The scientists and researchers who try to
build computers that attempt to duplicate some of the brain’s capabilities, even
crudely, have long faced a significant problem. Our best conventional technology
looks nothing like the biological system it attempts to emulate, nor is this
simply an issue of scale. Computer processors are built on 2D planar silicon,
connect via controller hubs (both on-die and across server nodes), and use a
simple binary system for determining whether or not a given transistor is on or
off.
People tend to think of the problem as
being one of scale. It isn’t. As John Hewitt explored in an article in January, the
problem isn’t that transistors are too big, it’s that neurons connect to each
other in 3D and are decidedly non-binary. The release of neurotransmitters in
the brain is governed by the movement of calcium ions across cell membranes.
Large influxes of calcium into the synapse produce larger downstream
effects.
IBM researchers have detailed a new
discovery that brings us one step closer to bridging the gap between synapses
and silicon. The research team has detailed a method for transforming an
insulative layer into a conductive material by exposing it to a charged fluid.
VO2 (vanadium(IV) oxide) is a compound with a particularly odd (and interesting)
habit. It transforms from an insulator to a conductor depending on its
temperature. That’s the sort of capability that makes scientists giddy, but it’s
just the starting point for what the IBM team found.
By exposing the VO2 thin film to an ionic
fluid, the scientists were able to stabilize the metallic phase of VO2 down to
five degrees Kelvin. Normally, VO2 is an insulator below 340K (68C) and
metallic/conductive at 68C or above. The previous explanation for this dramatic
change in behavior is called electrolyte gating. This theory posited that the
dramatic change in VO2′s transition capabilities was caused by the introduction
of the ionic liquid into the gate structure.
The research team tested this by cleaning
the heck out of their test substrate. The VO2 thin film was examined using X-ray
photoelectron spectroscopy (XPS) — no fluid was found. The treated VO2 film,
meanwhile, could still be flipped between low and high conductance by a
sufficient voltage change. The team confirmed its findings on VO2 thin films
over different substrates, to make certain that particular properties of the
underlying material weren’t the cause of the results.
Image credit: New
York Times
The next step, according to the team, is to
try and create larger fluidic circuits that flip on or off depending on local
fluid concentrations. “We could form or disrupt connections just in the same way
a synaptic connection in the brain could be remade, or the strength of that
connection could be adjusted,” Dr. Parkin told the New York Times. Parkin believes the team
will likely tackle a small memory array next.
What’s exciting about this work isn’t the
short-term implications, but the long-term goals. It’s extremely difficult
to model the behavior and function of a system if you can’t build a
representative model of it. The Blue Brain project is one of the world’s leading
efforts to simulate neuronal structure. The last major project milestone was the
simulation of a cellular mesocircuit with 100 neocortical columns and a million
cells in total. Doing so required the use of an IBM Blue Gene/P, one of the most
power-efficient supercomputers in existence. At present, simulating one
simplified component of a rat brain requires multiple orders of magnitude more
power than an organic brain uses.
And that’s why advances like this matter.
The ability to modify a material’s insulative properties without applying
electricity could be critical to future attempts to scale brain modeling
downward. Creating circuits that model synapse functions (even if they do so
imperfectly and very simply) can help us understand how their biological
counterparts function. It could dramatically reduce the power consumption (and
waste heat) generated by such attempts, just as the advent of modern
semiconductor manufacturing reduced computers from structures that fit into
warehouses to pockets.
It’s an exciting, if small, step in the
right direction.
By: Joel Hruska
Aucun commentaire:
Enregistrer un commentaire