Authors: Florentia Afentaki, Gurol Saglam, Argyris Kokkinis, Kostas Siozios, Georgios Zervakis, Mehdi B Tahoori

Published on: December 29, 2023

Impact Score: 8.3

Arxiv code: Arxiv:2312.17612

Summary

• What is new: First-time integration of Approximate Computing and Bespoke design in designing ultra-low power Multilayer Perceptron (MLP) classifiers for Printed Electronics.
• Why this is important: Printed Electronics, despite their advantages, struggle with complex circuit designs like machine learning classifiers due to large feature sizes.
• What the research proposes: An automated framework that employs Approximate Computing and Bespoke design to optimize all functions of MLP’s neurons for ultra-low power consumption.
• Results: Enabled battery-powered operation for complex MLP architectures, significantly outperforming current technologies in power efficiency.

Technical Details

Technological frameworks used: Approximate Computing and Bespoke design integrated automated framework

Models used: Multilayer Perceptron (MLP) classifiers

Data used: Various MLP sizes

Potential Impact

Ultra-low cost and flexible electronics markets, wearable tech, biomedical devices, and companies in the field of ubiquitous computing.

Want to implement this idea in a business?

We have generated a startup concept here: FlexiNeura.