Power Consumption when using AIModels on microcontrollers

This report is about the evaluation of different microcontrollers and their current con-sumption, specifically microcontrollers that will run AI models. The company Sensorbeeneeds a new microcontroller for their future projects. One of the areas of use will be AImodels. The most important parameter for Sensorbee is current consumption, minimizingcurrent consumption is a top priority as their products are powered by batteries or solarpanels. At the beginning of the project, the focus was on which microcontroller had thelowest current consumption. For AI models in particular, however, it turned out to be avery big difference in the optimization of the software. A well-optimized AI model canquickly classify data, which in turn leads to the microcontroller being able to spend muchof its time in sleep mode. This meant that not only the current consumption of the micro-controller could be taken into account, but how fast it ran through an AI model was at leastas important, if not more important. Even how easy it was for Sensorbee to get startedwith the new processor had to be evaluated, as Sensorbee is a small company that has littleresources to move around with. The candidate that was most promising in the beginningproved to be useless as the performance of the AI model was too poor and the general sup-port for the microcontroller was unsatisfactory. It turns out that there are a lot of variablesto consider when choosing a microcontroller and not just what is in the datasheet