Upgraded Pandora Possible?


rv6502

Member
Joined
Aug 9, 2009
Messages
153
http://www.samsung.com/global/business/semiconductor/products/dram/Products_MobileSDRAM.html

would the cpu support 512MB or even 1GB ?

shouldn't be too hard to swap the ram chip on top of the cpu with the right equipment.
or even do a few "special edition" boards.
 
Assuming the pinout is the same. I don't know enough about the chip packages to make a guess at that. The CPU should support 512MB at least, since dev boards using the OMAP3530 have been released with that much: http://www.igep-platform.com/index.php?option=com_content&view=article&id=46&Itemid=55
 
Um... Actually it should be very fucking hard, even with the best equipment you can get... I can't imagine how you would try to get the memory chip off the OMAP it is stacked onto without at least breaking the OMAP's connection to the board. Then getting the new chip soldered on cleanly looks even more impossible.

I'm pretty sure the SoC supports 512 MB. The Palm Pre Plus has that much RAM, for example. However, if I'm reading the OMAP TRM right, it only supports up to 2 Gbit, so 512 MB is the most you could put on there.
Edit: Wait, that makes no sense, because 2 Gbit is 256 Mbyte. So I'm probably looking at the wrong info...


Did you check if this would even fit physically?
 
OrR said:
Um... Actually it should be very fucking hard, even with the best equipment you can get... I can't imagine how you would try to get the memory chip off the OMAP it is stacked onto without at least breaking the OMAP's connection to the board. Then getting the new chip soldered on cleanly looks even more impossible.

Yeah, when I first read that my reaction was "you have got to be kidding me."
 
Last edited by a moderator:
Exophase said:
OrR said:
Um... Actually it should be very fucking hard, even with the best equipment you can get... I can't imagine how you would try to get the memory chip off the OMAP it is stacked onto without at least breaking the OMAP's connection to the board. Then getting the new chip soldered on cleanly looks even more impossible.

Yeah, when I first read that my reaction was "you have got to be kidding me."
Yeah, you and me both. Before I clicked on the link, I was assuming that it would be about using some of the unused headers to add support for something else. Which compared to desoldering and resoldering memory chips is quite sane. I assume that there are still unused connectors on the board that can be utilized.
 
Last edited by a moderator:
I'd like to know the motivation behind this question. It's not just MOR IZ BETAR is it?
 
I imagine it would be very difficult to swap out more RAM onto the chip, but perhaps someone with the right knowledge and equipment might (MIGHT) be able to.

I can see the motivation behind this... more applications running at the same time, for one. But if multitasking is something you want to do, maybe a netbook is a better idea. My phone has 256MB of RAM and it's plenty enough. I could have maybe 5 or 6 programs running simultaneously without problems, with maybe the occasional crash here and there, but ever since Opera Mobile hit version 10 I haven't once thought to myself "I wish my phone had more memory."

Anyway, that's my 2 cents. Mad props to you if you can do it though. :)
 
Isn't the Package-on-Package connection the sort of thing that would have to be wave-soldered? I can't imagine there'd be any way to redo that with the OMAP attached to a populated board, what with the, y'know, immersion in hot solder that the wave process requires.
 
We might make some boards with the 1ghz chip and more RAM as devkits for people in the industry, they would come with *no* support and be very expensive due to the custom nature of it, but there is a market for devkits with something like the Pandora case.
 
@QBJECT: Not a chance to do that in wave soldering! BGA has to be done in reflow, and both connections, OMAP to board and memory to OMAP, are BGA. The OMAP to board connection has a 0.4 mm ball pitch which IIRC took the factory months to figure out how to solder it reliably.
 
craigix said:
We might make some boards with the 1ghz chip and more RAM as devkits for people in the industry, they would come with *no* support and be very expensive due to the custom nature of it, but there is a market for devkits with something like the Pandora case.

are we talking like 400$ just for the board or much more ?

there might be a decent sub-market for baseprice+200$ pandoras if its not too expensive to order a few boards with higher-spec cpu & ram chips.
with the same chip footprint it shouldn't cause much trouble at the factory, no ?


... its easy to run out of ram with all the extra debug data, info & logs.


half-kidding for the next part,
my list for Santa:
http://www.zeph.com/bgarework_stations_systems_qfn_smd_hot_air_repair.htm

it looks almost too easy with this thing, a bit expensive for the few times I wished I had one, but still :D

as for DIY, I'm thinking epoxy-ing the cpu just a little bit so it stays in place while swapping the ram would be challenging but possible.
or building a small metal frame to align cpu&ram and then reflow them... but I'd be worried the metal would expand...
dunno. sounds like a fun challenge.
 
Last edited by a moderator:
craigix said:
We might make some boards with the 1ghz chip and more RAM as devkits for people in the industry, they would come with *no* support and be very expensive due to the custom nature of it, but there is a market for devkits with something like the Pandora case.
1GB RAM would be useful for developing. I have run out of RAM many times with only 256K.

I assume these would have to boot from SD since the larger RAM chips aren't available in the multi-chip package with flash. What kind of cost are we talking about here?
 
Last edited by a moderator:
Looking at the new Beagleboard-xM that should be out soon (if not already), they swapped the OMAP3530 for the 1GHz DM3730 with I believe 512MB RAM and no NAND for a modest $30 hike in price. The again I don't know what advantages they have in keeping costs down over the OPT.
 
Exophase said:
Yeah, when I first read that my reaction was "you have got to be kidding me."
well, OP did say 'with the right equipment', didn't he ; )

as a curiosity, i've heard of some ungodly machines used by silicon vendors (particularly AMD, but i'm sure all big fish have such) to literally patch a die post-production. to save time from re-spins of pre-production silicon, etc. think electronic microscope^2 -level of complexity & cost. also, i'm not sure if those would be still viable today as that was circa 130-90nm times.
 
