Intrinsic noise of a MOSFET

[bharath_k]

Hello everyone,

I want a random noise generator. I will feed the random noise to a comparator and then use this signal to trigger a oscillator.

So can I use the intrinsic noise a MOSFET as my noise source? if so how do I simulate this?

Thanks,
Bharath

 

[dick_freebird]

SPICE has some noise analyses but they may or may not be
useful to you.

The qualities of the noise product matter to some folks
more than others. Zener diodes are liked because they
make a pretty good "white" noise. Devices which have
other dominant noise mechanisms may give you some
"clicks, pops" (popcorn noise, shot noise, RTS) that
offend the ear. At the other end you have the kooks and
spooks at the NSA, who insisted their "randomizer" ICs
produce (at test) a very even distribution of the random
number output over time (which is more difficult than
you might imagine - the pursuit of perfect imperfection).

If you want to observe time-domain randomness then
you's probably need two things. One is a simulator with
'tnoise' (transient noise) analysis and the other is a set
of models with the pertinent noise model params in
place w/ sane values. You'd probably have to do some
reading to find out what noise behaviors are and are not
considered "well covered".

Frequency / phase noise domain, is more for getting a
rollup noise quantity than for "looking at" although you
can get some useful design insight from seeing how the
noise voltage increases along the lineup (weak link etc.).

For maximum noise you'd use the smallest gate area MOSFET
(and maybe extend this into any amplifier chain, more
noise there).

A "digital" technology may have poor or no noise model,
a RD or mixed signal CMOS is likelier to have accurate(-ish)
noise params and maybe even some simulation-vs-model
reportage in the PDK docs.

 

[bharath_k]

SPICE has some noise analyses but they may or may not be
useful to you.

The qualities of the noise product matter to some folks
more than others. Zener diodes are liked because they
make a pretty good "white" noise. Devices which have
other dominant noise mechanisms may give you some
"clicks, pops" (popcorn noise, shot noise, RTS) that
offend the ear. At the other end you have the kooks and
spooks at the NSA, who insisted their "randomizer" ICs
produce (at test) a very even distribution of the random
number output over time (which is more difficult than
you might imagine - the pursuit of perfect imperfection).

If you want to observe time-domain randomness then
you's probably need two things. One is a simulator with
'tnoise' (transient noise) analysis and the other is a set
of models with the pertinent noise model params in
place w/ sane values. You'd probably have to do some
reading to find out what noise behaviors are and are not
considered "well covered".

Frequency / phase noise domain, is more for getting a
rollup noise quantity than for "looking at" although you
can get some useful design insight from seeing how the
noise voltage increases along the lineup (weak link etc.).

For maximum noise you'd use the smallest gate area MOSFET
(and maybe extend this into any amplifier chain, more
noise there).

A "digital" technology may have poor or no noise model,
a RD or mixed signal CMOS is likelier to have accurate(-ish)
noise params and maybe even some simulation-vs-model
reportage in the PDK docs.

Thanks a lot for your detailed analysis. After a lot of search I found in latest version of Cadence we can do transient noise analysis. So I would probably try this.