16th International Forum on MPSoC for Software-defined Hardware
July 11-15, 2016, Nara Hotel, Nara, Japan
Slides available here!
Speaker:
Anca Molnos, CAE-LETI, France
Title:
Energy efficient SoCs: A reinforcement learning approach
Abstract:
This talk discusses the trends and challenges in energy consumption, temperature, and aging management of large Systems-on-Chip (SoC). We believe that, to achieve true energy efficiency, we still have several roadblocks to overcome. An important one is to deal with non-ideal cases, where: (1) application are too dynamic to construct accurate models, (2) application performance is not proportional to the operating point due to, e.g., contention at shared resources, and (3) the system software introduces further variations.
Consequently, future solutions will need to adapt their parameters and discover the dynamics of an application rather than to model them upfront. In this context, we propose and demonstrate a reinforcement-learning approach to the problem of energy minimisation on a test-chip with an host CPU and 16 worker processing elements.
Finally, we discuss new challenges in emerging technologies, e.g., FDSOI, arising from the fact that the best management strategy (e.g., supply voltage scaling vs. body-bias voltage scaling) is highly dependent on the temperature, and activity of the circuit.
Bio:
Anca Molnos received her M.Sc. degree in computer science from the "Politehnica" University of Bucharest, Romania and the Ph.D. degree in computer engineering from the Delft University of Technology, The Netherlands, in 2001 and 2009, respectively. Between 2006 and 2009 she was senior scientist at NXP Semiconductors, The Netherlands, working on low-power multi-processors and distributed real-time systems. From 2009 to 2012 she was a researcher with the Delft University of Technology, working on embedded multi-core resource management for low-power and quality of service. In 2013 she joined CEA LETI, where her research focuses on developing adaptive, energy-aware software, variability management frameworks, and adequate computing methods.
* If you wish to modify any information or update your photo, please contact the Publicity Chair at the following address.