Micron D9 and high voltage, electromigration effects

[saaya]

i dont think so, i think 2.4v is still "relatively" safe
if the mfg gives you warranty with 2.4v, then use 2.4v.... its rather unlikely the memory dies, and if it does, then the mfg will replace it.

and IF it dies, its most of the time one of the "usual suspects" motherboards... and on those boards micron ddr2 can die even with 2.1v... so... id say avoid those "special" boards and most likely youll be fine, even with 2.4

[Buckster]

my first set have keeled over

ran them 99.99% of their life at 2.10 - 2.15V

P5B Deluxe

[Rammsteiner]

Ive been running 4x 1GB D9GMH at 2.16Vdimm, sometimes upto 1.3Vdimm in BIOS. Although recent measurements by Tony showed the mobo overvolts a little:

2.16=2.23
2.26=2.32

2.3 wasnt even tested. But the RAM is still fine. It's about 3 months old now. I use Weak Drive Strength though, so maybe that might be the case. On the otherhand, I still didn't see D9GMH complains from AM2 systems. Maybe Intel memory controller messes something up?

PS: I know I posted a thread a while ago about my D9's, but didnt turn out to be the RAM but the CPU.

[naokaji]

since someone necro'd the thread allready I thought I'd put in my list:

2x Cellshock 8000 C4 (GKX), still alive after 24/7 usage with 2,4V
One module Team xtreem dead after benching at 3,3V, other module still alive (GMH).
2x Geil 8500 (GMH) require 2,5V to boot after benching at 3V+
3x Crucial (gmh) dead on 2,3V (overvolting board though) (4th module still works, but degraded to the point where its not suitable for 24/7 anymore) (run as 4x1 GB in DS4)
2x Patriot 8000 C5 (gmh) still fine even after benching at 2,9V.
1GB modules all the way.

that makes:
2x GKX working fine after 2,4V 24/7
1x GMH working fine after benching at 3V+
2X GMH degraded after benching at 3V+
1x GMH dead after after benching at 3V+
2x GMH still fine after benching at 2,9V
3x GMH dead after 2,3V 24/7 (on DS4 which overvolts)
1x GMH degraded after 2,3V 24/7 (on DS4 which overvolts)

[saaya]

i really wonder what micron is doing here...
is it the chips? the logic? the interface? the package?
it could really be any of the above...
what i dont get is how they cant implement something to prevent this from happening...

cant they add a fuse in front of the high voltage sensitive parts, and if the voltage goes too high the fuse blows, restricting the voltage flow to the logic behind it? that way the part degraded but it still works at stock and the sensitive logic, it just wont scale with higher volts anymore...

but i guess once the fuse blows the logic behind it will get damaged as well?
hmmm

[Rammsteiner]

From what I read Micron is the only IC manufacturer which implemented a lot of variables regarding ODT, ODT, Drive Strength, DQS, skew stuff etc.

Apparently, when one of those things isnt running in the 'perfect' ratio with other settings things might get messed up. When ODT settings are too strong the data signals will be dampened too strong. We won't notice untill everything runs out of order and we need to throw extra Vdimm to increase the signals again, but that results in a stronger degrade etc.

Im not sure if this is 100% correct, but it makes sense.

Also the worse part, we've no single method to test our own settings because it requires expensive lab equipment etc. Im just wondering what settings exactly might cause this though. I think that timings and Mhz them selves aren't going to have any influence where as the more sensitive parts, which might work perfect at our Mhz and timings, simply can't handle wrong setup ODT and signal/drive strengths and the signal gets dampened too much in the end.

On AM2 we can adjust some values as CPU-ODT, 6 specific drive strengths and one general Drive Strength. Fun, but Ive no clue when I set these settings in a harmfull way and thus slowly kill my RAM. I dont even know IF any of these settings is strong/weak enough to kill the RAM.

A lot of people have Intel based PC's now because of the bench results. So the fact a lot of Micron chips die on Intel boards isnt strange, although I saw no single form of this anywhere on AM2 PC's. So Im really wondering what the Intel boards apply for these variables to be sure Im in a safe range with my settings or not.

