frequency constraint changes routability

[John McMiller]

Hi.
I have a xc2v8000 design (70% utilization).

With the same EDIF netlist the Xilinx routability changes dramatically
with frequency:

Clock constraint: 25 MHz -> routed design
Clock constraint: 50 MHz -> 1200 un-routed wires
Clock constraint: 100 MHz -> 60000 un-routed wires

Unfortunately 100 MHz is my target frequency...

Is there a flag that tells the Xilinx P&R to prefer routing over
timing at the first phase, and do speed optimization only afterwards?

ThankX,
John

 

[Brian Drummond]

On 9 Sep 2003 00:20:14 -0700, john_mc_miller@yahoo.com (John McMiller)
wrote:

Hi.
I have a xc2v8000 design (70% utilization).

With the same EDIF netlist the Xilinx routability changes dramatically
with frequency:

Clock constraint: 25 MHz -> routed design
Clock constraint: 50 MHz -> 1200 un-routed wires
Clock constraint: 100 MHz -> 60000 un-routed wires

Unfortunately 100 MHz is my target frequency...

First run a timing analysis after mapping, before P&R, at 100 MHz (or
even faster). This tells you the maximum speed you can expect with
infinitely good routing. If there are any paths failing to meet timing,
you need to modify the design to remove the timing bottleneck (reduce
logic levels between registers), and generate a new EDIF file.

Improving the EDIF file will also work to improve the speed of
placed/routed designs, but you also have other options, such as
floorplanning, to improve the placement (which simplifies the routing
problem by ensuring related logic is placed together)

- Brian

 

[Andrew Paule]

sounds like new chip type is in order.

John McMiller wrote:

Hi.
I have a xc2v8000 design (70% utilization).

With the same EDIF netlist the Xilinx routability changes dramatically
with frequency:

Clock constraint: 25 MHz -> routed design
Clock constraint: 50 MHz -> 1200 un-routed wires
Clock constraint: 100 MHz -> 60000 un-routed wires

Unfortunately 100 MHz is my target frequency...

Is there a flag that tells the Xilinx P&R to prefer routing over
timing at the first phase, and do speed optimization only afterwards?

ThankX,
John