*banner
 

Process Network in Silicon: A High-Productivity, Scalable Platform for High-Performance Embedded Computing
Mike Butts

Citation
Mike Butts. "Process Network in Silicon: A High-Productivity, Scalable Platform for High-Performance Embedded Computing". Talk or presentation, 25, November, 2008.

Abstract
Conventional high-performance embedded system technology has reached fundamental scaling limits of single CPU/DSP performance and ASIC/FPGA hardware development productivity. Adapting the SMP multicore programming model from general-purpose computing has serious scaling and reliability issues as well, as shown by work here at Berkeley. Adapting the SIMD progamming model from HPC is not always a good match to increasingly irregular and complex algorithms in video, image and wireless processing. We defined a Structured Object Programming Model, based on the Kahn process network model of computation, specifically for embedded computing. Objects are strictly encapsulated software programs running concurrently on a MIMD array of processors and memories. They communicate and synchronize with one another through a structure of buffered channels, in which every data transfer is a synchronization event. An application is a hierarchy of objects and structure, programmed directly in high-level language, without need for automatic parallelizing compilers or synthesis tools. Having first defined this programming model, we defined a scalable hardware architecture, developed one teraOPS production silicon and an integrated development environment, and delivered it to a number of customers who are designing it into their products. University projects have been using it as well. Developers have found this programming model easy to learn and use, and two to three times as productive as their previous DSP, FPGA or ASIC platforms. A hardware and tools demo will be available after the seminar.

Electronic downloads

Citation formats  
  • HTML
    Mike Butts. <a
    href="http://chess.eecs.berkeley.edu/pubs/510.html"
    ><i>Process Network in Silicon: A
    High-Productivity, Scalable Platform for High-Performance
    Embedded Computing</i></a>, Talk or
    presentation,  25, November, 2008.
  • Plain text
    Mike Butts. "Process Network in Silicon: A
    High-Productivity, Scalable Platform for High-Performance
    Embedded Computing". Talk or presentation,  25,
    November, 2008.
  • BibTeX
    @presentation{Butts08_ProcessNetworkInSiliconHighProductivityScalablePlatform,
        author = {Mike Butts},
        title = {Process Network in Silicon: A High-Productivity,
                  Scalable Platform for High-Performance Embedded
                  Computing},
        day = {25},
        month = {November},
        year = {2008},
        abstract = {Conventional high-performance embedded system
                  technology has reached fundamental scaling limits
                  of single CPU/DSP performance and ASIC/FPGA
                  hardware development productivity. Adapting the
                  SMP multicore programming model from
                  general-purpose computing has serious scaling and
                  reliability issues as well, as shown by work here
                  at Berkeley. Adapting the SIMD progamming model
                  from HPC is not always a good match to
                  increasingly irregular and complex algorithms in
                  video, image and wireless processing. We defined a
                  Structured Object Programming Model, based on the
                  Kahn process network model of computation,
                  specifically for embedded computing. Objects are
                  strictly encapsulated software programs running
                  concurrently on a MIMD array of processors and
                  memories. They communicate and synchronize with
                  one another through a structure of buffered
                  channels, in which every data transfer is a
                  synchronization event. An application is a
                  hierarchy of objects and structure, programmed
                  directly in high-level language, without need for
                  automatic parallelizing compilers or synthesis
                  tools. Having first defined this programming
                  model, we defined a scalable hardware
                  architecture, developed one teraOPS production
                  silicon and an integrated development environment,
                  and delivered it to a number of customers who are
                  designing it into their products. University
                  projects have been using it as well. Developers
                  have found this programming model easy to learn
                  and use, and two to three times as productive as
                  their previous DSP, FPGA or ASIC platforms. A
                  hardware and tools demo will be available after
                  the seminar.},
        URL = {http://chess.eecs.berkeley.edu/pubs/510.html}
    }
    

Posted by Hiren Patel on 25 Nov 2008.
Groups: chess
For additional information, see the Publications FAQ or contact webmaster at chess eecs berkeley edu.

Notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright.

©2002-2018 Chess