Why must a V4 be configured within 10 minutes of power up?

I was looking at the Virtex4 specs and notices a spec (Tconfig) that says you must configure within 10 minutes of applying Vccint. I don't plan on violating this spec, but it struck me as odd. Just out of curiosity, what happens if you violate this spec? Can you damage the part if you leave power on for hours but don't configure it?

-Jeff

Reply to
Jeff Cunningham
Loading thread data ...

Hi Jeff, search :-

nbti site:xilinx.com

HTH., Syms.

Reply to
Symon

Jeff,

As is pointed out by others, the issue is the worry over NBTI.

Negative Bias Temperature Instability:

formatting link

Is an old issue, still around, where the pmos threshold voltage changes over time given different bias conditions (full ON, vs full OFF).

In V4, the DCM delay lines turned out to be the best ever NBTI shift characterization means ever designed. If you intentionally held the CLKIN high on one, vs. low on another, and didn't ever change it for perhaps three or four hundred hours, when hot, you would get vastly different pulse width out of the two lines. This led to concerns that if you didn't keep the DCM delay lines "busy" they might not be able to operate over all PVT, all frequencies, and all duty cycles.

Never proven in the lab (never have ever tested a single part that failed to meet specifications, no matter how stressed), but we decided to specify our way out of ever having to even deal with the issue, hence the strange specification.

As well, there is an "autocal" block that is inserted for every DCM, used or not, that keeps them all "busy" so that the NBTI effect occurs evenly, and no performance could be affected by the change in Vt (because the all happen symmetrically).

A bake at 150C for 48 hours brings all the Vt's back to normal (almost, not quite), so if you ever suspect something to not work from NBTI, a bake will then provide you with "proof" that it might have been NBTI.

Might have been HCI (hot carrier injection), too, so a bake is not conclusive. HCI should never be a problem with our parts, at least, the qualification reports show it isn't an issue...

Bottom line, I stopped worrying about NBTI three years ago, and in V5, all the NBTI stuff is completely hidden (and hardened) so there are no worries whatsoever. In V4, we have recommendations that we apply, but even those may be waived under specific circumstances.

I have written, and signed official variance letters, allowing customers to violate some of these overly severe NBTI precautions when we are informed of what the real stress can be, and have studied what the worst case NBTI can be (and found it to be no issue).

Please address these requests to me directly, as well as file a webcase at the same time (so everything gets tracked and officially recorded).

Austin

Reply to
austin

Austin,

Our freshly manufactured cards go through a test procedure where they are first powered up and all the voltages are manually measured, and then a boundary scan test is run. All of this takes substantially more than 10 minutes. Do I understand correctly that it is not really an issue? Or should I enforce some strict rules at the company preventing anyone from having a card powered up for more than 5 min at a time unless it has been configured?

Thanks, /Mikhail

Reply to
MM

Mikhail,

If this is done once: absolutely no problem.

Even if done a 100 times: no problem.

The concern is that you accumulate more than a few hundred hours at elevated temperature in a static (non-switching) condition.

And, as I said before, we were unable to actually see anything happen to the DCM.

MGTs are also affected by NBTI powered on but left unconfigured (on the V4 FX), and there, you will see variations after hundreds of hours of static "stress." But, again, it doesn't seem possible to accumulate such a stress, even for a MGT in your case.

Aust> Austin,

Reply to
austin

Thanks for the reassuring information, Austin!

/Mikhail

Reply to
MM

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.