syncing the CLK0 outputs of two DCMs if they use CLKIN_DIVIDE_BY_2

John,

There is no way to synchronize two separate DCM CLKDV outputs in phase, nor is there a way to synchronize the CLKDV with the CLKFX phase (where they are both say dividing by 1/2) that we are aware of.

This is a deficiency that we realized, and are planning to fix in subsequent generations.

The logic that generates the sequence has no ability to be reset or strobed by a user logic signal to cause them to be in sync.

Just to help us, why do you do the same thing twice? If you explain what you are doing, perhaps there is another way to do it?

Austin

John Providenza wrote:

Austin -

I'm trying to generate three phase adjustable clocks, although my example only talks about 2 DCMs for two clocks.

The current design uses one variable phase shift DCM to divide down the input 500MHz (using CLKIN_DIVIDE_BY_2) into a phase adjustable 250MHz. Incoming data is coming in sync'd to the 500MHz clock, so this phase adjustment lets me tweak the stup/hold time for the incoming data.

In the current design, 2 other DCMs are cascaded off the "primary" DCM to allow us to create 2 phase adjustable clocks. No clock manipulation, just CLKIN leading to CLK0 and CLK180. We need to adjust these two phases independently, thus 2 DCMs for these.

Our timing margins are tight, and I'd like to avoid cascading the jitter from the primary DCM into the 2 secondary DCMs.

I find little information about how the CLKIN_DIVIDE_BY_2 option effects timing, etc, so any insight is appreciated.

Thanks for your help.

John,

OK, got it.

The CLKIN_DIVIDE_BY_2 is a simple high speed toggle flip flop at the input of the DCM with a dummy FF of identical delay in the clock FB path (to allow for proper phase alignement).

This FF has no ability to be synchronized (reset).

I agree that you do not want to cascade DCMs as the p-p jitter does accumulate as the square root of the sum of the squares which takes away timing margin.

It is possible to multiplex two clocks by passing them to a BUFGMUX element, so taking the CLK0 and CLK180 to the BUFGNUX would allow selection of the "right" clock. You would have to take the CLKFB from the CLK2X back to the CLKIN, however (as if you make the CLK180 the feedback, it would then make CLK180 = CLK0!!!).

Resetting the DCM repeatedly until you get the phase you want is also problematical, as it may take many resets to accidentally get the "right" phase alignment from two separate DCMs.

Is it possible to change the design so that all you use is the phase shifted version of the clocks? The shifted CLK180 used instead of the non-shifted CLK0?

Aust> Austin -