Predicting the Performance of Application-Specific NoCs Implemented on FPGAs

Jason Lee and Lesley Shannon
Simon Fraser University


Abstract

Modern FPGAs are able to implement complex systems such as Systems-on-Chips (SoCs) and Networks-on-Chips (NoCs). Appropriate NoC topology choices for ASICs have been investigated and typically topologies that can be easily mapped to a two-dimensional fabric are used to reduce chip area and ensure electrical characteristics. However, for FPGAs, each device's size and routing fabric are fixed. Since these resources exist independent of use, the choice of topology is only limited by the performance of the NoC itself. In this work, we investigate how topology characteristics impact a NoC's performance on an FPGA. From this analysis, we have created an analytical model that describes the maximum operating frequency of a NoC as a function of the topology's network parameters. This model is in the form of a simple equation that is accurate to within 4.68% across a range of topologies, chip sizes, and device families. It enables us to explore how an FPGA's prefabricated routing interconnect provides increased freedom in the selection of application-specific topologies. The model can also be used by designers to explore the topology design before implementation.