Speaker:

Raphaël David, CEA LIST, France

Title:

Multi-criteria exploration framework for MPSOC

Abstract:

MPSOC becomes the mainstream in compute-intensive embedded systems to support the execution of massively parallel applications like image or computer vision applications. These applications exhibit more and more dynamic behavior with data-dependent execution times and control flows. In addition, MPSOC architectures are built on very advanced technologies in order to benefit from performance gains and power offered by the miniaturization of devices. This downscaling lead to increasing variability of transistors and bigger thermal gradients while amplifying the degradation mechanisms. To cope with these challenges, MPSoC comes with very advanced online resources management strategies to deal with energy, temperature and aging. This talk will present an exploration HW/SW framework dealing with multiple abstraction levels to cope with technological effects early in the MPSOC design flow. We will discuss to connection with third party tools to provide co-simulation properties with power, thermal and aging evaluation solutions. This framework will be illustrated with a homogeneous MPSOC architecture design example.

Bio:

Dr. Raphaël David is a senior expert at CEA LIST, in the field of embedded parallel architectures. He received his PhD degree in computer engineering for having designed the DART reconfigurable processor, from the University of Rennes I, France, in 2003. He has joined the CEA LIST in a post-doctoral position to study reconfigurable architectures benefits to reduce power consumption of embedded systems. Since 2004 he has proposed dynamic execution models for programmable and reconfigurable multi- and many-cores systems to support variable execution conditions, either coming from technology or data-dependant applications. He has managed these activities as the leader of the MPSOC design team in CEA LIST from 2006 to 2009 and has explored the architectures design space to support such advanced execution models in various national and European projects. His research interests also include the dynamically reconfigurable processors for image processing and low power design. Raphaël David took the lead of the Embedded Computing Laboratory of the CEA LIST in 2010 to design high performance embedded architectures for image processing and vision systems.