[Gta04-owner] GTA04A5/Letux 2804: status update

H. Nikolaus Schaller hns at goldelico.com
Wed Apr 6 20:08:46 CEST 2016

Hi all,

> Am 25.02.2016 um 14:22 schrieb H. Nikolaus Schaller <hns at goldelico.com>:
> Hi all,
> Am 19.02.2016 um 18:50 schrieb H. Nikolaus Schaller <hns at goldelico.com>:
>> Hi all,
>> I have promised to give more regular status updates on this project.

well, there wasn't much to report in the past weeks but yesterday and
today I spend several hours in the soldering lab and have some promising
news to share.

>> 2. Time plan
>> I had planned to make the 1GB RAM soldering experiments this week,
>> but the tool to help getting a tiny amount of solder paste on the chips did
>> not yet arrive. Looks as if it comes mid of next week.
> The tool we have designed was manufactured and did arrive in time. But it
> does not yet work as expected. So we will have to get a modified variant.

We had received a modified variant of that but it did not work well and
reliable enough.

So I have developed a new method how to fix the chip while being soldered.

After some first experiments it was much better than anything before. An already
used and desoldered OMAP was soldered perfectly in position, but did not work
(maybe the OMAP chip was broken).

> Some of you may be interested in what we are doing here, so let me explain
> a little:      
> The key task is to rework an existing GTA04A3 or A4 board with the 1 GByte
> RAM chip before we can have all GTA04A5 produced with that one.
> The good thing is that we can take a fully populated GTA04A3/A4 board and
> already can remove the OMAP and the PoP chips. This is a process which
> works fine without damages.
> The other way round is the difficult one: to solder a new chip onto an existing
> PCB where there is already some solder on the copper pads. Initial experiments
> were disappointing. The chip did swim on the solder flux and was soldered
> after moving several 100µm to the side.

> On a fresh PCB we use a laser cut stencil which allows to print a very tiny
> amount of solder paste on each of the contact points of the OMAP3 BGA.
> This solder paste "glues" the chip at the right position and there is not too
> much flux and therefore the chip does not swim away during reflow soldering.

This has happily been solved differently now the new method. The omap does
no longer swim away.

> But since we already have components on the boards for rework, it is
> impossible to use the stencil printer and print solder paste onto the PCB.
> So the idea is to print the solder paste on the OMAP chip (it is like if you
> want to glue two things together: it usually doesn't matter on which part
> you put the glue on).
> This is where our specially designed tool comes into the game. It should
> allow us to to place the OMAP CPU chip (which is just 12 mm  x 12 mm)
> in the stencil printer and keep it in place.
> Basically it works, but what does not yet work is to fine-adjust the stencil
> so that the tiny openings (ca. 300µm diameter) are exactly over the OMAP
> BGA balls. So we get the solder paste printed at the wrong locations.
> You may wonder how others are doing such rework. The answer is: they
> rarely can do it with such fine-pich and package-on-package chips (like the
> OMAP3 is) at all. It is a task that many SMD repair houses decline to do.
> And unless you have miniaturized systems like smartphones, chips are
> much bigger, less critical in placement, reballing and resoldering.
> There are even DIY videos on youtube. But that is for 1mm pitch (and not 0.4mm).
> Or they have a very expensive rework station with laser optics, vacuum nozzles
> and 4D positioning system. And everything in precision for 100µm.
> This is where we hope the redesigned tool will be better. Next week.

So what is the current status?

Today I managed with the new fixation method to solder a fresh DM3730CBP
chip to an old GTA04A4 board (a board which never had worked).

The result was this:

Texas Instruments X-Loader (MLO) 1.4.4ss modified for GTA04 (Feb  9 2015 - 14:32:20)
No NAND detected
OMAP3630/3730-GP ES2.1
Board detected: GTA04A4
Memory: Micron DDR 128MB/bank
Reading FAT boot sector

not there
Loading u-boot.bin from nand
Skipped bad block at 0x80000
Skipped bad block at 0xa0000
Skipped bad block at 0xc0000
Skipped bad block at 0xe0000
Skipped bad block at 0x100000
Skipped bad block at 0x120000
Skipped bad block at 0x140000
u-boot.bin not found or blank nand contents - attempting serial boot . . .
## Ready for binary (kermit) download to 0x80008000 at 115200 bps. Press 'M' to break into low level monitor.

Well, this can easily be explained: there is no RAM/NAND chip (Package on Package)
on top of the OMAP and the MLO boot loader (coming from MMC) detects this. At least

So after this success, I tried to solder a SAMSUNG 1GB RAM chip on top of
the DM3730CBP. Optically it looks perfect but the DM3730 did no longer boot :(

The reason turned out to be that now there is a short circuit under the DM3730 making VDD2
not available.

To give you an impression, here is the PCB layout. The VDD2 connections are highlighted:

Compare this with the GND balls and you can see that they are sometimes side-by-side. So a
small (the pads have a distance of 0.4mm) solder bridge can easily make a short circuit. We would
need to X-Ray the board to show where the short circuit is - but that would not help to fix it.

So what makes the solder bridges under the DM3730 appear just when soldering a PoP memory chip
on top (and not before)?

Well, it is the well known phenomenon of warpage [1]. The DM3730 simply warps when being heated
up. And warps differently if the PoP is on top or not on top.

These forces may squeeze BGA balls during the soldering process and there is a risk that a solder
bridge appears. Not always and not always at the same location.

How can we solve that?


while (!done) {
	unsolder memory chip
	unsolder dm3730
	clean solder pads
	if (dm3730 > 0 && memory > 0) {
		solder new dm3730
		solder new memory chip
		if (test() == successful)
			done = true;
	} else {
		get new dm3730 and memory;

As long as we have time and money for running this loop and nothing else on the PCB breaks we
can repeat that several times...

The main problem is finding the spare time to do this. Each run through the loop needs ca. 3 hours
and each set of dm3730+memory costs ca. 60 €. 

So to summarise the status:

We are almost done and only some millimeters are missing to the marathon.

It is only a matter of time until we have a board where we can test the DM3730 (and software) with
the 1GB RAM option. Which is a prerequisite to order the final GTA04A5 production at the external
assembly house (they can pick&place and solder the dm3730+memory combination automatically
in much higher quality than our manual rework approach), because we have to tell them which memory
chip to use.

So I am still sorry that I am not able to announce a final production date. Anyways you can place
your (pre)orders:



[1]: http://www.ti.com.cn/cn/lit/an/swpa182c/swpa182c.pdf (slide 56)

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.goldelico.com/pipermail/gta04-owner/attachments/20160406/a7ef4451/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PastedGraphic-1.tiff
Type: image/tiff
Size: 87398 bytes
Desc: not available
URL: <http://lists.goldelico.com/pipermail/gta04-owner/attachments/20160406/a7ef4451/attachment-0002.tiff>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PastedGraphic-2.tiff
Type: image/tiff
Size: 92754 bytes
Desc: not available
URL: <http://lists.goldelico.com/pipermail/gta04-owner/attachments/20160406/a7ef4451/attachment-0003.tiff>

More information about the Gta04-owner mailing list