Hector Martin

On one side, chips have been audited by highly skilled 3rd party lab and are designed for security On the other side, chips are pwned by design... Auditability is great if it improves the security, if no skilled people audit them and they are already pwned, what's the point?

On one side, firmware has not been audited by anyone competent (evidence: ROCA; it's clear that FIPS certification and such is useless) On the other, you can audit it yourself, and people will for a popular product.

FIPS is not very relevant... Common Criteria certification is. (and ROCA chip was CC certified). CC are not perfect (hence ROCA), but it doesn't mean they are useless... It remains far away more difficult to break a CC chip, than a STM32

The chip yes (probably), the firmware not so much. For the most part all those certification processes just slow things down and make it harder for people to actually implement modern best practices (instead of outdated ones). Silicon moves a lot slower than software.

It's not (probably). There are multiple articles showing that it costs a few dollars and require a few minutes to dump a general purpose MCU. The solution is more to try to open Secure chips, rather than using broken ones...

You have also historically been able to dump some secure MCUs with the same equipment. The problem is that NDAs stop those chips from being more widely audited, so we don't *know* if they really are more secure. In theory they are, but we don't have enough evidence.