Speaker:

Jan Madsen, DTU Informatics, Denmark

Title:

Mapping known applications onto unknown multicore platforms through probabilistic estimation

Abstract:

In this talk, we are interested in the mapping of embedded applications onto multi-ASIP platforms in the early stage of the design process. An Application Specific Instruction Set processor (ASIP) is a processor, which is tuned to a specific application or set of applications. When the program to be executed on the ASIP is known, the ASIP can be configured by selecting the appropriate micro-architecture components (such as number and type of functional units, exploring instruction level parallelism and memory configurations) and by extending the instruction set and hence, the compiler, accordingly. An optimized mapping can be obtained only when the number, interconnection and configuration of ASIPs have been fully determined. However, in order to find an optimized configuration of the platform and the individual ASIPs, the mapping has to be known. To address this “thicken and egg” problem, we propose a probabilistic estimation technique to capture the uncertainties of the detailed ASIP configurations and to identify the platform configuration and mapping which has the highest probability of meeting the requirements after a detailed ASIP optimization has been applied at a later stage.

Bio:

Jan Madsen is Professor in computer-based systems at DTU Informatics at the Technical University of Denmark. He is Deputy Head of DTU Informatics and Head of the Section on Embedded Systems Engineering. He is the leader of the Hardware Platforms and Multiprocessor System-on-Chip Cluster within the EU/IST Network of Excellence ArtistDesign and member of the Strategic Management Board of ArtistDesign. He is senior member of IEEE and is currently serving as Vice Chair of IEEE Denmark Section. His research interests are related to design of embedded computer systems. In particular system-level modeling and analysis of multiprocessor systems, including RTOS modeling and hardware/software codesign. He is generally interested in design methodologies (including CAD tools) and implementations of embedded systems, covering areas of adaptable systems, wireless sensor networks and biochips. He has published more than 110 publications in international journals and conferences as well as co-authored 11 book chapters. Jan Madsen is the lead delegate for Denmark in the Governing Board of the ARTEMIS Joint Undertaking, a pan-European research initiative for public-private partnership in Embedded Systems. He is on the steering committee of InfinIT, a national innovation network on ICT, where he is coordinating the strategic focus area on Embedded Systems. He is principle investigator in SYSMODEL and ASAM (both funded by ARTEMIS JU). He is participating in ProCell (NABIIT), programmable biochips, and in Wireless Sensor Network for Climate and Environmental Monitoring together with Delta. He is participating and member of the management board for IDEA4CPS, a new Chinese-Danish Basic Research Center for theoretical foundation for Cyber-Physical Systems (Danish Basic Research Foundation) He Co-General Chair of NOCS 2012, and he has been Program Chair of CODES+ISSS’11, DATE’07 and CODES’00, and General Chair of CODES’01. He is member of the Steering Committee of CODES+ISSS (ESWEEK). He is or has served on numerous program committees, including SIES, ARC, NOCS, LCTES, DAC, CODES+ISSS, ISSS, CODES, RTSS, DATE, SAC, and PARC.



* If you wish to modify any information or update your photo, please contact the web chairalain.fourmigue@polymtl.ca