Optimizing the software running on a ULP microcontroller in a bare-metal context for energy efficiency

: Principles and methods

Abstract

The climate change and the energy crisis become increasingly present these days and all sectors of the industry have a role to play with this issue. The IT industry has been addressing them through innovations at the hardware level while the issue is barely raising in the software sector. The embedded electronics is a trend today and, as such, has a major impact on the energy consumption of the IT sector.
In this thesis, we are searching for a mean to evaluate and to control the effects of the software on the consumed energy. To this end, we explore the source code, seen as a set of idioms in C, to evaluate how it affects the energy consumption. This consumption is further analyzed and an energy model is approached based on the internal signals of an MCU.
This model is based on the assumption that the energy consumption of the system is determined by the use it makes of the core, the volatile memory, and the non-volatile memory. A measurement bench is set up to allow, based on the execution of a given algorithm, to measure the energy consumption and to count the number of cycles related to each of the three above mentioned components. A series of experiments is then carried out using this environment, in order to feed the
model. The statistics are finally used to challenge the validity of the model.
If the approach appears to be relevant, further investigations are required to finalize an operational model. The goal would be, for a silicon vendor, to use it to start writing energy-efficient code as soon as the pre-silicon software development phase.

Date of Award	2021
Original language	English
Awarding Institution	University of Namur
Supervisor	Pierre Yves Schobbens (Supervisor) & James Jerson Ortiz Vega (Co-Supervisor)