[e-mail]

1
Sami.yehia_at_thalesgroup.com
Thales Research TechnologyEmbedded System Lab
- France

• Interconnect Cluster Meeting

2
ESL Mission

• High performance Embedded systems and tools
• for Aerospace, Defence and Security application
• Signal processing application
• Radar, Sonar, Electronic Warfare applications
• Telecommunications
• Data flow dominant with real time constraint
  and high reactivity requirements
• Image processing application
• Video surveillance
• Motion detection and pattern (facial/intruder)
  recognition
• Fine grain massively parallel SIMD processing

3
Interests in NOCs

• Research interests
• Multicore targeting signal and image processing
  applications
• Often hard real time constraints
• Reliability, fault tolerant for mission critical
  systems
• Specialization patterns and Programmable
  accelerators (SIMD,..)
• Current Multicore-related ongoing projects
  Ter_at_ops
• Multicore targeting embedded multimedia and
  signal processing applications
• Parallelism specialized units
• Common interface and programming tools targeting
  different types of specialized units.
• Started 2007, several industrial (TRT, Thomson,
  Arteris, AEDS, Esterel, Dassault aviation,) and
  academic partners ( INRIA, IEF, PRISM, ETIS,) .
  9m funding.
• NoC simulators and models

4
Research interests
Diversity of THALES applications distributed in
Business Units
Application Space
Tools
Architectural Space
Ter_at_ops
MORPHEUS
Ter_at_core
Cell
Diversity of architectural solutions

• Specialization how to parallelize ?

• Parallelism
• Use of existing programming models
• Homogeneous processing and load/task distribution

• Specialization
• Efficiency
• Performance

• Multipurpose multi-faceted solutions for cost
  effective specialization
• HW reconfiguration, polymorphic programming
  model, etc
• Benchmarking and workload characterization
• (bench suite for signal and image processing?)
• Multicore system simulation acceleration using
  FPGA

5
Proposed contribution

• Applications
• Image and video processing processing
• Currently working on a common benchmark suite for
  our application domain
• Realistic workloads
• Still to figure out how to give up the codes
  (easier through FP7 calls)