CPU death by electromigration

Electromigration was designed for 10 years endurance at intel. But "hot electron injection" from PMOS devices causes the MOSFET threshold voltages to change over years. That may be our problem. Electrons get stuck in oxide gate insulators.

Maybe we need to use proper lubricants when we consensually compute? P-Lube keeps your MOSFETs working like new! Now available as N-Lube so your NMOS devices don't feel left out.

What functions do you see disappearing?

My Packard Bell Win95 still boots up in 15 seconds and had an integrated Phone answering message system with a good modem, until I had to ship it to the "happy hunting ground" when I sold my other house. Almost 20 yrs a for a disk drive .

My current ASUS i7-4770K tower is 7 yrs old and won't need any more updates for another 7 yrs on Win7 x64.

Have you tried reviving it with supplements and antioxidants (lol)?

Exposing an IC with ultraviolet light will erase any hot electrons that got stuck under the gate and above the MOSFET channel. It is like erasing a UV EPROM chip through its window. Your CPU can be healed by UV light, like a cleansing. Electromigration can be accelerated by heat. Test it, report ack.

Electromigration is strongly dependent on temperature. If you run a chip at a lower temperature than the manufacturer used in electromigration calculations, then you will get a longer electromigration life.

Whether the manufacturer correctly simulated the current density present in all of the traces is a different matter. I am not aware of automated tools that can do this (other than for DC circuitry, which a CPU certainly isn't), though I have not dome any chip design for quite a few years.

There are other failure mechanisms, and I am not convinced that you would have enough information to identify the mechanism that you are seeing.

Synopsys Corporation offer CAD tools for electromigration evaluation of IC designs.

Underclock them. Heat production is directly connected to clock frequency. Every 6 degree celcius raise in temp halves the livetime of a silicon device. So lower clock and proper cooling will extend the lifetime. Most new processor chips are balancing on the edge of destruction, because highspeed advertizing sells.

At a lower clock frequency, many processors can run at a lower operating voltage. Since power dissipation is typically dependent of voltage squared, so a 10% drop n voltage will drop power dissipation by nearly 20 %.

When underclocking, it is important that tasks have carefully selected priorities, so that lower priority tasks run longer, without affecting high priority tasks too much.

Of course when the CPU load remains constantly at 100 % (the NULL task is seldomly executing), the power dissipation can not be reduced further by simply further reducing clock frequency or voltage.

snipped-for-privacy@gmail.com wrote in news: snipped-for-privacy@googlegroups.com:

General Instrument had Motorola build them a stereo audio chip for their set top satellite receiver and they did so, but forgot a mask layer inside the chip and it went un-noticed in test.

Anywhere from 30 to 800 hours after starting to utilize the chips in circuit, the audio channel output would exhibit noise. It seems the mask layer absence allowed for a slow build up of 'free' electrons within the chip which would then start injecting themselves into the audio. An Analog version.

So, GI had already integrated over a quarter million chips into receivers made both in Calif. and Puerto Rico. Some had actually already made it into customer installs. And had many more on hand which had to be trashed.

We did a massive re-work on thousands of receivers. The final price tag after not getting paid on the chip contract to start with, was over $10 million.

Motorola's answer was to simply buy the Uplink Encoder Division of General Instrument, if not the whole corp. I am not sure there.

All I know is that Motorola now owns GI and I am certain that chip flub was the trigger.

And you are sure it's not malware (or 'updates')

A CPU-year is not the same as a human year. Sort of like dogs.

In my experience, desktop PC's die a different death. First the covers come off while you tinker with some BS inside (video card, memory, fan, adding a new drive, etc..) and then never get replaced or you lose all the screws.

Then the power supply dies - but luckily the case is already open, so all y ou have to do is fiddle with the connectors on the replacement power supply not matching anything else in the build. You might even have to do this t wice as your original diagnosis of a bad power supply wasn't correct in the first place.

And by that time, none of your programs or browsing experiences work anymor e because the world has passed you by (in the pursuit of "better is the ene my of good enough"). Which is hardly ever the case, BTW - except maybe for USB.

You then buy a new PC, transfer or archive all of your files to a place you will never again think to look for them, and then add the original PC's ca rcass to the collection out in the garage.

So, by way of roundabout, the PC life is 7 year, but you only clean the gar age out every 20.

Hope that helps. :)

Where did you get a CPU lasts 7 years from?

Are you another Skybuck?

That doesn't answer the question. Try again.

Yes, first swap out all the supply-buss electrolytics.

I have an Intel Plato board with a Pentium 90 that was exchanged when the famous fp bug was found. It worked until quite recent and I stopped using it because of hard disk failure, not something with the board. That certainly is more than 20 years.

Groetjes Albert

I expect it depends on how hard it is worked and how hot the chip die gets in regular use. Some CPUs run *much* hotter than others.

My instinct would be that the faster Pentium III's would be vulnerable since some of them ran insanely hot in laptops. Hot enough to cause injury to the user if they were daft enough to use it on their lap.

Mine actually damaged the finish on a varnished table.

Since that era TDP has decreased as finer feature resolution and lower voltages became possible and heatsinks have also improved quite a bit.

in

CPU.

and the system much smarter preventing the cpu from getting as hot by throttling the clock speeds and adjusting the voltages