Xilinx DCM Post-Route Simulation Problem

I am attempting to perform a post-route simulation on a design containing a Xilinx DCM on a Virtex-II Pro used in Variable Phase Shift mode.

In behavioral simulation, when the DCM first locks and no phase shift is applied, the reference clock (clkin) and the clk0 (clkfb) of the DCM are in phase.

However, in the post-route simulation, when the DCM first locks and no phase shift is applied, the reference clock and the clk0 (clkfb) of the DCM are not at all in-phase.

I am resetting the DCM and applying phase shift exactly according to the spec sheet and Xilinx App Notes. Is this a known limitation of the Xilinx post-route DCM model, or does this reflect actual behavior? If it does reflect actual behavior, then how do I start the design with 0 phase shift between the reference clk (clkin) and clk0/clkfb?

Thanks in advance for any help you can provide.

What version of ISE is this? I vaguely recall having done a post-PAR sim of a DCM on a spartan-3e, and I didn't get any strange phase mismatches... I think that was on ISE 10.x or 11.x...

mrflibble said:

What version of ISE is this? I vaguely recall having done a post-PAR sim of a DCM on a spartan-3e, and I didn't get any strange phase mismatches... I think that was on ISE 10.x or 11.x...

Click to expand...

Hi mrfibble- this is the same ol' 9.1i SP3 we spoke about earlier (last version to support all features of V2P/ML310, 10.1i is more limited).

I am using the Variable Phase Shift feature on the DCM in the Virtex-II Pro. DCM documentation states that, in this mode, you can adjust the phase shift between (CLKIN) and (CLKFB/CLK0, CLK90, etc.)

My real question is this: How do I ensure I start with 0 phase shift between CLKIN and CLK0 before I begin shifting the phase with variable phase shift?

What you can do is check what value is needed to align CLKIN and CLK0 in simulation. Isn't it ooooh lets say 100....000? As in middle of the range results in 0 degree shift, and then increment/decrement value to increment/decrement phase shift. Isn't that sort of thing mentioned in the docs?

mrflibble- well that's the thing. I am using variable phase shift mode. In that mode, you can set an initial phase shift and then increment/decrement by one unit of phase shift from there. I was under the impression that, since this mode still uses feedback, once the DLL locks, if initial phase shift = 0, then CLKIN and CLK0 should be phase-aligned. In other words, I thought the DLL would be taking care of aligning the phase once it is locked. Is this not possible in variable phase shift mode? I have read all the docs and app notes on this and I am left confused.

I understood that. My point is .. do you KNOW a value of all 0 equates to a 0 phase shift or do you just assume this?

Case in point: you now use a value of 0, this gives you such and such amount of phase shift. How about you try some other cleverly chosen value and see how it changes? If you have a direct link to the pdf and the page number I'll take a look, beyond that I am too lazy right now.

OK well I figured out the answer, and it involves me being an idiot. The behavioral sim does not show any phase difference between CLKIN and CLK0 because the DCM has no delays for which to compensate. The post-route sim, on the other hand, shows what appears to be CLKIN and CLK0 out of phase, but this IS ACTUALLY PHASE-ALIGNED because the DCM is doing its job: using feedback to shift CLK0 forward in phase to compensate for delay in the feedback path so that CLKIN and CLK0 are phase-aligned.

So, to answer my question, when I asked "In other words, I thought the DLL would be taking care of aligning the phase once it is locked. Is this not possible in variable phase shift mode?"

Answer: The DLL IS taking care of aligning the phase once it is locked. This IS possible in variable phase shift mode.

Excellent. So just to be sure I understand you correctly, when you change the variable phase shift (in post-PAR simulation obviously you do see the phase changing accordingly?

mrflibble- yes, the phase changes as expected in post-P&R sim in variable phase shift mode.