Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

Automated Mapping for Heterogeneous Multiprocessor Embedded Systems

Abhijit Davare

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2007-115
September 7, 2007

http://www.eecs.berkeley.edu/Pubs/TechRpts/2007/EECS-2007-115.pdf

Increasing design complexity and time-to-market concerns have led to the increased prevalence of programmable processing elements for embedded systems. These platforms feature multiple processing elements, some of which may be customized for specific domains. Deploying applications typical of embedded system domains such as multimedia and transportation onto these platforms is difficult, not only due to the heterogeneous parallelism in the platforms, but also due to the performance constraints that typify embedded systems. This dissertation advocates and validates a design flow that enables designers to deploy applications onto this emerging class of embedded platforms. The design flow is based on the platform-based design methodology which initially separates the modeling of the application and architectural platform. The design flow advocates transforming these models in a structured manner such that both have compatible execution models and abstraction levels. The goal of this structured model transformation is to simplify the deployment challenge into a covering problem where portions of the application are assigned to processing elements in the architectural platform. The focus of this work is to validate the design flow by applying it to embedded systems from the multimedia and automotive domains. The case studies explore the tradeoffs inherent in modeling these systems as well as techniques for automatically solving the mapping problems. The automated techniques use mathematical programming approaches which have the flexibility to handle changes in the problem assumptions. We observe that regardless of the domain, some aspects of the design flow such as modeling and simulation are shared between systems. Based on the insights gained in applying this design flow to the case studies, the requirements and initial implementation for Metro II - a next-generation design framework for platform-based design - are described.

Advisor: Alberto L. Sangiovanni-Vincentelli


BibTeX citation:

@phdthesis{Davare:EECS-2007-115,
    Author = {Davare, Abhijit},
    Title = {Automated Mapping for Heterogeneous Multiprocessor Embedded Systems},
    School = {EECS Department, University of California, Berkeley},
    Year = {2007},
    Month = {Sep},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/2007/EECS-2007-115.html},
    Number = {UCB/EECS-2007-115},
    Abstract = {Increasing design complexity and time-to-market concerns have led to the increased prevalence of programmable processing elements for embedded systems. These platforms feature multiple processing elements, some of which may be customized for specific domains. Deploying applications typical of embedded system domains such as multimedia and transportation onto these platforms is difficult, not only due to the heterogeneous parallelism in the platforms, but also due to the performance constraints that typify embedded systems.

This dissertation advocates and validates a design flow that enables designers to deploy applications onto this emerging class of embedded platforms. The design flow is based on the platform-based design methodology which initially separates the modeling of the application and architectural platform. The design flow advocates transforming these models in a structured manner such that both have compatible execution models and abstraction levels. The goal of this structured model transformation is to simplify the deployment challenge into a covering problem where portions of the application are assigned to processing elements in the architectural platform.

The focus of this work is to validate the design flow by applying it to embedded systems from the multimedia and automotive domains. The case studies explore the tradeoffs inherent in modeling these systems as well as techniques for automatically solving the mapping problems. The automated techniques use mathematical programming approaches which have the flexibility to handle changes in the problem assumptions.

We observe that regardless of the domain, some aspects of the design flow such as modeling and simulation are shared between systems. Based on the insights gained in applying this design flow to the case studies, the requirements and initial implementation for Metro II - a next-generation design framework for platform-based design - are described.}
}

EndNote citation:

%0 Thesis
%A Davare, Abhijit
%T Automated Mapping for Heterogeneous Multiprocessor Embedded Systems
%I EECS Department, University of California, Berkeley
%D 2007
%8 September 7
%@ UCB/EECS-2007-115
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/2007/EECS-2007-115.html
%F Davare:EECS-2007-115