e91.waggin@gmail.com
Date: Wed, 22 Feb 2017 10:23:38 -0500
In Risks 30.15, Martin Ward wrote:
> A chip less than four years old is basically still in "alpha test"
That's not quite a fair characterization of this particular bug. I work for one of the companies significantly hit by this issue (*not* Intel), and I have many years background in hardware design so I've been messaging it to a lot of people in and around here lately.
The issue is a slight degradation of a small but critical circuit inside the chip that over a time measured in years will age a bit faster than expected. The years it takes the issue to even surface, coupled with the very small reduction in MTBF means it is not at all surprising that it took this long to find a couple of gates/wires that may not have been engineered quite as well as they should have been.
To characterize this as Alpha Test is not fair at all. All chips have problems throughout their life. Some are invisible, some are not. Some take a long time to discover, some surface very quickly. Sadly, the kinds of boards that use this chip are in very visible places thanks to them running the Internet. And that Internet itself has published this result far and wide. Irony at work.
That said, this is one of those risks of small embedded things out there that have latent issues and little ability to patch or service. In this case, there is no software remediation to patch, it requires a hardware fix. Major vendors like us will be repairing and upgrading boards. But how many small $100 appliances out there will just stop one day and be tossed in the trash?