4
\$\begingroup\$

I've been reading press articles and about electromigration and how it is associated with miniaturization. I have seen old computers working for decades, and I'm wondering how bad the electromigration may be nowadays. I wonder if the miniaturization is shortening the lifespan of electronics and to what extent this is happening.

Personally I have replaced most of my computers because they broke, not because they were too old to run what I needed. I wonder if this is going to become the norm, and we should not consider technical characteristics looking ahead in time, because anyway the computer will break before it's too slow.

\$\endgroup\$
  • 3
    \$\begingroup\$ Frankly, 90-95% of all electronic devices breaking are a result of either ESD events or electrolytic capacitors reaching their end of life. I would not be afraid of electromigration. \$\endgroup\$ – Laszlo Valko Nov 14 '13 at 14:04
  • 2
    \$\begingroup\$ Heat-related or mechanical issues are also common; especially to do with soldering BGA packages (e.g. xbox RROD) \$\endgroup\$ – pjc50 Nov 14 '13 at 14:10
  • 1
    \$\begingroup\$ @Trylks: Yes, you should put on your tinfoil hat and go back to using a abacus. Otherwise the military-industrial-Mircosoft-Intel-Google complex will suck your brain dry and take your soul. (Of course if you nothing to loose on either account, then go ahead and use a modern PC). \$\endgroup\$ – Olin Lathrop Nov 14 '13 at 15:04
  • 1
    \$\begingroup\$ This brings to mind the 2009 Wild And Crazy Ideas "BubbleWrap: Popping CMP Cores for Sequential Acceleration" where less power efficient cores would be overvolted to provide greater single-thread performance but reducing core life to, e.g., three months. \$\endgroup\$ – Paul A. Clayton Nov 14 '13 at 15:13
  • 1
    \$\begingroup\$ @Trylks The overvolted cores are disposable--3 months run time by the way--, but a number of "precious" cores are never overvolted and have standard 10 year durability. This is a response to dark silicon, where only a fraction of the chip can be active at one time and meet power constraints. It is also WACI. \$\endgroup\$ – Paul A. Clayton Nov 14 '13 at 18:34
5
\$\begingroup\$

I was hoping for a better response to this question. However somebody cancelled my upvote and some of the comments are verging on offensive; certainly less than helpful.

So my thanks to Vasily for starting to get the answers rolling.

I hesitate to answer because it's really not my area. I can recall conversations almost 30 years ago with an engineer from a III/V fab, where electromigration of gold at bonding pads was a concern, (and observable under microscope) but since the engineer was my brother, I can't help with references.

(I can also recall skepticism in other quarters, amid comparison with "the Atomic Theory" espoused by Flann o'Brien in "The Third Policeman", a most excellent tome)

Vasiliy's answer seems to cover the issue : it being a problem, due attention has mitigated the consequences so far.

Today's Deepchip mailing (scroll down to "AGING / RELIABILITY") mentions a paper on modelling aging in semiconductors; it appears that - whether electromigration or other - causes of parameter drift and end-of-life are of concern to process engineers. From the summary, it is unclear whether the writers at TSMC wear tinfoil hats.

\$\endgroup\$
5
\$\begingroup\$

One of the (basic) equations which models electromigration is Black's equation. It states that Mean Time to Failure reduces with cross-section of a wire and has strong dependence on temperature.

There are few factors which cause an increase in rates of electromigration (EM) when transistors become smaller and are placed more densely. The most significant are:

  • The width of the metal interconnects decrease exponentially
  • The total length of metal interconnects grows exponentially

The first factor leads to higher current densities (since the total current reduce over time, but not exponentially), and the second leads to an increase in space where a failure can occur (along the wires). Also both factors contribute to higher resistivity, thus increasing the amount of heat produced.

It seems like you might be right and EM failures are more likely today than they were 10 years ago. However, do not forget that design and manufacturing technologies are constantly evolving:

  • Today's ICs are checked with special tools for EM issues at layout level (some tools even use predictive models to find EM issues after synthesis)
  • Better materials are used (for example: copper instead of poly-silicon)
  • Manufacturing processes are more EM aware
  • Testing techniques and extrapolation models (for long time predictions) are better
  • Etc.

In fact, in light of the "power reduction trend" which emerged after smartphones and tablets were introduced, I think EM reliability today can be even better than before. Anyway, no manufacturer will sell a device which is known to fail for EM issues before other HW fails (see comments to your question). Not unless this manufacturer wants to stay on market in the future.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.