Jonathan Larochelle

I conduct research on low-power intelligent sensor systems for environmental and medical applications under the supervision of Prof. Peter Woias and Laura Comella at the Department of Microsystems Engineering, University of Freiburg, Germany. I have experience in low-power circuits and embedded machine learning.

While passionate about the dizzying technological progress of the current times, I am critical about many of its applications, especially with regard to its unforeseen negative consequences on our society and the environment. I am still figuring out how I can integrate stronger notions of sustainability, conviviality, and circularity in my design practice. Any pointers are more than welcome.

Outside of office hours, I enjoy biking in the german landscape and sailing on the Schluchsee.

Brain-MEP - Miniaturized Electrical Pulse Generator for Brain

This project aims to develop a minimally invasive implantable device for responsive neurostimulation used in epilepsy treatment. I am responsible for the optimization of machine learning algorithms for deployment in a ultra-low-power microcontroller. I have developed a hardware-aware neural architecture search pipeline in order to optimize a convolutional neural network for development on a specific microcontroller. Additionally, I have investigated the optimal computation of signal features and estimation methods for the inference energy of a model on a microcontroller.

This work is conducted in collaboration with the medical centers of the University of Freiburg and the University of Mannheim, as well as Precisis.

Light quality sensor with a variable field-of-view

Metrics such as the Leaf area index (LAI) and the Photosynthetically active radiation (PAR) are used by biologists to characterize plant metabolism and follow its evolution over time. The most common measurement systems use a combination of photodiodes to measure the incident light. To account for the irregular placement of leaves in a plant canopy, it is generally recommended to take many measurements at different positions around the plant, which prevents unattended continuous monitoring (in a forest, for example).

To solve this issue, I have developed a sensor concept which can "scan" its incident field-of-view, enabling identifying and processing the areas with lower or higher leaves density. This novel system uses a liquid crystal display to serve as an optical shutter. I have prototyped the system, developed a calibration procedure and set-up, and validated the system during a field measurement campaign. I have written my Master's thesis on this project. Additionnally, the system concept and a calibration procedure have been published in a peer-reviewed journal.

This work was conducted in collaboration with the Kompetenzzentrum Obstbau Bodensee.