AD_ward9

The existence of a 5V rail is due purely to the ADCs: we could not find decent Sigma Delta ADCs with enough inputs running off 3.3V. We use only Sigma Delta for this as they are perfectly linear by design, and have 24 bit resolution. The fact that the manufacturer does not have to step through ever single voltage on each input to test them also makes them cheaper than any other type of ADC - the silicon area is among the smallest too. The 3.3V Sigma Delta ADCs that are available have "issues" when tested, but sooner or later they will improve, and then the 5V rail will vanish.
Out of the thousands of op-amps available, for each application there is usually just one that is head and shoulders above all others. For example, for extremely low noise AC signals one particular Linear Technology op-amp takes the prize, but for near DC signals the AD8628 is par excellence, bar none. In a rebreather most signals are near DC, so voltage reference buffers and the CO2 sensor needs an op-amp that does not drift - the domain of the chopper amplifier. The temperature drift of 20nV/C worst case for the AD8628 is something no other op-amp comes close to. In fact, there was only a valve amp module in the 1950s that beat it: the AD8628 is otherwise an all time best for low noise performance with near DC signals. The AD8628 is not actually a chopper design, but is a chopper replacement, outperforming most choppers 10:1 on the relevant parameters.
We like old friends! We use the LM61 for the scrubber stick and the plug compatible LM35 for monitoring the temperature of the gas in the CO2 and He sensing path. We got there by trying all other avenues first: in the Rev A design, we used 0803 size NPT resistors, but during testing they were simply not rugged enough and meant fitting a fiddley heat sink, so we replaced them by digital sensors with an I2C interface - those failed the safety assessment because of their daisy chain and excess wiring, so Rev B build uses the LM35/61, as it tolerates multiple sensors failing and as their output goes via two MUXes into high precision ADCs which are there anyway, there is no advantage from digital output sensors. We oversample the sensors heavily, and noise is not a problem.
Maxim are not a favoured supplier: lead times and MOQs, as well as the performance of their parts simply are not good enough to replace the AD or Linear Tech parts we are using. Maxim's muxes are about the only thing that makes the relative pain of dealing with them worthwhile. For a low volume build, you are right that one supplier is preferable, but for higher volume build multiple suppliers are not a problem. We use all these suppliers for other projects, and the same manufacturer is involved, so it is not a problem to select the best part and treat the supply logistics as a very secondary item.
Actually, the opposite. We use SG-550CFs. We tested every component for helium susceptibility (pure He at 1400msw), and the results of that caused us to change clock components. We do not use a solenoid: solenoids have two failure states, neither being fail safe. The design uses a variable orifice valve, or rather two, so that if all the electronics dies (say, you ran a spike through the whole lot), then it still keeps you alive in the same manner as a KISS.

The power supplies are well regulated, and except for some cross talk between the injector motor and the Hall position sensor on the orifice, we do not see any correlation between noise on any signal and motor actuation. I will post some pictures later.
Commercial secrets Well, we have to have some ..

Thanks for your comments,

Alex