I meant the node itself was state of the art. The application reliability is important, but in practice will be balanced against cost of goods -- just like how consumer PCs use non-ECC memory and try to handle it in software (or just decide occasional crashes are acceptable), there would likely be additional failure compensation added in other areas to adjust for otherwise cheaper parts on a smaller process node.

And if you think a car company wouldn't use a less-reliable part because of safety concerns, look at the Pinto: Ford knew it had a tendency to, erm, explode, but calculated paying off lawsuits would be cheaper than fixing the problem. [1]

[1]: https://en.wikipedia.org/wiki/Ford_Pinto#Cost%E2%80%93benefi...