Modular SoC approach


sverm

Member
Joined
Aug 27, 2010
Messages
109
A while back, there were talks about a modular, upgradable SoC approach whereby the same "shell", i/o, and basic included peripherals could be re-used and the SoC somewhat simply upgraded.  Don't know if that's still in the cards (I couldn't find much talk about it in recent talks).

However the makers of RaspberryPi seem to have come up with an idea of their own in these regards, putting the SoC on a DIMM.  Perhaps some inspiration for Pyra design, if this is still something being thought of.

http://www.adafruit.com/blog/2014/04/07/breaking-news-raspberry-pi-compute-module-pi-on-a-dimm-raspberry_pi-raspberrypi/
 
Having a separate base PCB and CPU/RAM/PMIC PCB is still the current design, as far as what ED and gta04 have most recently said anyway.

To make this work they need to fit those components on the small PCB as tightly as possible, and use a custom socket that fits in this space without needing to make it bigger. I can tell you right away that SODIMM is going to be way too large for what they want. It's just not a good fit between what they need component-wise vs connector-wise, the PCB is not very compact at all. Instead of DIMM you'd prefer to have connectors on both sides and ones that plug in the bottom since an angle socket wastes more space on the base PCB. RPi's choice is nice because it's cheaper/simpler to design the DIMM PCB, it uses very easily available parts for connectors, and it's easy for users to replace, but it's just not size efficient.

HardKernel (makers of Odroid) have actually done this modular approach years ago, they got pretty compact results, eg like this: http://www.hardkernel.com/main/products/prdt_info.php?g_code=G129690436790 I think what Pyra uses will be much more similar to this than what RPi is doing.
 
The problem with these designs is that the communication buses between the modules evolve as fast as these modules. (Yay ! I can reuse my DDR2 ram ! Oh, it's DDR3 now...)
 
The problem with these designs is that the communication buses between the modules evolve as fast as these modules. (Yay ! I can reuse my DDR2 ram ! Oh, it's DDR3 now...)
As far as I know this daughterboard will have the SoC, System Memory and eMMC storage on the board. so there isn't much of a concern of the communication buses between memory and such getting outdated as it's all contained on the daughterboard.
 
Last edited by a moderator:
The problem with these designs is that the communication buses between the modules evolve as fast as these modules. (Yay ! I can reuse my DDR2 ram ! Oh, it's DDR3 now...)
As far as I know this daughterboard will have the SoC, System Memory and eMMC storage on the board. so there isn't much of a concern of the communication buses between memory and such getting outdated as it's all contained on the daughterboard.
USB 3, UHSII SDs, maybe higher resolution screen, etc.
 
The problem with these designs is that the communication buses between the modules evolve as fast as these modules. (Yay ! I can reuse my DDR2 ram ! Oh, it's DDR3 now...)
As far as I know this daughterboard will have the SoC, System Memory and eMMC storage on the board. so there isn't much of a concern of the communication buses between memory and such getting outdated as it's all contained on the daughterboard.
USB 3, UHSII SDs, maybe higher resolution screen, etc.
Well that wasn't what Gring was implying, also everything you are talking about would require a different peripheral board anyway if they change.
 
Last edited by a moderator:
Isn't the peripheral board much cheaper compared to daughter board? So if/when daughter board is updated, the peripheral board will probably be updated as well.

Instead, I would like to see ED using the same daughter board in other products. The daughter board seems to be costly to develop and setup production, selling more of them would provide better returns for all the investment he is making now.
 
Last edited by a moderator:
Isn't the peripheral board much cheaper compared to daughter board? 
Well kind of, the peripheral  PCB itself being most likely a 2 layer board would be cheaper to produce..  but I suppose It really depends on what you are changing, If it's a change on the peripheral board that would require an entire case redesign then it's not cheaper. 

So if/when daughter board is updated, the peripheral board will probably be updated as well. 
Doesn't that defeat the purpose, which is not have to replace the entire innards if you want to repair or upgrade an existing unit? The idea was to have a standardized connector that would allow a new SoC/Memory/Storage system to be easily dropped into an existing unit.

Instead, I would like to see ED using the same daughter board in other products. The daughter board seems to be costly to develop and setup production, selling more of them would provide better returns for all the investment he is making now.
I do like this idea.. but that can be done without having to change the peripheral board.
 
Last edited by a moderator:
So if/when daughter board is updated, the peripheral board will probably be updated as well. 
Doesn't that defeat the purpose, which is not have to replace the entire innards if you want to repair or upgrade an existing unit? The idea was to have a standardized connector that would allow a new SoC/Memory/Storage system to be easily dropped into an existing unit.
I though the idea was to allow design/development of the peripheral board to proceed without needing to prototype expensive 7-layer PCB boards, since an OMAP5 dev board can take the place of the daughter board meanwhile.

Also the smaller area of daughter board mean that production can be more efficient, can create more daughter boards from the same PCB area, thus reducing costs. The same GC machine could populate more daughter boards at the same time, because of reduced size.

From what I understand, even if the daughter board is always paired with the same peripheral board it would still be a win for ED&co.

Instead, I would like to see ED using the same daughter board in other products. The daughter board seems to be costly to develop and setup production, selling more of them would provide better returns for all the investment he is making now.
I do like this idea.. but that can be done without having to change the peripheral board.
Depends. For example it would be silly to try to use the same peripheral board to make a laptop like device which uses the same daughter board. The keyboard would be different, the layout ports at the back would be different, position of SD card slots would be different, there wouldn't be gaming controls, etc.
 
