Very often the voice recorders from plane crashes contain important clues (aside from what the pilots said). You can hear audio alerts from the aircraft, switches being flicked, lightning strikes, all sorts of things.

And in this case, we want to know what the heck was going on with the pilots. Did they go crazy, just go unconscious from drinking too much, was there a fire...

Imagine trying to update the system to use a different technology. Then imagine how amazingly bad it would be if it didn't work as well as the old system. Or if it lost data when it shouldn't have. And so on. It's the classic system rewrite conundrum. You can rewrite a system at fairly low cost, but in order to bring that system up to feature and reliability parity you need to put in an order of magnitude more effort. And in this case you don't really want to deal with "early adopter risk" so you need to dump all that effort (read: money) in ahead of time.

Except that if you want it to be a useful back box, it needs to survive Events. Like being blown up in midair. Or roasting in a jet fuel fire for half an hour. Or, random example, sitting at the bottom of the ocean for several years.

That multiplies the engineering challenge, which in turn reduces the amount of storage you can put into the thing.