Measuring Power Consumption of Processor

[Nilesh Khambekar]

Hi,

question about basics of power measurement...

In the power equation, we have
1) static power by = I X Vdd
[static]
2) switching power = 0.5 X C X (Vdd^2) X f
[dynamic]
3) short circuit power = Q X vdd X f X N
[dynamic]
4) leakage power = I(leak) X Vdd
[static]

How do we measure these components?

-For finding static power, we find total current flowing (from voltage
drop across register). and Multiply it by Supply voltage (core
voltage). right?

-How do we measure switching power of processor assuming maximum
switching activity? we know f and Vdd. What about C ?

-How do we measure short circuit power ?

-Can we disable clock to find leakage power component of static power?

Thanks,
Nilesh

 

[John Larkin]

On 18 Nov 2004 14:14:58 -0800, nilesh.khambekar@gmail.com (Nilesh
Khambekar) wrote:

Hi,


2) switching power = 0.5 X C X (Vdd^2) X f
[dynamic]

1.0, not 0.5. Every clock has two edges.

John

 

[john jardine]

"John Larkin" <jjlarkin@highSNIPlandTHIStechPLEASEnology.com> wrote in
message news:i88qp094q0fnm7dsu2pf8halve69vnfk9t@4ax.com...

On 18 Nov 2004 14:14:58 -0800, nilesh.khambekar@gmail.com (Nilesh
Khambekar) wrote:

Hi,


2) switching power = 0.5 X C X (Vdd^2) X f
[dynamic]

1.0, not 0.5. Every clock has two edges.

John

Does not, the bucket fill on the positive edge, using the top FET?. Bucket
contents dumped on neg edge.
regards
john

 

[John Larkin]

On 19 Nov 2004 13:21:20 -0800, nilesh.khambekar@gmail.com (Nilesh
Khambekar) wrote:

Hi,

Thanks.

Let me pinpoint my question...
What is practical procedure to measure these power components ?

1) How do I get C for finding switching power ?

Measure the slope of the Icc-versus-Fclock curve. Or look for it on a
datasheet, where it's sometimes specified. Note that "dynamic power
capacitance" isn't always real capacitance, as it may include
significant shoot-through current spikes.

2) How do I get I(leak) to meaure short circuit power

Short circuit power is zero.

3) How can one measure leakage power

Measure Icc at zero clock frequency. Multiply by Vcc.

John

 

[Nilesh Khambekar]

Hi,

Thanks!

Can you suggest some useful book related to this topic?
Or some good online references?

This will help me doing the actual measurements.

Basically I am CS student having some work experience in embedded systems.

again thanks,
Nilesh


John Larkin <jjlarkin@highSNIPlandTHIStechPLEASEnology.com> wrote in message news:<bessp0lghkndloh3rsggu4ekeegq4c5ltb@4ax.com>...

On 19 Nov 2004 13:21:20 -0800, nilesh.khambekar@gmail.com (Nilesh
Khambekar) wrote:

Hi,

Thanks.

Let me pinpoint my question...
What is practical procedure to measure these power components ?

1) How do I get C for finding switching power ?

Measure the slope of the Icc-versus-Fclock curve. Or look for it on a
datasheet, where it's sometimes specified. Note that "dynamic power
capacitance" isn't always real capacitance, as it may include
significant shoot-through current spikes.

2) How do I get I(leak) to meaure short circuit power

Short circuit power is zero.

3) How can one measure leakage power

Measure Icc at zero clock frequency. Multiply by Vcc.

John

 

[john jardine]

"Fred Bartoli"
<fred._canxxxel_this_bartoli@RemoveThatAlso_free.fr_AndThisToo> wrote in
message news:419da7a2$0$26032$626a14ce@news.free.fr...

"john jardine" <john@jjdesigns.fsnet.co.uk> a écrit dans le message de
news:cnj9tr$pud$1@newsg1.svr.pol.co.uk...

