R-tile Avalon® Streaming Intel® FPGA IP for PCI Express* User Guide

ID 683501
Date 10/06/2021
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

2.2.1. Clocking

In the PCIe Hard IP mode, the R-tile Avalon Streaming IP for PCI Express* has four primary clock domains:
  • PHY clock domain (i.e. core_clk domain): this clock is synchronous to the SerDes parallel clock.
  • EMIB/FPGA fabric interface clock domain (i.e. pld_clk domain): this clock is derived from the same reference clock (refclk0) as the one used by the SerDes. However, this clock is generated from a stand-alone core PLL.
  • Application clock domain (coreclkout_hip) for in-band signals: this clock is an output from the R-tile IP, and it has the same frequency as pld_clk.
  • Application clock domain (slow_clk) for sideband signals: this clock is another output from the R-tile IP. It is a divide-by-2/4 version of coreclkout_hip.
Figure 2. Clock Domains in PCIe Modes
Table 9.  PHY Clock and Application Clock Frequencies
Mode PHY Clock Frequency Application Clock Frequency
PCIe Gen1 1000 MHz Gen1 is supported only via link down-training and not natively. Hence, the application clock frequency depends on the configuration you choose in the IP Parameter Editor. For example, if you choose a Gen3 configuration, the application clock frequency is 250 MHz - 300 MHz.
PCIe Gen2 1000 MHz Gen2 is supported only via link down-training and not natively. Hence, the application clock frequency depends on the configuration you choose in the IP Parameter Editor. For example, if you choose a Gen3 configuration, the application clock frequency is 250 MHz - 300 MHz.
PCIe Gen3 1000 MHz 250 MHz - 300 MHz
PCIe Gen4 1000 MHz 250 MHz - 300 MHz
PCIe Gen5 1000 MHz 400 MHz - 500 MHz
Note: For a link down-training scenario when R-tile is configured at Gen3, Gen4 or Gen5 and the link gets down-trained to a lower speed, the application clock frequency will continue to run at the configured frequency set in the PLD Clock Frequency parameter. For example, when the PCIe Hard IP Mode parameter is set as Gen5 1x16 and the PLD Clock Frequency parameter as 500 MHz, the PLD clock frequency will continue to run at 500 MHz even if the link is down-trained to Gen4 or less.

R-tile has two reference clock inputs at the package level, refclk0 and refclk1. You must connect a 100 MHz reference clock source to these two inputs. Depending on the port mode, you can drive the two refclk inputs using either a single clock source or two independent clock sources.

In 1x16 and 4x4 modes, drive the refclk inputs with a single clock source (through a fanout buffer) as shown in the figure below.

Figure 3. Using a Single 100 MHz Clock Source in 1x16 and 4x4 Modes
In 2x8 mode, you can drive the refclk inputs with either a single 100 MHz clock source as shown above, or two independent 100 MHz sources (see Figure 4) depending on your system architecture. For example, if your system has each x8 port connected to a separate CPU/Root Complex, it may be required to drive these refclk inputs using independent clock sources. In that case, the refclk0 input for Port 0 must always be running because it feeds the reference clock for the R-tile core PLL that controls the data transfers between the R-tile and FPGA fabric via the EMIB. If this clock goes down, Port 0 link will go down and Port 1 will not be able to communicate with the FPGA fabric. Following are the guidelines for implementing two independent refclks in 2x8 mode:
  • If the link can handle two separate reference clocks, drive the refclk0 of R-tile with the on-board free-running oscillator.
  • If the link needs to use a common reference clock, then PERST# needs to indicate the stability of this reference clock. If this reference clock goes down, the entire R-tile must be reset.
Figure 4. Using Independent 100 MHz Clock Sources in 2x8 Mode