[saaya]

and its mostly 680 an 975 that kill micron ics at low vdimm, maybe they have a high preset drive strength which is causing the issues?

its too bad micron doesnt share more details in their specs.
their specs are very vague and ... not really helping that much.

[Rammsteiner]

Whether it's the fault of Intel memory controllers, Micron or both... I dont know. I think it's a little comparable to timings in general and auto settings (without SPD/EPP reading). If you OC RAM a little bit with auto settings some semi timings are set very very high because it's specified by JEDEC/whatever formula for x timing at y Mhz. At least, I thought some timings worked like that. Not sure. I think tRC is an example. At auto it's all the way up at 45, but 11 would work as well.

So maybe at x Mhz the auto settings for ODT and drive strenths are applied a lot stronger because Mhz decreases the signal strength a little, but this might be after all completely unnecesairy.

As I said before, Ive really no clue how exactly things work regarding IC's etc, but maybe that's an answer why Microns die en masse with certain Intel boards.

[zsamz_]

its gotta be memory controller
i think micron chips just need stable power

[saaya]

well even if you say its the mem controller, then other chips can clock almost as well and dont get damaged... so still its a weakness of the micron chips.
no matter how you twist and turn it, micron should find a way to prevent this from happening somehow. with micron ddr3 its even worse, most chips wont degrade, they just stop working... thats in my opinion a worse reaction to high vdimm than micron ddr2 which degraded so at least you knew you were pushing them too hard...

[camouflage]

well even if you say its the mem controller, then other chips can clock almost as well and dont get damaged... so still its a weakness of the micron chips.
no matter how you twist and turn it, micron should find a way to prevent this from happening somehow. with micron ddr3 its even worse, most chips wont degrade, they just stop working... thats in my opinion a worse reaction to high vdimm than micron ddr2 which degraded so at least you knew you were pushing them too hard...

Is there other chips than Micron..........

[Monstru]

At least Micron chips can take 3v and still go higher, which is not what I could say about other DDR2 chips

[saaya]

some elpida and powerchip are clocking almost as good as micron.
well... almost... but the point is, its possible to have the mem perform well and NOT have it degrade or die from unstable or high vdimm... if the other guys can do it i dont see why micron couldnt...

yeah well, 3.5v wrs are cool but... honestly... id have no problem giving that up and instead having memory not scale beyond 2.5v, but 2.5v is 100% safe for D9GKX instead... It would be better for us and for micron too...

[Monstru]

almost is a good chosen word

I think it all depends on what your purpose is. If you want good performance in daily use....there are many chips around. If you want extreme results...well....there are always risks involved, not only in Micron's case. That's something a hardcore overclocker has to take upon him

[Rammsteiner]

Well, as I said, memory controller being the fault here should be blamed as well for killing Microns.

The memory controller seems to set too strong drive strengths which is unneeded. Maybe the memory controllers which kill Microns cant even select other than auto Drive Strengths. If they can do, it's still the fault of the memory controller BUT I think with nowadays possibilities of EPP/SPD it should be possible to force lower Drive Strengths?

If those are the problem of course. I just dont hope my RAM to die cause that would throw about every bit I said on this page in a pile of crap.

[4Qman]

Micron D9GTR's

Died with 2.2v.

To be honest i use them every weekend for around 10hrs on my benching system. Never went over 2.2v

Lasted 4mths DEAD "Will boot but crash on boot or fail superPI instantly"

[mad_skills]

2 X 512Mb Team Xtreem PC6400 with D9GMH microns, one died at 2.75V, the other one alive even after 2.9V benching...
Seems to me that D9GKX take the higher voltage with CL4 overclock better than D9GMH do the same thing...With CL5 o/c D9GMH works better for me but for CL4 D9GKX gets the job done..

