/Novel chiplet-based architectures for event-based processing nodes at the edge

Novel chiplet-based architectures for event-based processing nodes at the edge

PhD - Leuven | More than two weeks ago

Design the next-generation of event-based neural network architectures for sensor fusion applications at the edge using chiplet technology exploring system partitioning and interconnect topologies

Chiplets technology is a promising concept to reduce design cost and time of advanced designs while maintaining scalability and flexibility. Moreover, the integration of sensor and compute chiplets within on die can alleviate the throughput bottleneck by using the die-to-die interconnect. Moreover, chiplet technology allows the combination of different technology, each optimized for the intended functionality, and can improve the yield compared to a traditional design implementing all the functionality in one SoC.

 

This project targets to develop chiplet-based designs for an event-based sensor node at the extreme edge. This includes a system-level exploration of different partitioning and interconnect schemes of such designs and the identification and benchmarking of suitable chiplet modules requiring co-optimization of the hardware and the targeted algorithms.


Required background: Electrical engineering with digital design experience

Type of work: 60% hardware implementation, 25% algorithm implementation, 15% benchmarking

Supervisor: Marian Verhelst

Daily advisor: Matthias Hartmann

The reference code for this position is 2023-099. Mention this reference code on your application form.

Who we are
Accept marketing-cookies to view this content.
Cookie settings
imec's cleanroom
Accept marketing-cookies to view this content.
Cookie settings

Send this job to your email