Electro-Migration in a Xerox Alto | Hacker News Reader

Electro-Migration in a Xerox Alto

engblg.livingcomputers.org

60 pointscoldnose8 comments

Electromigration is a major cause for failures of silicon components.

Over many years, metal in the very tiny vias between metal layers in the chip fail.

Indeed, this is one the largest challenges in semiconductor fabrication today. The narrower our wires become, the more susceptible they are to tiny voids formed by electromigration.

The article seems to imply that the effect is highly dependent on the "amount" of electricity going through a connection. With that in mind, how much does this occur on the scale of a CPU where the connection is only a few tens of atoms across but the power is measured in only a couple volts?

In microelectronics the voltage is low but the current density can be quite high, and that makes electromigration a more severe problem than in higher voltage macroscopic devices like motors.

Are you familiar with the field or have links to resources I can read? Basically what I want to know is how long will a modern core-i5 last before some transistor bridges and it dies? 10 years? 50?

This "Fundamentals of Electromigration-Aware Integrated Circuit Design" appears to be a decent starting point:

https://www.ifte.de/books/em/

https://www.ifte.de/books/em/em_chap1_figs.pdf

https://www.ifte.de/books/em/em_chap2.pdf

It won't answer your question directly (especially if you just read the free material instead of obtaining the whole book) but it should orient you for further reading. Note that electromigration is sensitive to temperature. Running cooler can extend the lifetime non-linearly, whether achieved via lower duty cycles or more effective cooling methods. You need to know how many years at what temperature the processor will run, not just the dimensions of the conductors.

When I saw the title, I thought a silicon chip in Xerox Alto has stopped functioning due to electromigration. Fortunately it was a macroscopic electromigration on its backplane and has an easy fix. I wonder if it could become a serious issue on computer history preservation in the future. Can a 7nm chip today survive an uptime of 30-50 years? Once it burns out, there would be no replacement... Or is this problem negligible compared to EEPROM losing critical firmware and data if you have proper core voltage and cooling?

Tangential, but is the spelling "buss rail" used in any widespread sense? Buss is affection, bus is shared transport.

Reply on news.ycombinator.com