analog failures

I've had designs where it was like your case and then my reset pin was introduced by means of a P-channel in the VCC line. However, than meant that the data lines had to be muxed and shut off as well or the chip could be supplied through its diode paths.

Only at the client's insistence. I prefer stuff that is more reliable. In this case, for example, the LM35. Unfortunely that one needs a negative supply if low temperatures have to be measured.

LM35 is one tricky dude. If you pull its output negative more than a couple tenths of a volt (or some spike does it for you) it will latch up. Never power them from more than 5 volts, or it will eventually latch up. It will oscillate if it sees a capacitive load, like a cable.

The datasheet recommend a resistor of such value, that just a few tens of uA are flowing. But that is only needed if low temps have to be monitored.

That's to be expected, opamps don't like that either.

They're talking about safety effects.

Anything that's untested will be unpredictable, for simple function or otherwise. The test criteria and methods differ between analog and digital. So, often to the more familiar fab processes for either.

I'd be more concerned with digital guys doing analog, than the other way around. I expect the former situation greatly outnumbers the reverse. In mixed signal, everything is analog, but they forget that.

What may be bugging them is precisely that an analog error involves a fairly small number of elements, where traditionally there are many more ways to 'fail'. Get used to it.

RL

I have already drawn my conclusions: Microchip is out. It was the same as with a DC/DC manufacturer. After the first heavy fail their sales called and I told him to come back when they have learned how to design. Never heard of him, maybe they have not yet learned it.

I have had quite good experience with Microchip. Opamps, even some LDO voltage regulators which were surprisingly stable (on the client's insistence since I never use LDO).

If that were true whole space programs would have to be scrapped.

I think you'll find that aerospace and zero-g electronics are tested as much as is practicable, prior to deployment. There are many failure avoidance or compensation techniques adopted at the design stage, for such hardware.

Again, in the article, they're talking about safety implications of component failure (not reliability) - approaching it as some kind of tail-end consideration entailed on recent attempts to account for software safety effects (long since due and very tricky) and from a perspective of relative ignorance.

RL

It can only be tested to some extent. Some things have to be "guaranteed by design". This can also work in electronics here on earth, provided that the designers truly know what they are doing.

Most everybody has good parts and buggy parts. I want to start a web site about IC bugs, sort of an Angie's List of semiconductors.

Not trying to be snarky, but most digital engineers work to eliminate noise and to have proper clocks without noise on the transition edge. Schmitt is only needed on the clock inputs, but even there proper design at the board level eliminates the need.

You think that doesn't happen?

And then there are companies that have good ICs but they fall flat on their faces when you want to order ... ... whole reels.

Schmitt is needed everywhere a signal connect to the outside world. A lot of uC design engineers do not understand this.

Define "outside world". Do you mean off board?

Yes, or via a noisy area on the board. Also if slow on-board sensors are involved.

... and then overlook that there can be external influences into boards and system. Such as EMI, magnetic fields or even weird stuff like that inexplicable double-pulse whenever somebody uses the freight elevator. But who am I to complain, this stuff brought me plenty of assignments.

I'm not overlooking anything. All that is part of the engineer's job. A Schmitt trigger would be a panacea for none of the above. A Schmitt trigger is useful when appropriate, but the original post was about "every SPI" having Schmitt inputs. First, its only useful on edges and so only the clock. Secondly it is not something you can't add externally. I don't know why a chip designer wouldn't include a Schmitt trigger on clock inputs, but I'm not going to second guess them on something as unimportant as this. I'd much prefer they got the circuit design right.