Use DCM and MMCM for Xilinx FPGA Clock Deskew

Make it to the Right and Larger Audience

Blog

Use DCM and MMCM for Xilinx FPGA Clock Deskew

Clock generation and distribution is a big concern in complex FPGA design. Improper design without consideration of clock phase alignment can cause design unable to close timing, or barely meet timing with unnecessary consumption of power and gate count, or extra design to accommodate async clocks. Here I am using Xilinx FPGA as an example to talk about my understanding of how to use DCM to achieve clock de-skew. DCM has been replaced by MMCM in latest Xilinx FPGA. The idea is still the same. So allow me to use DCM at first to my convenience.

 

fpga_clk_1

 

Fig1. Is a typical usage of DCM with internal feedback. Why use DCM and what is the issue here? Well, first internal logics need clock (clk C) to be balanced at all FF clk inputs. Normally we want to use global clock distribution network to achieve it. (For simplicity let me ignore BUFR and BUFH) So BUFG is instantiated to use global clk network. If the fpga clock domain doesn’t need to be synchronized with fpga input clock (clk A), you still don’t need DCM. Clk A just drives BUFG and fpga internal logic still works. But if fpga logic needs to exchange data with external logic which is in clk A domain, DCM is needed. This is because BUFG delay is normally large so clk A and clk C have large skew. With Fig1 scheme, DCM adjusts internal delay to make sure two clock inputs to DCM, clk B and clk C, are balanced (toggling at the same time). Assuming IBUFG delay is small, clk A and clk C are also balanced.

 

fpga_clk_2

 

In Fig2, the top DCM balances clk A and clk B using internal feedback and the bottom DCM balances clk A and clk C using external feedback. Therefore the component generating and using clk A, this fpga itself (using clk B for its internal logics), and the next component using clk C are all in the same clk domain (balanced). Note the board routing delays from point C to point D of the next component and from point C to point E of this fpga input need to be the same and SMALL. Small is important otherwise clk A and clk C are not balanced since DCM makes sure clk A and clk E are balanced while there is big delay between clk C and clk E.

The following is site premium content.
Use points to gain access. You can either purchase points or contribute content and use contribution points to gain access.
Highlights: 553 words, 3 images
 
Staff Engineer
Author brief is empty
Groups:

3 Comments
  1. hollis 3 years ago
    0
    -0

    Well presented. So IBUF and OBUF differences in timing can not be compensated. This limits how good is the balancing.

    5
    • ckegley 2 years ago
      0
      -0

      hollis, I think that’s the case. IBUF and OBUF insertion delay is small and therefore variation is small. BUFG reflects clock tree insertion delay. So if clock drives a large number of FFs, clock tree delay can be large too.

      0
  2. UManish 3 years ago
    0
    -0

    Jingzh, good writing about clk deskew. Maybe it is simplified that in your drawing BUFG drives OBUF directly. Actually ODDR is needed for this kind of clk forwarding. More detail is in my quick words.

    4

Contact Us

Thanks for helping us better serve the community. You can make a suggestion, report a bug, a misconduct, or any other issue. We'll get back to you using your private message ASAP.

Sending

©2018  ValPont.com

Forgot your details?