April 22 - 26, 2024
Seattle, Washington
May 7 - 9, 2024 (Virtual)

Event Supporters

2024 MRS Spring Meeting
SB03.10.03

High-Speed Edge Computing Implemented Using Reconfigurable Carbon Nanotube Transistor Memories

When and Where

Apr 25, 2024
11:15am - 11:30am
Room 436, Level 4, Summit

Presenter(s)

Co-Author(s)

Jingfang Pei1,Lekai Song1,Pengyu Liu1,Teng Ma2,Guohua Hu1

The Chinese University of Hong Kong1,Hong Kong Polytechnic University2

Abstract

Jingfang Pei1,Lekai Song1,Pengyu Liu1,Teng Ma2,Guohua Hu1

The Chinese University of Hong Kong1,Hong Kong Polytechnic University2
The constant data transportation from the edges to the centralized cloud computing infrastructure causes considerable constrains over the computational power and latency as well as the energy cost. Decentralization of the computation with the computational tasks distributed to the edges is an emergent solution to address the problem. Processing onsite data in local memories holds great promise to implement the edge computing. Here, we demonstrate edge computing using reconfigurable nonvolatile carbon nanotube transistor memory arrays, and prove high-speed, real-time video processing.<br/><br/>We fabricate the transistors from solution-sorted semiconducting single-walled carbon nanotubes. The transistors exhibit fast switching with the switching times and delays down to tens of nanoseconds, a large switching ratio of over 10<sup>5</sup>, and, particularly, a significant memory window of ~12 V arising from charge trapping in the sorting polymer. The above characteristics endow the transistors with highly-stabilized reconfigurable nonvolatile memory states and a high data processing speed. Owning to solution processing, the fabrication is wafer-scalable, the transistors exhibit uniform characteristic memory metrics, e.g. with 1.8% variation in the memory window, suggesting an industrial-scale manufacturing capability of the fabrication. Using the transistor memories, we design and implement an edge computing device with a convolution unit connecting to a differentiator, and demonstrate the application of the edge computing device in edge detection and motion track tasks of video streams. Particularly, the edge computing device successfully performs local video processing at a speed of 10,000 fps, exceeding the conventional high-speed cameras. Given the efficacy of the edge computing device, and the scalability of the fabrication, we envisage a promising prospect of realizing large-scale edge computing devices in implementing practical edge computing in, for instance, autonomous driving, virtual and augmented reality, and robotics.

Symposium Organizers

Dimitra Georgiadou, University of Southampton
Paschalis Gkoupidenis, Max Planck Institute
Francesca Santoro, Forschungszentrum Jülich/RWTH Aachen University
Yoeri van de Burgt, Technische Universiteit Eindhoven

Session Chairs

Paschalis Gkoupidenis
Jonathan Rivnay

In this Session