Related Experiment Video
Updated: Aug 15, 2026
14:37
Ambient Method for the Production of an Ionically Gated Carbon Nanotube Common Cathode in Tandem Organic Solar Cells
Published on: November 5, 2014
Analog Tensor Processing With Carbon Nanotube In-Memory Matrix Multiplications for Edge Computer Vision Acceleration
Jingfang Pei1, Lekai Song1,2, Songwei Liu1
1Department of Electronic Engineering, The Chinese University of Hong Kong, Hong Kong SAR, China.
Advanced Materials (Deerfield Beach, Fla.)
|August 14, 2026
Summary
This study introduces an analog tensor core using carbon nanotube memory for efficient edge computer vision. This innovation accelerates tasks like 3D spatial transformation and edge detection, crucial for AI applications.
Area of Science:
- Materials Science
- Computer Engineering
- Artificial Intelligence
Background:
- Computer vision relies heavily on tensor operations, particularly matrix multiplications, demanding significant computational resources.
- Existing hardware accelerators like GPUs and ASICs face limitations in edge applications due to architectural complexity and analog system incompatibility.
Purpose of the Study:
- To prototype an analog tensor core for edge computer vision acceleration.
- To leverage carbon nanotube charge-trapping nonvolatile memory for in-memory computing.
Main Methods:
- Developed an analog tensor core utilizing carbon nanotube nonvolatile memory.
- Integrated memory exhibiting linear, symmetric analog weights with fast programming and data processing capabilities.
- Enabled in-memory matrix multiplications for compact tensor core design.
Main Results:
- The analog tensor core demonstrated 3D spatial transformation with <2.81% error.
- Achieved edge detection with a signal-to-noise ratio >22 dB.
- Simulations showed viewfield distortion correction and edge detection for fisheye street view images.
Conclusions:
- The developed analog tensor core shows significant potential for accelerating computer vision tasks at the edge.
- This technology can benefit applications such as autonomous driving, VR/AR, robot navigation, and industrial automation.
- Parallel single-layer tensor processing using this core enables large-scale visual data analysis.
