Research on Analog Memristors with Large On/Off Ratios Using 2D Van der Waals Metallic Cathodes

Academic Background

With the rapid development of artificial intelligence (AI) applications, traditional Von Neumann architectures are facing performance bottlenecks in data-intensive computing tasks. Neuromorphic computing is an emerging paradigm capable of handling such tasks with higher speed and efficiency. In this field, memristors are attracting significant attention as they allow for in-memory and analog computing. In particular, analog memristors with multiple conductance states can significantly improve the efficiency of neuromorphic computing. However, current analog memristors typically have small on/off ratios, which limits their application in high-precision weight mapping.

To address this issue, researchers have been exploring ways to enhance the on/off ratio of analog memristors while maintaining their multi-conductance state capability. Traditional memristors are mainly categorized into two types: valence-change-mechanism (VCM) memristors and electrochemical-metallization (ECM) memristors. VCM memristors rely on the migration of anions (usually oxygen ions) and exhibit good analog resistive switching characteristics, but they tend to have low on/off ratios and high off-state currents, leading to higher power consumption. ECM memristors, on the other hand, rely on the migration of metal ions and feature higher on/off ratios and lower power consumption. However, their switching behavior is often abrupt, making it difficult to achieve multiple conductance states.

Research Motivation and Problem Statement

This research aims to design an analog memristor with a high on/off ratio and multiple conductance states by introducing two-dimensional (2D) van der Waals (vdW) metallic materials as the cathode. Traditional ECM memristors commonly use inert metals (e.g., gold, platinum) as cathodes, which block ion migration and lead to abrupt switching behavior. By leveraging 2D van der Waals metallic materials (e.g., graphene or platinum ditelluride) as the cathode, this study exploits their high diffusion barrier properties to enable reversible intercalation and de-intercalation of silver ions, thereby achieving high on/off ratios and multi-conductance-state analog resistive switching.

Source of the Paper

This study was collaboratively authored by Yesheng Li, Yao Xiong, Xiaolin Zhang, Lei Yin, Yiling Yu, Hao Wang, Lei Liao, and Jun He, affiliated with institutions such as the School of Physics and Technology at Wuhan University, Wuhan University of Technology, and Hunan University. The paper was published online in Nature Electronics on September 27, 2024.

Research Methodology and Experimental Design

1. Device Design and Fabrication

The research team designed a memristor based on 2D van der Waals metallic cathodes, with a structure comprising silver (Ag) as the top anode, indium phosphorus sulfide (In2P3S9, IPS) as the switching medium, and graphene (Gr) or platinum ditelluride (PtTe2) as the bottom cathode. The fabrication process includes the following steps: - Cathode Preparation: Mechanically exfoliate multilayer graphene or platinum ditelluride and transfer it onto a silicon dioxide substrate. - Switching Medium Preparation: Mechanically exfoliate IPS nanosheets and transfer them onto the cathode. - Anode Preparation: Deposit silver and gold as the top anode via an electron-beam thermal evaporation system.

2. Electrical Performance Testing

The team measured the current-voltage (I-V) characteristics of the memristors using semiconductor parameter analyzers and pulse measurement systems. The results demonstrated that memristors with graphene and platinum ditelluride cathodes exhibited analog resistive switching behavior with an on/off ratio as high as 10^8 and more than 8-bit conductance states.

3. Material Characterization

The microstructure of the memristors was characterized using transmission electron microscopy (TEM) and energy-dispersive X-ray spectroscopy (EDS). The results showed that the intercalation and de-intercalation of silver ions in graphene and platinum ditelluride are reversible and do not cause significant structural damage.

4. Theoretical Calculations

Using density functional theory (DFT), the team calculated the diffusion barriers of silver ions in graphene, platinum ditelluride, and IPS. The results revealed that silver ions faced higher diffusion barriers in 2D van der Waals metallic materials, limiting their migration speed and enabling analog resistive switching behavior.

Key Research Findings

1. High On/Off Ratio and Multi-Conductance States

Memristors with graphene and platinum ditelluride cathodes achieved on/off ratios of up to 10^8 and more than 8-bit conductance states. This significantly outperforms traditional ECM memristors, which typically exhibit lower on/off ratios and abrupt switching behavior.

2. Low Power Consumption

Pulse measurements revealed that the memristors consume extremely low power, with energy consumption as low as 56 attojoules (aJ) per spike. This makes them highly promising for low-power neuromorphic computing applications.

3. Emulation of Synaptic Behavior

The team demonstrated the memristors’ synaptic plasticity, including long-term potentiation (LTP) and long-term depression (LTD). The memristors exhibited linear and symmetrical LTP/LTD behavior, which is critical for weight updates in neuromorphic computing systems.

Conclusions and Implications

This research successfully designed an analog memristor with a high on/off ratio and multiple conductance states by introducing 2D van der Waals metallic materials as the cathode. The memristors not only achieved on/off ratios as high as 10^8 and more than 8-bit conductance states but also exhibited extremely low power consumption and excellent synaptic plasticity. The team further implemented a chip-level simulation of a convolutional neural network (CNN) based on these memristors, achieving an image recognition accuracy of 91% on the CIFAR-10 dataset.

This study provides a novel approach to high-performance analog memristor design and lays the groundwork for the development of neuromorphic computing hardware. The proposed strategy holds potential for future applications in artificial intelligence, the Internet of Things, and edge computing.

Research Highlights

1. High On/Off Ratio and Multi-Conductance States: Achieved an on/off ratio as high as 10^8 and more than 8-bit conductance states through the use of 2D van der Waals metallic cathodes.
2. Low Power Consumption: Achieved ultra-low energy consumption of 56 attojoules per spike, suitable for low-power neuromorphic computing.
3. Synaptic Behavior: Demonstrated linear and symmetrical LTP/LTD behavior, ensuring reliable weight updates in neuromorphic systems.
4. Chip-Level CNN Simulation: Successfully simulated a CNN for image recognition with an accuracy of 91%, highlighting the memristors’ potential for high-precision computing tasks.

Additional Insights

The researchers explored the influence of different cathode materials and thicknesses on the performance of the memristors. It was found that memristors based on platinum ditelluride exhibited lower operation voltages and higher on/off ratios compared to those using graphene. Additionally, the team validated the applicability of this technique across various switching mediums, demonstrating the broad versatility of the proposed strategy.

This research offers a groundbreaking pathway for designing high-performance analog memristors and contributes significantly to the advancement of neuromorphic computing hardware development.