[Tinkerphones] GTA04A5: good news and bad news

H. Nikolaus Schaller hns at goldelico.com
Wed Feb 22 17:55:30 CET 2017

Hi Paul,

> Am 22.02.2017 um 17:08 schrieb Paul Boddie <paul at boddie.org.uk>:
> On Wednesday 22. February 2017 15.49.53 H. Nikolaus Schaller wrote:
>> Sadly, this also makes it really the last GTA04 batch which will
>> ever be built (making these GTA04A5 boards rare collectibles).
>> Unless someone spends us enough money to make more experiments
>> with OMAP3 PoP soldering.
> Sorry to hear this particularly bad news! I wonder if anyone lurking around 
> the Pyra community might be able to bring some expertise to the matter.

Well, the Pyra is also made on the same production line of the same company...
They also built many OpenPandora devices, so it is probably the best expertise
we can find. An soldering OMAP5 is much easier since it is not package on package.

The key problem is the combination of the DM3730CBP + Samsung PoP memory which
seems to be "unsolderable" or at least not with the typical quality required
for financial success (>95% yield).

Thousands of OpenPandora with OMAP3530 and DM3730 and hundreds of GTA04A4
with DM3730 have been built before. But they used a Micron PoP memory
chip. This is unfortunately no longer in production.

The problem appears to be the thermal effects during heat up
and cool down how the DM3730 warps. And sometimes it warps 5% more and
squeezes the BGA balls. Sometimes it warps 5% less and the balls do not
give contact to the PCB.

> I noticed that certain technical issues, albeit in a different area, were 
> greatly assisted by people with access to tools and knowledge through their 
> workplaces or professional situations. Maybe some of those people, or others, 
> might have something to share about this.

We understand the problem fully and do not have a lack of information and it
is not a hobby production line but already a very professional one.

Two soultions would be easy:
* use a different memory chip (but it is not available)
* use a different processor (means redesign of the PCBs which is also beyond

So we are only trapped in fixed requirements...

>> This situation makes me assume a very high risk for the Neo900
>> project, since it uses exactly the same CPU and RAM/NAND PoP
>> chip as we have on the GTA04A5. And it was planned to use the
>> same production company and line. This means if we can't
>> get it soldered for the GTA04A5 by a company with >10
>> years experience and >150 projects per year, it is unlikely
>> that they will try again and be more successful for the
>> Neo900. They already nick-name this chip combination
>> "unsolderable chips". And it is also unlikely to find someone
>> else who can do it better without making ex(p+t)ensive experiments.
>> So the best for Neo900 would be to change the CPU/RAM chip.
>> But AFAIK, one of the key ideas is to use Maemo/Fremantle
>> which is closely tied to the OMAP3 architecture.
> Yes, this doesn't make me very comfortable about Neo900. Then again, maybe 
> someone involved there can figure this out and bring their experiences back to 
> this project.

Another thing is that the DM3730 is now ca. 10 years old and rarely used any
more. So it might become difficult to find people who still use it.

> Personally, I think some flexibility about the SoC would be acceptable for 
> Neo900, but I'm not invested in Maemo at all and have backed that project 
> because of the form factor, open hardware aspects, and Free Software support.

Me too. The issue is what the alternatives really are. Due to space constraints
it must be small chips. And PoP is the technology of first choice. Not many
hacker friendly ARM SoC are available as PoP.

Snapdragon: would be fine, but unobtainable in small quantities.
Mediatek: afaik the same
RockChip: I don't know if they have PoP mobile chips
Allwinner: AFAIK no PoP and not power optimized for mobile operation
Apple A10: unobtainable
Exynos: AFAIK not well supported

So TI did alway have the best support for the open/free soft/hardware
community (except the IMG PVR SGX). Hence the DM3730 would still be a very
good choice.

> [...]
>> 4. We hope that someone can help to fill the financial
>> gap. Well, this does not necessarily mean a donation
>> but if you know a project that is in need and can pay
>> an engineering team like we have, please contact me.
> On the topic of lucrative products, I guess you'll be selling the Letux 15 
> board that the Pyra uses as its CPU board, will you not?

Yes, that is planned as well, but selling hardware isn't very lucrative.
Especially where a Letux 15 will cost much more (ca. 150-200€)
than a Raspberry Pi 3.

And it needs projects that need the power of a Pyra CPU board (OMAP5 + 4GB RAM).
Maybe a Pyra based phone?

There is also the Letux Cortex 8 (AM3358 board) in the pipeline, which
is a stripped down and shrinked BeagleBone. But it is still a little more
expensive to build than buying a BeagleBone.

See: https://twitter.com/goldelico/status/816713817627492353

BR and thanks for encouragement,

More information about the Community mailing list