At least at one point in the past Tennessee's RNG definitely picked 3 and/or 4 individual digits. Maybe they make that information available somewhere, but even if they wanted to keep it a secret we know that's how they were doing it because for a short period the RNG was configured incorrectly and would not output a digit that had already been used for a specific drawing. Whatever digit came up as the first one couldn't come up as the 2nd one, and neither of the first two could come up in the 3rd position. I'm pretty sure it was just a P3 issue, which would suggest that they used separate RNGs for the two games, but I'm not positive. Of course the nature of a computerized RNG is such that you could easily use the same one for multiple games, and have the configuration set incorrectly for just one of the games. Human nature also means that it's easily possible to incorrectly configure all of any multiple RNGs used for a particular game, but just as there are multiple ball sets and machines for ball-drawn games there are also good reasons that there should be multiple RNGs for each game.
I don't see any reason that a computerized RNG couldn't also be set to output a 3 digit number, or impose all sorts of restrictions on what numbers are permissible (say simulating rolling multiple dice, for instance). Based on the TN mishap there would be a lot of logic in having a 3 digit number as the output instead of 3 individual digits, but I suspect that the underlying process might still be outputting individual digits that are simply strung together. IOW, instead of generating a number between 0 and 999 the RNG would generate 3 digits and then output them as a single number (with leading zeros as necessary). I think that would probably make it impossible to exclude digits just because they had come up in a previous position, though there might still be the potential to exclude a digit completely.
One thing I'd bet on is that if they do use the same RNG for both P3 and P4 they would treat them as completely independent drawings, so that the RNG would output B1, B2, and B3 and then B1, B2, B3, and B4. Although it shouldn't affect the results it would be so incredibly stupid to have it output a series of 7 digits as B1, B2, B3, B4, B5, B6, and B7 that I can't see it being done that way.
