Because of the holidays, TI E2E™ design support forum responses will be delayed from Dec. 25 through Jan. 2. Thank you for your patience.

TI-JESD204-IP: mgt_tx_usrclk2 cannot feed MMCM PLL?

Part Number: TI-JESD204-IP
Other Parts Discussed in Thread: DAC38RF82

Tool/software:

Hello,

I have created the following setup to drive the clocks of DAC38RF82 from a Kintex Ultrascale FPGA.

External Clock 50MHz is fed to FPGA to create a 200MHz clock reference for the DACClk.

From the 200MHz, I use the internal DAC PLL to create a 6600MHz sampling clock and output a divided reference version from the DAC to a 550MHz reference clock (I also use an external divider to achieve 6600/12=550).

I then feed the 550MHz to the MGT Clock reference of the TI-JESD204 core.

From there, I use the mgt_tx_userclk2 (68.75MHz) to feed an MMCM PLL and generate 275MHz for my logic and a copy of 68.75 MHz as tx_sys_clock for the JESD204 core. See the diagram below.

The documentation of the JESD204 core says that mgt_tx_userclk2 does not need a BUFG because it is included inside the core.

As VIVADO implementation complaint I explicitly used a BUFG_GT to drive the PLL.

VIVADO implementation continues to fail with the following message:

[DRC REQP-123] connects_CLKINSEL_VCC_connects_CLKIN1_ACTIVE: U_SysPLL/inst/mmcme3_adv_inst: The MMCME3_ADV with CLKINSEL tied high requires the CLKIN1 pin to be active.

Questions:

1. Is the clocking scheme presented acceptable?

2. Is there a problem driving a PLL from mgt_tx_usrclk2 internally?

3. If it is a problem can I route the signal to a pin and provide to the PLL input clock the same signal from external pins? 

4. Is there any reason that mgt_tx_usrclk2 has jitter that is not acceptable from MMCM?

Thanks,

Ilias

  • It seems that the reason behind that the clock is not accepted for the PLL in the implementation is related to the GTH IP. For some reason implementation in Vivado does not find the corresponding GTH module hence the clock tree is optimized away. The IP block is visible in my tree, but I will need to check the differences with the reference design of the JESD204 IP.