Worst-Case Latency Analysis for the Versal Network-on-Chip

The recent line of Versal FPGA devices from Xilinx Inc. includes a hard Network-On-Chip (NoC) embedded in the programmable logic, designed to be a high-performance system-level interconnect. While the target markets for Versal devices include applications with real-time constraints, such as automotive driver assist, the associated development tools only provide figures for "structural latencies" of data packets, which assume that the network is otherwise idle. In a realistic setting, this information is not enough to ensure deadlines are met, as different packets can contend for NoC switch outputs, which causes packet contents to be buffered while in transit, increasing their latency. In this work, we develop an approach for calculating upper bounds for such worst-case latencies (WCLs), assuming a model where system tasks release packets into the NoC periodically. In order to develop an accurate model for latencies in the network, we review the architecture and operation of the Versal NoC. We focus on a formal description of the NPS switches that compose the NoC from a flit arbitration perspective, based on study the available cycle-accurate switch simulation code. Working with the presented model, we propose an adaptation to an existing approach for WCL analysis in NoC, Recursive Calculus (RC), in order to apply it to the arbitration policy implemented in the Versal NoC. To evaluate the proposed approach, we implement a simulation experiment for the Versal NoC, with custom endpoints that allow for injecting packets programatically and measuring their latencies over the NoC. We simulate both a single NPS module and a complete NoC routing periodic workloads, in order to compare with the values given by the WCL approach and identify sources of pessimism