I'm sure I'm overlooking something, but what's the real use-case for true random number generation at that fast of a rate? Even a few Kb/s of random numbers is enough to continually reseed a cryptographic pseudo-random number generator that will generate as much output as you want that's indistinguishable from true randomness. I suppose you aren't reliant on the security of the underlying cryptographic primitives then, but you're still reliant on the particular hardware RNG chip being implemented in a way that's free of bias even if the underlying physics principle is sound.
A PN junction gives you only megabits/s of randomness at most.
This proposed method, if the article is correct, reaches gigabits/s.
But it could be because they are just using a large array.