[Cryptech Tech] Updated novena noise board

[Fredrik Thulin]

On Thursday, May 28, 2015 09:28:26 PM Jacob wrote:
> >>> If the lack of feedback on the list is to be taken as approval, I'm
> >>> ready
> >>> to order real PCBs.
> 
> Sorry for the delayed action on my part. It wasn't a good month.
> I attached my comments as a txt doc.

Thanks! I'll include the text and respond inline here:

> NOISE BOARD REV.03 COMMENTS BY JACOB, 28 MAY 2015
> ----------------------------------------------------------------------------
> --------------------- If I understand correctly, you had built a prototype
> based on the schematic and layout(?) that works. It that is the case, some
> of my comments, refering to comp values, may seem a bit out of place.
> Regardless, I have included all my comments below.

Correct, I've built three PCBs of this revision as well as ten or so of the 
earlier revisions for the Novena.
 
> Please also note that the comments are not exhaustive - there may be other
> detrimental issues that I haven't discovered yet.
> 
> Schematic:
> 
> 1. Capacitors with X5R composition are lower quality than X7R. I suggest to
> move all to the X7R type.

Ok. 

> 2. Due to the operational characteristics of
> Ceramic capacitors, it is common to select their voltage rating at least
> 80% higher than the max DC voltage they would encounter during service -
> for getting a reasonable life. Therefore, I recomment to change C12 to 10V
> voltage spec.

Ok. What I used in reality was the same as for C4 and C6 (25V) but I didn't 
know I ought to have 80% margin - I'll update the schematic.

> 3. Al caps like C1, C9 (and possibly others) are rated at
> 1000 hrs. This is considered below par for IT equipment. I suggest minimum
> 2000 hrs life spec as with the Panasonic EEE-FK1E100R series. 

Hmm, hrs as in hours? How is that measured? I must be reading that wrong - 
1000 hours is just 41 days =).

Anyway, thanks for recommending a specific series.

> 4. The ORANGE
> LED off the -15V rail in the DC-DC supply is not specced out, but I doubt
> that you would see any light out on a regular led at 1mA current. I suggest
> to reduce R3 value to 1.5 KOHM to get 8 mA or so through it.

I used to have orange LEDs of unknown spec, but ran out so started using amber 
LEDs instead - also of unknown spec. It does shine quite clearly with 15K, 
although not the brightest level I'm sure. You can see it in this picture:

  https://wiki.cryptech.is/attachment/wiki/WikiStart/trng_on_novena.jpg

> 5. C7, C10 can be 0603. In tight space 0805 is taxing.

Yes.
 
> Layout:
> 
> 1. There are no Ref Des silk in Gerbers. Also no comp outline in the Noise
> section.

Ok, I'm not used to creating Gerbers :(

> 2. I am not sure what the metal constructions (connected to AGND)
> on the periphery of the noise section (on all 4 sides) are. Are those for a
> shield mounting?

Correct. You can see the shield (which has a base part and a snap-on top part) 
in the picture I linked above.

> 3. Why are the footprint of JP1 and JP2 not in-line?

It simplifies mounting of 100-mil pin headers to have the holes slightly 
offset. The pins won't wobble around as much while soldering them. You can 
actually get the headers at something very close to 90 degrees like that, 
instead of 82-99 degrees ;).

The footprints are from the SparkFun Eagle library if I remember correctly.

> 4. Missing fiducials for assembly

Right. Can those be placed arbitrarily, or are there standard distances, 
regions etc.? Is three a good number? What size ought they be?

> 5. I would prefer to see wider traces going to PWR/GND pins

Ok

> 6. Should design in some openings in the soldemask of the connector pads. As
> it is now, there is a danger of shorts during assembly.

Sorry, this one I don't understand?

> 7. Since only lab
> measurements will determine the amount of noise coupling between the noise
> section AGND and the board GND, I suggest to also have an optional inductor
> (choke) placed as an experimental bridge between the 2 grounds. That
> inductor will be assembled only if the current ground topology will be too
> noisy. In that case, one can scrape off the current copper bridge between
> the 2 grounds and connect the grounds via the inductor. A 0805 footprint
> will be good.

Ok. I got some other feedback from Peter and Pavel earlier today so I'm going 
to rotate the AGND section 90 degrees clockwise too, to get shorter traces 
going into it.

> 8.I have some issues with the acute angle of some of the
> traces, but not to such a concern to flag them out now.

If it's not too much trouble I would appreciate an example pointing out good 
and bad ones. I'm always interested in improving and learing from those more 
experienced =).

Thanks again for your feedback, always appreciated.

/Fredrik