Having purchased the September edition of La Recherche because of its (disappointing!) coverage on black matter, I came by a short coverage on an Intel circuit producing “truly random” numbers… I already discussed this issue in an earlier post, namely that there is no reason physical generators are “more” random than congruential pseudo-random generators, but this short paper repeats the same misunderstanding on the role of “random” generators. The paper mentions dangers of pseudo-random generators for cryptography (but this is only when you know the deterministic function and the sequence of seeds used so far), while it misses the essential aspect of valid generators, namely that their distribution is exactly known (e.g., uniform) and, in the case of parallel generations, which seems to be the case for this circuit, that the generators are completely independent. La Recherche mentions that the entropy of the generator is really high, but this is more worrying than reassuring, as the Intel engineers do not have a more elaborate way to prove uniformity than a Monte Carlo experiment…

There is actually a deeper entry found on Technology Review. (Which may have been the source for the paper in the technology tribune of La Recherche.) The article mentions that the generator satisfied all benchmarks of “randomness” maintained by NIST. Those statistical tests sound much more reassuring than the entropy check mentioned by La Recherche, as they essentially reproduce Marsaglia’s DieHard benchmark… I remain rather skeptical about physical devices, as compared with mathematical functions, because of (a) non-reproducibility which is a negative feature despite what the paper says and of (b) instability of the device, which means that proven uniformity at time t does not induce uniformity at time t+1. Nonetheless, if the gains in execution are gigantic, it may be worth the approximation for most applications. But please stop using “true” in conjunction with randomness!!!