Project Number: 628.002.005
Project Manager: Dr.ir André Kokkeler
Faculty of Electrical Engineering, Mathematics and Computer Science
Power efficiency is a primary design goal in future High Performance Computing systems. We propose to exploit probabilistic- and approximate computing to improve power efficiency. Probabilistic computing increases energy efficiency by lowering supply voltages. If supply is reduced to extremely low voltages, this comes with probabilistic behaviour of digital circuits. Approximate computing uses deterministic digital blocks to produce results that have just-enough accuracy. If it can be guaranteed that probabilistic- and approximate computing blocks have a negligible effect on astronomical measurements, parts of Radio Astronomy applications can be mapped onto these blocks to dramatically increase power efficiency.
This project has a connection with the larger DOME project.
Project duration: 2014 - 2017
Project budget: 216.6 k-€ funding
Number of person/months: 1.6 fte/year
Involved groups: Computer Architecture for Embedded Systems
CTIT Research Centre: Centre for Green ICT