Abstract

This research explores the convergence of low-power microcontroller systems and binary event classification within the framework of carbon emission reduction efforts. The study proposes a novel methodology that utilizes microcontrollers for real-time event detection, implemented in a uniform hardware/firmware configuration under constrained computational resources. The approach demonstrates the capability of microcontrollers to efficiently process sensor inputs while significantly minimizing energy usage, all without reliance on large-scale training datasets. Two illustrative case studies one on CO₂ emissions from landfills and another on household energy consumption validate the practicality and efficacy of the method. The results underscore notable reductions in energy consumption, with data transmission during inactive periods reduced by 94.8% to 99.8%. Furthermore, the approach offers a sustainable alternative to conventional AI/ML platforms, which are substantially more resource-intensive consuming and emitting approximately 20,000 times more power and carbon, respectively.