i want my name is matt dispenser inaccurate a today to talk to about replacing
transistors with mechanical switches
upon hearing this
you might rightly think why on earth would you do that
i'm going to reply by taking a sort of roundabout route through history and pointing
out that there's some poetry to this
the first computers word fact mechanical the picture of data just difference engine
which one powered by hand crank hoods all fit order polynomials
if you best for the hundred years you get any which was the first fully
digital electronic computer
it way twenty tons consumed hundred fifty kilowatt power and perform the blazing five syllables
per second of floating point operations
now the difference between these two computers points out a tension in computer design that's
been around since the eighteen forties which is between
high powered high performance and lower power and lower performance
fortunately we broke some of the design tradeoffs in a pretty significant way since the
seventies
but this tension actually resurfaced in a very significant way around two thousand
this plot is a prediction of how computer power would increase from two thousand two
thousand and ten and you lotus people expected power would increase a lot
this obviously didn't happen as indicated by some of the annotations on the slide
we don't have nuclear reactors in our laptops
the question is how this happened
then is actually prepared is very well for this the idea is that transistors have
a property called their threshold voltage
and if you just the threshold voltage properly you can trade off between two kinds
of energy their dissipated in
a very necessary energy called dynamic energy which has to do with running computer and
weighted energy called leakage energy
and
by setting the threshold voltage properly you can actually find the minimum between them and
make them operate perfectly
this is actually what happened between two thousand two thousand ten in this why we
have many course not computers right now uptalk three that based on the cartoon on
the right the slide
the idea is that if you were operating at the one x point in that
cartoon you're consuming lots of dynamic energy above the optimum however you can slow yourself
down in order to save energy
and then stick to computers next to each other in order to recover your performance
and let the software engineers figure out what to do with two computers
you can do that again in this part you going from to x the four
x parallelism in order to save energy but once you're cores at that point
running it's lower won't save you any energy
and so as a result
you could this ceiling on parallelism which limits our ability to improve computing performance going
forward
now at this point my group likes it
turns sharply in the left field and says
the problem here is the transistor
if we can replace that with something that doesn't have this wasted leakage energy
then we can continue improving computing performance by scaling or voltage forever until we get
some other physical them
and what significant about this idea is that we've succeeded in building it
this is a cartoon of the device that we build the ideas you have a
piece of metal thing up near suspended by spring
when you put a voltage on that piece of metal at home towards the surface
and decorations on the bottom of connect different points on your chip
the other significant things
after we built that we measured it and found out that it has immeasurably low
leakage as near as we can tell it has not
so this means that we can replace the car drawing from the previous slide with
the drawing in the bottom corner of this one where there's no leakage energy and
we can just keep scaling forever
now does this mean it's a good idea not necessarily
we can call devices are big
and their slow compared to electrons
so there's a chance that we get a very energy efficient terribly performing computers if
we tried use these
however we've done a lot of very interesting work with circuit design in order to
mitigate that
problem in particular
by changing our design style from stacking series of gates next to each other making
very large distributed gates where all the input sit at the same time therefore all
mechanical delay is incurred at the same time
we can improve both are performance and advice count to save power energy and delay
this is even more significant because we built some of these things we demonstrated that
it is possible to get the functionality that we've been talking about and we don't
a lot of extensive simulations showing that we can improve performance
a result test case wasn't adder which we demonstrated twenty ten and simulations show that
you can get ten x m for improvement over the absolute best transistor could do
in terms of energy from the a ten x
delay penalty
we've also built a microprocessor and this other stuff coming the future including optimize memory
structures minorities