Feynman's ratchet used one sprung pawl on a ratchet wheel.
The spring biases the pawl towards the ratchet wheel so
mechanical pressure on the gentle slope of the ratchet
wheel drives the wheel the wrong way where it can rest
against the sharp or even overhanging slope. If the pawl
is then lifted by Brownian motion and the ratchet wheel
moves a little the wrong way when the pawl is high,
possible 50% of the time, than the wheel will rotate the
wrong way. If the ratchet wheel moves a little the right
way when the pawl is high, possible 50% of the time, then
the pawl will return to a low part of the gentle slope
near the sharp slope.
If there are many pawls on one ratchet wheel than they do
not have to be biased by springs because the probability
is high, and increases exponentially with the number of
pawls, that at least one pawl of a similar position group
will be in position to block counter rotation of the
ratchet wheel. This type of system should behave like a
larger scale mechanically rectified ratchet wheel at
thermal power levels.
I don't think Feynman tried hard enough to break the
Second Law. Fabricating a device that fails with
inadequate design doesn't prove that a better design won't
work.
Classical treatment of Feynman's ratchet:
http://www.eleceng.adelaide.edu.au/Groups/parrondo/ratchet.html
Aloha,
Charlie