A Comparative Review of Microcontroller Architectures in Embedded Machine Learning

This paper presents a systematic review of microcontroller architectures with a focus on their application in embedded machine learning, comparing the Raspberry Pi Pico (RP2040) and the Arduino UNO (ATmega328). Embedded Machine Learning (EML) is becoming increasingly crucial in low-power, cost-effective computing platforms, necessary for advancing smart, battery-powered devices with built-in learning capabilities. We explore the architectures, performance metrics, and ecosystem support for both microcontrollers to understand their suitability for various EML applications. The RP2040 microcontroller of the Raspberry Pi Pico is evaluated for its high performance and flexibility, supporting complex computational tasks with lower power consumption. In contrast, the Arduino UNO is analyzed for its accessibility, ease of use, and robust community support, making it ideal for simpler and cost-sensitive projects. Comparative analyses indicate that the choice between these platforms should be guided by specific project needs concerning computational intensity, operational environment, and developer expertise. This study aims to aid developers in selecting the appropriate microcontroller for EML projects by providing a clear comparison based on system architecture, energy efficiency, and overall performance.