[Arm-netbook] news page for allwinner a10

[lkcl luke]

On Fri, Dec 16, 2011 at 7:57 AM, Alejandro Mery <amery at geeks.cl> wrote:
> Hi Luke,
>
> On Thu, Dec 15, 2011 at 23:17, lkcl luke <luke.leighton at gmail.com> wrote:
>> i've created a news page, here:
>> http://rhombus-tech.net//allwinner_a10/news/
>>
>> i just had a discussion with the owner of the factory this afternoon:
>> he just needed clarification and reassurance that the PCB layout would
>> contain everything that he needs for his own plans, as well as
>> fulfilling ours as well.  i wasn't able to reassure him completely
>> because i thought that the 44-pin connector that i was cut/paste
>> copying to create the layout mock-up was a 2.54mm pin-spacing, meaning
>> that the connector would be some 66mm long: it turns out that it's not
>> - it was 2.0mm spacing i.e. only about 50mm long.  so, well, i may
>> have overdone it on getting everything in :)
>>
>> anyway, he is just going to double-check that in the ultra-low-cost
>> case, 16 pins of GPIO will be sufficient for the majority of purposes
>> that he needs, which it is highly unlikely to be in excess of that,
>> for PVRs, Tablets etc.  even if it is, the expansion header will have
>> extra pins available.
>
> this walked up some doubts on me, first considering this 44pin connector:
> * what is the minimum thickness you are aiming?

 for mass-volume purposes where user hot-swapping is planned, the PCB
ideally is to fit into the standard 3.5mm case, which typically has
very very thin stainless steel walls.  like... 0.15mm.  we might have
to go to 5mm "Type II" (particularly for double-sided PCBs), but
that's allowed-for in the spec.  5.5mm Type II is allowed.  8.0mm Type
III is not.

 clearly, in this case, the 44-pin connector would *NOT* even be populated.

 for "engineering purposes" and for ultra-low-cost applications (where
hot-swapping is *not* planned for the final product), the 44-pin
connector would be populated.

> * how will the mechanics of replacing the module work if the extension
> is connected?

 you would, clearly, a) not have the case on b) would be opening up
the device because a standard PCMCIA connector could not be used.

> * what worths has the standardised pcmcia connector has if each module
> will implement their own "extension" headers differently?

 it covers a hell of a lot of mass-volume cases.

 in the instances where extra GPIO and low-level functionality is
used, i'm recommending to the factories that they deploy a 48-pin or
possibly the 64-pin version of the STM32F.  it's about $1.50 in large
volume, and i've found prices as low as $1 for the really small
version.

 the nice thing about the STM32F is that it's the CPU that's used in
the arduino-like leafpad maple, and already has clones such as the
Olimexino, _and_ there's a GPLv3 library for it, libopenstm32f, _and_
i recently found a Micro-Kernel OS that had been ported to it, forget
what it's called now.

 anyway, yes: the STM32F family has a vast range of functions that
cover the majority of use-cases, such as battery monitor, mouse driver
(yes, really, it's one of the example source code apps on the
olimexino web site!), keyboard matrix, LCD backlight driver, resistive
touchpanel driver, stepper motor driver, you name it, it and any other
similar embedded CPU like it can do it.

 i just have a better feeling about the STM32F, especially because of
its low price, than any other Cortex M3 i looked at.


> and related to RALink, there is a firmware blob in the kernel patch
> and you mentioned another for a chip in the motherboard, doesn't this
> conflict with the ideals of the freedombox?

 *sigh* not entirely.  ok, there are two levels of "freedom" involved here.

 the first is absolute Software for absolutely every device that is
programmable.  this is the ideal.

 the second is a more practical approach, especially in the case where:

 a) there are laws covering certain types of firmware (particularly
involving R.F.).
 b) Certification costs, in some cases like WIFI which are in excess
of $100,000 *just* for FCC Certification, with an additional $5,000
*per country* that Certification is required
 c) the cost and risk associated with having an on-board PROM or ROM
area which contains the firmware are too great.  why increase the
cost, and risk having to spend another $2 million on another set of
mask charges for the silicon _just_ because there happened to be a bug
in the firmware that was discovered one year *after* release of the
product?
 d) in some cases, as turned out with the Marvell WIFI that was used
in the OLPC XO-1, marvell used someone *else's* ROM-based vector
table-based Microkernel OS that they did *not* own: thus, even if they
wanted to release the full source code of the modified WIFI that the
OLPC Corporation paid to have written (!) they couldn't do it!

 so it depends on what level of "freedom" you realistically want.

 l.