Last edited by a moderator:
I though the idea was to allow design/development of the peripheral board to proceed without needing to prototype expensive 7-layer PCB boards, since an OMAP5 dev board can take the place of the daughter board meanwhile.   Also the smaller area of daughter board mean that production can be more efficient, can create more daughter boards from the same PCB area, thus reducing costs. The same GC machine could populate more daughter boards at the same time, because of reduced size.
Well yeah, these are other benefits of this type of system.. but doesn't dismiss my points, the connector pin-out of the daughterboard could be standardized to a point where upgrading is possible even if a different SoC is chosen. Existing Pyras could essentially be upgraded. 

Depends. For example it would be silly to try to use the same peripheral board to make a laptop like device which uses the same daughter board. The keyboard would be different, the layout ports at the back would be different, position of SD card slots would be different, there wouldn't be gaming controls, etc.
Okay sure if ED wanted to get into other products, but that doesn't mean it requires changing the Pyra's peripheral board which is what I was getting at.
 
Last edited by a moderator:
Yeah sure, it would be nice to swap in an upgraded SoC module without having to change anything else.

Let's hope ED (or maybe someone else, since Pyra will be open hardware) can provide one at some point.
 
Last edited by a moderator:
This daughter-board...I've seen it mentioned in other threads but never fully understood it. Can anyone point me to the main thread that talks about it? 

Because from what little I'm able to understand at the moment, it sounds to me that the SoC (maybe even the RAM?) could potentially be replaced/upgraded to a newer, better one down the line after Pyra has finished development and been released.

If that's true, then I'd have no problem whatsoever buying a Pyra and upgrading it later. I'm already in love with almost every other aspect of it (still need to try out Linux on my desktop though, so that I can gain an understanding of how that OS actually works). Though, I probably shouldn't get my hopes up until I can get a proper explanation...
 
^

As mentioned, we're planning to put the CPU together with the memory and storage on a pluggable daughterboard. This saves money in production (since the mainboard will be two-layer and only the small CPU board needs to be more complex), and it adds the possibility to upgrade the Pyra in the future with whatever SoC / memory / storage will become available in the future.
From this thread: http://boards.openpandora.org/topic/15747-news-from-the-embedded-world/?p=318507/URL]
 
Last edited by a moderator:
^

As mentioned, we're planning to put the CPU together with the memory and storage on a pluggable daughterboard. This saves money in production (since the mainboard will be two-layer and only the small CPU board needs to be more complex), and it adds the possibility to upgrade the Pyra in the future with whatever SoC / memory / storage will become available in the future.
From this thread: http://boards.openpandora.org/topic/15747-news-from-the-embedded-world/?p=318507/URL]
 Thank you so very much!
 
Last edited by a moderator:
Not sure what threads talked about it - try the various news threads most likely.


But yes, the RAM will almost certainly be on there, assuming it'll be packaged on top of the CPU like it is on that OMAP3. I'm not personally expecting there to be much more than CPU and RAM on there, so the LCD-related chips, wifi, audio amps, USB controllers and so on will all be on the peripheral board but it depends on whether there's enough space on the 2-layer peripheral board to connect up all those chips (whereas the smaller CPU daughter board will be made of a more expensive multi-layer board to simplify routing).


As for learning Linux on your desktop first there are two trains of thought. Firstly, you don't need to have advance knowledge of Linux to use the Pandora, and hopefully not the Pyra either. I remember installing Linux for the first time on one of my PCs was a slightly fraught experience, and I wasn't at all convinced I'd got enough of the settings right to get it to install cleanly first time, though luckily I had, wheras the Pyra's OS will come preinstalled, so you don't need to worry about setting mount points and getting the dreaded wifi chipset working.


But on the other hand, learning Linux is a useful skill, and stuff you learn on PC is mostly useful on the Pyra. Knowing about the file system and what mount points you might want will come in handy if you choose to install your own OS on the Pyra down the line, for example.
 
But yes, the RAM will almost certainly be on there, assuming it'll be packaged on top of the CPU like it is on that OMAP3.
If it's OMAP5432, which is probably the only option, it won't be PoP. It will probably need to be on the SoC board anyway. Could be challenging though, or at least might not allow for too small of a board. Putting RAM on the bottom side of the daughter PCB doesn't sound that viable..
 
Last edited by a moderator:
As for learning Linux on your desktop first there are two trains of thought. Firstly, you don't need to have advance knowledge of Linux to use the Pandora, and hopefully not the Pyra either. I remember installing Linux for the first time on one of my PCs was a slightly fraught experience, and I wasn't at all convinced I'd got enough of the settings right to get it to install cleanly first time, though luckily I had, wheras the Pyra's OS will come preinstalled, so you don't need to worry about setting mount points and getting the dreaded wifi chipset working.
For the user-friendly distros, it's not like that anymore. I'd say the install for Fedora is probably easier now than the install for Windows.

-God Ginrai
 
Also this upgradable approach would allow for other things. Like even a little desktop version of the pyra. No screen or controls, just extra ports, maybe a few full sized ports. I'd buy that. A computer you could stash on a book shelve.

Sent from my Z665C using Tapatalk
 
Back
Top