[Oliver]

2 X 512Mb Team Xtreem PC6400 with D9GMH microns, one died at 2.75V, the other one alive even after 2.9V benching...
Seems to me that D9GKX take the higher voltage with CL4 overclock better than D9GMH do the same thing...With CL5 o/c D9GMH works better for me but for CL4 D9GKX gets the job done..

gkx is also higher binned and are really good at cas 3 and 4...

gmh is flying with cas 5, but high binned gmh can follow gkx close in the

[saaya]

gmh and gkx are identical, they are just binned different.
some batches of gmh and gkx clock better with cas5, 4 or 3 than others, but from what ive seen they dont differ in what latnencies they prefer, they are usually the same only that gkx clocks around 100mhz better on average than gmh

[Worthy]

What is it about the single sided ballistix 1066 modules they are selling now that are so bad?

I have returned three sets. All were bad in one day. Yet I have a set of double-sided 6400s that kill them and have lasted me two years without a problem?

Different revision chips, junk heatspreaders and thermal tape or what?

I have just thrown in 2 sticks of replacement modules at 2.10 volts and I have been running them throught the paces. No problems yet, but it seems that after about three weeks, they die.

They are cooled with their own 80mm fan in an opened case, so I don't get it.
When they fail, upping the volts does nothing so I don't believe it to be a case of electro migration (but I could be wrong)

I am being careful as I have a maximus formula that does indeed overvolt. I know that setting at 2.2 yields 2.19, so I am backed off to get an actual 2.10 volts. I see no prob with this since corsair uses the same chips and gets away with it.

Any take on this?

[Stelios]

So .. i'm asking again : Is there anything we can do to relax drive strength ? Mchbar , bios ???

The Clock twister (light , strong , moderate) in Asus bios has anything to do with drive strength ?

[mad_skills]

gmh and gkx are identical, they are just binned different.
some batches of gmh and gkx clock better with cas5, 4 or 3 than others, but from what ive seen they dont differ in what latnencies they prefer, they are usually the same only that gkx clocks around 100mhz better on average than gmh

Maybe you're right, i was just saying based on what i've experienced...

[saaya]

Maybe you're right, i was just saying based on what i've experienced...

sorry if i came off unfriendly, i know you only posted your experience, i just added my experience i used a lot of different chips from different batches and i was surprised as well, that the batches differ so much from each other.

What is it about the single sided ballistix 1066 modules they are selling now that are so bad?

I have returned three sets. All were bad in one day. Yet I have a set of double-sided 6400s that kill them and have lasted me two years without a problem?

Different revision chips, junk heatspreaders and thermal tape or what?

I have just thrown in 2 sticks of replacement modules at 2.10 volts and I have been running them throught the paces. No problems yet, but it seems that after about three weeks, they die.

They are cooled with their own 80mm fan in an opened case, so I don't get it.
When they fail, upping the volts does nothing so I don't believe it to be a case of electro migration (but I could be wrong)

I am being careful as I have a maximus formula that does indeed overvolt. I know that setting at 2.2 yields 2.19, so I am backed off to get an actual 2.10 volts. I see no prob with this since corsair uses the same chips and gets away with it.

Any take on this?

your fried 6 sticks by using high vdimm and voiding your warranty and NOW your thinking about what the problem could be?
man, im sorry for crucial to have you as a customer :P

i dont know what chips they are using, but i can guarantee you that some batches from micron are way more sensitive to vdimm than others.
there was a cellshock customer arguing with me about using high vdimm, he wouldnt accept that high vdimm simply kills the chips after some time...
he was using his cellshock kit together with a crucial kit at high vdimm, and wouldnt believe its the fault of the chips that his crucial kit worked at high vdimm for months, while his cellshock kit didnt. well, gmh isnt gmh, and gkx isnt gkx... there are some batches of chips that seem to be able to run with really high vdimm for a long time, while other batches seem to have a relatively high failure rate already at 2.2v...

i talked to an engineer whos in dram for ages, and he told me there used to be bonding problems. basically take a chip that fails and put it in the oven or heat it up with a heatgun, and voila, itll suddenly work fine again.
there are 2 common ways to bond memory or any chips nowadays that i know off.

