[Arm-netbook] EOMA68 card based on NXP i.MX7 (work in progress)

[Luke Kenneth Casson Leighton]

On Tue, May 16, 2017 at 8:28 PM, Vincent <ml.eoma68 at eml.cc> wrote:
> Hi Luke,
>
> Okay, it is good to see you on the same page when it comes to KiCad vs.
> Altium.

 yeah... ismo too.  he tried it... basically anyone who's ever used a
professional CAD package will scream in horror and try to dig out
their own eyes rather than use it :)

> Also, I wanted to give the community the chance to respond to
> this prior to making any design.

 awesome.

> I highly appreciate your effort in creating the template version of the
> i.MX7 design. I hope I can carry it on in your best interest without
> touching anything of the critical stuff.

 great.  i'll send it your way.  it should be pretty obvious what's
needed.  i *think* you might, unfortunately, need to rotate the eMMC
chip by 45 degrees (one way or the other) so as to make room for the
PMICs (without needing to do too much to the actual PMIC layout...
which i *don't* recommend altering if you can possibly help it).  but
to do _that_ you might also need to move one (or both) of the XTALs.
the "safest" one to move is the 32khz one - maybe replace that with an
MC149 footprint one (i think that's what it is: take a look at the
EOMA68-A20 board layout).  maybe MC-146?  anyway doesn't matter.

 one thing that will need sorting: the 5V DC input.  i haven't
completely finished that, i have however removed "battery" and other
stuff from the Reference Design, we'll need to go over that and make
sure it's clean/clear.

> Do you by accident know how critical the layer stack-up is in terms of
> material selection, e.g., in terms of using FR4 and having controlled
> impedances by the manufacturer?

 ok to give you some idea: DDR3 tracks which @ 3.2mil are nominally
100 ohms on a 6-layer stack on FR4 @ 1.6mm will drop to **HALF** that
amount if you convert to an 8 layer stack @ 1.2mm.

 *luckily* this is something that has been catered for in the actual
DDR3 standard, so you can have 40 ohm, 60 ohm, ... all the way up to
120 ohm impedance, and the DDR3 termination at *both* the PHY *and the
actual DDR3 ICs* can be changed in software.

 for the USB2 tracks, honestly they are so short that i wouldn't worry
about it.  just get them out on as few layers as you can with as
little modifications as you can, leave them @ 3.5 mil with spacing
around... 7 mil and be done with it.  just make sure that you set up
equally-spaced vias all the way.  general rule is: the more "artistic"
(i.e. the more "beautiful") the design, the greater the chances are
that it will work.  yes, really!  symmetry and beauty bizarrely have a
better chance of success than any amount of "engineering rules"!

 also *do not* put tracks close to high-speed signals, not even on
another layer.  if you absolutely absolutely have to put signals (any
signals) on neighbouring layers, make damn sure that they "cross" -
don't for goodness sake make them go "parallel".

>  For testing purposes etc., it would
> certainly be much easier to use a manufacturer that happens to be nearby.

 ... yyyeah... except i can guarantee they'll be charging you around
$2k to $4k for QTY 10 PCBs instead of only $USD 600 to 800, and god
knows what western PCBA (pcb assembly) rates are.  don't be surprised
if they quote you north of $USD 10,000.

 given that it's going to take at least 3-4 weeks for the PCBs and
assembly to be done _anyway_, what's adding the cost of a DHL courier
going to be?  $70 plus import duties?  is it *really* that important
to do "testing" that you don't want to save a whopping $8k difference
in the cost??


> Also, I wanted to point out that the i.MX7 security manual is only
> available under an NDA (one needs to sign up and request the file
> download). I'm actually wondering why this is the case because the stuff
> in there is not that special after all.

 bizarre!  well, one thing you might want to investigate: does that
NDA conflict with releasing source code or not?

> Thanks for summarizing all the other aspects (power profile, etc.).

 no problem vincent.

l.