Design and modeling of a sequential architecture of support vector machines in printed electronics
Loading...
Date
2025-07-14
Authors
Μπέσιας, Σπυρίδων
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Printed Electronics (PE) is a technology that has emerged as a viable alternative to traditional silicon-based computing. It offers ultra-low-cost on demand fabrication, flexibility and conformality. On the other hand, due to its additive manufacturing process, machine learning (ML) applications are quite prohibitive. In order to mitigate this designers use bespoke designs in order to simplify the circuits complexity and thus reduce its total area and power consumption by a significant margin. In this thesis I address this challenge by creating a Machine Learning model that retains high accuracy while being area and power efficient. To accomplish this, I employ Support Vector Machines (SVM), a powerful machine learning strategy known for its efficiency to classify and predict outcomes, making them the perfect candidate for the constraints presented by printed electronics. Furthermore, I utilize a sequential design approach that incorporates a bespoke control and storage unit along with a Multiply-Accumulate compute engine, enabling the implementation of previously prohibitive datasets on PE.
Description
Keywords
Support vector machines, Printed electronics, Sequential architecture, Digital design