Polymorphous Computing Architectures

main page downloads

The increasingly complex demands of embedded applications have driven significant innovation in many aspects of system design. Specifically, a research area in the field of computer architecture called polymorphous computing seeks to integrate powerful, configurable computing nodes into a single chip, facilitating tightly-coupled heterogeneous embedded computing applications. In contrast to other efforts in configurable and reconfigurable computing, polymorphous computing advocates a coarse-grained configurability (i.e. on the functional unit level, as opposed to the gate level), coupled with fast on-chip communication networks. This approach offers the benefits of a heterogeneous, multi-node computing platform, without the penalty of inter-chip communication delays.

The challenge with polymorphous computing lies in architecting a correct embedded application implementation. Correctness in this context not only implies correct functionality, but the timeliness of execution, and whether the implementation meets the design requirements. The difficulty of developing correct applications targeting static, discrete, multiprocessor platforms has been established and well-documented in the literature, along with a myriad of approaches for mitigating the associated complexity. Polymorphous computing further complicates the task of developing correct applications by integrating heterogeneity with architecture reconfiguration.

Model-Integrated Computing is seen as an enabling technology across several domains. To aid in the mitigation of the complexity of polymorphous system design, a model-integrated design approach and prototype toolset has been developed. A UML-based metamodel specifies a modeling language which facilitates the capture of embedded applications, polymorphous architectures, and performance metadata. The modeling language has been developed to offer a high-level, intuitive visual syntax, while retaining a rich semantics to facilitate automatic translations and analyses.

main page downloads


        For more information contact Ted Bapty