Late Breaking Results: Energy-Efficient Printed Machine Learning Classifiers with Sequential SVMs

TL;DR

This work targets the energy and area constraints of printed electronics for ML classifiers by proposing sequential printed bespoke SVMs that process one support vector per cycle. The architecture integrates control, storage, compute, and a voter to execute SVM inference with a linear kernel and low-precision weights, followed by post-training quantization. It achieves about $6.5\times$ energy savings while maintaining or improving accuracy relative to state-of-the-art fully parallel SVMs. The results suggest that such energy-efficient, multi-cycle SVMs can extend battery life in flexible, printed devices and enable practical battery-powered ML systems.

Abstract

Printed Electronics (PE) provide a mechanically flexible and cost-effective solution for machine learning (ML) circuits, compared to silicon-based technologies. However, due to large feature sizes, printed classifiers are limited by high power, area, and energy overheads, which restricts the realization of battery-powered systems. In this work, we design sequential printed bespoke Support Vector Machine (SVM) circuits that adhere to the power constraints of existing printed batteries while minimizing energy consumption, thereby boosting battery life. Our results show 6.5x energy savings while maintaining higher accuracy compared to the state of the art.

Figures (1)

• Figure 1: Abstract overview of our proposed sequential circuit design.