Building a discrete solution may be a viable option as individual transistors/devices can be characterized with this testing (and one can now know where failure occurs) and then used within a limited operating range.
Thanks, and from your kind post, i take it that the "jury is out" as to whether or not chips are more sensitive to radiation than discrete semiconductors.....it would need expensive equipment to ascertain and so we wont know.
You're reading a lot more into what I wrote than you should. There is a lot of information out there on radiation effects and how any semiconductor handles it, the "jury is out" is not the case. Any discrete device will be affected by radiation in a fashion similar to a full chip. It does take expensive equipment to ascertain radiation effects, but the difference is that once you have it figured out, a box full of discrete devices can be reconfigured in a lot of ways that a full chip cannot.
But from what you say, its cheaper to certify an all discrete design for radiation-environment use, (compared to a chip based design) and so there is going to be a definite market for all-discrete designs.
It is not necessarily cheaper to do, no. You still need to do the same testing either way. The difference is that once the testing is done and you know how things respond, you can build the discrete devices into something different and know how that circuit will respond. The same discrete devices can be used to make an amplifier, a regulator, or any of a number of circuits without needing testing as you know how the individual devices respond. The cost difference is that buying a radiation hardened chip has had the manufacturer do all the design and testing for you and you get all the data supporting that when you buy the parts. Buying commercial discrete devices may be cheaper, but you must now do all the testing and make models and then do the design yourself. How much does that cost?
"intuition" does tend to tell me that a SOT23 NPN is going to be more resilient to radiation damage than a tiny tiny NPN on a chip......there's more volume that needs destroying on the SOT23, so its bound to be more resilient to radiation(?). Like the difference between trying to smash a wall down with a hammer is going to take time.....trying to totally smash a bit of brick to dust is going to be quicker.
Spend time learning about radiation effects, please. Larger devices are not necessarily going to be more resilient. The larger volumes can actually work against you as particle strikes inject charge into a junction based on how far it goes through that junction. A 15 micron thick junction will get three times the charge as a 5 micron thick one does. Leakages may actually be worse in the larger device as there is a larger area between junctions where damage can occur. Remember, this isn't just about particle strikes, but ionizing radiation as well and a small NPN on a chip can have less than 1% of the junction area of a discrete device that can leak. That same size difference also comes into play with particle strikes as the rate at which you can expect a particle to strike the larger device as opposed to the smaller device is directly related to the size. When you get a charged particle every X number of days through a 1 sq. cm area, the larger device gets hit more often than the smaller one.
Also, ive worked on down-hole designs which were pretty much all discrete (apart from the famous UC2843 which everyone knows about) due to those discretes already having been certified.
I've worked with people doing down-hole designs as well. The biggest challenge there is temperature effects as the environment can run above 150C. The market for down-hole devices is so small that most chip manufacturers don't want to go after it. They're more than happy to rate devices to a 125C maximum junction temperature for commercial/industrial stuff and have been going to 150C for automotive stuff going under the hood. These markets are millions of dollars in size. The few times that I have seen chip manufacturers selling into down-hole designs is when they found that a circuit can withstand 175-185C and still function properly (a happy accident in their opinion).
And of course, bipolar stuff is always more resilient to anything than CMOS, because CMOS has gate junctions which are thin and delicate...this is why NPN's are more resilient than MOSFETs. And the UC2843 is one of the only power supply chips that is all bipolar, so it is of course, very resilient....(and very cheap). The introduction of the UCC28C43, (pin for pin with UC2843 but made of CMOS instead) which has far less bias current,, has in no way obseleted the UC2843...for the very reason i discuss here.
If a proton hits a FET junction, then that junction gets smashed up bad.....the smaller the junction, the worse it gets smashed up.(this is why chips are less resilient than discrete, which tend to be larger) ...if a proton hits a PN junction inside a BJT, the damage is nowhere near as bad.
Again, take the time to learn radiation effects. Bipolar devices are worse for some radiation effects compared to CMOS. This is why the radiation sensitive community has now switched to doing low dose rate testing vs. high dose rates for ionizing radiation. High ionizing radiation dose rates can get you a result in a short period of time, but don't reflect how dosing actually occurs in the environment. Any MOS device isn't really more fragile, it's different effects. A charged particle going through a gate oxide can create an ionized track that can allow current to flow if there is enough electric field across the oxide (usually seen with high drain voltages). That allows for the discharge of the capacitances through the channel; not a problem for a small device on a chip since it's a tiny capacitance, but a large MOS device can have enough capacitance to create high currents and destroy the oxide (single event gate rupture or SEGR). Bipolar vs. MOS devices change in different ways with ionizing radiation; bipolars tend to have beta drop, while MOS devices have a threshold shift from electrons getting knocked out of the gate oxide. N-channel devices have their gate thresholds drop and P-channel devices see an increase.
Having a particle strike isn't just 'smashing' junctions. Go back to look at the size of protons relative to the size of an atom--there's a lot of empty space in there. What charged particle strikes tend to do is create a track of electron-hole pairs in a material as their charge attracts and repels protons and electrons and any electric field causes those to drift in different directions. Hence you get a net electrical charge injected into the junction. Neutrons have no charge so they only cause an issue when they collide with electrons or the nucleus of an atom and that is rare given the space in an atom.
A discrete design can give options that a chip doesn't. Need everything to run low voltage except for one transistor? You don't need a high voltage process for everything on chip, you just buy the one device to be rated appropriately. However, when you look at it from a reliability standpoint, a discrete design is much worse overall because you have so many possible failure points as opposed to just the one chip. On top of all this, don't forget the fact that many chip functions aren't built in a radiation hardened solution. The number of radiation hardened products out there is tiny compared to the commercial and industrial spaces, and the number of companies and engineers actually designing specifically for that market is tiny (I might estimate a few hundred worldwide). So you may be stuck doing a discrete design in that environment because there might not be an existing solution.