Power-Efficient Invasive Loosely-Coupled MPSoCs (B03)

• Subject Area: Computer Architecture, Embedded and Massively Parallel Systems
• Term: from 2010 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 146371743

Machine-learning-based methods enable self-aware dark silicon management on massively parallel invasive computing platforms. The goal of the dark silicon management is to optimise for performance under power and thermal constraints. This approach will employ unsupervised-, supervised and reinforcement online learning methods in order to achieve self-awareness. It uses run-time sensory input provided by the iDoC hardware data accumulator that employs representational learning.

• DFG Programme: CRC/Transregios
• Subproject of: TRR 89: Invasive Computing
• Applicant Institution: Friedrich-Alexander-Universität Erlangen-Nürnberg
• Project Heads: Professor Dr.-Ing. Jörg Henkel; Professor Dr. Andreas Herkersdorf