| Clock | Source | Runs |
|---|---|---|
| data_clk_i | Lane clock from the bundled PHY: line rate / 20 (237.6 MHz at 4.752 Gb/s, 250 MHz at 5.0 Gb/s) | The link layer, the NATIVE output and the write side of the AXI4S FIFO |
| axi_aclk | Any free-running clock, typically the processor's fabric clock | The register file only |
| m_axis_aclk | The clock of the block that takes the stream, any frequency (AXI4S build) | The read side of the AXI4S FIFO and M_AXIS |
| m_axis_ebd_aclk | The clock of the block that takes the embedded data, any frequency; may be the same clock as m_axis_aclk (embedded data output built) | The read side of the EBD FIFO and M_AXIS_EBD |
- Hold
rstn_ilow until the PHY reports the lanes ready. The example designs show the connection. - Clock
axi_aclkfrom a free-running clock rather than the lane clock. The registers then stay readable while the link is down, and software can read ID and VERSION at boot before the sensor starts. - Every crossing between these clocks is inside the core and constrains itself. Do not add
set_clock_groupsor blanket false paths between these clocks: they would override the core's own constraints. - After reset the output stage needs a short settling time (about 40 lane clocks for NATIVE, 100 clocks of the slower clock for AXI4S). A frame that starts during that time is skipped, so a streaming sensor simply loses its first frame.