The design of Networks-on-Chip (NoCs) components implies a wide range of techniques and methods to address the microarchitecture of the packet-forwarding components, where routers and switches are the most complex… Click to show full abstract
The design of Networks-on-Chip (NoCs) components implies a wide range of techniques and methods to address the microarchitecture of the packet-forwarding components, where routers and switches are the most complex because they constitute the NoC’s backbone. Due to this complex design space, several works use approaches limiting architectural exploration, focusing only on achieving high-performance levels; therefore, they are inadequate for designing NoC components when particular functionalities are demanded, as in real applications with specific protocols and interfaces. This paper presents a design methodology based on a top-down approach with NoC-oriented abstraction levels to systematically generate a microarchitecture and its hardware description according to system requirements. The design flow transforms a high-level functional model into a microarchitecture model through a refinement process at each abstraction level. This structured approach involves integrating details on how the data is functionally managed within the component according to the system requirements and the processing granularity of each level, allowing testing alternatives in the early stages of the design when necessary. The models of each abstraction level can be described and simulated using the simulator OMNet++. Thus, the obtained microarchitecture model will be directly translated into a Hardware Description Language (HDL). The methodology is tested via the design of a NoC switch for a Software Defined Radio (SDR) system. Performance analysis and implementation results in a field-programmable gate array (FPGA) show that the proposed design is functional and comparable in both area and frequency to other similar state-of-the-art components, and it is also configurable to build star topologies of up to 16 nodes.
               
Click one of the above tabs to view related content.