Why not DDR in FPGAs?

to


I have not tried this with Synplicity yet.  With XST the Coolriunner dual
edge flip flop is inferred with the following VHDL construct:

   process(vclk)
   begin
      if (vclk'event) then
               ...
      end if;
   end process;

Which is exactly as one would expect.  To get this inference to work, if
running XST with a script, the part type has to be set to XBR rather than
XC2C64 (as one would expect from the documentation).  The flow engine sets
the part type correctly.


Regards,
Erik Widding.

The Xilinx Spartan-3, Virtex-II, and Virtex-II Pro FPGAs have DDR registers
as part of each I/O block.

For high-performance applications, you would want to use two outputs from
the Digital Clock Manager, one being the de-skewed, non-shifted clock output
and the de-skewed, 180 degrees phase-shifted clock output.  This technique
minimizes any potential duty-cycle distortion and gives you most of the
entire half-period for your logic application.

You might take a look at the following web site as it contains both
application notes and sample code.
http://www.xilinx.com/xlnx/xil_prodcat_systemsolution.jsp?title=xaw_memory_dram_ddr_page

I think the origional poster means the opposite, having the internal
registers in teh FPGA trigger on both the rising and falling clock
edge, to save power in clock distribution.

Unfortunatly, this would require even clock phases in all cases,
significantly more complex flip-flops, and I'm guessing the
power-savings in the clock distribution (now running at half speed)
would be swamped by the additional power draw within the flip-flops
themselves to allow them to trigger on both edges.

... but takes twice as many BUFGs.

Feature request: more BUFGs.

Grip: differential clocks take two global clock pins. It would be better
if they took one global clock pin and the other pin in the differential
pair was a standard I/O pin. I don't know if that is feasible, but
differential clocks chew IBUFGs quickly.

On the project I'm currently working on we ran out of global clock pins
this way and put one of our clocks on a non-global pin. Works fine.
Our experience shows that it's preferable to have differential clocks
even on non-clock pins than to use single-ended clocks.


Hamish