Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email firstname.lastname@example.org
From: coderman (codermangmail.com)
Date: Sun Dec 26 2010 - 00:45:22 CST
On Sat, Dec 25, 2010 at 2:43 PM, Dan Kaminsky <dandoxpara.com> wrote:
> Making noisy diodes isn't all that hard, AFAIK. You eliminate bias by only returning difference bits -- 01 is a 0, 10 is a 0. Whether the underlying silicon is in fact doing that...well, that's a question for the chip reversers.
noisy diodes, free spinning oscillators, ring oscillators, sub
samplings of above in complex structures, lots of options without lava
lamp or spinning platter craziness. 
as for eliminating bias, the von Neumann whitener as you describe
works well, but has unpredictable throughput. (that is, one
word/buffer may take longer to fill than the next depending on
generated bits, and at best you've got a significant reduction in
this is one reason it is preferable to read raw biased entropy at
maximum rate from the hardware source into an entropy daemon which
then validates hardware output before whitening, compressing, and/or
digesting read bits.
best regards, happy holidays, done beating this dead horse for now...
0. there's a nice survey/list in chapter 4 of Cryptographic Engineering.
Full-Disclosure - We believe in it.
Hosted and sponsored by Secunia - http://secunia.com/