Xilinx V4 Power Calculations

I am doing some evaluation on a possible XC4VFX20 design and had a
question about the Xilinx Power Calculator on the web.  This design is
over industrial temperature, and when I change the temperature in the
calculator some of the power values don't change.  For example,
changing the temperature alone (from 25 to 85 degrees C) causes the
VCCINT to go from 88mW to 226mW (as is expected) but VCCAUX stays at
88mW.  The PowerPC cores do the same thing (no change over temp).  Any
experiences with the accuracy of this are appreciated, as I need to
size this for power.  Even though the PowerPC is a hard core it should
still vary at least somewhat over temperature, correct?  I assumed the
same for VCCAUX as well.

Any previous experience and knowledge appreciated.

Thanks,

Jim Davis
JDDC

JD,

I'll look into it.

Austin

JD_Design wrote:

> I am doing some evaluation on a possible XC4VFX20 design and had a
> question about the Xilinx Power Calculator on the web.  This design is
> over industrial temperature, and when I change the temperature in the
> calculator some of the power values don't change.  For example,
> changing the temperature alone (from 25 to 85 degrees C) causes the
> VCCINT to go from 88mW to 226mW (as is expected) but VCCAUX stays at
> 88mW.  The PowerPC cores do the same thing (no change over temp).  Any
> experiences with the accuracy of this are appreciated, as I need to
> size this for power.  Even though the PowerPC is a hard core it should
> still vary at least somewhat over temperature, correct?  I assumed the
> same for VCCAUX as well.
> 
> Any previous experience and knowledge appreciated.
> 
> Thanks,
> 
> Jim Davis
> JDDC
> 

Austin,

Thanks for looking into it; power analysis and regulator sizing are not
exactly the FUN part of the design process, but they still must be done
:-(

Any info appreciated.

Jim Davis
JDDC

Austin,

One more thing; I noticed that the same is true for the DCM (no
variance over temperature; I did not have the data for the DCM
previously so I had not entered it).

Since it is about 38mW (3 for VCCINT, 5 for VCCAUX dynamic and 30 for
VCCAUX standby) it could be a factor at 85 degrees C if it does indeed
vary over temperature.

Thanks for any help!

JD

Jim,

The Vccaux has so little running off it, that there is hardly any 
significant difference in current with temperature.

The 405PPC is always there on die, so the static leakage is also there, 
if you use it or not.  The dynamic current of the 405PPC hardly varies 
with temperature:  the stronger and faster at cold balances with the 
weaker and slower at hot.  Faster edges spend less time not being a 0 or 
a 1, slower edges spend more time between 0 and 1.  Stronger transistor 
draw more current (but for less time).  Weaker transistors draw less 
current (for more time).

Yes, we could be more accurate for a particular process corner, but it 
would not be worth it:  we need to be safe for all process corners.

Austin

JD_Design wrote:

> Austin,
> 
> Thanks for looking into it; power analysis and regulator sizing are not
> exactly the FUN part of the design process, but they still must be done
> :-(
> 
> Any info appreciated.
> 
> Jim Davis
> JDDC
> 

Austin,

I guess I would expect VCCAUX power to scale at least somewhat with
VCCINT power over temp since at room temp they only differ by a factor
of two for current draw (since they have the same power at room
temperature but different voltage).

Understood about the leakage current on the PPC; of course, if we
didn't want to use it we could have looked at the LX :)

How about the DCM (I added that question in a later post)?  It also
doesn't vary over temperature in the calculator.

Thanks for the info,

JD

Hi Jim,

JD_Design wrote:

> Austin,
>
> One more thing; I noticed that the same is true for the DCM (no
> variance over temperature; I did not have the data for the DCM
> previously so I had not entered it).
>
> Since it is about 38mW (3 for VCCINT, 5 for VCCAUX dynamic and 30 for
> VCCAUX standby) it could be a factor at 85 degrees C if it does indeed
> vary over temperature.

Following on from Austin's i/p, I'd just like to add that our V4
concentration (in terms of what the Web Power Tool models) has been on
where we have seen the greatest variation, i.e., VCCINT quiescent.

Brendan

>
>
> Thanks for any help!
>
> JD

Comments below,

Austin

> 
> I guess I would expect VCCAUX power to scale at least somewhat with
> VCCINT power over temp since at room temp they only differ by a factor
> of two for current draw (since they have the same power at room
> temperature but different voltage).

Vccaux uses all thick oxide transistors.  Leakage varies imperceptibly 
with temperature compared to all the analog stuff which has a fixed bias 
always flowing.

> 
> Understood about the leakage current on the PPC; of course, if we
> didn't want to use it we could have looked at the LX :)
> 
> How about the DCM (I added that question in a later post)?  It also
> doesn't vary over temperature in the calculator.

It varies A LOT with frequency.

It runs (partially) from a regulator supplied by Vccaux, at a lower 
voltage than the core, so leakage is again something of little consequence.

The control logic runs from Vccint, but is so small in the overall 
scheme of things on Vccint, that again, it is lost in the Vccint leakage 
with temperature.

Austin,

I guess I am concerned that all of these "imperceptible" changes in
power over temperature that are calculated as ZERO in the power
calculator could actually add up to be something.  Are you saying that
the combination of VCCAUX, multiple DCMs and multiple processors will
still result in zero power difference between 25 degrees C and 85
degrees C for those functions?

Thanks,

JD

JD,

It is of course not identically 0, but it is small enough that 
variations from part to part will mask the effect.

So the process margin is much larger than the variation, so the 
variation is of no importance.

Austin

JD_Design wrote:

> Austin,
> 
> I guess I am concerned that all of these "imperceptible" changes in
> power over temperature that are calculated as ZERO in the power
> calculator could actually add up to be something.  Are you saying that
> the combination of VCCAUX, multiple DCMs and multiple processors will
> still result in zero power difference between 25 degrees C and 85
> degrees C for those functions?
> 
> Thanks,
> 
> JD
>