OpenWrt Forum Archive

Topic: AR9331 Piggy-Back memory Upgrade

The content of this topic has been archived on 6 Apr 2018. There are no obvious gaps in this topic, but there may still be some posts missing at the end.

I am about to upgrade my WR703N to 64MB with an Elpida EDD5116ADTA-6B-E RAM chip.

The chip comes from a 512MB sodimm and there will be 7 others left on it!

Does anyone know if these chips can be piggy-backed to get more than 64MB of ram? At a minimum I guess there is bound to be some electrical difficulties but is it feasible? Are there enough address lines to do that?

All input is welcome (I mean it!)

No way, the pin 17 of the ram footprint (address line A13, existing only in 1Gbit RAM chips) is isolated and unrouted, so I suppose that the AR9331 doesn't have such capability.

bummer :-(

I'm not a pro, but I guess matching a free dangling wire + resistor with one of the traces is out of the question! Even if the AR9331 had an A13 line.

Oh well... thanks for the reply nonetheless!

Well, I think the ram interface is hardware controlled, not by software, so we can't just wire a free GPIO to the new RAM CS pin (Chip Select) and let it work that easy.
Unless some developer say something, I think its impossible at the moment.
I'm also looking for a RAM upgrade for my 740n v4 (same hardware as your 703n) so that's why I searched if it could be done...

Unfortunately, you can't just  piggy-back an SDRAM chip on top of the existing one and expect it to work reliably sad

We are speaking here of 300 MHz here!

At this speed, the trace length has a magnitude comparable to the signal wavelength (inversely proportional to frequency, 1 meter or 3.3 ft), so having a mismatch will result in some bits of address/data/control buses arriving early/late and causing random read/write errors.

Do you see the small "snake" traces between the AR9331 and the SSRAM chip? These are done on purpose to match the different traces to be the exact same length...

Similarly, do you see all the resistor chips between the AR9331 and the SSRAM chip? There is one for every single trace. they are here to match the trace impedance between the 2 chips, so again at this high speed and for the same wavelength reason as above, there may be signal reflection if not impedance-matched.

I have done this on other devices and worked without problems, stacking a ran chip over the other, and just leaving CS pin alone and wired.
In this router the Ram works at 200MHz (400MHZ DDR), which is equal to 1.5meter wavelenght. So a 2-3 millimeters won't hurt tongue unless the signal is very exact, but as there are wait cycles (CAS, RAS, CL...) for accessing the RAM, I think there would be no problems.
For example this:

(Last edited by dabyd64 on 4 Sep 2012, 18:56)

I don't think you can guarantee the impedance in theses conditions lol

thats not mine! its an example taken from the web big_smile

The original questions were if there are some electrical difficulties and if it is feasible.

The answer to the first question is yes, there are some electrical problems, and up to 10 times the fundamental frequency minimum to have clean signal edges, so this normally requires both length and impedance matched traces. Of course, you can hack things like above, but don't expect it to work flawlessly: I doubt the manufacturer would have taken all these precautions if they were not necessary for reliability. Try by yourself to put a 2GSPS+ high-end scope in glitch capture mode if you have the opportunity, and you will see that these kind of events may actually happen with a bad design.

As for the answer to the second question, don't forget that addresses are multiplexed with RAS/CAS, so it looks like A13 is not required to double the capacity: the main Wiki page suggests to replace the original 16Mbit x 16bit chip by a 32Mbit x 16bit one, so this might be possible.

But in my opinion, replacing the SDRAM chip completely is a better option , but of course, this requires a hot air station.

(Last edited by Squonk on 5 Sep 2012, 01:33)

@Squonk: Thank you again for a very clear and concise response! It's always a pleasure reading your posts!

I have taken a soldering iron to my 703n again and replaced the memory chip. I used an Elpida EDD5116ADTA-6B chip salvaged from an old SODIMM.

I solder-bridged the whole thing and pried it carefully off the SODIMM. I did the same thing on the router, temporarily removing the uart connections.
Cleaned everything up with a copper wicker and some flux remover and soldered the new chip in.

It actually took me a few rounds of touch-ups and changing the iron tip to get it working but it looks ok so far :-)

Notes for the next time:

a) Liquid flux rocks!
b) Need to buy a hot-air station
c) NEVER, EVER again use cyanoacrylate to bond wires on a PCB for stress-relief!

Thank you!

+1 for hot air station and fux smile

Interesting topic!
But i'm not sure if we can got more than 64MB device for ar9331 with more than one ram chips,i will check it and see if we can do that.

The discussion might have continued from here.