Abstract
This paper is toward a promising solution to address the environmental sustainability challenge in computing by building brain-inspired and green non-Von Neumann systems with Resistive Random-Access Memory (ReRAM) made from natural organic materials, honey, for energy-efficient operation, renewable material resources, sustainable device manufacturing, and environmentally-friendly disposal. In this paper, honey-ReRAM and its arrays are firstly manufactured and tested. The resistance modulation mechanism of honey-ReRAM is analyzed and investigated. Then a Computing-in-Memory (CIM) architecture based on honey-ReRAM for edge AI and IoT applications is proposed and evaluated. The experimental results indicate that the proposed edge AI systems with the VGG8 and DenseNet-40 models have a high inference accuracy, more than 90%. If device variation and nonlinearity are considered, the VGG8 model is still effective, and the 16-level ReRAM enabled edge AI system has the best performance, including accuracy = 92%, TOPS = 0.864048, FPS = 701.431, energy efficiency = 45.819, and Peroverall= 558. Furthermore, four conductance drift scenarios and Analog-to-Digital Converter (ADC) quantization effects are also considered to further verify the proposed edge AI systems. It concludes: (1) the system has high immunity to the device drift effect and would be very reliable even with long duration of time; (2) as for the VGG8 models, when the ADC has 5-bit, all accuracies rise to over 90%; (3) as for DenseNet-40 models, a 7-bit ADC can ensure an accuracy of 88%. Finally, the honey-ReRAM enabled sustainable edge AI system is compared to the state-of-the-art and shows great merits.
Recommended Citation
J. Wang et al., "Honey-ReRAM Enabled Sustainable Edge AI System for IoT Applications," IEEE Access, Institute of Electrical and Electronics Engineers, Jan 2025.
The definitive version is available at https://doi.org/10.1109/ACCESS.2025.3636818
Department(s)
Electrical and Computer Engineering
Publication Status
Open Access
Keywords and Phrases
computing in-memory system; edge artificial intelligence (AI); honey-ReRAM; inference accuracy; Internet of Things (IOT); natural organic ReRAM; nonlinearity; sustainable; variation
International Standard Serial Number (ISSN)
2169-3536
Document Type
Article - Journal
Document Version
Final Version
File Type
text
Language(s)
English
Rights
© 2025 The Authors, All rights reserved.
Creative Commons Licensing
This work is licensed under a Creative Commons Attribution 4.0 License.
Publication Date
01 Jan 2025
Comments
National Science Foundation, Grant 2420994