[Arm-netbook] EOMA-68 Carrier Board Concept

[luke.leighton]

On Thu, Aug 15, 2013 at 9:56 PM, Christopher Thomas
<christopher at firemothindustries.com> wrote:
>
>
> On Thu, Aug 15, 2013 at 9:20 AM, Scott Sullivan <scott at ss.org> wrote:
>>
>> On 08/15/2013 05:51 AM, Christopher Thomas wrote:
>>>
>>> So, Ejectable PCMCIA, 4 Port USB Hub IC, with 2 available external
>>> ports, two ports are being utilized for the ATMEGA32U4, and USB Audio
>>> IC.
>>
>>
>> I'd like to catch this now. It has been a major sin of ARM board
>> manufactures to only provide USB ports with USB connectors. The Via APC
>> boards are particularly bad for this, especial because they have a mini-ITX
>> like form factor (called Neo ATX?)[1].
>>
>
> We are kind of “hamstrung” by the limitations of the 68pins.

 that's... a "downer" way to look at it.  the alternative - the
flip-side - is to look at that concept modular PC that was developed
by a taiwanese company: they did a custom module, custom 100-pin
interface, developed working prototype PC and carrier board.... total
investment required: $USD 10m.  when i heard this i was like WAARG!
and for a while i was really jealous...

... by contrast, by re-using PCMCIA the biggest cost we've had is only
$8k... and that was for redoing the u-shaped plastic surround (oh, and
redesigning the micro-usb so that the front metal edge was straight,
supported by the plastic).

 that $10m investment cost... it's going to have to be amortised
across the unit cost of their products, isn't it?   [that's if they
get any customers]

 .... so, chris: you choose :)  would you prefer to have really
expensive custom-designed modular systems where you'd have to put in a
MOQ of 100k units and it would _still_ be expensive... or would you
prefer a budget system done on a shoe-string where there are no
investors, no bank loans to repay, and no shareholders demanding their
dividends be added on to the costs?


> Most of the
> functionality that is being proposed on this board is duplicating features
> already present on the A20,

 ... and then you look at say an ingenic MIPS SoC, and it has a
different set of functions which are quotes duplicated quotes, and you
look at another SoC and it has a different set of functions which are
quotes duplicated quotes, and you look at a TI SoC and it has a
different set of functions which are quotes duplicated quotes and the
EOMA68 set is, after you've analysed over 50 SoCs, the best
lowest-common-denominator set of interfaces that suits general-purpose
mass-volume computing appliances.

 but there's another reason why you shouldn't be looking at this as
quotes duplication quotes.  a) because of the reasons above: it's the
closest portable mass-volume interface we could find: there aren't any
80-pin 2.5in mass-volume connectors around, and there certainly aren't
any 100-pin ones: out of those which have robust metal casework
already in mass-volume we're stuck with CF, PCMCIA and ExpressCard
[removable MiniPCI Type III _would_ have been perfect - 124 pins - but
it never really took off]

 b) once you've quotes duplicated quotes the functionality you are
FREE.  you are SoC-independent.  you are no longer dependent on one
manufacturer.  i cannot emphasise enough how critical this is.  the
vendor lock-in problem - the "vertical market" problem - is *really*
serious.  we've gone from an x86 PC "any-OS-works-even-microsoft-ones"
architecture which retailed at between $199 to $1999 to a "fuck you
jack, we're blatant copyright thieves if you want the GPL source code
you can suck my dick" situation with a NINETY EIGHT PERCENT gpl
violations rate... in hardware that retails for between $30 to $50!

 c) the actual cost of buying two EOMA products that share a single
CPU Card is *significantly* less than the cost of buying two
identically-spec'd non-EOMA-compliant products.  i had a non-technical
guest stay with us for a few days: when i mentioned this to them, as
well as the e-waste savings issues and the long-term cost savings, and
they were like "wow!  when can i get one?  ok, actually, when can i
get two??"



> but unfortunately, there’s no perfect solution
> to replicate ALL the functionality of a standard ATX Platform and still
> maintain an affordable and "profitable" system.

 i do have EOMA-200 for dealing with this scenario.  it's a 200-pin
interface (duh), is similar to Q-Seven except, like all EOMA
interfaces, NOTHING is optional.

