SLR Crossing for Wide Buses - 2023.2 English

UltraFast Design Methodology Guide for FPGAs and SoCs (UG949)

Document ID
Release Date
2023.2 English

When data flow requirements require that wide buses cross SLRs, use pipelining strategies to improve timing closure and alleviate routing congestion of long resources. For wide buses operating above 250 MHz, AMD recommends using at least three pipeline stages to cross an SLR: one at the top, one at the bottom, and one in the middle of the SLR. Additional pipeline stages might be required for very high clock frequency buses or when traversing horizontal as well as vertical distances.

The following figure illustrates a worst case crossing for a vu190-2 device. This example starts at an Interlaken dedicated block in the bottom left of SLR0 to a packet monitor block assigned to the top right of SLR2. Without pipeline registers for the data bus to and from the packet monitor, the design misses the 300 MHz timing requirement by a wide margin.

Figure 1. Data Path Crossing SLR without Pipeline Flip-Flop

However, adding seven pipeline stages to aid in the traversal from SLR0 to SLR2 allows the design to meet timing. It also reduces the use of vertical and horizontal long routing resources, as shown in the following figure.

Figure 2. Data Path Crossing SLR with Pipeline Flip-Flop Added

Tip: Use the AXI Register Slice IP or your custom auto-pipelining IP to close timing on wide buses across SLRs.