ECCS-EPSRC: Overcoming the Endurance Challenge in Energy-Efficient Atomic Memristors for AI and 6G Applications
Lead Research Organisation:
Liverpool John Moores University
Department Name: School of Engineering
Abstract
Achieving a low-carbon green footprint is currently one of the most urgent tasks in developing new ICT technologies, as energy required for the state-of-the-art artificial intelligence (AI), Internet-of-things (IoT), and next generations of mobile communications (6G) is skyrocketing. This is because the traditional von Neumann systems and CMOS technologies constantly need to shuttle massive amount of data between the physically-separated memory and processor units, hence consumes huge amounts of energy and data bandwidth for AI applications. Energy demand for mobile communication is also expected to increase exponentially, which needs to be substantially reduced for the 6G era.
In contrast to the conventional device technologies, memristors exhibit a programmable resistance with non-volatile memory and their memory market value has exceeded $621 million in 2021. Recent studies have shown that novel 2D memristive devices may also be exploited with significant advantages of 10,000x less energy consumption, both as synapse and neuron for advanced non-von Neumann AI computation and as RF switches for 6G mobile communication.
Our first demonstration in 2018 of non-volatile resistive switching (NVRS) in monolayer MoS2 and h-BN made a disruptive progress by substantially reducing the interelectrode distance to sub-nanometers and resulting in the thinnest memory cells with smaller switching voltages (~300 mV) and prospects of orders higher energy efficiency and density than existing memristors. We have since expanded the collection of two-dimensional (2D) atomic sheets showing NVRS to a dozen materials, indicating its potential universality in ultrathin non-metallic 2D materials for applications including high-density memory, neuromorphic computing, flexible nanoelectronics, and tera-hertz radio-frequency (RF) switches. The semiconductor industry can considerably benefit from the broad portfolio of 2D materials to advance electron devices, especially memory devices that is currently one of the key technology drivers.
One of the biggest challenges with atomic memristors is endurance. Typical values are less than 1000 cycles of resistance switching. This is reminiscent of the early days of oxide memristors. This challenge requires a rigorous experimental materials design and measurement study to gain operational insights that can translate to better engineered memristor devices (defects, interfaces, fields, etc.) and tailored testing protocol for orders of magnitude improved endurance.
We will achieve this ambitious goal through understanding the ageing, fatigue and degradation mechanisms in atomic memristors (atomristor) by using in-operando atomic resolution multi-probe scanning tunnelling microscope (STM) and developing novel electrical test protocols and characterization techniques, to provide the foundational knowledge for endurance improvement. We will also explore the use of interfacial layers and asymmetric electrodes as a novel degree of freedom that can control or impede the mobility of defects and result in more stable multi-level resistive switching. High-performance computing and communication devices such as synapses for neuromorphic computing, and zero-power non-volatile switches for 6G communication will be demonstrated. This project will lay the foundations for a new paradigm of deployable atomristors ubiquitously towards ultra-low energy AI and 6G systems with unprecedented efficiency and scalability.
In contrast to the conventional device technologies, memristors exhibit a programmable resistance with non-volatile memory and their memory market value has exceeded $621 million in 2021. Recent studies have shown that novel 2D memristive devices may also be exploited with significant advantages of 10,000x less energy consumption, both as synapse and neuron for advanced non-von Neumann AI computation and as RF switches for 6G mobile communication.
Our first demonstration in 2018 of non-volatile resistive switching (NVRS) in monolayer MoS2 and h-BN made a disruptive progress by substantially reducing the interelectrode distance to sub-nanometers and resulting in the thinnest memory cells with smaller switching voltages (~300 mV) and prospects of orders higher energy efficiency and density than existing memristors. We have since expanded the collection of two-dimensional (2D) atomic sheets showing NVRS to a dozen materials, indicating its potential universality in ultrathin non-metallic 2D materials for applications including high-density memory, neuromorphic computing, flexible nanoelectronics, and tera-hertz radio-frequency (RF) switches. The semiconductor industry can considerably benefit from the broad portfolio of 2D materials to advance electron devices, especially memory devices that is currently one of the key technology drivers.
One of the biggest challenges with atomic memristors is endurance. Typical values are less than 1000 cycles of resistance switching. This is reminiscent of the early days of oxide memristors. This challenge requires a rigorous experimental materials design and measurement study to gain operational insights that can translate to better engineered memristor devices (defects, interfaces, fields, etc.) and tailored testing protocol for orders of magnitude improved endurance.
We will achieve this ambitious goal through understanding the ageing, fatigue and degradation mechanisms in atomic memristors (atomristor) by using in-operando atomic resolution multi-probe scanning tunnelling microscope (STM) and developing novel electrical test protocols and characterization techniques, to provide the foundational knowledge for endurance improvement. We will also explore the use of interfacial layers and asymmetric electrodes as a novel degree of freedom that can control or impede the mobility of defects and result in more stable multi-level resistive switching. High-performance computing and communication devices such as synapses for neuromorphic computing, and zero-power non-volatile switches for 6G communication will be demonstrated. This project will lay the foundations for a new paradigm of deployable atomristors ubiquitously towards ultra-low energy AI and 6G systems with unprecedented efficiency and scalability.