> Right now, I’ve done some preliminary calculations and consulting with
> manufacturers on how much a complete Turn Key system would cost, and if the
> volume of the EOMA-68 Card can get down to the more competitive levels
> predicted at the onset of the EOMA-68 project, then I feel I could get this
> down to a retail cost to the consumer at roughly $70 (That's including the
> EOMA-68 CPU Card, obviously less if the CPU Card can be had for $15?!).

 don't start :)

> Adding in $5 for an extra USB hub IC might not seem like much, but it could
> mean the difference between someone choosing the EOMA Platform vs the
> cheaper rPI or Panda Board.

 no: i don't believe it would.  you remember that you had *more*
people buy MEBs than we'd sold EOMA68-A20 samples?  there *are* people
out there who "get it".  they look at an I/O board and they go "huh.
i'm going to buy this I/O board as a long-term investment.  over the
next 10 years i can keep upgrading the CPU Card, i will *not* have to
throw away the I/O board every time.  i'm going to save myself a
fortune".

... that's on one hand. on the other hand: i agree with you.  $5 is a
hell of a lot for a USB Hub IC.  plus the extra connectors.

> What you lose in USB, you gain in potential
> gains in exponential CPU advancement (Laws of thermodynamics
> not-withstanding) and the ability to upgrade, which is not something any
> current development system on the market can truly boast. Now, if we wanted
> to go with the full blown Development Kit Version of a fully spec’d system
> as described by the EOMA-68, on a full size ITX board, that is DEFINITELY an
> option, but beyond the scope of what I was proposing, which is an
> affordable, easy to produce, open development/experimentation/consumer-esq
> device.

 that makes sense.

>> Some USB ports should be brought out on the conventional 0.1" headers you
>> see on ATX motherboards (or the USB 3.0 equivalent 20pin connector). This is
>> especially important if making it ITX case compatible. Cases will have front
>> facing ports, and not including a header means any user can not use their
>> case to it's fullest.
>
>
> A potential compromise would be to have the ATMEGA32U4 as a jumper
> select-able option on the board, with the suggested headers available as
> desired, or maybe consider sacrificing some of the GPIO of the EOMA-68
> Specification and bring out the 3rd USB Host port of the A20 to the header.

 nope.  too late.  plus, many SoCs simply don't have 3 USBs, they only
have e.g. one USB-OTG and one USB-Host.  if there was an extra USB on
EOMA68 you'd be down to 6 GPIO pins (which is not enough), and then
some CPU Cards would have to sacrifice USB-OTG on the front edge,
making it not possible to power by USB-OTG as a stand-alone
computer.... *or* they'd have to put up the price by at least $1.50 by
putting on a USB Hub IC on the CPU Card.

 general badness all round - been through this already chris, over the
past 18-24 months :)

>
>
> One of the main reasons I chose to locate the holes via the ATX locations
> was due to the increase in length. I thought, if I was going to be
> increasing the width/length to the standard ITX I/O Panel, might as well
> make it mountable in an ITX case, that's not to say it is an ITX computer
> though.

 you're still going to get people going "awesome!  i want to put this
in an ITX case!" :)

>
>
> I think you are spot on in that some sort of power management is required
> (Luke’s Suggestion of the AXP is a good one), as I had intended parts of the
> board to be usable even without the EOMA-68 Card inserted.

 the AXP209 is around $1.50 and it saves a loooot of hassle.  but the
AXP221 is probably a better bet because it can handle more current.
don't know pricing though.  i can get you the datasheet, if it's not
around anywhere.


> The idea of
> having a hot-swappable option comes to mind, or programming the ATMEL chip
> via it's USB Bootloader.

 via it is USB bootloader.  doesn't make sense.  i believe you meant
to use the relative pronoun, "its" :)

> I know I'm missing something, but I think that's the gist of it so far.
> Thank you for your input though, this is exactly the dialogue I was hoping
> to achieve.

  please draw people back to this
http://rhombus-tech.net/community_ideas/carrier_board/ and do please
keep it up-to-date everyone.

l.