PhD - Leuven | More than two weeks ago
Autonomous vehciles, smart cities, indoor smart farming, connected wearables, etc. all have one thing in common: the problem of building accurate models is currently "solved" by adding more and more sensors, creating more and more data and eventually a massive data deluge in the cloud (or edge). A way out of this impasse is to design truly intelligent sensors that not only compress their raw data towards the cloud, but also learn which features are essential to communicate towards a central hub. Using spiking neural networks with onlne learning capabilities this can be achieved, making the downstream model building feasible without relying on high bandwidth datalinks. At imec, hardware and algorithm development for SNN's has been going hand in hand for different types of sensors, ranging from wearable/implantable ECG patches to radar sensors. In this PhD, we will build further on the results for purely digital implementations and explore mixed signal designs. You will be guided by experts in signal processing, SNN's and mixed signal low-power design, with the end oal of creating a sensor fusion IC for deeply fusing event-based radar and camera streams.
Required background: Electrical engineering, neuromorphic engineering, digital design
Type of work: 60% modeling/simulation, 30% experimental, 10% literature
Supervisor: Piet Wambacq
Co-supervisor: Jan Craninckx
Daily advisor: Ilja Ocket, Lars Keuninckx
The reference code for this position is 2021-120. Mention this reference code on your application form.