Computing Fabric for High performance Applications (COBRA-NL)

Introduction

We are living in an era where wireless connectivity has become an integral part of our lives. Wireless technology is penetrating into unanticipated segments of the economy beyond the traditional telecommunications and data-networking market (e.g. for monitor/control purposes in medical/healthcare, manufacturing, retail, as well as for security / transportation / identification / banking, etc.)

Future wireless systems need tremendous compute capacities, beyond 1 TeraFlops, but have a very restricted power budgets, resulting in energy requirements down to 1 pJ per operation. To enable this new processor architecture, application mapping, resource management and power saving techniques have to be researched and applied. This is the context of the European Catrene project COBRA-NL.

In the Dutch part of the COBRA-NL project, we will tackle this design problem using complex multi-core/manycore system-on-chips (SoCs) that enable the use of such leading-edge wireless technologies.

To stear the research COBRA considers two main application:

  • Cellular connectivity (4G and beyond i.e. LTE/LTE-Advanced)
  • Wireless connectivity for automotive (Car-infotainment, car-to-car and/or car-to-infrastructure communications).


Future wireless systems will have to run multiple radio applications simultaneously. The design of an architecture supporting this is quite a complex undertaking considering the variety of interdisciplinary challenges involved in the entire design process. Research challenges tackled within COBRA-NL are, among others,

  • Multi-core HW architecture, including vector DSPs, interconnect, memory hierarchy, interfaces, and HW support for Radio OS.
  • Radio OS: a streaming framework for running possibly multiple radios (SDR: software defined radio).
  • Efficient resource management (low energy, memory, CPU utilization etc.), possibly based on Scenario techniques.
  • Design flows and tools involved (Application analysis and parallelization using the dataflow paradigm, tools such as compilers, C-to-dataflow translation etc.).
  • Predictable and Composable design (with respect to real-time requirements), based on adequate models of computation (MoCs) and platform virtualization (giving guarantees on available resources).

 

Involved TU/e Research Groups

Within TU/e 2 research groups are joining COBRA-NL:
1) The Electronic Systems (ES) group
2) The System Architecture and Networking (SAN) group

Involved People

1) Prof. CH (Kees) van Berkel
2) Dr. Orlando Moreira
3) MSc. Hrishikesh Salunkhe
4) MSc. Alok Lele

Project Collaboration

The project is granted within the European Catrene context. COBRA-NL, which stands for "Computing Fabric for High Performance Applications", is a cooperation lead by STMicroelectronics, and having partners in France, Spain, and The Netherlands.

Within the Netherlands the partners are TUE, TUDelft, ST-Ericsson, NXP, Virage Logic, ACE and Compaan.

Links: