Superconducting synapse
Superconducting synapse.
Scientists have built a superconducting switch, that mimics the behaviour,
of a synapse between neurons.
This artificial synapse could help advance the development of computing systems,
that behave and learn like the human brain.
They say that neurons that fire together, wire together.
This is a nice rhyming way to say, that neurons that actively signal in a group,
tend to develop stronger connections with each other.
From an evolutionary view point, this makes sense, because it facilitates learning.
During new experiences, like learning to walk, speak a new language,
or making a mental map, neurons make new connections.
The more an activity is repeated, or the more learning involved,
the stronger these collections become.
The neurons don’t actually physically touch.
At the junction where the neurons meet there is a tiny gap called the synapse.
This tiny gap is a few nanometers.
In order for information flow from one neuron to another, neurotransmitters are released,
at the end of a long tail-like axon, of one neuron.
They flow across the gap and are picked up by receptors on the other neuron.
This in turn can trigger a electrical signal that will be transmitted along the second neuron.
If the signal from the first neuron isn’t very strong,
your second neuron won’t become activated.
Every connection is not going to be useful, and it is not in the brain’s best interest,
for every connection to be hypersensitive and easily activated.
However, neurons that frequently signal together,
become much more in tune with each other,
so much so that the threshold for activation becomes lower.
With repeated interactions,
the second neuron becomes much more sensitive to signals from the first neuron.
Consequently, well used neuronal pathways, that involve such sensitive neurons,
fire much more quickly.
You become better at walking, speaking a new language, or finding your way home.
Neuron pathways are able to adapt this way, because synapses can learn.
As firing becomes more frequent,
the first neuron can increase the number of neurotransmitters it sends,
and/or the second neuron becomes more sensitive to the neurotransmitters.
This involves changes in the way certain genes are regulated in response to regular use.
Researchers are working on building computers that function much like the brain.
To do that they want to mimic the way neurons behave and respond to one another.
Adaptable connections are important for this.
The problem is, how to achieve the same kind of learning mechanism in an artificial synapse.
It needs to have a lower threshold for activation, for connections used frequently.
The challenge is how to get an artificial synapse,
to learn without the benefit of finely tuned genetic and biochemical feedback mechanisms.
To address this scientists used what is called as a Josephson junction.
Instead of two electrically active neurons meeting at a synapse,
this junction involves two superconductors separated by a insulator.
The junction they built was extremely small, only 10 micrometers in diameter.
When a sufficient current runs through the junction, it produces low voltage spikes.
This is analogous to how a sufficient electrical signal,
in the first neuron can trigger a spike of electrical activity in the second neuron.
This is a first step in mimicking a neuronal connection.
But what about the synapse learning effect?
Here is where it gets really interesting.
The researchers filled the insulating gap with magnetic nanoclusters,
to the tune of 20000 per square micrometer.
Each of these nanoclusters behave like a tiny bar magnet polarised along an axis.
Filling the gap between the two superconductors, this multitude of tiny magnets,
begin in a disordered jumble, with their magnetic spins, pointing all over the place.
When they are in this magnetically disordered state, they tend to put a damper,
on any current trying to cross the gap between the superconductor.
Consequently the amount of current needed to trigger a voltage spike is quite high.
However, in the presence of a magnetic field, repeated pulses of current will cause,
all the nanoclusters to align in the same magnetic direction.
The more aligned they become, the lower is the amount of current,
needed to trigger a voltage spike.
In other words frequent electrical activity at this junction,
leads to a more sensitive interaction between the two superconductors.
When we have a 100watt light bulb in a lamp, and turn it on for one second,
we have just used 100 joules.
A neuron is so energy efficient that it would have to fire around 10 to the power of 17 times,
to match that amount of energy use.
This is because a single synaptic event only uses around 10 femtojoules of energy.
For comparison, the amount of energy needed for the artificial synapse to function,
is always less than one attojoule.
This is about 10000 times higher than the energy used by a natural synapse.
It is still pretty impressive.
In terms of speed though, the artificial synapse wins by a long shot.
A neuronal synapse may fire a few hundred times a second.
The artificial synapse was able to fire faster, at a rate of 100 million times a per second.
This coupled with a small size, and a design enabling 3D stacking of these junctions,
suggest that a significant amount of computational complexity,
could be achieved in a very small amount of space.
This leads to a lot of interesting possibilities.