Decoupled Access-Execute and Dynamic Voltage/Frequency Scaling Optimization for Energy Efficient tinyML Deployments on STM32 MCUs

Abstract

Over the last years the rapid growth Machine Learning (ML) inference applications deployed on the Edge is rapidly increasing. Recent Internet of Things (IoT) devices and microcontrollers (MCUs), be come more and more mainstream in everyday activities. In this work we focus on the family of STM32 MCUs. A DVFS capability for ARM Cortex M MCUs is implemented and integrated within a state-of the-art inference engine, along with a Decoupled Access Execute code optimization. A novel method ology is proposed for CNN deployment on the STM32 family, focusing on power optimization through effective clocking exploration and configuration and decoupled access-execute convolution kernel ex ecution. This approach is enhanced with optimization of the power consumption through Dynamic Voltage and Frequency Scaling (DVFS) under various latency constraints, composing an NP-complete optimization problem. We compare our approach against the state-of-the-art TinyEngine inference engine, as well as TinyEngine coupled with power-saving modes of the STM32 MCUs, indicating that we can achieve up to 25.2% less energy consumption for varying QoS levels.