Inverter Chain Synthesis Problem

I would use a counter or shifter to generate the delay.

-- Mike Treseler

Hi Mike,

I mean a comb logic delay.

Davy

A combinatorial chain of inverters is probably a bad idea. The timing isn't easily guaranteed within a tight range especially since routing isn't constrained to specific paths. If you *must* try to keep the inverters, synplify would use the syn_keep attribute on the combinatorial elements you don't want removed. Even then, you may need similar constraints in place & route.

Please don't go there.

actually whaen u r generating a netlist if u r using series of inverters but using only first input and the final output the synthesis tool will won't consider and it will think as redundant terms during translation phase ............... i think it's easy to use a counter to generate delay .............if there is any ways of generating delay just let me know

Hi,

Thank you for your help :-)

Do you think what's the best way to generate delay smaller than a clock period?

Best regards, Davy

Davy schrieb:

Hi Davy, unfortunately you didn't tell why you want to create such a delay. That would help to give you useful help.

One solution may be to feed the signal into a DFF that is clocked by the falling edge of your clock signal (assuming that the rest of your circuit uses the rising edge). Thus you get a delay of half a clock period.

Another way could be constraints. While max_delay constraints are available I'm not sure about min_delay constraints.

The problem with combinatorical delays in FPGAs is that they are somewhat meaningless, since the routing delays between the LUTs are much higher and (unless propperly constrained) not predictable. So this approach is probably the worst choice, sice it is wasting LUTs for no reason.

Have a nice synthesis

Eilert

If you have a steady clock in the right frequency range for your device, you can use a PLL, DLL, or DCM to get a different delay than 1.0 or 0.5 clock periods.

The DCM especially allows you to set up a shifted clock with 1/256th of a clock resolution (or about 50 ps, whichever is greater) of delay. This precisely delayed clock can give you the precision control over skew that today's designers need.

If you're in Virtex-4 devices, the inputs have per-pin controllable delays that are calibrated to give you 78 ps delay resolution on unclocked signals.

So - is your device able to supply the fun stuff? More application details can bring out the info on the "best" feature set to use.

For synplify tools,

wire [3:0] inv_chain/*synthesis syn_keep=1*/;

assign inv_chain[3:0] = ~{inv_chain[2:0], din}; assign dout = inv_chain[3];

I'm not use what XST uses in equivalence to syn_keep. It might be the same.

Like other ppl pointed out, you shouldn't expect to use invertor chains for any precision timing control. There are just too many uncertainties in FPGA. But if you just want to insert some combinational delay to solve some racing conditions, it might come in handy.

cheers,

jz

Again, noting other's comments about generating delays, the simplest way to do this is simply to instantiate them (and don't flatten).

not_gate u_not_1 ( .in (input), .out (sig1) );

not_gate u_not_2 ( .in (sig1), .out (output) );

etc.

There are some works about delay elements for asynchronous designs in Xilinx FPGA. Google for them.