cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
woko
Observer
Observer
2,420 Views
Registered: ‎09-13-2012

Protection for meeting the "Tvcco2vccaux" requirement

Jump to solution

Hi,

on our recently designed board (with Kintex 160T) we have a properly designed power-up sequence of the power supplies. Sadly, we didn't do that for the power-down. What we have read and heard is, that especially the Tvcco2vccaux condition is crucial to be met. As I said, power-up looks good, but during power-down, we have the Vcco3V3 to be around 200ms on where Vccaux1v8 is already off, violating the condition with 2,625V for 200ms (see screen). The datasheet says 300ms-800ms (85°C-120°C Tj). As we're facing some defects lately, we want to make this safe.

 

Here is what I want to ask you:

From your opinion, would it be feasible, to insert a series of three diodes (like 1N4148) between the two critical voltages, 3V3 and 1V8? This modification wouldn't let the difference of the two voltages exceed 1,5V (=3*0,6V), see sketch. Or is there another problem, which we don't see yet? This of course would be only a quick and dirty fix for existing hardware, before redesigning the board with a correct power-down.

 

Thank you, regards

diodes.png
TVCCO2VCCAUX powerdown gr3v3 ye1v8.png
0 Kudos
1 Solution

Accepted Solutions
austin
Scholar
Scholar
3,026 Views
Registered: ‎02-27-2008

No noise issue,

 

The "noise would have to beyond terrible to couple through three diodes not quite 'ON'.  Nominally, there is only 1.5 volts across the three diodes.  At -40C, the diodes will be ON strongly, so this diode clamp scheme will not work for cold environments.  I suspect, even for cold conditions, the system will warm up nicely pretty fast, so then the diode drop is > 0.7v, not ~ 0.5v (cold).

 

To force discharge of the Vcco 3.3v decoupling capacitors the diode clamp arrangement is just fine.  Unless it is operating with -40C air forced over the board.  Brrrr.

Austin Lesea
Principal Engineer
Xilinx San Jose

View solution in original post

8 Replies
klumsde
Moderator
Moderator
2,395 Views
Registered: ‎04-18-2011
Hi Woko,

It is important to understand that this is a lifetime spec.
so the number in the Data sheet is based off 240,000 power cycles.

Since you are not even at the maximum it would be probably more cycles until you saw a problem.
240,000 cycles is based on power cycling 32 times a day, every day for 20 years which is extremely unlikely

-------------------------------------------------------------------------
Don’t forget to reply, kudo, and accept as solution.
-------------------------------------------------------------------------
0 Kudos
woko
Observer
Observer
2,382 Views
Registered: ‎09-13-2012

Dear @klumsde,

thank you for the information about the lifetime spec.

 

The posted screenshot was measured in an laboratory-like environment and only on one single device. We'd like to make sure, that we are safe with this condition in every situation. That's why we thought about the fix with the diodes.

 

Do you think that modifying the board with the diodes mentioned would be an improve or do we run in possible other issues then?

 

Thanks,

0 Kudos
austin
Scholar
Scholar
2,366 Views
Registered: ‎02-27-2008

1N4148 diodes are likely to blow out (100 mA max!),

 

Choose diodes with sufficient forward current specification to discharge the 3.3v rail capacitance.  At least 1N400x series (1A diodes).

 

Assume resistance of diodes together is ~ 1 ohm,   If C=1,000 uF, then RC is ~1 millisecond

 

I = C dv/dt., dv=1.1v, t=1 ms, 1.1E-3 F/!E-3 = 1.1 ampere.

 

Typical of silicon diodes, they easily handle a surge of 10 times their rating.

 

So, 1N4148 may survive, or may not.  The resistance of three 1N4148's may just be high enough to save you .... but I would go with larger I diodes.

 

If the 3.3v C is smaller than 1,000 uF, 1N4148 are probably OK.

 

 

Austin Lesea
Principal Engineer
Xilinx San Jose
klumsde
Moderator
Moderator
2,363 Views
Registered: ‎04-18-2011

Good point Austin, 

 

I had a concern about coupling noise from VCCO to VCCAUX

VCCAUX is used by the PLL and MMCM so I was thinking this may not be a great idea from a performance point of view.

 

Do you have any thoughts??

-------------------------------------------------------------------------
Don’t forget to reply, kudo, and accept as solution.
-------------------------------------------------------------------------
0 Kudos
austin
Scholar
Scholar
3,027 Views
Registered: ‎02-27-2008

No noise issue,

 

The "noise would have to beyond terrible to couple through three diodes not quite 'ON'.  Nominally, there is only 1.5 volts across the three diodes.  At -40C, the diodes will be ON strongly, so this diode clamp scheme will not work for cold environments.  I suspect, even for cold conditions, the system will warm up nicely pretty fast, so then the diode drop is > 0.7v, not ~ 0.5v (cold).

 

To force discharge of the Vcco 3.3v decoupling capacitors the diode clamp arrangement is just fine.  Unless it is operating with -40C air forced over the board.  Brrrr.

Austin Lesea
Principal Engineer
Xilinx San Jose

View solution in original post

woko
Observer
Observer
2,307 Views
Registered: ‎09-13-2012

Dear @austin,

thank you so much for your detailed answering. The low-temperature issue makes sense, but our machines typically aren't placed in such a frosty environment ;-)

 

We will take this into account and look for some stronger diodes. As mentioned, this will only be a fix for our existing hardware, we will change the power-down sequence to a correct one in the upcoming release, and then the diodes aren't needed any more.

 

Thanks.

0 Kudos
jods
Observer
Observer
2,284 Views
Registered: ‎11-14-2008

When looking to your oscilloscope image, it looks that your 3V3 supply is still actively ON when your 1V8 supply is disabled. This would mean that the diodes will not 'pull down / discharge' the 3V3 supply, but rather that your 1V8 supply will be kept high (around 3.3-1.8=1.5V) until the 3V3 supply is effectively disabled. So the 3V3 will be powering your 1V8 and your diodes should be able to carry the current.

 

However the many voltage regulators contains a feature that during power off, their outputs are actively discharged. So your 1V8 regulator could potentially draw additional current through the diodes.

 

Another concern: some regulators do not like it when a low impedance power supply (the diodes) are imposed on their output.

 

My conclusion would be: before adding the diodes and creating a whole new bunch of potential problems, it might be useful to first investigate the root cause of the board problems you are experiencing.

0 Kudos
woko
Observer
Observer
2,271 Views
Registered: ‎09-13-2012

Yes @jods,

you're right, as the board is a PCIe addin card, we're powering the board from PCIe/ATX power supplies. The core voltages are derived from the 3V3PCIe and the 3V3 on our board are derived from the 12V (which is longer active on the mainboard than the 3V3PCIe). Thats why (and due to the lack of a proper power-down sequence) you see the 3V3 so much longer to be actively ON.

 

Thank you for your inputs about the diodes. We didn't thought about the 1V8 regulator, that's right. I've had some little bad feeling about it already, otherwise I wouldn't have started this thread. But you showed me the facts about it, thanks.

 

And another "yes", of course it would be the best to take more time to investigate the root cause of the defects... and we're up to it - but we are also under big pressure to deliver - you know, as always. For a small team like us, it's a complex challenge to find the root cause and of course we want to further use the existing hardware. No chance to wait for the redesign and trash all the existing boards... that's why we are looking for a quick fix. Just for your understanding.

 

But as @klumsde said, maybe this "Tvcco2vccaux" isn't such a big issue in our case as I thought.

0 Kudos