Dr. Marc Duranton is a senior member of the Embedded Computing Lab, part of the CEA LIST (http://www-list.cea.fr/gb/presentation/list_overview.htm). He previously spent more than 23 years in Philips, Philips Semiconductors and NXP Semiconductors. His research interests include parallel and highly efficient architectures, domain specific architecture, system modeling and validation, models of computation for real-time systems, emerging paradigms for computing systems. He is in charge of the roadmap activity in the Network of Excellence HiPEAC2 (http://www.hipeac.net/).
ICT and embedded electronic systems have had a tremendous impact on everyday life over the past decades in all domains. Their performance was fuelled by the “Moore’s law” that has driven the semiconductor industry, pushing towards fast processors, huge memory sizes and increasing communication bandwidth. But a major paradigm shift is taking place now. “Moore’s law”, while keeping its pace on transistor density, will only allow a minor increase of frequency and decrease of power dissipation per transistor. Even if it will still be feasible to pack more devices on a chip, the power dissipation of each device will not be reduced accordingly, and, as we are already at a limit of power dissipation or consumption, it will not be possible anymore to use all devices on a chip simultaneously. New technology nodes also add more leakage power, more dispersion and less reliability.
Higher energy efficiency and delivering reliable behavior from unreliable and highly disperse components lead to investigate new research directions at all level. This talk will focus mainly on solutions at the architecture level and their implications: highly parallel and specialized hardware, optical interconnect and 3-D stacking, permanent storage memories - allowing to change the memory hierarchy -, new computing paradigms, like bio-inspired systems, new ways of coding or processing information - changing the displacement of data and instructions -, reversible computing – reducing energy dissipation -.
We are entering a challenging period to explore a large spectrum of techniques that seem promising in achieving particular tasks at high efficiency level while decreasing the impact of the constraints of the new technology nodes.