density and neuromorphic synapses. However, high switching uniformity and controllable conductance of memristors during the conversion from a low (ON-state) to a high resistance state (OFFstate) have become essential for their implementation in neural networks. In this study, we fabricate a Pt/HfO 2 /HfAlO x /TiN memristor incorporating atomic-layer-deposited HfO 2 /HfAlO x high-k dielectric thin films as the active material to achieve excellent resistive switching performance with negligible parameter dispersion, multilevel conductance, and neuromorphic synapses for artificial intelligence (AI) systems. This two-terminal memristor exhibits a forming-free switching behavior with outstanding direct current endurance cycles (10 3 ), a high current ON/OFF ratio of >130, stable retention (10 4 s), and multilevel ON-and OFF-state, respectively. Also, memristor conductance/resistance could be modulated through current limits in the set-switching and stop voltage during the reset process, which is useful to acquire a trustworthy analogue switching conduct to mimic the biological neuromorphic synapses. The diverse features of synapses, such as potentiation, depression, spike-rate-dependent plasticity, paired-pulsed facilitation, and spike-time-dependent plasticity, are successfully mimicked in the Pt/HfO 2 /HfAlO x /TiN memristor. Furthermore, the experimental potentiation and depression data are employed for image processing of 28 × 28 pixels comprising 200 synapses. In the Modified National Institute of Standards and Technology database (MNIST), handwritten numbers can be successfully trained to recognize 6000 input images with a training accuracy of about 80%. This Hf-Al-O alloy-based memristor may enable high-density storage memory and realize controllable resistance/weight alteration as a neuromorphic synapse for AI systems.

Multi-level Cells and Quantized Conductance Characteristics of Al2O3-Based RRAM Device for Neuromorphic System

“…Figure 6 d shows MNIST pattern recognition simulation results by using the conductance results of Fig. 6 b and c [ 70 , 71 ]. The result of using Fig.…”

Enhanced Resistive Switching Performance through Air-Stable Cu2AgSbI6 Thin Films for Flexible and Multilevel Storage Application

“…It has been reported that the effective size of the conductive filaments plays an important role to achieve various LRS values at different I CC or stop voltages ( V stop ). , When the electric field effect is strengthened, the homogeneity of the charge carrier distribution in the switching layer improves resulting in the generation of stronger CFs, hence obtaining a degradation in the LRS values . Herein, the perovskites/ITO films are characterized by the conductive AFM (C-AFM) measurement to observe the morphology of the CFs at distinct bias voltages (a Pt-coated probe as the top electrode and ITO as ground) (Figure a–d).…”

“…34,35 When the electric field effect is strengthened, the homogeneity of the charge carrier distribution in the switching layer improves resulting in the generation of stronger CFs, hence obtaining a degradation in the LRS values. 36 Herein, the perovskites/ITO films are characterized by the conductive AFM (C-AFM) measurement to observe the morphology of the CFs at distinct bias voltages (a Pt-coated probe as the top electrode and ITO as ground) (Figure 4a−4d). As the positive bias voltage increases, the size and strength of the filaments are substantially improved (marked by green circles).…”

Radiation Response of HfOx-Based Resistive Random Access Memory (RRAM) Devices

“…The technology of RRAMs is being advanced to realize high-density (feature size in the range of 10–50 nm), high-speed, and low-power devices. , RRAM devices have a wide variety of applications, in various fields, as NVM elements with longer retention times − and as switches in digital logic systems for envisaging in-memory computation . The understanding of switching mechanisms , and the development of new methods to enhance switching speeds have been constantly evolving in recent times. , Based on the nature of switching mechanisms involved, these RRAM devices have also been used in neuromorphic systems ,− and as radiation detectors. − Radiation detectors have traditionally been employed to sense the nature of radiation and to quantify the absorbed dose . Metal-oxide-semiconductor field effect transistor (MOSFET) devices were first identified to have potential for realizing electronic sensors for detecting the radiation.…”