From: attila <attila@primenet.com>
Date: Thu, 18 Jan 1996 05:30:37 +0800
To: "Timothy L. Nali" <tn0s+@andrew.cmu.edu>
Subject: Re: Random Number Generators
In-Reply-To: <0kzHl6200bky0_dkQ0@andrew.cmu.edu>
Message-ID: <Pine.BSD.3.91.960117202900.3807G-100000@usr1.primenet.com>
MIME-Version: 1.0
Content-Type: text/plain


On Wed, 17 Jan 1996, Timothy L. Nali wrote:

> Hi all
> 
> 	For a class project, I will be designing a VLSI cmos chip to generate
> truly random numbers (The chip will be fabricated).  I'm limited to a
> 2-micron standard cmos technology (no fancy BiCMOS, MISC, or anthing
> else).  The most promising design I've seen so far (that I can actually
> do) is based on clocking a D flip-flop in the following way:
> 
> 
> 			         	-----
> 	8Khz clock ------	| 	|----- Random output
> 				|	|
> 				|	|
> (sloppy) slow clock	----  |>	|
> 				|	|
> 				-----
> 

	you can enhance the thermal stability with a temperature control
    scheme.

	and you can virtually eliminate voltage problems with separate
    regulation.

	use a digital oscillator and clean up the edges.

	would you not be better off for true randomness to use a) a
    > 8Mz clock, and b) to chain the output of one into the control gate
    of a second?  I think that gives you better spectral distribution
    presuming you use a second clock frequency.
 
	unlike most, I am still of the opinion that digital means of
    generating this should be more uniform. otherwise, use a 
    high-frequency diode, analog weight the curve, analog high-pass it,
    sample it, and go for it.  or take several TV stations, phase and mix
    the horizontal scans, etc.  --but I thought this was a digital project
    for CMOS  --actually CMOS can generate white noise, but you probably 
    will end up with a DSP on your chip! 

	biasing should be controllable with edge control.

	However, all of above needs to be bench tested for the practical
    results --keeping in mind measuring randomness of segments of a bit
    stream are "impossible"  --thoroughly frustating. 

	another schema is to play the old enigma game of lining up the
    spinning wheels -that works digitally, the gates on CMOS are not
    too hairy  -the question is how many wheels and their relative
    rotation (including direction)?  and, how many levels?

	how much real estate do you have at 2u? I ask because the
    use of the rotating wheels has been an old project I dumped since
    fab was far to expensive in the 70s --but it has held my interest.

	there has also been a thorough trashing or thrashing of RNG
    recently which should be in the archives.
 
> I'd also appreciate any other suggestions or advice you have on RNGs.
> 
> Thanks in advance.
> 

__________________________________________________________________________
    go not unto usenet for advice, for the inhabitants thereof will say:
      yes, and no, and maybe, and I don't know, and fuck-off.
_________________________________________________________________ attila__

    To be a ruler of men, you need at least 12 inches....