youd laugh, i actually didnt know this until i started working at cellshock, but memory chips have their own small pcb too
the package has the solder balls on one side to contact with the memory module, and contact pads on the other side to be connected to the actual silicon chip.

everything i wrote below might be entirely wrong, dont take it for granted
im just sharing what i know, or think to know...
if im wrong with anything i write, please let me know

1. the silicon is placed in the center of the package, belly up
so the contact pads are facing upwards. a machine then solders a very thin gold, silver or copper wire to the contact pad of the silicon, or a small drop which is then pulled quickly and forms a wire. the machine then draws the wire to the side of the package where the contact pads of the package are, and connects it there. there are different variations to this, you can solder the wire on both sides, or you solder it on one side and actually just stomp it onto the other pad and rip the wire off., or weld it with ultra sonic sound etc. in the stomp/wedge case the wire isnt actually soldered, but its deformed and pressed against the pad so it makes good contact. i dont know why, but this method of basically stomping the wire to the contact pad seems to be easier/cheaper in some way. it has slightly worse contact though and is prone to come off or have a working but bad quality contact. this happens really really fast, i think we all saw a clip of a chip getting bonded that way one day, looks a bit like a sewing machine

2. the silicon chip has contact pads that can take solder balls directly.
the chip is then soldered to the package directly... in theory you could even solder the silicon chip directly to the module pcb.

this isnt done on a chip by chip basis btw, you have one huge chunk of pcb, the package, where loads of chips get placed on, they all get soldered to the package or bonded. then the whole thing is submerged in the black plastic/ceramique stuff we all know, after it hardened the whole thing is then cut into small pieces, thats when it finally looks like the chip we all know from memory modules

the black stuff is used to protect the silicon, and, to make sure the heat can be transfaired properly.

blue= actual logic inside the silicon chip = heatsource


noooow, the problem that happened back then was that some of the bond, james bond , wires made bad contact or had come off.
well, the black cermaique stuff keeps the bond wires in place, so they cant really come off... but they arent making 100% contact either.

if you look at it its obvious that a bond wire has a smaller contact surface than a solder ball. smaller surface the current flows through... so... electromigration problems... im seriously starting to think that this is the achilles heel of micron chips or memory chips in general.

another thing about the bond wires is, once they are off, they are off.
theres no way to really fix this... solder has a really neat feature, its still kinda liquid, even in solid state. so even if youd have electromigration on it, its a massive chunk of metal that wont degrade that much that quickly, and, it will rebuild its shape to acertain degree.

its like a water drop, if some of it evaporates or is absorbed by the surface its on, itll still keep the same shape more or less. at room temperatures solder flows very slow, but we are talking about electromigration here which happens very slowly too... so i think solder might actually compensate for electromigration to some degree.

meh, i shouldve checked wikipedia before making my paint drawing, ah well
http://en.wikipedia.org/wiki/Flip_chip
http://en.wikipedia.org/wiki/Wire_bonding

disadvantages of direct silicon solder ball contact is that you need to reroute the traces more in the package/carrier... but if you do that i think you can get shorter traces than by using the bond wire technique...
but the problem seems to be cost, bond wires seem to be notably cheaper...

[mad_skills]

sorry if i came off unfriendly, i know you only posted your experience, i just added my experience i used a lot of different chips from different batches and i was surprised as well, that the batches differ so much from each other.

No unfriendliness what so ever just wanted to understand each other properly

your fried 6 sticks by using high vdimm and voiding your warranty and NOW your thinking about what the problem could be?
man, im sorry for crucial to have you as a customer :P

I would be asking myself if i'm doing something wrong...

[naokaji]

your fried 6 sticks by using high vdimm and voiding your warranty and NOW your thinking about what the problem could be?

I didnt see high voltages mentioned in his post, only 2,1V.


The weird thing is just the big difference in volts it takes to kill d9, I mean I had some die within rated volts, but I also have some that survived 3,3V.

The difference between gmh and gkx isnt too big, The voltage difference they survive might be comming from better pcb's being used with the gkx since they often end up in the expensive higher end models, just a thought though and far away from proven.