Micron D9 and high voltage, electromigration effects

[Micutzu]

Gautam's 500Mhz 3-3-3 club got me started on this, that thread is a graveyard for brave memory . There are already some threads on this out there and it's generally a common fact that high voltage helps your D9 RAM clock like mad, but it also kills it sooner or later. I just thought I would share my view on the matter as working in the field plus benching/abusing more than a fair share of memory got me some experience.

As you know, Micron DDR2 chips (aka "D9") are extremely responsive to voltage increase, especially at tight timings ... and alot of people are wondering what's the limit. Well, contrary to the old Winbond's BH-5 that were also very voltage responsive but given proper cooling could run 24/7 even at the top of the voltage range, the Micron D9 are much more decieveing. Before I go on though, let me introduce you to ...

Electromigration
It's a phenomena caused by the high current density in a solid material, that alters physically the cristalin structure of the material due to loss of cinetic energy of the electrons. Errr ... a little more explanation maybe ? :P. The solid metals are networks of bonded atoms that have shared a few electrons each. These electrons can move freely trough the metal's structure and transport, of course, electric charge, that's why the metals are so good electric conductors. Theoretically electrons have so small mass compared to the atom's nucleus that can be neglected, but when trough a very small piece of material you pass an immense quantity of current, the bilions of bilions of electrons forced to move at huge speeds hit the atoms in their way and in time they knock out of the otherwise nicely organised structure some atoms, creating voids in the places they were originally.

This phenomena cannot be avoided and it happens in any material the moment current passes trough it; there is a mathematical model called Black's equation that estimates the average life span of a material under known conditions. Of course, under regular conditions failure due to electromigration is very unlikely, but degradation rapidly accelerates under high current density / increased temperature. As manufacturers move to smaller and smaller manufacturing processes, the importance of electromigration grows larger, as the metallic structures are thinner and thinner and thus more sensitive to this kind of damage. It is true though that there are some technological methods to compensate a little the electromigration's effects by wisely placing the material during chip's construction, but they only delay what will happen anyway (hint - does this help to understand hardware burn-in better ? ).

Looking back at our overvolted Micron D9's, it's pretty easy to understand what is happening. High voltage creates high current, too much current starts damaging the internal chip architecture and soon your expensive overclocking memory will start malfunctioning or even stop working. The "voids" created in the material's structure won't let the current pass trough the area anymore (memory will need higher voltage to work) and will force the same current to pass trough an even narrower band of material, increasing the current density and amplifying the effects of electromigration. More and more voids will be created faster and faster (you will need more and more voltage to make the memory to work at all), until it will stop working alltogether.

Enough with the theoretical babble now, let's put some facts and numbers on the table. From what i've concluded so far, for Micron D9 chips 2.4V is the maximum allowable voltage for 24/7 use, considering decent cooling is provided. I know there are manufacturers that are ceritfying memory to work at more than that but hey, higher RMA rate for a highend low volume/high profit model is not a large price to pay if you want to have that PC9000 something to showoff.
From a bencher's point of view, i'd say 2.6V is OK and still safe, as long as you're not trying to break the record for "highest 24h memtest stable 3-3-3" or something. Short benches or those that don't stress memory too much can do OK with even a little more, even if i wouldn't run more than 2.8V not even for short periods trough a kit of expensive memory that i've paid for. And for you hardcore guys out there, 3.2V+ is the land of no return, the D9 that has been there and came back alive should be kissed every night before you go to sleep.

The first sign of electromigration damage is that the memory would not work anymore with very low voltage such as 1.8-2.0V ... if you notice this on your Micron D9 then you know it's already been affected, and using more voltage will shorten it's lifespan even more. A badly damaged chip can require up to 2.5-2.6V just to work properly, even if it can still work OK for a little more time at very high voltages.

[justin_c]

I have lost 2 sets to 2.3v and 2.4v. I wouldn't even go near 2.3v, more like 2.2v with droop.

[sedg.raven]

my memories dieds at 2,4 and the recomended voltage is 2,35 to 2,45 and have a 12cm 90cfm fan

[Gautam]

This is XS Micutzu. But your point is well noted. FYI, my sticks have done 2.4v 24/7 for a good 7 months or so and are no worse for the wear. When benching I give them 2.8V regularly, and 2.9 when I wanted 500MHz CAS 3. That was a very short period of time.

One thing you can notice about Black's equation is that its inverse exponential with respect to absolute temperature. Keep your semiconductor device cool, and its lifetime dramatically increases. 24/7 I've got a 90mm on my sticks, and when benching usually something pushing at least 110CFM if not close to 200.

Something else that I think is of concern is voltage regulation. Lots of boards these days IMHO regulate vdimm pretty crappily, and this is probably the cause of death at even lower vdimms. The Bad Axe 1 is the most notorious example of this, murdering ram past 2.2v set as it would periodically spike past 3V due to the low current mosfets used.

[ozzimark]

Originally Posted by Gautam

Something else that I think is of concern is voltage regulation.

you got it. i'll be willing to go as far as to say that everyone who has had ram die below 2.4v either:

1. weren't cooling them.
2. the mobo ate the ram.

[Tuppen^^]

Hmm.. creepy.. :P
I think I will go over 2,5/2,6 volts just for short bencing periods after reading this.. :P

[sedg.raven]

my problem was i680 http://www.evga.com/articles/350.asp

[Micutzu]

Gautam > I know it's XS, that doesn't mean one should go pedal to the metal with the voltages just to be Xtreme. IMO this stuff should be as documented as possible, that's the point of the thread, i'd still throw insane volts at some dimms if I consider it's worth it . You've touched two important things there:

1. It is true, poor Vdimm supply regulation can cause instability and can damage the RAM badly. The memory is a very nasty load for a voltage supply because it's consumption jumps up and down very very fast (technically called "transient loads"). These "jumps" happen that fast that a poorly designed/sized regulator remains a little behind and can overvolt temporarily when exiting large loads.

