In the near future, organic circuits that mimic biological neurons may have the power to boost processing speed and might even be able to connect directly to the brain.
There is no doubt that the human brain is an amazing computer. This tiny device weighs approximately three pounds, can process information a thousand times faster than supercomputers, can store a thousand times more information than powerful laptops, and uses no more energy than a 20-watt lightbulb.
Soft, flexible organic materials are being tested by researchers in an
attempt to replicate this success. They have the capability of operating like
biological neurons, and someday might even be able to interconnect with them.
It is likely that computer chips made of soft "neuromorphic" material
will be implanted directly into the brain in the future, allowing people to
think about controlling artificial arms and monitors using only their minds. In
contrast to conventional computer chips, these devices send and receive chemical
as well as electrical signals, just like real neurons do.
“Your brain works by releasing chemicals such as dopamine and serotonin
through neurotransmitters. We have developed materials that can interact
electrochemically with them,” says Alberto Salleo,
whose article in the 2021 Annual Review of Materials Research
discusses the possibilities for organic neuromorphic devices.
Using these soft organic materials, Salleo and other researchers have
developed electronic devices that work like transistors (which amplify
electrical signals and switch them) and memory cells (which store
information). Taking inspiration from the way human neuronal connections,
or synapses, work, the team developed neuromorphic computer circuits. It's more
like the brain's network of neurons than the circuits in digital computers,
whether they're made of silicon, metal, or organic materials.
There is a fundamental distinction between calculation and memory on
conventional digital computers since they work one step at a time. It creates
a bottleneck
for speed and energy consumption since one and zero have to be shuttled across
the processor.
The brain is responsible for various functions. The electrical state of one
neuron is affected by the signals received from many other neurons. By
integrating all the signals it has received, neurons serve both as calculating
devices and as memory devices: storing the value of all of these combined
signals as an infinitely variable analog value, rather than a zero or one value
like digital computers.
There have been a number of different "memristive"
devices developed by researchers that mimic this ability. Electrical resistance
is changed when current is passed through them. In the same way that biological
neurons calculate by adding up all the currents they have been exposed to,
these devices also do the same. As a result, they remember the value that their
resistance takes as a result.
For example, an organic memristor is made up of two layers of materials that
conduct electricity. In the presence of an electric current, positive ions are
driven from one layer into the other, thereby altering how easily the second
layer conducts electricity when faced with the same current again. "It
allows physics to perform the computation," explains Matthew Marinella, a computer
engineer at Arizona State University in Tempe who is researching neuromorphic
computing.
Moreover, the technique allows the computer to be liberated from the
constraints of strictly binary values. “In classical computer memory, there is
either a zero or a one. As a result of our work, we made a memory that can have
any value between zero and one. In this way, it can be tuned
analogically," Salleo explains. Most memristors and related devices are
made from silicon chips and do not contain organic materials. Artificial
intelligence programs even use some of these to speed up their performance. Yet
organic components can work faster and use less energy, Salleo says. It would
be even better if they were integrated into your brain. These materials are
soft and flexible, and they interact with biological neurons because they have
electrochemical properties.
French engineer Francesca Santoro, currently at RWTH Aachen University in
Germany, is creating a polymer device that
can learn from real cells. Unlike real neurons, the cells in her device are
separated from the artificial neuron by a small space, like synapses. In response
to the release of dopamine, a chemical that signals nerves, the artificial half of the device change its
electrical state. Similar to the electrical state of biological neurons, the
artificial neuron changes as the amount of dopamine produced by the cells
increases. “We are interested in designing an electronic device that looks like
a neuron and behaves like a neuron,” Santoro shared. It is possible to
employ this approach to improve the control of prosthetics or computer monitors using
brain activity. Electrodes on today's systems can only detect broad
patterns of electrical activity, as they only use standard electronics. To
operate the equipment, external computers are required. At least two ways could
be improved with flexible, neuromorphic circuits. Their ability to translate
neural signals at a much more granular level, responding to individual neurons,
would be greatly enhanced. Furthermore, Salleo says, the devices might also be
able to perform some computations themselves, saving energy and speeding up
processing. Salleo and Santoro believe that low-level, decentralized
neuromorphic computing systems are promising avenues for neuromorphic computing
involving small, neuromorphic computers. “Because they closely resemble the
electrical operation of neurons, they are excellent for electrically and
physically coupling with neuronal tissue,” Santoro says, “not to mention the
brain.”
An original version of this article was first published in Knowable Magazine, an independent publication from Annual Reviews.