A Sequential Approach to Realize Tiny Machine Learning in Hardware Platform
Abstract
Tiny machine learning (TinyML) represents a transformative frontier in artificial intelligence, enabling the deployment of machine learning models on resource-constrained devices. This technology embeds deep learning capabilities directly into billions of Internet of Things (IoT) devices and microcontrollers (MCUs), significantly expanding the scope of intelligent applications while enabling ubiquitous computing at the extreme edge. The implementation of TinyML presents substantial challenges due to severe hardware limitations, such as memory constraints, processing capabilities, and power requirements. These constraints were the motivations that led to the development of specialized processor architecture integrated in various hardware platforms designed specifically to support machine learning models for forecasting. This chapter comprehensively examines the current landscape of hardware platforms for TinyML implementations, analyzing their architectures, capabilities, and performance characteristics.
