I only just read this question and the comments. I understand that radiation data on the TPSA7A45xx series is a determining factor. I've also found at this link the following comment about making a Spice model:
We would possibly make one in the future. Issue is as TPS7A4501-SP is
targeted for space customers, Pspice model for TPS7A4501-SP needs to
be accurate - thus the reason for delay.
So it appears TI has definitely got you in their sights.
I'm not familiar with the TPSA7A45xx series and I've also not seen a schematic quite like the one they suggest. But from a glanced high level view I think the TPS7A4533 must be a fixed \$3.3 \:\text{V}\$ regulator and has its sense pin tied to the output to guarantee that voltage, plus a TPS7A4501 that allows the use of a voltage divider to set its output, plus an integrating opamp used to integrate the current (turned to a voltage via a small resistance) so that, on average, the two regulators share the load, equally. If I got that right, the whole idea of it makes sense and I kind of like it. More to their point, though, is that it maintains an LDO reality when looking for more current compliance. I think they did the right thing adding that schematic because it keeps the LDO reality while adding current compliance.
The problem gets way more complicated if you add more, though. A system of multiple integrators all working magically together to keep the currents though all the regulators balanced will be a monster of a project. It's non-trivial. Two is fine. But you immediately starting getting into trouble the moment you move off the reservation, so to speak.
Getting back to your question, though.
You do not specify your input source voltage. These regulators are called LDOs, with all that entails. Looking at the typical chart, these are LDO (PNP-style control for positive-side.)
The first schematic you have shown adds a PNP to an NPN-style voltage regulator. Using the same technique on these will add another \$V_{\small{BE}}\$ of voltage overhead, which in the of the TPSA7A45xx series is a significant add:
Are you willing to accept that?
I usually only consider an LDO in cases where it is actually needed. If my voltage overhead is high enough, I go for linear regulators of the NPN-style, not PNP-style. What's the point of an LDO if doesn't, in fact, matter as the input voltage is several volts higher?? It's kind of silly and it adds complications that may be avoided if you allow the regulator some overhead with which to do its magic.
So there are two possibilities:
- You need an LDO.
- You don't need an LDO.
Which is it?
If you need an LDO then the approach they give you is the right choice. You can't go gallivanting around and playing loose and easy with external bipolars, which will only add significant overhead voltage you can't afford being added.
If you do not need an LDO, but are using these LDOs simply because they qualify due to the CERN radiation data available on them and TI's focus on targeting space customers with them, then that's another thing altogether. In that case, then yes you can use external transistors.
(You will want to use large-feature-size transistors as they are more radiation resistant, almost by definition. Sadly, these take up space on a wafer, reducing the number of parts they get, increasing the cost of each, and therefore they aren't making nearly so many of them anymore. Only for high-current devices are they forced into somewhat larger feature sizes. I hate what this has done for me in creating low-noise, low-current front end designs. They've pretty much stopped making them and there's nothing today to replace what was once available to me. But in your case it may be just fine since higher power translates into larger feature size, by definition.)
But if you do that, I would strongly recommend that you add some added circuitry to foldback the current. To make that point, I will illustrate with a circuit:
This actively curtails the current if the load demands more than the design limit. You can see the foldback action in the curve shown.
You will need at least \$3\:\text{V}\$ of overhead. More will be better. Also, more parts to check for radiation data. Granted. But at least consider the idea.
