<ul><li>The rise of IoT has led to an increased demand for on-edge machine learning, with TinyML being a promising solution for resource-constrained devices like MCUs.</li><li>A new benchmarking methodology has been introduced to evaluate performance by integrating energy and latency measurements across three execution phases: pre-inference, inference, and post-inference.</li><li>The methodology ensures that devices operate independently, allowing for automated testing to enhance statistical significance. Testing on an STM32N6 MCU with high-performance and low-power configurations revealed insights on energy efficiency impact.</li><li>Findings showed that reducing core voltage and clock frequency improved pre- and post-processing efficiency without majorly affecting network execution performance, enabling cross-platform comparisons to determine the most efficient inference platform.</li></ul>

Benchmarking Energy and Latency in TinyML: A Novel Method for Resource-Constrained AI

Discover more