2. Temperature increases the effects of the electromigration, for once because higher temperature = higher molecular agitation / brownian movement, wich adds cinetic energy to the electrons. Heat is the biggest enemy, it's always important to keep things cool when overclocking.

sedg.raven > the only problem of the board is that it overvolts a little the Vdimm, around 2-3% or so; the regulation is decent upto 2.35-2.40V, then it gets worse. That is still OK, most Asus boards for example have way inferior regulation on both CPU and RAM.

[justin_c]

Well 2.3v d9gkx, gskill 8000zx, cooled by ocz xtc cooler (2x90mm).

[Tony]

Guys

Its not all about voltage...DRIVE is also real important, to much will do more damge than voltage ever will.

If you have a board with drive tweaks keep the drive as weak as you can keeping stability....the more drive you apply the more current you push thru the dimms and the more damage you do.

[ozzimark]

Originally Posted by Tony

Its not all about voltage...DRIVE is also real important, to much will do more damge than voltage ever will.

and by an interesting coincidence, lowering drive strength at high voltages on my dfi helps stability... by the time i've over 2.5-2.6v, it is around level 3 to level 7, weak

[ziddey]

Originally Posted by Tony

Guys

Its not all about voltage...DRIVE is also real important, to much will do more damge than voltage ever will.

If you have a board with drive tweaks keep the drive as weak as you can keeping stability....the more drive you apply the more current you push thru the dimms and the more damage you do.

This has to be a major reason. I'm curious about the drive strength in the ds3. It destroyed two very good sticks of gmh at 2.1v in 2 months. At first, they wouldn't post at 1.8v. Then they wouldn't even post 2.1v. So sad.

[Omastar]

I'm starting to think my BadAxe2 killed my first set of Ballistix, at 2.28v set in BIOS. Or it could be the fact that some sets of Ballistix work great at 2.3v/2.4v 24/7 with active cooling or like Gautam mentioned, the motherboard itself may be to blame.

Did they ever correct the voltage spikes to RAM in BadAxe2 as opposed to the original BadAxe?

[STEvil]

Originally Posted by Tony

Guys

Its not all about voltage...DRIVE is also real important, to much will do more damge than voltage ever will.

If you have a board with drive tweaks keep the drive as weak as you can keeping stability....the more drive you apply the more current you push thru the dimms and the more damage you do.

And just to expand on this, this is why some slots seem to make a memory module hotter than others.

[cadaveca]

For those unaware, drivestrength=current=AMPs. That's why it's so important. I've known this stuff for a while, after losing a few sets what was happening was pretty obvious.

Thanks for the ocnfirmation, Tony, I had thought as much, but good to see you of all people acknowledging this.

[IluvIntel]

How on earth do we know what drive strenght I have if say I use the Asus P5B Deluxe board? It does'nt have the drive strength controls from bios like I have in my DFI 975X mobo...

[lowfat]

Originally Posted by justin_c

Well 2.3v d9gkx, gskill 8000zx, cooled by ocz xtc cooler (2x90mm).

they are actually 60mm fans

Originally Posted by Tony

Guys

Its not all about voltage...DRIVE is also real important, to much will do more damge than voltage ever will.

If you have a board with drive tweaks keep the drive as weak as you can keeping stability....the more drive you apply the more current you push thru the dimms and the more damage you do.

wow, nice to know. I always had them maxed with my BH-5 and DFI NF2 board

[cadaveca]

D9's have much higher leakage than BH- or CH-(most was CH, like most D9's are GMH), and I'm pretty sure this is why they behave differently.

[SMa]

intresting read

but what do you guys mean by "drive strength"?


Can electromigration be stopped or slowed down by cooling? (water or subzero)

[cmos.gr]

Originally Posted by IluvIntel

How on earth do we know what drive strenght I have if say I use the Asus P5B Deluxe board? It does'nt have the drive strength controls from bios like I have in my DFI 975X mobo...

and what if you know?...You cant change it so....

[p0tter]

thats why I buy DFI boards.

Originally Posted by IluvIntel

How on earth do we know what drive strenght I have if say I use the Asus P5B Deluxe board? It does'nt have the drive strength controls from bios like I have in my DFI 975X mobo...

[Eldonko]

Originally Posted by ozzimark

you got it. i'll be willing to go as far as to say that everyone who has had ram die below 2.4v either:

1. weren't cooling them.
2. the mobo ate the ram.

My experience is the opposite Mark. 10gb of dead D9 and counting, alot of it never saw above 2.2v. All active cooled with 120mm, 3 different boards.

Good theory/explanation btw Micutzu.

[lowfat]

Originally Posted by SMa

intresting read

but what do you guys mean by "drive strength"?

its an option in the memory section of a few motherboards, mostly DFIs.

[itznfb]

commando must regulate vdimm better than p5b deluxe. i killed 6 sets of D9 on my P5B, with the same benches i've killed none on the commando.

[cadaveca]

Originally Posted by SMa

intresting read

but what do you guys mean by "drive strength"?


Can electromigration be stopped or slowed down by cooling? (water or subzero)

Drive strength is quite literally the amperage provided @ the specified voltage. You can look at it this way...@ 400mhz, that's 400cycles/second.

Let's say you get punched 400 times, by an smurf. Annoying, but it isn't gonna hurt you.

Now let's replace that smurf with GodZilla. Let him hit ya a couple hundred times. Nothing left of ya but mush, eh?