[Cryptography] New way to produce random bits much faster

[Jeff Harvey]

A long time lurker, I am pleased to have something to contribute to this
discussion as a physicist and as an advocate for open access to research
papers.

I find the quality of Phys.org articles to be routinely sub-par. The Yale
press release does a much better job, but is still mostly just a press
release. Fortunately in physics it has long been a nearly universal
practice to publish on the arxiv preprint server before ending up paywalled
in Science. You can find that article here: https://arxiv.org/abs/2004.07157

My synopsis would be that they have a device that makes a laser beam
version of static on the TV screen, and they've used the geometry of the
space to refine the focusing of the static so that it is less smeared out
or otherwise repetitive in time and for that reason less random. But
they're a ways off from truly having a device they can sell you or
something that is as fast as the headlines might claim.

Figure 2 in the paper shows the improvements in focusing of their speckle
pattern by tuning the shape of the cavity, which you can see pretty clearly
with your eyes without understanding too much of the underlying physics.
The 'quantum' buzzword may get thrown around, along with stimulated
emission. Because the laser is being reflected off of a material, atoms are
absorbing the light and then re-emitting the light (this is what reflection
is), and the exact timing of this is random. The reflecting laser light
interferes and produces a speckle pattern that is constantly shimmering and
changing due to the underlying quantum mechanical processes. This all seems
physically sound.

Figure 3 is giving you a picture of what their speckle pattern data might
look like, and how that changes over time.

Figure 4 is what I think might be of most interest to this group, in that
this is the roundup of the High-speed random bit generation data. Fig. 4A
has a pretty convincing signal to noise ratio, with the background in
black. Their indication of tau, the time resolution looks a little strange
to me, but I would agree that there is a real signal showing a spot getting
bright, then not bright, then brighter again. Fig. 4B is showing that they
don't actually have a symmetric distribution in the graph of the
probability density of the intensity (brightness), but if they instead look
at the differential intensity which looks at samples around that spot too,
then they do in fact have a symmetric graph. Fig. 4E is showing that they
get that bit back to where it was before in about 5 picoseconds, which
would get at a maximum speed limit for how fast this could work. Fig. 4F is
showing you the limitations to physical scale, which seems to suggest that
the channels need to be separated by about 1 micron, maybe a little less. I
find it important to note that these limitations are not being claimed to
have been met when actually recording data or engineering a device.

Finally, in the Discussion and Conclusions section, they get to the details
of what they have done in experiment, which is "parallel RBG in 243
channels with 820 Gb/s per channel." They go on to mention that they
anticipate a technological limitation in the photodetectors being able to
see fast enough to keep up with the actual physical phenomena of speckle
pattern laser show. They have ideas for how to make the most of this, but
that seems like something that would scale the cost. In this paper the
authors state that currently their random bit streams are generated by a
computer through off-line post-processing. Which I take to mean they get
pictures of the speckle pattern and then analyze them much more slowly.

It's nice that this paper also includes a healthy supplemental section with
technical details on their device and its testing. I don't have much
experience with the kind of streak camera that they mention using, nor what
kind of photodetectors might be of use in a consumer device that the
phys.org article is hyping in our imaginations. Other experts may be
interested in the final section, S.7 NIST randomness test where they show
the results of using the NIST SP800-22 Random Bit Generator test suite
(Fig. S13.B). This is not my usual cup of tea :)

Many thanks to the members of the list for the interesting information, and
apologies for any mistakes or misunderstandings.

-JSH
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.metzdowd.com/pipermail/cryptography/attachments/20210228/42e68c60/attachment.htm>