Beware of Vref pins becoming "unused" (Xilinx)

Gabor,

Do not use a high impedance reference generator.

Aust> I've run into an issue with the Bitgen software and

I'm getting awful confused....

1) I don't understand what a "high impedance reference generator" is. I can just about understand what a "reference generator with limited output current" is, but a reference generator with high impedance output sounds to me about as useful as a chocolate teapot. Do we, or do we not, want a fixed voltage? 2) With that doubt in mind, I looked up the cited LP2995. Its quoted maximum output current is no less than 1.5 *amps* (not a typo; amps, not milliamps) on its Vtt output pin. 3) The Vtt output is quoted as having a rather high output impedance of 5k ohms. On closer inspection, though, we find that this is its slope resistance over the rather narrow crossover region where its Vtt output current is only a few microamps. Outside this crossover region, the output slope resistance appears to be in the milliohm region. Its regulation of Vtt remains excellent.

So, chaps, just what is going on? Is Gabor mistakenly trying to use the Vref output of his LP2995 (which is feeble, and can source/sink hardly any current at all)? Or is Austin hiding from us the unpleasant fact that the reference pins need over an amp to keep them happy? (Only joking - I hope!) Or have I missed some important piece of hidden folklore?

Gabor, are you trying to use Vref instead of Vtt from the LP2995? And if you're using Vtt, did you remember to tie the Vsense input to the same place?

--
Jonathan Bromley, Consultant

DOULOS - Developing Design Know-how

Jonathon,

All we need is the leakage (less than 10uA + or -) per Vref pin.

But I agree that a high impedance Vref generator is not very useful (or desirable).

Aust> >

(or

High impedance is always relative to something. In this case I am using the Vref pin of the LP2995, which was designed to create a more stable reference voltage than the Vtt (termination) output. The Vtt is used to source and sink current from a lot of 50 ohm resistors and has a good deal of noise on it. The point I was making in the post is that the tools make assumptions about the use of Vref pins based on the IOBs used in that bank, and free them up if there are no inputs in the bank which require Vref. The default action of Bitgen is then to add pull-downs to every pin, so in my case I had 6 "weak" pull-downs going against my reference generator. If I had instead used a resistor divider for Vref, I may not have seen such a dramatic change in the reference voltage, however there would be some sag and therefore the reference voltage would no longer be centered in the signal swing. Removing the default pull-down of unused IOB's fixes this problem.

Gabor,

Ahhh. Now I understand, unused pins in banks get tied if there are no Vref standards in the bank.

Got it.

Good point.

Austin

OK. I misunderstood your post - I thought that the LP2995 was dedicated to providing your reference. If it's biasing a bunch of terminations then it will, as you say, be noisy. Your heads-up is useful - thanks.

However, I still take issue with your use of the word "impedance", which seems to me to describe small-signal behaviour, whereas you are using it to describe the current-limited large-signal behaviour of a reference output. My guess is that the LP2995's Vref output exhibits a small-signal impedance of an ohm or less.

Yes, I know I'm being pedantic.

Sorry, it's the way I'm made.

--
Jonathan Bromley, Consultant

DOULOS - Developing Design Know-how

(snip)

As far as I know, it can be either V/I or dV/dI, and you just have to know which one it is. For real measurements you have to change enough to measure the change big or small.

-- glen