I am aware of that.

However, the question then is, whether you (or they) know, how "good" that randomness is, i.e. what is the percentage of the "unbiased" SRAM cells in your particular array, how well "unbiased" they are, and what are all the potential factors influencing this.

Also, too many applications in this field play on the appearance of randomness. If you take say any well-known poem, say a thousand of letters (which in itself are undoubtedly not very well random), subject it some sort of encryption, the result may be a thousand of bytes appearing completely random even when subjected to "industry grade" tests of randomness (they would probably fail tests invented by cryptographers like Marsaglia's "Diehard" suite, but it again takes a great deal of understanding to be able to properly interpret the results of those, which makes them practially unusable for various regulatory and testing agencies). Yet the true randomness in them is maybe half an order of magnitude less (given to limited character set and limited ways of grouping of letters); and if the "opponent" knows your "method" (i.e. that you've chosen a literary work as the "seed"), then even much less.

The key word here is "understanding", which cannot be simplified to a mere application of a set of formulae.

JW