CPU burn in...fact or fiction?

[phil]

Well, this age old topic has reared it's head again. I regularly frequent various overclocking sites and regularly see people recommending a "burn-in" on new CPU's to enable them to run higher overclocks. I can't say if this is true or not as I usually hammer my CPU's from day 1 :o. These guys seem to think it works though. I remember a while back when reading the A.C.H.O newsgroup that somebody posted a link to an article dis-proving the "burn-in" theory but I can't for the life of me find it. Anyway, I found a decent enough guide on how to burn-in (or "condition") the CPU. Check it here (http://people.freenet.de/s.urfer/conditioning.htm). If it works, let us know :baa:.

[MikeTimbers]

Electrically, no it doesn't work. But there are lots of things that shouldn't work but do. Burn-in is most likely explained through heat conditioning the power regulators and settling the thermal compounds.

<off-topic>
Here is another example of what shouldn't work but does:

After multiple letters on the subject, one of the UK hi-fi magazines did a blind (deaf?) sound test of plain audio CDs and CDs that had been frozen and then de-frosted. In nearly every case, the listeners said the frozen ones sounded better. Sony said this was nonsense but people still believe it.

</off-topic>

[Ercewee]

Good question.

Normal integrated circuits already get a burn-in in the factory. After the fabrication there is still some movement (mostly because of stress) in the IC. The burn-in does a rapid aging to make sure that the circuit will behave in the same way during the years of use. This also weeds out specimen that come through the first quality check, but fail after a very short time of use.

However if your device is very critical such a burn-in can destroy your correct working devices and your yield drops tremendously.

Unfortunatly I have no idea how a burn in can improve your oc-performance. I'll think about it. :)

edit:

Phil, with respect to the article you gave, I don't believe hot-electron effect can be the cause of beter oc-performance. Hot elektron effects leads to a lowering of your Threshold Voltage in the device. Therefore on a rising edge your transistor can be on slightly earlier then under normal conditions. However since the steepness of your device is still the same, it takes the same time to reach the level which would be a '1' or '0' for the next stage. The speed of your device is independend of the Threshold Voltage.
Hot electron effects only lead to degrading of your noise-margins in your device. Not to a speed improvement.

This is just my first reaction, I haven't checked this in literature, so I might refine my conclusion later. :)

[Nofinger]

Originally posted by MikeTimbers
<off-topic>
Here is another example of what shouldn't work but does:

After multiple letters on the subject, one of the UK hi-fi magazines did a blind (deaf?) sound test of plain audio CDs and CDs that had been frozen and then de-frosted. In nearly every case, the listeners said the frozen ones sounded better. Sony said this was nonsense but people still believe it.</off-topic>

Thank for sharing this info :cool:

All my cd's are now in the freezer :D :rolleyes:

[MikeTimbers]

Let us know if it's true ;)

[ohms18k]

It's something I believe works. At my old job we use to burn in CPU.s for 72 hrs under 50% duty cycle in a heat room. What we found was a failure rate of less than 1% and the CPU's not burned in had a failure rate around 20% we were sending out about 300 controls a week and the heat room was costing time and money so they stopped that process to the disagreement of the engineer. Now we were not overclocking the CPU’s but I can tell you the burn-in process help the stability of the controls shipped. There are some papers written up on the burn-in process and the effects on electronic components at my old job which I don’t have access to right now but I will try to get hold of this info. Why they stopped the burn-in process was this; since the burn in process was not producing failures in the initial process why do it? Well after the burn-in stopped we found out in the long run.

[MikeTimbers]

I don't think that's comparing like with like tho'. When I used to build PCs for a living we burned them in for two days as well but that was to weed out the ones that couldn't cope with full load and heat. The ones which passed are statistically more likely to last than batches which have not been pre-tested - that goes without saying.

But to claim that running over-volted at stock will then magically allow a higher overclock at a later date because it "widens the electrical path" (not quoting anyone, but I swear I've seen this written) seems like BS to me.

[zhotfire]

Hmmm... is it possible there's a tempering of the die going on making it more durable/conductive?

[Ercewee]

I just found a mistake in my post. I state that the Threshold Voltage goes down, that should be up offcouse. :banghead:

zhotfire:

What kind of mechanism are you thinking about for making it more durable/conductive?

We can't change the conductivity of the metals.

We could change the conductivity of the Silicon, this by changing the concentration of dopants, however this would also change the properties of your devices, so that's not a good idea. To change these you would need diffusion, and for that, to have some effect, you need a higher temperature then you reach in a burn-in. :)

Further if we could influence the conductance of the junctions between the interconnect and the devices. These are 'ohmic contacts', usually consisting of a salicide (silicon metal alloy) or some special barrier layers (also alloyed to the silicon). However during the fabrication an alloying step (high temp.) is already done. And the contacts between the devices and the interconnect should be fine (otherwise your ic wouldn't function at all). Secondly, I don't think the temperature reached during a burn-in is high enough for this. There are even some more drawbacks, I won't go into those.

[zhotfire]

Originally posted by Ercewee
.... What kind of mechanism are you thinking about for making it more durable/conductive?
I was thinking about how metals are tempered... but if a high temp is used in fabrication then i can't see any benefit of cooking at lower temps. Of course, i don't know how the mechanics of tempering work. Does the metal become more dense, do the atomic components absorb the energy(overclocked ;))? I really should look that up.... :rolleyes: I was thinking tempering would strengthen the atomic bonds? I don't know where i get these ideas.... :)

[Ercewee]

Originally posted by zhotfire

I was thinking about how metals are tempered... but if a high temp is used in fabrication then i can't see any benefit of cooking at lower temps. Of course, i don't know how the mechanics of tempering work. Does the metal become more dense, do the atomic components absorb the energy(overclocked ;))? I really should look that up.... :rolleyes: I was thinking tempering would strengthen the atomic bonds? I don't know where i get these ideas.... :)

Sorry, I wasn't familiar with the term termpering of metals. :)

I asked a material science student, and according to him the crystal structure of the material changes: From body centered cubic to face centered cubic. This has direct influence on the mechanical properties, however he was unsure how this would influence the electrical properties. Secondly according to him, this is primarily of influence in steel, not in Aluminum or Copper used in processors.

[phil]

I think what the article I linked to was trying to say is that heat isn't required (and is in fact a negative effect) for the "burn-in". What was mentioned was low speeds/ mults/ temp and a *high* vcore @ full load. It is this extra vcore that gives the required effect (whatever that may be).

[X-Calibur]

I think it pleases the Rat-God of Overclocking, that and the monthly sacrifice in blood.

[ohms18k]

I didn't read that article phil. but why would running lower temp/mults with higher vcore @ full load be any different than than just running it in heat?