[Gta04-owner] CPU, Memory

Dr. H. Nikolaus Schaller hns at goldelico.com
Mon Mar 7 08:45:36 CET 2011


Hi Glenn,

Am 05.03.2011 um 13:33 schrieb Glenn Moeller-Holst:

> At 11:55 +0100 02/03/11, Dr. H. Nikolaus Schaller wrote:
> ...
>> Unfortunately we did not get beyond approx. 95% completion
>> so far. This means that approx. 90 connections
>> are still unrouted. And routing them manually with the tools
>> we have (we use EAGLE professional) is quite painful and
>> very time consuming.
> 
> How is that possible ;-) - you have an 8-layer PCB and PoP memory? (ironically)

Well, PoP allows us to reduce from 16 to 8 layers :)

> Now serious:
> 
> * Have you tried to rotate one or more chips X degrees?

Yes.

> * Maybe some chips are forgiving if you swap pin-connection - if also done in software/firmware?

No room for optimization.

> Rule of thumb/hints - maybe tested in arbitrary order - see if any of these can help you in any way - or just temporarily change your state of mind to look at the problem from a different angle/perspective of how to place and position the components:
> * keep HF wires short

Yes. But not the problem.

> * keep high current wires short

and wide.

This is part of the problem. Since wide enough power traces leave
almost no space for other signals. So we have to find the best
compromise between wide power traces and fan-out of the other
signals.

The key problem is connecting the 284 balls TPS65950 with the 512 balls
OMAP3. Both have a pitch of 0.4mm and the width of the wires don't allow
to add traces that can pass between two pads. I.e. any inner row must
go within the pad (Via-In-Pad technology) to one of the inner layers.

Even more critical is the B2B connector which has 0.3mm pitch. So when
we look at the unwired traces most of them are either TPS, OMAP or
B2B connector. Only a very small handful of others has different reasons
and the reasons were easily identified/solved.

> * keep LF wires short
> 
> * Just a few extra wires: Maybe place a PCB FM-antenna around the PCB perimeter with 90° diversity connections ;-) Sorry I could not resist...

Yes! Well - you can solder a wire for the FM antenna and wind it around the display
module.

> * Maybe add one more Micro-SDHC slot/connector somewhere on the board? - I already have problem with only one 16GB and I would like a big local backup partition.

We have no interface available and no space. But the MMC slot should be able to
handle up to 32 GB like the BeagleBoard.

> In some later revision Micro-SDXC might be supported? I do not know if the microcontroller can handle it.
> SDXC:
> http://de.wikipedia.org/wiki/SD_Memory_Card#SDXC_.28SD_3.0.29

I have not checked but most likely not. AFAIK, the SDXC protocols were announced
beginning of 2009 and the original OMAP3 design is older. So I think the MMC controller
(a hardware module within the OMAP3) is not prepared. There may be tricks (by doing
more in software, i.e. the Linux MMC driver) that may reduce bandwidth. But I don't know.

Maybe you could dig into the data sheets of the OMAP3530 or someone else has
thought about this?

Here is a discussion by others on this topic:

http://e2e.ti.com/support/dsp/omap_applications_processors/f/447/t/96917.aspx

Finally you may connect an external SD card reader/hard drive through USB - although that
isn't quite portable any more (the USB Host power output is specified for ~100-150 mA).

> 
> -
> 
> Do you place components on both sides?

Yes. Where possible.

If we remove the display we have plenty of space to fill. But I am sure that
this is not your intention...

> 
> 
>> 
>> We also have looked into KiCAD but it did not appear to
>> be mature enough for such a complex and miniaturized
>> project and would run into the same mode of manually
>> editing.
>> 
>> Well, there are high price commercial tools (Altium, Cadence, Mentor, ...)
>> which could be better supporting this work, but converting
>> the existing EAGLE project into a file compatible to those
>> may loose important information/knowledge/design rules
>> and may introduce new costly errors.
>> 
>> Therefore, we are continuing the slow manual path staying
>> with EAGLE... The good side is that we already know
>> that there are no problems in post-processing and
>> production based on the EAGLE files because we have
>> done this several times with our PCB supplier. This
>> is also an area were experiments are quite costly.
>> 
>> I hope you stay with us and support our project,
>> Nikolaus
> ...
> 
> At 14:18 +0000 02/03/11, Neil Jerram wrote:
> ...
>> And another one: please take your time, and don't take on too much
>> stress for yourselves.
> ...
> 
> Me too.

Thanks!

We have relaxed a little and got new energy to solve the remaining
issues (other teams and companies have shown that it can be solved,
so there is no technical reason that we can't).

BR,
Nikolaus



More information about the Gta04-owner mailing list