Last edited by a moderator:
Ari64 said:
craigix said:
We might make some boards with the 1ghz chip and more RAM as devkits for people in the industry, they would come with *no* support and be very expensive due to the custom nature of it, but there is a market for devkits with something like the Pandora case.
1GB RAM would be useful for developing. I have run out of RAM many times with only 256K.

I assume these would have to boot from SD since the larger RAM chips aren't available in the multi-chip package with flash. What kind of cost are we talking about here?

256K? Damn. Get a Pandora soon. ;)

-God Ginrai
 
Last edited by a moderator:
God Ginrai said:
256K? Damn. Get a Pandora soon. ;)
Oops... heh. Beagleboard has 256M like the Pandora.

Hopefully Craig will get Pandora production straightened out soon.
 
Last edited by a moderator:
Ari64 said:
Hopefully Craig will get Pandora production straightened out soon.
Agreed. Too much time passing without any significant progress. So much nice software is being released now, and still no Pandora to test it with :(
 
Last edited by a moderator:
Caine said:
Ari64 said:
Hopefully Craig will get Pandora production straightened out soon.
Agreed. Too much time passing without any significant progress. So much nice software is being released now, and still no Pandora to test it with :(
OPTeam already mentioned accelerated and optimized Production processes nearly 1 Month ago but somehow it feels like everything is getting slower and slower. ^^""
I really hope this will change "soon".
 
Last edited by a moderator:
Well, as someone who has reworked a few BGAs in his time... I figure I can pop in here and show you guys a link to my other post here: http://www.gp32x.de/board/index.php?/topic/52920-some-questions-about-the-hardware/page__view__findpost__p__838902

No sense in typing that out again since it's all still there... so yeah, it's doable if you're REEAAALLY good at soldering, have the right equipment, and are willing to take the risk.

I wouldn't do it, and I'm the only person I know (in person, duh) who's ever actually been successful in replacing a BGA... just the normal kind of BGA too, not one of these package-on-package ones.

I can comfortably claim that I may be the most skilled solderer on these boards and not be excessively "puffing my chest" so to speak. I know I'm the most skilled at my workplace so this claim isn't too far out there. Notice I say that I MAY be the best... because I really don't know who else is out there or what their skill with liquid tin and lead is and the last thing I'm trying to do is offend, challenge, or make light of anyone else's talents. :D

Hopefully you get my point as I intended it: if a guy who's good enough to make a claim of possibly being the best soldering hand on the boards AND gets paid good money to be that good at soldering wouldn't try replacing the CPU/RAM BGA package-on-package assembly on his Pandora, then neither should you unless YOU'RE good enough at soldering to make similar claims.

QBJECT said:
Isn't the Package-on-Package connection the sort of thing that would have to be wave-soldered? I can't imagine there'd be any way to redo that with the OMAP attached to a populated board, what with the, y'know, immersion in hot solder that the wave process requires.

Actually, wave soldering is used for through-hole only... these parts are surface-mount. Additionally, during the wave solder process, printed circuit boards do NOT actually get submersed. Instead, a conveyor that only holds the board by the edges brings the boards down to the surface level of the molten pool of solder and glides it across the surface with the bottom of the circuit board just barely touching the solder. Surface tension then soaks the liquid solder up into the holes and accomplishes all that soldering for you. If the circuit board were to actually be submersed in the liquid solder, all the components on it would burn up.

All the parts for our Pandora motherboards are surface-mount and likely assembled by pick-and-place machines. Which means they "silkscreen" a sticky solder paste onto the surface of the circuit board and then run it through a pick-and-place machine which uses air suction to pick up each component from a tape-and-reel and then place it onto the appropriate place on the motherboard. Then after the machine places all the components onto the board it will send the board on to an oven which will bake the entire board slowly and evenly (to avoid warping of the board from the heat) up to just hot enough to melt the solder paste. The paste is sticky enough to hold the components in place until the boards go though the oven.

BGAs usually come "pre-balled" meaning the balls of solder that act as pins are already attached to the bottom of the chip. A pick-and-place machine still places them onto the circuit board but you don't silkscreen solder paste onto the pads first. Instead it would just be some tacky flux to hold the BGA in place until it gets to the oven.

At my company, we have our blank circuit boards go thru the pick-and-pull machines first to get only the BGAs attached. After the BGAs are done baking onto the board, the board is cleaned before being sent through the pick-and-place machine again for all the rest of the parts to be attached with the solder paste. Then after baking and cleaning one more time any through-hole parts are attached by hand. This is industry standard manufacturing techniques so I don't see why the Texas factory would be doing it any different.

Sorry to be so wordy... hopefully now you know a little more about how our Pandoras (and other electronic gadgets) are made and found what I had to say interesting... my apologies if it is not so. :D

darkblu said:
as a curiosity, i've heard of some ungodly machines used by silicon vendors (particularly AMD, but i'm sure all big fish have such) to literally patch a die post-production. to save time from re-spins of pre-production silicon, etc. think electronic microscope^2 -level of complexity & cost. also, i'm not sure if those would be still viable today as that was circa 130-90nm times.

I have a friend who works at Intel... he gets to play with one of those on a regular basis and oh yeah they still use them for the latest and greatest cutting edge stuff. I forget what they're called exactly, but these machines are used during the prototyping stage more than the actual chip manufacturing process. The chip developers leave extra silicon in there and the machine can focus a laser into the right spot to turn non-conductive paths into conductive ones in the silicon... my friend told me that during prototyping sometimes half of what's in the chip is dead silicon so they have the flexibility to make modifications with these machines. It's a lot cheaper to do a little laser surgery than it is to have a whole new prototype silicon die made.
 
Last edited by a moderator:
Back
Top