Virtex2 (500) DCM Frequency Synthesize

Hello
I've got following problem:
I need 27MHz and 54MHz clocks, input freq. is 20MHz.
Till today I used DCM FS 27/20 next to DLL and DLL*2.
Theoreticly it should works OK, but doesn't. From time to time after
reboot or after clok stop, it works strange and completly bad. Today
I've read about jitter on output of DCM's CLKFX, I check on "Virtex-II
CLKFX Jitter Calculator" that in this conditions I have over 2 ns
jitter on CLKFX output, which is to big for DLL input CLKIN.
So I try to do this way:
DCM FS 27/5 - I have 108MHz next to DLL/4 and DLL*2
The jitter on CLKFX should be below 1ns and now it must work but
doesn't.
What should I check else?
Please help me, I lose my hair :>

Best regards
Jerzy Gbur

Jerzy,

Since it sounds like you are cascading two DCMs, you must hold the 
second DCM in reset until the first has locked.

Cascading two DCMs from a CLKFX to a CLKIN to a CLKFX is not 
recommended, but has been characterized to work under most situations 
where the jitter from the first DCM does not exceed the requirement of 
the second DCM.

It is not recommended to have a CLKFX out to a CLKIN to a CLK0 (CLKDV, 
CLK2X) out (use of DLL) as this is a very poor cascade.  The CLK0 should 
always come from the first DCM to the CLKIN of the second DCM which 
generates the CLKFX (if the frequencies work out).

It would be better to just take the 108 MHz from the first DCM and 
divide it down with a synchronous counter to get the 27 and 54 MHz 
signals, why use another DCM?  This also reduces the jitter as a 
percentage of the period.

Not that I do not appreciate designers using every DCM they can (after 
all, I was on the team that designed it), the DCM is a short-cut to 
doing true synchonous design (with clock enables)!

Do not forget using the old simple techniques from years past, either.

Austin

Jerzy wrote:
> Hello
> I've got following problem:
> I need 27MHz and 54MHz clocks, input freq. is 20MHz.
> Till today I used DCM FS 27/20 next to DLL and DLL*2.
> Theoreticly it should works OK, but doesn't. From time to time after
> reboot or after clok stop, it works strange and completly bad. Today
> I've read about jitter on output of DCM's CLKFX, I check on "Virtex-II
> CLKFX Jitter Calculator" that in this conditions I have over 2 ns
> jitter on CLKFX output, which is to big for DLL input CLKIN.
> So I try to do this way:
> DCM FS 27/5 - I have 108MHz next to DLL/4 and DLL*2
> The jitter on CLKFX should be below 1ns and now it must work but
> doesn't.
> What should I check else?
> Please help me, I lose my hair :>
> 
> Best regards
> Jerzy Gbur

I would use two independent DCMs:
first one multiplies by 27 and divides by 10, generates 54 MHz
second one multiplies by 27 and divides by 20, generates 27 MHz

You can of course also generate 27 MHz by flip-flop-dividing the 54 MHz.

That gets you out of the cascading dilemma, when the output jitter is more
than the input to a DCM tolerated.
Always try to avoid cascading DCMs!
Peter Alfke, Xilinx Applications

> From: furia1024@wp.pl (Jerzy)
> Organization: http://groups.google.com
> Newsgroups: comp.arch.fpga
> Date: 7 May 2004 07:49:57 -0700
> Subject: Virtex2 (500) DCM Frequency Synthesize
> 
> Hello
> I've got following problem:
> I need 27MHz and 54MHz clocks, input freq. is 20MHz.
> Till today I used DCM FS 27/20 next to DLL and DLL*2.
> Theoreticly it should works OK, but doesn't. From time to time after
> reboot or after clok stop, it works strange and completly bad. Today
> I've read about jitter on output of DCM's CLKFX, I check on "Virtex-II
> CLKFX Jitter Calculator" that in this conditions I have over 2 ns
> jitter on CLKFX output, which is to big for DLL input CLKIN.
> So I try to do this way:
> DCM FS 27/5 - I have 108MHz next to DLL/4 and DLL*2
> The jitter on CLKFX should be below 1ns and now it must work but
> doesn't.
> What should I check else?
> Please help me, I lose my hair :>
> 
> Best regards
> Jerzy Gbur

First Thank All Of You, very much for answers

Peter Alfke wrote:

>I would use two independent DCMs:
>first one multiplies by 27 and divides by 10, generates 54 MHz
>second one multiplies by 27 and divides by 20, generates 27 MHz
>
>You can of course also generate 27 MHz by flip-flop-dividing the 54 MHz.
>
>That gets you out of the cascading dilemma, when the output jitter is more

Yes, I'll try that after weekend. It will be great that phase between
54MHz and 27MHz will be constans. Maybe dividing by flip-flop helps.
Another thing, that 54MHz must drive external ZBT SRAM, so I'd like to
use external feedback.
And how to do it without DCM?



Best regards
Jerzy Gbur