"John Larkin" <jjlarkin@highSNIPlandTHIStechPLEASEnology.com> wrote in
message news:i88qp094q0fnm7dsu2pf8halve69vnfk9t@4ax.com...
On 18 Nov 2004 14:14:58 -0800, nilesh.khambekar@gmail.com (Nilesh
Khambekar) wrote:

Hi,


2) switching power = 0.5 X C X (Vdd^2) X f
[dynamic]

1.0, not 0.5. Every clock has two edges.

John



Does not, the bucket fill on the positive edge, using the top FET?.
Bucket
contents dumped on neg edge.

Sure, but... 1/2 C V^2 is just the energy stored into the capacitor and
doesn't account for the energy wasted into the switches.

For a cap charged from a supply U :

Wu = U int(i*dt)

and also U = int(i/C*dt)

which gives Wu = C U^2.

So, 1/2 C U^2 dissipated into the CPU at the rising edge, and the same at
the falling edge.


--
Thanks,
Fred.


Thanks Fred.

I'd come in from a simple SMPS 'transfer of energy' POV, assuming zero
switch losses and neglecting that it needs the averaged work to charge the
cap (*0.5) and then averaged work to discharge it (*0.5).
regards
john

 

[Fred Bartoli]

"john jardine" <john@jjdesigns.fsnet.co.uk> a écrit dans le message de
news:cnj9tr$pud$1@newsg1.svr.pol.co.uk...

"John Larkin" <jjlarkin@highSNIPlandTHIStechPLEASEnology.com> wrote in
message news:i88qp094q0fnm7dsu2pf8halve69vnfk9t@4ax.com...
On 18 Nov 2004 14:14:58 -0800, nilesh.khambekar@gmail.com (Nilesh
Khambekar) wrote:

Hi,


2) switching power = 0.5 X C X (Vdd^2) X f
[dynamic]

1.0, not 0.5. Every clock has two edges.

John



Does not, the bucket fill on the positive edge, using the top FET?. Bucket
contents dumped on neg edge.

Sure, but... 1/2 C V^2 is just the energy stored into the capacitor and
doesn't account for the energy wasted into the switches.

For a cap charged from a supply U :

Wu = U int(i*dt)

and also U = int(i/C*dt)

which gives Wu = C U^2.

So, 1/2 C U^2 dissipated into the CPU at the rising edge, and the same at
the falling edge.


--
Thanks,
Fred.

 

[Nilesh Khambekar]

Hi,

Thanks.

Let me pinpoint my question...
What is practical procedure to measure these power components ?
1) How do I get C for finding switching power ?
2) How do I get I(leak) to meaure short circuit power
3) How can one measure leakage power

thanks,
Nilesh

"Fred Bartoli" <fred._canxxxel_this_bartoli@RemoveThatAlso_free.fr_AndThisToo> wrote in message news:<419da7a2$0$26032$626a14ce@news.free.fr>...

"john jardine" <john@jjdesigns.fsnet.co.uk> a écrit dans le message de
news:cnj9tr$pud$1@newsg1.svr.pol.co.uk...

"John Larkin" <jjlarkin@highSNIPlandTHIStechPLEASEnology.com> wrote in
message news:i88qp094q0fnm7dsu2pf8halve69vnfk9t@4ax.com...
On 18 Nov 2004 14:14:58 -0800, nilesh.khambekar@gmail.com (Nilesh
Khambekar) wrote:

Hi,


2) switching power = 0.5 X C X (Vdd^2) X f
[dynamic]

1.0, not 0.5. Every clock has two edges.

John



Does not, the bucket fill on the positive edge, using the top FET?. Bucket
contents dumped on neg edge.

Sure, but... 1/2 C V^2 is just the energy stored into the capacitor and
doesn't account for the energy wasted into the switches.

For a cap charged from a supply U :

Wu = U int(i*dt)

and also U = int(i/C*dt)

which gives Wu = C U^2.

So, 1/2 C U^2 dissipated into the CPU at the rising edge, and the same at
the falling edge.