Electromigration in CPUs

[Shindyo180]

Hello Everyone, I would like to ask if "High Temperatures" will actually cause Electromigration or will increase the rate of Electromigration in Processors provided that I am not overclocking and increasing the Core Voltage.

 

[mrmez]

From my understanding, the shrinking gate size in particular makes electro-migration more likely.
Excess heat and voltage will speed this up, still, file it under "shit that will never concern you".

 

[Dogsnake]

No Electromigration is voltage dependent. It occurs no matter what but at such a slow rate that electronic equipment last a very long time. The increased voltages used during O.C. do increase the rates but still the life of a O.C. cpu is usually longer than the time before it is upgraded. Now you might consider that the higher voltages do cause increased heat to be produced so the heat is a factor in the migration process but not an accelerant.

 

[Pinhedd]

Hello Everyone, I would like to ask if "High Temperatures" will actually cause Electromigration or will increase the rate of Electromigration in Processors provided that I am not overclocking and increasing the Core Voltage.

Yes it does.

Metallic atoms are more tightly bound to their lattice structures at lower temperatures; it requires less force to tear them free at higher temperature. Furthermore, resistivity increases as temperature increases. Neither of these are particularly relevant at the operating temperatures of most consumer electronics though.

 

[Shindyo180]

How will I know if the effects of "Electromigration" to my CPU is already significant and what are the signs and symptoms to look about when my CPU significantly degrades overtime?

 

[Pinhedd]

How will I know if the effects of "Electromigration" to my CPU is already significant and what are the signs and symptoms to look about when my CPU significantly degrades overtime?

The largest symptom of electromigration is increased resistivity of the metallic interconnects. As resistivity increases, electrons lose more kinetic energy as they pass through the material. The net effect of this is increased switching time of combinational circuits and degraded signal quality. The resolution for both of these effects is usually to crank up the supply voltage, which in turn accelerates electromigraiton. It's a battle that can't be won.

CPUs are shipped from the manufacturer with a VID that is calibrated to provide sufficient overhead supply voltage to ensure stability for many years of regular operation. Overclockers often [unwisely] reduce this overhead to nearly zero in order to reduce heat ouptut at the overclocked configuration. Over time a previously stable overclock will become unstable which requires the voltage to be increased; rinse and repeat.

If electromigration progresses far enough, the metallic interconnects will degrade completely to the point that one or more become open circuits (infinite resistance) at which point the electronic device becomes non-functional.

An extreme overclock can destroy a CPU via electromigration in a matter of days, but a modest overclock will last many years before the effects become noticeable. I have one QX9650 that will no longer operate stably at its factory VID.

@mrmez: What you're thinking of is called hot-carrier injection. It's closely related to electromigration but is a different phenomena.

 

[Shindyo180]

Does a damaged CPU through Electromigration becomes slower, produce any sorts of errors, blue screen of death and sometimes it even fails to boot up?

 

[Pinhedd]

Does a damaged CPU through Electromigration becomes slower, produce any sorts of errors, blue screen of death and sometimes it even fails to boot up?

It won't operate any slower, but it can crash, freeze, etc...

 

[Shindyo180]

The only I know before asking this question is that Electromigration only becomes a major concern when overclocking a CPU or GPU especially when increasing the Core Voltage and Current. I did'nt know that it may or may not happen to a CPU or GPU running at Default Clock Speed with Default Core Voltage and Current applied.

 

[Dogsnake]

My understanding is that it happens continuously but since we are talking molecular/atomic level here, the rate of change is so slow that is is just not important in real life. Even with an O.C.'d cpu, you would only see noticeable damage in in an extreme overclock that persisted for a long duration. The migration effect is just not an issue in the every day use of performance systems. So ya you could say that with a maxed out system stable in every day use with either water or air cooling the life span before unusable levels could be shortened by a year or two. But the life of the system in any case would exceed the users keeping it. We got an O.C.' P4 still running the same as it did almost 10 yrs ago. The thing about heat is that yes as temps of a device go up resistance usually does also. Higher resistance requires higher voltage and that gives more heat raising the resistance...rinse and repeat. Migration is a fun theoretical consideration to talk about but it just is not a major concern in performance systems for every day use. It is not a major concern and is you think it is I think you are chasing your tail for nothing.

 

[Digitrax]

It seems that electromigration has become a big buzzword for overclockers in the last 5 years, sparked in part by Intel's study on the increase of its effects due to die-shrinkage. In point of fact, though, TDDB ("Time-Dependent Dielectric Breakdown"), and (to a less extent) HCI ("Hot Carrier Injection) have more significant effects on the aging of MOS circuits, particularly as you scale <30nm.

see here: http://ieee-cas.org/wp-content/uploads/2011/03/DLP_LectureSummaryGilsonWirth.pdf

That having been said, any "enthusiast" likely to endanger their CPU to the point of conspicuously shortening its life — (we're talking 15%+ overvolting here) — is also likely to replace that CPU before the problem can reach a point where it manifests itself in any real way. i.e. We're looking at perhaps a 7 year vs. 8 or 9 year CPU lifespan difference, not the difference between 6 years and 2.