1 / 14

Architecture exploration and optimisation of a flexible signal processing unit

Architecture exploration and optimisation of a flexible signal processing unit. Jean BERTRAND, Jérémie POULY – CNES Axel BONESS, François BERTRAND – CEA LETI. Goals. Evaluate SoCKET tools capacity to support architecture exploration Constraints Need to achieve meaningful technical work

talasi
Download Presentation

Architecture exploration and optimisation of a flexible signal processing unit

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Architecture exploration and optimisation of a flexible signal processing unit Jean BERTRAND, Jérémie POULY – CNESAxel BONESS, François BERTRAND – CEA LETI

  2. Goals • Evaluate SoCKET tools capacity to support architecture exploration • Constraints • Need to achieve meaningful technical work • Specific tools required • Small team involved • Choice of « easy » exploration • For verification of results • To ease success Workshop - November 2011

  3. Design under improvement VECTOR MAGNETOMETER NULL FIELD MAGNETOMETER YOU NAME IT Workshop - November 2011

  4. Design overview • Design with large datapath • 48 bit endpoint, 104 at most • Multiplier/accumulator/shifter structure • Micro-programmable • Task scheduling • BUT: • Computational performance depend on I/O throughput • HENCE : • Good arbiter (also) needed Workshop - November 2011

  5. Constraints • As our design prones reconfigurability : • any application is candidate • Model • Input • Arbiter (to score) • Application • Ouput • Metrics Workshop - November 2011

  6. SystemC-TLM-TAC needs • To be correctly installed and configured • Some parts are not painless • Users have to • Master C++ • Understand SystemC/TLM/TAC hierarchy and components • Users should be seasoned in these fields • Users must be confident with the design Workshop - November 2011

  7. TAC & architecture exploration • Design of model is the important stage • Data collection is not built-in • We had to implement our specific outputs • Model is communication based therefore good candidate for TLM-TAC modeling • Model evolution easy (arbiter switch) • Simulation campaign easy with unix scripting Workshop - November 2011

  8. What was scored ? • Tough I/O application identified • In two different kind of applications • Three different arbiters • Round Robin • Time slot FIFO • Allows different parameters • Priorized FIFO Workshop - November 2011

  9. Results RECS1 time slot arbiter Workshop - November 2011

  10. Results RECS2-4 time slot arbiter Workshop - November 2011

  11. Scoring of arbiters Workshop - November 2011

  12. Results : analysis • Technical result : priorized FIFO arbiter is the best • Modelisation results easy to obtain • Ease of characterization of several arbiters Workshop - November 2011

  13. Synthesis • Architecture exploration succeeded • Results were meaningful and relevant • When TAC API is mastered, development is an easy process • Excellent ratio time taken/risks leveraged • given initial training and seasoning • Modelisation environment mastering as lengthy ROI Workshop - November 2011

  14. Questions Workshop - November 2011

More Related